Visual Speech Animation: Lei Xie, Lijuan Wang, and Shan Yang

Visual Speech Animation
Lei Xie, Lijuan Wang, and Shan Yang
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
A Typical VSA System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Face/Mouth Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Input and Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Mapping Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
A Deep BLSTM-RNN-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
LSTM-RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
The Talking Head System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Performances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Selected Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Karaoke Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Technology Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
L. Xie (*)
School of Computer Science, Northwestern Polytechnical University (NWPU), Xi’an, P. R. China
e-mail: lxie@nwpu.edu.cn; xielei21st@gmail.com
L. Wang
Microsoft Research, Redmond, WA, USA
e-mail: lijuanw@microsoft.com
S. Yang
School of Computer Science, Northwestern Polytechnical University, Xi’an, China
e-mail: syang@nwpu-aslp.org
# Springer International Publishing AG 2016 1

B. Müller, S.I. Wolf (eds.), Handbook of Human Motion,
DOI 10.1007/978-3-319-30808-1_1-1
2 L. Xie et al.
Abstract
Visual speech animation (VSA) has many potential applications in human-
computer interaction, assisted language learning, entertainments, and other
areas. But it is one of the most challenging tasks in human motion animation
because of the complex mechanisms of speech production and facial motion. This
chapter surveys the basic principles, state-of-the-art technologies, and featured
applications in this area. Specifically, after introducing the basic concepts and the
building blocks of a typical VSA system, we showcase a state-of-the-art approach
based on the deep bidirectional long short-term memory (DBLSM) recurrent
neural networks (RNN) for audio-to-visual mapping, which aims to create a
video-realistic talking head. Finally, the Engkoo project from Microsoft is
highlighted as a practical application of visual speech animation in language
learning.
Keywords
Visual speech animation • Visual speech synthesis • Talking head • Talking face •
Talking avatar • Facial animation • Audio visual speech • Audio-to-visual map-
ping • Deep learning • Deep neural network
Introduction
Speech production and perception are both bimodal in nature. Visual speech, i.e.,
speech-evoked facial motion, plays an indispensable role in speech communication.
Plenty of evidence shows that voice and face reinforce and complement each other in
human-human communication (McGurk and MacDonald 1976). By viewing the
speaker’s face (and mouth), valuable information is provided for speech perception.
Visible speech is particularly effective when auditory speech is degraded or contam-
inated, due to acoustic noise, bandwidth limitation, or hearing impairment. In an
early study, Breeuwer et. al. (Breeuwer and Plomp 1985) have already shown that
the recognition of short sentences that have been band-pass filtered improves
significantly when subjects are allowed to watch the speaker. The same level of
improvement can be observed from hearing-impaired listeners and cochlear implant
patients (Massaro and Simpson 2014). In the experiments, lipreading provides
essential speech perceptual information. The influence of visual speech is not only
limited to situations with degraded auditory input. In fact, Sumby and Pollack found
that seeing the speaker’s face is equivalent to about 15dB signal-to-noise ratio (SNR)
improvement of acoustic signal (Sumby and Pollack 1954).
Due to the influence of visual speech in human-human speech communication,
researchers have shown their interest in its impact in human-machine interaction.
Ostermann and Weissenfeld (Ostermann and Weissenfeld 2004) have shown that
trust and attention of humans toward machines increase by 30% when communicat-
ing with a talking face instead of text only. That is to say, visual speech is able to
attract the attention of a user, making the human-machine interface more engaging.
Visual Speech Animation 3
Hence, visual speech animation (VSA)1 aims to animate the lips/mouth/articulators/

face synchronizing with speech for different purposes. In a broad sense, VSA may
include facial expressions (Jia et al. 2011; Cao et al. 2005) and visual prosody
(Cosatto et al. 2003) like head (Ben Youssef et al. 2013; Le et al. 2012; Busso
et al. 2007; Ding et al. 2015; Jia et al. 2014) and eye (Le et al. 2012; Dziemianko
et al. 2009; Raidt et al. 2007; Deng et al. 2005) motions, which are naturally
accompanied with human speech. Readers can read Chapter Eye Motion and
Chapter Head Motion Generation for more details.
Applications of VSA can be found across many domains, such as technical support
and customer service, communication aids, speech therapy, virtual reality, gaming,
film special effects, education, and training (Hura et al. 2010). Specific applications
may include a virtual storyteller for children, a virtual guider or presenter for personal
or commercial Web site, a representative of user in computer games, and a funny
puppetry for computer-mediated human communications. It is clearly promising that
VSA will become an essential multimodal interface in many applications.
Speech-evoked face animation is one of the most challenging tasks in human
motion animation. Human face has an extremely complex geometric form (Pighin
et al. 2006), and the speech-originated facial movements are the result of a compli-
cated interaction between a number of anatomical layers that include the bone,
muscle, fat, and skin. As a result, humans are extremely sensitive to the slightest
artifacts in an animated face, and even the small subtle changes can lead to unreal-
istic appearance. To achieve realistic visual speech animation, tremendous efforts
from speech, image, computer graphics, pattern recognition, and machine learning
communities have been made since several decades ago (Parke 1972). Those efforts
have been summarized in proceedings of visual speech synthesis challenge (LIPS)
(Theobald et al. 2008), surveys (Cosatto et al. 2003; Ostermann and Weissenfeld
2004), featured books (Pandzic and Forchheimer 2002; Deng and Neumann 2008),
and several journal special issues (Xie et al. 2015; Fagel et al. 2010). This book
chapter aims to introduce the basic principles, survey the state-of-the-art technolo-
gies, and discuss featured applications.
State of the Art
After decades of research, a state-of-the-art visual speech animation system currently

can realize lifelike or video-realistic performance through 2D, 2.5D, or 3D face
modeling and a statistical/parametric text/speech to visual mapping strategy. For
instance, in Fan et al. (2016), an image-based 2D video-realistic talking head is
introduced. The lower face region of a speaker is modeled by a compact model learned
from a set of facial images, called active appearance model (AAM). Given pairs of the
audio and visual parameter sequence, a deep neural network model is trained to learn
1
Also called visual speech synthesis, talking face, talking head, talking avatar, speech animation,
and mouth animation.
4 L. Xie et al.
Video/ Visual
Sensor Data Face/Mouth Parameters
Model
Data Learn
Collection Mapping
Feature
Audio/Text Extraction Audio/Text
Feature
Audio/Text
Feature
New Feature Face/Mouth
Mapping Animation
Audio/Text Extraction Model
Visual
Parameters
Fig. 1 The building blocks of a typical VSA system
the sequence mapping from audio to visual space. To further improve the realism of the
proposed talking head, the trajectory tiling method is adopted to use the predicted AAM
trajectory as a guide to select a smooth real sample image sequence from the recorded
database. Based on the similar techniques, Microsoft has released an online visual
speech animation system that can help users to learn English (Wang et al. 2012c).
A Typical VSA System
As shown in Fig. 1, a typical visual speech animation system is usually composed of

several modules: data collection, face/mouth model, feature extraction, and learning
a mapping model.
Data Collection
According to the source of data used for face/mouth/articulator modeling, a VSA

system can be built from images, video recordings, and various motion capture
equipments like mocap, electromagnetic articulography (EMA), magnetic resonance
imaging (MRI), and X-ray. What type of data are collected essentially depends on
the costs, the desired appearance of the face/head, and the application needs. Many
approaches choose the straightforward way for data collection: videos of a speaker
are recorded by a camera, and the image sequences are used as the source for 2D or
3D face/head modeling (Theobald et al. 2008; Bregler et al. 1997; Cosatto et al.
2003; Cosatto and Graf 1998; Xie and Liu 2007a; Wang et al. 2010a; Cosatto and
Graf 2000; Anderson et al. 2013; Ezzat et al. 2002; Ezzat and Poggio 2000; Xie and
Liu 2007b), as shown in Fig. 2a. A recent trend to produce quality facial animation is
to use 3D motion-captured data (Deng and Neumann 2008), which have been
successfully used in movie special effects to drive a virtual character. As shown in
Fig. 2b, to record facial movements, an array of high-performance cameras is utilized
a b c
Camera Video Motion Capture EMA
Fig. 2 Various data collection methods in building a visual speech animation system. (a) Camera
video (Theobald et al. 2008), (b) motion capture (Busso et al. 2007), and (c) EMA from http://www.
gipsa-lab.grenoble-inp.fr/
to reconstruct the 3D marker locations on a subject’s face. Although the mocap

system is quite expensive and difficult to set up, the reconstructed data provide
accurate timing and motion information. Once the data are collected, facial anima-
tion can be created by controlling underlying muscle structure or blend shapes (see
Chapter Blendshape Facial Animation for details).
Another data collection system, EMA, as shown in Fig. 2c, is often used to record
the complex movements of the lips, jaw, tongue, and even intraoral articulators
(Richmond et al. 2011). The sensors, called coils, are attached to different positions
on a speaker’s face or in the mouth, and the 3D movements of the sensors are
collected in a high frame rate (e.g., 200 frames per second) during the speaker’s
talking. Visual speech animation generated by the EMA data is usually used for
articulation visualization (Huang et al. 2013; Fagel and Clemens 2004; Wik and
Hjalmarsson 2009). In Wang et al. (2012a), an animated talking head is created
based on the EMA articulatory data for the pronunciation training purposes.
Face/Mouth Model
The appearance of a visual speech animation system is determined by the underlying

face/mouth model, while generating animated talking heads that look like real
people is challenging. The existing approaches to talking heads use either image-
based 2D models (Seidlhofer 2009; Zhang et al. 2009) or geometry-based 3D ones
(Musti et al. 2014).Cartoon avatars are relatively easier to build. The more human-
like, realistic avatars, which can be seen in some games or movies, are much harder
to build. Traditionally, expensive motion capture systems are required to track the
real person’s motion or, in an even more expensive way, have some artists to
manually hand touch every frame. Some desirable features of the next generation
avatar are as follows: it should be a 3D avatar to be integrated easily into a versatile
3D virtual world; it should be photo-realistic; it can be customized to any user; last
but not least, an avatar should be automatically created with a small amount of
recorded data. That is to say, the next generation avatar should be 3D, photo-realistic,
personalized or customized, and easy to create with little bootstrapping data.
6 L. Xie et al.
In facial animation world, a great variety of different animation techniques based on

3D models exist (Seidlhofer 2009). In general, these techniques first generate a 3D face
model consisting of a 3D mesh, which defines the geometry shape of a face. For that a
lot of different hardware systems are available, which range from 3D laser scanners to
multi-camera systems. In a second step, either a human-like or cartoon-like texture may
be mapped onto the 3D mesh. Besides generating a 3D model, animation parameters
have to be determined for the later animation. A traditional 3D avatar requires a highly
accurate geometric model to render soft tissues like lips, tongue, wrinkles, etc. It is both
computationally intensive and mathematically challenging to make or run such a model.
Moreover, any unnatural deformation will make the resultant output fall into the
uncanny valley of human rejection. That is, it will be rejected as unnatural.
Image-based facial animation techniques achieve great realism in synthesized videos
by combining different facial parts of recorded 2D images (Massaro 1998; Zhang et al.
2009; Eskenazi 2009; Scott et al. 2011; Badin et al. 2010). In general, image-based
facial animations consist of two main steps: audiovisual analysis of a recorded human
subject and synthesis of facial animation. In the analysis step, a database with images of
deformable facial parts of the human subject is collected, while the time-aligned audio
file is segmented into phonemes. In the second step, a face is synthesized by first
generating the audio from the text using a text-to-speech (TTS) synthesizer. The TTS
synthesizer sends phonemes and their timing to the face animation engine, which
overlays facial parts corresponding to the generated speech over a background video
sequence. Massaro (1998), Zhang et al. (2009), Eskenazi (2009), and Scott et al. (2011)
show some image-based speech animation that cannot be distinguished from recorded
video. However, it is challenging to change head pose freely or to render different facial
expressions. Also, it is hard to blend it into 3D scenes seamlessly. Image-based
approaches have their advantages that the photo realistic appearance is guaranteed.
However, a talking head needs to be not just photo-realistic in a static appearance but
also exhibit convincing plastic deformations of the lips synchronized with the
corresponding speech, realistic head movements, and natural facial expressions.
An ideal 3D talking head can mimic realistic motion of a real human face in 3D
space. One challenge for rendering realistic 3D facial animation is on the mouth area.
Our lip, teeth, and tongue are mixed with nonrigid tissues, and sometimes with
occlusions. This means accurate geometric modeling is difficult, and also it is hard to
deform them properly. Moreover, they need to move together in sync with spoken audio;
otherwise, people can observe the asynchrony and think it unnatural. In real world, when
people talk, led by vocal organs and facial muscles, both the 3D geometry and texture
appearance of the face are constantly changing. Ideally, we can capture both geometry
change and texture change simultaneously. There is lot of ongoing research for solving
this problem. For example, with the help of Microsoft Kinect kinds of motion sensing
device, people try to use the captured 3D depth information to better acquire the 3D
geometry model. On the other hand, people try to recover the 3D face shape from single
or multiple camera views (Wang et al. 2011; Sako et al. 2000; Yan et al. 2010).
In 2.5D talking head as above, as there is no captured 3D geometry information
available, they adopt the work in Sako et al. (2000) which reconstructs a 3D face model
from a single frontal face image. The only required input to the 2D-to-3D system is a
frontal face image of a subject with normal illumination and neutral expression. A
semi-supervised ranking prior likelihood models for accurate local search and a robust
parameter estimation approach are used for face alignment. Based on this 2D alignment
algorithm, 87 key feature points are automatically located, as shown in Fig. 3. The
feature points are accurate enough for face reconstruction in most cases. A general 3D
face model is applied for personalized 3D face reconstruction. The 3D shapes have
been compressed by the PCA. After the 2D face alignment, the key feature points are
used to compute the 3D shape coefficients of the eigenvectors. Then, the coefficients
are used to reconstruct the 3D face shape. Finally, the face texture is extracted from the
input image. By mapping the texture onto the 3D face geometry, the 3D face model for
the input 2D face image is reconstructed. They reconstruct a 3D face model for each 2D
image sample in recordings, as examples shown in Fig. 3. Thus a 3D sample library is
formed, where each 3D sample has a 3D geometry mesh, a texture, and the
corresponding UV mapping which defines how a texture is projected onto a 3D
model. After 2D-to-3D transformation, original 2D sample recordings turn into 3D
sample sequences, which consist of three synchronous streams: geometry mesh
sequences for depicting the dynamic shape, texture image sequences for the changing
appearance, and the corresponding speech audio. This method combines the best of
both 2D image sample-based and 3D model-based facial animation technologies. It
renders realistic articulator animation by wrapping 2D video images around a simple
and smooth 3D face model. The 2D video sequence can capture the natural movement
of soft tissues, and it helps the new talking head to bypass the difficulties in rendering
occluded articulators (e.g., tongue and teeth). Moreover, with the versatile 3D geom-
etry model, different head poses and facial expressions can be freely controlled. The
2.5D talking head can be customized to any user by using the 2D video of the user.
Techniques based on 3D models impress by their great automatism and flexibility
while lacking in realism. Image-based facial animation achieves photo-realism while
having little flexibility and lower automatism. The image-based techniques seem to be
the best candidates for leading facial animation to new applications, since these tech-
niques achieve photo-realism. The image-based technique combined with a 3D model
generates photo realistic facial animations, while providing some flexibility to the user.
Input and Feature Extraction
According to the input signal, a visual speech animation system can be driven by
text, speech, and performance. The simplest VSA aims to visualize speech pro-
nunciations by an avatar from tracked makers of human performance. Currently,
performance-based facial animation can be quite realistic (Thies et al. 2016; Wang
and Soong 2012; Weise et al. 2011). The aim of such a system is usually not only for
speech visualization. For example, in Thies et al. (2016), an interesting application
for real-time facial reenactment is introduced. Readers can go through
Chapter Video-based Performance Driven Facial Animation for more details.
During the facial data collection process, speech and text are always collected as
well. Hence, visual speech can be driven by new voice or text input, achieved by a
8 L. Xie et al.
Fig. 3 Auto-reconstructed 3D face model in different mouth shapes and in different view angles
(w/o and w/ texture)
learned text/audio-to-visual mapping model that will be introduced in the next

section. To learn such a mapping, a feature extraction module is firstly used to obtain
representative text or audio features. The textual feature is often similar to that used
for TTS system (Taylor 2009), which may include information about phonemes,
syllables, stresses, prosodic boundaries, and part-of-speech (POS) labels. Audio
features can be typical spectral features (e.g., MFCC (Fan et al. 2016)), pitch, and
other acoustic features.
Mapping Methods
Both text- and speech-driven visual speech animation systems desire an input to
visual feature conversion or mapping algorithm. That is to say, the lip/mouth/facial
movements must be naturally synchronized with the audio speech.2 The conversion
is not trivial because of the coarticulation phenomenon of the human speech
production mechanism, which causes a given phoneme to be pronounced differently
depending on the surrounding phonemes. Due to this phenomenon, learning an
audio/text-to-visual mapping becomes an essential task in visual speech animation.
Researchers have devoted much effort in this task, and the developed approaches can
be roughly categorized into rule based, concatenation, parametric, and hybrid.
Rule Based
Due to the limitations of data collection and learning methods, early approaches are
mainly based on hand-crafted mapping rules. In these approaches, the counterpart of
audio phoneme, viseme, is defined as the basic visual unit. Typically, visemes are
manually designed as key images of mouth shapes, as shown in Fig. 4, and empirical
smooth functions or coarticulation rules are used to synthesize novel speech anima-
tions. Ezzat and Poggio propose a simple approach by morphing key images of
visemes (Ezzat and Poggio 2000). Due to the coarticulation phenomenon, morphing
between a set of mouth images is apparently not natural. Cohen and Massaro (Cohen
and Massaro 1993) propose a coarticulation model, where a viseme shape is defined
via dominance functions that are defined in terms of each facial measurement, such
as the lips, tongue tip, etc., and the weighted sum of dominance values determines
the final mouth shapes. In a recent approach, Sarah et. al. (Taylor et al. 2012) argue
that static mouth shapes are not enough, so they redefine visemes as clustered
temporal units that describe distinctive speech movements of the visual speech
articulators, called dynamic visemes.
Concatenation/Unit selection
To achieve photo- or video-realistic animation effect, concatenation of real video clips
from a recorded database has been considered (Bregler et al. 1997; Cosatto et al. 2003;
Cosatto and Graf 1998; 2000). The idea is quite similar with that in concatenative TTS
(Hunt and Black 1996). In the off-line stage, a database of recorded videos is cut into
sizable clips, e.g., triphone units. In the online stage, given a novel text or speech
target, a unit selection process is used to select appropriate units and assemble them in
an optimal way to produce the desired target, as shown in Fig. 5.
To achieve speech synchronization and a smooth video, the concatenation algo-
rithm should be elaborately designed. In Cosatto et al. (2003), a phonetically labeled
target is first produced by a TTS system or by a labeler or an aligner from the
recorded audio. From the phonetic target, a graph is created with states
corresponding to the frames of the final animation. Each state of the final animation
2
Sometimes this task is called lip synchronization or lip sync for short.
10 L. Xie et al.
Fig. 4 Several defined visemes from Ezzat and Poggio (2000)
(a video frame) is populated with a list of candidate nodes (a recorded video sample
from the database). Each state is fully connected to the next, and concatenation costs
are assigned for each arc, while target costs are assigned to each node. A Viterbi
search on the graph finds the optimal path, i.e., the path that generates the lowest
total cost. The balance between the two costs is critical in the final performance, and
its weighting is empirically tuned in real applications.
The video clips for unit selection are usually limited to lower part of the face that
has most speech-evoked facial motions. After selection, the concatenated lower face
clips are stitched to a background whole face video, resulting in the synthesized
whole face video, as shown in Fig. 6. To achieve seamless stitches, much efforts
have been made on image processing. With a relatively large video database, the
concatenation approach is able to achieve video-realistic performance. But it might
be difficult to add different expressions, and the flexibility of the generated visual
speech animation is also limited.
w ih n d ow
Reconstructed Face Images Frame
Image Candidates
Fig. 5 Unit selection approach for visual speech animation (Fan et al. 2016)
Sample
Lower Background Stitched

Face Image Image/Sequence Image/Sequence
Mask
Fig. 6 Illustration of the image stitching process in a video-realistic talking head (Fan et al. 2016)
Parametric/Statistical
Recently, parametric methods have gained much attention due to their elegantly
automatic learned mappings from data. Numerous attempts have been made to
model the relationship between audio and visual signals, and many are generative
probabilistic model based, where the underlying probability distributions of audio-
visual data are estimated. Typical models include Gaussian mixture model (GMM),
hidden Markov model (HMM) (Xie and Liu 2007a; Fu et al. 2005), dynamical
Bayesian network (DBN) (Xie and Liu 2007b), and switching linear dynamical
system (SLDS) (Englebienne et al. 2007).
12 L. Xie et al.
The hidden Markov model-based statistical parametric speech synthesis (SPSS)

has made a significant progress (Tokuda et al. 2007). Hence, the HMM approach was
also intensively investigated for visual speech synthesis (Sako et al. 2000; Masuko
et al. 1998). In HMM-based visual speech synthesis, auditory speech and visual
speech are jointly modeled in HMMs, and the visual parameters are generated from
HMMs by using the dynamic (“delta”) constraints of the features (Breeuwer and
Plomp 1985). Convincing mouth video can be rendered from the predicted visual
parameter trajectories. This approach is called trajectory HMM. Usually, maximum
likelihood (ML) is used as the criterion for HMM model training. However, ML
training does not optimize directly toward visual generation error. To compensate
this deficiency, a minimum generated trajectory error (MGE) method is proposed in
Wang et al. (2011) to further refine the audiovisual joint modeling by minimizing the
error between the generation result and the real target trajectories in the training set.
Although HMM can model sequential data efficiently, there are still some limi-
tations, such as the wrong model assumptions out of necessity, e.g., GMM and its
diagonal covariance, and the greedy, hence suboptimal, search-derived decision-
tree-based contextual state clustering. Motivated by the superior performance of
deep neural networks (DNN) in automatic speech recognition (Hinton et al. 2012)
and speech synthesis (Zen et al. 2013), a neural network-based photo-realistic
talking head is proposed in Fan et al. (2015). Specifically, a deep bidirectional
long short-term memory recurrent neural network (BLSTM-RNN) is adopted to
learn a direct regression model by minimizing the sum of square error (SSE) of
predicting visual sequence from label sequence. Experiments have confirmed that
the BLSTM approach significantly outperforms the HMM approach (Fan et al.
2015). The BLSTM approach will be introduced in detail later in this chapter.
Hybrid
Although parametric approaches have many merits like small footprint, flexibility,
and controllability, one obvious drawback of those approaches is the blurring
animation due to the feature dimension reduction and the non-perfect learning
method. So there are some hybrid visual speech animation approaches that use the
predicted trajectory to guide the sample selection process (Wang et al. 2010b), which
combines the advantages of both the video-based concatenation and the parametric
statistical modeling approaches. In a recent approach (Fan et al. 2016), visual
parameter trajectory predicted by a BLSTM-RNN is used as a guide to select a
smooth real sample image sequence from the recorded database.
A Deep BLSTM-RNN-Based Approach
In the past several years, deep neural networks (DNN) and deep learning methods
(Deng and Yu 2014) have been successfully used in many tasks, such as speech
recognition (Hinton et al. 2012), natural language processing, and computer vision.
For example, the DNN-HMM approach has boosted the speech recognition accuracy
significantly (Deng and Yu 2014). Deep neural networks have been investigated for
regression/mapping tasks, e.g., text to speech (Zen et al. 2013), learning clean speech
from noisy speech for speech enhancement (Du et al. 2014), and articulatory
movement prediction from text and speech (Zhu et al. 2015). There are several
advantages of the DNN approaches: it can model long-span, high-dimensional, and
the correlation of input features; it is able to learn nonlinear mapping between input
and output with a deep-layered, hierarchical, feed-forward, and recurrent structure; it
has the discriminative and predictive capability in generation sense, with appropriate
cost function(s), e.g., generation error.
Recently, recurrent neural networks (RNNs) (Williams and Zipser 1989) and their
bidirectional variant, bidirectional RNNs (BRNNs) (Schuster and Paliwal 1997),
become popular because they are able to incorporate contextual information that is
essential for sequential data modeling. Conventional RNNs cannot well model the
long-span relations in sequential data because of the vanishing gradient problem
(Hochreiter 1998). Hochreiter and Schmidhuber (1997) found that the LSTM archi-
tecture, which uses purpose-built memory cells to store information, is better at
exploiting long-range context. Combining BRNNs with LSTM gives BLSTM,
which can access long-range context in both directions. Speech, both in auditory
and visual forms, is typical sequential data. In a recent study, BLSTM has shown
state-of-the-art performance in audio-to-visual sequential mapping (Fan et al. 2015).
RNN
Allowing cyclical connections in a feed-forward neural network, a recurrent neural

network (RNN) is formed (Williams and Zipser 1989). RNNs are able to incorporate
contextual information from previous input vectors, which allows them to remember
past inputs and allows them to persist in the network’s internal state. This property
makes them an attractive model for sequence-to-sequence learning. For a given input
vector sequence x = (x1,x2...,xT), the forward pass of RNNs is as follows:
ht ¼ HðWxh xt þ Whh ht1 þ bh Þ, (1)
yt ¼ Why ht þ by , (2)
where t = 1,...,T and T is the length of the sequence; h = (h1,...,hT) is the hidden state
vector sequence computed from x; y = (y1,...,yT) is the output vector sequence; W is
the weight matrices, where Wxh, Whh, and Why are the input-hidden, hidden-hidden,
and hidden-output weight matrices, respectively. bh and by are the hidden and output
bias vectors, respectively, and H denotes the nonlinear activation function in the
output layer.
For the visual speech animation task, because of the speech coarticulation
phenomenon, a model is desired to have access to both the past and future contexts.
Bidirectional recurrent neural networks (BRNNs), as shown in Fig. 7, fit this task
14 L. Xie et al.
Outputs
Backward Layer
Forward Layer
Inputs
Fig. 7 Bidirectional recurrent neural networks (BRNNs)
!
well. A BRNN computes both forward state sequence h and backward state
!
sequence h , as formulated below:
! ! !
h t ¼ H Wx !h xt þ W !
h
h h t1 þ b ! ,
h
(3)
!!
h t ¼ H Wx h x t þ W h h
hh þ bh , (4)
t1
! !
yt ¼ W ! h þ W h y h t þ by :
hy t
(5)
LSTM-RNN
Conventional RNNs can access only a limited range of context because of the
vanishing gradient problem. Long short-term memory (LSTM) uses purpose-built
memory cells, as shown in Fig. 8, to store information, which is designed to
overcome this limitation. In sequence-to-sequence mapping tasks, LSTM has been
shown capable of bridging very long time lags between input and output sequences
by enforcing constant error flow. For LSTM, the recurrent hidden layer function H is
implemented as follows:
it ¼ σ ðWxi xt þ Whi ht1 þ Wci ct1 þ bi Þ, (6)

f t ¼ σ Wxf xt þ Whf ht1 þ Wci ct1 þ bf , (7)
at ¼ τðWxc xt þ Whc ht1 þ bc Þ, (8)
ct ¼ f t ct1 þ it at , (9)
Fig. 8 Long short-term

memory (LSTM)
ht
ot Output
gate
ct Forget
ft
Cell gate
it Input
gate
Memory
Block
xt
ot ¼ σ ðWxo xt þ Who ht1 þ Wco ct þ bo Þ, (10)
ht ¼ ot θðct Þ, (11)
where σ is the sigmoid function; i, f, o, a, and c are input gate, forget gate, output
gate, cell input activation, and cell memory, respectively. τ and θ are the cell input
and output nonlinear activation functions; generally tanh is chosen. The multiplica-
tive gates allow LSTM memory cells to store and access information over long
periods of time, thereby avoiding the vanishing gradient problem.
Combining BRNNs with LSTM gives rise to BLSTM, which can access long-
range context in both directions. Motivated by the success of deep neural network
architectures, deep BLSTM-RNNs (DBLSTM-RNNs) are used to establish the
audio-to-visual mapping for visual speech animation. Deep BLSTM-RNN is created
by stacking multiple BLSTM hidden layers.
The Talking Head System
Figure 9 shows the diagram of an image-based talking head system using DBLSTM
as the mapping function (Fan et al. 2015). The diagram actually follows the basic
structure of a typical visual speech animation system in Fig. 1. The aim of the system
is to achieve speech animation with video-realistic effects. Firstly, an audio/visual
database of a subject talking to a camera with frontal view of his/her face is recorded
as the training data. In the training stage, the audio is converted into a sequence of
16 L. Xie et al.
contextual phoneme labels L using forced alignment, and the corresponding lower
face image sequence is transformed into active appearance model (AAM) feature
vectors V. Then a deep BLSTM neural network is used to learn a regression model
between the two audio and visual parallel sequences by minimizing the SSE of the
prediction, in which the input layer is the label sequence L and the output prediction
layer is the visual feature sequence V. In the synthesis stage, for any input text with
natural or synthesized speech by TTS, the label sequence L is extracted from them
and then the visual AAM parameters V ^ are predicted using the well-trained deep
BLSTM network. Finally, the predicted AAM visual parameter sequence V ^ can be
reconstructed to high-quality photo-realistic face images and rendering the full-face
talking head with lip-synced animation.
Label Extraction
The input sequence L and output feature sequence V are two time-varying parallel
sequences. The input of a desired talking head system can be any arbitrary text along
with natural audio recordings or synthesized speech by TTS. For natural recordings,
the phoneme/state time alignment can be obtained by conducting forced alignment
using a trained speech recognition model. For TTS-synthesized speech, the pho-
neme/state sequence and time offset are a by-product of the synthesis process.
Therefore, for each speech utterance, the phoneme/state sequence and their time
offset are converted into a label sequence, denoting as L = (i1,...,it,...,iT), where T is
the number of frames in the sequence. The format of the frame-level label it uses the
one-hot representation, i.e., one vector for each frame, shown as follows:
2 3
40, . . . , 0, . . . , 1; 1, . . . , 0, . . . , 0 ; 0, 0, 1, . . . , 0 ; 0, 1, 0 5,
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl}
K K K 3
where K denotes the number of phonemes. In Fan et al. (2015), triphone and the
information of three states to identify it are used. The first three K-element sub--
vectors denote the identities of the left, current, and right phonemes in the triphone,
respectively, and the last three elements represent the phoneme state. Please note that
the contextual label can be easily extended to contain richer information, like
position in syllable, position in word, stress, part of speech, etc. But if the training
data is limited, we may consider the phoneme and state level labels only.
Face Model and Visual Feature Extraction

In the system (Fan et al. 2015), the visual stream is a sequence of lower face images
which are strongly correlated to the underlying speech. As the raw face image is hard
to model directly due to the high dimensionality, active appearance model (AAM)
(Cootes et al. 2001) is used as face model for visual feature extraction. AAM is a
joint statistical model compactly representing both the shape and the texture varia-
tions and the correlation between them.
Text &
Audio Label Extraction L
Deep BLSTM
A/V NN
Training
Database Face Visual Feature Model
images Extraction
V
Text &
Audio Label L V̂ Image
Prediction
Extraction Reconstruction
Fig. 9 Diagram of an image-based talking head system using DBLSTM-RNN as the mapping (Fan
et al. 2015)
Since the speaker moves his/her head naturally during recording, head pose
normalization among all the face images is performed before AAM modeling.
With the help of an effective 3D model-based head pose tracking algorithm, the
head pose in each image frame is normalized to a fully frontal view and further
aligned. Facial feature points and the texture of a lower face used in Fan et al. (2015)
are shown in Fig. 10.
The shape of the jth lower face, sj, can be represented by the concatenation of the
x and y coordinates of N facial feature points:

sj ¼ xj1 , xj2 , . . . , xjN , yj1 , yj2 ,:::, yjN , (12)
where j = 1,2,..., J and J is the total number of the face images. In this work, a set of
51 facial feature points is used, as shown in Fig. 10a. The mean shape is simply
defined by
XJ
s0 ¼ s =J:
j¼1 j
(13)
Applying principal component analysis (PCA) to all J shapes, sj can be given

approximately by
XNshape
sj ¼ s0 þ i¼1
aji~s i ¼ s0 þ aj Ps , (14)
18 L. Xie et al.
Fig. 10 Facial feature points and the texture of a lower face used in Fan et al. (2015). (a) 51 facial
feature points. (b) The texture of a lower face
T
where Ps ¼ ~s 1 , ~s 2 , . . . , ~s i , . . . ¼ ~s Nshape denotes the eigenvectors corresponding to

the N shape largest eigenvalues and aj ¼ aj1 , aj2 , . . . , aji , . . . , ajN shape is the jth shape
parameter vector.
Accordingly, the texture of the jth face image, tj, is defined by a vector
concatenating the R/G/B value of every pixel that lies inside the mean shape, so

tj ¼ r j1 , . . . , r jU , gj1 , . . . , gjU , bj1 , . . . , bjU , (15)
where j = 1,2,..., J and U is the total number of pixels.

As the dimensionality of the texture vector is too high to use PCA directly,
EM-PCA (Roweis 1998) is used instead to all J textures. As a result, the jth texture
tj can be given approximately by
XNtexture
tj ¼ t0 þ i¼1
bjt~t i ¼ t0 þ bj Pt , (16)
where t0 is the mean texture. Pt contains the eigenvectors corresponding to the

Ntexture largest eigenvalues, and bj is the jth texture parameter vector.
The above shape and texture models can only control the shape and texture
separately. In order to recover the correlation between the shape and the texture, aj
and bj are combined in another round of PCA:
XNappearance
aj , b j ¼ i¼1
vji v~i ¼ vj Pv , (17)
and assuming that Pvs and Pvt are formed by extracting the first Nshape and the last
Ntexture values from each component in Pv. Simply combining the above equations
gives
sj ¼ s0 þ vj Pvs Ps ¼ s0 þ vj Qs , (18)
tj ¼ t0 þ vj Pvt Pt ¼ t0 þ vj Qt : (19)
Now, the shape and texture of the jth lower face image can be constructed by a
single vector vj. vj is the jth appearance parameter vector which is used as the AAM
visual feature. Subsequently, the lower face sequence with T frames can be
represented by the visual feature sequence V = (v1,...,vt,...,vT).
DBLSTM-RNN Model Training

In the training stage, multiple sequence pairs of L and V are available. As both
sequences are represented as continuous numerical vectors, the network is treated as
a regression model minimizing the SSE of predicting V from L. In the synthesis
stage, given any arbitrary text along with natural or synthesized speech, they are
firstly converted into a sequence of input features and then fed into the trained
network. The output of the network is the predicted visual AAM feature sequence.
After reconstructing the AAM feature vectors to RGB images, photo-realistic image
sequence of the lower face is generated. Finally, the lower face is stitched to a
background face and the facial animation of the talking head is rendered.
Learning deep BLSTM network can be regarded as optimizing a differentiable
error function:
XMtrain
EðwÞ ¼ k¼1
Ek ðwÞ, (20)
where Mtrain represents the number of sequences in the training data and w denotes
the network internode weights. In the task, the training criterion is to minimize the
SSE between the predicted visual features V ^ ¼ ð^v 1 , ^v 2 ,:::, ^v T Þ and the ground truth
V = (v1, v2, ... , vT). For a particular input sequence k, the error function takes the
form
XT k 1 XT k
E k ðw Þ ¼ E ¼ ^v k vk 2 , (21)
t¼1 kt 2 t¼1 t t
where Tk is the total number of frames in the kth sequence. In every iteration, the
error gradient is computed with the following equation:
@Eðwðr ÞÞ
Δwðr Þ ¼ mΔwðr 1Þ α , (22)
@wðr Þ
20 L. Xie et al.
where 0 α 1 is the learning rate, 0 m 1 is the momentum parameter, and w

(r) represents the vector of weights after rth iteration of update. The convergence
condition is that the validation error has no obvious change after R iterations.
Backpropagation through time (BPTT) algorithm is usually adopted to train the
network. In the BLSTM hidden layer, BPTT is applied to both forward and back-
ward hidden nodes and back-propagates layer by layer, taking the error function
derivatives
with respect to the output of the network as an example. For
k k k
^v ¼ ^v , . . . , ^v , . . . , ^v k
in the kth V^ , because the activation function
t t1 tj tN appearance
used in the output layer is an identity function, we have
X
^v ktj ¼ woh zkht , (23)
h
where o is the index of the an output node, zkht is the activation of a node in the hidden
layer connected to the node o, and woh is the weight associated with this connection.
By applying the chain rule for partial derivatives, we can obtain
k
@Ekt XNappearance @Ekt @^v tj
¼ , (24)
@woh j¼1 @^v ktj @woh
and according to Eqs. (21) and (23), we can derive
@Ekt XNappearance
k k k
¼ ^v tj v tj zht , (25)
@woh j¼1
@Ekt XT @Ekt
¼ : (26)
@woh t¼1 @w
oh
Performances
The performances of the DBLSTM-based talking head are evaluated on an A/V

database with 593 English utterances spoken by a female in a neutral style (Fan et al.
2015). The DBLSTM approach is compared with the previous HMM-based
approach (Wang and Soong 2015). The results for FBB128 DBLSTM3 and HMM
are shown in Table 1. We can clearly see that the deep BLSTM approach out-
performs the HMM approach by a large margin in terms of the four objective
metrics.
Subjective evaluation is also carried out in Fan et al. (2015). Ten sequences of
labels are randomly selected from the test set as the input. The deep BLSTM-based
3
FBB128 means two BLSTM layers sitting on top of one feed-forward layer and each layer has
128 nodes.
Table 1 Performance RMSE RMSE RMSE

comparison between deep Comparison (shape) (texture) (appearance) CORR
BLSTM and HMM
HMM 1.223 6.602 167.540 0.582
DBLSTM 1.122 6.286 156.502 0.647
45.7% 28.6% 25.7%

DBLSTM-RNNs Neutral HMM
Fig. 11 The percentage preference of the DBLSTM-based and HMM-based photo-real talking
heads
and the HMM-based talking head videos are rendered, respectively. For each test
sequence, the two talking head videos are played side by side randomly with original
speech. A group of 20 subjects are asked to perform an A/B preference test according
to the naturalness. The percentage preference is shown in Fig. 11. It can be seen
clearly that the DBLSTM-based talking head is significantly preferred to the
HMM-based one. Most subjects prefer the BLSTM-based talking head because its
lip movement is smoother than the HMM-based one. Some video clips of the
synthesized talking head can be found from Microsoft Research (2015).
Selected Applications
Avatars, with lively visual speech animation, are increasingly being used to com-
municate with users on a variety of electronic devices, such as computers, mobile
phones, PDAs, kiosks, and game consoles. Avatars can be found across many
domains, such as customer service and technical support, as well as in entertainment.
Some of the many uses of avatars include the following:
• Reading news and other information to users

• Guiding uses through Web sites by providing instructions and advice
• Presenting personalized messages on social Web sites
• Catching users attention in advertisements and announcements
• Acting as digital assistants and automated agents for self-service, contact centers,
and help desks
• Representing character roles in games
• Training users to perform complex tasks
• Providing new branding opportunities for organizations
Here, we focus on one application that uses talking head avatar for audio/visual
computer-assisted language learning (CALL).
Imagine a child learning from his favorite TV star who appears to be personally
teaching him English on his handheld device. Another youngster might show off her
own avatar that tells mystery stories in a foreign language to her classmates. The
22 L. Xie et al.
speech processing technologies “talking head” are notable in its potential for
enabling such scenarios. These features have been successfully tested in a large-
scale DDR project called Engkoo (Wang et al. 2012c), from Microsoft Research
Asia. It is used by ten million English learners in China per month and was the
winner of the Wall Street Journal 2010 Asian Innovation Readers’ Choice Award
(Scott et al. 2011).
Talking head generates karaoke-style short synthetic videos demonstrating oral
English. The videos consist of a photo-realistic person speaking English sentences
crawled from the Internet. The technology leverages a computer-generated voice
with native speaker-like quality and synchronized subtitles on the bottom of the
video; it emulates popular karaoke-style videos specifically designed for a Chinese
audience in order to increase user engagement. Compared to using prerecorded
human voice and video in English education tools, these videos not only create a
realistic look and feel but also greatly reduce the cost of content creation by
generating arbitrary content sources synthetically and automatically. The potential
for personalization is there as well. For example, a choice of voice based on preferred
gender, age, speaking rate, or pitch range and dynamics can be made, and the
selected type of voice can be used to adapt a pre-trained TTS such that the
synthesized voice can be customized.
Motivation
Language teachers have been avid users of technology for a while now. The arrival
of the multimedia computer in the early 1990s was a major breakthrough because it
combined text, images, sound, and video in one device and permitted the integration
of the four basic skills of listening, speaking, reading, and writing. Nowadays, as
personal computers become more pervasive, smaller, and more portable, and with
devices such as smartphones and tablet computers dominating the market, multime-
dia and multimodal language learning can be ubiquitous and more self-paced.
For foreign language users, learning correct pronunciation is considered by many
to be one of the most arduous tasks if one does not have access to a personal tutor.
The reason is that the most common method for learning pronunciation, using audio
tapes, severely lacks completeness and engagement. Audio data alone may not offer
users complete instruction on how to move their mouth/lips to sound out phonemes
that may be nonexistent in their mother tongue. And audio as a tool of instruction is
less motivating and personalized for learners. As supported by studies in cognitive
informatics, information is processed by humans more efficiently when both audio
and visual techniques are utilized in unison.
Computer-assisted audiovisual language learning increases user engagement
when compared to audio alone. There are many existing bodies of work that use
visualized information and talking head to help language learning. For example,
Massaro (1998) used visual articulation to show the internal structure of the mouth,
enabling learners to visualize the position and movement of the tongue. Badin et. al
(2010) inferred learners’ tongue position and shape to provide visual articulatory
corrective feedback in second language learning. Additionally, a number of studies

done in Eskenazi (2009) focused on overall pronunciation assessment and segmen-
tal/prosodic error detection to help learners improve their pronunciation with com-
puter feedback.
In the project in Wang et al. (2012c), the focus is on generating a photo-realistic,
lip-synced talking head as a language assistant for multimodal, web-based, and
low-cost language learning. The authors feel that a lifelike assistant offers a more
authoritative metaphor for engaging language learners, particularly younger demo-
graphics. The long-term goal is to create a technology that can ubiquitously help
users anywhere, anytime, from detailed pronunciation training to conversational
practice. Such a service is especially important as a tool for augmenting human
teachers in areas of the world where native, high-quality instructors are scarce.
Karaoke Function
Karaoke, also known as KTV, is a major pastime among Chinese people, with
numerous KTV clubs found in major cities in China. A karaoke-like feature has
been added to Engkoo, which enables English learners to practice their pronuncia-
tion online by mimicking a photo-realistic talking head lip-synchronously within a
search and discovery ecosystem. This “KTV function” is exposed as videos gener-
ated from a vast set of sample sentences mined from the web. Users can easily launch
the videos with a single click at the sentence of their choosing. Similar to the karaoke
format, the videos display the sentence on the screen, while a model speaker says it
aloud, teaching the users how to enunciate the words, as shown in Fig. 12. Fig. 13
shows the building blocks of the KTV system.
While the subtitles of karaoke are useful, it should be emphasized that the pacing
offered is especially valuable when learning a language. Concretely, the rhythm and
the prosody embedded in the KTV function offer users the timing cues to utter a
given sentence properly. Although pacing can be learned from listening to a native
speaker, what is offered uniquely in this system is the ability to get this content at
scale and on demand.
The KTV function offers a low-cost method for creating highly engaging,
personalizable learning material utilizing the state-of-the-art talking head rendering
technology. One of the key benefits is the generation of lifelike video as opposed to
cartoon-based animations. This is important from a pedagogical perspective because
the content appears closer in nature to a human teacher, which reduces the perceptive
gap that students, particularly younger pupils, need to make from the physical
classroom to the virtual learning experience.
The technology can drastically reduce language learning video production costs
in scenarios where the material requires a human native speaker. Rather than
repeatedly taping an actor speaking, the technique can synthesize the audio and
video content automatically. This has the potential for further bridging the classroom
and e-learning scenarios where a teacher can generate his talking head for students to
take home and learn from.
24 L. Xie et al.
Fig. 12 Screenshots of Karaoke-like talking heads on Engkoo. The service is accessible at http://
dict.bing.com.cn
Fig. 13 Using talking head synthesis technology for KTV function on Engkoo
Technology Outlook
The current karaoke function, despite its popularity with web users, can be further
enhanced to reach the long-term goal, the vision being that of creating an
indiscernibly lifelike computer assistant, at low cost and web based, helpful in
many language learning scenarios, such as interactive pronunciation drills and
conversational training.
To make the talking head more lifelike and natural, a new 3D photo-realistic, real-
time talking head is proposed with a personalized appearance (Wang and Soong
2012). It extends the prior 2D photo-realistic talking head to 3D. First, approxi-
mately 20 minutes of audiovisual 2D video is recorded with prompted sentences
spoken by a human speaker. A 2D-to-3D reconstruction algorithm is adopted to
automatically wrap the
3D geometric mesh with 2D video frames to construct a training database, as
shown in Fig. 14. In training, super feature vectors consisting of 3D geometry,
texture, and speech are formed to train a statistical, multi-streamed HMM. The
model is then used to synthesize both the trajectories of geometry animation and
dynamic texture.
As far as the synthesized audio (speech) output is concerned, the research
direction is to make it more personalized, adaptive, and flexible. For example, a
new algorithm which can teach the talking head to speak authentic English sentences
which sound like a Chinese ESL learner has been proposed and successfully tested.
Also, synthesizing more natural and dynamic prosody patterns for ESL learners to
mimic is highly desirable as an enhanced feature of the talking head.
The 3D talking head animation can be controlled by the rendered geometric
trajectory, while the facial expressions and articulator movements are rendered
with the dynamic 2D image sequences. Head motions and facial expressions can
also be separately controlled by manipulating corresponding parameters. A talking
head for a movie star or celebrity can be created by using their video recordings.
With the new 3D, photo-realistic talking head, the era of having lifelike, web-based,
and interactive learning assistants is on the horizon.
The phonetic search can be further improved by collecting more data, both in text
and speech, to generate the phonetic candidates to cover the generic and localized
spelling/pronunciation errors committed by language learners at different levels.
26 L. Xie et al.
Fig. 14 A 3D photo-realistic talking head by combining 2D image samples with a 3D face model
When such a database is available, a more powerful LTS can be trained

discriminatively such that the errors observed in the database can be predicted and
recovered gracefully.
In future work, with regard to the interactivity of the computer assistant, it can
hear (via speech recognition) and speak (TTS synthesis), read and compose, correct
and suggest, or even guess or read the learner’s intention.
Summary
This chapter surveys the basic principles, state-of-the-art technologies, and featured
applications in the visual speech animation area. Data collection, face/mouth model,
feature extraction, and learning a mapping model are the central building blocks of a
VSA system. The technologies used in different blocks depend on the application
needs and affect the desired appearance of the system. During the past decades,
much effort in this area has been devoted to the audio/text-to-visual mapping
problem, and approaches can be roughly categorized into rule based, concatenation,
parametric, and hybrid. We showcase a state-of-the-art approach, based on the deep
bidirectional long short-term memory (DBLSM) recurrent neural networks (RNN)
for audio-to-visual mapping in a video-realistic talking head. We also use the
Engkoo project from Microsoft as a practical application of visual speech animation
in language learning. We believe that with the fast development of computer
graphics, speech technology, machine learning, and human behavior studies, the
future visual speech animation systems will become more flexible, expressive, and
conversational. Subsequently, applications can be found across many domains.
References
Anderson R, Stenger B, Wan V, Cipolla R (2013) Expressive visual text-to-speech using active
appearance models. In: Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, IEEE. p 3382
Badin P, Ben Youssef A, Bailly G et al (2010) Visual articulatory feedback for phonetic correction
in second language learning. In: Proceedings of Second Language learning Studies: Acquisition,
Learning, Education and Technology, 2010
Ben Youssef A, Shimodaira H, Braude DA (2013) Articulatory features for speech-driven head
motion synthesis. In: Proceedings of the International Speech Communication Association,
IEEE, 2013
Breeuwer M, Plomp R (1985) Speechreading supplemented with formant frequency information
from voiced speech. J Acoust Soc Am 77(1):314–317
Bregler C, Covell M, Slaney M (1997) Video rewrite: driving visual speech with audio. In:
Proceedings of the 24th annual conference on Computer graphics and interactive techniques,
ACM Press, p 353
Busso C, Deng Z, Grimm M, Neumann U et al (2007) Rigid head motion in expressive speech
animation: Analysis and synthesis. IEEE Trans Audio, Speech, Language Process 15
(3):1075–1086
Cao Y, Tien WC, Faloutsos P et al(2005) Expressive speech-driven facial animation. In: ACM
Transactions on Graphics, ACM, p 1283
Cohen MM, Massaro DW (1993) Modeling coarticulation in synthetic visual speech. In: Models
and techniques in computer animation. Springer, Japan, p 139
Cootes TF, Edwards GJ, Taylor CJ (2001) Active appearance models. IEEE Trans Pattern Anal
Mach Intell 23(6):681–685
Cosatto E, Graf HP (1998) Sample-based synthesis of photo-realistic talking heads. In: Proceedings
of Computer Animation, IEEE, p 103
Cosatto E, Graf HP (2000) Photo-realistic talking-heads from image samples. IEEE Trans Multimed
2(3):152–163
Cosatto E, Ostermann J, Graf HP et al (2003) Lifelike talking faces for interactive services. Proc
IEEE 91(9):1406–1429
Deng Z, Neumann U (2008) Data-driven 3D facial animation. Springer
Deng L, Yu D (2014) Deep learning methods and applications. Foundations and Trends in Signal
Processing, 2014
Deng Z, Lewis JP, Neumann U (2005) Automated eye motion using texture synthesis. IEEE
Comput Graph Appl 25(2):24–30
Ding C, Xie L, Zhu P (2015) Head motion synthesis from speech using deep neural networks.
Multimed Tools Appl 74(22):9871–9888
Du J, Wang Q, Gao T et al (2014) Robust speech recognition with speech enhanced deep neural
networks. In: Proceedings of the International Speech Communication Association, IEEE, p 616
Dziemianko M, Hofer G, Shimodaira H (2009). HMM-based automatic eye-blink synthesis from
speech. In: Proceedings of the International Speech Communication Association, IEEE, p 1799
Englebienne G, Cootes T, Rattray M (2007) A probabilistic model for generating realistic lip
movements from speech. In: Advances in neural information processing systems, p 401
Eskenazi M (2009) An overview of spoken language technology for education. Speech Commun 51
(10):832–844
Ezzat T, Poggio T (2000) Visual speech synthesis by morphing visemes. Int J Comput Vision 38
(1):45–57
28 L. Xie et al.
Ezzat T, Geiger G, Poggio T (2002) Trainable videorealistic speech animation. In: ACM
SIGGRAPH 2006 Courses, ACM, p 388
Fagel S, Clemens C (2004) An articulation model for audiovisual speech synthesis: determination,
adjustment, evaluation. Speech Commun 44(1):141–154
Fagel S, Bailly G, Theobald BJ (2010) Animating virtual speakers or singers from audio:
lip-synching facial animation. EURASIP J Audio, Speech, Music Process 2009(1):1–2
Fan B, Wang L, Soong FK et al (2015) Photo-real talking head with deep bidirectional LSTM. In:
IEEE International Conference on Acoustics, Speech, and Signal Processing, IEEE, p 4884
Fan B, Xie L, Yang S, Wang L et al (2016) A deep bidirectional LSTM approach for video- realistic
talking head. Multimed Tools Appl 75:5287–5309
Fu S, Gutierrez-Osuna R, Esposito A et al (2005) Audio/visual mapping with cross-modal hidden
Markov models. IEEE Trans Multimed 7(2):243–252
Hinton G, Deng L, Yu D, Dahl GE et al (2012) Deep neural networks for acoustic modeling in
speech recognition: The shared views of four research groups. IEEE Signal Process Mag 29
(6):82–97
Hochreiter S (1998) The vanishing gradient problem during learning recurrent neural nets and
problem solutions. Int J Uncertain Fuzz 6(02):107–116
Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Comput 9(8):1735–1780
Huang D, Wu X, Wei J et al (2013) Visualization of Mandarin articulation by using a physiological
articulatory model. In: Signal and Information Processing Association Annual Summit and
Conference, IEEE, p 1
Hunt AJ, Black AW (1996) Unit selection in a concatenative speech synthesis system using a large
speech database. In: IEEE International Conference on Acoustics, Speech, and Signal Pro-
cessing, IEEE, p 373
Hura S, Leathem C, Shaked N (2010) Avatars meet the challenge. Speech Technol, 303217
Jia J, Zhang S, Meng F et al (2011) Emotional audio-visual speech synthesis based on PAD. IEEE
Trans Audio, Speech, Language Process 19(3):570–582
Jia J, Wu Z, Zhang S et al (2014) Head and facial gestures synthesis using PAD model for an
expressive talking avatar. Multimed Tools Appl 73(1):439–461
Kukich K (1992) Techniques for automatically correcting words in text. ACM Comput Surv 24
(4):377–439
Le BH, Ma X, Deng Z (2012) Live speech driven head-and-eye motion generators. IEEE Trans Vis
Comput Graph 18(11):1902–1914
Liu P, Soong FK (2005) Kullback-Leibler divergence between two hidden Markov models.
Microsoft Research Asia, Technical Report
Massaro DW (1998) Perceiving talking faces: from speech perception to a behavioral principle. Mit
Press, Cambridge
Massaro DW, Simpson JA (2014) Speech perception by ear and eye: a paradigm for psychological
inquiry. Psychology Press
Masuko T, Kobayashi T, Tamura, M et al (1998) Text-to-visual speech synthesis based on parameter
generation from HMM. In: IEEE International Conference on Acoustics, Speech, and Signal
Processing, IEEE, p 3745
McGurk H, MacDonald J (1976) Hearing lips and seeing voices. Nature 264:746–748
Microsoft Research (2015) http://research.microsoft.com/en-us/projects/voice_driven_talking_head/
Mikolov T, Chen K, Corrado G et al (2013) Efficient estimation of word representations in vector
space. arXiv preprint arXiv:1301.3781
Musti U, Zhou Z, Pietikinen M (2014) Facial 3D shape estimation from images for visual speech
animation. In: Proceedings of the Pattern Recognition, IEEE, p 40
Ostermann J, Weissenfeld A (2004) Talking faces-technologies and applications. In: Proceedings of
the 17th International Conference on Pattern Recognition, IEEE, p 826
Pandzic IS, Forchheimer R (2002) MPEG-4 facial animation. The standard, implementation and
applications. John Wiley and Sons, Chichester
Parke FI (1972) Computer generated animation of faces. In: Proceedings of the ACM annual
conference-Volume, ACM, p 451
Peng B, Qian Y, Soong FK et al (2011) A new phonetic candidate generator for improving search
query efficiency. In: Twelfth Annual Conference of the International Speech Communication
Association
Pighin F, Hecker J, Lischinski D et al (2006) Synthesizing realistic facial expressions from
photographs. In: ACM SIGGRAPH 2006 Courses, ACM, p 19
Qian Y, Yan ZJ, Wu YJ et al (2010) An HMM trajectory tiling (HTT) approach to high quality TTS.
In: Proceedings of the International Speech Communication Association, IEEE, p 422
Raidt S, Bailly G, Elisei F (2007) Analyzing gaze during face-to-face interaction. In: International
Workshop on Intelligent Virtual Agents. Springer, Berlin/Heidelberg, p 403
Microsoft Research (2015) http://research.microsoft.com/en-us/projects/blstmtalkinghead/
Richmond K, Hoole P, King S (2011) Announcing the electromagnetic articulography (day 1)
subset of the mngu0 articulatory corpus. In: Proceedings of the International Speech Commu-
nication Association, IEEE, p 1505
Roweis S (1998) EM algorithms for PCA and SPCA. Adv Neural Inf Process Syst:626–632
Sako S, Tokuda K, Masuko T et al(2000) HMM-based text-to-audio-visual speech synthesis. In:
Proceedings of the International Speech Communication Association, IEEE, p 25
Schuster M, Paliwal KK (1997) Bidirectional recurrent neural networks. IEEE Trans Signal Process
45(11):2673–2681
Scott MR, Liu X, Zhou M (2011) Towards a Specialized Search Engine for Language Learners
[Point of View]. Proc IEEE 99(9):1462–1465
Seidlhofer B (2009) Common ground and different realities: World Englishes and English as a
lingua franca. World Englishes 28(2):236–245
Sumby WH, Pollack I (1954) Erratum: visual contribution to speech intelligibility in noise
[J. Acoust. Soc. Am. 26, 212 (1954)]. J Acoust Soc Am 26(4):583–583
Taylor P (2009) Text-to-speech synthesis. Cambridge university press, Cambridge
Taylor SL, Mahler M, Theobald BJ et al (2012) Dynamic units of visual speech. In: Proceedings of
the 11th ACM SIGGRAPH/Eurographics conference on Computer Animation, ACM, p 275
Theobald BJ, Fagel S, Bailly G et al (2008) LIPS2008: Visual speech synthesis challenge. In:
Proceedings of the International Speech Communication Association, IEEE, p 2310
Thies J, Zollhfer M, Stamminger M et al(2016) Face2face: Real-time face capture and reenactment
of rgb videos. In: Proceedings of Computer Vision and Pattern Recognition, IEEE, p 1
Tokuda K, Yoshimura T, Masuko T et al (2000) Speech parameter generation algorithms for
HMM-based speech synthesis. In: IEEE International Conference on Acoustics, Speech, and
Signal Processing, IEEE, p 1615
Tokuda K, Oura K, Hashimoto K et al (2007) The HMM-based speech synthesis system. Online:
http://hts.ics.nitech.ac.jp
Wang D, King S (2011) Letter-to-sound pronunciation prediction using conditional random fields.
IEEE Signal Process Lett 18(2):122–125
Wang L, Soong FK (2012) High quality lips animation with speech and captured facial action unit
as A/V input. In: Signal and Information Processing Association Annual Summit and Confer-
ence, IEEE, p 1
Wang L, Soong FK (2015) HMM trajectory-guided sample selection for photo-realistic talking
head. MultimedTools Appl 74(22):9849–9869
Wang L, Han W, Qian X, Soong FK (2010a) Rendering a personalized photo-real talking head from
short video footage. In: 7th International Symposium on Chinese Spoken Language Processing,
IEEE, p 129
Wang L, Qian X, Han W, Soong FK (2010b) Synthesizing photo-real talking head via trajectory-
guided sample selection. In: Proceedings of the International Speech Communication Associ-
ation, IEEE, p 446
30 L. Xie et al.
Wang L, Wu YJ, Zhuang X et al (2011) Synthesizing visual speech trajectory with minimum
generation error. In: IEEE International Conference on Acoustics, Speech and Signal Pro-
cessing, IEEE, p 4580
Wang L, Chen H, Li S et al (2012a) Phoneme-level articulatory animation in pronunciation training.
Speech Commun 54(7):845–856
Wang L, Han W, Soong FK (2012b) High quality lip-sync animation for 3D photo-realistic talking
head. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, IEEE, p
4529
Wang LJ, Qian Y, Scott M, Chen G, Soong FK (2012c) Computer-assisted Audiovisual Language
Learning, IEEE Computer, p 38
Weise T, Bouaziz S, Li H et al (2011) Realtime performance-based facial animation. In: ACM
Transactions on Graphics, ACM, p 77
Wik P, Hjalmarsson A (2009) Embodied conversational agents in computer assisted language
learning. Speech Commun 51(10):1024–1037
Williams RJ, Zipser D (1989) A learning algorithm for continually running fully recurrent neural
networks. l Comput 1(2):270–280
Xie L, Liu ZQ (2007a) A coupled HMM approach to video-realistic speech animation. Pattern
Recogn 40(8):2325–2340
Xie L, Liu ZQ (2007b) Realistic mouth-synching for speech-driven talking face using articulatory
modelling. IEEE Trans Multimed 9(3):500–510
Xie L, Jia J, Meng H et al (2015) Expressive talking avatar synthesis and animation. Multimed
Tools Appl 74(22):9845–9848
Yan ZJ, Qian Y, Soong FK (2010) Rich-context unit selection (RUS) approach to high quality TTS.
In: IEEE International Conference on Acoustics, Speech and Signal Processing, IEEE, p 4798
Zen H, Senior A, Schuster M (2013) Statistical parametric speech synthesis using deep neural
networks. In: IEEE International Conference on Acoustics, Speech and Signal Processing,
IEEE, p 7962
Zhang LJ, Rubdy R, Alsagoff L (2009) Englishes and literatures-in-English in a globalised world.
In: Proceedings of the 13th International Conference on English in Southeast Asia, p 42
Zhu P, Xie, L, Chen Y (2015) Articulatory movement prediction using deep bidirectional long
short-term memory based recurrent neural networks and word/phone embeddings. In: Sixteenth
Annual Conference of the International Speech Communication Association
Blendshape Facial Animation
Ken Anjyo
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Blendshape Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Blendshape Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Examples and Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Techniques for Efficient Animation Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Direct Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Use of PCA Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Blendshape Creation, Retargeting, and Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Abstract
Blendshapes are a standard approach for making expressive facial animations in
the digital production industry. The blendshape model is represented as a linear
weighted sum of the target faces, which exemplify user-defined facial expressions
or approximate facial muscle actions. Blendshapes are therefore quite popular
because of their simplicity, expressiveness, and interpretability. For example,
unlike generic mesh editing tools, blendshapes approximate a space of valid
facial expressions.
This article provides the basic concepts and technical development of the
blendshape model. First, we briefly describe a general face rig framework and
thereafter introduce the concept of blendshapes as an established face rigging
approach. Next, we illustrate how to use this model in animation practice, while
K. Anjyo (*)
OLM Digital, Setagaya, Tokyo, Japan
e-mail: anjyo@acm.org

DOI 10.1007/978-3-319-30808-1_2-1
2 K. Anjyo
clarifying the mathematical framework for blendshapes. We also demonstrate a

few technical applications developed in the blendshape framework.
Keywords
Computer facial animation • Face rig • Blendshapes • Retarget • Deformer • Facial
motion capture • Performance capture
State of the Art
Digital characters now appear not only in films and video games but also in various
digital contents. In particular, facial animation of a digital character should then
convey emotions to it, which plays a crucial role for visual storytelling. This requires
a digital character animation process as well as its face rigging process (i.e., the setup
process) to be very intensive and laborious.
In this article, we define face rig as the pair of a deformer and its user interface
(manipulator). The deformer means a mathematical model of deforming a face
model’s geometry for making animation. The user interface provides animators a
toolset of manipulating the face model, based on the deformer. In a production
workplace, however, they usually use several deformers at a time, so that the user
interface in practice should be more complicated, yet sophisticated, rather than the
user interface that we will mention in later sections for blendshapes.
A variety of the face rig approaches have been developed. Physics-based models
provide the rigorous and natural approaches, having several applications not only in
the digital production industry but also in medical sciences, including surgery
simulations. The physics-based approaches for computer graphic applications
approximate the mechanical properties of the face, such as skin layers, muscles,
fatty tissues, bones, etc. Although the physics-based methods may be powerful in
making realistic facial animations, artists are then required to have a certain amount
of knowledge and experiences regarding background physics. This is not an
easy task.
On one hand, several commercial 3D CG packages provide proprietary face rig
approaches, such as “cluster deformers” (see Tickoo (2009)), which allow the artist
to specify the motion space using a painting operation for making 3D faces at key
frames.
The blendshapes offer a completely different face rig approach. A blendshape
model generates face geometry as a linear combination of a number of face poses,
each of which is called a blendshape target. These targets typically mean individual
facial expressions, shapes that approximate facial muscle actions or FACS (Facial
Action Coding System (Ekman and Friesen 1978)) motions. These targets are
predefined (designed) by the artist. The blendshapes are therefore parameterized
with the weights of the targets, which gives an intuitive and simple way for the artist
to make animation. The interface is called sliders and used to control the weights.
Blendshape Facial Animation 3
Fig. 1 Blendshapes user interface example. Left: The slider box and a 3D face model under editing,
where the slider box gives a partial view of the blendshape sliders. This is because, in general, the
number of sliders is too large to see all sliders at a time. Instead a desired slider can be reached by
scrolling the slider box. The 3D face model shows an edited result with the slider operation for right
eye blink, right: the face model before the slider operation
Figure 1 presents such a slider interface example and a simple editing result for a
blendshape model.
The use of motion capture data has become a common approach to make
animation of a digital character. As is well known, the original development of
motion capture techniques was driven by the needs of life science community, where
the techniques are mainly used for the analysis on a subject’s movement. In the
digital production industry, facial motion capture data may be used as an input for the
synthesis of realistic animations. The original data will then be converted to a digital
face model and edited to obtain desired facial animations. Some of the face rig
techniques are therefore indispensable in the converting (retargeting) and editing
processes.
Blendshape Applications
As mentioned earlier, several face rig techniques are used together for practical
situations. Even when more sophisticated approaches to facial modeling are used,
blendshapes are often employed as a base layer over which physically based or
functional (parameterized) deformations are layered.
4 K. Anjyo
In digital production studios and video game companies, they need to develop a
sophisticated system that should fully support the artists for efficient and high-
quality production of visual effects and character animation. The role of blendshape
techniques may therefore be a small portion of the system, but is still crucial. Here
we briefly describe a few state-of-the-art applications that use blendshape
techniques:
• Voodoo. This system has been developed in Rhythm & Hues Studios over years,
which deals mainly with animation, rigging, matchmove, crowds, fur grooming,
and computer vision (see Fxguide (2014)). The system provides several prodi-
gious face rigging tools using blendshapes. For example, many great shots in the
film Life of Pi in 2012 were created with this system.
• Fez. This is the facial animation system developed in ILM (Bhat et al. 2013;
Cantwell et al. 2016; CGW 2014), which involves an FACS implementation
using blendshape techniques. It has contributed to recent films, such as Warcraft
and Teenage Mutant Ninja Turtles, in 2016.
• Face Plus. This is a plug-in for Unity, which is a cross-platform game engine.
This plug-in enables us to construct a facial capture and animation system using a
web camera (see Mixamo (2013) for details). Based on the blendshape character
model created by an artist, the system provides real-time facial animation of the
character.
In the following sections, we describe the basic practice and mathematical

background of the blendshape model.
Blendshape Practice
The term “blendshapes” was introduced in the computer graphics industry, and we
follow the definition: blendshapes are linear facial models in which the individual
basis vectors are not orthogonal but instead represent individual facial expressions.
The individual basis vectors have been referred to as blendshape targets and morph
targets or (more roughly) as shapes or blendshapes. The corresponding weights are
often called sliders, since this is how they appear in the user interface (as shown in
Fig. 1). Creating a blendshape facial animation thus requires specifying weights for
each frame of the animation, which has traditionally been achieved with key frame
animation or by motion capture.
In the above discussion, we use a basic mathematical term “vectors.” This section
starts with explaining what the vectors mean in making 3D facial models and
animations. We then illustrate how to use the blendshapes in practice.
Formulation
We represent the face model as a column vector f containing all the model vertex
coordinates in some order that is arbitrary (such as xyzxyzxyz, or alternately
xxxyyyzzz) but consistent across the individual blendshapes. For example, let us
consider the face model composed of n = 100 blendshapes, each having p = 1000
vertices, with each vertex having three components x, y, z. Similarly, we denote the
blendshape targets as vectors bk, so the blendshape model is represented as
X
n
f¼ wk bk , (1)
k¼0
where f is the resulting face, in the form of a m = 30,000 1 vector (m = 3p); the
individual blendshapes b0, b1, , bn are 30,000 1 vectors; and wk. denotes the
weight for bk (1 k n). We then put b0 as the neutral face. Blendshapes can
therefore be considered simply adding vectors.
Equation (1) may be referred to as the global or “whole-face” blendshape
approach. The carefully sculpted blendshape targets appeared in Eq. (1) then serve
as interpretable controls; the span of these targets strictly defines the valid range of
expressions for the modeled face. These characteristics differentiate the blendshape
approach from those that involve linear combinations of uninterpretable shapes (see
a later section) or algorithmically recombine the target shapes using a method other
than that in Eq. (1). In particular, from an artist’s point of view, the interpretability of
the blendshape basis is a definitive feature of the approach.
In the whole-face approach, scaling all the weights by a multiplier causes the
whole head to scale, while scaling of the head is more conveniently handled with a
separate transformation. To eliminate undesired scaling, the weights in Eq. 1 may be
constrained to sum to one. Additionally the weights can be constrained to the interval
[0,1] in practice.
In the local or “delta” blendshape formulation, one face model b0 (typically the
resting face expression) is designated as the neutral face shape, while the remaining
targets bk(1 k n) in Eq. (1) are replaced with the difference bk – b0 between the
k-th face target and the neutral face:
X
n
f ¼ b0 þ wk ðbk b0 Þ: (2)
k¼1
Or, if we use matrix notation, Eq. (2) can be expressed as:
f5Bwþb0 , (3)
where B is an m n matrix having bk – b0 as the k-th column vector, and w = (w1,

w2,...,wn)T is the weight vector.
6 K. Anjyo
Fig. 2 Target face examples. From left: neutral, smile, disaffected, and sad
In this formulation, the weights are conventionally limited to the range [0,1],
while there are exceptions to this convention. For example, the Maya blendshape
interface allows the [0,1] limits to be overridden by the artist if needed. If the
difference between a particular blendshape bk and the neutral shape is confined to
a small region, such as the left eyebrow, then the resulting parameterization offers
intuitive localized control.
The delta blendshape formulation is used in popular packages such as Maya (see
Tickoo (2009)), and our discussion will assume this variant if not otherwise speci-
fied. Many comments apply equally (or with straightforward conversion) to the
whole-face variant.
Examples and Practice
Next, we show a simple example of the blendshape model, which has 50 target faces.
The facial expressions in Fig. 1 were also made with this simple model. A few target
shapes of the model are demonstrated in Fig. 2, where the leftmost image shows its
neutral face. Using the 50 target shapes, the blendshape model provides a mixture of
such targets.
As mentioned above, the blendshape model is conceptually simple and intuitive.
Nevertheless, professional use of this model further requires a large and labor-
intensive effort of the artists, some of which are listed as follows:
• Target shape construction

– To express a complete range of realistic expressions, digital modelers often
have to create large libraries of blendshape targets. For example, the character
of Gollum in The Lord of the Rings had 946 targets (Raitt 2004). Generating a
reasonably detailed model can be as much as a year of work for a skilled
modeler, involving many iterations of refinement.
– A skilled digital artist can deform a base mesh into the different shapes needed
to cover the desired range of expressions. Alternatively, the blendshapes can
be directly scanned from a real actor or a sculpted model. A common template
model can be registered to each scan in order to obtain vertex-wise correspon-

dences across the blendshape targets.
• Slider control (see Fig. 1)
– To skillfully and efficiently use the targets, animators need to memorize the
function of 50 to 100 commonly used sliders. Then locating a desired slider
isn’t immediate.
– A substantial number of sliders are needed for high-quality facial animation.
Therefore the complete set of sliders does not fit on the computer display.
• Animation editing
– As a traditional way, blendshapes have been animated by key frame animation
of the weights. Commercial packages provide spline curve interpolation of the
weights and allow the tangents to be specified at key frames.
– Performance-driven facial animation is an alternative way to make animation.
Since blendshapes are the common approach for realistic facial models,
blendshapes and performance-driven animation are frequently used together
(see section “Use of PCA Models,” for instance). We then may need an
additional process where the motion captured from a real face is “retargeted”
to a 3D face model.
Techniques for Efficient Animation Production
In previous sections, we have shown that the blendshapes are a conceptually simple,
common, yet laborious facial animation approach. Therefore a number of develop-
ments have been made to greatly improve efficiency in making blendshape facial
animation. However, in this section, let us restrict ourselves to describe only a few of
our work, while we also mention some techniques related to blendshapes and facial
animation. To know more about the mathematical aspect of blendshape algorithms,
we would recommend referring to the survey (Lewis et al. 2014).
Direct Manipulation
In general, interfaces should provide both direct manipulation and editing of under-
lying parameters. While direct manipulation usually provides more natural and
efficient results, parameter editing can be more exact and reproducible. Artists
might therefore prefer it in some cases.
While inverse kinematic approaches to posing human figures have been used for
many years, analogous inverse or direct manipulation approaches for posing faces
and setting key frames have emerged quite recently. In these approaches, the artist
directly moves points on the face surface model, and the software must solve for the
underlying weights or parameters that best reproduce that motion, rather than tuning
the underlying parameters.
Here we consider the cases where the number of sliders is considerably large (i.e.,
well over 100) for a professional use of the blendshape model. Introducing a direct
8 K. Anjyo
Fig. 3 Example of direct manipulation interface for blendshapes
manipulation approach would then be a legitimate requirement. To achieve this, we

solve the inverse problem of finding the weights for given point movements and
constraints.
In Lewis and Anjyo (2010), this problem is regularized by considering the fact
that facial pose changes are proportional to slider position changes. The resulting
approach is easy to implement and can cope with existing blendshape models.
Figure 3 shows such a direct manipulation interface example, where selecting a
point on the face model surface creates a manipulator object termed a pin, and the
pins can be dragged into desired positions. According to the pin and drag operations,
the system solves for the slider values (the right panel in Fig. 3) for the face to best
match the pinned positions. It should then be noted that the direct manipulation
developed in Lewis and Anjyo (2010) can interoperate with the traditional
parameter-based key frame editing. As demonstrated in Lewis and Anjyo (2010),
both direct manipulation and parameter editing are indispensable for blendshape
animation practice. There are several extensions of the direct manipulation approach.
For instance, a direct manipulation system suitable for use in animation production
has been demonstrated in Seo et al. (2011), including treatment of combination
blendshapes and non-blendshape deformers. Another extension in Anjyo et al.
(2012) describes a direct manipulation system that allows more efficient edits
using a simple prior learned from facial motion capture.
Use of PCA Models
In performance-driven facial animation, the motion of human actor is used to derive

the face model. Whereas face tracking is a key technology for the performance-
driven approaches, this article focuses on performance capture methods that drive a
face rig. The performance capture methods mostly use PCA basis or blendshape
basis.
We use principal component analysis (PCA) to obtain a PCA model for the given
database of facial expression examples. As usual each element of the database is
represented as an m 1 vector x. Let U be an m r matrix consisting of the
r eigenvectors corresponding to the largest eigenvalues of the data covariance
matrix. The PCA model is then given as:
f5Ucþe0 , (4)
where the vector c means the coefficients of those eigenvectors and e0 denotes the
mean vector of all elements x in the database. Since we usually have r m, the PCA
model gives a good low-dimensional representation of the facial models x. This also
leads us to solutions to statistical estimation problems in a maximum a posteriori
(MAP) framework. For example, in Lau et al. (2009), direct dragging and stroke-
based expression editing are developed in this framework to find an appropriate c in
Eq. (4).
The PCA approaches are useful if the face model is manipulated only with direct
manipulation. Professional animation may also require slider operations, so that the
underlying basis should be of blendshapes, rather than PCA representation. This is
due to the lack of interpretability of the PCA basis (Lewis and Anjyo 2010).
A blendshape representation (3) can be equated to a PCA model (4) that spans the
same space:
Bw þ bo5Ucþe0 : (5)
We know from Eq. (5) that the weight vector w and the coefficient vector c can be
interconverted:
1
w5 BT B BT ðUcþe0 b0 Þ (6)
c5UT ðBwþb0 e0 Þ, (7)
where we use the fact that UTU is an r r unit matrix in deriving the second
Eq. (7). We note that the matrices and vectors in Eqs. (6) and (7), such as
(BTB)1BTU and (BTB)1BT(e0 – b0), can be precomputed. Converting from
weights to coefficients or vice versa is thus a simple affine transform that can easily
be done at interactive rates. This will provide us a useful direct manipulation method
for a PCA model, if the model can also be represented with a blendshape model.
Blendshape Creation, Retargeting, and Transfer
Creating a blendshape model for professional animation requires sculpting on the

order of 100 blendshape targets and adding hundred more shapes in several ways
10 K. Anjyo
(see Lewis et al. (2014), for instance). Ideally, the use of dense motion capture of a
sufficiently varied performance should contribute to efficiently create such a large
number of blendshape targets. To achieve this, several approaches have been
proposed, including a PCA-based approach (Lewis et al. 2004) and a sparse matrix
decomposition method (Neumann et al. 2013).
Expression cloning approaches (Noh and Neumann (2001); Sumner and Popović
(2004), for instance) are developed for retargeting the motion from one facial model
(the “source”) to drive a face (the “target”) with significantly different proportions.
The expression cloning problem was posed in Noh and Neumann (2001), where the
solution was given as a mapping by finding corresponding pairs of points on the
source and target faces using face-specific heuristics. The early expression cloning
algorithms do not consider adapting the temporal dynamics of the motion to the
target, which means that they work well if the source and target are of similar
proportions. The movement matching principle in Seol et al. (2012) provides an
expression cloning algorithm that can cope with the temporal dynamics of face
movement by solving a space-time Poisson equation for the target blendshape
motion.
Relating to expression cloning, we also mention model transfer briefly. This is the
case where the source is a fully constructed blendshape model and the target consists
of only a neutral face (or a few targets). Deformation transfer (Sumner and Popović
2004) then provides a method of constructing the target blendshape model, which is
mathematically equivalent to solving a certain Poisson equation (Botsch et al. 2006).
We also have more recent progresses for the blendshape model transfer, including
the one treating with a self-collision issue (Saito 2013) and the technique allowing
the user to iteratively add more training poses for blendshape expression refinement
(Li et al. 2010).
Conclusion
While the origin of blendshapes may lie outside academic forums, blendshape
models have evolved over the years along with a variety of advanced techniques
including those described in this article. We expect more scientific insights from
visual perception, psychology, and biology will strengthen the theory and practice of
the blendshape facial models.
In a digital production workplace, we should also promote seamless integration of
the blendshape models with other software tools to establish a more creative and
efficient production environment.
Acknowledgments I would like to thank J. P. Lewis for mentoring me over the years in the field of
computer facial animation research and practice. Many thanks go to Ayumi Kimura for her fruitful
discussions and warm encouragements in preparing and writing this article. I also thank Gengdai
Liu and Hideki Todo for their helpful comments and creation of the images in Figs. 1, 2, and 3.
References
Anjyo K, Todo H, Lewis JP (2012) A practical approach to direct manipulation blendshapes. J
Graph Tools 16(3):160–176
Bhat K, Goldenthal R, Ye Y, Mallet R, Koperwas M (2013) High fidelity facial animation capture
and retargeting with contours. In: Proceedings of the 12th ACM SIG-GRAPH/Eurographics
Symposium on Computer Animation, 7–14
Botsch M, Sumner R, Pauly M, Gross M (2006) Deformation transfer for detail-preserving surface
editing. In: Proceedings of Vision, Modeling, and Visualization (VMV), 357–364
Cantwell B, Warner P, Koperwas M, Bhat K (2016) ILM facial performance capture, In ACM
SIGGRAPH2016 Talks, 26:1–26:2
CGW web page (2014) http://www.cgw.com/Publications/CGW/2014/Volume-37-Issue-4-Jul-
Aug-2014-/Turtle-Talk.aspx
Ekman P, Friesen W (1978) Facial action coding system: manual. Consulting Psychologists Press,
Palo Alto
Fxguide web page (2014) https://www.fxguide.com/featured/voodoo-magic/
Lau M, Chai J, Xu Y-Q, Shum H-Y (2009) Face poser: interactive modeling of 3D facial
expressions using facial priors. ACM Trans Graph 29(1), 3:1–3:17
Lewis JP, Anjyo K (2010) Direct manipulation blendshapes. IEEE Comput Graph Appl 30
(4):42–50
Lewis JP, Mo Z, Neumann U (2004) Ripple-free local bases by design. In: Proceedings of
International Conference on Acoustics, Speech and Signal Processing (ICASSP), 684–688
Lewis JP, Anjyo K, Rhee T, Zhang M, Poghin F, Deng Z (2014) Practice and theory of blend-shape
facial models. Eurographics 2014 (State of the Art Reports), 199–218
Li H, Weise T, Pauly M (2010) Example-based facial rigging. ACM Trans Graph 29(3), 32:1–32:6
Mixamo web page (2013) https://www.mixamo.com/faceplus
Neumann T, Varanasi K, Wenger S, Wacker M, Magnor M, Theobalt C (2013) Sparse localized
deformation components. ACM Trans Graph 32(6), 179:1–179:10
Noh J, Neumann U (2001) Expression Cloning. In: SIGGRAPH2001, Computer Graphics Pro-
ceedings, ACM Press/ACM SIGGRAPH, 277–288
Raitt B (2004) The making of Gollum. Presentation at U. Southern California Institute for Creative
Technologies’s Frontiers of Facial Animation Workshop, August 2004
Saito J (2013) Smooth contact-aware facial blendshape transfer. In: Proceedings of Digital Produc-
tion Symposium 2013 (DigiPro2013), ACM. 7–12
Seo J, Irving J, Lewis JP, Noh J (2011) Compression and direct manipulation of complex
blendshape models. ACM Trans Graph 30(6), 164:1–164:10
Seol Y, Lewis JP, Seo J, Choi B, Anjyo K, Noh J (2012) Spacetime expression cloning for
blendshapes. ACM Trans Graph 31(2), 14:1–14:12
Sumner RW, Popović J (2004) Deformation transfer for triangle meshes. ACM Trans Graph 23
(3):399–405
Tickoo S (2009) Autodesk maya 2010: a comprehensive guide. CADCIM Technologies,
Schererville
Eye Animation
Andrew T. Duchowski and Sophie Jörg
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Eye Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Listing’s and Bonders’ Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Modeling Physiologically Plausible Eye Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Modeling Induced Torsion During Vergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Fixations and Saccades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Parametric Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Modeling Microsaccadic Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Modeling Saccadic Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Pupil Dilation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Procedural Eye Movement Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Periocular Motions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Running the Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Summary: Listing the Sources of Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Abstract
The synthesis of eye movements involves modeling saccades (the rapid shifts of
gaze), smooth pursuits (object tracking motions), binocular rotations implicated
in vergence, and the coupling of eye and head rotations. More detailed move-
ments include dilation and constriction of the pupil (pupil unrest) as well as small
fluctuations (microsaccades, tremor, and drift, which we collectively call
A.T. Duchowski (*)

Clemenson University, Clemson, SC, USA
e-mail: duchowski@clemson.edu
S. Jörg
School of Computing, Clemson University, Clemson, SC, USA
e-mail: sjoerg@clemson.edu

DOI 10.1007/978-3-319-30808-1_3-1
2 A.T. Duchowski and S. Jörg
microsaccadic jitter) made during fixations, when gaze is held nearly steady. In
this chapter, we focus on synthesizing physiologically plausible eye rotations,
microsaccadic jitter, and pupil unrest. We review concepts relevant to the anima-
tion of eye motions and provide a procedural model of gaze that incorporates
rotations adhering to Donders’ and Listing’s laws, the saccadic main sequence,
along with gaze jitter and pupil unrest. We model microsaccadic jitter and pupil
unrest by 1/f α or pink noise.
Keywords
Eye movements • Saccades • Fixations • Microsaccadic jitter
Introduction
Engaging virtual characters are highly relevant in many applications, from enter-
tainment to virtual reality (e.g., training). Realistic eye motions are important for
increasing the perceived believability of virtual actors (avatars) and physical human-
oid robots. Looser and Wheatley [36] show that people are influenced by the eyes
more than by other facial features when rating the animacy of virtual characters.
In film, eye movements are important for conveying the character’s emotions and
thoughts, e.g., as in Neil Burger’s feature film Limitless. In virtual environments, eye
gaze is of vital importance for the correct identification of deictic reference – what is
being looked at [42]. As characters become more realistic and humanlike, correct
gaze animation becomes even more important. Garau et al. [20] found a strong
subjective interaction effect between the realism of a character and their gaze; for a
more realistic character, more elaborate gaze behavior is preferred, whereas for a less
realistic character, random gaze patterns received better ratings.
The dynamics of eye motions, however, have received little attention since Lee
et al.’s [35] seminal Eyes Alive model which focused largely on saccadic eye
movements, implementing what is known as the saccadic main sequence [5]. The
rapid advancement of eye tracking technology has revitalized interest in recording
eye movements for inclusion in computer graphics and interactive systems [17, 57,
65].
Ruhland et al. [53] survey the state of the art in eye and gaze animation, where
efforts aimed at modeling the appearance and movement of the human eye are
reviewed. Beyond the synthesis of saccades (the rapid shifts of gaze), their report
also considers tracking motions known as smooth pursuits, binocular rotations
implicated in vergence (used for depth perception), and the coupling of eye and
head rotations (e.g., the vestibulo-ocular reflex (VOR)). Ruhland et al. [53] further-
more review high-level aspects of gaze behavior including past efforts to model
visual attention, expression of emotion, nonverbal interaction, conversation and
listening behavior, verbal interaction, and speech-driven gaze.
In this chapter, we extend their review by focusing on several important aspects
missing from their survey, namely, oculomotor rotations of the eyeball adhering to
Eye Animation 3
Donders’ and Listing’s laws [59], the detailed motions of the eye during fixations
[38] that can be modeled with microsaccadic jitter, and rapid fluctuations of the pupil
(pupil unrest) [54]. We present previous findings on these topics and derive a
physiologically plausible procedural eye movement model where microsaccadic
jitter and pupil unrest are modeled by 1/fα pink noise. Our chapter contribution is
based on prior publications presented at Computer Graphics International (CGI)
[14], Motion in Games (MIG) [15], and the Symposium on Eye Tracking Research
& Applications (ETRA) [18].
Eye Rotation
Almost all normal primate eye movements used to reposition the fovea result as
combinations of five basic types: saccadic, smooth pursuit, vergence, vestibular, and
small movements associated with fixations [51]. These smaller motions consist of
drift, tremor, and microsaccades [52]. Other movements such as adaptation and
accommodation refer to nonpositional aspects of eye movements (i.e., pupil dilation,
lens focusing).
In general, the eyes move within six degrees of freedom: three translations within
the socket and three rotations, although physical displacement is required for trans-
lations to occur (e.g., a push of a finger). There are six muscles responsible for
movement of the eyeball: the medial and lateral recti (sideway movements), the
superior and inferior recti (up/down movements), and the superior and inferior
obliques (twist) [12]. The neural system involved in generating eye movements is
known as the oculomotor plant [51]. Eye movement control signals emanate from
several functionally distinct brain regions. Areas in the occipital cortex are thought
to be responsible for high-level visual functions such as recognition. The superior
colliculus bears afferents emanating directly from the retina, particularly from
peripheral regions conveyed through the magnocellular pathway. The semicircular
canals react to head movements in three-dimensional space. All three areas (i.e., the
occipital cortex, the superior colliculus, and the semicircular canals) convey effer-
ents to the eye muscles through the mesencephalic and pontine reticular formations.
Classification of observed eye movement signals relies in part on the known
functional characteristics of these cortical regions [16].
Eye movement models typically do not consider the oculomotor plant for the
purposes of animation; rather, signal characteristics are of greater importance. For
example, Komogortsev et al. have developed a sophisticated model of the oculomo-
tor plant but for biometric identification purposes rather than for animation [30, 31].
Prior models of eye rotation have been developed from the perspective of
capturing and modeling observed gaze behavior but do not necessarily take into
account their synthetic reproduction, i.e., animation [24, 50]. What are also often
overlooked are constraints of orbital rotations following Listing’s and Donders’
laws. In this chapter we discuss previous models based on quaternion rotation and
show how they can be implemented in a straightforward manner to ensure physio-
logically plausible eye rotation.
Listing’s and Bonders’ Laws
Listing’s and Donders’ laws state that eyeball rotations can effectively be modeled as
compositions of rotations exclusively about the vertical and horizontal axes, with
negligible torsion when head direction is fixed. Further implications arise during
ocular vergence movement, as discussed below.
Using recorded gaze movements from monkeys and humans, Tweed et al. [59]
define the eyeball’s primary position as one in which the gaze vector is orthogonal to
Listing’s plane, the plane of displacement, which essentially models the tilt of the
head. Listing’s law states that the eye assumes only those orientations that can be
reached from the primary position by a single rotation about an axis lying in Listing’s
plane. For modeling purposes, Listing’s law, a refinement of Donders’ law, states
that in the absence of head tilt and with static visual surrounding, we can effectively
ignore the eyeball’s torsional component when modeling saccades [48]. In practice,
torsion fluctuations of up to about 5 have been observed [19].
Interestingly, Tweed and Vilis [60] show that the primary gaze direction varies
between primates (e.g., preferred head tilt varies within humans as well as within
monkeys, and between the two groups, with monkeys generally carrying their heads
tilted slightly more back than humans, on average). We believe that this variability in
preferred primary gaze direction is a factor in believability of virtual actors which
may not have been previously exploited. Ma and Deng [37], for example, describe a
model of gaze driven by head direction/rotation, but their gaze-head coupling model
is designed in a fashion that seems contrary to physiology: head motion triggers eye
motion. Instead, because the eyes are mechanically “cheaper” and faster to rotate, the
brain usually triggers head motion when the eyes exceed a certain saccadic ampli-
tude threshold (about 30 ; see Murphy and Duchowski [41] for an introductory note
on this topic). Nevertheless, Ma and Deng and then later Peters and Qureshi [47]
both provide useful models of gaze/head coupling with a good “linkage” between
gaze and head vectors.
Our model currently focuses only on gaze direction and assumes a stationary head
but eventually could offer extended functionality, expressed in terms of quaternions,
which are generally better suited for expressing rotations than vectors and Euler
angles. In our model, the eyes are the primary indicators of attention, with head
rotation following gaze rotation when a rotational threshold (e.g., 30 ) is exceeded.
Modeling Physiologically Plausible Eye Rotation
A coordinated model of the movement of the head and eyes relies on a plausible
description of the eyeball’s rotation within its orbit. Eyeball rotation, at any given
instant in time t, is described by the deviation of the eyeball from its primary
position. This rotation can be described by the familiar Euler angles used to denote
roll, pitch, and yaw. Mathematically, a concise and convenient representation of all
Eye Animation 5
three angles in one construct is afforded by a quaternion that describes the eyeball’s
orientation, which describes the direction of the vector ge emanating from the eyeball
center and terminating at the 3D position of gaze in space, pt ¼ ðxt , yt , zt Þ.
Tweed et al. [59] specify the quaternion in question in relation to the bisector Ve
of the current reference gaze direction ge and primary gaze vector gp. To precisely
model Listing’s law, assuming Ve is a normalized forward-pointing vector orthogo-
nal to the displacement plane (which may be tilted back), the quaternion
q expressing the rotation between gp and ge is q ¼ ½Ve ge , ðVe ge Þ in a right-
handed coordinate system. The quaternion q is a vector with four components, q
¼ ðq0 , qτ , qV , qH Þ , with q0 the scalar part of q and qτ, qV, and qH the torsional,
vertical, and horizontal rotational components, respectively.
Listing’s and Donders’ laws are important for setting up traditional computer
graphics rotations of the eyeball because together they not only specify a convenient
simplification but they also specify a physiologically observable constraint, namely,
qτ ¼ 0 , negligible visible torsion. Moreover, head/eye movement coordination is
made implicit since Listing’s plane can be used to model the orientation of the head
with quaternion q fixed to lie in Listing’s plane. This is accomplished by first setting
the gaze direction vector gr to point at reference point pt, then specifying the rotation
quaternion’s plane by parameters f, fV, and fH, which are used to express qτ as a
function of qH and qV : qT ¼ f þ f V qV þ f H qH . If f is not 0, then the reference
position pt does not satisfy Listing’s law (see Fig. 1). Quaternion
pffiffiffiffiffiffiffiffiffiffiffiffi
ffi
e¼ 1 f 2 , f , 0, 0 , however, does and has the same direction as gr. To force
pffiffiffiffiffiffiffiffiffiffiffiffiffi
q to adhere to Listing’s law, we set up quaternion e1 ¼ 1 f 2 , f , 0, 0 and
right-multiply q. This fixes the reference position adjusting the quaternion’s tor-
sional component. To find Listing’s plane, the normal vector is computed by
specifying the quaternion of the primary position relative to e as
p ¼ ðV 1 , V 0 , V 3 , V 2 Þ where V ¼ ð1, f V , f H Þ=jð1, f V , f H Þj. Quaternion
q is then left-multiplied by p1 giving p1 qe1 as the corrected rotation quaternion
satisfying Listing’s law such that qτ ¼ 0. It is important to note that gr describes the
orientation of the head. That is, gr can be used to model an avatar’s preferred head
tilt, and thus primary gaze direction, or rotation undergone during vergence eye
movements (see below).
Modeling Induced Torsion During Vergence
Torsional eye movements are associated with vergence or can occur in response to
specific visual stimuli [62]. Tweed et al.’s [59] quaternion framework models
Listing’s law by negating cyclotorsion (qτ ¼ 0). The shape of the surface to which
the eye position quaternions are constrained resembles a plane when only the eyes
move and the head is stationary. When gaze shifts involve both the eye and head
a b c d
looking ahead no torsion slight torsion 180° torsion
Fig. 1 Eye rotation from primary position (a): (b) no torsion modeling Listing’s and Donders’
laws; (c) slight torsion due to Listing’s plane tilt; (d) implausible torsion though mathematically
possible if quaternions are not constrained
(e.g., during VOR), the rotation surface twists and becomes non-planar [21]. This
twist is similar to that produced by a Fick gimbal model of rotations in which the
horizontal axis is nested within a fixed vertical axis.1
Tweed et al.’s [59] quaternion framework does not explicitly consider vergence
movements. Mok et al. [39] suggest that eye positions during vergence remain
restricted to a planar surface (Listing’s plane), but that surface is rotated relative to
that observed for far targets. The rotation is such that during convergence both eyes
undergo extorsion during downward gaze shifts and intorsion during upward gaze shifts,
i.e., Listing’s plane of the left eye is rotated to the left about the vertical axis (qV > 0)
while that of the right is rotated to the right (qV < 0).
For example, vergence of degree θ can be modeled by constructing quaternion
b
q ¼ ð cos ðθ=2Þ, sin ðθ=2ÞvÞ where v denotes the vertical axis (0, 1, 0) and then
1
rotating gr via bqgr b
q , as illustrated by Fig. 2. During convergence, primary position
is rotated temporally by approximately two-thirds the angle of vergence.
Our model currently does not take into account refinements concerning VOR
within the basic quaternion framework but produces correct vergence eye move-
ments in the context of head-free (instead of head-fixed) rotations and allows
targeting of the eye at a look point pt. This point may in turn be animated to rotate
the eye, e.g., via a procedural model (see below).
Fixations and Saccades
Fixations, the stationary eye movements required to maintain steady gaze when
looking at a visual target, are never perfectly still but instead are composed of small
involuntary eye movements, composed of microsaccades, drift, and tremor
1
The non-commutativity of rotations leads to false torsions from equivalent rotations around
eye-fixed and head-fixed axes; under normal circumstances, the eye assumes orientations given
by Euler rotations satisfying Listing’s law [50].
Eye Animation 7
a b c
looking in, level looking in and up looking in and down
Fig. 2 Eye rotation during 30 vergence (qV ¼ 0:07): (a) no torsion (f ¼ 0:00) when gaze level;
(b) slight intorsion (f ¼ 0:06) looking up; (c) slight extorsion (f ¼ 0:06) looking down. The plane
drawn behind the (left) eye is a visualization of the rotation of gaze direction gr during convergence
[52]. Microsaccades play a vital role in maintaining visibility during fixation but are
perhaps the least understood of all eye movement types, despite their critical
importance to normal vision [38]. If microsaccadic eye movements were perfectly
counteracted, visual perception would rapidly fade due to adaptation [22,
27]. Microsaccades contribute to maintaining visibility during fixation by shifting
the retinal image in a fashion that overcomes adaptation, generating neural responses
to stationary stimuli in visual neurons.
Martinez-Conde et al. [38] note that microsaccades are unnoticed, but this
generally refers to oneself – it is not possible to detect one’s own eye movements
when looking in a mirror. Because the perceptual system is sensitive to, and
amplifies, small fluctuations [61], noticing others’ eye movements, even subtle
ones, may be important, especially during conversation, turn-taking, etc. (see
Vertegaal [63]).
Even though microsaccades are the largest and fastest fixational eye movement,
they are relatively small in amplitude, carrying the retinal image across a range
of several dozen to several hundred photoreceptor widths [38]. Microsaccades
and saccades share many physical and functional characteristics suggesting that
both eye movements have a common oculomotor origin, i.e., a common
neural generator for both (current evidence points to a key role of the superior
colliculus).
While microsaccades play a crucial role in maintaining visibility during fixation,
they may also betray our emotional state, as they reflect human brain activities
during attention and cognitive tasks [28]. Laretzaki et al. [34] show that fixational
distribution is more widespread during periods of psychological threat versus
periods of psychological safety. That is, the dispersion of microsaccades is larger
under perception of threat than under perception of safety. DiStasi et al. [13] also
note that saccadic and microsaccadic velocity decrease with time-on-task, whereas
drift velocity increases, suggesting that ocular instability increases with mental
fatigue. Thus, dispersion of the microsaccadic position distribution can be made to
increase with (simulated) increased fatigue.
Microsaccades and saccades follow the main sequence that describes the rela-
tionship between their amplitude (θ) and duration (Δt) and can be modeled by the
linear equation
Δt ¼ 2:2θ þ 21 ðmillisecondsÞ (1)
for saccadic amplitudes up to about 20 [5, 29].2 The main sequence gives us a
plausible range of durations and corresponding eyeball rotations that are intuitively
understood: the larger the eye rotation (θ), the more time required to rotate that eye.
All these insights can be used to develop parametric models for the synthesis of
fixations and saccades.
Parametric Models
Animating gaze shifts of virtual humans often involve the use of parametric models
of human gaze behavior [4, 46]. While these types of models enable virtual humans
to perform natural and accurate gaze shifts, signal characteristics, and in particular
noise, are rarely addressed, if at all. Noise, however, although a nuisance from a
signal processing perspective, is a key component of natural eye movements.
To generate a stream of synthetic gaze points resembling captured data pt ¼
ðxt , yt Þ (the z-coordinate can be dropped if the points are projected on a 2D viewing
plane), a reasonable strategy is to guide synthetic gaze to a sequence of known
points, i.e., a grid of points that is used to calibrate the eye tracker to human viewers,
or, e.g., a model of reading behavior where gaze is directed toward as yet unread or
previously read words (regressions or revisits) and to lines above and below the
current line of text [10], or a set of points selected by an artist. Given such a sequence
(e.g., see Fig. 3), several characteristics need to be added, namely:
1. A model of the spatiotemporal fixation perturbation (microsaccadic jitter) [15]

2. A model of saccadic velocity (i.e., position and duration)
3. Control of the simulation time step and sampling rates (see section “Running the
Simulation”)
We suggest to model the spatiotemporal perturbation of gaze points at a fixation,

which arises from microsaccades, drift, and tremor, with 1/f α or pink noise, which
we call microsaccadic jitter.
2
In their development of Eyes Alive, Lee et al. [35] (see also Gu et al. [23]) expressed the main
sequence as Δt ¼ dθ þ D0 (milliseconds) with d [2,2.7] ms/deg and D0 [20,30] ms.
Eye Animation 9
Modeling Microsaccadic Jitter
A key aspect for the realistic synthesis of eye motion is the inclusion of
microsaccadic gaze jitter. While the recorded eye movement signal is well under-
stood from the point of view of analysis, surprisingly little work exists on its detailed
synthesis. Most analytical approaches are concerned with gaze data filtering, e.g.,
signal smoothing and/or processing for the detection of specific events such as
saccades, fixations, or more recently further distinction between ambient and focal
fixations [32]. During analysis of recorded eye movements, gaze data is commonly
filtered, especially when detecting fixations (e.g., see Duchowski et al. [18], who
advocate the use of the Savitzky-Golay filter for signal analysis and detection of
fixations). The signal processing approach (filtering) still dominates even in very
recent approaches to synthesis, e.g., Yeo et al.’s [65] Eyecatch simulation, where the
simulation used the Kalman filter to produce gaze, but focused primarily on saccades
and smooth pursuits (see below). Microsaccades, tremor, or drift were not modeled.
As noted by Yeo et al., simulated gaze behavior looked qualitatively similar to gaze
data captured by an eye tracker, but comparison of synthesized trajectory plots
showed absence of gaze jitter that was evident in the raw data.
The distribution of microsaccade amplitudes tends to a 1 asymptote, making it a
convenient upper amplitude threshold, although microsaccade amplitude distribu-
tion tends to peak at about 12 arcmin [38]. Amplitude distribution can be modeled
by the Poisson probability mass function P ðx, λÞ ¼ λx eλ =x! with x ¼ 6, shifted by
x 8:5 and scaled by 5.5 and approximated by a normal distribution 5.5
pffiffiffi
N x 8:0, μ ¼ λ, σ ¼ λ . The resultant normal distribution resembles the
microsaccade distribution reported by Martinez-Conde et al. [38] and provides a
starting point for modeling of microsaccadic jitter, suggesting that perturbation
about the point of fixation can be modeled by the normal distribution
N ðμ ¼ 0, σ ¼ 12=60Þ (arcmin) for each of the x- and y-coordinate offsets to the
fixation modeled during simulation (setting σ ¼ 0 yields no jitter during fixation and
can be used to simulate keyframed saccades). Modeling microsaccadic jitter by the
normal distribution yields white noise perturbation.
White noise perturbation is a logical starting point for modeling microsaccadic
jitter, but it is uncorrelated and therefore not necessarily a good choice. Recorded
neural spikes are superimposed with noise that exhibits non-Gaussian characteristics
and can be approximated as 1/f α noise [64].
Pulse trains of nerve cells belonging to various brain structures have been
observed and characterized as 1/f noise [61]. The 1/f regime accomplishes a trade-
off: the perceptual system is sensitive to and amplifies small fluctuations; simulta-
neously, the system preserves a memory of past stimuli in the long-time correlation
tails. The memory inherent in the 1/f system can be used to achieve a priming effect:
the succession of two stimuli separated by 50–100 ms at the same location results in
a stronger response to the second stimulus.
actual gaze data Gaze point data

a 900
800
700
y-coordinate (pixels)
600
500
400
300
200
100
0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
x-coordinate (pixels)
synthetic for rendering Gaze point data

b 900
800
700
600
500
400
300
200
100
0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
synthetic data with noise Gaze point data

c 900
800
700
600
500
400
300
200
100
0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Eye Animation 11
To model microsaccades by 1/f α (pink) noise, the white noise perturbations

modeled by normal distribution N ð0, σ ¼ 12=60Þ (arcmin visual angle) are digitally
filtered by a digital pink noise filter with system function:
1 n zq n a þ ak
H n ðzÞ ¼ ∏ k
, Gn ðαÞ ¼ ∏ (2)
Gn ðαÞ k¼1 z þ pk k¼1 ak α þ 1
where Gn(α) is the nth order approximation to an ideal analog pink noise filter with
pffiffi a
system function GðsÞ ¼ 1=ð sÞ , ak ¼ tan2 ðkθÞ, and
1 αak α αk
qk ¼ , pk ¼
1 þ αak α þ αk
with f0 the unity gain frequency, α ¼ tan ðπ f 0 T Þ, and T the sampling period for
filter order n ℤ. With α ¼ 1:0, the filter produces pink noise given white noise as
input [25]. For other values of a, α very good approximation for θ is
θ ¼ π=ð2n þ 2 2αÞ.
We chose a 4th order filter for reshaping of the microsaccadic jitter modeled by
Gaussian noise, N ð0, σ ¼ 12=60Þ arcmin visual angle. More formally, we define the
pink noise filter as a function of two parameters, P ðα, f 0 Þ, where 1/f α describes the
pink noise power spectral distribution and f0 the filter’s unity gain frequency
(or more simply its gain). Setting α ¼ 1 produces 1/f noise. Setting α ¼ 0 produces
white, uncorrelated noise, with a flat power spectral distribution, likely a poor
choice for modeling biological motion such as microsaccades. We found α ¼ 0:6
and f 0 ¼ 0:85 gave fairly natural microsaccadic jitter [15].
In practice, a look point drives the rotation of the eyeball. We can therefore model
microsaccades as separate x- and y-directional offsets to the main view vector. This
then requires two pink noise filters, one for each of the two dimensions. Setting the
simulation up this way allows independent control of horizontal and vertical
microsaccades so that, for example, by controlling α, horizontal microsaccades can
be made to be more noisy (more noise devoted to the high-frequency portion of the
spectrum) than vertical microsaccades.
The above model of microsaccadic jitter does not consider where fixations are
made, i.e., the above model can be used to add perturbations to randomly distributed
fixations. We can guide gaze to a sequence of fixation points, specified as a series of
2D look point coordinates. Microsaccadic jitter is then used to perturb the look point
about each fixation point. More formally, we simulate a sequence of look points via
the following fixation simulation, developed by Duchowski et al. [18]:
ptþh ¼ pt þ P ðα, f 0 Þ (3)

ä
Fig. 3 Generation of a sequence of synthetic gaze points based on raw gaze data captured by an
eye tracker in (a): (b) microsaccadic jitter at fixation points identified from raw gaze data; (c)
addition of simulated eye tracker noise which obscures microsaccadic jitter but produces gaze
distributed about calibration points resembling raw gaze data
Saccade acceleration, velocity, and position profiles

1
Parameterized curve (arbitrary units)
0.5
-0.5
position
velocity
acceleration
-1
0 0.2 0.4 0.6 0.8 1
Time (normalized, arbitrary units)
Fig. 4 Parametric saccade position model derived from an idealized model of saccadic force-time
function assumed by Abrams et al.’s [1] symmetric-impulse variability model: scaled position 60H
€ ðtÞ
(t), velocity 31:H ðtÞ, and acceleration 10H
where pt is the look point at simulation time t and h is the simulation time step.
Modeling Saccadic Velocity
To effect movement of the look point pt between fixation points, a model of saccades
is required, specifying both movement and duration of the gaze point. We start with
an approximation to a force-time function assumed by a symmetric-impulse vari-
ability model [1]. This function, qualitatively similar to symmetric limb movement
trajectories, describes an acceleration profile that rises to a maximum, returns to zero
about halfway through the movement, and then is followed by an almost mirror-
image deceleration phase.
To model a symmetric acceleration function, we can choose a combination of
Hermite blending functions h11(t) and h10(t), so that H € ðtÞ ¼ h10 ðtÞ þ h11 ðtÞ where
h10 ðtÞ ¼ t 2t þ t, h11 ðtÞ ¼ t t , t ½0, 1, and H
3 2 3 2 € ðtÞ is acceleration of the gaze
point over normalized time interval t ½0, 1 . Integrating acceleration produces
velocity, H_ ðtÞ ¼ 12 t4 13 þ 12 t2 which when integrated once more produces posi-
tion HðtÞ ¼ 10 t 14 t4 þ 16 t3 on the normalized interval t ½0, 1 (see Fig. 4).
1 5
Given an equation for position over a normalized time window (t ½0, 1), we can
now stretch this time window at will to any given length t ½0, Δt. Because the
distance between gaze target points is known a priori, we can use these distances
(pixel distances converted to amplitudes in degrees visual angle) as input to the main
sequence (1) to obtain saccade length.
Eye Animation 13
Assuming data collected from the eye tracker does not deviate greatly from the
main sequence found in the literature [5, 29], we set the expected saccade duration to
that given by (1) but augmented with a 10 targeting error. We also add in a slight
temporal perturbation to the predicted saccade duration, based on empirical obser-
vations. Saccade duration is thus modeled as
Δt ¼ 2:2N ðθ, σ ¼ 10 Þ þ 21 þ N ð0, 0:01Þ ðmillisecondsÞ (4)
Pupil Dilation
Pupil dilation is controlled by top-down and bottom-up processes. There is evidence

that it responds to cognitive load [2, 6], ambient light changes [7], and visual
detection [49]. However, for the purposes of animation, pupil unrest, the slight
oscillation between pupil dilation and constriction, is perhaps more interesting to
model. Stark et al. [54] describe pupil diameter fluctuations as noise in a biological
system with the major component of the pupil unrest as random noise in the
0.05–0.3 Hz range, with transfer function GðsÞ ¼ :16expð: 18sÞ=ð1 þ 0:1sÞ3 and
gain equal to 0.16. This transfer function can be modeled by a third-order
Butterworth filter with system function G3 ðsÞ ¼ 1=ðs3 þ 2s2 þ 2s þ 1Þ with cutoff
frequency set to 1.5915 (see Hollos and Hollos [26]). Such a filter can thus be used to
smooth Gaussian noise (e.g., N ð0, 0:5Þ) but will result in uncorrelated noise.
In recent work on eye capture, Bérard et al. [9] model pupil constriction/dilation
but only via linear interpolation of keyframes in response to light conditions. They
did not, however, procedurally animate the pupil as a function of pupil unrest.
Pamplona et al. [45] modeled pupil unrest (hippus) but via small random variations
to light intensity, likely to be white noise although they did not specify this directly.
We model pupil unrest directly, via pink noise perturbation.
We can model pupil diameter oscillation with pink noise by once again filtering
white noise with the same digital pink noise filter as for microsaccadic perturba-
tions. For pupil oscillation, we chose a 4th order filter for reshaping pupil oscilla-
tion modeled as Gaussian noise, N ð0, σ ¼ 0:5Þ. We found pink noise parameters
α ¼ 1:6 and f 0 ¼ 0:35 produced fairly natural simulation of pupil unrest [15].
Procedural Eye Movement Model
In the previous sections, we reviewed important concepts relevant to the animation

of eye movements and presented components of a procedural model for eye move-
ment synthesis, consisting of rotations that adhere to Donders’ and Listing’s laws,
and a model of gaze that incorporates the saccadic main sequence, along with gaze
jitter and pupil unrest.
We model both microsaccades and pupil unrest by 1/f α or pink noise, where
0 < α < 2, with exponent α usually close to 1. The use of 1/f α pink noise to model
microsaccadic jitter and pupil oscillation is a key aspect of our simulation. Various
colors of noise have appeared in the computer graphics literature, e.g., blue noise
for anti-aliasing, green noise for halftoning, and white noise for common random
number generation [66]. Pink noise is regarded as suitable for describing physical
and biological distributions, e.g., plants [11, 43] and galaxies [33], as well as the
behavior of biosystems in general [56]. Aks et al. [3] suggest that memory across
eye movements may serve to facilitate selection of information from the visual
environment, leading to a complex and self-organizing (saccadic) search pattern
produced by the oculomotor system reflecting 1/f pink noise.
To complete our model, we add periocular motions to our model and run our
simulation.
Periocular Motions
The motions of the upper and lower eyelids comprise saccadic lid and smooth
pursuit movements, where the eyelid motion is closely related to the motion of the
corresponding eyeball and blinks. We create the saccadic lid and smooth pursuit
movements by rigging the eyelids to the eyeball. To model blinks, we approximate
the eyelid closure function proposed by Trutoiu et al. [58]. We use a piecewise
function to model the eyelid blink in two temporal components, the faster closure
followed by a slower opening:

a ðt μ Þ2 , tμ
CðtÞ ¼ (5)
b eclogðtμþ1Þ , otherwise
with C ¼ 1 indicating lid fully open and C ¼ 0 lid fully closed, where t ½0, 100
represents normalized percent blink duration (scalable to an arbitrary duration), μ
¼ 37 the point when the lid should reach full (or nearly-full) closure, a ¼ 0:98
indicating percent lid closure at the start of the blink, with b ¼ 1:18 and c ¼ μ=100
parameters used to shape the asymptotic lid opening function.
Trutoiu et al. [58] recorded blink frequencies from their actors of 6.6, 8.2, and
27.0 blinks per minute, or 14 blinks per minute on average. These rates appear to be
within normal limits reported by Bentivogolio et al. [8], namely, 17 blinks per
minute, ranging from 4.5 while reading to 26 during conversation. Simulating
conversation, our procedural model uses 25 blinks per minute as the average with
a mean duration of 120 ms. Unlike Steptoe et al. [55], we do not use kinematics to
model blinks; rather we use a simplified stochastic model of blink duration.
Running the Simulation
When running the simulation, it is important to keep the simulation time step (h)
small, e.g., h ¼ 0:0001. When about to execute a saccade, set the saccade clock t ¼ 0,
and then while t < Δt, perform the following simulation steps:
Eye Animation 15
1: t ¼ t=Δt ðscale interpolant to time windowÞ

2: pt ¼ Ci1 þ H ðtÞCi ðadvance positionÞ
3: t ¼ t þ h ðadvance time by the time step hÞ
where Ci denotes the ith 2D look point sequence coordinates and pt is the saccade
position, both in vector form.
Setting time step h to an arbitrarily small value allows dissociation of the
simulation clock from the sampling rate. We can thus sample the synthetic eye
tracking data at arbitrary sampling periods, e.g., d ¼ 1, d ¼ 16, or d ¼ 33 ms for
sampling rates of 1000 Hz, 60 Hz, or 30 Hz, respectively.
Unfortunately, eye trackers’ sampling rates are not precise, or rather, eye trackers’
sampling periods are generally non-uniform, most likely due to competing processes
on the computer used to run the eye tracking software and/or due to network
latencies. The simulation sampling period can be modeled by adding in a slight
random temporal perturbation, e.g., N ð0, σ ¼ 0:5Þ milliseconds.
Summary: Listing the Sources of Variation
To recount, the stochastic model of eye movements is based on infusion of proba-

bilistic noise at various points in the simulation:
• Fixation durations, modeled in this instance by N ð1:081, σ ¼ 2:9016Þ (seconds),

the average and standard deviation from Duchowski et al. [18],
• Microsaccadic fixation jitter, modeled by pink noise P ðα ¼ :6, f 0 ¼ :85Þ (degrees
visual angle),
• Saccade durations, modeled by (4), and
• Sampling period N ð1, 000=F , σ ¼ 0:5Þ (milliseconds), with F the sampling
frequency (Hz).
For rendering purposes, the eye movement data stream is appended with:
• Blink duration, modeled as N ð120, σ ¼ 70Þ (ms), and

• Pupil unrest, modeled by pink noise P ðα ¼ 1:6, f 0 ¼ :35Þ (relative diameter).
Collectively, the above sources of error can be considered as stochastic perturba-

tion of the gaze point about its current location, i.e.,
ptþh ¼ pt ¼ P ðα, f 0 Þ þ η (6)
where the primary source of microsaccadic jitter is represented by pink noise P

ðα, f 0 Þ and η represents the various sources of variation, above. See Duchowski et al.
[18] for further details.
For affective eye movement synthesis, modulating α will result in modulation of
jitter. Modulation of f0 controls the amount of dispersion of fixational points. Both
model parameters can thus be used to control the expected appearance of emotional
state. What remains is tuning of these parameters to effect emotional expression.
Results from our perceptual experiments thus far have showed that animations based
on the procedural model with pink noise jitter were consistently evaluated and
perceived as highly natural in the presence of other animations as alternatives
[15]. We have also found that some jitter, but not too much, captures visual attention
better than when jitter is excessive or not present.
Conclusion
In this chapter, we have summarized some of the physiological characteristics of eye

motions and presented a physiologically plausible procedural model of eye move-
ments, complete with blinks, saccades, and fixations augmented by microsaccadic
jitter and pupil unrest, both modeled by 1/f α or pink noise.
Our procedural model of gaze motion constitutes the basis of a “bottom-up” model
of gaze rotations, modeled by quaternions which are used to effect eyeball rotation in
response to a “look point” in space projected onto a 2D plane in front of the eye. The
location of this look point is determined by the procedural model simulated over time,
which is tasked with producing a characteristic fixation/saccade signal that models
recorded gaze data. The procedural model differs from others, e.g., those driven by
saliency such as Oyekoya et al.’s [44], or those driven by head movement propensity
[47]. These latter models can be considered “top-down” models as they are more
concerned with prediction of locations of gaze that, say, an autonomous avatar is
likely to make. Our model is concerned with low-level signal characteristics of the
fixation point regardless of how it was determined.
Subjective evaluations have shown that the absence of noise is clearly unnatural.
Microsaccadic jitter therefore appears to be a crucial ingredient in the quest toward
natural eye movement rendering. Gaze jitter is naturally always present since the
eyes are never perfectly still. We believe that correctly modeling the jitter that
characterizes gaze fixation is a key factor in promoting the believability and accep-
tance of synthetic actors and avatars, thereby bridging the Uncanny Valley [40].
Acknowledgments This material is based in part upon work supported by the US National
Science Foundation under Grant No. IIS-1423189. Any opinions, findings, and conclusions
expressed in this material are those of the authors and do not necessarily reflect the views of the
NSF.
References
1. Abrams RA, Meyer DE, Kornblum S (1989) Speed and accuracy of saccadic eye movements:
characteristics of impulse variability in the oculomotor system. J Exp Psychol Hum Percept
Perform 15(3):529–543
2. Ahern S, Beatty J (1979) Pupillary responses during information processing vary with scholas-
tic aptitude test scores. Science 205(4412):1289–1292
Eye Animation 17
3. Aks DJ, Zelinsky GJ, Sprott JC (2002) Memory across eye-movements: 1 / f dynamic in visual
search. Nonlinear Dynamics Psychol Life Sci 6(1):1–25
4. Andrist S, Pejsa T, Mutlu B, Gleicher M (2012) Designing effective gaze mechanisms for
virtual agents. In: Proceedings of the 2012 ACM Annual Conference on Human Factors in
Computing Systems, CHI’12. ACM, New York, pp 705–714. doi:10.1145/2207676.2207777,
http://doi.acm.org/10.1145/2207676.2207777
5. Bahill AT, Clark M, Stark L (1975) The main sequence. A tool for studying human eye
movements. Math Biosci 24(3/4):191–204
6. Beatty J (1982) Task-evoked pupillary responses, processing load, and the structure of
processing resources. Psychol Bull 91(2):276–292
7. Beatty J, Lucero-Wagoner B (2000) The pupillary system. In: Cacioppo JT, Tassinary LG,
Bernston GG (eds) Handbook of psychophysiology, 2nd edn. Cambridge, Cambridge Univer-
sity Press, pp 142–162
8. Bentivoglio AR, Bressman SB, Cassetta E, Carretta D, Tonali P, Albanese A (1997) Analysis of
blink rate patterns in normal subjects. Mov Disord 12(6):1028–1034
9. Bérard P, Bradley D, Nitti M, Beeler T, Gross M (2014) High-quality capture of eyes. ACM
Trans Graph 33(6):2231–22312. doi:10.1145/2661229.2661285
10. Campbell CS, Maglio PP (2001) A robust algorithm for reading detection. In: ACM workshop
on perceptive user interfaces. ACM Press, New York, pp 1–7
11. Condit R, Ashton PS, Baker P, Bunyavejchewin S, Gunatilleke S, Gunatilleke N, Hubbell SP,
Foster RB, Itoh A, LaFrankie JV, Lee HS, Losos E, Manokaran N, Sukumar R, Yamakura T
(2000) Spatial patterns in the distribution of tropical tree species. Science 288(5470):1414–1418
12. Davson H (1980) Physiology of the eye, 4th edn. Academic, New York, NY
13. Di Stasi LL, McCamy MB, Catena A, Macknik SL, Cañas JJ, Martinez-Conde S (2013)
Microsaccade and drift dynamics reflect mental fatigue. Eur J Neurosci 38(3):2389–2398
14. Duchowski A, Jörg S (2015) Modeling physiologically plausible eye rotations: adhering to
Donders’ and listing’s laws. In: Proceedings of computer graphics international (short papers)
(2015)
15. Duchowski A, Jörg S, Lawson A, Bolte T, Świrski L, Krejtz K (2015) Eye movement synthesis
with 1/f pink noise. In: Motion in Games (MIG) 2015, Paris, France
16. Duchowski AT (2007) Eye tracking methodology: theory & practice, 2nd edn. Springer,
London, UK
17. Duchowski AT, House DH, Gestring J, Wang RI, Krejtz K, Krejtz I, Mantiuk R, Bazyluk B
(2014) Reducing visual discomfort of 3D stereoscopic displays with gaze-contingent depth-of-
field. In: Proceedings of the ACM symposium on applied perception, SAP’14. ACM, New York,
NY, pp 39–46. doi:10.1145/2628257.2628259, http://doi.acm.org/10.1145/2628257.2628259
18. Duchowski AT, Jörg S, Allen TN, Giannopoulos I, Krejtz K (2016) Eye movement synthesis.
In: Proceedings of the ninth biennial acm symposium on eye tracking research & applications,
ETRA’16. ACM, New York, NY, pp 147–154. doi:10.1145/2857491.2857528, http://doi.acm.
org/10.1145/2857491.2857528
19. Ferman L, Collewijn H, Van den Berg AV (1987) A direct test of listing’s law – I. human ocular
torsion measured in static tertiary positions. Vision Res 27(6):929–938
20. Garau M, Slater M, Vinayagamoorthy V, Brogni A, Steed A, Sasse MA (2003) The impact of
avatar realism and eye gaze control on perceived quality of communication in a shared
immersive virtual environment. In: Human factors in computing systems: CHI 03 conference
proceedings. ACM Press, New York, pp 529–536
21. Glenn B, Vilis T (1992) Violations of listing’s law after large eye and head gaze shifts. J
Neurophysiol 68(1):309–318
22. Grzywacz NM, Norcia AM (1995) Directional selectivity in the cortex. In: Arbib MA (ed) The
handbook of brain theory and neural networks. Cambridge, MA, The MIT Press, pp 309–311
23. Gu E, Lee SP, Badler JB, Badler NI (2008) Eye movements, saccades, and multi-party
conversations. In: Deng Z, Neumann U (eds) Data-driven 3D facial animation. Springer,
London, UK, pp 79–97. doi:10.1007/978-1-84628-907-1_4
24. Haslwanter T (1995) Mathematics of three-dimensional eye rotations. Vision Res 35

(12):1727–1739
25. Hollos S, Hollos JR (2015) Creating noise. Exstrom Laboratories, LLC, Longmont, CO,
http://www.abrazol.com/books/noise/ (last accessed Jan. 2015). ISBN
9781887187268 (ebook)
26. Hollos S, Hollos JR (2015) Recursive Digital Filters: A Concise Guide. Exstrom Laboratories,
LLC, Longmont, CO, http://www.abrazol.com/books/filter1/ (last accessed Jan.
2015). ISBN 9781887187244 (ebook)
27. Hubel DH (1988) Eye, brain, and vision. Scientific American Library, New York, NY
28. Kashihara K, Okanoya K, Kawai N (2014) Emotional attention modulates microsaccadic rate
and direction. Psychol Res 78:166–179
29. Knox PC (2012) The parameters of eye movement (2001). Lecture Notes, URL: http://www.liv.
ac.uk/~pcknox/teaching/Eymovs/params.htm (last accessed November 2012)
30. Komogortsev OV, Karpov A (2013) Liveness detection via oculomotor plant characteristics:
attack of mechanical replicas. In: Proceedings of the IEEE/IARP international conference on
biometrics (ICB), pp 1–8
31. Komogortsev OV, Karpov A, Holland CD (2015) Attack of mechanical replicas: liveness
detection with eye movements. IEEE Trans Inform Forensics Secur 10(4):716–725
32. Krejtz K, Duchowski AT, Çöltekin A (2014) High-level gaze metrics from map viewing:
charting ambient/focal visual attention. In: Kiefer P, Giannopoulos I, Raubal M, Krüger A
(eds) 2nd international workshop in eye tracking for spatial research (ET4S)
33. Landy SD (1999) Mapping the universe. Sci Am 224:38–45
34. Laretzaki G, Plainis S, Vrettos I, Chrisoulakis A, Pallikaris I, Bitsios P (2011) Threat and trait
anxiety affect stability of gaze fixation. Biol Psychol 86(3):330–336
35. Lee SP, Badler JB, Badler NI (2002) Eyes alive. ACM Trans Graph 21(3):637–644.
doi:10.1145/566654.566629, http://doi.acm.org/10.1145/566654.566629
36. Looser CE, Wheatley T (2010) The tipping point of animacy. How, when, and where we
perceive life in a face. Psychol Sci 21(12):1854–62
37. Ma X, Deng Z (2009) Natural eye motion synthesis by modeling gaze-head coupling. In: IEEE
virtual reality, pp 143–150. Lafayette, LA
38. Martinez-Conde S, Macknik SL, Troncoso Xoana G, Hubel DH (2009) Microsaccades: a
neurophysiological analysis. Trends Neurosci 32(9):463–475
39. Mok D, Ro A, Cadera W, Crawford JD, Vilis T (1992) Rotation of listing’s plane during
vergence. Vision Res 32(11):2055–2064
40. Mori M (1970) The uncanny valley. Energy 7(4):33–35
41. Murphy H, Duchowski AT (2002) Perceptual gaze extent & level of detail in VR: looking
outside the box. In: Conference abstracts and applications (sketches & applications), Computer
graphics (SIGGRAPH) annual conference series. ACM, San Antonio, TX
42. Murray N, Roberts D, Steed A, Sharkey P, Dickerson P, Rae J, Wolff R (2009) Eye gaze in
virtual environments: evaluating the need and initial work on implementation. Concurr Comput
21:1437–1449
43. Ostling A, Harte J, Green J (2000) Self-similarity and clustering in the spatial distribution of
species. Science 27(5492):671
44. Oyekoya O, Steptoe W, Steed A (2009) A saliency-based method of simulating visual attention
in virtual scenes. In: Reality V (ed) Software and technology. New York, ACM, pp 199–206
45. Pamplona VF, Oliveira MM, Baranoski GVG (2009) Photorealistic models for pupil light reflex
and iridal pattern deformation. ACM Trans Graph 28(4):106:1–106:12. doi:10.1145/
1559755.1559763, http://doi.acm.org/10.1145/1559755.1559763
46. Pejsa T, Mutlu B, Gleicher M (2013) Stylized and performative gaze for character animation. In:
Navazo I, Poulin P (eds) Proceedings of EuroGrpahics. EuroGraphics
47. Peters C, Qureshi A (2010) A head movement propensity model for animating gaze shifts and
blinks of virtual characters. Comput Graph 34:677–687
Eye Animation 19
48. Porrill J, Ivins JP, Frisby JP (1999) The variation of torsion with vergence and elevation. Vision
Res 39:3934–3950
49. Privitera CM, Renninger LW, Carney T, Klein S, Aguilar M (2008) The pupil dilation response
to visual detection. In: Rogowitz BE, Pappas T (eds) Human vision and electronic imaging, vol
6806. SPIE, Bellingham, WA
50. Quaia C, Optican LM (2003) Three-dimensional Rotations of the Eye. In: Kaufman PL, Alm A
(eds) Adler’s phsyiology of the eye: clinical application, 10th edn. C. V. Mosby Co., St. Louis,
pp 818–829
51. Robinson DA (1968) The oculomotor control system: a review. Proc IEEE 56(6):1032–1049
52. Rolfs M (2009) Microsaccades: Small steps on a long way. Vision Res 49(20):2415–2441.
doi:10.1016/j.visres.2009.08.010, http://www.sciencedirect.com/science/article/pii/
S0042698909003691
53. Ruhland K, Andrist S, Badler JB, Peters CE, Badler NI, Gleicher M, Mutlu B, McDonnell R
(2014) Look me in the eyes: a survey of eye and gaze animation for virtual agents and artificial
systems. In: Lefebvre S, Spagnuolo M (ed) Computer graphics forum. EuroGraphics STAR –
State of the Art Report. EuroGraphics.
54. Stark L, Campbell FW, Atwood J (1958) Pupil unrest: an example of noise in a biological
servomechanism. Nature 182(4639):857–858
55. Steptoe W, Oyekoya O, Steed A (2010) Eyelid kinematics for virtual characters. Comput
Animat Virtual World 21(3–4):161–171
56. Szendro P, Vincze G, Szasz A (2001) Pink-noise behaviour of biosystems. Eur Biophys J 30
(3):227–231
57. Templin K, Didyk P, Myszkowski K, Hefeeda MM, Seidel HP, Matusik W (2014) Modeling
and optimizing eye vergence response to stereoscopic cuts. ACM Trans. Graph 33(4):8 Article
145 (July 2014), DOI = http://dx.doi.org/10.1145/2601097.2601148
58. Trutoiu LC, Carter EJ, Matthews I, Hodgins JK (2011) Modeling and animating eye blinks.
ACM Trans Appl Percept (TAP) 2(3):17:1–17:17
59. Tweed D, Cadera W, Vilis T (1990) Computing three-dimensional eye position quaternions and
eye velocity from search coil signals. Vision Res 30(1):97–110
60. Tweed D, Vilis T (1990) Geometric relations of eye position and velocity vectors during
saccades. Vision Res 30(1):111–127
61. Usher M, Stemmler M, Olami Z (1995) Dynamic pattern formation leads to 1/f noise in neural
populations. Phys Rev Lett 74(2):326–330
62. van Rijn LJ (1994) Torsional eye movements in humans. Ph.D. thesis, Erasmus Universiteit
Rotterdam, Rotterdam, The Netherlands
63. Vertegaal R (1999) The GAZE groupware system: mediating joint attention in mutiparty
communication and collaboration. In: Human factors in computing systems: CHI’99 conference
proceedings. ACM Press, New York, pp 294–301
64. Yang Z, Zhao Q, Keefer E, Liu W (2009) Noise characterization, modeling, and reduction for
in vivo neural recording. In: Bengio Y, Schuurmans D, Lafferty J, Williams CKI, Culotta A
(eds) Advances in neural information processing systems, vol 22., pp 2160–2168
65. Yeo SH, Lesmana M, Neog DR, Pai DK (2012) Eyecatch: simulating visuomotor coordination
for object interception. ACM Trans Graph 31(4):42:1–42:10
66. Zhou Y, Huang H, Wei LY, Wang R (2012) Point sampling with general noise spectrum. ACM
Trans Graph 31(4):76:1–76:11. doi:10.1145/2185520.2185572, URL: http://doi.acm.org/10.
1145/2185520.2185572
Head Motion Generation
Najmeh Sadoughi and Carlos Busso
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Role of Head Motion in Human Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Relation Between Head Motion and Speech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Head Movement Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Rule-Based Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Data-Driven Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Hybrid Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Open Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Speech-Driven Models Using Synthetic Speech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Exploring Entrainment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Modeling Personality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Joint Models to Integrate Head Motion with Other Gestures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Abstract
Head movement is an important part of body language. Head motion plays a role
in communicating lexical and syntactic information. It conveys emotional and
personality traits. It plays an important role in acknowledging active listening.
Given these communicative functions, it is important to synthesize Conversation
Agents (CAs) with meaningful human-like head motion sequences, which are
timely synchronized with speech. There are several studies that have focused on
synthesizing head movements. Most studies can be categorized as rule-based or
data-driven frameworks. On the one hand, rule-based methods define rules that
map semantic labels or communicative goals to specific head motion sequences,
N. Sadoughi (*) • C. Busso (*)

Multimodal signal Processing Lab, University of Texas at Dallas, Dallas, TX, USA
e-mail: nxs137130@utdallas.edu; busso@utdallas.edu
# Springer International Publishing Switzerland 2016 1

DOI 10.1007/978-3-319-30808-1_4-1
2 N. Sadoughi and C. Busso
which are appropriate for the underlying message (e.g., nodding for affirmation).
However, the range of head motion sequences that are generated by these systems
are usually limited, resulting in repetitive behaviors. On the other hand, data-
driven methods rely on recorded head motion sequences which are used either to
concatenate existing sequences creating new realizations of head movements or
to build statistical frameworks that are able to synthesize novel realizations of
head motion behaviors. Due to the strong correlation between head movements
and speech prosody, these approaches usually rely on speech to drive the head
movements. These methods can capture a broader range of movements displayed
during human interaction. However, even when the generated head movements
may be tightly synchronized with speech, they may not convey the underlying
discourse function or intention in the message. The advantages of rule-based and
data-driven methods have inspired several studies to create hybrid methods that
overcome the aforementioned limitations. These studies have been proposed to
generate the movements using parametric or nonparametric approaches,
constraining the models not only on speech, but also on the semantic content.
This chapter reviews most influential frameworks to generate head motion. It also
discusses open challenges that can move this research area forward.
Keywords
Conversational agent • Rule-based animation • Data-driven animation • Speech-
driven animation • Head movement generation • Semantic content • Back-
channel • Nonverbal behaviors • Expressive head motion • Rapport • Embodied
conversational agents • Visual prosody
Introduction
Head movement is an integral part of our body language used during human
interactions. Head motion can play a communicative role displaying emblems (i.e.,
gestures conveying specific meaning) and regulators (i.e., gestures to control turn-
taking sequence) (Heylen 2005). It is also important to establish rapport by provid-
ing suitable backchannels while listening to conversational partners (Gratch et al.
2006; Huang et al. 2011). Having rhythmic head movements coupled with speech
prosodic features increases speech intelligibility by signaling syntactic boundaries
(Munhall et al. 2004). Head movements are also used to convey the mood of the
speaker (Busso et al. 2007a, b). It can also be used to express uncertainty (Marsi and
van Rooden 2007). Given the key role of head motion during human interaction, it is
not surprising that there is a need to model and capture all these aspects to generate
believable conversational agents (CAs). CAs without head movements are per-
ceived less warm (Welbergen et al. 2015) and less natural (Busso et al. 2007a;
Mariooryad and Busso 2012). This chapter describes influential frameworks pro-
posed to synthesize head motion sequences for CAs, describing the main challenges.
For head movement generation, previous studies have proposed frameworks that
can be categorized into two main approaches: rule-based methods (Cassell et al.
Head Motion Generation 3
1994; Kopp et al. 2006; Liu et al. 2012; Pelachaud et al. 1996) and data-driven
frameworks (Busso et al. 2005; Chiu and Marsella 2011; Chuang and Bregler 2005;
Deng et al. 2004; Mariooryad and Busso 2012; Sadoughi et al. 2014; Taylor et al.
2006). The predominant approach to generate head motion is rule-based systems,
where hand crafted head movements such as shaking and nodding are carefully
selected and stored. The semantic content is then analyzed generating head motion
sequences following the selected rules. These methods usually define several heu-
ristic rules derived from previous psychological and observational studies, mapping
the syntactic and semantic structure of the utterance to prototypical head motion
sequences. These methods also define rules to synchronize head movements with the
underlying speech. The second category corresponds to data-driven methods, where
head motion sequences are generated from existing recordings. These data-driven
methods either concatenate existing head motion sequences according to given
criteria or learn statistical models to capture the distribution of head motion. A
prevalent modality used in previous data-driven studies is a set of speech prosodic
features, leveraging the strong coupling between head motion and speech (Busso
and Narayanan 2007).
The two main approaches for head movement generation have advantages and
disadvantages. On the one hand, rule-based systems have the advantage of
considering the meaning of the message to choose appropriate movements.
However, the head movements may seem repetitive, since the range and vari-
ability of head motions are usually limited to the predefined sequences per type
of movement stored in the system. Under similar conditions, the system will tend
to generate similar behaviors oversimplifying the complex relationship between
verbal and nonverbal information. Furthermore, forcing synchronization between
behaviors and speech is challenging (e.g., the coupling between speech and head
motion). On the other hand, data-driven frameworks have the potential to capture
the range of behaviors seen in real human interaction, creating novel realizations
that resemble natural head motions. When speech features are used to generate
head motion, the models can automatically learn the synchronization between the
prosodic structure and head movements. However, using solely data-driven
models may disregard the semantic content of the message, resulting in move-
ments that are not aligned with the message. These systems may generate
perfectly synchronized emblems contradicting the message (e.g., shaking the
head during affirmation). To balance the tradeoff between naturalness and appro-
priateness, studies have attempted to bridge the gap between both methods
creating hybrid approaches that leverage the advantages of both methods, over-
coming their limitations (Chiu et al. 2015; Sadoughi and Busso 2015; Sadoughi
et al. 2014; Stone et al. 2004).
This chapter describes the role of head motion in human interaction, emphasiz-
ing the importance of synthesizing behaviors that properly convey the relation
between head motion and other verbal and nonverbal channels. We review influ-
ential studies proposing rule-based, data-driven, and hybrid frameworks. The
chapter also discusses open challenges that can lead to new advances in this
research area.
Table 1 Some of the head motion roles identified by Heylen (2005)

Head motion functions
Show affirmation or negation
Show inclusivity or intensification
Organize the interaction
Mark the listing
Mark the contrast
Show the level of understanding
Mark uncertain statements
Facilitate turn taking/giving
Signal ground holding
Signal the mood
Signal shyness and hesitation
Backchannel requests
State of the Art
Head motion plays an important role during human communication. This section
summarizes relevant studies describing the function of head motion during human
interaction (section “Role of Head Motion in Human Interaction”), emphasizing the
strong relationship with other verbal and nonverbal channels (section “Relation
between Head Motion and Speech”).
Role of Head Motion in Human Interaction
Heylen (2005) surveyed studies analyzing the role of head motion during human
conversation, listing 25 different roles, including enhancing communicative atten-
tion, marking contrast between sentences, and communicating the degree of under-
standing. Table 1 lists some of these functions. It is clear that head movement is an
essential part of body language, which facilitates human-human interaction not only
while speaking, but also while listening (McClave 2000).
Speakers use head movements to reinforce the meaning of the message. We often
use emblems such as head nods for affirmations, head shakes for negation, and head
tilt with words like “um,” “uh,” and “well” (Lee and Marsella 2006; Liu et al. 2012).
Lee and Marsella (2006) investigated the nonverbal behaviors of individuals during
dyadic interactions. They annotated the videos in terms of a set of discourse
functions including affirmation, negation, contrast, intensification, inclusivity, obli-
gation, listing, assumption, possibility, response, request, and word search. They
found that generally there are nonverbal behavior patterns related to head motion
accompanying these labels (e.g., head shake during negation, head nod during
affirmation, head shake during the use of words such as “really,” and lateral head
sweep during the use of words such as “everything,” “all,” and “whole”). Head
motion is also used to parse syntactic information, creating visual markers to

segment phrases within an utterance. Hadar et al. (1983) recorded head movements
from four subjects during conversation, reporting that after removing pauses of more
than 1 s, 58.8% of the still head poses occurred during speech pauses. Graf et al.
(2002) observed frequently an initial head movement after a pause, which is
followed by speech. Another important communicative function of head motion is
to stress words, functioning as a visual marker for intonation (Graf et al. 2002;
Moubayed et al. 2010). These aspects are important during human interaction. For
example, Munhall et al. (2004) studied the effect of head movements on speech
intelligibility. They conducted an evaluation where an animated face replicated the
original head and face motions of recorded sentences. The task was to recognize
speech in the presence of noisy audio. They evaluated two conditions for the
animated face: with head motion and without head motion. They count the number
of correctly identified syllables, showing improved performance when the animated
face contained head motion.
Head motion also plays a key role while listening, where people provide nonver-
bal feedback to the speaker using primarily their head movements. A common
behavior is to nod to acknowledge active listening (McClave 2000). Ishi et al.
(2014) analyzed the occurrences of head movements during the listeners’s back-
channels such as “yes” and “uhm,” observing one or multiple head nods which were
timely synchronized with the verbal backchannel.
Head movements also convey the affective state of the speaker (Busso et al.
2007b; Busso and Narayanan 2007). In our previous work, we studied the displace-
ment of head motions of an individual expressing different emotions: happiness,
anger, sadness, and neutrality (Busso and Narayanan 2007). Our results showed
significant difference in head motions across all emotions, except between happiness
and anger. In another study (Busso et al. 2007a), we demonstrated that head motion
behaviors are discriminative for emotion recognition. Using only global statistics
derived from head motion trajectories at the sentence level, we were able to
recognize between these four emotional states with 65.5% accuracy (performance
at chance was 25%). This study also demonstrated the contribution of head motion
on emotional perception. We generated animations of expressive sentences. The
novelty of the approach was that we purposely created mismatches between the
emotion in the sentence (e.g., happiness) and the emotion on the head motion
sequence (e.g., sadness). The corpus used in this study has sentences read by an
actor conveying the same emotions. Therefore, we were able to create these mis-
matches by temporally aligning the corresponding frames across emotions. The
evaluators rated the emotional content in terms of activation, valence and domi-
nance, using a five Likert-like scale for each emotional dimension. Figure 1 shows
the results for valence (1– very positive; 5– very negative). The first bar in each of
the plot represents the matched condition, where the emotion on the sentence match
the emotion of the head motion sequence. The next three bars provide the perception
achieved in mismatched conditions by changing the emotion of the head motion
sequences. The bar “FIX” represents the perception achieved without any head
motion. Finally, the bar “WAV” gives the perception achieved when the stimuli
a b
5 5
4 4
3 3
2 2
1 1
HAP NEU SAD ANG FIX WAV ANG NEU SAD HAP FIX WAV
Happiness Anger
c d
5 5
4 4
3 3
2 2
1 1
SAD NEU HAP ANG FIX WAV NEU SAD HAP ANG FIX WAV
Sadness Neutral
Fig. 1 These figures show the perceived valence (1: positive, 5: negative) for four emotional
categories, when the head movements are generated with the same emotional class (i.e., matched
condition, first bar), with three other emotions (i.e., mismatched condition, second to fourth bars),
when the head is fixed (FIX), and when the evaluators only listened to the audio (WAV)
only included speech. These figures show that expressive head motion sequences
change the emotional perception of the animation. For the neutral sentences, adding
an angry head motion sequence makes the animation more negative, and adding a
happy head motion sequence makes the animation more positive. Similarly, Lance
and Marsella (2007) proposed to include emotional head movements during gaze
shift when synthesizing the animations. The result of their study showed that people
distinguished differences between high versus low level of arousal and high versus
low level of dominance. These results indicate that modeling expressive behaviors in
synthesizing head motion sequences is important.
Head motion also affects perception of personality traits. Arellano et al. (2011)
performed a perceptual evaluation on static images of a character with various head
orientations and gazes. The result showed that people’s perception of personality
traits such as agreeableness and emotional stability was affected by head orientation,
while for gaze no significant difference was revealed. Arya et al. (2006) analyzed the
effect of head movement and facial actions on the perceived personality by others.
They performed perceptual evaluations on a set of animated videos to see the effect
of visual cues on the perception of personality based on two parameters: affiliation
and dominance. They created several videos, each with a specific facial action, and
asked the evaluators to rank them with a set of attributes. The result of this study
showed that dynamic head movements such as head tilt and eye gaze aversion
communicate a sense of dominance for the character. Moreover, the results showed
that the higher the frequency and intensity of the head movements are, the higher is
their perceived level of dominance.
Relation Between Head Motion and Speech
Data-driven models have the potential of capturing naturalistic variations of the

behaviors (Foster 2007). One useful and accessible modality that can be used to
drive facial behaviors is speech. Spoken language carries important information
beyond the verbal message that a CA engine should capitalize on. Therefore, this
chapter focuses the discussion on data-driven frameworks relying on speech fea-
tures. Head motion and speech are intrinsically connected at various levels (Busso
and Narayanan 2007). As mentioned in section “Role of Head Motion in Human
Interaction,” head motion conveys visual markers of intonation, defining syntactic
boundaries and stressed segments. As a result, speech features and head motion
sequences are highly correlated.
Several studies have reported a strong correlation between speech prosody
features and head movements. Munhall et al. (2004) analyzed the correlation
between head motion and the prosodic features including fundamental frequency
and RMS energy. The study focused on recordings from a single subject. The
correlation at the sentence level between head motion and the fundamental fre-
quency was ρ ¼ 0:63 , and between head motion and the RMS energy was ρ ¼
0:324. Kuratate et al. (1999) showed a correlation of ρ ¼ 0:88 between head motion
and the fundamental frequency for an American English speaker. We also reported
similar results in Busso et al. (2005) using pitch, intensity, and their first and second
order derivatives as speech prosodic features. We evaluate the relationship between
head and speech features using canonical correlation analysis (CCA). CCA projects
two modalities with similar or different dimensions into a common space where their
correlation is maximized. The CCA for head and speech features was ρ ¼ 0:7 at the
sentence level highlighting the strong connection between them. The study was
further extended in Busso and Narayanan (2007), observing similar results.
Studies have also shown co-occurrence of head movements and speech prosody
events. Graf et al. (2002) showed that although the amplitude and direction of the
movements may vary according to idiosyncratic characteristics, semantic content of
the message, and affective state of the speaker, there is a common synchrony of the
timings between pitch accents and head events. Mcclave (2000) reported that there is
Table 2 Perception of CAs synthesized with and without head motion reported in previous studies
ranging from 1 (bad) to either 5 or 7 (great). Some of these values are approximated from figures in
their corresponding publications
With head Without head
Study Criterion movement movement
Busso et al. (2007a) Naturalness (1–5) 3.61 3.10
Mariooryad and Busso Naturalness (1–5) ~2.90 2.32
(2012)
Welbergen et al. (2015) Warmth (1–7) ~5.10 ~4.55
Competence (1–7) ~6.10 ~5.90
Human-likeness ~4.50 ~4.50
(1–7)
co-occurrence between head movement patterns and the meaning of speech. For
instance, head shakes happen during expressions of inclusivity and intensification.
Lee and Marsella (2009) proposed a hidden Markov model (HMM) classifier to
detect head nods based on features selected from speech including part of speech
(PoS) (e.g., conjunction, proper noun, adverb, and interjection), dialog acts (e.g.,
backchannel, inform, suggest), phrases, and verb boundaries. Their classifier showed
high performance, indicating a close connection between head nods, dialog acts, and
timing of the uttered words.
Head Movement Synthesis
It is important to generate head movements for CAs, due to its role in conveying the
intended message, emotion, and level of rapport displayed by speakers. Table 2 lists
some studies which have compared facial animations synthesized with and without
head motion, in terms of naturalness, warmth, competence, and human-likeness.
These studies show the clear benefit of using head motion.
Head motion has three degrees of freedom (DOF) for rotation and three DOF for
translation. While some methods consider all six DOF, most studies rely only on the
three rotation angles (Fig. 2). The studies focused on the synthesis of head motion
are usually based on rule-based or data-driven methods. Rule-based systems define
several rules about the shape and timing of the head movements and use a predefined
handcrafted dictionary of head gestures to synthesize them. While these gestures are
usually selected based on the meaning of the message, their variations are limited to
the list of gestures defined in the system. Also, local synchronization and timing of
these gestures with speech is challenging. Data-driven methods utilize prerecorded
motion databases. These methods usually concatenate the prerecorded motions to
create a new realization or create them by sampling from the models trained on the
recordings. Due to the correlation between head movements and speech prosody
features, these methods usually consider speech prosody features in generating the
Fig. 2 Three degrees of freedom for head motion rotation. Some studies also include three degrees
of freedom for head translation
movements. This approach also facilitates the synchronization between speech and
gestures, capturing subtle timing relations between prosody and head motion. Also,
these methods have the potential to capture the range of motions seen in real
recordings. However, their main drawback is that these methods disregard the
meaning of the message while creating the movements. Therefore, the movements
are not constrained to convey the same meaning as the speech and may even
contradict the message (e.g., head nods for negations).
Foster (2007) compared rule-based and data-driven generation of head move-
ments. The result of this evaluation showed that people preferred facial animations
generated with data-driven methods more than rule-based methods; however, the
difference was not statistically significant. The study also concluded that the range of
the displays for data-driven method was more similar to the original recordings than
the displays obtained with rule-based systems.
Rule-based systems and data-driven methods have key features that are ideal to
synthesize human-like head motions. This section describes influential studies for
rule-based systems (section “Rule-Based Methods”) and data-driven methods (sec-
tion “Data-Driven Models”). It also summarizes efforts to create hybrid approaches
which leverage the benefits of both of these methods (section “Hybrid Approaches”).
Rule-Based Methods
Rule-based methods define rules for head movements to communicate the meaning
more clearly. Table 3 summarizes some of the rules defined by previous studies. One
of the most influential studies on rule-based systems was presented by Cassell et al.
(1994). They designed a system to generate synchronized speech, intonations, and
gestures (including head motion) by defining rules. For example, they generated
head nods during emphatic segments or backchannels. Their system approximated
gaze with head orientation, and, therefore, all the rules for gaze behaviors involved
specific head motions. For example, the CA would look up during question, look at
the listener during answer, look away at the beginning of a long turn, and look at the
Table 3 This table provides a brief summary of the rules proposed in the previous studies which
provide a mapping from the discourse function or intention to specific head movements
Study Mapping
Lexical/emotional affiliate Head movement/pose
Cassell et al. Backchannel Head nods
(1994)
Emphasis Head nods
Question Look up
Answer Look away
Beginning of turn Look away
Turn request Look up
Pelachaud et al. Anger Forward pose
(1996)
Sadness Downward
Disgust Backward and up
Fear Backward
Sadness Downward
Surprise Backward
Liu et al. (2012) Backchannel Head nod
End of question Head nod
End of turn when giving the turn to the Head nod
interlocutor
Keeping a turn by a short pause Head nod
Thinking, but keeping the turn Head tilt
Thinking and preparing the next utterance, Head tilt
e.g., “uhmm”
Gratch et al. Lowering of pitch of interlocutor Head nod
(2006)
Raised loudness of interlocutor Head nod
Speech disfluency of interlocutor Posture/gaze shift
Posture/gaze shift of interlocutor Mimic
Nods or shakes of interlocutor Mimic
Marsella et al. Affirmation Big nod, tilt left nod, tilt
(2013) right nod
Negation Shake, small shake
Contrast Tilt right, tilt left
Mental state Tilt half nod left
Emphasis Small nod
listener for short turns. The extension of this framework resulted in REA, a CA
which responded with gestures to different discourse functions (Cassell et al. 1999).
DeCarlo et al. (2004) presented RUTH, a platform architecture for embodied
CAs. The inputs of this platform are enriched transcriptions with prosodic and
gestural markers at the word level. The prosodic markers correspond to the tones
and break indices (ToBI), which define pitch accents and boundary tones (Silverman
et al. 1992). The gestural markers are predefined behaviors. For head motion, they
defined 14 types of head motions. They considered variations of head nods (upward,
downward, upward with some rightward, upward with some leftward, downward
with some rightward, and downward with some leftward) and head tilts (clockwise,
counter clockwise, clockwise with downward nodding, counter clockwise with
downward nodding). They also defined gestures to move (forward, backward) or
turn (to the right or to the left) the head. These gestures are then rendered, synchro-
nizing the behaviors at the points specified by the tags.
There are studies that have attempted to incorporate head motions conveying
emotions using rule-based systems. Pelachaud et al. (1996) developed a system to
generate expressive facial and head movements, as well as eye movements. As input,
they used the transcriptions tagged with accents, the desired emotional state, and its
intensity. They used head and eye movements as regulators, which facilitate the
communication between the speaker and listener, in a rule-based manner. They also
defined rules specifying the head direction depending on the target emotional state,
following the results from previous psychological studies. The emotional head
motion rules include moving forward during anger, downward during sadness, and
backward during surprise. Marsella et al. (2013) proposed an emotionally aware
rule-based system. Their framework relied on syntactic and acoustic analysis on the
input speech consisting of either natural or synthesized speech. They used the
acoustic analysis to find the emotional state and word emphasis and the syntactic
analysis to find the appropriate category of behaviors to be synthesized for each
communicative goal. Their proposed system also handles co-articulation for consec-
utive behaviors that are close in time, leading to novel realizations while trans-
itioning from one gesture to another.
To define the rules, there are studies investigating video recordings of human
interaction, aiming to identify consistent patterns between movements (including
head motion) and discourse features. Kipp (2003) analyzed human gestures in
23 clips of a TV show. They developed ANVIL, a video annotation toolkit to
annotate the gestures. They found a common set of 15 gestures occurring across
the two speakers. They defined these gesture profiles, including the position and
orientation of the head and hands during the gestures. For synthesis, they automat-
ically annotate the transcription with words, PoS, what the utterance is about (theme,
rheme, and focus), and discourse relations (opposition, repetition, and listing). They
used carefully designed rules to map these tags to a set of semantic tags. Using these
semantic tags, and the statistics derived from their annotated corpus for these tags,
they choose the most probable gesture considering local and global constraints.
Following a similar approach, Liu et al. (2012) proposed a rule-based approach to
appropriately generate head tilts and head nods, where the rules were derived by
observing and analyzing human interaction data. First, they annotated the phrases in
their database with a list of dialog acts, along with head nods and tilts. Second, they
created a mapping between dialog acts and the corresponding head movements.
They found frequent occurrences of head nods during backchannels and the last
syllable of strong phrase boundaries. They also found head tilts during weak phrase
boundaries and segments when the individual was either thinking or embarrassed.
They exploited these relations in the generation of head nods and tilts for human-
robot interaction (HRI). They used a fixed shape trajectory for head nod and tilt,
driven by the rules learnt from their corpus. They used perceptual evaluational to
measure the perceived naturalness of the head movements visualized on robots.
Their results showed improved naturalness when both head nods and tilts are
incorporated in the system compared with the case when they used only head nods
and when they used the original sequences.
Generating head motion for CAs is not only important while speaking but also
while listening. Showing rapport is one of the aspects that needs to be considered for
generating believable CAs. Gratch et al. (2006) proposed a virtual listener called
virtual rapport, which aims to create a sense of rapport with the user, by defining
heuristic rules from previous psychological studies. For example, the CA nods
whenever it senses that the users lower their pitch or raise their loudness. It also
nods when the speaker nods.
Similar rule-based systems include the work of Rickel and Johnson (1998)
(Steven) André et al. (1996) (the DFKI Persona) Beskow and McGlashan (1997)
(Olga), Lester et al. (1999) (pedagogical agents), and Smid et al. (2004). An
important drawback of rule-based systems is that they cannot easily capture the
rich and complex variability observed in natural recordings, resulting many times in
repetitive behaviors (Foster 2007).
Data-Driven Models
The second category of approaches to generate head motion corresponds to data-

driven frameworks. Data-driven approaches usually utilize motion capture record-
ings of head motion trajectories. Table 4 summarizes some of these frameworks,
highlighting the input that drives the approaches.
Studies have created head movements by blending segments of recorded motion
capture data. Chuang and Bregler (2005) designed a system to create emotional
facial gestures. For head motion, they stored segments of pitch contours and their
corresponding head motions. During synthesis, they searched for a combination of
pitch contours in their stored libraries, finding the best matching contours. Then,
they connected the corresponding sequence of head motion segments, re-sampling
the sequence to match the timing of the input. Deng et al. (2004) proposed a similar
approach using K Nearest Neighbors (KNNs). They stored the training audio and
head motion trajectories indexed by the audio features and used dynamic program-
ming to search for the most appropriate set of motion segments using seven nearest
neighbors. They allowed the user to specify head poses for key frames, which were
added as constraints in the dynamic programming search. They considered smooth-
ness of the trajectory as one of the factors in their optimization process, avoiding
sudden transitions.
Le et al. (2012) proposed a framework based on Gaussian mixture model (GMM)
to generate head motion driven by prosodic features (loudness and fundamental
frequency). Their framework learns three separate joint GMMs, modeling the
Table 4 Brief summary of data-driven methods proposed in previous studies. The table lists the
corresponding input during testing and the approach used to synthesize the head motion sequences
Study Input Method
Chuang and Bregler Pitch, target expressive style KNN, path searching
(2005)
Deng et al. (2004) Pitch, five formants, 13-MFCC, KNN, path optimization
12-LPC
Busso et al. (2005) Pitch, intensity HMMs
Busso et al. (2007a) Pitch, intensity, emotion HMMs
Sargin et al. (2008) Pitch, intensity PHMMs
Mariooryad and Busso Pitch, intensity DBNs
(2012)
Chiu and Marsella (2011) Pitch, intensity CRBMs
Levine et al. (2010) Pitch, intensity, syllable length HCRFs, reinforcement
learning
Le et al. (2012) Pitch, intensity GMMs
relation between speech prosody features and (1) head poses, (2) the velocity of head
motion, and (3) the acceleration of head motion. They approximate the joint distri-
bution of them, by assuming that they are independent (product of the probabilities
provided by the GMMs). Having the head pose at the two previous frames, and the
prosodic features for the current frame, they find the current head poses by maxi-
mizing the final joint distribution using gradient descent. A key advantage of this
approach is that it can run online facilitating real-time implementation.
There are other data-driven studies that use probabilistic modeling strategies that
capture the temporal dynamic of head motion. Examples of these frameworks
include HMMs and dynamic Bayesian networks (DBNs). For example, we presented
a framework based on HMM for modeling the relationship between speech prosodic
features and head motion (Busso et al. 2005). We used vector quantization to
quantize the space of head motion and designed HMMs to learn the joint represen-
tation of head movements and speech prosodic features. Figure 3 gives the block
diagram of this study, where the RMS energy, the fundamental frequency, and their
first and second order derivatives were used to create a 6D feature vector. The
HMMs represents head poses where their transitions were learned during training.
The HMMs decode the most likely sequence of head poses given a speech signal,
where head poses transitions that are common are rewarded and uncommon transi-
tions are penalized. Given the discrete representation of head poses used in this
study, we smoothed the angular trajectories of the generated head poses. This
framework was very effective in generating head motion sequences that are timely
aligned with prosodic information. Following this study, we extended the HMM
approach to incorporate the relationship between prosody and head motion under
different emotional states (Busso et al. 2007a, b). The results showed that the models
were able to generate expressive head motions accompanying speech. Other studies
have also proposed speech-driven models to synthesize head motion. Sargin et al.
(2008) used parallel HMMs (PHMMs) to jointly model speech and head movements
Feature
Extraction
HMM Sequence Generator
Spherical
Head Motion Vector
Synthesis
Cubic + Quantization
Interpolation
Noise
Generation
Fig. 3 The block diagram of the approach proposed by Busso et al. (2005), where HMMs are used
to synthesize head motion sequences driven from speech
a b c
(Sargin et al. 2008) (Mariooryad and Busso 2012) (Taylor et al. 2006)
Fig. 4 (a) PHMM proposed by Sargin et al. (2008) to model the relationship between head
movement primitives and speech prosody, (b) The DBN model proposed by Mariooryad and
Busso (2012) to jointly model the relationship between head and eyebrow movements with speech
prosody features, and (c) CRBM proposed by Taylor et al. (2006) to learn the human motion
by simultaneously clustering and segmenting the two modalities. PHMM consists of

several left-to-right HMMs, where each branch models a head motion primitive
automatically extracted from the data (see Fig. 4a). PHMM jointly solves the
segmentation and clustering of head motion sequences. In their study, they found
the most probable state sequence and their corresponding head motion values for a
given speech signal.
DBNs are another suitable framework to capture the relation between head
motion and speech. DBN is a generative model that provides the flexibility to impose
different structures by introducing nodes representing variables and direct links
representing conditional dependencies between the variables. Therefore, it can
model the dependencies between two temporal sequences in a principled way.
Notice that HMM is a particular type of DBN. We have demonstrated the potential
of DBNs to model the relation between head motion and speech. Mariooryad and
Busso (2012) designed several structures of DBNs to capture the joint representation
of speech with not only head movements but also eyebrow movements. Figure 4b
shows an example, where the Head&Eyebrow node represents a jointly discrete state
describing eyebrow and head motion. During training, all the variables are available.
During synthesis, the Head&Eyebrow node is not available but is approximated by
propagating the evidences from the Speech node. The animations generated with this
method were compared with subjective and objective metrics demonstrating the
need to jointly model eyebrow and head motion together.
There are data-driven methods relying on conditional restricted Boltzmann
machines (CRBMs) (Chiu and Marsella 2011). CRBM provides an efficient non-
linear tool for modeling the global dynamics and local constraints of a temporal
signal (see Fig. 4c). Given N þ 1 frames, the model learns the mapping between the
visible units and hidden layers, which will reconstruct missing observation during
synthesis. For synthesis, the model takes the first N frames, aiming to estimate the
N þ 1 frame. In addition, the auto regressive connections between the previous
frames and the current frame learn the temporal constraints of the data. During
synthesis, this model generates the ðN þ 1Þth sample using contrastive divergence,
based on the previous N frames. These properties make CRBM very useful for
predicting and generating temporal sequences. Taylor et al. (2006) demonstrated the
benefits of using this framework for modeling human motion trajectories (e.g.,
walking). They used CRBM with auto regressive connections for predicting the
human motion pose for the next frame, given the previous N frames. Following this
study, Taylor and Hinton (2009) proposed to add an extra variable to constrain the
CRBM’s generation based on specific stylized walk sequences such as drunk, strong,
and graceful. The success of this framework in this domain motivated Chiu and
Marsella (2011) to use a variation of CRBM to generate head motion sequences.
They proposed a hierarchical factored conditional RBM (HFCRBM), which predicts
the current head pose based on the previous two poses, constrained on speech
prosody features.
The aforementioned studies utilized either the concatenation approach or statis-
tical models to learn the relation between head motion and speech. Levine et al.
(2010) combined both strategies by using hidden conditional random fields
(HCRFs) to model the relationship between a set of kinematic features of joint
movements and speech prosodic features. The premise of the study is that prosody
is related to the head motion kinematic rather than the actual head motion. For
synthesis, they inferred the kinematic features based on the prosodic features. Next,
they searched through the recordings, in an online manner (forward path), using a
cost function that incorporates the inferred kinematic features. They used Markov
decision process (MDP) to ensure smoothness in the head motion trajectories.
Hybrid Approaches
Combining rule-based and data-driven approaches to exploit the benefits from each
method results in an enhanced system. Several studies have focused on bridging the
gap between these two methods (Huang et al. 2011; Sadoughi and Busso 2015;
a b
(Sadoughi and Busso 2014) (Sadoughi et al. 2015)
Fig. 5 Dynamic Bayesian Network of the proposed constrained models (Sadoughi and Busso
2015; Sadoughi et al. 2014). The systems generate behaviors constrained by (a) the underlying
discourse function or (b) the target gesture
Sadoughi et al. 2014; Stone et al. 2004). We describe some of these studies in this
section.
Stone et al. (2004) proposed a system to generate head and body movements,
by concatenating prerecorded audio and motion units. The key aspect of the
approach is that the units are associated with communicative functions or intents.
Therefore, they can decompose any new utterance into units and solve a dynamic
search through their data to find the best combination matching the intended
communicative function. Since the approach uses a concatenative framework, the
dynamic search also smoothes transition between speech segments and between
motion sequences. It also synchronizes emphasis points across speech and ges-
tures. To achieve this goal, they annotate the emphasis segments on their record-
ings. During testing, they provide tags describing emphasis on the transcriptions,
coordinating the emphasis on motion sequences to start and end at the
corresponding frames. The intermediate frames, which are the frames in between
the emphatic points, are derived by interpolation. The limitation of this work is
that the variations of speech and motion sequences are limited to the indexed
phrases found in the recordings.
There are other studies that have combined rule-based and data-driven
approaches by adding meaningful constraints to their models. We have designed a
speech-driven model to synthesize eyebrow and head movements constrained by
discourse functions (Sadoughi and Busso 2015; Sadoughi et al. 2014). Figure 5a
describes the structure of our first model (Sadoughi et al. 2014), built upon the DBN
model proposed by Mariooryad and Busso (2012) (see Fig. 4b). In this structure, the
Constraint node is added as a child of the hidden state Hh &e, which controls the
dependency between speech and head and eyebrow motion. During training and
synthesis, the Constraint node is given as input, which dictates the behaviors
generated by the system. This study used the IEMOCAP database (Busso et al.
2008), which consists of dyadic interaction between two actors. We manually
annotated two discourse functions corresponding to affirmation and question. The
models were trained and tested with data from a single subject. The results showed
Fig. 6 MSP-AVATAR, a corpus designed to generate behaviors constrained by the communicative

function of the message. The figure shows the placement of the reflective markers, the skeleton used
to reconstruct the data, and the setting of the recordings
that evaluators preferred the constrained models for questions. For affirmation, the
results were not conclusive.
A challenge in creating models constrained by the semantic meaning of the
sentence is the lack of motion capture databases with appropriate annotations for
discourse functions. To overcome this limitation, we recorded the MSP-AVATAR
corpus (Sadoughi et al. 2015), a motion capture corpus of dyadic interactions,
capturing facial expressions and upper body motion, including head motion
(Fig. 6). In each session, two actors improvised scenarios carefully designed to
include a set of the following communicative functions: contrast, affirmation,
negation, question, uncertainty, suggest, warn, inform. We also considered scenarios
to include iconic gestures for words such as large and small, and deictic gestures for
pronouns such as you and I. This corpus contains the audio and video of both authors
and motion capture recordings from one of them. Figure 6 shows the placement of
the markers, the marker’s skeleton, and the setting of the recordings of the corpus.
Using this corpus, we are currently extending our framework to combine rule-based
and data-driven models by considering these discourse functions.
An alternative framework to bridge rule-based and data-driven models is to
generate the behaviors dictated by the predefined rules using data-driven models.
To understand how this framework works, consider the SAIBA framework proposed
by Kopp et al. (2006). SAIBA is a behavior generation framework for embodied
conversational agents (ECAs) composed of three layers: intent planning, behavior
planning, and behavior realization. The first two layers define the intent of the
message and the gestures required to convey the communicative goal. We envision
rule-based systems to create these layers. The last layer generates the intended
behavior by setting the amplitude and timing constraints. We envision data-driven
models to create this layer. Data-driven models will generate novel realization of
specific gestures defined by the behavior planning layer. We have explored this
approach for head and hand gestures (Sadoughi and Busso 2015). Figure 5b illus-
trates the proposed system, where the Constraint node is placed as a parent of the
Fig. 7 This figure shows the overall block diagram of the method proposed by Sadoughi and Busso
(2015) to retrieve arbitrary prototypical head movements. The approach only requires few examples
of the target behaviors
hidden state Hh &e. The key novelty is that the constraints correspond to specific
behaviors. For head motion, we only considered head nods and head shakes, but the
system is flexible to incorporate other behaviors. Notice that we need several
examples of the target behaviors to train the proposed model. We addressed this
key problem with a semi-supervised approach to retrieve examples of the target
behaviors from the database. This framework, illustrated in Fig. 7, requires few
samples for training, which are used to automatically retrieve similar examples
from the database. The first step searches for possible matches using one-class
support vector machine (SVM). We use temporal reduction and multi-scale win-
dows to handle similar gestures with different durations. The classifiers are fast.
They are set to identify many candidates segments conveying the target gesture.
The second step uses dynamic time alignment kernel (DTAK) to improve the
precision of the system by removing samples that are not similar to the given
examples. We use the retrieved samples to train the speech-driven framework
described in Fig. 5b, generating novel data-driven realization of the target behav-
iors (e.g., head shakes and head nods).
Another interesting domain to combine rule-base and data-driven systems is in
the generation of behaviors while the CAs are listening. As we mentioned in section
“Rule-Based Methods,” Gratch et al. (2006) proposed a rule-based systems to
generate a virtual rapport. Their team extended their framework, entitled virtual
rapport 2.0, by shifting their approach towards a more data-driven approach (Huang
et al. 2011). The approach relies on an interesting data collection design described in
Huang et al. (2010) to analyze human responses during an interaction. They col-
lected data from subjects watching a story teller in a prerecorded video. Their task
was to press a key each time they felt backchannel (verbal and nonverbal feedbacks,
such as head nods, “uh-huh,” or “OK”) were appropriate. The subjects were
informed of the interaction goal, which was to promote rapport. They collect
multiple subjects under the same interaction to separate the idiosyncratic responses
from essential responses. Using these recordings, they train a conditional random
field (CRF) model, which uses these recorded videos to predict when and how to
generate backchannels. The input of their system includes pause and the user’s eye
gaze, generating different types of nodding as output. For predicting the end of the
speaking turn, they defined rules using the verbal and nonverbal cues observed in
their data. To make the CA system more friendly, they embedded a smile detector,
smiling whenever the system detects that the user smiles. This version created a
higher sense of rapport in the speakers.
Open Challenges
Generating meaningful head motion sequences conveying the range of behaviors

observed during human interaction is an important problem. This area offers inter-
esting challenges that future research should address. We describe some of these
challenges in this section.
Speech-Driven Models Using Synthetic Speech
An important limitation for speech-driven methods is the assumption that natural

speech is available to synthesize head motion. Having prerecorded audio for each
sentence spoken by the CA is not realistic in many domains. Instead, text-to-speech
(TTS) systems provide the flexibility to scale the system beyond prerecorded
sentences. An advantage for rule-based systems is that the rules are generally derived
from transcriptions instead of speech features. Therefore, they can easily handle CAs
using synthetic speech. For speech-driven frameworks, however, the models rely on
acoustic features derived from natural speech. Using synthetic speech is a major
limitation. There are very few studies that have addressed this problem. Welbergen
et al. (2015) provided a framework to generate head movements for a CA driven by
synthetic speech. They used the probabilistic model proposed by Le et al. (2012).
They tested their framework with synthetic speech, performing subjective evaluation
to assess the warmth, competence, and human-likeness of their animations. The
results showed that adding head movements by using an online implementation of
their framework increases the perception level of these social attributes. Although
the approach proposed by Welbergen et al. (2015) uses synthetic speech, their
system has a mismatch between train and test conditions. During training, the
models are built with original speech. During synthesis, the models are driven by
features extracted from synthetic speech. Features extracted from synthetic speech
do not have the same dynamic range as features derived from original speech. Given
these clear differences in the feature space between natural and synthetic speech, this
mismatch produces very limited range of behaviors. We are investigating systematic
approaches to address this problem by using adaptation techniques that reduce the
mismatch between train and test conditions and increase the range of behaviors
generated by the models (Sadoughi and Busso 2016). Solving this problem can
dramatically increase the application domain where speech-driven animation can be
used.
Exploring Entrainment
As discussed in section “Role of Head Motion in Human Interaction,” head motion

conveys the emotional state of the message (Busso et al. 2007a). An open challenge
is to identify effective frameworks to generate head motion sequences that elicit a
target emotion. While predefined rules can be used (Pelachaud et al. 1996), data-
driven frameworks may provide more realistic sequences (Busso et al. 2007a). These
systems, which are able to convey expressive behaviors, open opportunities to explore
entrainment effects between the user and the CA. Entrainment is the phenomenon
during human interaction where interlocutors mirror the behaviors of other. This
phenomenon, which affect lexical, prosodic, and gestural cues, has been also observed
during human computer/robot interaction (Bell et al. 2003; Breazeal 2002). Interest-
ingly, we have also observed entrainment effects on emotional behaviors during
dyadic interactions (Mariooryad and Busso 2013). Can a CA manipulate its emotional
reactions to affect the affective state of the user? The head movement of the CA can be
modulated with the appropriate emotional cues to increase the emotional entrainment
with the user. Attempting to capture this subtle communicative aspect can lead to more
effective CAs with better rapport with the user. Although Huang et al. (2011) proposed
to make the CA more friendly by producing smiles as a response of smiles from users,
this area offers opportunities to systematically design interfaces beyond that, leverag-
ing the findings from entrainment studies (Jakkam and Busso 2016). The first step in
this direction requires investigation of emotional entrainment in human conversations
(Mariooryad and Busso 2013; Xiao et al. 2015). The investigation can be used to
create appropriate affective cues for the CAs, increasing the emotional entrainment by
developing models that incorporate relevant factors.
Modeling Personality
Since individual differences play an important role in the range of head motion
shown during human interaction (Youssef et al. 2013), the model to synthesize head
motion sequences should carefully consider personality and idiosyncratic differ-
ences. Note that personality and emotional displays are interconnected. For instance,
an introvert and an extrovert person will express their emotions differently under the
same circumstance. There are studies proposing framework to incorporate person-
ality traits in their CAs, especially for rule-based methods. Using a rule-based
strategy, Poggi et al. (2005) proposed to modulate the goals of the message
according to the personality traits of the ECA. Kipp (2003) proposed to investigate
the gesture profiles displayed by two speakers. They found that the gestures of the
speakers were different in important ways. There were some gestures only used by
one of the speakers. They also found important differences in the frequency of their
gestures, the timing patterns, and the mapping functions used to link semantic tag to
actual gestures. They utilized all these aspects to personalize their animated charac-
ters. However, this was a limited study, and more effort is required to generalize the
models to a broader range of personalities.
Joint Models to Integrate Head Motion with Other Gestures
Another remaining challenge is how to integrate the generated head movements with
the movements of other parts of the body. There is a high synchrony between head
movements and facial gestures. For example, we have reported a high CCA between
head and eyebrow movements ( ρ ¼ 0:89 ) (Mariooryad and Busso 2012). When
individual speech-driven models are separately used to synthesize individual behav-
iors, the relationship between these behaviors may not be preserved. For example,
we can perfectly capture the timing relationship between speech and head motion
and between speech and eyebrow motion. However, the generated head and eyebrow
motion may not be perceived realistic when rendering the CA, as these behaviors
may fail to capture the relation between head and eyebrow motion. In Mariooryad
and Busso (2012), we proposed to jointly model head and eyebrow motion in a
speech-driven framework. The result of this study showed that people preferred the
animations synthesized by the joint model rather than the ones where the behaviors
were independently generated. Capturing these subtle dependencies is not only
important for generating realistic behaviors but also for conveying synthesized
behaviors that increase speech intelligibility (Munhall et al. 2004). Having a data-
driven model which incorporates all these relations will result in a more convincing
animation. The challenge is that modeling more modalities will increase the com-
plexity of the model. Extending data-driven approaches without significantly
increasing their complexity is an open challenge.
Conclusions
This chapter gives an overview on the studies relevant to head motion generation.
We started by reviewing the importance of head motion in human interactions. Head
movements play an important role in face to face communication. They provide
semantic and syntactic cues while speaking. We use head motion as a backchannel
while listening to other. They play an important role in conveying personality and
emotional traits. The functions are important for communication, so realistic CAs
should have well-designed head motion sequences that are timely synchronized with
speech.
The chapter overviewed different methods to generate head movements, which
can be categorized into two main approaches: rule-based and data-driven frame-
works. Rule-based methods rely on heuristic rules to generate the head movements
based on the underlying communicative goal of the message. Data-driven methods
rely on recorded head motion sequences to generate new instances. Within data-
driven methods, we focused the review on speech-driven frameworks which lever-
ages the close relationship between prosody and head motion. Rule-based and data-
driven methods have their own advantages and disadvantages. We reviewed hybrid
approaches which have attempted to bridge the gap between these methods, over-
coming their limitations.
There are still open challenges in generating realistic head motion sequences. We
discussed opportunities which we believe can result in head motion sequences that
are more effective and engaging. Previous studies have built the foundation to
understand better the role of head motion. They have also provided convincing
frameworks to generate human-like head motion sequences. They offer a perfect
platform for future studies to advance even more this research area.
Acknowledgments This work was funded by National Science Foundation under grant
IIS-1352950.
References
André E, Müller J, Rist T (1996) The PPP persona: a multipurpose animated presentation agent. In:
Workshop on advanced visual interfaces, Gubbio, pp 245–247
Arellano D, Varona J, Perales FJ, Bee N, Janowski K, André EE (2011) Influence of head
orientation in perception of personality traits in virtual agents. In: The 10th international
conference on autonomous agents and multiagent systems-Volume 3, Taipei, pp 1093–1094
Arya A, Jefferies L, Enns J, DiPaola S (2006) Facial actions as visual cues for personality. Comput
Anim Virtual Worlds 17(3–4):371–382
Bell L, Gustafson J, Heldner M (2003) Prosodic adaptation in human-computer interaction. In: 15th
international congress of phonetic sciences (ICPhS 03), Barcelona, pp 2453–2456
Beskow J, McGlashan S (1997) Olga – a conversational agent with gestures. In: Proceedings of the
IJCAI 1997 workshop on animated interface agents: making them intelligent, Nagoya
Breazeal C (2002) Regulation and entrainment in human-robot interaction. Int J Robot Res 21
(10–11):883–902
Busso C, Narayanan S (2007) Interrelation between speech and facial gestures in emotional
utterances: a single subject study. IEEE Trans Audio, Speech Lang Process 15(8):2331–2347
Busso C, Deng Z, Neumann U, Narayanan S (2005) Natural head motion synthesis driven by
acoustic prosodic features. Comput Anim Virtual Worlds 16(3–4):283–290
Busso C, Deng Z, Grimm M, Neumann U, Narayanan S (2007a) Rigid head motion in expressive
speech animation: analysis and synthesis. IEEE Trans Audio, Speech Lang Process 15
(3):1075–1086
Busso C, Deng Z, Neumann U, Narayanan S (2007b) Learning expressive human-like head motion
sequences from speech. In: Deng Z, Neumann U (eds) Data-driven 3D facial animations.
Springer-Verlag London Ltd, Surrey, pp 113–131
Busso C, Bulut M, Lee C, Kazemzadeh A, Mower E, Kim S, Chang J, Lee S, Narayanan S (2008)
IEMOCAP: Interactive emotional dyadic motion capture database. J Lang Resour Eval 42
(4):335–359
Cassell J, Pelachaud C, Badler N, Steedman M, Achorn B, Bechet T, Douville B, Prevost S, Stone
M (1994) Animated conversation: rule-based generation of facial expression gesture and spoken
intonation for multiple conversational agents. In: Computer graphics (Proc. of ACM
SIGGRAPH’94), Orlando, pp 413–420
Cassell J, Bickmore T, Billinghurst M, Campbell L, Chang K, Vilhjalmsson H, Yan H (1999)
Embodiment in conversational interfaces: Rea. In: International conference on human factors in
computing systems (CHI-99), Pittsburgh, pp 520–527
Chiu C-C, Marsella S (2011) How to train your avatar: a data driven approach to gesture generation.
In: Intelligent virtual agents, Reykjavik, pp 127–140
Chiu C-C, Morency L-P, Marsella S (2015) Predicting co-verbal gestures: a deep and temporal
modeling approach. In: Intelligent virtual agents, Delft, pp 152–166
Chuang E, Bregler C (2005) Mood swings: expressive speech animation. ACM Trans Graph 24
(2):331–347
DeCarlo D, Stone M, Revilla C, Venditti JJ (2004) Specifying and animating facial signals for
discourse in embodied conversational agents. Comput Anim Virtual Worlds 15(1):27–38
Deng Z, Busso C, Narayanan S, Neumann U (2004) Audio-based head motion synthesis for avatar-
based telepresence systems. In: ACM SIGMM 2004 workshop on effective telepresence (ETP
2004). ACM Press, New York, pp 24–30
Foster ME (2007) Comparing rule-based and data-driven selection of facial displays. In: Workshop
on embodied language processing, association for computational linguistics, Prague, pp 1–8
Graf HP, Cosatto E, Strom V, Huang FJ (2002) Visual prosody: facial movements accompanying
speech. In: Proceedings of IEEE international conference on automatic faces and gesture
recognition, Washington, DC, pp 396–401
Gratch J, Okhmatovskaia A, Lamothe F, Marsella S, Morales M, van der Werf R, Morency L (2006)
Virtual rapport. In: 6th international conference on intelligent virtual agents (IVA 2006), Marina
del Rey
Hadar U, Steiner TJ, Grant EC, Rose FC (1983) Kinematics of head movements accompanying
speech during conversation. Hum Mov Sci 2(1):35–46
Heylen D (2005) Challenges ahead head movements and other social acts in conversation. In:
Artificial intelligence and simulation of behaviour (AISB 2005), social presence cues for virtual
humanoids symposium, page 8, Hertfordshire
Huang L, Morency L-P, Gratch J (2010) Parasocial consensus sampling: combining multiple
perspectives to learn virtual human behavior. In: Proceedings of the 9th international conference
on autonomous agents and multiagent systems: volume 1-volume 1, Toronto, pp 1265–1272
Huang L, Morency L-P, Gratch J (2011) Virtual rapport 2.0. In: Intelligent virtual agents, Reykjavik,
pp 68–79
Ishi CT, Ishiguro H, Hagita N (2014) Analysis of relationship between head motion events and
speech in dialogue conversations. Speech Commun 57:233–243
Jakkam A, Busso C (2016) A multimodal analysis of synchrony during dyadic interaction using a
metric based on sequential pattern mining. In: IEEE international conference on acoustics,
speech and signal processing (ICASSP 2016), Shanghai, pp 6085–6089
Kipp M (2003) Gesture generation by imitation: from human behavior to computer character
animation. PhD thesis, Universität des Saarlandes, Saarbrücken
Kopp S, Krenn B, Marsella S, Marshall AN, Pelachaud C, Pirker H, Thórisson KR, Vilhjálmsson H
(2006) Towards a common framework for multimodal generation: the behavior markup lan-
guage. In: International conference on intelligent virtual agents (IVA 2006), Marina Del Rey, pp
205–217
Kuratate T, Munhall KG, Rubin PE, Vatikiotis-Bateson E, Yehia H (1999) Audio-visual synthesis of
talking faces from speech production correlates. In: Sixth European conference on speech
communication and technology, Eurospeech 1999, Budapest, pp 1279–1282
Lance B, Marsella SC (2007) Emotionally expressive head and body movement during gaze shifts.
In: Intelligent virtual agents, Paris, pp 72–85
Le BH, Ma X, Deng Z (2012) Live speech driven head-and-eye motion generators. IEEE Trans Vis
Comput Graph 18(11):1902–1914
Lee J, Marsella S (2006) Nonverbal behavior generator for embodied conversational agents. Intell
Virtual Agents 4133:243–255
Lee JJ, Marsella S (2009) Learning a model of speaker head nods using gesture corpora. In:
Proceedings of the 8th international conference on autonomous agents and multiagent
systems-volume 1, volume 1, Budapest, pp 289–296
Lester J, Stone B, Stelling G (1999) Lifelike pedagogical agents for mixed-initiative problem
solving in constructivist learning environments. User Model User-Adap Inter 9(1–2):1–44
Levine S, Krähenbühl P, Thrun S, Koltun V (2010) Gesture controllers. ACM Trans Graph 29
(4):1–124
Liu C, Ishi CT, Ishiguro H, Hagita N (2012) Generation of nodding, head tilting and eye gazing for
human-robot dialogue interaction. In: Human-Robot interaction (HRI), 2012 7th ACM/IEEE
international conference on, Boston, pp 285–292
Mariooryad S, Busso C (2012) Generating human-like behaviors using joint, speech-driven models
for conversational agents. IEEE Trans Audio, Speech Lang Process 20(8):2329–2340
Mariooryad S, Busso C (2013) Exploring cross-modality affective reactions for audiovisual emo-
tion recognition. IEEE Trans Affect Comput 4(2):183–196
Marsella S, Xu Y, Lhommet M, Feng A, Scherer S, Shapiro A (2013) Virtual character performance
from speech. In ACM SIGGRAPH/Eurographics symposium on computer animation (SCA
2013), Anaheim, pp 25–35
Marsi E, van Rooden F (2007) Expressing uncertainty with a talking head. In: Workshop on
multimodal output generation (MOG 2007), Aberdeen, pp 105–116
McClave EZ (2000) Linguistic functions of head movements in the context of speech. J Pragmat 32
(7):855–878
Moubayed SA, Beskow J, Granström B, House D (2010) Audio-visual prosody: perception,
detection, and synthesis of prominence. In: COST 2102 training school, pp 55–71
Munhall KG, Jones JA, Callan DE, Kuratate T, Vatikiotis-Bateson E (2004) Visual prosody and
speech intelligibility: head movement improves auditory speech perception. Psychol Sci 15
(2):133–137
Pelachaud C, Badler N, Steedman M (1996) Generating facial expressions for speech. Cognit Sci 20
(1):1–46
Poggi I, Pelachaud C, de Rosis F, Carofiglio V, de Carolis B (2005) Greta. a believable embodied
conversational agent. In: Stock O, Zancanaro M (eds) Multimodal intelligent information
presentation, Text, speech and language technology. Springer Netherlands, Dordrecht, pp 3–25
Rickel J, Johnson WL (1998) Task-oriented dialogs with animated agents in virtual reality. In:
Workshop on embodied conversational characters, Tahoe City, pp 39–46
Sadoughi N, Busso C (2015) Retrieving target gestures toward speech driven animation with
meaningful behaviors. In: International conference on Multimodal interaction (ICMI 2015),
Seattle, pp 115–122
Sadoughi N, Busso C (2016) Head motion generation with synthetic speech: a data driven
approach. In: Interspeech 2016, San Francisco, pp 52–56
Sadoughi N, Liu Y, Busso C (2014) Speech-driven animation constrained by appropriate discourse
functions. In: International conference on multimodal interaction (ICMI 2014), Istanbul, pp
148–155
Sadoughi N, Liu Y, Busso C (2015) MSP-AVATAR corpus: motion capture recordings to study the
role of discourse functions in the design of intelligent virtual agents. In: 1st international
workshop on understanding human activities through 3D sensors (UHA3DS 2015), Ljubljana
Sargin ME, Yemez Y, Erzin E, Tekalp AM (2008) Analysis of head gesture and prosody patterns for
prosody-driven head-gesture animation. IEEE Trans Pattern Anal Mach Intell 30(8):1330–1345
Silverman K, Beckman M, Pitrelli J, Ostendorf M, Wightman C, Price P, Pierrehumbert J,
Hirschberg J (1992) ToBI: a standard for labelling english prosody. In: 2th international
conference on spoken language processing (ICSLP 1992), Banff, pp 867–870
Smid K, Pandzic I, Radman V (2004) Autonomous speaker agent. In: IEEE 17th international
conference on computer animation and social agents (CASA 2004), Geneva, pp 259–266
Stone M, DeCarlo D, Oh I, Rodriguez C, Stere A, Lees A, Bregler C (2004) Speaking with hands:
creating animated conversational characters from recordings of human performance. ACM
Trans Graph (TOG) 23(3):506–513
Taylor GW, Hinton GE (2009) Factored conditional restricted Boltzmann machines for modeling
motion style. In: Proceedings of the 26th annual international conference on machine learning,
Montreal, pp 1025–1032
Taylor GW, Hinton GE, Roweis ST (2006) Modeling human motion using binary latent variables.
Adv Neural Inf Process Syst 1345–1352
Welbergen H, Ding Y, Sattler K, Pelachaud C, Kopp S (2015) Real-time visual prosody for
interactive virtual agents. In: Intelligent virtual agents, Delft, pp 139–151
Xiao B, Georgiou P, Baucom B, Narayanan S (2015) Modeling head motion entrainment for
prediction of couples’ behavioral characteristics. In: Affective computing and intelligent inter-
action (ACII), 2015 international conference on, Xi’an, pp 91–97
Youssef AB, Shimodaira H, Braude DA (2013) Head motion analysis and synthesis over different
tasks. Intell Virtual Agents 8108:285–294
Hand Gesture Synthesis for Conversational
Characters
Michael Neff
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gesture Generation Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gesture Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gesture Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Additional Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Abstract
This chapter focuses on the generation of animated gesticulations, co-verbal
gestures that are designed to accompany speech. It begins with a survey of
research on human gesture, discussing the various forms of gesture, their struc-
ture, and timing requirements relative to speech. The two main problems for
synthesizing gesture animation are determining what gestures a character should
perform (the specification problem) and then generating appropriate motion (the
animation problem). The specification problem has used a range of input, includ-
ing speech prosody, spoken text, and a communicative intent. Both rule-based
and statistical approaches are employed to determine gestures. Animation has
also used a range of procedural, physics-based, and data-driven approaches in
order to solve a significant set of expressive and coordination requirements. Fluid
gesture animation must also reflect the context and include listener behavior and
floor management. This chapter concludes with a discussion of future challenges.
M. Neff (*)
Department of Computer Science & Program for Cinema and Digital Media, University of
California – Davis, Davis, CA, USA
e-mail: mpneff@ucdavis.edu

DOI 10.1007/978-3-319-30808-1_5-1
2 M. Neff
Keywords
Gesture • Character animation • Nonverbal communication • Virtual agents •
Embodied conversational agents
Introduction
Do gestures communicate? Yes, they do. This has been the conclusion of several
meta-studies on the impact of gesture (Goldin-Meadow 2005; Hostetter 2011;
Kendon 1994). It is also one of the distinguishing features of gestures in animation.
While all movement communicates to some degree, gestures often play a role that is
explicitly communicative. Another distinguishing feature for the gestures that we are
most often interested in is that they are co-verbal. That is, they occur with speech and
they are inextricably linked to that speech in both content and timing. McNeill
argues that gestures and language are not separate, but gestures are part of language
(McNeill 2005).
There are different forms of movement that can broadly be called “gesture.”
Building on the categories of Kendon (1988), McNeill defined “Kendon’s Contin-
uum” (McNeill 1992, 2005) to capture the range of gesture types people employ:
• Gesticulation: gesture that conveys a meaning related to the accompanying

speech.
• Speechlike gestures: gestures that take the place of a word(s) in a sentence.
• Emblems: conventionalized signs, like a thumbs-up.
• Pantomime: gestures with a story and are produced without speech.
• Sign language: signs are lexical words.
As you move along the continuum, the degree to which speech is obligatory
decreases, and the degree to which gestures themselves have the properties of a
language increases. This chapter will focus on gesticulations, which are gestures that
co-occur with speech as they are most relevant to conversational characters. Syn-
thesis of the whole spectrum, however, presents worthwhile animation problems.
Emblems and pantomimes are useful in situations where speech may not be possible.
Sign languages are the native language of many members of the deaf community,
and sign synthesis can increase their access to computational sources. The problems
of gesticulations are unique, however, since they are co-present with speech and do
not have linguistic structure on their own.
Kendon introduced a three-level hierarchy to describe the structure of gestures
(Kendon 1972). The largest structure is the gesture unit. Gesture units start in a
retraction or rest pose, continue with a series of gestures, and then return to a rest
pose, potentially different from the initial rest pose. A gesture phrase encapsulates
an individual gesture in this sequence. Each gesture phrase can in turn be broken
down into a sequence of gesture phases. A preparation is a motion that takes the
hands to the required position and orientation for the start of the gesture stroke. A
Hand Gesture Synthesis for Conversational Characters 3
prestroke hold is a period of time in which the hands are held in this configuration.
The stroke is the main meaning carrying movement of the gesture and has the most
focused energy. It may be followed by a poststroke hold in which the hands are held
at the end position. The final phase is a retraction that returns the hands to a rest
pose. All phases are optional except the stroke. There are some gestures in which the
stroke does not involve any movement (e.g., a raised index finger). These are
variously called an independent hold (Kita et al. 1998) or a stroke hold (McNeill
2005). The pre- and poststroke holds were proposed by Kita (1990) and act to
synchronize the gesture with speech. The prestroke hold delays the gesture stroke
until the corresponding speech begins, and the poststroke hold occurs while the
corresponding speech is completing. Much like they allow mental processing in
humans, they can be used in synthesis systems to allow time for planning or other
processing to take place.
The existence of gesture units is important for animation systems as it indicates a
potential need to avoid generating a sequence of singleton gestures that return to a
rest pose after each gesture. While this would offer the simplest synthesis solution,
people are quite sensitive to the structure of gestural communication. A study (Kipp
et al. 2007) showed that people found a character that used multiple phrase gesture
units more natural, friendly, and trustworthy than a character that performed single-
ton gestures, which was viewed as more nervous. These significant differences in
appraisal occurred despite only 1 of 25 subjects being able to actually identify the
difference between the multiphrase g-unit clips and single phrase g-unit clips. This
illustrates what appears to be a common occurrence in our gesture research: people
will react to differences in gesture performance without being consciously aware of
what those differences are.
Gestures are synchronized in time with their co-expressive speech. About 90 %
of the time, the gesture occurs slightly before the co-expressive speech (Nobe 2000)
and rarely occurs after (Kendon 1972). Research on animated characters does indicate a
preference for this slightly earlier timing of gesture, but also suggests that people may
not be particularly sensitive to errors in timing, at least within a +/ .6 second range
(Wang and Neff 2013).
A number of categorizations of gesture have been proposed. One of the best
known is from McNeill and Levy (McNeill 1992; McNeill and Levy 1982) and
contains the classes iconics, metaphorics, deictics, and beats. Iconic gestures create
images of concrete objects or actions, such as illustrating the size of a box. Meta-
phorics create images of the abstract. For instance, a metaphoric gesture could make
a cup shape with the hand, but refer to holding an idea rather than an actual object.
Metaphoric gestures are also used to locate ideas spatially, for instance, putting
positive things on the left and negative to the right and then using this space to
categorize future entities in the conversation. Deictics locate objects and entities in
space, as with pointing, creating a reference and context for the conversation. They
are often performed with a hand that is closed except for an extended index finger,
but can be performed with a wide range of body parts. Deixis can be abstract or
concrete. Concrete deixis points to an existing reference (e.g., an object or person) in
space, whereas abstract deixis creates a reference point in space for an idea or
4 M. Neff
concept. Beats are small back-and-forth or up-and-down movements of the hand,

performed in rhythm to the speech. They serve to emphasize important sections of
the speech.
In later work, McNeill (2005) argued that it is inappropriate to think of gesture in
terms of categories, but the categories should instead be considered dimensions. This
reflects the fact that any individual gesture may contain several of these properties
(e.g., deixis and iconicity). He suggests additional dimensions of temporal highlight-
ing (the function of beats) and social interactivity, which helps to manage turn taking
and the flow of conversation.
State of the Art
Generation of conversational characters has achieved substantial progress, but the

bar for success is extremely high. People are keen observers of human motion and
will make judgments based on subtle details. By way of analogy, people will make
judgments between good and bad actors, and actors being good in a particular role,
but not another – and actors are human, with all the capacity for naturalness and
expressivity that comes with that. The bar for conversational characters is that of a
good actor, effectively performing a particular role. The field remains a long way
from being able to do this automatically, for a range of different characters and over
prolonged interactions with multiple subjects.
Gesture Generation Tasks
Gesture Specification
When generating virtual conversational characters, one of the primary challenges is

determining what gestures a character should perform. Different approaches have
trade-offs in terms of the type of input information they require, the amount of
processing time needed to determine a gesture, and the quality of the gesture
selection, both on grounds of accurately reflecting a particular character personality
and being appropriate for the co-expressed utterance.
One approach is to generate gestures based on prosody variations in the spoken
audio signal. Prosody includes changes in volume and pitch. Such approaches have
been applied for head nods and movement (Morency et al. 2008), as well as gesture
generation (Levine et al. 2009, 2010). A main advantage of the approach is that
good-quality audio can be highly expressive, and using it as an input for gesture
specification allows the gestures to match the expressive style of the audio. Points of
emphasis in the audio appear to be good landmarks for placing gesture, and their use
will provide uniform emphasis across the channels. Prosody-based approaches have
been used to generate gesture in real time as a user speaks (Levine et al. 2009, 2010).
The drawback of only using prosody is that it does not capture semantics, so the
gestures will likely not match the meaning of the audio and certainly not supplement
the underlying meaning that is being conveyed in the utterance with information not
present in the audio. This concern can be at least partially addressed by also parsing
the spoken text (Marsella et al. 2013). It is believed that in human communication,
the brain is co-planning the gesture and the utterance (McNeill 2005), so approaches
that do not use future information about the planned utterance may be unlikely to
match the sophistication of human gesture-speech coordination.
Another approach generates gesture based on the text of the dialogue that is to be
spoken. A chief benefit of these techniques is that text captures much of the
information being conveyed, so these techniques can generate gestures that aid the
semantics of the utterance. Text can also be analyzed for emotional content and
rhetorical style, providing a rich basis for gesture generation. Rule-based approaches
(Cassell et al. 2001; Lee and Marsella 2006; Lhommet and Marsella 2013; Marsella
et al. 2013) can determine both the gesture locations and the type of gestures to be
performed. Advantages of these techniques are that they can handle any text covered
by their knowledge bases and are extensible in flexible and straightforward ways.
Disadvantages include that some amount of manual work is normally required to
create the rules and it is difficult to know how to author the rules to create a particular
character, so behavior tends to be generic. Other work uses statistical approaches to
predict the gestures that a particular person would employ (Bergmann et al. 2010;
Kipp 2005; Neff et al. 2008). These techniques support the creation of individualized
characters, which are essential for many applications, such as anything involving
storytelling. Individualized behavior may also outperform averaged behavior
(Bergmann et al. 2010), as would be contained in generic rules. These approaches,
however, are largely limited to reproducing characters like the subjects modeled and
creating arbitrary characters remains an open challenge. Recent work has begun
applying deep learning to the mapping from text and prosody to gesture (Chiu
et al. 2015). This is a potentially powerful approach, but requires a large quantity
of data, and ways to produce specific characters must be developed. While the divide
between prosody-driven and rule-based approaches is useful for understanding
techniques, current approaches are increasingly relying on a combination of text
and prosody information (e.g., (Lhommet and Marsella 2013; Marsella et al. 2013)).
Techniques based on generating gesture from text are limited to ideas expressed
in the text. The information we convey through gesture is sometimes redundant with
speech, although expressed in a different form, but often expresses information that
is different to that in speech (McNeill 2005). For example, I might say “I saw a
[monster.],” with the square brackets indicating the location of a gesture that holds
my hand above my head, with my fingers bent 90 at the first knuckle and then held
straight. The gesture indicates the height of the monster, information completely
lacking from the verbal utterance. Evidence suggests that gestures are most effective
when they are nonredundant (Goldin-Meadow 2006; Hostetter 2011; Singer and
Goldin-Meadow 2005). This implies the need to base gesture generation on a deeper
notion of a “communicative intent,” which may not solely be contained in the text
and describes the fully message to be delivered.
The SAIBA (situation, agent, intention, behavior, animation) framework repre-
sents a step toward establishing a computational architecture to tackle the
6 M. Neff
fundamental multimodal communication problem of moving from a communicative

intent to output across the various agent channels of gesture, text, prosody, facial
expressions, and posture (SAIBA. Working group website 2012). The approach
defines stages in production and markup languages to connect them. The first
stage is planning the communicative intent. This is communicated using the Func-
tion Markup Language (Heylen et al. 2008) to the behavior planner, which decides
how to achieve the desired functions using the agent modalities available. The final
behavior is then sent to a behavior realizer for generation using the Behavior Markup
Language (Kopp et al. 2006; Vilhjalmsson et al. 2007). Such approaches echo, at
least at the broad conceptual level, theories of communication like McNeill’s growth
point hypothesis that argue gesture and language emerge in a shared process from a
communicative intent (McNeill 2005). Recent work has sought to develop cognitive
(Kopp et al. 2013) and combined cognitive and linguistic models (Bergmann
et al. 2013) to explore the distribution of communicative content across output
modalities.
Gesture Animation
Generation of high-quality gesture animation must satisfy a rich set of requirements:
• Match the gesture timing to that of the speech.

• Connect individual gestures into fluent gesture units.
• Adjust the gesture to the character’s context (e.g., to point to a person or object in
the scene).
• Generate appropriate gesture forms for the utterance (e.g., show the shape of an
object, mime an action being performed, point).
• Vary the gesture based on the personality of the character.
• Vary the gesture to reflect the character’s current mood and tone of the speech.
While a wide set of techniques have been used for gesture animation, the need for
precise agent control, especially in interactive systems, has often favored the use of
kinematic procedural techniques (e.g., (Chi et al. 2000; Hartmann et al. 2006; Kopp
and Wachsmuth 2004)). For example, Kopp and Wachsmuth kopp04 present a
system that uses curves derived from neurophysiological research to drive the
trajectory of gesturing arm motions. Procedural techniques allow full control of
the motion, making it easy to adjust the gesture to the requirements of the speech,
both for matching spatial and timing demands.
While gesture is less constrained by physics than motions like tumbling, physical
simulation has still been used for gesture animation and can add important nuance to
the motion (Neff and Fiume 2002, 2005; Neff et al. 2008; Van Welbergen
et al. 2010). These approaches generally include balance control and a basic approx-
imation to the muscle, such as a proportional derivative controller. The balance
control will add full-body movement to compensate for arm movements, and the
controllers can add subtle oscillations and arm swings. These effects require proper
tuning.
Motion capture data has seen increasing use in an attempt to improve the realism
of character motion. These techniques often employ versions of motion graphs
(Arikan and Forsyth 2002; Kovar et al. 2002; Lee et al. 2002) which concatenate
segments of motion to create a sequence, such as in Fernández-Baena et al. (2014)
and Stone et al. (2004). The motion capture data can provide very high-quality
motion, but control is more limited, so it can be a challenge to adapt the motion to
novel speech or generated different characters. Gesture relies heavily on hand shape,
and it can be a challenge to capture good-quality hand motion while simultaneously
capturing body motion. Some techniques seek to synthesize acceptable hand motion
using the body motion alone (Jörg et al. 2012). For a fuller discussion of the issues
around hand animation, please refer to (Wheatland et al. 2015).
As part of the SAIBA effort, several research groups have developed “behavior
realizers,” animation engines capable of realizing commands in the Behavior
Markup Language (Vilhjalmsson et al. 2007) that is supplied by a higher level in
an agent architecture. These systems emphasize control and use a combination of
procedural data and motion clips (e.g., (Heloir and Kipp 2009; Kallmann and
Marsella 2005; Shapiro 2011; Thiebaux et al. 2008; Van Welbergen et al. 2010)).
The SmartBody system, for example, uses a layering approach based on a hierarchy
of controllers for different tasks (e.g., idle motion, locomotion, reach, breathing).
These controllers may control different or overlapping parts of the body, which
creates a coordination challenge. They can be combined or one controller may
override another (Shapiro 2011).
Often gesture specification systems will indicate a particular gesture form that is
required, e.g., a conduit gesture in which the hand is cupped and moves forward.
Systems often employ a dictionary of gesture forms that can be used in syntheses.
These gestures have been encoded using motion capture clips, hand animation, or
numerical spatial specifications. Some techniques (Kopp et al. 2004) have sought to
generate the correct forms automatically, for example, based on a description of the
image trying to be created by the gesture.
Gesture animation is normally deployed in scenarios where it is desirable for the
characters to portray clear personalities and show variations in emotion and mood.
For these reasons, controlling expressive variation of the motion has been an
important focus. A set of challenges must be solved. These include determining
how to parameterize a motion to give expressive control, understanding what aspects
of motion must be varied to generate a desired impact, ensuring consistency over
time, determining how to expose appropriate control structures to the user or
character control system, and, finally, synthesizing the motion to contain the desired
properties. Chi et al. (2000) use the Effort and Shape components of Laban Move-
ment Analysis to provide an expressive parameterization of motion. Changing any of
the four effort qualities (Weight, Space, Time, and Flow) or the Shape Qualities
(Rising-Sinking, Spreading-Enclosing, Advancing-Retreating) will vary the timing
and path of the gesture, along with the engagement of the torso. Hartmann
8 M. Neff
et al. (Hartmann 2005) use tension, continuity, and bias splines (Kochanek and
Bartels 1984) to control arm trajectories and provide expressive control through
parameters for activation, spatial and temporal extent, and fluidity and repetition.
Neff and Fiume (2005) develop an extensible set of movement properties that can be
varied and a system that allows users to write character sketches that reflect a
particular character’s movement tendencies and then layer additional edits on top.
While gestures are often largely thought of as movements of the arms and hands
and often represented this way in computational systems, they can indeed use the
whole body. A character can nod its head, gesture with its toe, etc. More importantly,
while arms are the dominant appendages for a motion, engaging the entire body can
lead to more clear and effective animation. Lamb called this engagement of the
whole body during gesturing Posture-Gesture Merger and argued that it led to a more
fluid and attractive motion (Lamb 1965).
Additional Considerations
Conversations are interactions between people and this must be reflected in the
animation. Both the speaker(s) and listener(s) have roles to play. Visual attention
must be managed through appropriate gaze behavior to indicate who is paying
attention and how actively, along with indicating who is thinking or distracted.
Attentive listeners will provide back channel cues, like head nods, to indicate that
they are listening and understanding. These must be appropriately timed with the
speaker’s dialogue. Holding the floor is also actively managed. Speakers may decide
to yield their turn to another. Listeners may interrupt, and the speaker may yield in
response or refuse to do so. Floor management relies on both vocal and gestural cues.
Proxemics are also highly communicative to an audience and must be managed
appropriately. This creates additional animation challenges in terms of small-scale
locomotion in order to fluidly manage character placement.
Gestural behavior must adapt to the context. Gestures will be adjusted based on
the number of people in the conversation and their physical locations relative to one
another. As characters interact, they may also begin to mirror each other’s behavior
and postures. Gestures are also often used to refer to items in the environment and
hence must be adapted based on the character’s location. Finally, characters will
engage in conversations while also simultaneously performing other activities, such
as walking, jogging, or cleaning the house. The gesture behavior must be adapted to
the constraints of this other behavior, for example, gestures performed while jogging
tend to be done with more bent arms and are less frequent than standing gestures
(Wang et al. 2016).
Future Directions
While significant progress has been made, the bar for conversational gesture anima-
tion is very high. We are a long way from being able to easily create synthetic
characters that match the expressive quality, range, and realism of a skilled actor, and
applications that rely on synthetic characters are impoverished by this gap. Some of
the key issues to address include:
Characters with large gesture repertoires: It currently takes a great deal of
work to build a movement set for a character, generally involving recording,
cleaning, and retargeting motion capture or hand animating movements. This places
a practical limitation on the number of gestures that they can perform. Methods that
allow large sets of gestures to be rapidly generated are needed. A particular challenge
is being able to synthesize novel gestures on the fly to react to the character’s current
context.
Motion quality: While motion quality has improved, it remains well short of
photo-realism, particularly for interactive characters. Hand motion remains a partic-
ular challenge, as is appropriate full-body engagement. Most systems focus on
standing characters, whereas people engage in a wide range of activities while
simultaneously gesturing. A significant challenge is correctly orchestrating a perfor-
mance across the various movement modalities (breath, arm movements, body
movements, facial expressions, etc.), especially when the motion diverges from
playback of a recording or hand-animated sequence.
Planning from communicative intent: Systems that can represent an arbitrary
communicative intent and can distribute it across various communication modes,
and do so in different ways for different speakers, remain a long-term goal. This will
likely require both improved computational models and a more thorough under-
standing of how humans formulate communication.
Customization for characters and mood: While people tend to have their own,
unique gesturing style, it is a challenge to imbue synthetic characters with this
expressive range without an enormous amount of manual labor. It is also a challenge
to accurate reflect a character’s current mood; anger, sadness, irritation,
excitement, etc.
Authoring controls: If a user wishes to create a particular character with a given
role, personality, etc., there must be tools to allow this to be authored. Substantial
work is required to allow authors to go from an imagined character to an effective
realization.
References
Arikan O, Forsyth DA (2002) Interactive motion generation from examples. ACM Trans Graph 21
(3):483–490
Bergmann K, Kopp S, Eyssel F (2010) Individualized gesturing outperforms average gestu-
ring–evaluating gesture production in virtual humans. In: International conference on intelli-
gent virtual agents. Springer, Berlin/Heidelberg, pp 104–117
Bergmann K, Kahl S, Kopp.S (2013) Modeling the semantic coordination of speech and gesture
under cognitive and linguistic constraints. In: Intelligent virtual agents. Springer, Berlin,
Heidelberg, pp 203–216
Cassell J, Vilhjálmsson H, Bickmore T (2001) BEAT: the behavior expression animation toolkit. In:
Proceedings of SIGGRAPH 2001. ACM, New York, NY, pp 477–486
Chi DM, Costa M, Zhao L, Badler NI (2000) The EMOTE model for effort and shape. In:
Proceedings of SIGGRAPH 2000. ACM, New York, NY, pp 173–182
10 M. Neff
Chiu C-C,Morency L-P, Marsella S (2015) Predicting co-verbal gestures: a deep and temporal
modeling approach. In: International conference on intelligent virtual agents. Springer, Cham,
pp 152–166.
Fernández-Baena A, Montaño R, Antonijoan M, Roversi A, Miralles D, Alas F (2014) Gesture
synthesis adapted to speech emphasis. Speech Comm 57:331–350
Goldin-Meadow S (2005) Hearing gesture: how our hands help us think. Harvard University Press,
Massachusetts
Goldin-Meadow S (2006) Talking and thinking with our hands. Curr Dir Psychol Sci 15(1):34–39
Hartmann B, Mancini M, Pelachaud C (2006) Implementing expressive gesture synthesis for
embodied conversational agents. In Proc. Gesture Workshop 2005, vol 3881 of LNAI. Springer,
Berlin\Heidelberg, pp 45–55
Heloir A, Kipp M (2009) EMBR–A Realtime Animation Engine for Interactive Embodied Agents.
In: Intelligent virtual agents 09. Springer, Berlin, Heidelberg, pp 393–404
Heylen D, Kopp S, Marsella SC, Pelachaud C, Vilhjálmsson H (2008) The next step towards a
function markup language. In: International workshop on intelligent virtual agents. Springer,
Berlin, Heidelberg, pp 270–280
Hostetter AB (2011) When do gestures communicate? A meta-analysis. Psychol Bull 137(2):297
Jörg S, Hodgins J, Safonova A (2012) Data-driven finger motion synthesis for gesturing characters.
ACM Trans Graph 31(6):189
Kallmann M, Marsella S (2005) Hierarchical motion controllers for real-time autonomous virtual
humans. In: Proceedings of the 5th International working conference on intelligent virtual
agents (IVA’05), pp 243–265, Kos, Greece, 12–14 September 2005
Kendon A (1972) Some relationships between body motion and speech. Stud dyadic commun 7
(177):90
Kendon A (1988) How gestures can become like words. Cross-cult perspect nonverbal commun
1:131–141
Kendon A (1994) Do gestures communicate? A review. Res lang soc interact 27(3):175–200
Kipp M (2005) Gesture generation by imitation: from human behavior to computer character
animation. Universal-Publishers, Boca Raton, Fl, USA
Kipp M, Neff M, Kipp K, Albrecht I (2007) Towards natural gesture synthesis: evaluating gesture
units in a data-driven approach to gesture synthesis. In Proceedings of intelligent virtual agents
(IVA07), vol 4722 of LNAI, Association for Computational Linguistics, Berlin, Heidelberg, pp
15–28
Kita S (1990) The temporal relationship between gesture and speech: a study of Japanese-English
bilinguals. MS Dep Psychol Univ Chic 90:91–94
Kita S, Van Gijn I, Van Der Hulst H (1998) Movement phase in signs and co-speech gestures, and
their transcriptions by human coders. In: Proceedings of the International Gesture Workshop on
Gesture and Sign Language in Human-Computer Interaction. Springer-Verlag, Berlin, Heidel-
berg, pp 23–35
Kochanek DHU, Bartels RH (1984) Interpolating splines with local tension, continuity, and bias
control. Comput Graph 18(3):33–41
Kopp S, Wachsmuth I (2004) Synthesizing multimodal utterances for conversational agents.
Comput Anim Virtual Worlds 15:39–52
Kopp S, Tepper P, Cassell J (2004) Towards integrated microplanning of language and iconic
gesture for multimodal output. In: Proceedings of the 6th international conference on multi-
modal interfaces. ACM, New York, NY, pp 97–104
Kopp S, Krenn B, Marsella S, Marshall AN, Pelachaud C, Pirker H, Thórisson KR, Vilhjálmsson H
(2006) Towards a common framework for multimodal generation: the behavior markup lan-
guage. In: International workshop on intelligent virtual agents. Springer, Berlin, Heidelberg, pp
205–217
Kopp S, Bergmann K, Kahl S (2013) A spreading-activation model of the semantic coordination of
speech and gesture. In: Proceedings of the 35th annual conference of the cognitive science
society (CogSci 2013). Cognitive Science Society, Austin (in press, 2013)
Kovar L, Gleicher M, Pighin F (2002) Motion graphs. ACM Trans Graph 21(3):473–482
Lamb W (1965) Posture and gesture: an introduction to the study of physical behavior. Duckworth,
London
Lee J, Marsella S (2006) Nonverbal behavior generator for embodied conversational agents. In:
Intelligent virtual agents. Springer, Berlin, Heidelberg, pp 243–255
Lee J, Chai J, Reitsma PSA, Hodgins JK, Pollard NS (2002) Interactive control of avatars animated
with human motion data. ACM Trans Graph 21(3):491–500
Levine S, Theobalt C, Koltun V (2009) Real-time prosody-driven synthesis of body language.
ACM Trans Graph 28(5):1–10
Levine S, Krahenbuhl P, Thrun S, Koltun V (2010) Gesture controllers. ACM Trans Graph 29
(4):1–11
Lhommet M, Marsella SC (2013) Gesture with meaning. In: Intelligent Virtual Agents. Springer,
Berlin, Heidelberg, pp 303–312
Marsella S, Xu Y, Lhommet M, Feng A, Scherer S, Shapiro A (2013) Virtual character performance
from speech. In: Proceedings of the 12th ACM SIGGRAPH/Eurographics Symposium on
Computer Animation, ACM, New York, NY, pp 25–35
McNeill D (1992) Hand and mind: what gestures reveal about thought. University of Chicago Press,
Chicago
McNeill D (2005) Gesture and thought. University of Chicago Press, Chicago
McNeill D, Levy E (1982) Conceptual representations in language activity and gesture. In: Jarvella
RJ, Klein W (eds) Speech, place, and action. Wiley, Chichester, pp 271–295
Morency L-P, de Kok I, Gratch J (2008) Predicting listener backchannels: a probabilistic multi-
modal approach. In: International workshop on intelligent virtual agents. Springer, Berlin/
Neff M, Fiume E (2002) Modeling tension and relaxation for computer animation. In Proc. ACM
SIGGRAPH Symposium on Computer Animation 2002, ACM, New York, NY, pp 81–88
Neff M, Fiume E (2005) AER: aesthetic exploration and refinement for expressive character
animation. In: Proceeding of ACM SIGGRAPH / Eurographics Symposium on Computer
Animation 2005, ACM, New York, NY, pp 161–170
Neff M, Kipp M, Albrecht I, Seidel H-P (2008) Gesture modeling and animation based on a
probabilistic re-creation of speaker style. ACM Trans Graph 27(1):5:1–5:24
Nobe S (2000) Where do most spontaneous representational gestures actually occur with respect to
speech. Lang gesture 2:186
SAIBA. Working group website, 2012. http://wiki.mindmakers.org/projects:saiba:main
Shapiro A (2011) Building a character animation system. In: International conference on motion in
games, Springer, Berlin\Heidelberg, pp 98–109
Singer MA, Goldin-Meadow S (2005) Children learn when their teacher’s gestures and speech
differ. Psychol Sci 16(2):85–89
Stone M, DeCarlo D, Oh I, Rodriguez C, Stere A, Lees A, Bregler C (2004) Speaking with hands:
creating animated conversational characters from recordings of human performance. ACM
Trans Graph 23(3):506–513
Thiebaux M, Marshall A, Marsella S, Kallman M (2008) Smartbody: behavior realization for
embodied conversational agents. In: Proceedings of 7th International Conference on autono-
mous agents and multiagent systems (AAMAS 2008), International Foundation for Autono-
mous Agents and Multiagent Systems Richland, SC, pp 151–158
Van Welbergen H, Reidsma D, Ruttkay Z, Zwiers J (2010) Elckerlyc-A BML realizer for contin-
uous, multimodal interaction with a virtual human. Journal on Multimodal User Interfaces 4
(2):97–118
Vilhjalmsson H, Cantelmo N, Cassell J, Chafai NE, Kipp M, Kopp S, Mancini M, Marsella S,
Marshall A, Pelachaud C et al (2007) The behavior markup language: recent developments and
challenges. In: Intelligent virtual agents. Springer, Berlin/New York, pp 99–111
Wang Y, Neff M (2013) The influence of prosody on the requirements for gesture-text alignment. In:
Intelligent virtual agents. Springer, Berlin/New York, pp 180–188
12 M. Neff
Wang Y, Ruhland K, Neff M, O’Sullivan C (2016) Walk the talk: coordinating gesture with
locomotion for conversational characters. Comput Anim Virtual Worlds 27(3–4):369–377
Wheatland N, Wang Y, Song H, Neff M, Zordan V, Jörg S (2015) State of the art in hand and finger
modeling and animation. Comput Graphics Forum. The Eurographs Association and John
Wiley & Sons, Ltd., Chichester, 34(2):735–760
Depth Sensor-Based Facial and Body
Animation Control
Yijun Shen, Jingtian Zhang, Longzhi Yang, and Hubert P. H. Shum
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Extracting Facial and Body Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Facial Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Body Posture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Human Environment Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Dealing with Noisy Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Face Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Posture Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Prior Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Depth Camera-Based Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Abstract
Depth sensors have become one of the most popular means of generating human
facial and posture information in the past decade. By coupling a depth camera and
computer vision based recognition algorithms, these sensors can detect human
facial and body features in real time. Such a breakthrough has fused many new
research directions in animation creation and control, which also has opened up
new challenges. In this chapter, we explain how depth sensors obtain human
facial and body information. We then discuss on the main challenge on depth
sensor-based systems, which is the inaccuracy of the obtained data, and explain
how the problem is tackled. Finally, we point out the emerging applications in the
Y. Shen (*) • J. Zhang (*) • L. Yang (*) • H.P.H. Shum (*)

Northumbria University, Newcastle upon Tyne, UK
e-mail: yi.shen@northumbria.ac.uk; jingtian.zhang@northumbria.ac.uk;
longzhi.yang@northumbria.ac.uk; hubert.shum@northumbria.ac.uk

DOI 10.1007/978-3-319-30808-1_7-1
2 Y. Shen et al.
field, in which human facial and body feature modeling and understanding is a
key research problem.
Keywords
Depth sensors • Kinect • Facial features • Body postures • Reconstruction •
Machine learning • Computer animation
Introduction
In the past decade, depth sensors have become a very popular mean of generating
character animation. In particular, since these sensors can obtain human facial and
body information in real time, it is used heavily in real-time graphics and games.
While it is expensive to use human motion to interact with computer applications
using traditional motion capture system, depth sensors provide an affordable alter-
native. Due to the low cost and high robustness of depth sensors, it can be applied in
a wide application domain with easy setup. Apart from popular applications such as
motion-based gaming, depth sensors are also applied in emerging applications such
as virtual reality, sport training, serious games, smart environments, etc.
In order to work with depth sensors, it is important to understand their working
principle, as well as their strength and weakness. In this chapter, we provide compre-
hensive information on how depth sensors track human facial and body features using
computer vision and pattern recognition based techniques, and identify their strength
in computational cost and robustness. Then, we focus on the major weakness of depth
sensors, that is, the low accuracy that happens during occlusion, and explain possible
solutions to improve recognition quality in detail. In particular, we discuss in depth on
machine learning-based reconstruction method that utilize prior knowledge to correct
corrupted data obtained by the sensors. Finally, we give some examples on depth
sensors-based application, especially in the field of animation creation, to show how
these sensors can improve existing methods in human-computer interaction.
In the rest of this chapter, we review the state of the art in section “State of the
Art.” We explain in more details about how depth sensors obtain and process human
facial and body movement information in section “Extracting Facial and Body
Information.” We then discuss on the main challenge on depth sensor-based systems,
which is about the relatively low accuracy of the data obtained, and explain how this
challenge can be tackled in section “Dealing with Noisy Data.” We finally point out
various emerging applications developed with depth sensors in section “Depth
Camera-Based Applications” and conclude this chapter in section “Conclusion.”
State of the Art
Typical depth sensors utilize a depth camera to obtain a depth image. The main
advantage of the depth camera over traditional color cameras is that instead of
obtaining color information, it estimates the distance of the objects seen by the
Depth Sensor-Based Facial and Body Animation Control 3
camera using an infrared sensor. The images taken from a depth camera are called
depth images. In these images, the pixels represent distance instead of color.
The nature of depth images provides a huge advantage in automatic recognition
using computer vision and machine learning algorithm. With traditional color
images, recognizing objects requires segmenting them based on color information.
This is challenging under situations in which the background has a similar color as
the foreground objects (Fernandez-Sanchez et al. 2013). Moreover, color values are
easily affected by lighting conditions, which reduces the robustness of object
recognition (Kakumanu et al. 2007). On the contrary, with depth images, since the
pixel value represents distance, automatic object segmentation becomes independent
of the color of the object. As long as the object is geometrically separated from the
background, accurate segmentation can be performed. Followed by such an
improved segmentation process is an improved object recognition system, which
identifies the nature of the objects using accurate geometric features. Such an
advancement in accuracy and robustness allows depth sensors to become a popular
commercial product that leads to many new applications.
The Microsoft Kinect (https://developer.microsoft.com/en-us/windows/kinect),
which utilizes both color and depth cameras, is one of the most popular depth
sensors. Due to the uses of both color and depth cameras, Kinect can create a 3D
point cloud based on the obtained images. Figure 1 shows the images obtained by the
two cameras, as well as two views of the corresponding point cloud. Kinect gaming
usually involves players controlling the gameplay with body movement. Virtual
characters in the game are then synthesized on the fly based on the movement
information obtained. Such a kind of application involves different domains of
research. First, computer vision and machine learning techniques are applied to
analyze the depth images obtained by the depth sensor. This typically involves
recognizing different human features, such as the human body parts (Shotton
et al. 2012). Then, human computer interaction researches are applied to translate
to control signals from the body movement into gameplay controls. Computer
graphics and animation algorithms are used to create real-time rendering, which
usually includes character animation synthesized from the movement of the player.
In some situations, virtual reality (Kyan et al. 2015) or augmented reality (Vera
et al. 2011) research is adapted to enhance the immersiveness of the game.
However, depth sensors are not without their weaknesses. Comparing to tradi-
tional capturing devices such as accelerometers, the accuracy of depth sensors is
Fig. 1 (From left to right) The color and depth images obtained by a Microsoft Kinect, as well as
two views of 3D point cloud rendered by combing the color and depth information
4 Y. Shen et al.
considerably lower. This is mainly because these sensors usually consist of a single
depth camera. When occlusions occur, the sensors cannot obtain information from
the shielded area. This results in a significant drop in recognition accuracy. While it
is possible to utilize multiple depth cameras to obtain better results, one has to deal
with the cross-talk, interference of infrared signals, among multiple cameras (Alex
Butler et al. 2012). It also deficits the advantage of using depth sensors in terms of
easy setup and efficient capture. Therefore, it is preferable to enhance the sensor
accuracy using software algorithms, instead of introducing more hardware.
To enhance the quality of the obtained data, machine learning approaches using
prior knowledge of the face and body features have shown great success (Shum
et al. 2013). The main idea is to apply prior knowledge onto the tracked data and
correct the less reliable parts or introduce more details onto the data. Such knowl-
edge can either be defined manually or learned from examples. The key here is to
represent the prior knowledge in a way that is efficient and effective to be used
during run-time.
Extracting Facial and Body Information
There is a large body of research in obtaining facial and body information from depth
cameras. In this section, we explain some of the main methods and discuss on their
performances.
Facial Feature
Facial feature detection usually involves face segmentation and landmark detection.
The former segments the face from the background, while the latter detect key
regions and feature points. To segment the face area from the background and the
rest of the human body, one can detect the skin color and perform segmentation
(Bronstein et al. 2005). However, such a method is easily affected by illumination.
Using the histogram of depth information from the depth image can improve the
system robustness (Segundo et al. 2010). Since human faces have the same topology,
it is possible to apply geometric rules to identify landmarks on the face. A simple
example is to approximate the face with an ellipse and divide the ellipse into
different slices based on predefined angles (Segundo et al. 2010). For each slices,
corresponding features can be searched based on the 3D height of the face. For
example, the eyes are the lowest point on the corresponding slice, while the nose is
the highest. Similarly, it is possible to use local curvature to represent different
features on the face, so as to determine different facial regions (Chang et al. 2006).
For example, the eye regions are usually a valley and can be represented by a specific
value of mean curvature and Gaussian curvature. The disadvantage for these
methods is that manually defined geometric rules may not be robust for different
users, especially for users coming from different countries. A better solution is to
apply a data-driven approach. For example, one can construct a database with
Fig. 2 3D facial landmark identified by Kinect overlapped on 2D color images
segmented facial regions and train a random forest that can automatically identify
facial regions on a face (Kazemi et al. 2014).
Another direction of representing facial features is to use predefined facial
template (Li et al. 2013; Weise et al. 2011). Such a template is a high-quality 3D
mesh with controllable parameters. During run-time, the system deforms the 3D
template to align with the geometry structure of the segmented face from the depth
image. Such a deformation process is usually done by numerical optimization due to
the high degree of freedom. Upon successful alignment, the systems can understand
the observed face in the depth image with the deformed template. It can also
represent the face with a set of deformation parameters so as to control animation
in real-time.
Microsoft Kinect also provides support for 3D landmark detection, as shown in
Fig. 2. Different expressions can be identified based on the arrangements of 3D
landmarks. Such understanding of facial orientation and expression is useful for real-
time animation control.
Body Posture
The mainstream of depth sensors-based body recognition system is to apply pattern

recognition and machine learning techniques to identify the human subject. By
training a classifier that can identify how individual body part appears in the depth
image, one can recognize these parts using real-time depth camera input (Girshick
et al. 2011; Shotton et al. 2012; Sun et al. 2012). There are several major challenges
in this algorithm. Chief among them is the availability of training data. In order to
train a classifier, a large number of depth images with annotation indicating the body
parts are needed. Since body parts appear differently based on the viewing angle, the
training database should capture such parts in different viewpoints. Moreover, since
users of different body sizes appear differently in depth images, to train a robust
classifier that can handle all users, training images consisting of body variation are
6 Y. Shen et al.
Fig. 3 (Left) The 3D skeleton obtained by Microsoft Kinect with the corresponding depth and
color images. (Middle and Right) Two views of the 3D point cloud together with the obtained 3D
skeleton
needed. As a result, hundreds of thousands of annotated depth images will be

required, which exceeds what human labors can generate. To solve the problem, it
is proposed to synthesize depth images using different humanoid models and 3D
motion capture data. Since the body part information of these humanoid models is
known in advance, it becomes possible to automatically annotate the position of the
body parts in the synthesized depth image. With these training images, one can train
a decision forest to classify depth pixels into the corresponding body parts. Different
designs of decision forest have resulted in different level of success, and they are all
capable of identifying body parts in real-time.
Microsoft Kinect also applied a pattern recognition approach to recognize body
parts with the depth images (Shotton et al. 2012). Figure 3 shows the results of
Kinect posture recognition, which is shown as the yellow skeleton. By overlapping
the skeleton with the 3D point cloud, it can be observed that the Kinect performs
reasonably accurate under normal circumstances.
Another stream of method in body identification and modeling is to take advan-
tage of the geometry of human body and utilize body template model (Liu
et al. 2016a; Zhang et al. 2014a). First, the pixels in the depth image that belongs
to the human body are extracted. Since the pixel value represents distance, one can
project them into a 3D space and create a point cloud of human body. Then, the
system fits a 3D humanoid mesh model into such a point cloud, so as to estimate the
body posture. This process involves deforming the 3D mesh model such that the
surface of the model aligns with the point cloud. Since the template model contains
human information such as body parts, when deforming the model to fit the point
cloud, we identify the corresponding body information in the point cloud. The main
challenge in this method is to deform the mesh properly to avoid unrealistic postures
and over-deformed surfaces, which is still a challenging research problem. Physics-
based motion optimization can ensure the physical correctness of the generated
postures (Zhang et al. 2014a). Utilizing simplified, intermediate template for defor-
mation optimization can enhance the optimization performance (Liu et al. 2016a).
This method can potentially provide richer body information depending on the
template used. However, a major drawback of such an optimization-based approach
is the higher run-time computational cost, making it inefficient to be applied in real-
time systems.
Human Environment Interaction
The depth images captured do not only contain information about the user but also
the surrounding environment. Therefore, it is possible to identify high-level infor-
mation about how the user interacts with the environment.
Unlike the human body, the environment does not have a uniform structure, and
therefore it is not possible to fit a predefined template or apply prior knowledge.
Geometry information based on planes and shapes become the next available infor-
mation to extract. The RANdom SAmple Consensus (RANSAC) algorithm can be
used to identify planer objects in the scenes such as walls and floors, which can help to
understand how the human user moves around in the open areas (Mackay et al. 2012).
It is also possible to compare successive depth images to identify the moving parts, in
order to understand how the user interacts with external objects (Shum 2013).
Depth cameras can be used for 3D scanning in order to obtain surface information
of the environment or even the human user. While one depth image only provide
information about a partial surface, which we called a 2.5D point cloud, multiple
depth images taken from different viewing angle can combine and form a full 3D
surface. One of the most representative systems in this area is called KinectFusion
(Newcombe et al. 2011). Such a system requires the user to carry a Kinect and
capture depth images continuously over a static environment. Real-time registration
is performed to understand the 3D translation and rotation movement of the depth
camera. This allows alignment of multiple depth images to form a complete 3D
surface. Apart from scanning the environment, it is possible to scan the face and
body of a human user (Cui et al. 2013) and apply real-time posture deformation on
the Kinect tracked skeleton (Iwamoto et al. 2015). Finally, because single view depth
cameras suffer from the occlusion problem, it is proposed to capture how human
users interact with objects by combining KinectFusion, color cameras, and
accelerometer-based motion capture system (Sandilands et al. 2012, 2013).
Since depth sensors can obtain both environment and human information, it facili-
tates the argument that human information can enhance understanding of unstructured
environment (Jiang and Saxena 2013; Jiang et al. 2013). Using a chair as an example. A
chair can come with different shapes and designs, which makes recognition extremely
difficult. However, the general purpose of a chair is for human to rest on. Therefore, with
the human movement information obtained by depth cameras, we can identify a chair
not just by its shapes but also by the way the human interacts with it. Similarly, human
movement may be ambiguous sometimes. Understanding the environment helps us to
identify the correct meaning of the human motion. Depth sensors open up new directions
on recognition by considering human and environment information together.
Dealing with Noisy Data
The main problem of using depth sensors is to deal with the noisy data obtained. In
particular, most depth sensor-based applications rely on a single point of view to
obtain the depth image. As a result, the quality of the detected face and posture are of
8 Y. Shen et al.
Fig. 4 An overview of depth sensors data enhancement
low resolution and suffer heavily from occlusion. It is possible to apply machine
learning algorithms to enhance the quality of the data. The idea is to introduce a
quality enhancement process that considers prior knowledge of the human body,
which is typically a database of high-quality faces or postures, as shown in Fig. 4. In
this section, we discuss how body and facial information can be reconstructed from
noisy data.
Face Enhancement
While depth sensors can obtain facial features, due to the relatively low resolution,
the quality of the features is not always satisfying. The 3D face obtained is usually
missing details and unrealistic. In this section, we explain how we can enhance the
quality of 3D faces obtained from depth sensors.
Since the quality of a single depth image is usually noisy with low resolution, the
3D facial surface generated is rough. By obtaining high-quality 3D faces through 3D
scanners and their corresponding color texture, one can construct a face database and
extract the corresponding prior knowledge (Liang et al. 2014; Wang et al. 2014).
These faces in the database are divided into patches such as eyes, nose, etc. Since
color texture is available, one can take advantage of color features to enhance the
segmentation accuracy. Given the low quality depth and color images of a face
obtained from sensors, facial regions are obtained in run-time. For each region
obtained, a set of similar patches is found in the database. Such a region is then
approximated by a weight sum of the database patches. By replacing different parts
of the run-time face image with their corresponding approximation, a high-quality
3D face surface can be generated. This method depends heavily on the quality and
variety of face in the database, as well as the way we abstract those faces to represent
the one observed by depth sensors in run-time.
Constructing a database for prior knowledge is costly. It is therefore proposed to
scan the face of the user in different angles, and apply such a face to enhance the
run-time detected face (Zollhöfer et al. 2014). The system first requests the user to
rotate around a depth sensor and obtain a higher quality 3D mesh, using registration
methods similar to the KinectFusion mentioned in the last section (Newcombe
et al. 2011). Then, given a run-time lower quality depth image of the face, the
system deforms the high quality 3D mesh such that it aligns with the depth image
pixels. As a result, high quality mesh with run-time facial expression can be
generated. The core problem here is to deform the high quality facial mesh nicely
and avoid generating visual artifact. It is shown that by dividing the face into
multiple facial regions to strengthen the feature correspondence, deformation quality
can be improved (Kazemi et al. 2014).
Posture Enhancement
The body tracked by depth sensors may contain inaccurate body parts due to
different types of error. Simple sensor error can be caused by geometry shape of
body parts and viewing angles. It is proposed to apply Butterworth filter (Bailey and
Bodenheimer 2012) or a simple low-pass filter (Fern’ndez-Baena et al. 2012) to
smooth out the vibration effect of tracked positions due to this type of error.
However, when occlusions occur, in which a particular body part is shield from
the camera, the tracked body position would contain a large amount of error. Simple
filter will not be sufficient to correct these postures.
As a solution, it is proposed to utilize accurately captured 3D human motion as
prior knowledge and reconstruct the inaccurate postures from the depth sensor. In
this method, a motion database is constructed using carefully captured 3D motion,
usually with optical motion capture systems. Given a depth sensor posture, one can
search for a similar posture in the database. The missing or error body parts from the
depth sensors can be replaced by those in the corresponding database posture (Shum
and Ho 2012). However, such a naive method cannot perform well for complex
posture, as using only one posture from the database cannot always generalize the
posture performed by the user, and therefore cannot effectively reconstruct the
posture.
More advanced posture reconstruction algorithms utilize machine learning to
generalize posture information from the motion database (Chai and Hodgins 2005;
Liu et al. 2011; Tautges et al. 2011). In particular, the motion database is used to
create a low dimensional latent space by dimensionality reduction techniques. Since
the low dimensional space is generated using data from real human, each point in the
space represents a valid natural posture. Given a partially mistracked posture from a
depth camera, one can project the posture into the learned low dimensional space and
apply numerical optimization to enhance the quality of the posture. The optimized
result is finally back-projected into a full body posture. Since the optimization is
performed in the low dimensional latent space, the solution found should also be a
natural posture. In other words, the unnatural elements due to sensor error can be
removed. The major problem of this method is that the system has no information
about which part of the body posture is incorrect. Therefore, while one would expect
the system to correct the error parts of the posture using information from the
accurate parts, the actual system may perform vice versa. As a result, the optimized
posture may no longer be similar to the original depth sensor input.
To solve the problem, optimization process that considers the reliability of
individual body part is proposed (Shum et al. 2013). The major difference from
10 Y. Shen et al.
Fig. 5 Applying posture reconstruction to enhance the quality of the obtained data
this method comparing with prior ones is that it divide the posture reconstruction
process into two steps. In the first step, a procedural algorithm is used to evaluate the
degree of reliability of individual body parts. This is by accessing the behavior of a
tracked body part to see if the position of the part is inconsistent, as well as accessing
the part with respect to its neighbor body parts to see if it creates inconsistent bone
length. In the second step, posture reconstruction is performed with reference to this
reliability information, such that the system relies on the more parts with higher
reliability. Essentially, the reliability information helps the system to explicitly use
the correct body parts and reconstruct the incorrect ones. Such a system can be
further improved by using Gaussian process to model the motion database, which
helps to reduce the amount of motion data needed to reconstruct the posture (Liu
et al. 2016b; Zhou et al. 2014). Better rules to estimate the reliability of the body
parts can also enhance the system performance (Ho et al. 2016).
Figure 5 shows the result of applying posture reconstruction. The color and depth
images show that the user is occluded by a chair and the surrounding environment.
The yellow skeleton on the left is the raw posture obtained by Kinect, in which less
reliable body parts are highlighted in red. The right character shows the
reconstructed posture using the method proposed in (Shum et al. 2013). The
awkward body parts are identified and corrected using the knowledge learned from
a motion database.
Prior Knowledge
The major research focus of face and posture enhancement is to apply appropriate
prior knowledge to improve data obtained in rum-time. In machine learning-based
algorithms, such prior knowledge is usually learned from a database and represented
in a format that can be efficiently used in run-time.
For motion enhancement, since human-motion is highly nonlinear with large
variation, it is not effective to represent the database using a single model. Instead,
many of the existing research apply multiple local models to represent the database,
such as using a mixture of Gaussian model (Liu et al. 2016b). It is also proposed to
apply deep learning to learn a set of manifolds that represents a motion database
(Holden et al. 2015). Precomputing these models and manifolds are time-consuming,
as it involves abstracting the whole database. Therefore, lazy learning algorithm is
adapted, in which modeling of the database is not done as a preprocesss but as a
run-time process using run-time information (Chai and Hodgins 2005; Shum
et al. 2013). During run-time, based on the user-performed posture, the system
retrieves a number of relevant postures from the database and models such a subset
of postures only. This method has two advantages. First, by modeling only a small
number of postures that are relevant to the performed posture, one can reduce the
computational cost of constructing a latent space. Second, since the subset of
postures are relatively similar, one can assume that they all lay in a locally linear
space and apply simpler linear dimensionality reduction to generate the latent space.
This allows real-time generation of the latent space. With improved database orga-
nization, the database search time can be further reduced and the relevancy of the
retrieved results can be enhanced (Plantard et al. 2016a, b), such that real-time
ergonomic and motion analysis applications can be preformed (Plantard
et al. 2016b).
Figure 6 visualizes how prior knowledge can be estimated from database. Each
blue circle in the figure represents a database entry, and the filling color represents its
value. The obtained prior knowledge from the scattered database entries is
represented by the shaded area, which enables one to understand the change of
value within the considered space. The left figure shows a traditional machine
learning algorithm, in which prior knowledge is obtained as a preprocess, consider-
ing all database entries. During run-time, when a query arrives, the system uses the
knowledge to estimate the corresponding value of the query. The right figure shows
the case of lazy learning, in which prior knowledge is obtained during run-time. This
allows the system to extract database entries that are more similar to the query and
estimate the prior knowledge with only such a subset of data.
Depth Camera-Based Applications
With depth sensors, it becomes possible to consider the user posture as part of an
animation system and create real-time animation. Here, we discuss on some depth
sensors-based animation control systems and point out the challenges and solutions.
Producing real-time facial animation with depth sensors is efficient. By
representing the facial features with a deformed template, it is possible to drive the
facial expression of virtual 3D faces (Li et al. 2013; Weise et al. 2011). Due to the
different dimensions between the faces of the user and the character, directly
12 Y. Shen et al.
Fig. 6 (Upper) Traditional machine learning that represents the prior from the whole database.
(Lower) Lazy learning that represents the prior from a subset of the database based on the online
query
applying facial features such as landmark locations generates suboptimal results. The
proposed common template acts as a bridge to connect the two ends. Such a template
is a parametric representation of the face, which is more robust against difference in
dimensions. With the template, it becomes possible to retarget the user’s expression
onto the character’s face.
Typical real-time animation systems such as games utilize a motion database to
understand what the user performs and renders the scenario accordingly. For exam-
ple, one can compare the user performed motion obtained from depth sensors with a
set of motion in the database and understand the nature of the motion as well as how
it should affect the real-time render (Bleiweiss et al. 2010). Alternatively, with an
interaction database, one can generate a virtual character that acts according to the
posture of the user, in order to create a two character dancing animation, which is
difficult to be captured due to hardware limitation (Ho et al. 2013).
While it is possible to utilize the posture captured by depth sensors for driving the
animation of virtual characters, the generated animation may not be physically
correct and dynamically plausible. On the one hand, since the depth sensors track
kinematic positions only, there is no information about the forces exerted. It is
proposed to combine the uses of depth cameras with pressure sensors and estimate
the internal joint torque using inverse dynamics (Zhang et al. 2014b). This allows
simulating virtual characters with physically correct movement. On the other hand,
while depth sensors can track the body parts positions, it is relatively difficult to track
Fig. 7 Real-time dynamic deformed character generated from Kinect postures
how the body deforms dynamically during the movement. Therefore, it is proposed
to enhance the realism of the generated character by applying real-time physical
simulation onto Kinect postures (Iwamoto et al. 2015). This allows the system to
synthesize real-time dynamic deformation, such as the jiggling of fresh, based on the
movement obtained in real-time, as shown in Fig. 7.
Utilizing depth cameras, user can interact with virtual objects with body motion.
On the one hand, predefined hand and arm gestures can be used to control virtual
objects. Once the Kinect has detected a set of specific gestures, a 3D virtual object
can be fitting onto the user and move according to the user’s gesture (Soh
et al. 2013). On the other hand, the virtual objects can be attached on the user’s
body and move with the user’s posture, such as carrying a virtual handbag (Wang
et al. 2012).
Depth sensors fuse a new application known as the virtual fitting, in which the
shopping experience can be facilitated by letting customers to try on virtual clothing.
This allows mix-and-match of clothes and accessories in real-time without being
physically present in the retail shop. The system involves building a 2D segmented
clothing database indexed by the postures of the user. During run-time, the system
searches for suitable database entries and overlays them on the customers fitting
image (Zhou et al. 2012). Another clothes fitting method is to utilize a 3D clothing
database and obtain 3D models of the user with depth sensors. This allows the
system to recommend items that fits with the user’s body (Pachoulakis and
Kapetanakis 2012).
Conclusion
In this chapter, we explained how depth sensors are applied to gather human facial
and body posture information in generating and controlling animations. Depth
sensors obtain human information in real-time and provide a cheaper alternative
for human motion capturing. However, there are still rooms to improve the quality of
the obtained data. In particular, depth sensors suffer heavily from occlusions, in
which part of the human body is shield. Machine learning algorithms can reconstruct
the data and improve the quality, but more research is needed to solve the problem.
Depth sensors fuse many interesting applications in the computer animation and
14 Y. Shen et al.
games domains, providing real-time control on animation creation. Given the rate of
new depth sensors research and applications, such a technology can become an
important part of the daily life in the near future.
Acknowledgment This work is supported by the Engineering and Physical Sciences Research
Council (EPSRC) (Ref: EP/M002632/1).
References
Alex Butler D, Izadi S, Hilliges O, Molyneaux D, Hodges S, Kim D (2012) Shake’n’sense: reducing
interference for overlapping structured light depth cameras. In: Proceedings of the SIGCHI
conference on human factors in computing systems, CHI’12. ACM, New York, pp 1933–1936
Bailey SW, Bodenheimer B (2012) A comparison of motion capture data recorded from a vicon
system and a Microsoft Kinect sensor. In: Proceedings of the ACM symposium on applied
perception, SAP’12. ACM, New York, pp 121–121
Bleiweiss A, Eshar D, Kutliroff G, Lerner A, Oshrat Y, Yanai Y (2010) Enhanced interactive
gaming by blending full-body tracking and gesture animation. In: ACM SIGGRAPH ASIA
2010 Sketches. Seoul, South Korea. ACM, p 34
Bronstein AM, Bronstein MM, Kimmel R (2005) Three-dimensional face recognition. Int J Comput
Vision 64(1):5–30
Chai J, Hodgins JK (2005) Performance animation from low-dimensional control signals. In
SIGGRAPH’05: ACM SIGGRAPH 2005 Papers. ACM, New York, pp 686–696
Chang KI, Bowyer KW, Flynn PJ (2006) Multiple nose region matching for 3d face recognition
under varying facial expression. IEEE Trans Pattern Anal Mach Intell 28(10):1695–700
Cui Y, Chang W, Nöll T, Stricker D (2013) Kinectavatar: fully automatic body capture using a single
Kinect. In: Proceedings of the 11th international conference on computer vision, vol
2, ACCV’12. Springer-Verlag, Berlin/Heidelberg, pp 133–147
Fern’ndez-Baena A, SusÃn A, Lligadas X (2012) Biomechanical validation of upper-body and
lower-body joint movements of Kinect motion capture data for rehabilitation treatments. In:
Intelligent Networking and Collaborative Systems (INCoS), 2012 4th International Conference
on, pp 656–661
Fernandez-Sanchez EJ, Diaz J, Ros E (2013) Background subtraction based on color and depth
using active sensors. Sensors 13(7):8895–915
Girshick R, Shotton J, Kohli P, Criminisi A, Fitzgibbon A (2011) Efficient regression of general-
activity human poses from depth images. In: Computer Vision (ICCV), 2011 I.E. international
conference on. Barcelona, Spain. pp 415–422
Ho ESL, Chan JCP, Komura T, Leung H (2013) Interactive partner control in close interactions for
real-time applications. ACM Trans Multimedia Comput Commun Appl 9(3):21:1–21:19
Ho ES, Chan JC, Chan DC, Shum HP, Cheung YM, Yuen PC (2016) Improving posture classifi-
cation accuracy for depth sensor-based human activity monitoring in smart environments.
Comput Vis Image Underst 148:97–110. doi:10.1111/cgf.12735
Holden D, Saito J, Komura T, Joyce T (2015) Learning motion manifolds with convolutional
autoencoders. In ACM SIGGRAPH ASIA 2015 technical briefs. ACM, Kobe, Japan. 2015
SIGGRAPH ASIA
Iwamoto N, Shum HPH, Yang L, Morishima S (2015) Multi-layer lattice model for real-time
dynamic character animation. Comput Graph Forum 34(7):99–109
Jiang Y, Saxena A (2013) Hallucinating humans for learning robotic placement of objects. In:
Proceedings of the 13th international symposium on experimental robotics. Springer Interna-
tional Publishing, Heidelberg, pp 921–937
Jiang Y, Koppula H, Saxena A (2013) Hallucinated humans as the hidden context for labeling 3d
scenes. In: Proceedings of the 2013 I.E. conference on computer vision and pattern recognition,
CVPR’13. IEEE Computer Society, Washington, DC, pp 2993–3000
Kakumanu P, Makrogiannis S, Bourbakis N (2007) A survey of skin-color modeling and detection
methods. Pattern Recogn 40(3):1106–22
Kazemi V, Keskin C, Taylor J, Kohli P, Izadi S (2014) Real-time face reconstruction from a single
depth image. In: 3D Vision (3DV), 2014 2nd international conference on, vol 1. IEEE, Lyon,
France. 2014 3DV. pp 369–376
Kinect sdk. https://developer.microsoft.com/en-us/windows/kinect
Kyan M, Sun G, Li H, Zhong L, Muneesawang P, Dong N, Elder B, Guan L (2015) An approach to
ballet dance training through ms Kinect and visualization in a cave virtual reality environment.
ACM Trans Intell Syst Technol (TIST) 6(2):23
Li H, Yu J, Ye Y, Bregler C (2013) Realtime facial animation with on-the-fly correctives. ACM
Trans Graph 32(4):42–1
Liang S, Kemelmacher-Shlizerman I, Shapiro LG (2014) 3d face hallucination from a single depth
frame. In: 3D Vision (3DV), 2014 2nd international conference on, vol 1. IEEE, Lyon, France.
2014 3DV. pp 31–38
Liu H, Wei X, Chai J, Ha I, Rhee T (2011) Realtime human motion control with a small number of
inertial sensors. In: Symposium on interactive 3D graphics and games, I3D’11. ACM,
New York, pp 133–140
Liu Z, Huang J, Bu S, Han J, Tang X, Li X (2016a) Template deformation-based 3-d reconstruction
of full human body scans from low-cost depth cameras. IEEE Trans Cybern PP(99):1–14
Liu Z, Zhou L, Leung H, Shum HPH (2016b) Kinect posture reconstruction based on a local
mixture of gaussian process models. IEEE Trans Vis Comput Graph 14 pp. doi:10.1109/
TVCG.2015.2510000
Mackay K, Shum HPH, Komura T (2012) Environment capturing with Microsoft Kinect. In:
Proceedings of the 2012 international conference on software knowledge information manage-
ment and applications, SKIMA’12. Chengdu, China. 2012 SKIMA
Newcombe RA, Izadi S, Hilliges O, Molyneaux D, Kim D, Davison AJ, Kohli P, Shotton J,
Hodges S, Fitzgibbon A (2011) Kinectfusion: real-time dense surface mapping and tracking.
In: Proceedings of the 2011 10th IEEE international symposium on mixed and augmented
reality, ISMAR’11. IEEE Computer Society, Washington, DC, pp 127–136
Pachoulakis I, Kapetanakis K (2012) Augmented reality platforms for virtual fitting rooms. Int J
Multimedia Appl 4(4):35
Plantard P, Shum HP, Multon F (2016a) Filtered pose graph for efficient kinect pose reconstruction.
Multimed Tools Appl 1–22. doi:10.1007/s11042-016-3546-4
Plantard P, Shum HPH, Multon F (2016b) Ergonomics measurements using Kinect with a pose
correction framework. In: Proceedings of the 2016 international digital human modeling
symposium, DHM ’16, Montreal, 8 p
Sandilands P, Choi MG, Komura T (2012) Capturing close interactions with objects using a
magnetic motion capture system and a rgbd sensor. In: Proceedings of the 2012 motion in
games. Springer, Berlin/Heidelberg, pp 220–231
Sandilands P, Choi MG, Komura T (2013) Interaction capture using magnetic sensors. Comput
Anim Virtual Worlds 24(6):527–38
Segundo MP, Silva L, Bellon ORP, Queirolo CC (2010) Automatic face segmentation and facial
landmark detection in range images. Systems Man Cybern Part B Cybern IEEE Trans 40
(5):1319–30
Shotton J, Girshick R, Fitzgibbon A, Sharp T, Cook M, Finocchio M, . . . Blake A (2013) Efficient
human pose estimation from single depth images. IEEE Trans Pattern Anal Machine Intell 35
(12):2821–2840
Shum HPH (2013) Serious games with human-object interactions using rgb-d camera. In: Pro-
ceedings of the 6th international conference on motion in games, MIG’13. Springer-Verlag,
Berlin/Heidelberg
16 Y. Shen et al.
Shum HPH, Ho ESL (2012) Real-time physical modelling of character movements with Microsoft
Kinect. In: Proceedings of the 18th ACM symposium on virtual reality software and technology,
VRST’12. ACM, New York, pp 17–24
Shum HPH, Ho ESL, Jiang Y, Takagi S (2013) Real-time posture reconstruction for Microsoft
Kinect. IEEE Trans Cybern 43(5):1357–69
Soh J, Choi Y, Park Y, Yang HS (2013) User-friendly 3d object manipulation gesture using Kinect.
In: Proceedings of the 12th ACM SIGGRAPH international conference on virtual-reality
continuum and its applications in industry, VRCAI’13. ACM, New York, pp 231–234
Sun M, Kohli P, Shotton J (2012) Conditional regression forests for human pose estimation. In:
Computer Vision and Pattern Recognition (CVPR), 2012 I.E. conference on. Providence, Rhode
Island. pp 3394–3401
Tautges J, Zinke A, Krüger B, Baumann J, Weber A, Helten T, Müller M, Seidel H-P, Eberhardt B
(2011) Motion reconstruction using sparse accelerometer data. ACM Trans Graph 30
(3):18:1–18:12
Vera L, Gimeno J, Coma I, Fernández M (2011) Augmented mirror: interactive augmented reality
system based on Kinect. In: Human-Computer Interaction–INTERACT 2011. Springer, Lisbon,
Portugal. 2011 INTERACT. pp 483–486
Wang L, Villamil R, Samarasekera S, Kumar R (2012) Magic mirror: a virtual handbag shopping
system. In: Computer Vision and Pattern Recognition Workshops (CVPRW), 2012 I.E. com-
puter society conference on. IEEE, Rhode Island. 2012 CVPR. pp 19–24
Wang K, Wang X, Pan Z, Liu K (2014) A two-stage framework for 3d facereconstruction from rgbd
images. Pattern Anal Mach Intell IEEE Trans 36(8):1493–504
Weise T, Bouaziz S, Li H, Pauly M (2011) Realtime performance-based facial animation. ACM
Trans Graph (TOG) 30:77, ACM
Zhang P, Siu K, Jianjie Z, Liu CK, Chai J (2014a) Leveraging depth cameras and wearable pressure
sensors for full-body kinematics and dynamics capture. ACM Trans Graph 33(6):221:1–221:14
Zhang P, Siu K, Jianjie Z, Liu CK, Chai J (2014b) Leveraging depth cameras and wearable pressure
sensors for full-body kinematics and dynamics capture. ACM Trans Graph (TOG) 33(6):221
Zhou Z, Shu B, Zhuo S, Deng X, Tan P, Lin S (2012) Image-based clothes animation for virtual
fitting. In: SIGGRAPH Asia 2012 technical briefs. ACM, Singapore. 2012 SIGGRAPH ASIA.
p 33
Zhou L, Liu Z, Leung H, Shum HPH (2014) Posture reconstruction using Kinect with a probabi-
listic model. In: Proceedings of the 20th ACM symposium on virtual reality software and
technology, VRST’14. ACM, New York, pp 117–125
Zollhöfer M, Nießner M, Izadi S, Rehmann C, Zach C, Fisher M, Wu C, Fitzgibbon A, Loop C,
Theobalt C et al (2014) Real-time non-rigid reconstruction using an rgb-d camera. ACM Trans
Graph (TOG) 33(4):156
Real-Time Full-Body Pose Synthesis
and Editing
HO Edmond S. L. and Yuen Pong C.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Editing and Synthesizing Full-Body Pose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Editing Poses by Inverse Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Data-Driven Pose Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
User Interface for Full-Body Posing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
Posing character has always been playing an important role in character anima-
tion and interactive applications such as computer games. However, such a task is
time-consuming and labor-intensive. In order to improve the efficiency in char-
acter posing, researchers in computer graphics have been working on a wide
variety of semi- or fully automatic approaches in creating full-body poses,
ranging from traditional approaches like inverse kinematics (IK), data-driven
approaches which make use of captured motion data, as well as direct pose
manipulation through intuitive interfaces. In this book chapter, we will introduce
the aforementioned techniques and also discuss their applications in animation
production.
H. Edmond S. L.
Department of Computer and Information Sciences, Northumbria University, Newcastle upon
Tyne, UK
e-mail: e.ho@northumbria.ac.uk
Y. Pong C. (*)
Department of Computer Science, Hong Kong Baptist University, Kowloon Tong, Hong Kong
e-mail: pcyuen@comp.hkbu.edu.hk

DOI 10.1007/978-3-319-30808-1_8-1
2 H. Edmond S. L. and Y. Pong C.
Keywords
Pose synthesis • Pose editing • Inverse kinematics • Motion retargeting • Motion
blending • Jacobian-based IK • Cyclic coordinate decent • Collision avoidance •
Data-driven pose synthesis
Introduction
A recent study Kyto et al. (2015) found that 61 % of time in animation production
was spent on character posing by professional animators. Such a time-consuming
task motivated tremendous effort in improving the efficiency of human character
posing over the last three decades. Producing full-body poses has always been
playing an important role in character animation. For example, a long motion
sequence can be represented by key poses in keyframe animation, and the
in-between poses can be interpolated from the key poses. Such an approach is still
widely used in animation production nowadays.
While only a small number of poses have to be created in keyframe animation,
creating a full-body pose manually (e.g., controlling the locations of all body parts)
is a tedious and time-consuming task. As results, researchers have been working on
fully or semiautomatic methods in full-body pose creation. In this chapter, we will
review methods in character posing, including those creating new pose from scratch
and creating new pose by modifying existing motion data. In addition to introducing
the pose creation techniques, the applications of those techniques in animation
production, such as reaching, retargeting, foot-skate cleanup, etc., will be discussed.
State of the Art
Jacobian-based inverse kinematics (IK) algorithms have been widely used in pose
editing over the last three decades in computer animation. A recent work by Harish
et al. (2016) presents a new parallel numerical IK approach for multithreaded
architectures. The new approach can handle complex articulated structures (e.g.,
with more than 600 degrees of freedom) with multiple constraints at real time.
Besides the traditional approaches, many intuitive real-time posing approaches
have been proposed. Guay et al. (2013) proposed to synthesize 3D pose of a
character by drawing a single aesthetic line called line of action (LOA). An example
is shown in Fig. 1. The main idea of the new method is to edit the pose of the
character such that the selected body line becomes similar to the sketched simple
curve given by the user. Ten body lines which connect every pair of end effectors
(e.g., head to right hand, left hand to right leg, etc.) are defined, and the most
appropriate one will be selected according to the location of the line sketched and
the viewpoint.
Another interesting real-time posing approach proposed by Rhodin et al. (2014)
recently enables the user to control the poses of a virtual character by a wide range of
Real-Time Full-Body Pose Synthesis and Editing 3
Fig. 1 3D poses generated by a single aesthetic line (colored in red) proposed by Guay et al. (2013)
(Reproduced with permission from Guay et al. (2013))
Fig. 2: Posing characters with different articulated structures with the tangible input device
proposed in Jacobson et al. (2014) (Reproduced with permission from Jacobson et al. (2014))
input motion such as full-body, hand, or face motion from any motion-sensing
devices. Instead of controlling the skeleton of the characters, the new method defines
the vertex-to-vertex correspondence between the meshes representing the input
motion and the virtual character and deforms the mesh of the character accordingly.
Furthermore, researchers have been trying to provide the users with a more
natural way in posing characters. In Jacobson et al. (2014), a tangible puppet-like
input device is proposed for interactive pose manipulation. By measuring the relative
bone orientation on the input device, the pose of the virtual character will be updated
accordingly. Different from previously developed puppet-like input devices (e.g.,
Esposito et al. 1995), the new device can be used for articulated structure with
different topologies as the device is assembled from modular, interchangeable, and
hot-pluggable parts. Examples of posing are shown in Fig. 2.
To further enhance the realism of the resultant animation, physics-based
approaches have been adapted in animation production. However, due to the high
computational cost in previous approaches, most of them cannot be used in interac-
tive applications. A recent work by Hämäläinen et al. (2015) synthesizes physically
valid poses of humanlike characters at interactive frame rate in performing a wide
variety of motions such as balancing on a ball, recovering from disturbances,
reaching and juggling a ball, etc.
Editing and Synthesizing Full-Body Pose
In this section, techniques for synthesizing and editing full-body character pose will
be introduced. Firstly, a traditional posing approach inverse kinematics (IK) which is
widely used in computer animation and robotics will be discussed in Section 3.1. In
particular, different types of IK solvers (Sections 3.1.1 and 3.1.3) and examples of
their applications (Section 3.1.2) will be presented. Secondly, data-driven
approaches which make use of existing motion data to produce natural-looking
pose will be explained in Section 3.2. Finally, direct manipulation of the pose of
the character using different kinds of intuitive interfaces such as puppet-based,
natural user interface and sketch-based approaches will be reviewed in Section 3.3.
Editing Poses by Inverse Kinematics
With the advancement of motion capture technology, more and more motion data are
available nowadays. Reusing the captured or existing poses in new applications can
be more efficient (in terms of production time and labor cost) than creating new
poses from scratch. However, the number of captured or existing poses is limited,
and it is necessary to edit the available poses according to the user’s new require-
ments. In addition, there are demands in editing the full-body pose at runtime in
interactive applications such as computer games.
In computer animation, characters are usually represented as articulated structures
in which the body segments connected at the joint locations. The pose of the
character is then controlled by the joint parameters (e.g., joint angles) and the global
translation of the root joint (e.g., the hip of a humanlike character). Given the skeletal
structure (e.g., bone lengths), the joint parameters of the current pose, and the
changes in the joint parameters, the changes in location or orientation of every
joint can be computed:
x_i ¼ J i θ_ (1)
where x_i can be the changes in positions or orientations of the i-th joint and Ji is the
Jacobian matrix for mapping the changes in joint parameters θ_ to the changes in x_i of
the i-th joint:

@xi
Ji ¼ (2)
@θ
However, it is a tedious task for the user to control the pose of a character by
specifying that all joint parameters as a humanlike character usually composed of
more than 40 joints. Instead of editing a pose by specifying all of the joint parameters
as in Eq. 1, the new pose can be produced by specifying the target location and/or the
orientation of the selected joint(s) and the required changes in the joint parameters
will be computed by inverse kinematics (IK).
IK has been widely used in robotics and computer animation for controlling
robots and characters. Poses of articulated characters such as human and animal
figures can be edited by such an approach for producing computer animations. IK
can be applied to a wide range of applications in animation production, for example,
keyframe postures editing, interactive postures editing, and editing pre-captured
motion data.
An early analytic approach proposed by Lee and Shin (1999) determines the
posture based on the positions of the hands and feet relative to the positions of
shoulder and hips. High performance in pose editing has been demonstrated in their
experiments. However, one of the major limitations of analytic approaches is that the
analytic solvers must be designed specifically to each individual system which
cannot be applied to arbitrary articulated structures. On the other hand, numerical
IK solvers are more general and will be discussed below.
Numerical Inverse Kinematics Approaches

Numerical solvers linearize the relationship of joint parameters and the positions
and/or orientations of the end effectors around the current posture to obtain the IK
solution for new position and/or orientation of the end effectors close to the current
position and/or orientation:
θ_ ¼ J 1
i x_i (3)
where J 1
i is the inverse of the Jacobian matrix (Eq. 2) of the i-th joint. Since the
Jacobian matrix may not be a square matrix, the pseudoinverse of the Jacobian
matrix can be used for solving the IK problem. The pseudoinverse J þ i can be
calculated by:
1
Jþ T T
i ¼ Ji JiJi (4)
and the IK problem can be solved by:
θ_ ¼ J þ
i x_i (5)
There are three main advantages for numerical solvers:
• They can be applied to arbitrary chain structures.

• Various types of constraints, such as positional or planar constraints, can be
handled in the same platform.
• Constraints can be easily switched on and off.
Therefore, many numerical IK approaches have been proposed. The most prac-
tical and commonly used numerical solver is based on the least square methods
(Whitney 1969). One of the major problems in the original least squares method is
that it becomes unstable near the singularity points, in which results in large changes
in the solution (i.e., the joint parameters ̇ θ). The singularity problem usually occurs
when there is no physically feasible solution to the IK problem; for example, the
target location of a controlled joint is unreachable or there are multiple constraints
which are conflicting with each other.
To tackle this problem, various methodologies such as singularity-robust
(SR) inverse (Nakamura and Hanafusa 1986) have been developed to stabilize the
system near such singularity postures for generating full-body postures in graphics
applications (Yamane and Nakamura 2003). The main idea of using SR inverse is to
introduce a weighting parameter to balance between satisfying constraints and
stabilizing the changes in joint parameters. The SR inverse J i is calculated by:
1
J i ¼ J Ti J i J Ti þ kI (6)
where k is the weighting parameter and I is the identity matrix. The larger the value
of k, the larger the error in satisfying constraints, but the solution becomes more
stable. The bottleneck of these methods is the cost of computing the pseudoinverse
matrix, which grows cube proportional to the number of constraints.
Baraff (1996) proposes a method of forward dynamics for articulated body
structures, which can be used for solving IK problems. Instead of calculating the
pseudoinverse matrix, an equation of Lagrange multipliers is solved. Since the
matrix used in his method is sparse, efficient solvers for sparse matrix can be used.
However, the method can only handle equality constraints, and the cost still
increases cubic proportional to the number of auxiliary constraints.
Application of Numerical Inverse Kinematics Approaches

As numerical IK approaches can be applied to arbitrary articulated structures, a wide
range of applications has been developed for real-time full-body pose creation. Four
types of applications are briefly discussed below.
Footskate Cleanup
In animation production, the motion sequence can be produced by interpolating key
poses or directly using MOCAP data. However, when interpolating the key poses,
artifacts such as footskate can be produced. Footskate occurs when the motion
cannot reproduce the footplants the animation intended to create. For example, in
a walking motion, footskates occurs when the character slides over the surface of the
ground. When reusing MOCAP data, footskate will occur if the motion data were not
well captured due to the noise or tracking error. This problem can be solved by
analyzing the motion sequence (e.g., the state in a walk cycle, ankle height and
rotation, etc.) and determine the position of the feet at every pose (in each frame).
Then, IK can be applied to edit the poses accordingly (Kovar et al. 2002; Kulpa et al.
2005; Lee and Shin 1999; Lu and Liu 2014).
Character Retargeting
While reusing existing motions, including both MOCAP data and previously created
motions, is common practice in animation production, it is a very challenging task.
Fig. 3 Examples of retargeting the original Judo poses (colored in red) to characters with different
sizes using interaction mesh (Ho et al. 2010b) (Reproduced with permission from Ho et al. (2010b))
This is because the articulated character used in existing motions may differ from the
new character(s) in body segment lengths and sizes. As a result, animating new
characters with existing motion data may result in loss of contacts (e.g., wrong
footplates, unable to reach an object). To tackle this problem, IK can be applied at
every pose (in each frame) to constrain some of the body parts (such as hands and
feet) to preserve the contacts as in the original motion while editing the other parts of
the body accordingly (Gleicher 1998). Motion retargeting can also be applied to
transferring the live performance of a human subject to the movement of a virtual
character (Shin et al. 2001).
Ho et al. (2010b) proposed interaction mesh, which represents the spatial rela-
tions between closely interacting body parts. By preserving the shape of the inter-
action mesh while retargeting characters to new sizes, penetration-free postures will
be produced. Examples of retargeting closely interacting characters to different sizes
are illustrated in Fig. 3. The spatial relation-based representation can also be used for
controlling the movement of humanoid robots in highly constrained environment
(Ho and Shum 2013) and synthesizing virtual partner in VR applications (Ho et al.
2013a).
Collision Avoidance in Pose Editing

Collision and interpenetration of body segments can significantly degrade the
realism of the resultant character animation. Collisions may occur when the charac-
ter is interacting with other objects and characters in the scene. To solve this
problem, collision detection algorithms will be applied to determine which body
part(s) are colliding with other body parts or objects. Next, the colliding segments
will be moving away from each other to avoid the interpenetration by applying IK to
edit the poses. Various approaches have been proposed (Kallmann 2008; Lyard and
Magnenat-Thalmann 2008).
Editing Poses for Interaction Between Characters

For interactive applications such as computer games and virtual reality application,
virtual characters will response to the avatars or objects controlled by the user. When
handling virtual scenes with close interactions between the characters such as
fighting and dancing, the poses of the characters have to be edited at runtime to
preserve the context of the interaction. Imagine that the attacking character is trying
to punch the head of the defending characters. The punching trajectories have to be
edited in order to reach the target (i.e., the head) when the defending characters is
moving around (e.g., controlled by the user). For example, Shum et al. (Shum et al.
2007; Shum et al. 2008) edit the positions of the interacting body parts at every
frame, and Ho and Komura (Ho and Komura 2009; Ho and Komura 2011; Ho et al.
2010a) edit the way the body parts tangling with each other. IK can also be used for
creating the reactive motion when external perturbation is applied on the character
(Komura et al. 2005).
Heuristic Inverse Kinematics Approach

One of the representative heuristic search approaches in solving IK problems is the
cyclic coordinate descent (CCD) (Wang and Chen 1991) method. The CCD method
is a simple and fast approach for computing the joint parameters to satisfying the
constraints iteratively. The IK problem is solved in cycles in which each joint
parameter will be computed in a cycle. Starting with the outermost joint (i.e., closet
to the end effector) (Welman 1993) in the articulated structure, the joint parameters
are updated sequentially to bring the end effector E to the target location T. When
editing a joint parameter on joint i, the positions of joint i, E, and T in Cartesian
coordinates are projected to Pi, Ea, and Ta according to the axis of rotation a. Next,
the rotation Δθjointi required to bring Ea closer to Ta can be found by calculating the
angle between the two vectors originated from Pi to Ea and Pi to Ta:

ð E a Pi Þ ð T a P i Þ
Δθjointi ¼ arccos (7)
kðEa Pi Þk kðT a Pi Þk
The direction of rotation will be determined by the cross product of the two
vectors Pi to Ea and Pi to Ta. By iteratively updating the joint parameters in every
cycle, the difference between the positions of E and T can be minimized. Since the
joint parameters can be computed analytically as in Eq. 7 in each step, computa-
tionally expansive calculations such as matrix manipulations are not needed and
result in less computational costly than numerical approaches introduced in 3.1.1.
This makes the CCD method suitable for real-time full-body pose editing
applications.
Summary of IK Approaches
A general problem of traditional IK algorithms is the difficulty to ensure the
naturalness of the synthesized motion. This is because natural human motion
involves a lot of subtle behaviors such as balancing and correlation of body parts,
which are difficult to be modeled mathematically. In the next section, we will
introduce methodologies in using precaptured human motion to improve the solution
for pose editing.
Fig. 4 Results obtained in Rose et al. (1998), the sample motions (green) and the blended motions
(yellow) (Reproduced with permission from Rose et al. (1998))
Data-Driven Pose Synthesis
The idea of data-driven motion synthesis is to make use of captured motion data to
create the required postures, such that natural and humanlike movement can be
created by specifying a relatively small number of constraints. An early work by
Rose et al. (1998) edits poses by interpolating collected poses to satisfy the con-
straints. This is based on an old technique called motion blending in computer
animation. In Rose et al. (1998), a concept of verbs and adverbs is proposed to
generate new poses from examples. In their work, verbs refer to parameterized
motions constructed from sets of similar motions, and adverbs are parameters that
control the verbs. For each verb, the sample motions are time aligned by manually
specifying the key-time for every motion. Then, the motions clips are placed on a
parameter space based on the characteristics of the motion clips. Motion blending is
done by computing the weights of the sample motions in the corresponding verb
using radial basis functions (RBF). By specifying the adverbs, new motion will be
created. In addition, users can create a verbgraph so that transition motions between
verbs can be generated. Figure 4 shows the sample motions (green) and the blended
motions (yellow) created by their method.
In the approaches proposed in recent years, the collected human motions have to
be analyzed first, and machine learning tasks are often required to learn a model for
pose synthesis in later stage. In the rest of this section, we roughly divide the data-
driven approaches into two categories: offline training and online modeling
approaches.
Offline Training Approaches

Synthesizing a natural-looking pose can be viewed as finding a solution (i.e., joint
parameters) from a natural movement space created using captured motions.
Grochow et al. (2004) propose to use scaled Gaussian process latent variable
model (SGPLVM) (Lawrence 2004) to create such a natural pose space. While the
process of learning the pose model is done offline, the learned model can be used for
real-time full-body pose synthesis. By specifying constraints such as the positions of
the hands and feet, natural-looking full-body pose can be synthesized. However, due
to the complexity of the learning process, the model cannot be trained with large
number of poses. In Wu et al. (2011), Wu et al. further propose to select a subset of
distinctive postures in a large pose database for learning a natural pose space for pose
synthesis. Wei and Chai (2011) solved the same problem by constructing a mixture
of factor analysis. The algorithm segments the motion database into local regions
and models each of them individually. Nevertheless, the training cost and system
complexity increases with the amount of source data, and the effectiveness of
dimensional reduction reduces with the increase of motion data variety.
Online Modeling Approaches

As opposed to offline training approaches, online modeling has shown to be
effective for real-time application with large motion dataset. The idea is to select a
small subset of posture based on run-time information to synthesize the required
posture. For example, Chai and Hodgins (2005) use a lazy learning approach to learn
low-dimensional local linear models (principal component analysis (PCA)) to
approximate the high-dimensional manifold which contains the natural and valid
poses during runtime. Given the current pose of the character and the target positions
of the selected joint(s) as constraints, a set of postures that are similar to the current
one is used to learn the local linear model. By interpolating the poses in the low-
dimension space while minimizing the energy terms to ensure that the synthesized
pose is smooth (i.e., joint velocities), and satisfy the constraints given by the user and
the probability distribution of the captures motion in the training data, natural-
looking full-body motion can be synthesized.
Liu et al. extended the idea by using the maximum a posteriori framework to
reconstruct the motion, which enhanced the consistency of the movement in the
temporal domain (Liu et al. 2011). The general problem of these methods is that it is
difficult to ensure the set of extracted postures to be logically similar as kinematics
metric is used. Ho et al. (2013b) also use a lazy learning approach to learn local
linear models while taking into account the spatial relationship between body parts.
The topology-based approach computes and represents the tangling of body parts
using a subset of topology coordinates (Ho and Komura 2009). As interpolating
Fig. 5 Posing characters with close interactions while avoiding penetration of body parts by the
method proposed in Ho et al. (2013b) (Reproduced with permission from Ho et al. (2013b))
poses with significant difference can easily result in interpenetrates of the body parts,
their method only selects topologically similar poses to learn the local model and
ensures the changes in spatial relationship to be small when editing the pose. By this,
penetration-free postures can be created (Fig. 5).
User Interface for Full-Body Posing
Beside the pose synthesis and editing approaches introduced above, another stream
of research lies in providing a more natural and intuitive interface for the users to
pose the characters. In this subsection, three types of character posing interfaces,
puppet-based interface, natural user interface, and sketch-based interface, will be
introduced.
Puppet-Based Interface
An early work by Esposito et al. (1995) provides the users with a puppet called
Monkey as shown in Fig. 6. Monkey is a humanlike puppet with 32 degrees of
freedom, 1800 tall, and about 600 wide. Rotational sensors are located at the joints of
the puppet to measure the joint angles when the user manipulates the puppet. The
pose sequence produced by puppet-based input devices can also be used as example
motion to retrieve similar movement from the database (Numaguchi et al. 2011). A
recently proposed tangible input device (Jacobson et al. 2014) introduced in
Section 2 further enables users to create articulated characters with different topol-
ogy for real-time full-body posing.
Natural User Interface

With the advancement in motion-sensing technologies, a wide variety of natural user
interface (NUI) applications have been proposed. In human posing, a recent work by
Fig. 6 An input device,

called Monkey, for interactive
pose manipulation
(Reproduced with permission
from Esposito et al. (1995))
Oshita et al. (2013) captures the movements of the fingers and hands of the user
when manipulating an intangible puppet. The design of the puppet control is inspired
by traditional puppet controlling mechanism, in which the head/body rotation and
body translation are controlled by the right hand while the legs of the character are
controlled by the left hand. The hand and finger movement is captured at runtime
using leap motion controller (Motion 2016). Unlike previous sensor-based
approaches like the data glove-based methods (Isrozaidi et al. 2010; Komura and
Lam 2006), no sensor or marker is required to be attached on the hand as the leap
motion controller tracks the finger and hand movement by emitting infrared
(IR) light and analyzing the reflected IR light to calculate the 3D positions of
different parts of the hand(s) over time (Fig. 7).
Sketch-Based Interface
Another type of popular intuitive posing interfaces is sketch-based approaches
which are inspired by the pose design process in traditional 2D hand-drawn anima-
tion production. Mapping 2D sketch to 3D pose is an under-constrained problem. An
early work proposed by Igarashi et al. (1999) enables user to sketch the 2D silhouette
of the character, and the corresponding 3D mesh model will be generated. In
additional, methods for posing 3D characters using sketches of the skeleton in 2D
Fig. 7 An input device, called Monkey, for interactive pose manipulation (Reproduced with
permission from Oshita et al. (2013))
stick figures (Choi et al. 2012; Davis et al. 2003; Wei and Chai 2011) are also an
active research topic. In Lin et al. (2012), the user can sketch the sitting pose of a
stick figure in 2D. By taking into account the interaction between the sketched pose
and the environment and preserving the physical correctness such as balancing of the
character, the 3D pose will be produced at interactive rate (with GPU
implementation).
To further simplify the input from the user, highly abstracted sketch such as the
line of action approach (Guay et al. 2013) introduced in Section 2 has been proposed.
A recent work proposed by Hahn et al. (2015) further proposed a method to allow the
user to define a custom sketch abstraction, for example, by sketching the outline or
the skeleton, and the system will map the sketch to the rigging parameters to edit the
pose of the 3D character by deforming the mesh model.
Conclusion
In this chapter, various kinds of real-time full-body posing approaches have been
discussed. Traditional posing approaches such as IK automatically creates new pose
according to a small number of constraints which reduces the workload of the
animators in posing characters. Data-driven approaches produce natural-looking
poses by limiting the poses to be produced lie in the natural pose space. Finally, a
wide range of intuitive interfaces for directly manipulating the pose of the character
to further simplify the posing process. While the research interests in full-body
posing have been changing from traditional methods to more intuitive controls, we
believe IK will still play an important role in character posing.
References
Baraff D (1996) Linear-time dynamics using lagrange multipliers. In: SIGGRAPH ’96: Proceedings
of the 23rd annual conference on computer graphics and interactive techniques. ACM,
New York, pp 137–146. doi:10.1145/237170.237226
Chai J, Hodgins JK (2005) Performance animation from low-dimensional control signals. In:
SIGGRAPH ’05: ACM SIGGRAPH 2005 papers. ACM, New York, pp 686–696.
doi:10.1145/1186822.1073248
Choi MG, Yang K, Igarashi T, Mitani J, Lee J (2012) Retrieval and visualization of human motion
data via stick figures. Comput Graph Forum 31(7pt1):2057–2065. doi:10.1111/j.1467-
8659.2012.03198.x
Davis J, Agrawala M, Chuang E, Popović Z, Salesin D (2003) A sketching interface for articulated
figure animation. In: Proceedings of the 2003 ACM SIGGRAPH/eurographics symposium on
computer animation, SCA ’03. Eurographics Association, Aire-la-Ville, pp 320–328 http://dl.
acm.org/citation.cfm?id=846276.846322
Esposito C, Paley WB, Ong J (1995) Of mice and monkeys: a specialized input device for virtual
body animation. In: Proceedings of the 1995 symposium on interactive 3D graphics, I3D ’95.
ACM, New York, p 109–ff. doi:10.1145/199404.199424
Gleicher M (1998) Retargeting motion to new characters. In: SIGGRAPH ’98: Proceedings of the
25th annual conference on computer graphics and interactive techniques. ACM Press,
New York, pp 33–42. doi:10.1145/280814.280820
Grochow K, Martin SL, Hertzmann A, Popović Z (2004) Style-based inverse kinematics. ACM
Trans Graph 23(3):522–531. doi:10.1145/1015706.1015755
Guay M, Cani MP, Ronfard R (2013) The line of action: an intuitive interface for expressive
character posing. ACM Trans Graph 32(6):205:1–205:8. doi:10.1145/2508363.2508397
Hahn F, Mutzel F, Coros S, Thomaszewski B, Nitti M, Gross M, Sumner RW (2015) Sketch
abstractions for character posing. In: Proceedings of the 14th ACM SIGGRAPH/eurographics
symposium on computer animation, SCA ’15. ACM, New York, pp 185–191. doi:10.1145/
2786784.2786785
Hämäläinen P, Rajamäki J, Liu CK (2015) Online control of simulated humanoids using particle
belief propagation. ACM Trans Graph 34(4):81:1–81:13. doi:10.1145/2767002
Harish P, Mahmudi M, Callennec BL, Boulic R (2016) Parallel inverse kinematics for multithreaded
architectures. ACM Trans Graph 35(2):19:1–19:13. doi:10.1145/2887740
Ho ESL, Komura T (2009) Character motion synthesis by topology coordinates. In: Dutr’e P,
Stamminger M (eds) Computer graphics forum (Proceedings of Eurographics 2009), Munich,
vol 28, pp 299–308
Ho ESL, Komura T (2011) A finite state machine based on topology coordinates for wrestling
games. Comput Animat Virtual Worlds 22(5):435–443. doi:10.1002/cav.376
Ho ESL, Shum HPH (2013) Motion adaptation for humanoid robots in constrained environments.
In: Robotics and automation (ICRA), 2013 I.E. international conference on, pp 3813–3818.
doi:10.1109/ICRA.2013.6631113
Ho ESL, Komura T, Ramamoorthy S, Vijayakumar S (2010a) Controlling humanoid robots in
topology coordinates. In: Intelligent robots and systems (IROS), 2010 IEEE/RSJ international
conference on, pp 178–182. doi:10.1109/IROS.2010.5652787
Ho ESL, Komura T, Tai CL (2010b) Spatial relationship preserving character motion adaptation.
ACM Trans Graph 29(4):1–8. doi:10.1145/1778765.1778770
Ho ESL, Chan JCP, Komura T, Leung H (2013a) Interactive partner control in close interactions for
real-time applications. ACM Trans Multimed Comput Commun Appl 9(3):21:1–21:19.
doi:10.1145/2487268.2487274
Ho ESL, Shum HPH, Ym C, PC Y (2013b) Topology aware data-driven inverse kinematics.
Comput Graph Forum 32(7):61–70. doi:10.1111/cgf.12212
Igarashi T, Matsuoka S, Tanaka H (1999) Teddy: a sketching interface for 3d freeform design. In:
Proceedings of the 26th annual conference on computer graphics and interactive techniques,
SIGGRAPH ’99. ACM Press/Addison-Wesley, New York, pp 409–416. doi:10.1145/
311535.311602
Isrozaidi N, Ismail N, Oshita M (2010) Data glove-based interface for real-time character motion
control. In: ACM SIGGRAPH ASIA 2010 Posters, SA ’10. ACM, New York, p 5:1.
doi:10.1145/1900354.1900360
Jacobson A, Panozzo D, Glauser O, Pradalier C, Hilliges O, Sorkine-Hornung O (2014) Tangible

and modular input device for character articulation. ACM Trans Graph 33(4):82:1–82:12.
doi:10.1145/2601097.2601112
Kallmann M (2008) Analytical inverse kinematics with body posture control. Comput Animat
Virtual Worlds 19(2):79–91
Komura T, Lam WC (2006) Real-time locomotion control by sensing gloves. Comput Animat
Virtual Worlds 17(5):513–525. doi:10.1002/cav.114
Komura T, Ho ESL, Lau RW (2005) Animating reactive motion using momentum-based inverse
kinematics: motion capture and retrieval. J Vis Comput Animat 16(3–4):213–223. doi:10.1002/
cav.v16:3/4
Kovar L, Schreiner J, Gleicher M (2002) Footskate cleanup for motion capture editing. In: SCA ’02:
Proceedings of the 2002 ACM SIGGRAPH/Eurographics symposium on Computer animation,
pp 97–104. doi:10.1145/545261.545277
Kulpa R, Multon F, Arnaldi B (2005) Morphology-independent representation of motions for
interactive human-like animation. Computer Graphics Forum 24(3):343–351. doi:10.1111/
j.1467-8659.2005.00859.x
Kyto M, Dhinakaran K, Martikainen A, Hamalainen P (2015) Improving 3d character posing with a
gestural interface. IEEE Comput Graph Appl. doi:10.1109/MCG.2015.117
Lawrence ND (2004) Gaussian process latent variable models for visualisation of high dimensional
data. In: Advances in neural information processing systems (Proceedings of NIPS 2003). MIT
Press, Cambridge, MA, pp 329–336
Lee J, Shin SY (1999) A hierarchical approach to interactive motion editing for human-like figures.
In: Proceedings of the 26th annual conference on computer graphics and interactive techniques,
SIGGRAPH ’99. ACM Press/Addison-Wesley Publishing, New York, pp 39–48. doi:10.1145/
311535.311539
Lin J, Igarashi T, Mitani J, Liao M, He Y (2012) A sketching interface for sitting pose design in the
virtual environment. IEEE Trans Vis Comput Graph 18(11):1979–1991. doi:10.1109/
TVCG.2012.61
Liu H, Wei X, Chai J, Ha I, Rhee T (2011) Realtime human motion control with a small number of
inertial sensors. In: Symposium on interactive 3D graphics and games, I3D ’11. ACM,
New York, pp 133–140. doi:10.1145/1944745.1944768
Lu J, Liu X (2014) Foot plant detection for motion capture data by curve saliency. In: Computing,
Communication and Networking Technologies (ICCCNT), 2014 international conference on,
pp 1–6. doi:10.1109/ICCCNT.2014.6963001
Lyard E, Magnenat-Thalmann N (2008) Motion adaptation based on character shape. Comput
Animat Virtual Worlds 19(3–4):189–198. doi:10.1002/cav.v19:3/4
Leap Motion (2016, n.d.) https://www.leapmotion.com/
Nakamura Y, Hanafusa H (1986) Inverse kinematics solutions with singularity robustness for robot
manipulator control. J Dyn Syst Meas Control 108:163–171
Numaguchi N, Nakazawa A, Shiratori T, Hodgins JK (2011) A puppet interface for retrieval of
motion capture data. In: Proceedings of the 2011 ACM SIGGRAPH/Eurographics symposium
on computer animation, SCA ’11. ACM, New York, pp 157–166. doi:10.1145/
2019406.2019427
Oshita M, Senju Y, Morishige S (2013) Character motion control interface with hand manipulation
inspired by puppet mechanism. In: Proceedings of the 12th ACM SIGGRAPH international
conference on virtual-reality continuum and its applications in industry, VRCAI ’13. ACM,
New York, pp 131–138. doi:10.1145/2534329.2534360
Rhodin H, Tompkin J, Kim KI, Kiran V, Seidel HP, Theobalt C (2014) Interactive motion mapping
for real-time character control. Comput Graph Forum (Proc Eurograph) 33(2):273–282.
doi:10.1111/cgf.12325
Rose C, Cohen MF, Bodenheimer B (1998) Verbs and adverbs: multidimensional motion interpo-
lation. IEEE Comput Graph Appl 18:32–40. doi:10.1109/38.708559
Shin HJ, Lee J, Shin SY, Gleicher M (2001) Computer puppetry: an importance-based approach.
ACM Trans Graph 20(2):67–94. doi:10.1145/502122.502123
Shum HPH, Komura T, Yamazaki S (2007) Simulating competitive interactions using singly
captured motions. In: Proceedings of ACM virtual reality software technology 2007, pp 65–72
Shum HPH, Komura T, Yamazaki S (2008) Simulating interactions of avatars in high dimensional
state space. In: ACM SIGGRAPH symposium on interactive 3D graphics (i3D) 2008, pp
131–138
Wang LCT, Chen CC (1991) A combined optimization method for solving the inverse kinematics
problems of mechanical manipulators. IEEE Trans Robot Autom 7(4):489–499. doi:10.1109/
70.86079
Wei XK, Chai J (2011) Intuitive interactive human-character posing with millions of example
poses. IEEE Comput Graph Appl 31:78–88. doi:10.1109/MCG.2009.132
Welman C (1993) Inverse kinematics and geometric constraints for articulated figure manipulation.
Master’s thesis, Simon Frasor University
Whitney D (1969) Resolved motion rate control of manipulators and human prostheses.
Man-Machine Syst IEEE Trans 10(2):47–53. doi:10.1109/TMMS.1969.299896
Wu X, Tournier M, Reveret L (2011) Natural character posing from a large motion database. IEEE
Comput Graph Appl 31(3):69–77. doi:10.1109/MCG.2009.111
Yamane K, Nakamura Y (2003) Natural motion animation through constraining and deconstraining
at will. IEEE Trans Vis Comput Graph 9(3):352–360. doi:10.1109/TVCG.2003.1207443
Real-Time Full Body Motion Control
John Collomosse and Adrian Hilton
Abstract
This chapter surveys techniques for interactive character animation, exploring data-
driven and physical simulation-based methods. Interactive character animation is
increasingly data driven, with animation produced through the sampling, concat-
enation, and blending of pre-captured motion fragments to create movement. The
chapter therefore begins by surveying commercial technologies and academic
research into performance capture. Physically based simulations for interactive
character animation are briefly surveyed, with a focus upon technique proven to
run in real time. The chapter focuses upon concatenative synthesis approaches to
animation, particularly upon motion graphs and their parametric extensions for
planning skeletal and surface motion for interactive character animation.
Keywords
Animation • 3D motion capture • Real-time motion • Virtual reality • Augmented
reality • 4D mesh calculation • Parametric motion
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Commercial Technologies for Performance Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Marker-Less Human Motion Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Interactive Character Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Real-Time Physics-Based Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Concatenative Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Summary and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
J. Collomosse (*) • A. Hilton

Centre for Vision Speech and Signal Processing (CVSSP), University of Surrey, Surrey, UK
e-mail: j.collomosse@surrey.ac.uk; a.hilton@surrey.ac.uk

DOI 10.1007/978-3-319-30808-1_9-1
2 J. Collomosse and A. Hilton
Introduction
Compelling visuals and high-quality character animation are cornerstones of modern

video games and immersive experiences. Yet character animation remains an expen-
sive process. It can take a digital artist weeks to skin (design the 3D surface
representation) of a character model and then rig it with a skeleton to facilitate full
body control and animation. Animation is often expedited by retargeting human
performance capture data to drive the character’s movement. Yet creativity and
artistic input remains in the loop, blending handcrafted animation with motion
capture data which itself may be an amalgam of multiple takes (e.g., it is common
for separate passes to be used for the face, head, and hands). Performance capture
itself is expensive; equipment hire, operation, and studio/actor time can approach
millions of US dollars on a high-end production. The recent resurgence of virtual and
augmented reality (VR/AR) experiences, in which character interaction takes place
at very close quarters, is further driving up expectations of visual realism.
Creating believable interactive digital characters is therefore a trade-off between
project budget and quality. Better tool support inevitably leads to efficiency and so a
rebalancing toward higher quality. In this chapter, we survey state-of-the-art tech-
nologies and algorithms (as of 2015) for efficient interactive character animation.
While a common goal is a drive toward increased automation, which in some cases
can produce interactive characters with near-complete automation, one should not
lose sight that these are tools only and the need for the creative artist in the loop
remains essential to reach the high-quality bar demanded by modern production. As
such this chapter takes a practical view on animation, first surveying the commercial
technologies and academic research into performance capture and then surveying the
two complementary approaches to real-time animation – physically based
approaches (examined further in chapter C-2) and data-driven approaches.
Although character animation is frequently used within other domains with the
Creative Industries (movies, broadcast), its use within games requires new animation
sequences to be generated on the fly, responding in real time to user interaction and
game events. This places some design restrictions on the underpinning algorithms
(efficient data structures, no temporal look ahead for kinematics). This chapter
therefore focuses upon algorithms for interactive, rather than more general offline,
character animation covered elsewhere in this book.
State of the Art
Historically character animation has been underpinned by meticulous observations

of movement in nature, for example, the gait cycles of people or animals. This link
has been made explicit by contemporary character animation, which is trending
toward a data-driven process in which sampled physical performance is the basis for
synthesizing realistic movement in real time. This chapter therefore begins by
surveying commercial technologies, and state-of-the-art Computer Vision algo-
rithms, for capturing human motion data.
Real-Time Full Body Motion Control 3
Commercial Technologies for Performance Capture
Motion Capture (mocap) technology was initially developed within the Life Sci-
ences for human movement analysis. The adoption of mocap for digital entertain-
ment, commonly referred to as performance capture (PC), is now widespread. PC
accounts for 46% of the total 3D motion capture system market which is growing
annually at a rate of around 10% and expected to reach 142.5 million US dollars by
2020 (Rohan 2015). Indeed many of the innovations in mocap (e.g., marker-less
capture) are now being developed within the Creative Industries and transferred back
into domains such as biomechanics and healthcare.
PC systems enable sequences of skeletal joint angles to be recorded from one or
several actors. The key distinction between PC systems is the kind of physical
marker or wearable device (if any) required to be attached to the actors.
The predominant form of PC in the Creative Industries is marker based, using
passive markers that are tracked visually using synchronized multiple viewpoint
video (MVV). Popular systems for passive marker PC are manufactured by Vicon
(UK) and Optitrack (US), which require the actor to wear retroreflective spheres
(approximately 20–30 are typically used for full body capture). A region of the
studio (capture volume) is surrounded by several infrared (IR) cameras in known
locations and illuminated by several diffuse IR light sources. Prior to capture of
performance data, a calibration process is performed to learn the relative locations
(extrinsic parameters) of the cameras. This enables the world location of the markers
attached to the actor to be triangulated, resulting in a 3D point cloud from which a
skeletal pose is inferred using physical and kinematic constraints. Modern software
(e.g., Blade or MotionBuilder) can perform this inference in real time providing
immediate availability of a pose estimate for each actor in the scene. PC service
providers (e.g., The Imaginarium Studios, London UK) have begun to harness this
technology to pre-visualize the appearance of digital characters for movie or game
production during live actor performance. Such facilities provide immediate visual
feedback to both the actor and director onset removing the trial and error and so
improving efficiency in the capture process (Fig. 1, left). Other forms of passive PC
in regular use include the fractal suits patented by Industrial Light and Magic for full
body motion capture (Fig. 1, right). The suits are tracked using visible light and so
are more amenable to deployment in outdoor sets where strong natural light makes
IR impractical.
Active marker-based systems include offerings from CodaMotion (UK) and
PhaseSpace (US). Markers are bright IR light-emitting diodes (LEDs) that pulse
with unique signatures that identify the marker to one or several observing cameras.
Since markers are uniquely labeled at source, automated tracking of markers is trivial
making marker confusion highly unlikely. By contrast, the labeling of triangulated
markers in a passive system is performed during pose inference and may be incorrect
in the presence of clutter (e.g., multiple occluding actors). Marker mislabeling causes
errors in pose estimation, which can only be removed through addition of more
witness cameras so reducing the chance of occlusion or manually correcting the data
post-capture. An advantage of active marker-based systems is therefore the need for
Fig. 1 Performance capture technologies. Top: Vicon IR-based system being used to pre-visualize
character performance in real time within the UNREAL Games engine (EPIC). Bottom: Industrial
Light and Magic’s fractal suit enabling visual light-based tracking outdoors
fewer cameras and reduced data correction. Active systems tend to perform better
outdoors, again due to obviating the need for large area IR illumination. The
disadvantage is the additional expense and time required for actor setup (wires and
batteries) due to the complexity of the markers. The workflow to produce a skeletal
pose estimate from active marker data is identical to passive systems, since the
capture again results in a sequence of 3D point movements.
Inertial motion capture system uses inertial motion-sensing units (IMUs) to detect
changes in joint orientation and movement, providing an alternative to visual
tracking and so removing the problem of marker occlusion. IMUs are worn on each
limb (around 12–14 for full body capture) and connected wirelessly to a hub which
forwards the data for software processing. Common IMU captures solutions include
AnimeZoo (UK), XSens (Netherlands), and most recently the crowd-funded
PerceptionNeuron (US) system. All of these solutions again rely upon a robust
back-end software product to infer a skeletal pose estimate using physical and
kinematic constraints. The disadvantage of inertial capture is drift, since the IMUs
output only a stream of relative joint angles. For this reason, IMU mocap is
sometimes combined with a secondary modality, e.g., laser ranging or passive
video, to capture the world position of the actor.
An emerging form of PC is marker-less mocap, using Computer Vision to track
the actor without the need for wearables. Although the accuracy of commercial
marker-less systems has yet to reach parity with marker-based solutions, the greatly
reduced setup time and flexibility to use only regular video cameras for capture
makes such systems a cost-effective option. For the purposes of teaching data-driven
animation production, marker-less technologies are therefore attractive. Solutions
include the OrganicMotion stage (US), a cube arrangement of around 20 machine
vision cameras that calculates human pose using the silhouette of the performer
against a uniform background from the multiple camera angles. More recently The
Captury (Germany) launched a software-only product for skeletal PC that estimates
pose against an arbitrary background using a possibly heterogeneous array of
cameras. Yet although commercial solutions to marker-less PC remain in their
infancy, academic research is making good progress as we next discuss.
Marker-Less Human Motion Estimation
Passive estimation of human pose from video is a long-standing Computer Vision

challenge, particularly when visual fiducials (markers) are not present. Methods can
be partitioned into those considering monocular (single-view) video or multiple
viewpoint video.
Monocular Human Pose Estimation

Human pose estimation (HPE) often requires the regions of interest (ROIs)
representing people to be identified within the video. This person localization
problem is can be solved using background (Zhao and Nevatia 2003) or motion
(Agarwal and Triggs 2006) subtraction, in the cases of simple background. In more
cluttered scenarios, supervised machine can be applied to detect the presence of a
person within a sliding window swept over the video frame. Within each position of
the window, pretrained classifiers based on Histogram of Gradient (HoG) descriptors
can robustly identify the torso (Eichner and Ferrari 2009), face (Viola and Jones
2004), or entire body (Dalal and Triggs 2005).
Once the subject is localized within the frame, the majority of monocular HPE
algorithms attempt to infer only a 2D, i.e., apparent pose of the performer. These
adopt either (a) top-down fitting of a person model, optimizing limb parameters and
projecting to image space to evaluate correlation with image data, or (b) individually
segmenting parts and integrating their positions in a bottom-up manner to produce a
maximal likelihood pose.
Bottom-up approaches dominated early research into HPE, over one decade ago.
Srinivasan and Shi (2007) used an image segmentation algorithm (graph cut) to
parse a subset of salient shapes from an image and group these into a shape
resembling a person using a set of learned rules. However the approach was limited
to a single person, and background clutter was reported to interfere with the initial
segmentation and so the eventual accuracy of the approach. Ren et al. proposed an
alternative algorithm in which Canny edge contours were recursively split into
segments, each of which was classified as a putative body part using shape cues
such as parallelism (Ren et al. 2005). Ning et al. (2008) similarly attempted to label
body parts individually, applying a Bag of Visual Words (BoVW) framework to
learn codewords for body zone labeling – segmenting 2D body parts to infer pose.
Mori and Malik described the first bottom-up algorithm capable of estimating a 3D
pose in world space, identifying the position of individual joints in a 2D image using
scale and symmetry constraints, and then matching those 2D joint positions to a set
of many “training images” each of which had been manually annotated a priori with
2D joint positions (Mori et al. 2004) and was associated also with a 3D ground truth.
Once the closest training image had been identified by matching query and training
joint positions in 2D, the relevant 3D pose was returned as the result.
Top-down approaches, in which the entire 2D image is used as evidence to fit a
model, are more contemporary. The most common form of model fitted to the image
is a “pictorial structure,” essentially a collection of 2D limbs (regions) articulated by
springs, that can be iteratively deformed to fit to evidence in the image under an
optimization process (Andriluka et al. 2009; Eichner and Ferrari 2009). However
such approaches do not yield recover a 3D pose estimate or if so are unstable due to
ambiguity in reasoning from a single image.
Multi-View Human Pose Estimation

A 3D estimate of human pose may be inferred with less ambiguity using footage
captured from multiple viewpoints. In such a setup, a configuration of cameras
(typically surround a subject in a 180 or 360 arc) observes a capture volume within
which a performance is enacted. The cameras are typically calibrated, i.e., for a
subject observed by C camera views c = [1,C] the extrinsic parameters
{Rc,COPc}
(camera orientation and focal point) and intrinsic parameters f c , oxc , oyc (focal
length and 2D optical center) are known.
Two categories of approach exist: (a) those estimating 2D pose from each view
independently and fusing these to deduce a 3D pose and (b) those inferring a 3D
pose from 3D geometric proxy of the performer recovered through volumetric
reconstruction.
Computer Vision has undergone a revolution in recent years, with deep
convolutional neural networks (CNNs) previously popular in text recognition
being extended and applied to solve many open problems including human pose
Fig. 2 Convolutional neural networks (CNNs) used for pose estimation in multi-viewpoint video.
(a) Using 2D detections of body parts fused in a 3D probabilistic model (from (Elhayek et al.
2015)), (b) recognition of pose from 3D volumetric data recovered from multiple views (from
(Trumble et al. 2016))
estimation. CNNs have shown particularly strengths in general object detection, with
some state-of-the-art networks, e.g., GoogLeNet (Google Inc.), surpassing human
performance in certain scenarios. Most recently CNNs have also been used to detect
human body parts in single and multiple viewpoint video and infer from these human
pose. Elhayek et al. (2015) estimate human body parts from individual video
viewpoints using CNN detectors and then fuse these under a probabilistic model
fusing color and motion constraints from a body part tracker to create a 3D pose
estimate. The CNN detection step is robust to clutter, making the system suitable for
estimation of 3D pose in complex scenes including outdoors (Fig. 2a).
In volumetric approaches, a geometric proxy of the performer is built using a
visual hull (Grauman et al. 2003) computed from foreground mattes extracted each
camera image Ic using a chroma key or more sophisticated image segmentation
algorithm. To compute the visual hull, the capture volume is coarsely decimated into
a set of voxels at locations V = {V1, . . ., Vm}; a resolution of 1 cm3 is commonly used

for a capture volume of approximately 6 2 6 m. The probability of the voxel
being part of the performer in a given view c is:
pðVj cÞ ¼ BðI c ðx½V i , y½V i ÞÞ, (1)
where B(.) is a simple blue dominance term derived from the RGB components of
B
I c ðx, yÞ, i:e:1 RþGþB and (x, y) is the point within Ic that Vi projects to:
f c vx f vy
x½V i ¼ þ oxc and y½V i ¼ c þ oyc , where, (2)
vz vz

vx vy vz ¼ COPc R1
c Vi: (3)
The overall probability of occupancy for a given voxel p(V ) is:
C
pðV i Þ ¼ ∏ 1= 1 þ epðV j cÞ : (4)
i¼1
We compute p(Vi) for all Vi V to create a volumetric representation of the

performer for subsequent processing. An iso-contour extraction algorithm such as
marching cubes (Lorensen and Cline 1987) is used to extract a triangular mesh
model from the voxel-based visual hull (Fig. 3). The result is a topologically
independent 3D mesh for each frame of video. This can be converted into a
so-called “4D” representation using a mesh tracking process to conform these
individual meshes to a single mesh that deforms over time (Budd et al. 2013).
Once obtained, it is trivial to mark up a single frame of the performance to embed
a skeleton (e.g., marking each joint limb as an average of subsets of mesh vertices)
and have the skeleton track with the performance as the mesh deforms. As we
explain in subsection “Concatenative Synthesis,” either the skeletal or surface
representations from such a 4D performance capture may be used to drive character
animation interactively.
CNNs have also been applied to volumetric approaches, with a spherical histo-
gram (c.f. subsection “Surface Motion Graphs”) derived from the visual hull being
fed into a CNN to directly identify human pose (Trumble et al. 2016). The system
contrasts with Elhayek et al. (2015) where the CNN operates in 2D rather than 3D
space and similarly adds robustness to visual clutter in the scene.
Interactive Character Animation
Interactive character animation often takes place within complex digital environ-
ments, such as games, in which multiple entities (characters, moveable objects, and
static scene elements) interact continuously. Since these interactions are a function of
user input, they cannot be predicted or scripted a priori, and enumerating all possible
Fig. 3 4D performance capture. Multiple video views (top) are fused to create a volumetric
representation of the performance which is meshed (bottom). The per-frame meshes are conformed
to a single deforming mesh over time, into which a skeleton may be embedded and tracked (right)
eventualities is intractable. It is therefore necessary to plan animation in real time

using fast, online algorithms (i.e., algorithms using data from the current and
previous timesteps only). Two distinct categories of algorithm exist.
First, algorithms drawing upon pre-supplied database of motion for the character,
usually obtained via PC and/or manual scripting. Several fragments of motion data
(“motion fragments”) are stitched and blended together to create a seamless piece
animation. A trivial example is a single cycle of a walk, which can be repeatedly
concatenated to create a character walking forward in perpetuity. However more
complex behavior (e.g., walks along an arbitrary path) can be created by carefully
selecting and interpolating between a set of motion fragments (e.g., three walk
cycles, one veering left, one veering right, and one straight-ahead) such that no
jarring movement occurs. This form of motion synthesis, formed by concatenating
(and in some cases interpolating between) several motion fragments, is referred to as
“concatenative synthesis.” The challenge is therefore in selecting and sequencing
appropriate motion fragments to react to planning requirements (move from A to B)
under environmental (e.g., occlusion) and physical (e.g., kinematic) constraints. This
is usually performed via a graph optimization process, with the motion fragments
and valid transitions between these encoded in the nodes and edges of a directed
graph referred to as a “move tree” or “motion graph” (Kovar et al. 2002). The key
advantages of a motion graph are predictability of movement and artistic control
over the motion fragments that are challenging to embody within a physical simu-
lation. The disadvantage is that motion cannot generalize far beyond the motion
fragments, i.e., character movement obtained via PC in the studio. We discuss
concatenative synthesis in detail within subsection “Concatenative Synthesis.”
Second, algorithms that do not require prescripted or directly captured animation

but instead simulate the movement under physical laws. Physics simulation is now
commonly included within games engines (e.g., Havoc, PhysX) but used primarily
to determine motion of objects or particles or animation of secondary characteristics
such as cloth attached to characters (Armstrong and Green 1985). Yet more recently,
physics-based character animation has been explored integrating such engines into
the animation loop of principal characters (Geijtenbeek et al. 2010). Physics-based
simulation offers the significant advantage of generalization; characters modeled in
this manner can react to any situation with the virtual world and are not bound to a
database on preordained movements. Nevertheless, the high computational cost of
simulation forces accuracy-performance trade-offs for real-time use. Simplifying
assumptions such as articulated rigid bodies for skeletal structure is very common.
It is therefore inaccurate to consider physically simulated animation as being more
“natural”; indeed the tendency of simulation to produce “robotic” movements
lacking expressivity has limited practical uptake of these methods for interactive
character animation until comparatively recently. We briefly discuss physics-based
character control in the next section, restricting discussion to the context of real-time
animation for interactive applications. A detailed discussion of physics-based char-
acter animation in a broader context can be found in chapter C-2.
Real-Time Physics-Based Simulation
Physically simulated characters are usually modeled as a single articulated structure

of rigid limb components, interconnected by two basic forms of joint mimicking
anatomy in nature. Characters modeled under physical simulation are typically
humanoid (Hodgins 1991; Raibert and Hodjins 1991) or animal (Wampler and
Popovic 2009) consisting predominantly of hinge joints, with hip and shoulder
joints implemented as ball-socket joints. Depending on the purpose of the simula-
tion, limbs may be amalgamated for computational efficiency (e.g., a single compo-
nent for the head, neck, and torso) (Tin et al. 2008). More complex simulations can
include sliding joints in place of some hinge joints that serve to model shock
absorption within the ligaments of the leg (Kwon and Hodgins 2010).
Character Model Actuation

The essence of the physical simulation is to solve for the forces and torques that
should be applied to each limb, in order to bring about a desired motion. This solve is
performed by a “motion controller” algorithm (subsection “Character Motion Con-
trol”). The locations at each limb where forces are to be applied are a further design
consideration of the modeler. The most common strategy is to consider torque about
each joint (degree of freedom), a method known as servo actuation. While intuitive,
servo actuation is not natural – effectively assuming each joint to contain a motor
capable of rotating its counterpart – careful motion planning is necessary to guard
against unnatural motion arising under this simplified model.
Biologically inspired models include simulated muscles that actuate through

tendons attached to limbs, effecting a torque upon the connected joints. Muscle-
actuated models are more challenging to design motion controllers for, since the
maximum torque that can be applied by a muscle is limited by the turning moment of
the limb which is dependent on the current pose of the model. Furthermore the
number of degrees of freedom in such models tends to be higher than servo-actuated
models, since muscles tend to operate in an antagonist manner with a pair of muscles
per joint enabling “push” and “pull” about the joint. Moreover such models cannot
be considered natural unless the tendons themselves are modeled as nonrigid
structures, capable of stretch and compressing to store and release energy in the
movement. The high computational complexity of motion controllers to solve for
muscle-actuated models therefore remains a barrier to their use in real-time character
animation, whose applications to digital entertainment (rather than say, biomechan-
ics) rarely require biologically accurate simulation. We therefore do not consider
them further in this chapter.
Character Motion Control

Use cases for character animation rarely demand direct, fine-grain control of each
degree of freedom in the model. Rather, character control is directed at a higher level,
e.g., “move from A to B at a particular speed, in a particular style.” Such directives
are issued by game AI, narrative, or other higher level controllers. Motion controllers
are therefore a mid-layer component in the control stack bridging the semantic gap
between high-level control and low-level actuation parameters. In interactive sce-
narios, simple servo-based actuation (i.e., independent, direct control over joint
torques) is adopted to ensure computation of the mapping is tractable in real time.
Solving for the movement is performed iteratively, over many small timesteps,
each incorporating feedback supplied by the physics engine from each actuation of
the model at the previous timestep under closed-loop control. This obviates the need
to model complex outcomes of movements within the controller itself. Feedback
comprises not only global torso position and orientation but also individual joint
orientation and velocity post-simulation of the movement. It is common for control-
lers to reason about the stability (balance) of the character when planning movement.
The center of mass (COM) of the character should correspond to the zero-moment
point (ZMP), i.e., the point at which the reaction force from the world surface results
in a zero net moment. When the COM and ZMP coincide, the model is stable.
We outline two common strategies to motion control that are applicable to
physically based real-time interactive character animation.
Control in Joint-Space via Pose Graphs Some of the earliest animation engines
comprised carefully engineered software routines, to procedurally generate motion
according to mechanics models embedded within kinematics solvers and key-framed
poses. This approach is derived from real-time motion planning in the robotics
literature.
Such approaches model the desired end-position of a limb (or limbs) as a “key-
pose.” Using a kinematics engine, the animation rig (i.e., joint angles) is gradually
adjusted to bring the character’s pose closer to that desired key-pose. The adjustment
is an iterative process of actuation and feedback from the environment to determine
the actual position of the character and subsequent motions. For example, the COM
and ZMP as well as the physical difference between the current and intended joint
positions are monitored to ensure that the intended motion does not unbalance the
character unduly and that progress is not impeded by itself or other scene elements.
A sequence of such key-poses is defined within a “Pose Space Graph” (PSG),
where the nodes in the graph are procedurally defined poses, i.e., designed by the
animator, but the movements between poses are solved using an inverse kinematics
engine (IK). A motion, such as a walk, is performed by transitioning through states
in the PSG (Fig. 4 illustrates a walk cycle in a PSG). Due to physical or timing
constraints, a character often will not reach a desired pose within the PSG before
being directed toward the next state. Indeed it is often unhelpful for the character to
decelerate and pause (i.e., obtain ZMP of zero) and become stable at a given state
before moving on the next; a degree of perpetual instability, for example, exists
within the human walk cycle. Therefore key-poses in the PSG are often exaggerated
on the expectation that the system will approximate rather than interpolate the
key-poses within it.
The operation of PSGs is somewhat analogous to motion graphs (c.f. subsection
“Skeletal Motion Graphs”), except that IK is used to plan motion under physical
models, rather than pre-captured performance fragments concatenated and
played back.
Control via Machine Learned Models Although expensive to train, machine

learning approaches offer a run-time efficiency unrivaled by other real-time motion
controller strategies. Most commonly, neural networks (NN) are used to learn a
mapping between high-level control parameters and low-level joint torques, rather
than manually identified full body poses (subsection “Character Motion Control”).
Notwithstanding design of the fitness function of the network and its overall
architecture, the training process itself is fully automatic using a process trial and
error via feedback from the physics simulation. A further appeal of such approaches
is that such training is akin to the biological process of learned actuation in nature.
Networks usually adopt a shallow feed-forward network such as a multi-layer
perceptron (MLP) (Pollack et al. 2000), although the growing popularity of deeply
learned networks has prompted some recent research into their use as motion
controllers (Holden et al. 2016). Training the MLP proceeds from an initially
randomized (via Gaussian or “white” noise) set of network weights, using a function
derived from some success metric typically the duration that the controlled model
can execute movement (e.g., walk) for without destabilizing and falling over. Many
thousands of networks (weight configurations) are evaluated to drive character
locomotion, and the most successful networks are modified and explored further in
an iterative optimization process to train the network (Sims 1994).
Optimization of the NN weights is commonly performed by an evolutionary
algorithm (EA) in which populations of network configurations (i.e., sets of weights)
are evaluated in batches. The more successful configurations are propagated with the
Fig. 4 Physically based interactive character animation. (a) Pose Space Graph used to drive high-
level goals for kinematics solvers which direct joint movements (from (Laszlo and Fiume 1996));
(b) ambulatory motion of a creature and person learned by optimization processes mimicking nature
(from (Holden et al. 2016; Sims 1994)), respectively
subsequent batch and spliced with other promising configurations, to produce

batches of increasingly fitter networks (Yao 1999). In complex networks with
many weights and complex movements, it can be challenging for EAs to converge
during training. In such cases, weight configurations for the NN can be bootstrapped
by training the same network over simpler problems. This improves up white noise
initialization for more complex tasks. In practice, training a NN can take tens of
thousands of iterations to learn an acceptable controller (Sims 1994) for even very
simple movements. Yet, once learned the controller is trivial to evaluate quickly and
can be readily deployed into a real-time system. Even with bootstrapped training
however, NN cannot learn complex movement, and it was not until the recent advent
of more sophisticated (deeper) NNs that locomotion of a fully articulated body was
demonstrated using an entirely machine-learned motion controller (Holden et al.

2016).
Concatenative Synthesis
Motion concatenation is a common method for synthesizing interactive animation

without the complexity and computational expense of physical simulation. In a
concatenative synthesis pipeline, short fragments of motion capture are joined (and
often blended) together to create a single seamless movement. In the simplest
example, a single walk cycle may be repeated with appropriately chosen in-out
points to create a perpetual gait. A more complex example may concatenate walk
cycles turning slightly left, slightly right, or advancing straight-ahead to create
locomotion along an arbitrary path.
Skeletal Motion Graphs

Concatenative synthesis is dependent on the ability to seamlessly join together pairs
of pre-captured motion fragments – subsequences of performance capture – to build
complex animations. An initial step when synthesizing animation is therefore to
identify the transition points within performance captured footage, at which motion
fragments may be spliced together. Typically the entire capture (which may in
practice consist of several movements, e.g., walking, turning) is considered as a
single long sequence of t = [1, N] frames, and pairs of frames {1..N, 1..N} identified
that could be transitioned between to without the view perceiving a discontinuity.
A measure of similarity is defined, computable from and to any time instant in the
sequence, and that measure thresholded to identify all potential transition points.
Figure 5 visualizes both the concept and an example of such a comparison computed
exhaustively over all frames of a motion capture sequence – brighter cells indicating
closer matching frame pairs.
Measures of Pose Similarity Pose similarity measures (which, in practice, often

compute the dissimilarity between frames) should exhibit three important properties:
1. Be invariant to global rotation and translation – similar poses should be identified

as similar regardless of the subject’s position in world space at both time instants.
Otherwise, few transition points will be detected.
2. Exhibit spatiotemporal consistency – poses should not only appear similar at the
pair of time instants considered but also move similarly. Otherwise, motion will
appear discontinuous.
3. Reflect the importance of certain joints over others. Otherwise, a difference in
position of, e.g., a finger might outweigh a difference in position of a leg.
Common similarity measures include direct comparison of joint angles
(in quaternion form) or, more commonly, direct comparisons of limb spatial
position in 3D. A set of 3D points p1..m is computed either from limb
end-points or from the vertices of a coarse mesh approximating the form of the
Hit
Stand Walk Jog
Fig. 5 An example of an animation (top) generated by a motion graph (left) comprising four
actions. A visualization of the inter-frame distance comparison used to compute a motion graph
(right)
model and a sum of squared differences used to evaluate the dissimilarity D( p, p0 )

between point sets from a pair of frames p and p0 at times t and t0 , respectively:
X
m 2
Dðp, p0 Þ ¼ minθ, x0 , z0 ωi pi Mθ, x0 , z0 p0i : (5)
i¼1
where |.| is the Euclidean distance in world space, pi is the ith point in the set, and
M is a Euclidean transformation best aligning the two sets of point clouds, via a
translation on the ground plane (z0, z0) and a rotation θ about the vertical ( y) axis – so
satisfying property (1). In order to embed spatiotemporal coherency (2), the score is
computed over point sets not just from a given pair of times {t,t0 } but for a k frame
window t 2k , t þ 2k . This is effectively a low-pass filter over time and explains the
blurred appearance of Fig. 5 (right). For efficiency, pair-wise scores are computed
and the resulting matrix low-pass filtered. The relative importance of each point
(associated with the limb from which the point was derived) is set manually via ωi
satisfying property (3).
Motion Graph Construction Local thresholding is applied to the resulting simi-

larity matrix, identifying nonadjacent frames (t,t0 ) that could be concatenated
together to produce smooth transitions according to properties (1–3). For example,
if the mocap sequence contains several cycles of a walk, it is likely that
corresponding points in the walk cycles (e.g., left foot down at the start of each
cycle) would be identified as transitions. Playing one walk cycle to this time t and
then “seeking” forward or backward by several hundred frames to the corresponding
time t0 in another walk cycle will not produce a visual discontinuity despite the
nonlinear temporal playback.
The “in” (t) and “out” (t0 ) frames of these transition points are identified and
represented as nodes in a graph structure (the motion graph). Edges in the graph
correspond to clips of motion, i.e., motion fragments between these frames in linear
time. Additional edges are introduced to connect the “in” and “out” frames of each
transition. Usually the pose of the performer at the “in” and “out” points differs
slightly, and so this additional edge itself comprises a short sequence of frames
constructed by interpolating the poses at “in” and “out,” respectively, e.g., using
quarternion-based joint angle interpolation.
Motion Path Optimization Random walks over the motion graph representation
can provide a useful visualization to confirm that sensible transitions have been
identified. However interactive character control requires careful planning of routes
through the motion graph, to produce movement satisfying constraints the most
fundamental of which are the desired end pose (and position in the world, pv), the
distance that the character should walk (dv), and the time it should take the character
to get there (tv). Under the motion graph representation, this corresponds to com-
puting the optimal path routing us from the current frame of animation (i.e., the
current motion capture frame being rendered) to a frame corresponding to the target
key-pose elsewhere in the graph. Since motion graphs are often cyclic, there is
potentially unbounded number of possible paths. The optimal path is the one
minimizing a cost function, expressed in terms of these four animation constraints
(Ctrans, Ctime, Cdist, and Cspatial):
CðPÞ ¼ Ctrans ðPÞ þ ωtime Ctime ðPÞ þ ωdist Cdist ðPÞ þ ωspace Cspace ðPÞ: (6)
Studying each of these terms in turn, the cost of a path P is influenced by Ctrans,
reflecting the cost of all performing animation transitions along the path P. Writing
the sequence of Nf edges (motion fragments) along this path as {fj} where j = [1, Nf],
this cost is a summation of the cost of transitioning at each motion graph node along
that path:
X
Ctrans ðPÞ ¼ N f 1D f j 7! f jþ1 , (7)
j¼1
where D ( fj 7! fj+1) expresses the cost of transitioning from the last frame of fj to the
first frame of fj+1, computing by measuring the alignment of their respective point
clouds p and p0 via D( p, p0 ) (Eq. 5).
The timing cost Ctime(P) is computed as the absolute difference between the target
time tv for the animation sequence and the absolute time time(P) taken to transition
along the path P:
Ctime ðPÞ ¼ jtimeðPÞ tv j: timeðPÞ ¼ N f Δt, (8)
where Δt is the time take to display a single frame of animation, e.g., Δt ¼ 25

1
for
25 frames per second.
Similarly, the cost Cdist( p) is computed as the absolute difference between the
target distance dv for the character to travel and the absolute distance traveled dist(P)
computed by summing the distance traveled for each frame comprising P.
X

Cdist ðPÞ ¼ jdistðPÞ tv j: distðPÞ ¼ N f 1P f j P f jþ1 , (9)
j¼1
where P is a 2D projection operation, projecting the 3D points clouds p and p0

corresponding to the end frame of fj and start frame of fj+1, respectively, to the 2D
ground (x z) plane and computing the centroid.
The final cost Cspatial is computed similarly via centroid projection of the anima-
tion end-point, penalizing a large distance between the target end-point of the
character and end-point arising from the animation described by P.

Cspace ðPÞ ¼ P f N f 1 pv : (10)
The three parameters ωtime, ωdist, and ωspace are normalizing weights typical
values of which are ωtime = 1/10, ωdist = 1/3, and ωspace = 1 (Arikan and Forsyth
2002).
Finding the optimal path Popt for a given set of constraints C. . . is found by
minimizing the combined cost function C(P) (Eq. 6):
Popt ¼ argmin C ðPÞ: (11)

P
An efficient approach using integer programming to search for the optimal path
that best satisfies the animation constraints can be found in Huang et al. (2009) and is
capable of running in real time for motion fragment datasets of several minutes. Note
that Ctrans is be precomputed for all possible motion fragment pairs, enabling
run-time efficiencies – the total transition cost for a candidate path P is simply
summed during search.
Surface Motion Graphs

Surface motion graphs (SMGs) extend the skeletal motion graph concept beyond
joint angles, to additionally consider the 3D volume of the performer. This is
important since the movement of 3D surfaces attached to the skeleton (e.g., hair or
flowing clothing) is often complex, and simple concatenation of pre-animated or
captured motion fragments without considering the movement of this surface geom-
etry can lead to visual discontinuities between motion fragments.
Consideration of surfaces, rather than skeletons, requires the motion graph
pipeline to change only in two main areas. First, the definition of frame similarity,
i.e., Eq. 5 must be modified to consider volume rather than joint positions. Second,
the algorithm for interpolating similar frames to create smooth transitions must be
substituted for a surface interpolation algorithm.
Fig. 6 Visualization of a spherical histogram computed from a character volume. Multiple video
views (left) are combined to produce a volumetric estimate of the character (middle) which is
quantized into a spherical (long-lat) representation at multiple radii from the volume centroid
3D Shape Similarity To construct a SMG, an alternative measure of frame simi-

larity using 3D surface information is adopted, reflecting the same three desirable
properties of similarity measures outlined in subsection “Skeletal Motion Graphs.”
A spherical histogram representation is calculated from the 3D character volume
within the frame. The space local to the character’s centroid is decimated into
sub-volumes, divided by equispaced lines of longitude and latitude – so yielding a
2D array encoding the histogram that encodes the volume occupied by the character.
Spherical histograms are computed over a variety of radii, as depicted in Fig. 6
(right) yielding a three-dimensional stack of 2D spherical histograms.
The SMG is computed as with skeletal motion graphs, through an optimization
process that attempts to align each video to every other – resulting in a matrix of
similarity measurements between frames. The similarity between the spherical
histograms Hr (.) at radius r of the 3D character meshes Qa and Qb is computed by:
1X R

DðQa , Qb Þ ¼ min ωr H r ðQa , 0Þ H r Qb , ϕ , (12)
ϕ R r¼1
where H(x, φ) indicates a spherical histogram computed over a given mesh x, rotated
about the y axis (i.e., axis of longitude) by φ degrees. In practice this rotation can be
performed by cycling the columns of the 2D histogram obviating any expensive
geometric transformations; an exhaustive search across φ = [0, 359] degrees is
recommended in Huang et al. (2009). The use of the model centroid, followed by
optimization for φ, fulfills property (1), i.e., rotational and translational invariance in
the comparison. The resulting 2D matrix of inter-frame comparisons is low-pass
filtered as before to introduce temporal coherence, satisfying property (2).
Weightings set for each radial layer of the spherical histogram ωr weight the
importance of detail as distance increases, satisfying (3).
Transition Generation Due to comparatively high number of degrees of freedom

on a 3D surface, it is much more likely that the start and end frames of a pair of
motion fragments fj and fj+1 selected on an optimal path Popt will not exactly match.
To mitigate any visual discontinuities on playback, a short transition sequence is
introduced to morph the former surface (Sj) into the latter (Sj+1). This transition
sequence is substituted in for small number of frames (L ) before and after the
transition point. Writing this time interval k = [L, L], a blending weight α(k) is
computed:
kþL
αðkÞ ¼ , (13)
2L
and a nonlinear mesh blending algorithm (such as the Laplacian deformation scheme
of Tejera et al. (2013)) is applied to blend S j 7! S j+1 weighted by al pha(k).
Parametric Motion Graphs

Parametric motion graphs (PMG) extend classical motion graphs (subsection “Skel-
etal Motion Graphs”) by considering not only the concatenation but also the
blending of motion fragments to synthesize animation. A simple example is a
captured sequence of a walk and a run cycle. By blending these two motion
fragments together, one can create a cycle of a walk, a jog, a run, or anything
in-between. Combined with the concatenation of cycles, this leads to a flexibility and
storage efficiency not available via classic methods – a PMG requires only a single
example of each kind of motion fragment, whereas a classical approach would
require pre-captured fragments of walks and runs at several discrete speeds.
Parametric extensions have been applied to both skeletal (Heck and Gleicher
2007) and surface motion graphs (Casas et al. 2013). Provided a mechanism exists
for interpolating a character model (joint angles or 3D surface) between two frames,
the method can be applied. Without loss of generality, we consider surface motion
graphs (SMGs) here.
SMGs assume the availability of 4D performance capture data, i.e., a single 3D
mesh of constant topology deforming over time to create character motion (subsec-
tion “Multi-View Human Pose Estimation”). We consider a set of N temporally
aligned 4D mesh sequences Q = {Qi(t)} for i = [1, N] motion fragments. Since
vertices are in correspondence, it is possible to interpolate frames from such
sequences directly by interpolating vertex positions in linear or piecewise linear
form. We define such an interpolating function b(Q, w) yielding an interpolated mesh
QB(t, w) at time t given a vector of weights w expressing how much influence each of
the meshes from the N motion fragments at time t should contribute to that interpo-
lated mesh.
QB ðt, wÞ ¼ bðQ, wÞ, (14)
where w = {wi} for normalized weights w [0, 1] that drives a mesh blending
function b(.) capable of combining meshes at above 25 frames per second for
interactive character animation.
Fig. 7 Three animations each generated from a parametric motion graph (Casas et al. 2013). First
row: walk to run (speed control). Second row: short to long horizontal leap (distance, i.e., leap
length control). Third row: short to high jump (height control). Final row: animation from a
parametric motion graph embedded within an outdoor scene under interactive user control (Casas
et al. 2014) (time lapse view)
Several steps are necessary to deliver parametric control of the motion fragments:
time warping to align pairs of mesh sequences (which may different in length) so that
they can be meaningfully interpolated, the blending function b(.) to perform the
interpolation, and mapping between high-level “user” parameters from the motion
controller to low-level blend weights w. Considerations such as path planning
through the graph remain, as with classical motion graphs, but must be extended
since the solution space now includes arbitrary blendings of motion fragments, as
well as the concatenated of those blended fragments. Exhaustively searching this
solution space is expensive, motivating real-time methods to make PMG feasible for
interactive character animation.
Mesh Sequence Alignment Mesh sequences are aligned using a continuous time
warping function t = f(tu) where the captured timebase tu is mapped in nonlinear
fashion to a normalized range t = [0, 1] so as to align poses. The technique is
described in Witkin and Popovic (1995). Although coarse results are obtainable
without mesh alignment, failure to properly align sequences can lead to artifacts such
as foot skate.
Real-Time Mesh Blending Several interpolation schemes can be employed to

blend a pair of captured poses. Assuming a single mesh has been deformed to
track throughout the 4D performance capture source data (i.e., all frames have
constant topology), a simple linear interpolation between 3D positions of
corresponding vertices is a good first approximation to a mesh blend. Particularly
in the presence of rotation, such approximations yield unrealistic results. A high-
quality solution is to use differential coordinates, i.e., a Laplacian mesh blend
(Botsch and Sorkine 2008); however solution of the linear system comprising a
3v 3v matrix of vertex positions, where v is of the order 105, is currently
impractical for interactive animation. Therefore a good compromise can be obtained
using a piecewise linear interpolation (Casas et al. 2013), which precomputes offline
a set of nonlinear interpolated meshes (e.g., via Botsch and Sorkine (2008)), and any
requested parametric mesh is produced by weighted linear blending of the closest
two precomputed meshes. The solution produces more realistic output, in general,
that linear interpolation at the same computational cost.
High-Level Control High-level parametric control is achieved by learning a map-

ping function f(w) between the blend weights w and the high-level motion param-
eters p, e.g., from the motion controller. A mapping function w = f1( p) is learned
from the high-level parameter to the blend weights required to generate the desired
motion. This is necessary as the blend weights w do not provide an intuitive
parameterization of the motion. Motion parameters p are high-level user-specified
controls for a particular class of motions such as speed and direction for walk or run
and height and distance for a jump. The inverse mapping function f1 from param-
eters to weights can be constructed by a discrete sampling of the weight space w and
evaluation of the corresponding motion parameters p.
Parametric Motion Planning PMGs dispense of the notion of precomputed tran-

sition points, since offline computation of all possible transition and blend possibil-
ities between, e.g., a pair of mesh sequences would yield an impractical number of
permutations to permit real-time path finding.
We consider instead a continuous “weight-time” space with the weight modeling
the blend between one mesh sequence (e.g., a walk) and another (e.g., a run). We
consider motion planning as the problem of finding a route through this space, taking
us from a source time and pose (i.e., weight combination) to a target time and pose.
Figure 8 illustrates such a route finding process. The requirement for smooth motion
dictates we may only modify the weight or time in small steps, yielding to a “fanning
out” or trellis of possible paths from the source point in weight-time space. The
ws wd
Qsi Qdi
1 1
Msj Mdj
0 0
0 t 1 0 t 1
source parametric space target parametric space
Fig. 8 Real-time motion planning under a parametric motion graph. Routes are identified between
trellises fanned out from the source pose Qs and end pose Qd. The possible (red) and optimal (green)
paths are indicated (illustration only)
optimal path Popt between two parametric points in that space is that minimizing a
cost function balancing mesh similarity ES(P) and time taken, i.e., latency EL(P) to
reach that pose:
Popt ¼ argmin ES ðPÞ þ λEL ðPÞ, (15)

PΩ
where λ defines the trade-off between transition similarity and latency. The transition
path P is optimized over a trellis of frames as in Fig. 7 starting at frame Qs(ts, ws)
ending at Qd(td wd) where Qs and Qd are interpolated meshes (Eq. 14).
The trellis is sampled forward and backward in time at discrete intervals in time Δt
and parameters Δw up to a threshold depth in the weight-time space. This defines a
set of candidate paths P comprising the transitions between each possible pair of
frames in the source and target trellis.
For a candidate path P, the latency cost EL(P) is measured as the number of frames
in the path P between the source and target frames. The transition similarity cost
ES(P) is measured as the similarity in mesh shape and motion at the transition point
between the source and target motion space for the path P, computable via Eq. 12 for
mesh data (or if using purely skeletal mocap data, via Eq. 5). Casas et al. (2012)
proposed a method based on precomputing a set of similarities between the input
data and interpolating these at run-time to solve routing between the two trellises at
interactive speeds. Figure 7 provides examples of animation generated under this
parametric framework.
Summary and Future Directions
This chapter has surveyed techniques for interactive character animation, broadly
categorizing these as either data driven or physical simulation based. Arguably the
major use cases for interactive character animation are video games and immersive
virtual experiences (VR/AR). In these domains, computational power is at a pre-

mium – developers must seek highly efficient real-time algorithms, maintaining high
frame rates (especially for VR/AR) without compromising on animation quality.
This has led interactive animation to trend toward data-driven techniques that
sample, blend, and concatenate fragments of performance capture rather than
spend cycles performing expensive online physical simulations.
This chapter has therefore focused upon synthesis techniques that sample, con-
catenate, and blend motion fragments to create animation. The chapter began by
surveying commercial technologies and academic research into performance capture.
Although commercial systems predominantly focus upon skeletal motion capture,
research in 4D performance capture is maturing toward practical solutions for
simultaneous capture of skeletal and surface detail. The discussion of motion graphs
focused upon the their original use for skeletal data and their more recent extensions
to support not only 4D surface capture but also their parametric variants that enable
blending of sequences in addition to their concatenation. Physical simulation-based
approaches for character animation were examined within the context of interactive
animation, deferring broader discussion of this topic to chapter C-2.
Open challenges remain for interactive character animation, particularly around
expressivity and artistic control. Artistic directors will often request editing of
animation to move in a particular style (jaunty, sad), adjustments that can be
performed manually by professional tools such as Maya (Autodesk) or
MotionBuilder (Vicon) but that cannot be applied automatically in an interactive
character animation engine. While work such as Brand et al.’s Style Machines
(Brand and Hertzmann 2000) enables stylization of stand-alone skeletal mocap
sequences, algorithms have yet to deliver the ability to modulate animation interac-
tively, e.g., to react to emotional context in a game. An interesting direction for
future research would be to integrate stylization and other high-level behavioral
attributes into the motion graph optimization process.
Cross-References
▶ Biped Controller for Character Animation

▶ Blendshape Facial Animation
▶ Data-Driven Character Animation Synthesis
▶ Data-Driven Hand Animation Synthesis
▶ Depth Sensor Based Facial and Body Animation Control
▶ Example-Based Skinning Animation
▶ Eye Animation
▶ Hand Gesture Synthesis for Conversational Characters
▶ Head Motion Generation
▶ Laughter Animation Generation
▶ Physically-Based Character Animation Synthesis
▶ Real-Time Full Body Motion Control
▶ Real-Time Full Body Pose Synthesis and Editing
▶ Video-Based Performance Driven Facial Animation

▶ Visual Speech Animation
References
Agarwal A, Triggs B (2006) Recovering 3d human pose from monocular images. IEEE Trans PAMI
28(1):44–58
Andriluka M, Roth S, Schiele B (2009) Pictorial structures revisited: people detection and articu-
lated pose estimation. In: Proceedings of computer vision and pattern recognition, IEEE
Arikan O, Forsyth D (2002) Synthesizing constrained motions from examples. ACM Trans Graph
21(3):483–490
Armstrong W, Green M (1985) The dynamics of articulated rigid bodies for purposes of animation.
Vis Comput 4(1):231–240
Botsch M, Sorkine O (2008) On linear variational surface deformation methods. IEEE Trans Vis
Comput Graph 14(1):213–230
Brand M, Hertzmann A (2000) Style machines. In: Proceedings of ACM SIGGRAPH, ACM Press,
pp 183–192
Budd C, Huang P, Klaudiny M, Hilton A (2013) Global non-rigid alignment of surface sequences.
Int J Comput Vis 102(1):256–270
Casas D, Tejera M, Guillemaut JY, Hilton A (2012) 4d parametric motion graphs for interactive
animation. In: Proceedings of Symposium on Interactive 3D Graphics and Games (I3D), IEEE
Casas D, Tejera M, Guillemaut JY, Hilton A (2013) Interactive animation of 4d performance
capture. IEEE Trans Vis Comput Graph (TVCG) 19(5):762–773
Casas D, Volino M, Collomosse J, Hilton A (2014) 4d video textures for interactive character
appearance. In: Proceedings of Computer Graphics Forum (Eurographics 2014), IEEE
Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: Proceedings of
Computer Vision and Pattern Recognition, IEEE, vol 3. pp 886–893
Eichner M, Ferrari V (2009) Better appearance models for pictorial structures. In: Proceedings of
British Machine Vision Conference (BMVC), IEEE
Elhayek A, Aguiar E, Tompson J, Jain A, Pishchulin L, Andriluka M, Bregler C, Schiele B,
Theobalt C (2015) Efficient convnet-based marker-less motion capture in general scenes with
a low number of cameras. In: Proceedings of computer vision and pattern recognition, IEEE
Geijtenbeek T, Bogert AJVD, Basten B, Egges A (2010) Evaluating the physical realism of
character animations using musculoskeletal models. In: Proceedings of Conference on Motion
in Games, vol 5. Springer-Verlag, Heidelberg, pp 11–22
Grauman K, Shakhnarovich G, Darrell T (2003) A Bayesian approach to image-based visual hull
reconstruction. In: Proceedings of CVPR, IEEE
Heck R, Gleicher M (2007) Parametric motion graphs. In: Proceedings of Symposium on Interac-
tive 3D Graphics and Games (I3D), IEEE, pp 129–136
Hodgins J (1991) Biped gait transition. In: Proceedings of Conference on robotics and automation,
IEEE, pp 2092–2097
Holden D, Saito J, Komura T (2016) A deep learning framework for character motion synthesis and
editing. In: Proceedings of ACM SIGGRAPH, ACM
Huang P, Hilton A, Starck J (2009) Human motion synthesis from 3d video. In: Proceedings of
CVPR, IEEE
Kovar L, Gleicher M, Pighin F (2002) Motion graphs. ACM Trans Graph 21(3):473–482
Kwon T, Hodgins J (2010) Control systems for human running using an inverted pendulum model
and a reference motion capture sequence. In: Proceedings of Eurographics Symposium on
Computer Animation (SCA), Blackwell
Laszlo J, Fiume EVD (1996) Limit cycle control and its application to the animation of balancing
and walking. ACM Trans on graphics, ACM
Lorensen W, Cline H (1987) Marching cubes: a high resolution 3d surface construction algorithm.
ACM Comput Graph 21(4):163–169
Mori G, Ren X, Efros A, Malik J (2004) Recovering human body configurations: combining
segmentation and recognition. In: Proceedings of computer vision and pattern recognition,
IEEE, pp 326–333
Ning H, Xu W, Gong Y, Huang T (2008) Discriminative learning of visual words for 3D human
pose estimation. In: Proceedings of CVPR, IEEE
Pollack J, Lipson H, Ficici S, Funes P (2000) Evolutionary techniques in physical robotics. In:
Proceedings of International Conference Evolvable Systems (ICES). Springer-Verlag
Raibert M, Hodjins K (1991) Animation of dynamic legged locomotion. ACM Comput Graph 25
(4):349–358
Ren X, Berg E, Malik J (2005) Recovering human body configurations using pairwise constraints
between parts. In: Proceedings of International Conference on computer vision, IEEE, vol 1. pp
824–831
Rohan A (2015) 3D motion capture system market – global forecast to 2020. Tech. rep. Markets and
Markets Inc., Vancouver
Sims K (1994) Evolving virtual creatures. ACM Trans. on Graphics, ACM
Srinivasan P, Shi J (2007) Bottom-up recognition and parsing of the human body. In: Proceedings of
computer vision and pattern recognition, IEEE, pp 1–8
Tejera M, Casas D, Hilton A (2013) Animation control of surface motion capture. ACM Trans. on
Graphics, IEEE, pp 1532–1545
Tin K, Coros S, Beaudoin P (2008) Continuation methods for adapting simulated skills. ACM
Trans. on Graphics, IEEE, vol 27(3)
Trumble M, Gilbert A, Hilton A, Collomosse J (2016) Learning markerless human pose estimation
from multiple viewpoint video. In: Proceedings of ECCV workshops, ACM
Viola P, Jones M (2004) Robust real-time object detection. Int J Comput Vis 2(57):137–154
Wampler K, Popovic Z (2009) Optimal gait and form for animal locomotion. ACM Trans. on
Graphics, ACM, vol 28(3)
Witkin A, Popovic Z (1995) Motion warping. ACM Trans. on Graphics, ACM, pp 105–108
Yao X (1999) Evolving artificial neural networks. In: Proceedings of the IEEE, IEEE, vol 87
Zhao T, Nevatia R (2003) Bayesian human segmentation in crowded situations. In: Proceedings of
computer vision and pattern recognition, IEEE, vol 2. pp 459–466
Physically Based Character Animation
Synthesis
Jie Tan
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Physical Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Simulation in Maximal Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Simulation in Generalized Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Contact Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Simulation Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Motion Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Trajectory Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Reinforcement Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Improving Realism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Reducing Prior Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Bringing the Character to the Real World . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Abstract
Understanding and synthesizing human motions are an important scientific quest.
It also has broad applications in computer animation. Research on physically
based character animation in the last two decades has achieved impressive
advancement. A large variety of human activities are synthesized automatically
in a physically simulated environment. The two key components of physically
based character animation are (1) physical simulation that models the dynamics of
humans and their environment and (2) controller optimization that optimizes the
character’s motions in the simulation. This approach has an inherent realism
because we all live in a world that obeys physical laws, and we evolved to survive
J. Tan (*)
Georgia Institute of Technology, Atlanta, GA, USA
e-mail: jtan34@gatech.edu

DOI 10.1007/978-3-319-30808-1_11-1
2 J. Tan
in this physical environment. In this chapter, we will review the state of the art of
physically based character animation, introduce a few established methods in
physical simulation and motor control, and discuss promising future directions.
Keywords
Character animation • Physical simulation • Trajectory optimization • Reinforce-
ment learning
Introduction
Mother Nature has created a diverse set of awe-inspiring motions in the animal
kingdom: Birds can fly in the sky, fishes can swim in the water, geckos can crawl on
vertical surfaces, and cats can reorient themselves in midair. Similarly, human’s
motions exhibit efficiency (locomotion), agility (kung fu), gracefulness (ballet), and
dexterity (hand manipulation). Studying these motions is not only a scientific quest
that quenches our curiosity but also an important step toward synthesizing them in a
way that can fundamentally change our life. Character animation aims to faithfully
synthesize these motions of humans and animals and display them to an audience for
the purpose of entertainment, storytelling, and education. The synthesized motions
need to appear realistic to give the audience an immersive experience.
In the last few decades, we have seen tremendous advance in character animation.
Some of the most breathtaking movies, such as Harry Potter, Avatar, and Life of Pi,
rely heavily on computer-generated animations. Nowadays, it is almost impossible
for the audience to tell apart the computer-synthesized motions from the real footage.
Behind these realistic animations lie countless hours of tedious manual work of
highly specialized experts. For example, to produce a 100-min feature film at Pixar
can take dozens of artists and engineers more than 5 years of development. In today’s
animation pipeline, the most popular techniques are key frames or motion capture,
both of which require artistic expertise and laborious manual work. Even worse, the
knowledge and efforts that are put into one animation sequence are not necessarily
generalizable to other motions. In my point of view, these are not efficient or
principled ways of animation synthesis.
A principled way to synthesize character animation is to study the funda-
mental factors that have shaped our motions. Instead of focusing on the
appearance of our motions, we need to dig deeper to understand why we move
in the way that we are doing today. After understanding the root causes that have
shaped our movements, we can then synthesize them naturally and automatically.
Our motions are shaped through millions of years of optimization (evolution) in a
world that obey physical laws. This insight has motivated a new paradigm of
physically based character animation. The two key components of this paradigm
are physical simulation and motion control. We first build a physical simulation to
model the physical world and then perform optimization to control the motions of
characters so that they can move purposefully, naturally, and robustly in the
simulated environment.
Physically Based Character Animation Synthesis 3
Although we often take our motions for granted since we can perform them so
effortlessly, physically based character animation is a notoriously difficult problem
because our motions involve sophisticated neuromuscular control, sensory informa-
tion processing, motion planning, coordinated muscle activation, and complicated
interactions with the surrounding physical environment. Even though we are still far
from fully understanding the underlying control mechanisms that govern our
motions, two decades of research in physically based character animation has
brought us new insights, effective methodologies, and impressive results. The
purpose of this chapter is to review the state of the art (section “State of the Art”),
introduce some of the established algorithms (sections “Physical Simulation” and
“Motion Control”), and discuss promising future research directions (section “Future
Directions”) in physically based character animation.
State of the Art
Starting from the seminal work of Hodgins et al. (1995), controlling a physically
simulated human character has been extensively studied in computer animation. A
wide variety of human activities, including walking (Yin et al. 2007), running (Kwon
and Hodgins 2010), swimming (Kwatra et al. 2009; Si et al. 2014), biking (Tan
et al. 2014), dressing (Clegg et al. 2015), gymnastics (Hodgins et al. 1995), reacting
to perturbations (Wang et al. 2010), falling and landing (Ha et al. 2012) and
manipulating objects with hands (Liu 2009; Ye and Liu 2012; Bai and Liu 2014)
are realistically synthesized in physically simulated environments (Fig. 1).
Two widely used techniques in physically based character animation are trajec-
tory optimization and reinforcement learning. Trajectory optimization formulates a
constrained optimization to minimize a task-related objective function subject to
physical constraints. It has been applied to control the iconic jumping Luxo Jr lamp
(Witkin and Kass 1988), humanoid characters (Liu and Popovic´ 2002; Jain
et al. 2009; Ye and Liu 2010), and characters with arbitrary morphologies (Wampler
and Popovic´ 2009). The resulting motions are physically plausible and follow the
animation principles such as anticipation and follow-through (Thomas and Johnston
1995).
Reinforcement learning algorithms solve a Markov decision process (MDP) to
find optimal actions at different states. When the MDP has moderate dimensions,
(fitted) value function iteration has been successfully applied to generalize motion
capture data (Treuille et al. 2007; Levine et al. 2012), to carry out locomotion tasks
(Coros et al. 2009), and to manipulate objects with hands (Andrews and Kry 2013).
When the dimensionality is high, policy search (Ng and Jordan 2000) can directly
search for a control policy without the need to construct a value function. Many
studies on locomotion control (Yin et al. 2008; Wang et al. 2009, 2012; Coros
et al. 2011; Geijtenbeek et al. 2013) performed policy search on parameterized
controllers.
Although we have seen impressive advances for the last two decades, the
gracefulness, agility, and versatility of real human motions remain unmatched.
4 J. Tan
Fig. 1 Various human activities, such as running, swimming, dressing, performing bicycle stunts,
interacting with the environment, and manipulating clothes, are modeled in a physically simulated
environment (Image courtesy of (Hodgins et al. 1995; Si et al. 2014; Clegg et al. 2015; Tan
et al. 2014; Coros et al. 2010; Bai and Liu 2014))
There are challenges in physically based character animation that need further
investigation. First, controlling balance is a key problem of synthesizing human
motions in a physically simulated environment. Balance can be maintained by
exerting virtual forces (Pratt et al. 2001; Coros et al. 2010), applying linear feedback
(Laszlo et al. 1996; Yin et al. 2007; da Silva et al. 2008; Coros et al. 2010), using
nonlinear control policies (Muico et al. 2009), planning the contact forces (Muico
et al. 2009; Tan et al. 2012b), employing reduced models (Tsai et al. 2010; Kwon and
Hod- gins 2010; Mordatch et al. 2010; Coros et al. 2010; Ye and Liu 2010), and
training in stochastic environments (Wang et al. 2010). Although the balance
problem in simple locomotion tasks, such as walking and running, has been solved,
maintaining balance in tasks that require agile motions remains an open problem.
Another challenge is to effectively plan the contacts. We human can only move
ourselves and other objects through contacts. However, contact events (contact
breakage, sliding, etc.) introduce unsmooth forces to the dynamics, which breaks
the control space into fragmented feasible regions. As a result, a small change in
control parameters can easily generate bifurcated consequences. For this reason,
many previous methods explicitly assumed that the contacts remain static (Abe
et al. 2007; Jain et al. 2009; Kim and Pollard 2011) while optimizing controllers.
This assumption significantly restricts the effectiveness of the controller because the
controller is not allowed to actively exploit contact breakage, slipping contacts, or
rolling contacts to achieve control goals. Three promising research directions to
tackle this challenge are contact-invariant optimization (Mordatch et al. 2012, 2013),
QPCC (Tan et al. 2012b) and policy search with stochastic optimization (Wu and
Popovic´ 2010; Wang et al. 2010; Mordatch et al. 2010).
An important criterion in character animation is the realism of the synthesized
motions. There is still large room to improve the quality of physically based
character animation. One possible cause of the unnatural motions is the vast simpli-
fication of the human models. To improve the realism, prior work has simulated the
dynamics of muscles and demonstrated complex interplay among bones, muscles,
ligaments, and other soft tissues for individual body parts, including the neck (Lee
and Terzopoulos 2006), upper body (Zordan et al. 2006; DiLorenzo et al. 2008; Lee
et al. 2009), lower body (Wang et al. 2012), and hands (Tsang et al. 2005; Sueda
et al. 2008). However, constructing such a sophisticated biological model for a full-
human character is computationally prohibitive. An alternative solution is to aug-
ment a physically controlled character with realistic motion capture streams (da Silva
et al. 2008; Muico et al. 2009; Liu et al. 2010).
Physical Simulation
Physically based character animation consists of two parts, simulation and control.
This section concentrates on the simulation part while the next section on control.
Although the majority of research in physically based character animation
focuses on control, a good understanding of physical simulation is essential for
designing effective controllers because complex human behaviors often require
sophisticated controllers that exploit the dynamics of a multi-body system.
In physically based character animation, a human character is often represented as
an articulated rigid-body system (Fig. 2 left), a group of rigid bodies chained
together through rotational joints. These joints can have different number of degrees
of freedom (DOF). For example, the shoulder is a ball joint (three DOFs), the wrist is
a universal joint (two DOFs), and the elbow is a hinge joint (one DOF). In some
cases, if the character’s motion involves dexterous hand manipulation, a detailed
hand model (Fig. 2 right) is attached to each wrist. Note that the articulated rigid-
body system is a dramatic but necessary simplification since simulating each bone,
muscle, and tendon that a real human has would require a prohibitively huge amount
of computational resources. In the simulation, the articulated figure is represented as
a tree structure. Each node is a rigid body and each edge is a joint. One node can have
multiple children but at most one parent. The root node has no parent. While any
body can be selected as the root node, a common choice is to use the torso as the root.
In this tree structure, loops are not allowed. Although it is possible to simulate loops,
such case is rare in character animation and will not be discussed here. There are two
major methods to simulate the dynamics of an articulated rigid-body system:
simulation in maximal coordinates (Cartesian space) and simulation in generalized
coordinates (joint space).
6 J. Tan
Fig. 2 Articulated figures in character animation to represent a human character (left) and a hand
(right)
Simulation in Maximal Coordinates
In maximal coordinates, the physical states of the articulated figures are defined for
each node (rigid body). Each body has six degrees of freedom: three translational
and three rotational. The dynamics of each rigid body is considered independently. A
list of additional joint constraints is imposed to ensure that the two adjacent bodies
will stick together at the joint location. The dynamic equation for each body is

m I33 0 v_ mg f 0
¼ _ þ þ a (1)
0 I ω_ Iω τ τ
where m and I are the mass and the inertia tensor of the body; I33 is a 3 3 identity
matrix; v and ω are the linear and angular velocities; [f, τ]T are the passive forces and
torques from joint constraints, contacts, gravity, and other external sources; and τ a
are the torques actively exerted by the controllers, which is the focus of section
“Motion Control.”
Joints that connect two rigid bodies constrain their relative motions. Different
number of constraints is imposed according to the type of joints. For example, a
hinge joint has only one DOF. Thus it has five constraints that eliminate all but the
rotation along the hinge axis. A ball joint has three DOFs. Its constraints eliminate
the relative translation at the joint location.
Suppose a joint connects body A and body B, the translational constraints are:

I33 vA I33 vB
¼0
½rA ωA ½rB ωB
where [r] is the skew symmetric matrix of r, which is the vector from the center of
mass (COM) of the body to the joint location. The rotational constraints are
dTi ðωA ωB Þ ¼ 0
where di is an axis perpendicular to the rotational DOFs and i could be a subset of {0,
1, 2} depending on the type of joints.
To allow a character to actively control its motion, actuators are attached to the
joints. According to Newton’s third law, the two bodies connected to a common
actuator should receive equal and opposite joint torques.
τaA τaB ¼ 0
where τA and τB are the torques exerted by the actuator to body A and B, respectively.
Although simple to understand and implement, simulating characters in maximal
coordinates has a few drawbacks. First, the state representation is redundant. It is not
efficient to use all six DOFs of a rigid body and then eliminate most of them with
joint constraints. Second, the accumulating numerical errors in simulation would
cause the joint constraints not satisfied exactly. Eventually, joints will dislocate and
adjacent bodies can drift apart. Both of these shortcomings can be overcome by
simulation in generalized coordinates.
Simulation in Generalized Coordinates
In generalized coordinates, the physical states q, q_ of the articulated figure, are

defined on the edges of the tree (joint angles). Note that the root node is attached to
the world space via a 6-DOF joint that can translate and rotate freely. Each DOF is
one component of q. The number of DOFs equals the dimensionality of q. In other
words, there is no redundancy in this representation.
The dynamic equation in generalized coordinates for an articulated rigid-body
system is
MðqÞq€ þ Cðq,q_ Þ ¼ Q þ τa (2)
where q, q_ , and q€ are the position, velocity, and acceleration in generalized

coordinates; M(q) is the mass matrix; C(q, q_ ) accounts for the Coriolis and
centrifugal force; Q is the external generalized force, including gravity and contact
force; and τa is the generalized force exerted by the controller (section “Motion
Control”). This equation can be derived from Lagrangian dynamics. Detailed deri-
vation is omitted here but can be found in Liu and Jain (2012).
When the articulated rigid bodies are simulated in generalized coordinates, it is
often necessary to convert physical quantities back and forth between generalized
and maximal coordinates. For example, we need to compute the velocity at certain
point on the articulated figure in Cartesian space for collision detection. We also need
8 J. Tan
to convert the forces from the Cartesian space to generalized coordinates to apply
contact forces. Jacobian matrix J bridges these two different coordinate systems.
@x
J¼ (3)
@q
It represents the relation how much a point x moves in the Cartesian space if the
joint angles q changes slightly. Here are the two most frequently used formulas that
convert velocities and forces, and more conversion formulas can be found in Liu and
Jain (2012).
v ¼ Jq_
Q ¼ JT f
Simulation in generalized coordinates is widely used in physically based charac-

ter animation. Although it takes more effort to walk through the derivation, it has
important advantages over simulation in maximal coordinates. Apparently, the
representation is more compact. There is no redundancy and thus no need to use
constraints to eliminate redundant states. More importantly, it ensures that the joints
are satisfied exactly. Two connecting bodies can never drift apart even with numer-
ical errors because the states of dislocated joints are not part of the state space in
generalized coordinates.
Contact Modeling
Most of our daily activities, such as locomotion and hand manipulation, involve
interacting with our surrounding environments through contacts. Accurately simu-
lating contacts and computing contact forces are crucial to physically based character
animation. Penalty method and linear complementarity problem (LCP) are two
widely used methods to model contact.
Penalty Method
When a body A penetrates another body B, a repulsive penalty force fc is exerted to
separate these two bodies.

kdn if d > 0,
fc ¼ (4)
0 otherwise:
where k is the stiffness, d is the penetration depth, and n is the contact normal.
Penalty method is trivial to implement. However, to make it work properly,
tedious manual tuning is often needed. While too small k cannot effectively stop
the penetration, too large k would lead to undesired bouncy collision response. Even
worse, when simulating with large time steps, penalty method could make the
simulation unstable. In addition, it is not clear how to accurately model static friction
using penalty methods.
Linear Complementarity Problem

Linear complementarity problem (LCP) is a more accurate and stable method to
model contacts. A contact force fc can be decomposed into the normal and the
tangential (frictional) forces.
f c ¼ f ⊥ n þ Bf jj
where n is the contact normal; f⊥ and fk are the normal and tangential components,
respectively; and B is a set of bases that span the tangential plane (Fig. 3). The more
bases bi are used, the more accurate the approximation of the friction cone is, but
more computation is needed to solve the resulting LCP.
LCP imposes a set of constraints to satisfy the conditions of Coulomb friction:
1. In the normal direction, only repulsive forces are exerted to stop penetration.
2. In a static contact situation, the contact force lies within the friction cone.
3. In a sliding contact situation, the contact force lies at the boundary of the friction
cone and the friction direction is opposite to the sliding direction.
I will illustrate the concept of LCP using the formulation in the normal direction.
The formulation in the tangential directions is beyond the scope of this chapter. It can
be found in the following tutorials (Lloyd 2005; Tan and et al. 2012a).
In a physical simulation, after the collisions are resolved, the relative velocity
between the contact points of two colliding bodies can only be either zero (resting) or
positive (separating), but not negative (penetrating):
v⊥ 0 (5)
Fig. 3 A linearized friction cone used in LCP formulation. Left: a foot is in contact with the ground.
Right: the friction cone at the contact point. n is the contact normal, and bi are a set of tangential
bases
10 J. Tan
Similarly, the normal contact force can be zero (no force) or positive (repulsive
force), but not negative (sticking force):
f⊥ 0 (6)
The repulsive normal force exists (f ⊥ > 0) if and only if the two bodies are in
contact (v⊥ ¼ 0). In contrast, when they are separating (v⊥ > 0), there is no contact
force (f ⊥ ¼ 0). In other complementarity condition needs to be satisfied:
v⊥ f ⊥ ¼ 0 (7)
Combining the dynamic equations (Eq. 1 or 2) and the LCP constraints (Eqs. 5, 6,
and 7) forms a mixed LCP problem. It can be solved efficiently by direct (Lloyd
2005) or iterative solvers (Erleben 2007; Kaufman et al. 2008; Otaduy et al. 2009).
Simulation Software
There is a growing need for simulation software that can accurately simulate the
complex dynamics of virtual humans and their interactions with their surrounding
environment. A number of open-source physical simulators are readily available for
research in physically based character animation. The popular ones include Open
Dynamic Engine (ODE) (http://www.ode.org/), Bullet (http://www.bulletphysics.
org/), Dynamic Animation and Robotics Toolkit (DART) (http://dartsim.github.io/)
and MuJoCo (http://www.mujoco.org). All of them can simulate articulated rigid
bodies with an LCP-based contact model in real time. These simulators allow the
user to specify the structure of the articulated figure, the shape and the physical
properties of each body, the type of joints, and other parameters describing the
environment. Different simulators may offer different features, speed, and accuracy.
Erez et al. (2015) provided an up-to-date review and an in-depth comparison among
these modern physical engines.
Motion Control
We human can carefully plan our motion, coordinately and purposefully, and
exercise our muscles to achieve a wide variety of high-level tasks, ranging from
simple locomotion, dexterous hand manipulation, to highly skillful stunts. To model
them in animation, simulating the passive dynamics is not enough. The key chal-
lenge that motion control tackles is to find controllers that can achieve high-level
motion tasks (e.g., walk at 1 m/s, grasp a bottle and open the cap, etc.). In character
animation, a controller is the character’s “algorithmic brain” that decides how much
torque (τ a in Eqs. 1 and 2) are needed at each joint to successfully fulfill the task in a
way that mimics human behavior. Optimization-based motion control is the most
extensively researched topic in physically based character animation. The
Fig. 4 Different stages of

walking in SIMBICON
(Image courtesy of (Yin left stance
et al. 2007))
0.3 s
0 1
left foot right foot

strike strike
3 2
0.3 s
right stance
optimization searches for a controller that minimizes a task-related cost function,

subject to dynamical constraints.
One common misunderstanding is that one can formulate a large optimization for
arbitrary tasks. Due to complexity of human motions and nonlinearity of the
dynamics, a large optimization may have competing objectives and many local
minima. Up till today, there are no efficient optimization algorithms that can reliably
find meaningful controllers in such cases. For this reason, a common practice in this
field is to decompose a high-level task into multiple lower-level subtasks and
formulate smaller optimization for each of the simpler subtask. For example, in
SIMBICON (Yin et al. 2007), a walking cycle is decomposed into multiple stages
(Fig. 4). Within each stage, separate optimizations can be used for controlling the
two legs, the upper body, the balance, and the style. After solving all the optimiza-
tions, these low-level controllers can be combined so that the character can walk
naturally and robustly. Controller decomposition depends on the task and requires
domain knowledge. We refer the readers to the research literature to learn controller
decomposition on a case-by-case basis. In this section, we will discuss two generic
optimization-based methods of motion control: trajectory optimization and rein-
forcement learning.
Trajectory Optimization
Starting from the classical paper “Spacetime Constraints” (Witkin and Kass 1988),
trajectory optimization has become a mainstream technique in physically based
character animation. It searches for a controller that minimizes a cost function
subject to physical constraints. The general form of the optimization is
12 J. Tan
minx, u XN
gðxt , ut Þ
subject to t¼0 (8)
xtþ1 ¼ hðxt Þ þ Bt ut
where x is the physical states and u is the control. In character animation, the states
are usually defined as x :¼ ½q,q_ T , and the control u :¼ τa . g is the cost function,
which is handcrafted to reflect the high-level task. For example, if the task is to walk
at 1 m/s, one term in the cost function could be the distance between the current
COM of the character and a desired COM position that moves at 1 m/s. The
constraint usually consists of the dynamic equation h. Note that in most applications
of character animation, the dynamics are nonlinear to the state x but linear to the
control u (see Eqs. 1 and 2). In addition, the constraints can also include joint limits,
torque limits, and other task-related requirements.
To make it more concrete, we will revisit the simple example in the original
“Spacetime Constraints” paper: controlling a single particle. The task of this particle
is to fly from point a to point b in T seconds using a time-varying jet force f(t). The
dynamics of the particle is m€ x f mg ¼ 0, where x is its position, m is its mass,
and g is gravity. The goal of this flight is to minimize the total fuel consumption
ÐT 2
0 jfj dt. After discretization along time, the optimization has the following form:
minx, f XN
jf t j 2
subject to t¼0
xtþ1 ¼ 2xt xt1 þ Δtm2 f t þ Δt2 g

x0 ¼ a
xN ¼ b
It is not too difficult to extend the above derivations to control a human character.
We will need to change the control force f(t) to joint torques τa (t), the physical
constraint to the dynamic equation of articulated rigid bodies (Eq. 1 or 2), and the
cost function to a relevant function specific to the task.
There are different options to solve the optimization according to the structure
of the problem. Assuming the cost function and the dynamic equations are smooth,
Witkin and Kass (1988) applied a generic nonlinear optimizer, sequential qua-
dratic programming (SQP), to solve the problem. It is an iterative optimization
technique that solves a sequence of quadratic programs that approximate the
original problem. The solution of SQP is an optimal trajectory of state x(t) and
control u(t). Note that this method produces a feedforward (open-loop) controller:
a trajectory over time. It cannot be generalized to the neighboring regions of the
state space. As a result, the controller will fail the task even with a slight
disturbance to the motion.
When the cost function is quadratic and the dynamic equation is linear,
minx, u XN
xTt Qt xt þ uTt Rt ut
subject to t¼0 (9)
xtþ1 ¼ At xt þ Bt ut
the trajectory optimization is called an LQ problem. This problem can be solved very
efficiently by linear-quadratic regulator (LQR). The derivation of LQR can be found
in most of optimal control textbooks (Todorov 2006). Thus we will not repeat it here.
The solution is a feedback (close-loop) controller ut ¼ Kt xt. Although the require-
ment of linear dynamics seems restrictive, LQR still plays an important role in
physically based character animation. One important application is to design a
physically based controller to track motion captured data, which is an effective
way to increase the realism of the synthesized motions. Given a motion capture
sequence x, we can linearize the dynamic equation at its vicinity:
@h
Δ xtþ1 ¼ Δ xt þ Bt ut þ hðxt Þ xtþ1
@x
where Δ x ¼ x x . This gives an LQ problem that seeks a feedback controller
ut = KtΔxt that minimizes the difference between the actual and the reference
motion over the entire trajectory.
More importantly, LQR is a building block to solve the more general trajectory
optimization problem (Eq. 8). Given an initial trajectory x0, u0, x1, u1, . . ., uN, xN, we
can perform the following steps iteratively:
1. Compute the LQ approximation of the original problem (Eq. 8) around the current
trajectory by computing a first-order Taylor expansion of the dynamics and a
second-order expansion of the cost function.
2. Use LQR to solve the LQ approximation to get an optimal controller.
3. Apply the current optimal controller to generate a new trajectory.
4. Go to step 1 until convergence.
This iterative-LQR process is similar to the core idea behind differential dynamic
programming (DDP). We refer the interested reader to Todorov (2006) for a more
thorough discussion about LQR and DDP. The key advantage of DDP is that it not
only provides a feedforward trajectory, but also an optimal feedback controller near
that trajectory.
To sum up, trajectory optimization is an effective way to synthesize character
animation. The synthesized motion is not only physically correct but, more impor-
tantly, demonstrates an important animation principle, anticipation. This is because
the objective of trajectory optimization is to minimize a long-term cost. Thus, the
character can move intelligently that gives up short-term gains to minimize the long-
term cost. For example, for a jump-up task, the trajectory optimization could produce
a controller that sacrifices the height of the character’s COM first for a much higher
jump later. However, there are a few shortcomings of trajectory optimization. First, it
often leads to a high-dimensional optimization that is expensive to solve. Another
14 J. Tan
problem of high-dimensional nonlinear optimization is that the solver is more likely

to get stuck at bad local minima. Thus, a good initialization is extremely important.
Last but not least, trajectory optimization exploits the mathematical form of dynamic
equations to design the optimal controller. If the dynamics is not smooth, too
complicated, or unknown, it is not clear how to apply trajectory optimization
methods.
Reinforcement Learning
Reinforcement learning is motivated by the learning process of humans. It optimizes

a controller by interacting with the physical environment with numerous trials.
Initially, the controller tries out random moves. If a desired behavior is observed, a
reward is provided as the positive reinforcement. This reward system will gradually
shape the controller until eventually it successfully fulfills the high-level task.
Reinforcement learning is an active research area, with a large number of algorithms
proposed every year. We refer readers to Kaelbling et al. (1996) and Kober
et al. (2013) for a thorough review. We will focus on policy search in the remaining
of this section. Policy search is a popular reinforcement learning algorithm in
physically based character animation. It performs extremely well in this field
because it can solve control problems with high-dimensional continuous state and
action spaces, which is essential to control a human character.
Mathematically, reinforcement learning solves a Markov decision process
(MDP). An MDP is a tuple (S, A, R, D, γ, Psas0 ), where S is the state space and
A is the action space. The states reflect what the current situation is and the actions
are what a character can perform to achieve the specified task. R is the reward
function. A reward function is a mapping from the state-action space to a real
number: R : S A 7! R, which evaluates how good the current state and action
are. D is the distribution of the initial state s0 D and γ [0, 1] is the discount
factor of the reward over time. Psas0 is the transition probability. It outputs the
probability that the next state is s0 if an action a is taken at the current state s. In
physically based character animation, the transition probability is computed by
physical simulation. Although most of the physical simulations are deterministic,
random noise can be added in simulation to increase the robustness of the learned
controller (Wang et al. 2010).
The solution of an MDP is a control policy π that maps the state space to the
action space π : S 7! A. It decides what actions to take at different situations. The
return of a policy is the accumulated rewards along the state trajectory starting at s0
by following the policy π for N steps.
X
N
V π ðs0 Þ ¼ γ Ni Rðsi , π ðsi ÞÞ
i¼0
The reward at earlier states can be exponentially discounted over time. The value
of a policy is the expected return with respect to the random initial state s0 drawn
from D.
V ðπ Þ ¼ Es0 D ½V π ðs0 Þ (10)
Note that the goal of MDP is not to maximize the short-term reward R at the next
state, but a long-term value function V. Using V instead of R as the optimization
target prevents the controller from applying short-sighted greedy strategies. This
agrees with our ability of long-term planning when we are executing our motions.
To formulate an MDP, we need to design the state space S, the action space A, and
the reward function R for a given task. Ideally, the state space should contain all the
possible states of the articulated rigid-body system, including the joint angles q, joint
velocities q_ , and time t, and the actions should include all the joint torques τa.
However, this means that the state and the action space can have hundreds of
dimensions. Due to the curse of dimensionality, solving an MDP in such a high-
dimensional continuous space is computationally infeasible. In practice, researchers
often carefully select states and actions specifically for the tasks in question to make
the computation tractable. For example, if the task is to keep balance while standing,
the state space only needs to include important features for balance, such as the
character’s COM and the center of the ground contact polygon. Similarly, given the
well-known balance strategies, such as ankle strategy and hip strategy, the action
space can be as simple as a few torques at lower body joints. Using prior knowledge
of the task can greatly simplify the problem, which is a common practice in
physically based character animation.
A reward function for character animation usually consists of two parts, a task-
related component that measures how far the current state is from the goal state and a
component that evaluates the naturalness of the motions. Designing a good reward
function is essential to the success of the entire learning algorithm. A good design of
the reward should be a smooth function that gives continuous positive reinforce
whenever progress is made. Mathematically, this design will present gradient infor-
mation that can guide optimization solvers. In contrast, a common mistake is to give
reward only when the task is achieved, which makes the reward function a narrow
spike and flat zero elsewhere. This should be avoided because nearly all the
optimization algorithms would have trouble finding such a spike.
To apply policy search to solve the MDP, we need to parameterize the policy. A
policy can be any arbitrary function. A practical way to optimize it is to parameterize
it and then search for the optimal policy parameters. Commonly used parameteriza-
tions include lookup tables, linear functions, splines, and neural networks. Parame-
terization determines the potential quality of the final policy. However, there is no
consensus what the best way is to parameterize a policy. It is decided on a case-by-
case basis. Once the policy parameterization is decided, an initial policy is iteratively
evaluated and improved until the optimization converges or a user-specified maxi-
mum number of iterations is reached.
16 J. Tan
To evaluate a policy, we can execute the policy in the simulation for N time steps
with different initial states s0 D, accumulate the rewards, and average the returns
to compute the value of the policy. Policy improvement adjusts the policy parameters
to increase its value. A straightforward way is to follow the policy gradient (Ng and
Jordan 2000). However, in many character animation tasks, such as locomotion and
hand manipulation, contact events happen frequently, which introduces unsmooth
contact forces that invalidate policy gradient. For this reason, sample-based stochas-
tic optimization techniques are particularly suited for physically based character
animation. Covariance matrix adaptation evolution strategy (CMA-ES) (Hansen
2006) is the most frequent applied optimization methods for motion control. CMA
can work as long as we can evaluate the value of the policy. It does not need to
compute gradient and does not rely on good initializations. More importantly, CMA
is a “global” search algorithm that explores multiple local optima. Although there is
no guarantee that CMA will converge at the global optimum, in practice, it often
finds good local optima in moderately high-dimensional control spaces (e.g., 20–30
dimensions). For the completeness of the chapter, we will briefly describe the CMA
algorithm. Readers can refer to the original paper (Hansen 2006) for additional
details.
CMA starts with an initial underlying Gaussian distribution in the policy param-
eter space with a large covariance matrix. A population of samples is drawn from this
distribution. The first generation of samples is not biased in the parameter space due
to the large covariance matrix. Each CMA sample represents a control policy. The
policies are evaluated through simulation. They are sorted according to their values
and a certain percentage of the inferior samples are discarded. The underlying
Gaussian distribution is updated according to the remaining good samples and is
used to generate the next generation of the samples. This process is performed
iteratively. Over iterations, the underlying distribution is shifted and narrowed.
Eventually, the distribution converges to a good region of the policy space. The
best CMA sample throughout all the iterations is selected as the optimal control
policy.
In summary, reinforcement learning is a generic method for motion control. It can
automatically learn a wide range of behaviors through simulation trials. Reinforce-
ment learning, more specifically, policy search, is becoming one of the most popular
approaches in character animation synthesis. Reinforcement learning does not
assume any mathematical forms of the dynamic equation. It treats the physical
simulation as a black box, as long as it can output the next state given the current
state and action. Thus, it is not bounded to a particular dynamics model. The same
learning algorithm can still work even if the simulation software is upgraded. The
main challenge of reinforcement learning is to design the states, the actions, the
reward, and the policy parameterization. We need to inject enough prior knowledge
into the design so that the search space is small enough to be computationally
feasible, but not too small to contain effective policies. This requires a lot of
experience and manual tweaking, especially for challenging motion tasks.
Future Directions
The research on physically based character animation has achieved stunning results
in the past two decades. However, we are still far from the ultimate goal of character
animation: a fully automatic system that can synthesize visually realistic motions
that are comparable to those of real humans. Many interesting problems remain to be
challenging research problems. Here I list a few promising future research directions.
Improving Realism
One of the biggest issues of the physically based character animation is its quality.
The synthesized motions are not yet realistic enough for broader applications. It is
currently not comparable to the quality of the animations that are hand-tuned by
artists. One important reason is that the articulated rigid-body systems that are
widely used today are a vast simplification of the real human model. A real human
has 206 bones and over 600 muscles, which has far more degrees of freedom. In
addition, the joints of the articulated rigid body are controlled independently, but
human joints move in coordination due to the intricate arrangement of muscles and
tendons. A recent trend is to build more sophisticated human models based on
biological musculotendon structures (Lee and Terzopoulos 2006; Lee et al. 2009;
Wang et al. 2012). It has been demonstrated that an accurate human model can
dramatically improve the realism of the synthesized motions. As more computa-
tional power is made available to us in the next decade, I expect that we can soon
afford to use highly detailed human models to synthesize character animation with
high fidelity.
Another source of the unnaturalness is due to the handcrafted objective function
in motion control. The objective functions focus mostly on energy efficiency aspect
of motions. However, efficient motions do not equal natural motions. Although
minimizing energy expenditure is one important factor that governs our motion, it
is not the only factor. Our motions are also governed by personal habits, emotion,
task, environment, and many other external factors. It is extremely challenging to
handcraft objective functions for all of them. Assuming that we have abundant of
motion data, which is a realistic assumption given the large volume of motion
sensors installed in phones and other wearable computing devices, it is promising
to extract objective functions from these data using inverse reinforcement learning
(Ng and Russell 2000).
Reducing Prior Knowledge
Although physically based character animation frees us from much manual work of
traditional animation pipelines, it still requires some high-level prior knowledge to
work effectively. For example, we know that regulating the COM of a character
relative to the contact points is important for balance tasks. We can inject this prior
18 J. Tan
knowledge by manually choosing the COM and the ground contact points as
features and include them into the state space of reinforcement learning. Selecting
the right features (prior knowledge) is crucial for many of the current control
algorithms. However, good features for one task may not carry over to different
tasks. Manually selecting features would not scale to more sophisticated characters,
more complicated environments, or more challenging tasks. We need an algorithm
that can discover control strategies with less or even no prior knowledge. This
reminds me of the recent success of deep learning. The way that we are using
hand-engineered features in reinforcement learning today is analogue to using
HoG or SIFT features in computer vision a few years ago. Recent advance in
computer vision has demonstrated that deep neural networks, such as autoencoder
(Vincent et al. 2008) or restricted Boltzmann machine (Hinton 2012), can learn
features automatically. I believe that the next breakthrough in reinforcement learning
is to employ similar techniques to automatically discover important features for
different motion tasks.
Bringing the Character to the Real World
The recent development in physically based character animation has introduced a set
of powerful computational tools. With these tools, natural, agile, and robust motions
can be synthesized efficiently and autonomously in a simulation. However, creating
lifelike robots is still an extremely challenging, trial-and-error process that is
restricted to experts. The fast evolution of 3D printing technology will soon trigger
a shift in the robotics industry from mass production to personalized design and
fabrication, which will result in an immediate need for a faster, cheaper, and more
intuitive way to design robotic controllers. I believe that the computational tools that
are developed in physically based character animation can potentially automate and
streamline the process if we can transfer the controllers from the virtual simulation to
the real world.
Transferring controllers optimized in a simulation onto a real robot is a nontrivial
task. An optimal controller that works in a state-of-the-art simulation often fails in a
real environment. This is known as the Reality Gap. This gap is caused by various
simplifications in the simulation, including inaccurate physical models, unmodeled
actuator dynamics, assumptions of perfect sensing, and zero latency. To fully tap into
the power of the computational tools, we need to develop more accurate physical
simulations to bridge the Reality Gap. Researchers in physically based character
animation have started to investigate this problem (Bharaj et al. 2015; Megaro
et al. 2015). I believe that with further research and development, the Reality Gap
will shrink rapidly, which will make it easier to transfer controllers from the
simulation to the real world. As a result, I envision that the two separate research
fields of character animation and robotics will eventually start to merge. This will
inevitably trigger a fundamental revolution in both character animation and robotics.
References
Abe Y, da Silva M, Popovic’ J (2007) Multiobjective control with frictional contacts. In: Pro-
ceedings of the 2007 ACM SIGGRAPH/Eurographics symposium on Computer animation.
SCA”07. pp 249–258
Andrews S, Kry P (2013) Goal directed multi-finger manipulation: control policies and analysis.
Bai Y, Liu CK (2014) Coupling cloth and rigid bodies for dexterous manipulation. In: Proceedings
of the seventh international conference on motion in games. MIG”14. ACM, pp 139–145
Bharaj G, Coros S, Thomaszewski B, Tompkin J, Bickel B, Pfister H (2015) Computational design
of walking automata. In: Proceedings of the 14th ACM SIGGRAPH/eurographics symposium
on computer animation. SCA”15. ACM, pp 93–100
Clegg A, Tan J, Turk G, Liu CK (2015) Animating human dressing. ACM Trans Graph 34
(4):116:1–116:9
Coros S, Beaudoin P, van de Panne M (2009) Robust task-based control policies for physics-based
characters. ACM Trans Graph 28(5):170:1–170:9
Coros S, Beaudoin P, van de Panne M (2010) Generalized biped walking control. ACM Trans
Graph 29(4):130, Article 130
Coros S, Karpathy A, Jones B, Reveret L, van de Panne M (2011) Locomotion skills for simulated
quadrupeds. ACM Trans Graph 30(4):59
Da Silva M, Abe Y, Popovic’ J (2008) Interactive simulation of stylized human locomotion. In:
ACM SIGGRAPH 2008 Papers. SIGGRAPH”08. ACM, pp 82:1–82:10
DiLorenzo PC, Zordan VB, Sanders BL (2008) Laughing out loud: control for modeling anatom-
ically inspired laughter using audio. In: ACM SIGGRAPH Asia 2008 papers. SIGGRAPH
Asia”08. pp 125:1–125:8
Erez T, Tassa Y, Todorov E (2015) Simulation tools for model-based robotics: comparison of bullet,
havok, mujoco, ode and physx. In: ‘ICRA’, IEEE. pp 4397–4404
Erleben K (2007) Velocity-based shock propagation for multibody dynamics animation. ACM
Transactions on Graphics (TOG), 26(2), Article No. 12.
Geijtenbeek T, van de Panne M, van der Stappen AF (2013) Flexible muscle-based locomotion for
bipedal creatures. ACM Trans Graph 32(6)
Ha S, Ye Y, Liu CK (2012) Falling and landing motion control for character animation. ACM Trans
Graph 31(6):1
Hansen N (2006) The cma evolution strategy: a comparing review. In: Towards a new evolutionary
computation. Springer, New York, pp 75–102
Hinton GE (2012) A practical guide to training restricted Boltzmann machines. In: Montavon G,
Orr GB, Mller K-R (eds) Neural networks: tricks of the trade, 2nd edn, Lecture notes in
computer science. Springer, New York, pp 599–619
Hodgins JK, Wooten WL, Brogan DC, O’Brien JF (1995) Animating human athletics. In:
SIGGRAPH. pp 71–78
Jain S, Ye Y, Liu CK (2009) Optimization-based interactive motion synthesis. ACM Trans Graph 28
(1):1–10
Kaelbling LP, Littman ML, Moore AP (1996) Reinforcement learning: a survey. J Artif Intell Res
4:237–285
Kaufman DM, Sueda S, James DL, Pai DK (2008) Staggered projections for frictional contact in
multibody systems. ACM Trans Graph 27:164:1–164:11
Kim J, Pollard NS (2011) Direct control of simulated non-human characters. IEEE Comput Graph
Appl 31(4):56–65
Kober J, Bagnell JAD, Peters J (2013) Reinforcement learning in robotics: a survey. Int J Robot Res
32:1238
Kwatra N, Wojtan C, Carlson M, Essa I, Mucha P, Turk G (2009) Fluid simulation with articulated
bodies. IEEE Trans Vis Comput Graph 16(1):70–80
20 J. Tan
Kwon T, Hodgins J (2010) Control systems for human running using an inverted pendulum model and
a reference motion capture sequence. In: Proceedings of the 2010 ACM SIGGRAPH/eurographics
symposium on computer animation. SCA”10. Eurographics Association, pp 129–138
Laszlo J, van de Panne M, Fiume E (1996) Limit cycle control and its application to the animation
of balancing and walking. In: Proceedings of the 23rd annual conference on computer graphics
and interactive techniques. SIGGRAPH”96. ACM, pp 155–162
Lee S-H, Terzopoulos D (2006) Heads up! Biomechanical modeling and neuromuscular control of
the neck. ACM Trans Graph 25(3):1188–1198
Lee S-H, Sifakis E, Terzopoulos D (2009) Comprehensive biomechanical modeling and simulation
of the upper body. ACM Trans Graph 28:99:1–99:17
Levine S, Wang JM, Haraux A, Popovic’ Z, Koltun V (2012) Continuous character control with
low-dimensional embeddings. ACM Trans Graph 31(4):28:1–28:10
Liu CK (2009) Dextrous manipulation from a grasping pose. ACM Transactions on Graphics
(TOG) - Proceedings of ACM SIGGRAPH 2009, 28(3), Article No. 59.
Liu CK, Jain S (2012) A short tutorial on multibody dynamics, Technical report GIT-GVU-15-01-1,
Georgia Institute of Technology, School of Interactive Computing
Liu CK, Popovic’ Z (2002) Synthesis of complex dynamic character motion from simple anima-
tions. In: Proceedings of the 29th annual conference on computer graphics and interactive
techniques. SIGGRAPH”02. ACM, pp 408–416
Liu L, Yin K, van de Panne M, Shao T, Xu W (2010) Sampling-based contact-rich motion control.
ACM Trans Graph 29(4), Article 128
Lloyd J (2005) Fast implementation of Lemke’s algorithm for rigid body contact simulation. In:
Proceedings of the 2005 I.E. international conference on robotics and automation. ICRA 2005.
pp 4538–4543
Megaro V, Thomaszewski B, Nitti M, Hilliges O, Gross M, Coros S (2015) Interactive design of
3d-printable robotic creatures. ACM Trans Graph 34(6):216:1–216:9
Mordatch I, de Lasa M, Hertzmann A (2010) Robust physics-based locomotion using
low-dimensional planning. In: ACM SIGGRAPH 2010 papers. SIG- GRAPH”10. ACM, pp
71:1–71:8
Mordatch I, Popovic’ Z, Todorov E (2012) Contact-invariant optimization for hand manipulation.
In: Proceedings of the ACM SIGGRAPH/eurographics symposium on computer animation.
SCA”12. Eurographics Association, pp 137–144
Mordatch I, Wang JM, Todorov E, Koltun V (2013) Animating human lower limbs using contact-
invariant optimization. ACM Trans Graph 32(6):203:1–203:8
Muico U, Lee Y, Popovic’ J, Popovic’ Z (2009) Contact-aware nonlinear control of dynamic
characters. In: ACM SIGGRAPH 2009 papers. SIGGRAPH”09. ACM, pp 81:1–81:9
Ng AY, Jordan M (2000) Pegasus: a policy search method for large MDPs and POMDPs. In:
Proceedings of the sixteenth conference on uncertainty in artificial intelligence, UAI’00.
Morgan Kaufmann Publishers, San Francisco, pp 406–415
Ng AY, Russell SJ (2000) Algorithms for inverse reinforcement learning. In: Proceedings of the
seventeenth international conference on machine learning, ICML”00. Morgan Kaufmann Pub-
lishers, San Francisco, pp 663–670
Otaduy MA, Tamstorf R, Steinemann D, Gross M (2009) Implicit contact handling for deformable
objects. Comput Graph Forum (Proc. of Euro- graphics) 28(2):559-568
Pratt JE, Chew C-M, Torres A, Dilworth P, Pratt GA (2001) Virtual model control: an intuitive
approach for bipedal locomotion. Int J Robot Res 20(2):129–143
Si W, Lee S-H, Sifakis E, Terzopoulos D (2014) Realistic biomechanical simulation and control of
human swimming. ACM Trans Graph 34(1):10:1–10:15
Sueda S, Kaufman A, Pai DK (2008) Musculotendon simulation for hand animation. ACM Trans
Graph 27:83:1–83:8
Tan J, Siu K, Liu CK (2012a) Contact handling for articulated rigid bodies using lcp. Technical
report GIT-GVU-15-01-2, Georgia Institute of Technology, School of Interactive Computing
Tan J, Turk G, Liu CK (2012a) Soft body locomotion. ACM Trans Graph 31(4):26:1–26:11
Tan J, Gu Y, Liu CK, Turk G (2014) Learning bicycle stunts. ACM Trans Graph 33(4):50:1–50:12
Thomas F, Johnston O (1995) The illusion of life: Disney animation, Hyperion. Abbeville Press,
New York, NY.
Todorov E (2006) Optimal control theory. In: Bayesian brain: probabilistic approaches to neural
coding. MIT Press, Cambridge, MA. pp 269–298
Treuille A, Lee Y, Popovic’ Z (2007) Near-optimal character animation with continuous control.
Tsai Y-Y, Lin W-C, Cheng KB, Lee J, Lee T-Y (2010) Real-time physics-based 3D biped character
animation using an inverted pendulum model. IEEE Trans Vis Comput Graph 16(2):325–337
Tsang W, Singh K, Eugene F (2005) Helping hand: an anatomically accurate inverse dynamics
solution for unconstrained hand motion. In: Proceedings of the 2005 ACM SIGGRAPH/
eurographics symposium on computer animation. SCA”05. pp 319–328
Vincent P, Larochelle H, Bengio Y, Manzagol P-A (2008) Extracting and composing robust features
with denoising autoencoders. In: Proceedings of the 25th international conference on machine
learning. ICML”08. ACM, pp 1096–1103
Wampler K, Popovic’ Z (2009) Optimal gait and form for animal locomotion. ACM Trans Graph 28
(3):60:1–60:8
Wang JM, Fleet DJ, Hertzmann A (2009) Optimizing walking controllers. ACM Trans Graph 28
(5):168:1–168:8
Wang JM, Fleet DJ, Hertzmann A (2010) Optimizing walking controllers for uncertain inputs and
environments. ACM Trans Graph 29(4):73:1–73:8
Wang JM, Hamner SR, Delp SL, Koltun V (2012) Optimizing locomotion controllers using
biologically-based actuators and objectives. ACM Trans Graph 31(4):25:1–25:11
Witkin A, Kass M (1988) Spacetime constraints. In: Proceedings of the 15th annual Conference on
computer graphics and interactive techniques. SIG- GRAPH”88. ACM, pp 159–168
Wu J-C, Popovic’ Z (2010) Terrain-adaptive bipedal locomotion control. In: ACM SIGGRAPH
2010 papers. SIGGRAPH”10. ACM, pp 72:1–72:10
Ye Y, Liu CK (2010) Optimal feedback control for character animation using an abstract model. In:
SIGGRAPH”10: ACM SIGGRAPH 2010 papers. ACM, New York, pp 1–9
Ye Y, Liu CK (2012) Synthesis of detailed hand manipulations using contact sampling. ACM Trans
Graph 31(4):41:1–41:10
Yin K, Loken K, van de Panne M (2007) SIMBICON: simple biped locomotion control. In: ACM
SIGGRAPH 2007 papers. SIGGRAPH”07
Yin K, Coros S, Beaudoin P, van de Panne M (2008) Continuation methods for adapting simulated
skills. ACM Trans Graph 27(3):81
Zordan VB, Celly B, Chiu B, DiLorenzo PC (2006) Breathe easy: model and control of human
respiration for computer animation. Graph Models 68:113–132
Data-Driven Hand Animation Synthesis
Sophie Jörg
Abstract
As virtual characters are becoming more and more realistic, the need for record-
ing and synthesizing detailed animations for their hands is increasing. Whether
we watch virtual characters in a movie, communicate with an embodied conver-
sational agent in real time, or steer an agent ourselves in a virtual reality
application or in a game, detailed hand motions have an impact on how we
perceive the character. In this chapter, we give an overview of current methods to
record and synthesize the subtleties of hand and finger motions. The approaches
we present include marker-based and markerless optical systems, depth sensors,
and sensored gloves to capture and record hand motions and data-driven algo-
rithms to synthesize movements when only the body or arm motions are known.
We furthermore describe the complex anatomy of the hand and how it is being
simplified and give insights on our perception of hand motions to convey why
creating realistic hand motions is challenging.
Keywords
Hand motions • Fingers • Character animation • Data-driven animation • Virtual
characters • Motion capture
Introduction
In recent years, character animation has made tremendous steps toward realistic
virtual agents, with increasingly better solutions for motion capturing body motions,
creating highly realistic facial animation, and simulating cloth and hair. With these
more and more realistic components, providing plausible hand and finger motions
S. Jörg (*)
School of Computing, Clemson University, Clemson, SC, USA
E-Mail: sjoerg@clemson.edu

DOI 10.1007/978-3-319-30808-1_13-1
2 S. Jörg
has become highly important. We use our hands to explore our environment by
touching and manipulating objects, to conduct basic tasks such as eating or writing,
to handle complex tools, to create art pieces, or to perform musical instruments
(Napier 1980).
Hand movements also play a crucial role in communicating information and can
even take a main role in conveying meaning in sign languages. However, the
complexity of the hand anatomy and our sensitive perception of small details in
hand motions make it challenging to record or synthesize accurate hand motions.
Furthermore, the differences in size of the hand compared to the body motions
complicate the process of capturing both at the same time. Therefore, finger motions
are typically animated manually, which is a cumbersome and time-intensive process
for the animator.
This chapter describes how detailed hand movements can be recorded and
synthesized. We first explain why we need detailed hand motions for virtual char-
acters and why these subtle motions are challenging to create, giving further details
on the anatomy of the hands and on our perception of hand motions. After a brief
state of the art, we then delve into the different ways to record and to synthesize hand
motions, focusing on optical systems, depth sensors, and sensored gloves to capture
hand motions and on data-driven algorithms to synthesize hand movements
depending on the motions of the body. We conclude by describing some of the
next challenges in the field.
Applications
Why do we need detailed hand motions for virtual characters? Hand and finger
motions are ubiquitous in our lives. They are such an integral part of our daily lives
that we take those intricate motions for granted. We see and interpret them effort-
lessly without the need to think about them. However, even small differences in hand
motions can change our interpretation of a scene (Jörg et al. 2010). Especially as the
realism of virtual characters is increasing, the lack of detailed hand motions becomes
disturbing. More details on our perception of hand motions are given in section
“Perception of Hand Motions.” The importance of hand motions varies with the
application, the task, and the type of character. We start by describing the most
common applications and tasks for which we need virtual hand motions.
Of course virtual hand motions are typically required for animated characters, be
it for entertainment, education, or any other application. Specific situations when
hand motions are crucial are during conversations, when manipulating objects, when
playing instruments, or when using American Sign Language (ASL). Approaches
have been suggested to create finger motions for each of these tasks. Jörg et al.
(2012) synthesize hand motions for known body movements for conversational
characters with a data-driven approach. Their method is presented in more detail
in section “Synthesizing Data-driven Hand Motions.” Manipulation tasks are bound
by the physical constraints of the manipulated objects. Physics-based and data-
driven approaches and combinations of both are very effective in that area
Data-Driven Hand Animation Synthesis 3
(Liu 2009; Pollard and Zordan 2005; Ye and Liu 2012). Algorithms have been
developed to allow virtual characters to play the guitar (ElKoura and Singh 2003) or
the piano (Zhu et al. 2013), and techniques have been refined to improve how to
capture ASL (Huenerfauth and Lu 2010; Lu and Huenerfauth 2009).
For the previously listed applications, hand and finger animations are mostly
created without a time constraint. Real-time applications present additional chal-
lenges. For embodied conversational agents (ECAs), hand motions are particularly
important. In conversations, finger motions can convey meaning and emphasis
(McNeill 1992) and even personality and mood (Wang et al. 2016). In many
approaches generating gestures for conversational characters, finger motions are
not considered separately from the hand, and only a small number of noticeable
hand shapes are synthesized, such as pointing with the index finger. As the quality of
animation for embodied conversational agents is continuing to rise, the need for
more accurate hand motions is increasing.
The commercialization of new technologies in virtual reality (VR) produced
further applications. Hand motions need to be tracked in real time so that a person
in VR can see and use their hands. Furthermore, if multiple persons communicate in
VR we need methods to create movements that accurately convey the meaning of
their conversations. Finally, more realistic hand animations in games could increase
immersion and presence. If a virtual character has to grab and manipulate a wide
range of different objects, not all necessary hand shapes can be created in advance,
and adjustments are needed in real time.
After describing in which applications and for which tasks detailed hand and
finger motions are most important, we will review what the challenges are when
creating them.
Challenges
What are the challenges when capturing or synthesizing hand and finger motions?
The main difficulties come from the complex structure of the hand allowing for
intricate motions, the smaller size relative to the body, and people’s impressive
abilities to recognize and interpret subtleties in hand and finger motions.
Structure of the Hand

In their reference work on anatomy and human movement, Palastanga and Soames
(2012) characterize “the development of the hand as a sensitive instrument of
precision, power and delicacy” as “the acme of human evolution.” The intricate
structure of the hand that allows us to perform a wide range of actions is often taken
for granted. The hand consists of 27 bones, not counting sesamoid bones. The
arrangement of muscles, tendons, and ligaments enables a large variety of possible
poses. The phalanges of the five digits – index, middle, ring, and little or pinky finger
and the thumb – with their 14 bones in total are connected to the palm. The five
metacarpal bones form the palm, and the eight carpal bones connect the metacarpals
to the wrist (Jörg 2011; Napier 1980; Palastanga and Soames 2012).
4 S. Jörg
For animation, in many cases, this skeleton is simplified. The eight carpal bones
at the wrist are summarized into one wrist joint, and the metacarpals that form the
palm might be represented by a simple, rigid structure. A further approximation
concerns the joints. The complex spinning, rolling, and sliding motions of the joints
are typically approximated with rotations around a point. The number of degrees of
freedom of those joints can also be reduced. For example, the metacarpophalangeal
joints are represented with Hardy Spicer joints with two degrees of freedom. In
reality, a slight medial or lateral rotation of the fingers, so a rotation around the axis
that goes through the length of the finger, is possible with the largest angle achiev-
able passively for the little finger. A resulting simple hand skeleton has around 24
degrees of freedom (Jörg 2011; Liu 2008; Parent 2012). However, there is no single
standard skeleton when it comes to character animation, and the exact characteristics
vary widely based on the application and the desired level of realism.
As a consequence of the anatomy of the hand and the arrangement of the tendons,
different joints tend to be moved together such as the interphalangeal joints of each
finger or the ring finger and the little finger (Häger-Ross and Schieber 2000; Jörg and
O’Sullivan 2009). When animating or synthesizing hand motions, we can take
advantage of this property by representing hand motions in a reduced set of
dimensions (Braido and Zhang 2004; Ciocarlie et al. 2007; Santello et al. 1998).
The joints do not move in perfect synchrony, so the dimensionality reduction is not
lossless. How much a motion can be simplified and the dimensionality reduced
depends on the required quality and detail of the motions.
In summary, the structure of the hand is very complex, but can be simplified for
animation depending on how accurate and natural the resulting motion should look.
How much simplification is possible also depends on our ability to perceive and
interpret hand motions.
Perception of Hand Motions

People are extremely skilled when it comes to recognizing and interpreting human
full body motions. We can recognize friends from far away by their walks and
posture before we can see their faces, and we can make a reasonable guess of
characteristics such as the sex of a person just based on their motions (Cutting and
Kozlowski 1977; Kozlowski and Cutting 1977). This process is effortless; it happens
automatically without actively analyzing the motion.
In a similar way, our interpretation of gestures during communication is mostly
automatic – as is our usage of those gestures. A wide range of insights has been
gained on the meaning and interpretation of hand and arm motions, mostly by
observation of how we gesture when we communicate (Kendon 2004; McNeill
1992). The subtle motions of the fingers are an inherent part of gestures, but their
exact use and perception are rarely examined separately. But as the detailed finger
motions are difficult to capture and might be created separately from the body
motions, it becomes important to find out when these details affect our impression
of a character or a scenario.
A perceptual experiment investigating the effect of small delays in finger motion
compared to body motion found that viewers could even detect small
synchronization errors of 0.1 s in short motion clips of a few seconds. A 0.5 s delay
in finger motion altered the interpretation of a 30 s long scenario (Jörg et al. 2010). It
has been shown that animated hands and handlike structures can convey emotions
(Samadani et al. 2011). More interestingly, hand poses and motions influence the
perceived personality of a virtual character with and without the presence of body
motions. For example, spreading motions are seen as more extraverted and open than
flexion, and small hand motions are regarded as more emotionally stable and
agreeable than large motions (Wang et al. 2016). Hand animation is thus essential
when conveying meaning and creating convincing virtual characters.
Our perception of virtual characters also depends on their appearance. The same
body motions on a more realistic humanlike character are rated to be biological
(in contrast to artificial) less often than when they are shown on a less detailed and
more abstract character (Chaminade et al. 2009). There are less studies on this
subject when it comes to hand motions, but it has been shown for hands as well
that different brain areas are activated when we watch real and when we watch
virtual hand actions (Perani et al. 2001).
The appearance of virtual hands also has an impact on our perception in virtual
reality applications, notably on the virtual hand illusion. The virtual hand illusion is a
body ownership illusion: When one sees a virtual hand in virtual reality that is
controlled by and moves in synchrony with one’s own hand, after a short condition-
ing phase, a threat to the virtual hand can trigger an affective response as if the virtual
hand was seen as a part of one’s own body. That means that if a virtual knife hits the
virtual hand, a user can get startled and quickly pulls away their hand even if the
virtual knife cannot do any real damage. Users feel to a large degree as if the virtual
hand was their own. This illusion can be induced for a surprisingly large variety of
models. It has been shown to occur for realistic hands, cartoony hands, a zombie and
a robot hand, an abstract hand made of a torus and ellipsoids, for a cat claw, and for
objects or shapes such as a square, a sphere, a balloon, and a wooden block.
However, the illusion is much weaker for objects and strongest for realistic hands
(Argelaguet et al. 2016; Lin and Jörg 2016; Ma and Hommel 2015a, b; Yuan and
Steed 2010; Zhang and Hommel 2015). While the motion of the hand plays a crucial
role in inducing the virtual hand illusion, it is not known as of yet how much offset,
latency, or errors are possible without destroying the illusion of ownership.
While many questions remain to be solved when it comes to our perception of
hand motions, evidence suggests that viewers are able to notice small details that can
be crucial for our interpretation of a situation. These perceptual skills contribute to
the challenges when creating detailed, realistic finger motions.
State of the Art
Motion capturing has become a standard technique when it comes to creating highly
realistic body motions for movies, games, or similar applications. While more
effective and less expensive systems are still being developed, it has been possible
for about two decades to capture body motions with sufficient accuracy for typical
6 S. Jörg
applications with virtual characters. For finger motions, however, it is still not
possible to capture accurate data in real time without restrictions. Many technologies
and algorithms have been suggested to create detailed finger motions such as
automatically synthesizing them based on the body motions or using sensored gloves
to measure the finger’s joint angles. With the rise of virtual reality applications in
recent years, new demand for capturing hand motions has emerged. Current devices
to capture only the hands for the consumer market have been developed, but these
devices still lack a high reliability and accuracy. A recent, detailed survey of research
literature on finger and hand modeling and animation has been compiled by Wheat-
land et al. (2015).
Capturing Hand Motions
Motion capture has become a widely adopted technology to animate realistic virtual
characters for movies and games (Menache 1999). The detailed movements of the
fingers are, however, challenging to capture. Several methods have been developed
to accomplish this task, each with their advantages and drawbacks. Optical marker-
based motion capture, sensored gloves, markerless optical motion capture, and depth
sensors are the most popular solutions and are described in this section.
Optical Marker-Based Motion Capture
Optical marker-based motion capture records the positions of retroreflective or LED

markers and computes a skeleton based on that data. It provides a high accuracy
compared to other capturing techniques. The optical motion capturing of finger
movements requires careful planning. The denser the markers are placed, the smaller
the markers need to be to avoid mislabelings and occurrences where multiple
markers are mistaken as one by the cameras. For a rather comprehensive marker
set with 20 markers (three on each digit and five on the palm), 6.5 mm spherical or
3/4 spherical retroreflective markers are recommended. The optical cameras used in
such a system need to be placed closer to the performer than if only body motions
were captured as a higher resolution is needed. Furthermore, cameras on the ground
level should be added as the back of the hand where the markers are placed is
directed downward during many gestures (Jörg 2011).
Even with a careful setup, occlusions, where individual markers are hidden from
the cameras, cannot be avoided. For example, when the hand forms a fist, the ends of
the fingertips are hidden. Therefore, adequate post-processing is required, which
typically involves arduous manual labor. These hurdles are the reason that in
applications that do not require real-time animations, such as for animated movies,
finger animations are typically created manually. A small number of markers or
sensored gloves can be used for previsualization (Kitagawa and Windsor 2008).
Several approaches have been suggested to compute an optimal sparse marker set
to ensure better marker separation and identification. The detailed finger motions are
reconstructed based on that subset of markers, thus reducing the time required for
post-processing. It is possible to use dimensionality reduction techniques such as
principal component analysis to take advantage of the approximate redundancy in
hand motions (Wheatland et al. 2013). Another approach tests which markers can be
left out by reconstructing the hand pose finding similar poses in a database and
verifying how different the resulting pose is (Mousas et al. 2014). The computed
optimal marker sets vary with the approach and the example databases used.
Common to most of them is a marker on the thumb and one marker at most on
each finger except for the index finger where there might be two markers (Kang et al.
2012; Mousas et al. 2014; Schröder et al. 2015; Wheatland et al. 2013). Once the
motions with the reduced marker set are recorded, the full-resolution motions are
reconstructed using a database and searching for correspondences.
Hoyet et al. (2012) evaluate the perception of a diverse set of finger motions
including grasping, opening a bottle, and playing the flute, recorded with reduced
marker sets. They use simple methods such as inverse kinematics and interpolation
techniques to reconstruct the hand motions. They show that the required number of
markers depends on the type of motion. For motions where the fingers only display
secondary motions, a static hand pose might be good enough. For a majority of
cases, a simple eight-marker hand model with six markers to capture the four fingers
(four on the fingertips and two on the finger base of the index and pinky) and two
markers to capture the thumb produces motions with sufficiently high quality so that
viewers do not realize a difference with a full marker set. Still, for some motions a
full marker set using forward kinematics is needed.
Sensored Gloves
As recording finger motions with optical, marker-based motion capture systems is

challenging, alternative methods have been developed. Sensored gloves directly
measure the bending angle of different joints (Sturman and Zeltzer 1994). The
number and configuration of bend sensors vary between 5 and 24. Commercially
available technologies include fiber optic sensors, piezoresistive sensors, and inertial
sensors. The main advantage of sensored gloves is that they create a continuous
signal without any occlusions that can be used in real time. However, the accuracy is
lower than for marker-based optical motion capture systems. The number of sensors
of the gloves is typically small compared to the number of degrees of freedom of the
hand. Furthermore, the sensors do not measure the global position of the hand. The
gloves might move compared to the skin during the capture; therefore regular
recalibrations are necessary if a high accuracy is required, which might interrupt
capture sessions. Finally, cross-coupling between sensors adds further challenges,
and more complex calibration methods have been developed (Kahlesz et al. 2004;
Wang and Neff 2013; Wheatland et al. 2015). Sensored gloves are therefore most
useful in applications where a continuous signal is important but accuracy is not
crucial. These properties explain why gloves were used, for example, for virtual
reality applications or as a baseline for movies where the motions can be adjusted in
8 S. Jörg
post-processing. A survey of glove-based systems has been compiled by Dipietro

et al. (2008).
Markerless Optical Systems and Depth Sensors
Markerless optical systems and depth sensors have become very popular in the past
years. These systems, examples are the Microsoft Kinect or the Leap Motion sensor,
only cover a small capture volume, but they are small, light, and inexpensive. Their
accuracy depends largely on the captured hand poses and the hand orientations in
relation to the sensor. Detailed silhouettes can be recognized with a much higher
accuracy than poses where fingers are touching each other and might be hidden from
the sensor by the palm or by other fingers or the thumb. Fast motions might also not
be recognized accurately. However, algorithms for these systems are currently
developed at a high speed, so that further progress is likely in the near future.
The Microsoft Kinect is an RGB-D camera, which records color like a standard
camera and depth information in addition to it. The Leap Motion sensor on the other
hand uses only depth information. Other approaches only use information from a
regular camera. Wang and Popović presented a method where the user wears a cloth
glove with a colored pattern (Wang and Popović 2009). With the pattern, the pose of
the hand can be estimated in single frames by looking for the nearest neighbor hand
pose in a database. As a result of several improvements such as approximating the
nearest neighbor lookup, the search can be conducted in real time. Many suggestions
for algorithms using only monocular videos exist (de La Gorce et al. 2008). They are
typically not very accurate and computationally too intensive for real-time tasks as of
now. When the recognition problem with or without depth information is reduced to
recognizing a specified subset of poses and gestures in a controlled environment,
more reliable approaches exist.
Synthesizing Data-Driven Hand Motions
An alternative approach to capturing finger motions is to synthesize the complete

hand movements. Once a database of motions has been created, data-driven methods
can learn from this data and reuse and adapt it as required. An example of how to
synthesize hand and finger motions based on the method developed by Jörg et al.
(2012) is elaborated in the next paragraph.
The goal of the approach is to automatically create motions for all fingers and the
thumb, not including the wrist, for conversational situations with a full body motion
clip without hand movements as input motion and an available database of motions
that includes both hand and body movements. To this aim, the algorithm finds the
best hand motion clip from the database taking into account features such as the
similarity of the arm motions and the smoothness of consecutive finger motions. The
synthesis process consists of the following steps: First, the input motion and the
database are segmented based on the wrist velocity. Second, the algorithm searches
the database for segments with similar wrist motions than the ones from the input
motion applying dynamic time warping to adapt the length of each segment within
certain limits. Third, a weighted motion graph is computed. The start node of the
graph is connected to the k segments from the database that are most similar to the
first input motion segment. Each of these k segments is then connected to the k
segments that are most similar to the second input motion segment and so on. For
each transition, a cost is calculated by comparing the orientations and angular
velocities of the fingers at the last frame of a segment and at the first frame of the
next segment. A weighted sum of the corresponding transition and segment costs is
applied to each connection, and the shortest path is computed with Dijkstra’s
algorithm resulting in a choice of motion segments. Finally, when combining these
segments, transitions are created where necessary. This algorithm creates plausible
finger motions for conversational situations but excludes any interactions with
objects or self-collisions with the virtual character itself and does not take into
account any partial information of finger positions. Enhancements to this approach
have been developed by Mousas et al. (2015).
Ye and Liu’s approach (Ye and Liu 2012) in contrast creates detailed and
physically plausible hand manipulations when presented with a full body motion
and the movements of objects that are being manipulated. They determine feasible
hand-object contact points using a database and create the hand movements
according to these contact positions. They select visually diverse solutions that result
in intricate motion strategies such as the relocation of contact points.
Many further methods have been suggested to create hand and finger motions
when starting with body motions, such as capturing them separately from the body
and synchronizing them in a post-processing step (Majkowska et al. 2006), proce-
dural algorithms using databases (Aydin and Nakajima 1999), determining key hand
poses based on the body motion with a support vector machine (Oshita and Senju
2014), or approaches taking advantage of data-driven and physics-based methods
(Kry and Pai 2006; Neff and Seidel 2006), to cite just a few examples. Each method
has their advantages and drawbacks. A more exhaustive review of the research
literature on hand and finger motion synthesis can be found in Wheatland et al.’s
state-of-the-art report (Wheatland et al. 2015).
Conclusion and Future Directions
Detailed hand motions are highly important, especially as our expectations toward
realistic virtual characters increase. But high-quality, accurate hand motions are still
time consuming to capture or synthesize. While methods and techniques in this field
are improving at a fast pace, there are still many open questions and processes which
need improvements. Future directions include the following topics:
• Approaches for a variety of applications: Many approaches that have been

suggested specialize on a specific task. One next step would be to develop
10 S. Jörg
approaches that are effective for multiple tasks or to combine approaches and use
the optimal approach based on an automatic assessment of the situation.
• Interpretation of subtle hand motions: Many questions are unsolved when it
comes to our interpretation of subtle hand motions. How do details in finger
motions influence our understanding and our interpretation when communicat-
ing? How much error or latency in hand motions is tolerable?
• Efficient methods for real-time capturing or synthesis: Recording or synthesizing
movement in real time has its own challenges, for example, optimizations over
longer segments of motion are not possible. The accuracy and reliability of
current devices need to be improved.
• Improve our understanding of the virtual hand illusion: How much error is
allowed? Why are some people more prone to the illusion than others? Our
understanding of the reasons for the illusion and the conditions in which it occurs
is still limited. Insights could allow for more appropriate feedback to improve
communication and manipulation in virtual reality applications (Ebrahimi et al.
2016; Prachyabrued and Borst 2014).
• Details of hand motions: For animations that aim to look realistic, details such as
skin deformations and wrinkles based on the anatomy of the hand or on contacts
need to be synthesized. While progress has been made in this area (Andrews et al.
2013; Li and Kry 2014), automatic photo-realism has not been reached yet.
• Control tools for animators: Finally, methods need to be made accessible and
usable for animators, which includes the development of intuitive controls to
allow for a more efficient workflow.
Cross-References
▶ 3D Dynamic Pose Estimation Using Cameras and No Markers

▶ 3D Dynamic Pose Estimation Using Reflective Markers or Electromagnetic
Sensors
▶ 3D Dynamic Probabilistic Pose Estimation Using Cameras and Reflective
Markers
▶ Body Movements in Music Performances: On the Example of Clarinet Players
▶ Movement Efficiency in Piano Performances
▶ Perceptual Evaluation of Human Animation
▶ Postural Movements of Violin Players
References
Andrews S, Jarvis M, Kry PG (2013) Data-driven fingertip appearance for interactive hand
simulation. In: Proceedings of motion on games, MIG ‘13, Dublin, pp 155:177–155:186
Argelaguet F, Hoyet L, Trico M, Lecuyer A (2016) The role of interaction in virtual embodiment:
effects of the virtual hand representation. In: IEEE virtual reality (VR), Greenville, pp 3–10
Aydin Y, Nakajima M (1999) Database guided computer animation of human grasping using
forward and inverse kinematics. Comput Graph 23(1):145–154. doi:10.1016/S0097-8493(98)
00122-8
Braido P, Zhang X (2004) Quantitative analysis of finger motion coordination in hand manipulative
and gestic acts. Hum Mov Sci 22(6):661–678. doi:10.1016/j.humov.2003.10.001
Chaminade T, Hodgins J, Kawato M (2009) Anthropomorphism influences perception of computer-
animated characters’ actions. Soc Cogn Affect Neurosci 2(3):206–216
Ciocarlie M, Goldfeder C, Goldfeder C (2007) Dimensionality reduction for hand-independent
dexterous robotic grasping. In: IEEE/RSJ international conference on intelligent robots and
systems, IROS 2007, San Diego, pp 3270–3275
Cutting J, Kozlowski L (1977) Recognizing friends by their walk: gait perception without famil-
iarity cues. Bull Psychon Soc 9(5):353–356
de La Gorce M, Paragios N, Fleet DJ (2008) Model-based hand tracking with texture, shading and
self-occlusions. In: IEEE conference on computer vision and pattern recognition, Anchorage,
pp 1–8
Dipietro L, Sabatini A, Dario P (2008) A survey of glove-based systems and their applications.
IEEE Trans Syst Man Cybern Part C Appl Rev 38(4):461–482
Ebrahimi E, Babu SV, Pagano CC, Jörg S (2016) An empirical evaluation of visuo-haptic feedback
on physical reaching behaviors during 3D interaction in real and immersive virtual environ-
ments. ACM Trans Appl Percept 13(4):19:1–19:21
ElKoura G, Singh K (2003) Handrix: animating the human hand. In: Proceedings of the ACM
SIGGRAPH/Eurographics symposium on computer animation, San Diego, pp 110–119
Häger-Ross C, Schieber MH (2000) Quantifying the independence of human finger movements:
comparisons of digits, hands, and movement frequencies. J Neurosci 20(22):8542–8550
Hoyet L, Ryall K, McDonnell R, O’Sullivan C (2012) Sleight of hand: perception of finger motion
from reduced marker sets. In: Proceedings of the ACM SIGGRAPH symposium on interactive
3D graphics and games, I3D ‘12, Costa Mesa, pp 79–86
Huenerfauth M, Lu P (2010) Accurate and accessible motion-capture glove calibration for sign
language data collection. ACM Trans Access Comput 3(1):2:1–2:32
Jörg S (2011) Perception of body and hand animations for realistic virtual characters. Ph thesis,
University of Dublin, Trinity College, Dublin
Jörg S, O’Sullivan C (2009) Exploring the dimensionality of finger motion. In: Proceedings of the
9th Eurographics Ireland workshop (EGIE 2009), Dublin, pp 1–11
Jörg S, Hodgins J, O’Sullivan C (2010) The perception of finger motions. In: Proceedings of the 7th
symposium on applied perception in graphics and visualization (APGV 2010), Los Angeles, pp
129–133
Jörg S, Hodgins JK, Safonova A (2012) Data-driven finger motion synthesis for gesturing charac-
ters. ACM Trans Graph 31(6):189:1–189:7
Kahlesz F, Zachmann G, Klein R (2004) Visual-fidelity dataglove calibration. In: Computer
graphics international. IEEE Computer Society, Crete, pp 403–410
Kang C, Wheatland N, Neff M, Zordan V (2012) Automatic hand-over animation for free-hand
motions from low resolution input. In: Motion in games. Lecture notes in computer science, vol
7660. Springer, Berlin/Heidelberg, pp 244–253
Kendon A (2004) Gesture – visible action as utterance. Cambridge University Press, Cambridge
Kitagawa M, Windsor B (2008) MoCap for artists: workflow and techniques for motion capture.
Focal Press, Amsterdam/Boston
Kozlowski LT, Cutting JE (1977) Recognizing the sex of a walker from a dynamic point-light
display. Percept Psychophys 21(6):575–580
Kry PG, Pai DK (2006) Interaction capture and synthesis. ACM Trans Graph 25(3):872–880
Li P, Kry PG (2014) Multi-layer skin simulation with adaptive constraints. In: Proceedings of the
7th international conference on motion in games, MIG ‘14, Playa Vista, pp 171–176
12 S. Jörg
Lin L, Jörg S (2016) Need a hand?: how appearance affects the virtual hand illusion. In: Proceedings
of the ACM symposium on applied perception, SAP ‘16, Anaheim, pp 69–76
Liu CK (2008) Synthesis of interactive hand manipulation. In: Proceedings of the ACM
SIGGRAPH/Eurographics symposium on computer animation, Dublin, pp 163–171
Liu CK (2009) Dextrous manipulation from a grasping pose. ACM Trans Graph 28(3):3:1–3:6
Lu P, Huenerfauth M (2009) Accessible motion-capture glove calibration protocol for recording
sign language data from deaf subjects. In: Proceedings of the 11th international ACM SIGAC-
CESS conference on computers and accessibility, pp 83–90
Ma K, Hommel B (2015a) Body-ownership for actively operated non-corporeal objects. Conscious
Cogn 36:75–86
Ma K, Hommel B (2015b) The role of agency for perceived ownership in the virtual hand illusion.
Conscious Cogn 36:277–288
Majkowska A, Zordan VB, Faloutsos P (2006) Automatic splicing for hand and body animations.
In: Proceedings of the ACM SIGGRAPH/Eurographics symposium on computer animation.
Boston, MA, USA, pp 309–316
McNeill D (1992) Hand and mind: what gestures reveal about thought. The University of Chicago
Press, Chicago
Menache A (1999) Understanding motion capture for computer animation and video games.
Morgan Kaufmann Publishers Inc., San Francisco
Mousas C, Newbury P, Anagnostopoulos CN (2014) Efficient hand-over motion reconstruction. In:
Proceedings of the 22nd international conference in Central Europe on computer graphics,
visualization and computer vision, WSCG ‘14. Plzen, Czech Republic, pp 111–120
Mousas C, Anagnostopoulos CN, Newbury P (2015) Finger motion estimation and synthesis for
gesturing characters. In: Proceedings of the 31st spring conference on computer graphics, SCCG
‘15. Smolenice, Slovakia, pp 97–104
Napier J (1980) Hands. Pantheon Books, New York
Neff M, Seidel HP (2006) Modeling relaxed hand shape for character animation. In: Articulated
Motion and deformable objects. Lecture notes in computer science, vol 4069. Springer, Berlin/
Oshita M, Senju Y (2014) Generating hand motion from body motion using key hand poses. In:
Proceedings of the 7th international conference on motion in games, MIG ‘14. Playa Vista, CA,
USA, pp 147–151
Palastanga N, Soames R (2012) Anatomy and human movement – structure and function, 6th edn.
Butterworth Heinemann/Elsevier, Edinburgh/New York
Parent R (2012) Computer animation: algorithms and techniques, 3rd edn. Morgan Kaufmann,
Burlington
Perani D, Fazio F, Borghese NA, Tettamanti M, Ferrari S, Decety J, Gilardi MC (2001) Different
brain correlates for watching real and virtual hand actions. Neuroimage 14:749–758
Pollard NS, Zordan VB (2005) Physically based grasping control from example. In: Proceedings of
the ACM SIGGRAPH/Eurographics symposium on computer animation. Los Angeles, CA,
USA, pp 311–318
Prachyabrued M, Borst CW (2014) Visual feedback for virtual grasping. In: IEEE symposium on
3D User Interfaces, 3DUI, 2014. Minneapolis, MN, USA, pp 19–26
Samadani AA, DeHart BJ, Robinson K, Kulic D, Kubica E, Gorbet R (2011) A study of human
performance in recognizing expressive hand movements. In: IEEE international symposium on
robot and human interaction communication. Atlanta, GA, USA
Santello M, Flanders M, Soechting JF (1998) Postural hand synergies for tool use. J Neurosci 18
(23):10,105–10,115
Schröder M, Maycock J, Botsch M (2015) Reduced marker layouts for optical motion capture of
hands. In: Proceedings of the 8th ACM SIGGRAPH conference on motion in games, MIG ‘15.
Paris, France, pp 7–16
Sturman DJ, Zeltzer D (1994) A survey of glove-based input. IEEE Comput Graph Appl 14
(1):30–39
Wang Y, Neff M (2013) Data-driven glove calibration for hand motion capture. In: Proceedings of
the 12th ACM SIGGRAPH/Eurographics symposium on computer animation, SCA ‘13. Ana-
heim, CA, USA, pp 15–24
Wang RY, Popović J (2009) Real-time hand-tracking with a color glove. ACM Trans Graph 28
(3):63
Wang Y, Tree JEF, Walker M, Neff M (2016) Assessing the impact of hand motion on virtual
character personality. ACM Trans Appl Percept 13(2):9:1–9:23
Wheatland N, Jörg S, Zordan V (2013): Automatic hand-over animation using principle component
analysis. In: Proceedings of motion on games, MIG ‘13. Zürich, Switzerland, pp
175:197–175:202. ACM
Wheatland N, Wang Y, Song H, Neff M, Zordan V, Jörg S (2015) State of the art in hand and finger
modeling and animation. Comput Graph Forum 34(2):735–760
Ye Y, Liu CK (2012) Synthesis of detailed hand manipulations using contact sampling. ACM Trans
Graph 31(4):245–254
Yuan Y, Steed A (2010) Is the rubber hand illusion induced by immersive virtual reality? Virtual
Reality Conference (VR). IEEE Computer Soc. Waltham, MA, USA, pp 95–102
Zhang J, Hommel B (2016) Body ownership and response to threat. Psychol Res 80(6):1020–1029
Zhu Y, Ramakrishnan AS, Hamann B, Neff M (2013) A system for automatic animation of piano
performances. Comput Anim Virtual Worlds 24(5):445–457
Example-Based Skinning Animation
State of the Art
Tomohiko Mukai
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State-of-the-Art Skinning Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Example-Based Skinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Linear Blend Skinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Skinning Weight Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Skinning Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Example-Based Helper Bone Rigging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Per-Example Optimization of Helper Bone Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Helper Bone Controller Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Abstract
The skinning technique has been widely used for synthesizing the natural skin
deformation of human-like characters in a broad range of computer graphics
applications. Many skinning methods have been proposed to improve the defor-
mation quality while achieving real-time computational performance. The design
of skinned character models, however, requires heavy manual labor even for
experienced digital artists with professional software and tools. This chapter
presents an introduction to an example-based skinning method, which builds a
T. Mukai (*)
Tokai University, Tokyo, Japan
e-mail: tmki@acm.org; mukai@tokai-u.jp

DOI 10.1007/978-3-319-30808-1_14-1
2 T. Mukai
skinned character model using an example sequence of handcrafted or physically

simulated skin deformations. Various types of machine learning techniques and
statistical analysis methods have been proposed for example-based skinning.
In this chapter, we first review state-of-the-art skinning techniques, especially
for a standard skinning model called linear blend skinning that uses a virtual
skeleton hierarchy to drive the skin deformation. Next, we describe several
automated methods for building a skeleton-based skinned character model
using example skin shapes. We introduce skinning decomposition methods that
convert a shape animation sequence into a skinned character and its skeleton
motion. We also explain a practical application of skinning decomposition, which
builds a so-called helper bone rig from an example animation sequence. We
finally discuss the future directions of example-based skinning techniques.
Keywords
Animation • Rigging • Linear blend skinning • Helper bone
Introduction
In computer animation, natural skin deformation is vital for producing lifelike

characters, and there are many techniques and professional software packages for
creating expressive skin animations. For example, a physics-based volumetric sim-
ulation is a typical approach for generating physically valid skin deformations. A
physics simulation, however, requires large computational costs and careful the
design of a character’s musculoskeletal model. Moreover, a numerical simulation
is inferior to kinematic or analytical models in terms of controllability and stability.
The animation of a human character in interactive graphics applications is often
created using a skeleton-based skinning method that embeds a virtual skeletal
structure into the solid geometry of the character’s surface model. This technique
deforms the skin surface according to the movement of the virtual skeleton on the
basis of simple geometric operations and has become a de facto standard skinning
method because of its simplicity and efficiency. The typical production procedure of
skinning animations is composed of the following three processes:
Modeling A surface geometry of a character model is created as a polygonal mesh or

parametric surface, and its material appearance is designed via shading models
and texture images. The character model is often created in the rest pose such as
the so-called T-stance shown in Fig. 1a. In this chapter, we assume that the skin
geometry is constructed as a polygonal mesh.
Rigging A virtual skeleton hierarchy, typically composed of rigid bones and rota-
tional joints, is bound to the character geometry. This process requires specifying
the skinning weights that describe the relative influence of each bone over the
vertex of skin mesh (Fig. 1b). This process is called rigging, character setup, or
simply setup.
Example-Based Skinning Animation 3
a b c
Skeleton
Skin
Modeling in the T-stance Rigging Animation
Fig. 1 Typical process for creating skeleton-driven skin animation
Animation An animation of the skinned character is created as a time series of

rotations of the skeleton joints. The deformation of the skin surface is driven by
the joint movements (Fig. 1c).
In this chapter, we especially focus on the rigging of the skeleton-driven charac-

ter. Building a good character rig is a key requirement for synthesizing natural and
fine skin deformation while providing a full range of motion control for animators,
i.e., an ill-designed skeleton and rig cause unnatural deformation even if the artists
carefully design the character’s geometry and skeleton movement. Moreover, the
skeleton rig should be as simple and intuitive to manipulate as possible so that the
animators can easily create the character motion. Owing to this trade-off between
quality and manipulability, the rigging of a complex human-like character is still a
challenging task even for skilled riggers and animators using professional tools,
which requires trial and error and artistic experience and intuition.
Many researchers have tried to develop automatic and semiautomatic methods of
building a character rig. In particular, recent studies have primarily focused on data-
driven approaches for constructing an optimal skeleton-based character rigs by using
example data. Example-based skinning methods optimize the structure of the skel-
eton hierarchy and skinning weights so that the skinned animation approximates the
example shapes well. Various types of machine learning techniques and statistical
analysis methods have been proposed for stably and efficiently obtaining an accurate
approximation.
The rest of this chapter is structured as follows. In the following section, we
review state-of-the-art skinning techniques. Next, we explain skinning decomposi-
tion methods that build skinned character models from example skin shapes. We also
introduce a practical skinning model called a helper bone rig and its example-based
construction algorithm based on the skinning decomposition method. Finally, we
will discuss the future prospects of example-based skinning techniques.
4 T. Mukai
State-of-the-Art Skinning Techniques
Accurate skin deformation is often generated using physics-based musculoskeletal

(Li et al. 2013) or volumetric (Fan et al. 2014) simulations. These simulation-based
methods are unsuitable for intuitively or intentionally changing the style of defor-
mations and real-time applications because of their high computational cost. More-
over, the manual rigging of a character’s musculoskeletal model is a very
challenging task. A data-driven approach learns the dynamical properties of soft-
tissue materials from the example data (Shi et al. 2008) but still suffers from the
computational complexity.
The kinematic skinning method efficiently computes the vertex positions of the
skin mesh on the basis of the pose of the internal skeleton structure. Linear blend
skinning (LBS) is a standard technique for synthesizing skin deformation in real-
time applications, which computes a deformed vertex position by transforming each
vertex through a weighted combination of bone transformation matrices (Magnenat-
Thalmann et al. 1988). Multiweight enveloping (Merry et al. 2006; Wang and
Phillips 2002) extends the LBS model by adding a weight to each matrix element.
The nonlinear skinning technique uses a dual quaternion instead of a transformation
matrix to overcome LBS artifacts, but a side effect called a bulging artifact is still
caused while bending (Kavan et al. 2007). Several hybrid approaches have been
proposed to blend bone transformations with fewer artifacts. Kavan and Sorkine
(2012) proposed a blending scheme that decomposes a bone rotation into swing and
twist components and separately blends each component using different algorithms.
The stretchable and twistable bone model (Jacobson and Sorkine 2011) uses differ-
ent weighting functions for scaling and bone twisting, respectively. These methods
successfully synthesize artifact-free skin deformation and do not discuss stylized
skin deformation such as muscle–skin deformation. EigenSkin constructs an effi-
cient model of the additive vertex displacement for LBS using a principal component
analysis (Kry et al. 2002). These methods successfully synthesize artifact-free skin
deformation. Naive LBS, however, is still a de facto standard skinning model in
interactive graphics applications because of its efficiency and simplicity.
One practical solution for minimizing LBS artifacts is to add extra bones called
helper bones. The helper bone rig has become a practical real-time technology for
synthesizing stylized skin deformation based on LBS. The helper bone is a second-
ary rig that influences skin deformation, and its pose is procedurally controlled
according to the pose of the primary skeleton. Mohr and Gleicher (2003) first
introduced the basic concept of a helper bone system. In their work, helper bones
are generated by subdividing primary bones, and a scaling parameter is procedurally
controlled according to the twist angle of the primary bone thus minimizing the
candy-wrapper artifact. This technique has been widely used in many products
because of its efficiency, flexibility, and compatibility with the standard graphics
pipeline (Kim and Kim 2011; Parks 2005). Although this technique provides a
flexible yet efficient synthesis of a variety of expressive skin deformations, rigging
with helper bones is still a labor-intensive process. We have developed an example-
based technique to build helper bone rigs, as explained in Section “Example-Based

Helper Bone Rigging.”
Scattered data interpolation such as pose-space deformation (PSD) is another
approach for synthesizing skin deformation from example shapes (Kurihara and
Miyata 2004; Lewis et al. 2000; Sloan et al. 2001). PSD uses radial basis function
interpolation for blending example shapes according to the skeleton pose. This
technique produces high-quality skin animation via intuitive designing operations.
However, the PSD model requires a runtime engine to store all example data in
memory. Furthermore, the computational cost of PSD increases in proportion to the
number of examples. Consequently, many example shapes cannot be used in real-
time systems with a limited memory capacity, such as mobile devices.
The machine learning-based approach for skin deformation analyzes the relation-
ship between a skeleton pose and its corresponding skin shape using a large set of
samples. A regression technique was proposed to estimate a linear mapping from the
skeletal pose to the deformation gradient of the skin surface polygons (Pulli and
Popović 2007). The seminal work of Park and Hodgins (2008) predicts an optimal
mapping from the skeletal motion to the dynamic motion of several markers placed
on a skin surface. Neumann et al. (2013) proposed a statistical model of skin
deformation that is learned from human skin shapes captured with range scan
devices. These methods construct a regression model from a set of example skeleton
poses and skin shapes.
Example-Based Skinning
Linear Blend Skinning
LBS (Magnenat-Thalmann et al. 1988) is a standard method that has been widely
used for a broad range of interactive applications such as games. Most real-time
graphics engines support the LBS-based rig because of its simplicity and efficiency.
The LBS model computes a deformed vertex position by transforming each vertex
through a weighted combination of bone transformation matrices. Given a skeleton
with P bones, the global transformation matrix of the p-th bone is denoted as a 4 4
homogeneous matrix Gp, p {1, , P}. Let Ḡp and vj , j {1, , J} denote
the matrix and positions of the j-th vertex on a skin in the initial T-stance pose, and
let the skinning transformation matrix be represented by Mp ¼ Gp G 1. The global
p
transformation matrix Gp can be decomposed into a product of the local transfor-
mation matrix Lp and the parent’s global transformation matrix as Gp = Gϕ(p) Lp
where ϕ( p) {1, , P} is the parent of the p-th bone. The deformed vertex position
vj is computed with nonnegative skinning weights wj,p as
X
vj ¼ wj, p Mp vj ; (1)
p
6 T. Mukai
where the affinity constraint p wj, p = 1, 8j is satisfied. The number of nonzero

skinning weights at each vertex is assumed to be less than a constant number k,
which is expressed as p |wj, p|1 k, 8p, where ||α denotes the Lα norm. This
sparsity assumption can be interpreted as that each vertex moves according to the
transformations of a few spatially neighboring bones. Moreover, this constraint
ensures the efficient computation of a skin animation regardless of the total number
of skeleton bones since the uninfluenced bones can be eliminated in the computation
of the vertex deformation.
Skinning Weight Optimization
An LBS rig is built by designing a skeleton hierarchy, including the initial bone
transformation Ḡp and the parent–child relation ϕ( p) of each bone and the
corresponding skinning weights wj,p of each vertex. Since the skinning weights are
more difficult to design than the skeleton structure owing to the larger number of free
parameters, several methods have been proposed to optimize the skinning weights
for an arbitrarily skeleton structure and skin geometry. The Pinocchio system (Baran
and Popović 2007) uses an analogy to heat diffusion over the skin mesh for
estimating shape-aware skinning weights. The bounded biharmonic weight model
(Jacobson et al. 2011) produces a smooth distribution of the skinning weights that
minimizes the Laplacian energy over the character’s volumetric structure. The
deformation-aware method (Kavan and Sorkine 2012) optimizes the skinning
weights to minimize an elastic deformation energy over a certain range of skeletal
poses. These methods make an assumption regarding the material properties of the
skin surface, e.g., they assume that the physical properties of the character skin, such
as the elasticity, stiffness, and friction, do not change over the entire body surface.
However, this assumption is somewhat optimistic because many characters should
have heterogeneous distribution of material properties that include those of bones,
muscles, and fats. Further investigation is therefore demanded to improve the quality
of the weight optimization.
Alternative approach uses a set of example data of the skin shape and skeleton
poses (Miller et al. 2011; Mohr and Gleicher 2003; Wang and Phillips 2002).
Given the prior information of the skeleton hierarchy, and N examples of a pair
of skeleton pose M b p, n and skin shape bvj, n , n {1, , N}, the optimal skinning

weights wj,p are obtained by solving a constrained problem to minimize the squared
error of the vertex positions between the example shape b vj, n and the deformed skin
vj as
2
X X X
fw g ¼ argminj ^vj , n ^ p, n vj ;
wj, p M (2)
w n j
p

2
X
subject to wj, p ¼ 1, 8j; (3)
p
wj, p 0, 8j, p; (4)

X
jwj, p j k, 8j; (5)
1
p
where the three constraints are (Eq. 3) the affinity constraint, (Eq. 4) nonnegativity
constraint, and (Eq. 5) sparsity constraint. This constrained least-squares problem
can be approximately solved using a quadratic programming (QP) solver by relaxing
the sparsity constraint, as detailed in Section “Weight Update Step.” We can alter-
natively use the simpler nonnegative least-squares method (James and Twigg 2005)
if no sparsity constraint is imposed.
These weight optimization techniques produce a good initial guess for the
skinning weights for an arbitrary shape of the skin mesh and skeleton hierarchy.
Manual refinement, however, is still required in practice to eliminate undesirable
artifacts and to add artist-directed stylized skin behavior.
Skinning Decomposition
Several algorithms have been proposed to extract both the skinning weights and
optimal bone transformations from a set of example shapes, which is called the
skinning decomposition problem. In other words, the goal of skinning decomposi-
tion is to convert a shape animation into a bone-or skeleton-based skinned anima-
tion. Given a set of N example shapes bvj, n , the goal of skinning decomposition is to

find the optimal skinning weights wj,p and skinning matrix Mp,n to best approximate
the example shapes in a least-squares sense as
2
X X X

fw , M g ¼ argmin ^vj , n wj, p Mp vj (6)
w, M n j
p

2
subject to the affinity, nonnegativity, and sparsity constraints on the skinning

weights.
The skinning mesh animation algorithm (James and Twigg 2005) uses a mean-
shift clustering algorithm for identifying rigid or near-rigid bone transformations and
applies the nonnegative least-squares method to estimate the skinning weights.
Kavan et al. (2010) proposed a least-squares method with a dimensionality reduction
to efficiently extract the nonrigid bone movements and skinning weights from an
example shape animation. The smooth skinning decomposition with rigid bones
(SSDR) algorithm (Le and Deng 2012) introduced a rigidity constraint on the bone
8 T. Mukai
transformation M, which requires that M be a product of the rotation matrix R and

translation matrix T as M = T R, where RT R = I and det(R) = 1 are satisfied.
This algorithm was later extended to identify hierarchically structured bone trans-
formations from a shape animation sequence (Le and Deng 2014). The SSDR
algorithm is designed to meet the requirements of the production of interactive
graphics systems: Three types of constraints on the skinning weights are always
assumed in many graphics software packages, and the rigidity constraint on the bone
transformations is also necessary for most game engines.
The SSDR algorithm uses a block coordinate descent algorithm that optimizes the
transformation of each bone or skinning weight at each subiteration while fixing the
other variables. For instance, the weight update step optimizes the skinning weights
while fixing all bone transformations, and the transformation update step optimizes
the transformation of each bone while fixing the transformations of the other
remaining bones and the skinning weights. These alternative processes are repeated
until the objective function converges. The details of each subiteration are described
below.
Initialization
In the first step, each vertex of the skin mesh is bound to one bone with a skinning
weight of one. The initialization problem then becomes the clustering of vertices into
the specified number of clusters, where vertices in the same cluster have similar rigid
transformations. For each cluster, a rigid bone transformation is fitted to relate the
vertex positions in the rest pose to the vertex positions at each example. Since the
quality of this motion-driven clustering has a great effect on the remaining skinning
decomposition steps, several clustering algorithms have been explored, such as
mean shift clustering (James and Twigg 2005), K-means clustering (Le and Deng
2012), and Linde–Buzo–Gray algorithm (Le and Deng 2014), for stably obtaining an
accurate result. It is possible to apply a more sophisticated algorithm for enhancing
the stability and efficiency of the clustering, which is an important open question.
Weight Update Step

The optimal skinning weights wj,p are updated while fixing all bone transformations
Mp,n in Eq. 6. The resulting optimization problem is rewritten as the following
per-vertex constrained least squares problem:

wj ¼ argmin Awj b2 ;
2 (7)
Wj

subject to wj 0 wj ¼ 1 wj k, 8j;
1 0
Where
2 3
M v ... Mp, 1 vj h iT
T ^ 4 1, 1 j
wj ¼ wj, 1 wj, p , A ¼ ⋮ ⋱ ⋮ 5, b ¼ ^vTj, 1 ^vTj, N :
M1, N vj Mp, N vj
This problem is difficult to directly solve using a standard numerical solver owing
to the L0-norm constraint |wj|0 k. Hence, an approximate solution is used to relax
the sparsity constraint (Le and Deng 2012). Specifically, the L0-norm constraint is first
excluded from Eq. 7, and the resulting QP problem is solved using a stock numerical
solver. When the solution does not satisfy the L0-norm constraint, the k bones requiring
the most effort are selected, and the weights for other bones are set to zero. The final
solution is obtained by solving the QP problem again with the selected k bones.
Bone Transformation Update Step

The transformation of the p-th bone for each example shape is optimized while fixing
the skinning weights and the transformations of remaining P 1 bones at each
subiteration. Each subproblem becomes a per-example weighted absolute orienta-
tion problem given by
2
n o X X

Rp, n, Tp, n ¼ argmin ^vj , n wj, p T p, n Rp, n vj ; (8)
Rp, n , T p, n j
p

2

RTp, n Rp, n ¼ I, det Rp, n ¼ 1, 8p, n;

^ are obtained by the closed-form method. Please refer
where the optimal Rp,n and T p, n
to Le and Deng (2012) for further details.
Applications
The skinning decomposition algorithm allows for the conversion of any type of
shape animation sequence into a skeleton-or bone-based skinned animation. For
example, the animation of soft body objects is often created using numerical
simulations such as a finite element method or mass–spring network models. A
facial expression is created using a blendshape animation technique. These advanced
techniques, however, are not supported by some graphics engines, especially those
for mobile devices. Therefore, a complex skin behavior created using professional
content creation tools is converted into an LBS-based skinning animation that is
widely supported by most engines. This procedure is fully compatible with a
standard production workflow. The main drawback is the lack of detail in the skin
deformation caused by wrinkles, self-collisions, etc., because the sparse set of rigid
bones merely linearly approximates the example deformations.
Example-Based Helper Bone Rigging
The helper bone rig has become a practical real-time technology for synthesizing
stylized skin deformation based on LBS (Kim and Kim 2011; Mohr and Gleicher
2003; Parks 2005). The helper bone is a secondary rig that influences the skin
10 T. Mukai
deformation, and its pose is procedurally controlled according to the pose of the
primary skeleton as illustrated in Fig. 2. Although the helper bone rig is manually
designed in common practice, it requires a labor-intensive process for developing a
procedural bone controller and the skinning weights. We have proposed an example-
based rigging method (Mukai 2015). Our method uses a two-step algorithm to add
helper bones to a predesigned primary skeleton rig using example pairs of the
primary skeletal pose and desirable skin shape. In the first step, our system estimates
the optimal skinning weights and helper bone transformation for each example. We
used a modified version of the SSDR algorithm to incrementally insert rigid helper
bones into the character rig. In the second step, the helper bone controller is
constructed as a polynomial function of the primary skeleton movement.
Here, we first formulate LBS with P primary bones and H helper bones as
!
X X
vj ¼ wj, p Mp þ wj, h Sh vj ; (9)
p h
where Sh and wj,h denote the skinning matrix of the h-th helper bone and the
corresponding skinning weight, respectively. The first term represents the skin
deformations driven by the primary skeleton, and the second term contributes
additional control of the deformations using helper bones. The number of helper
bones H is manually set to balance the deformation quality and computational cost.
Helper bones are procedurally controlled with simple expressions according to the
pose of the primary skeleton in common practice. We use a polynomial function fh
that maps the primary skeleton pose Lp to the helper bone transformation Sh as
Initial pose
Skin mesh Helper bones
Primary bone Primary bone
Deformation
a b
Skin deformation
Rotation of
primary bone
Procedural
Candy-wrap artifact control
Linear blend skinning Linear blend skinning + Helper bones
Fig. 2 Linear blend skinning with procedural control of helper bones

Sh f h ðL1 , L2 , , LP Þ: (10)
Our helper bone rigging technique builds the regression function fh and skinning
weights using example shapes and skeleton poses. Given a set of N pairs of an
example shape and a primary skeleton pose ^ ^ p, n, our problem is formulated as a
vj, n, M
constrained least-squares problem that minimizes the squared reconstruction error
between the example shape and the skin mesh with respect to the skinning weights
wj,p, wj,h and the skinning matrices Sh,n as
2
X X X X

fw , S g ¼ argmin ^vj, n ^ p, n vj
wj, p M wj, h Sh, n vj (11)
w, S n j
p h

2
subject to the affinity, nonnegativity, and sparsity constraints on the skinning weights
and the rigidity constraint on the skinning matrix Sh, n.
Per-Example Optimization of Helper Bone Transformations
The optimal rigid transformations of the helper bones are first estimated for each
example shape using the optimization procedure summarized in Algorithm 1. Our
system inserts the specified number of helper bones into the character rig in an
incremental manner. Then, the helper bone transformations for each example and the
skinning weights are optimized using an iterative method. The overall procedure is
similar to the SSDR algorithm, where the skinning weights and bone transformations
are alternately optimized by subdividing the optimization problem (Eq. 11) into
subproblems of bone insertion, skinning weight optimization, and bone transforma-
tion optimization. We used the optimization techniques from the SSDR algorithm to
solve these three subproblems. The main difference here is that the SSDR algorithm
does not have the prior information about the transformable bones but only the
information about the number. Hence, the SSDR algorithm applies a clustering
technique to simultaneously estimate an initial bone configuration. In our method,
the primary skeleton and its example poses are given in the problem. This method
inserts helper bones using incremental optimization with a hard constraint on the
primary bone transformation.
Algorithm 1 Optimization of helper bone transformations and skinning weights

Input: {^
vj}, {Ḡp¯}, {^
vp,n}, {Ĝp,n}, H
Output: {Sh,n}, {wj,p}, {wj,h}
1: {Sh,n} = I, 8h, n, {wj,h} = 0, 8j, h

2: Initialize {wj,p}
3: repeat
4: Insert a new helper bone
5: Update helper bone transformations {Sh,n}
12 T. Mukai
6: Update skinning weights {wj,p} and {wj,h}

7: Remove insignificant helper bones
8: until The number of inserted helper bones is reached
9: repeat
10: Update helper bone transformations {Sh,n}
11: Update skinning weights {wj,p} and {wj,h}
12: until The error threshold is reached
Incremental bone insertion Our technique uses an incremental method to insert a

new helper bone into the region where the largest reconstruction errors occur. For
example, if the new LBS causes an elbow-collapse artifact, a helper bone is
generated around the elbow to minimize this artifact. First, our system searches for
a vertex with the largest reconstruction error, which is computed as
2
X X X

j ¼ argmin ^vj , n ^ p, n vj
wj, p M wj, h Sh, n vj (12)
j n
p h

2
Second, we compute a rigid transformation that closely approximates the dis-

placement of the identified vertex and its one-ring neighbors from their initial
position by solving an absolute orientation problem (Horn 1987). Then, a new
helper bone is generated using the estimated transformation as its own transforma-
tion. Next, the skinning weights wj,p and wj,h and the transformation matrix of all
helper bones are updated by solving constrained least-squares problems. Finally, the
system removes the insignificant helper bones that have little influence on the skin
deformation. Our current implementation removes the helper bones that influence
less than four vertices. This process is repeated until the specified number of helper
bones is achieved.
Weight and bone transformation update After inserting the specified number
of helper bones, the skinning weight update step (Section “Weight Update Step”)
and bone transformation update step (Section “Bone Transformation Update Step”),
except for the primary bones, are alternately iterated until the approximation error
converges.
Helper Bone Controller Construction
The helper bone controller is constructed by learning a mapping from the primary
bone transformations to the helper bone transformations. We use a linear regression
model to represent the mapping from the local transformation of primary bones Lp to
that of helper bones Lh.
Transformation parameterization The local transformation matrix Lh is
extracted from the estimated skinning matrix Sh. By definition, Sh is decomposed
into a product of the transformation matrices as
1 ;
Sh ¼ GϕðhÞ Lh G (13)
h
where ϕ(h) {1, , P} is the parent primary bone of the h-th helper bone, which is
selected to minimize the approximation error as detailed later. The initial transfor-
mation matrix Gh is an unknown rigid transformation matrix. Assuming that the
local transformation matrix of the helper bones is the identity matrix in the initial
stance pose, Ḡh is equal to that of the parent primary bone Ḡϕ(h) by the definition of
forward kinematics. Therefore, we can uniquely extract the local transformation
matrix by
Lh ¼ G1
ϕðhÞ Sh GϕðhÞ : (14)
The extracted local matrix is parameterized with fewer variables to reduce the
dimensionality of the regression problem. Using a rigid transformation, Lh can be
parameterized using a combination of a translation vector th ℜ3 and bone rotation
variables rh SO(3). We used exponential maps for rh (Grassia 1998). This results
T
in the transformation of Lh into a six-dimensional vector form tTh rTh ℜ6 . In
addition, the local transformation of the primary bone Lp is parameterized by its
animating variables. For simplicity, we have assumed that each primary bone does
not have a translation or scale key and that a bone rotation is always represented by
exponential maps.
Regression model construction We have used a χ-th-order polynomial function
as a regression model. The transformation parameter of each helper bone is approx-
imated by

th
f n L 1 , L 2 , , Lp
rh
h iT (15)
¼ Fh xT1 xTp 1 ;
H
1
where xp ℜ4 χ becomes an independent variable vector that is composed of all
of the variables of the χ-th-order polynomial of rp. For example, if we take χ = 2,
the independent variable vector from r = [r1, r2, r3] is x = [r1, r2, r3, r21, r22, r23, r1r2,
r1r3, r2r3]. The regression matrix for the h-th helper bone Fh ℜ6(1+p dim(xp)) is
estimated from examples using the least-squares technique. In addition, we add a
sparsity constraint to minimize the number of nonzero regression coefficients to
generate a simpler model. The least-squares problem with the sparsity constraint can
be formulated as a Lasso problem (Tibshirani 2011) given by
Fh ¼ argmin jYh Fh j22 þ λjFh j1 ; (16)

Fh
where
14 T. Mukai
2 3
X1, 1 X1, N
6 ... ⋱ ... 7
X¼6 4 XP, 1 XP, N 5
7
1 1

t th, N
Y h ¼ h, 1
rh, 1 rh, N
and λ is the positive shrinkage parameter that controls the trade-off between the
model accuracy and the number of nonzero coefficients. Using a stock Lasso solver,
we can efficiently solve this problem.
Parent bone selection There is only one problem that remains: the selection of an
appropriate parent bone ϕ(h) for each helper bone. This is a discrete optimization
problem. Generally, since the number of primary bones is smaller, we can implement
an exhaustive search to find the optimal one. Further, the best parent bone ϕ(h) can
be identified by evaluating Eqs. 14 and 15. Here, each primary bone can be used as
ϕ(h), and the best one that minimizes the objective function (Eq. 16) can be selected.
Experimental Results
We evaluated the approximation capability and computational performance of our

helper bone rigging system. For all experiments, the parameter k, which is the
maximum number of transformations to be blended, was fixed at 4. The reconstruc-
tion error was evaluated using root mean square (RMS) error of the vertex position.
The optimization procedure of the helper bone transformation is parallelized over
vertices, helper bones, or examples using Intel Threading Building Blocks. The
computational timing was measured at 3.4 GHz on a Core i7-4770 CPU (eight
logical processors) with 16 GB of RAM.
Test Dataset
We used a muscle function from Autodesk Maya to synthesize an example skin
shape from a skeleton pose. The muscle system emulates static muscle–skin defor-
mation with a skeletal pose. The muscle system also produces a dynamic deforma-
tion that is caused by bone acceleration and the inertia of the muscles. For our
experiment, we used only static deformation because our method supports only static
mapping from a skeleton pose to a skin shape. The test character model is a sample
asset of a Maya tutorial, as shown in Fig. 3. The height of the leg model is 200 cm in
the initial pose. The skeleton has P = 3 animating bones and five degrees of freedom
(DOFs) including hip swing and twist (3 DOFs), knee bend (1 DOF), and ankle bend
(1 DOF). The eleven muscles expand and contract according to the movement of the
primary skeleton. They drive the deformation of 663 vertices using a proprietary
algorithm.
A test dataset was created by uniformly sampling the bone rotation of the primary

skeleton every 20 within each range of joint motion. Consequently, we created 6750
a b c
Primary skeleton Muscle Skin mesh
Fig. 3 Character model used to create an example pose and skin shape. The skin deformation is
driven by a primary skeleton and virtual muscle, which is a built-in function of Autodesk Maya
example pairs consisting of a skeleton pose and skin shape by discretizing the DOFs
of the hip swing, hip twist, knee bend, and ankle bend into 6 6, 9, 5, and 5 levels,
respectively.
Evaluating the Optimized Bone Transformations

In the first experiment, different numbers of helper bones were inserted into the
character rig while fixing the polynomial order χ = 2 and the shrinkage parameter
λ = 0. Figure 4 shows the convergence of the reconstruction error with the number
of helper bones and the number of iterations. The reconstruction error decreased
according to the number of helper bones. In addition, there were no significant
differences between the reconstruction error of four helper bones and that of five
helper bones. This result indicates that the approximation almost converged at four
helper bones. The reconstruction error monotonically decreased with the number of
iterations, which demonstrates the stability of our SSDR-based rigging system.
Figure 5 shows optimized models using different numbers of helper bones. The
center image of each screen shot shows the initial pose, and the left and right images
show a leg stretching pose and bending pose, respectively. The helper bones a, b, and
c are located near the hip, knee, and ankle to minimize LBS artifacts. Helper bone
d is located in the thigh to emulate the muscle bulge. The skinning weight map for
each helper bone is visualized in Fig. 6. Helper bone a had a significant influence on
a large area of the thigh, whereas the other helper bones had a lesser influence. This
is the inevitable result of our incremental bone insertion algorithm where the first
helper bone is inserted to offset the largest reconstruction error.
16 T. Mukai
Fig. 4 Convergence of the RMS error [cm]

reconstruction error according
to the numbers of helper Without helper bone
bones and iterations 5.0
4.0
3.0
2.0 H =1
2
3
4
5
0
5 10 15 20 iteration
a b
a a
a a
a a
b
b
b
(a1) Leg extension (a2) Initial pose (a3) Leg bend

(b1) Leg extension (b2) Initial pose (b3) Leg bend
d
c
a a a
a a d
a d b
b b b
b b
c c
c c
(c1) Leg extension (c2) Initial pose (c3) Leg bend (d1) Leg extension (d2) Initial pose (d3) Leg bend
Fig. 5 Optimized character rigs using different numbers of helper bones. Each model shows a
different helper bone behavior
Fig. 6 Skinning weight map for each helper bone. The larger weight is indicated by a darker area
To build the rig with one helper bone, our system consumed 0.17, 0.51, and 0.17 s
per iteration for the bone insertion step, weight update step, and transformation
update step, respectively. The total optimization time was about 15 s for 20 iterations.
For the rig with four helper bones, the time recorded was 0.17, 0.82, and 0.72 s per
iteration. The total time was about 32 s.
Evaluating the Accuracy of the Bone Controller

In the second experiment, we examined the approximation capability of the helper
bone controller. We evaluated the increase in the RMS error caused by approximat-
ing the bone transformations with the regression model. We also counted the number
of nonzero polynomial coefficients using different polynomial orders X and shrink-
age parameters λ while fixing H = 4.
The experimental results are summarized in Table 1. The baseline RMS recon-
struction error, which was measured after the per-example transformation optimiza-
tion, was 1.36 cm. The ratio of the increase in the approximation error was within the
range of 150–190 %. The reconstruction error decreased according to the polyno-
mial order, and there was no significant difference between the quadratic and cubic
polynomials. On the other hand, the redundant polynomial terms were removed
through the shrinkage parameter λ while minimizing the increase in the approxima-
tion error. In this experiment, our prototype system consumed about 5 μs per frame
to compute all skinning matrices Ŝh from the primary skeleton pose Lp. In detail,
1 μs was consumed to compose the independent variables xp from the local trans-
formation matrices Lp, and the computation of the regression model using Eq. 15
consumed 1 μs for each helper bone. The former time increases in proportion to the
number of primary bones, and the latter increases with the number of helper bones.
The computational speed is sufficiently fast, although we could further improve the
performance by parallelizing the execution of bone controllers.
18 T. Mukai
Table 1 Statistics of reconstruction error and the number of nonzero polynomial coefficients with
respect to the polynomial order w and the shrinkage parameter l
Average # of nonzeros RMS error [cm]
λ 0 10 20 0 10 20
Linear 6 5.3 5.1 2.57 2.58 2.59
Quadratic 14 10.9 9.1 2.11 2.12 2.17
Cubic 26 18.4 15.1 2.03 2.07 2.11
Limitations
Currently, our system does not provide any guidelines for creating an example
dataset. Even though we have used uniform sampling of the joint DOFs to create
example poses in the experiments, this simple method might generate many redun-
dant examples. This method may even possibly fail to sample important poses and
shapes. We plan to perform further studies to identify a more artist-friendly workflow
that can create a minimal example dataset. We believe that an active learning method
(Cooper et al. 2007) could be a possible solution that allows artists to design example
shapes in a step-by-step manner.
Our method does not ensure global optimality for the skinning weights and helper
bone controller. We have found that an increase in the number of helper bones often
degrades the reconstruction, because numerical errors are separately accumulated
when solving the optimizations for the skinning decomposition and bone controller
construction.
Future Directions
In this chapter, we have described an example-based rigging technique for building

an LBS rig using the example data of skin deformation. Although the example-based
method requires a large amount of example data to construct a character rig, several
simulation methods have been proposed to generate physically valid skin deforma-
tion using a heavy computation burden. Moreover, the recent development of 3D
scanning devices allow for the acquisition of the skin deformations of actual human
beings. These state-of-the-art shape acquisition techniques will enable the mass
production of example skin shapes within a short period of time and significantly
reduce the amount of unartistic manual labor for rig construction.
Although most skinning decomposition techniques and helper bone rigs have
been adopted for the LBS-based technique, popular nonlinear blend skinning tech-
niques such as dual quaternion skinning (Kavan et al. 2007), stretchable and
twistable bones (Jacobson and Sorkine 2011), and elasticity-inspired joint deformers
(Kavan and Sorkine 2012) are worth investigating, which is an interesting open
question.
Dynamic skinning is another promising future direction. The kinodynamic skin-
ning technique (Angelidis and Singh 2007) provides volume-preserving deformation
based on proxy muscles. A rig-space physics technique optimizes the free parame-
ters of the handcrafted kinematic rig to approximate physically simulated skin
deformation (Hahn et al. 2012, 2013). Position-based dynamics method (or PBD)
was used to synthesize the skin deformation caused by self-collisions and the
secondary effects of soft tissues (Rumman and Fratarcangeli 2015). This method
provides a plausible and stable soft body motion at interactive rates but requires the
elaborate construction of a PBD-based rig. The MoSh model (Loper and Black
2014) estimates the dynamic skin deformation from a sparse set of motion capture
markers using a statistical model of human skin shapes. This method generates the
skin shape in a low-dimensional subspace to meet the movement of the markers. The
Dyna model (Pons-Moll et al. 2015) also constructs a dynamic skin deformation
model using a subspace analysis of 4D scans of human subjects. This skin defor-
mation is generated using a second-order autoregressive with an exogenous input
model in the low-dimensional subspace. The SMPL model (Loper et al. 2015) learns
corrective blendshape models from shape samples and has been extended to syn-
thesize dynamic deformation by incorporating the autoregressive model. These
methods successfully produce realistic deformation of human skin. We have pro-
posed an example-based method for controlling the helper bones to mimic the
secondary dynamics of soft tissues (Mukai and Kuriyama 2016) while neglecting
the effect of gravity and the interactions with other objects.
References
Angelidis A, Singh K (2007) Kinodynamic skinning using volume-preserving deformations. In:
Proceedings of ACM SIGGRAPH/Eurographics symposium on computer animation 2007, pp
129–140
Baran I, Popović J (2007) Automatic rigging and animation of 3d characters. ACM Trans Graph
26(3):72:1–72:8
Cooper S, Hertzmann A, Popović Z (2007) Active learning for real-time motion controllers. ACM
Trans Graph 26(3):5
Fan Y, Litven J, Pai DK (2014) Active volumetric musculoskeletal systems. ACM Trans Graph
33(4):152
Grassia FS (1998) Practical parameterization of rotations using the exponential map. Graph Tool
3(3):29–48
Hahn F, Martin S, Thomaszewski B, Sumner R, Coros S, Gross M (2012) Rigspace physics. ACM
Trans Graph 31(4):72:1–72:8
Hahn F, Thomaszewski B, Coros S, Sumner R, Markus G (2013) Efficient simulation of secondary
motion in rig-space. In: Proceedings of ACM SIGGRAPH/Eurographics symposium on com-
puter animation 2013, pp 165–171
Horn BKP (1987) Closed-form solution of absolute orientation using unit quaternions. J Opt Soc
Am A 4(4):629–642
Jacobson A, Sorkine O (2011) Stretchable and twistable bones for skeletal shape deformation.
ACM Trans Graph 30(6):Article 165
Jacobson A, Baran I, Popović J, Sorkine O (2011) Bounded biharmonic weights for real-time
deformation. ACM Trans Graph 30(4):78:1–78:8
James DL, Twigg CD (2005) Skinning mesh animations. ACM Trans Graph 24(3):399–407
Kavan L, Sorkine O (2012) Elasticity-inspired deformers for character articulation. ACM Trans
Graph 31(6):Article 196
20 T. Mukai
Kavan L, Collins S, Zara J, O’Sullivan C (2007) Skinning with dual quaternions. In: Proceedings of
ACM SIGGRAPH symposium on interactive 3D graphics 2007, pp 39–46
Kavan L, Sloan PP, O’Sullivan C (2010) Fast and efficient skinning of animated meshes. Comput
Graph Forum 29(2):327–336
Kim J, Kim CH (2011) Implementation and application of the real-time helperjoint system. In:
Game developers conference 2011
Kry PG, James DL, Pai DK (2002) Eigenskin: real time large deformation character skinning in
hardware. In: Proceedings of ACM SIGGRAPH/Eurographics symposium on computer anima-
tion 2002, pp 153–159
Kurihara T, Miyata N (2004) Modeling deformable human hands from medical images. In: Pro-
ceedings of ACM SIGGRAPH/Eurographics symposium on computer animation 2004, pp
355–363
Le BH, Deng Z (2012) Smooth skinning decomposition with rigid bones. ACM Trans Graph 31(6):
Article 199
Le BH, Deng Z (2014) Robust and accurate skeletal rigging from mesh sequences. ACM Trans
Graph 33(4):1–10
Lewis JP, Cordner M, Fong N (2000) Pose space deformation: a unified approach to shape
interpolation and skeleton-driven deformation. In: Proceedings of SIGGRAPH 2000, pp
165–172
Li D, Sueda S, Neog DR, Pai DK (2013) Thin skin elastodynamics. ACM Trans Graph 32(4):49
Loper M, Black NMMJ (2014) Motion and shape capture from sparse markers. ACM Trans Graph
33(6):220:1–220:13
Loper M, Mahmood N, Romero J, Pons-Moll G, Black MJ (2015) SMPL: a skinned multi-person
linear model. ACM Trans Graph 34(6):248:1–248:16
Magnenat-Thalmann N, Laperrière R, Thalmann D (1988) Joint-dependent local deformations for
hand animation and object grasping. In: Proceedings on graphics interface’88, pp 26–33
Merry B, Marais P, Gain J (2006) Animation space: a truly linear framework for character
animation. ACM Trans Graph 25(6):1400–1423
Miller C, Arikan O, Fussell DS (2011) Frankenrigs: building character rigs from multiple sources.
IEEE Trans Vis Comput Graph 17(8):1060–1070
Mohr A, Gleicher M (2003) Building efficient, accurate character skins from examples. ACM Trans
Graph 22(3):562–568
Mukai T (2015) Building helper bone rigs from examples. In: Proceedings of ACM SIGGRAPH
symposium on interactive 3D graphics and games 2015, pp 77–84
Mukai T, Kuriyama S (2016) Efficient dynamic skinning with low-rank helper bone controllers.
Neumann T, Varanasi K, Hasler N, Wacker M, Magnor M, Theobalt C (2013) Capture and statistical
modeling of arm-muscle deformations. Comput Graph Forum 32(2):285–294
Park SI, Hodgins JK (2008) Data-driven modeling of skin and muscle deformation. ACM Trans
Graph 27(3):Article 96
Parks J (2005) Helper joints: advanced deformations on runtime characters. In: Game developers
conference 2005
Pons-Moll G, Romero J, Mahmood N, Black MJ (2015) Dyna: a model of dynamic human shape in
motion. ACM Trans Graph 33(4):120:1–120–10
Pulli RYWK, Popović J (2007) Real-time enveloping with rotational regression. ACM Trans Graph
26(3):73
Rumman NA, Fratarcangeli M (2015) Position-based skinning for soft articulated characters.
Comput Graph Forum 34(6):240–250
Shi X, Zhou K, Tong Y, Desbrun M, Bao H, Guo B (2008) Example-based dynamic skinning in real
time. ACM Trans Graph 27(3):29:1–29:8
Sloan PPJ, Rose CF, Cohen MF (2001) Shape by example. In: Proceedings of ACM SIGGRAPH
symposium on interactive 3D graphics 2011, pp 135–143
Tibshirani R (2011) Regression shrinkage and selection via the lasso: a retrospective. J R Stat Soc
Ser B (Stat Methodol) 73(3):273–282
Wang XC, Phillips C (2002) Multi-weight enveloping: least-squares approximation techniques for
skin animation. In: Proceedings of ACM SIGGRAPH/Eurographics symposium on computer
animation, pp 129–138
Crowd Formation Generation and Control
Jiaping Ren, Xiaogang Jin, and Zhigang Deng
Abstract
Crowd formation transformation simulates crowd behaviors from one formation
to another. This kind of transformation has often been used in animation films,
group calisthenics performance, video games, and other special effect applica-
tions. Given a source formation and a target formation, one intuitive approach to
achieve this kind of transformation between two formations is to establish the
source point and the destination point for each agent and plan the trajectory for
each agent while maintaining collision free maneuvers. Crowd formation gener-
ation and control usually consists of five different parts: formation sampling, pair
assignment, trajectory generation, motion control, and evaluation. In this chapter,
we will describe the involved techniques from abstract user input to collective
crowd formation transformations.
Keywords
Crowd simulation • Motion control • Motion transition • Evaluation
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Formation Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Formation Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Pair Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Trajectory Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Motion Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
J. Ren (*) • X. Jin

State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China
e-mail: ren_jia_ping@qq.com; jin@cad.zju.edu.cn
Z. Deng
Department of Computer Science, University of Houston, Houston, TX, USA
e-mail: zdeng4@uh.edu

DOI 10.1007/978-3-319-30808-1_15-1
2 J. Ren et al.
Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Introduction
In recent years, simulation of group formation transformation has been increasingly

used in feature animation films, video games, mass performance rehearsal, tactical
arrangements of players for sports teams training, and so on. Furthermore, group
formation generation and control can also find its wide applications in many other
scientific and engineering fields including but not limited to robot control, multi-
agent systems, and behavioral biology.
In most crowd simulation systems, each agent intelligently moves toward its
destination through navigational pathfinding algorithms and avoids collisions with
other agents and obstacles through local behavior control models. In collective
crowd formation transformation, we should also consider the behaviors of the entire
group.
When simulating a crowd, one intuitive way to form a target formation is to
provide each agent’s desired position at a particular moment and generate transitions
between that position and the destination. However, users must manually specify
many spatial, temporal, and correspondence constraints, which is time-consuming
and nontrivial, particularly when the crowd includes many agents that change their
locations frequently. Similarly, when a group of people perform a collective action
simultaneously in the real world, it is generally impossible for their commander or
team leaders to convey detailed movement information such as every group mem-
ber’s position at every time instance. So there is a need to find automatically methods
to solve these problems.
How to transfer abstract user inputs to the expression that we can deal with easily?
How to choose the destination for each agent? When finding the path for each agent,
which path is the best? And, how to control the formation efficiently? To answer
these questions, we decompose collective crowd formation transform into five
different steps. Suppose that the user specifies the source and target formation shapes
with sketches or contours, we first sample the agents in the formations. Then, we
assign the source position and the target position for each agent. We use crowd
simulation methods to generate realistic collision-avoidance trajectories of agents.
We also introduce some methods to control the movements of agents and high-level
transformation for crowds. We consider the approaches to measure the crowd
simulation and transformation results.
Organization The rest of the chapter is organized as follows. We give an overview

of prior work in section “State of the Art.” In section “Formation Sampling,” we
introduce how to sample the source and target formations given by users. In section
“Pair Assignment,” we assign corresponding positions in both source and target
Crowd Formation Generation and Control 3
formations for each agent. In section “Trajectory Generation,” we generate trajecto-

ries for each agent from the source formation to the target formation. In section
“Motion Control,” five methods about motion control are introduced. In section
“Evaluation,” we give six approaches to evaluate the results of crowd formation
transformation. We conclude and discuss future work in section “Summary.”
State of the Art
Numerous crowd simulation and modeling approaches have been developed during
the past several decades. Here we briefly review recent efforts about crowd simula-
tion, formation transformation, and evaluation.
For crowd simulation, there are two major kinds of models: rule-based models
and force-based models. Rule-based crowd models are flexible to simulate various
crowd agents through a set of delicately designed rules. The seminal work by
Reynolds (1987) presented the concept of Boids that simulates flocks of birds and
schools of fishes via several simple yet effective steering behavioral rules to keep the
group cohesion, alignment, and separation, as well as avoid collisions between
group members. Recently, Klotsman and Tal (2012) provided a biologically moti-
vated rule-based artificial bird model, which produces plausible and realistic line
formations of birds.
A distinct research line of crowd simulations is force-based models, originally
developed from human social force study by Helbing and Molnr (1995). Later, it was
further applied and generalized to other simulation scenarios such as densely
populated crowds (Pelechano et al. 2007), simulation of pedestrian evolution
(Lakoba et al. 2005), and escape panic (Helbing et al. 2000).
Group formation control is a vital collective characteristic of many crowds.
Existing approaches typically combine heuristic rules with explicit hard constraints
to produce and control sophisticated group formations. For example, Kwon et al.
(2008) proposed a framework to generate aesthetic transitions between key crowd
formation configurations. A spectral-based group formation control scheme
(Takahashi et al. 2009) was also proposed. However, in these approaches, exact
agent groups distributions at a number of key frames need to be specified by users.
Gu and Deng (Gu and Deng 2011, 2013; Xu et al. 2015) proposed an interactive and
scalable framework to generate arbitrary group formations with controllable transi-
tions in a crowd. Henry and colleagues (Henry et al. 2012, 2014) proposed a single-
pass algorithm to control crowds using a deformable mesh, and this approach can be
used to control crowd environment interaction and obstacle avoidance. In addition,
they proposed an alternative metric for use in a pair assignment approach for
formation control that incorporates environment information. These approaches
either need nontrivial manual involvements (Kwon et al. 2008; Takahashi et al.
2009) or are focused on intuitive user interfaces for formation control and interaction
(Gu and Deng 2011, 2013; Henry et al. 2012, 2014; Xu et al. 2015).
Many approaches are proposed to evaluate the results or improve the accuracy of
multiagent and crowd simulation algorithms. Most of them perform evaluation by
4 J. Ren et al.
comparing the algorithms’ output with real-world sensor data. Pettré et al. (2009)
compute appropriate parameters based on Maximum Likelihood Estimation. Lerner
et al. (2009) annotate pedestrian agent trajectories with action-tags to enhance their
natural appearance or realism. Guy et al. (2012) propose an entropy-based evaluation
approach to quantify the similarity between real-world and simulated trajectories.
Formation Generation
Formation Sampling
In this section, we describe how to generate the source and target formation based on
agents from user inputs. Different assumptions about the inputs by users correspond
to different sampling strategies. Here we mainly introduce two formation sampling
strategies based on different user inputs.
One assumption supposes that the user gives contour shapes, such as squares or
circles, or brush paintings (Gu and Deng 2013). This work proposes a unified
formation shape representation called a formation template, i.e., an oversampled
point space with a roughly even distribution, and it offers an interactive way for users
to draw the input formations. In this case, to generate a visually balanced target
formation template, it first evenly samples the points on the boundaries, and then fills
the area between the inclusive and exclusive boundaries through an extended
scanline flood-fill algorithm. In addition, it uses a filling algorithm to discretize the
grid space. To avoid sampling points too close to the boundaries, it checks four
points with constant offsets (top, bottom, left, and right) to the current checkpoint.
Another assumption supposes that the user specifies the source and target forma-
tion shapes with sketches and the number of agents in the formation (Xu et al. 2015).
In order to automatically use a user-specified number of agents to form a specified
formation shape with a visually natural distribution, the sampling process from the
formation shape is mainly divided into the following two stages. First, it uses a
simple way to tentatively sample the approximation of the parameterized number of
formation points in the formation shape with a roughly even distribution. However,
the sampled result is very rigid, lacking the aesthetic effect. More importantly, it is
difficult for this method to accurately sample a user-specified number of formation
points. Therefore, users have to tune the number of sampled points to be exactly
equivalent to the user-specified number by randomly deleting some sampled points
or adding some formation points according to the Roulette Wheel Selection strategy.
Then a corresponding agent is located at the location of a formation point.
Pair Assignment
After the sampling process, to generate the formation transformation, we need to pair
the agents in the source formation with those in the target formation. We introduce
two methods to solve this pair assignment problem in this section.
Fig. 1 Generating arbitrary formations from random agent distributions (left-most) through inclu-
sion sketch (white boundaries) and exclusion sketch (red boundaries) in Gu and Deng (2011)
The method presented in (Gu and Deng 2011, 2013) estimates the agent distri-
bution in the target formation to find the correspondence between any agent in the
initial formation and its appropriate candidate position in the formation template (see
Fig. 1). Using formation coordinates, it designs a pair assignment algorithm based on
two key heuristics. First, in the target formation, boundary agents should closely fit
the boundary curves to clearly exhibit the user-specified formation shape. Second,
each nonboundary agent should keep its adjacency condition as much as possible.
This algorithm first finds correspondences for the boundary agents, and then finds
correspondences for the nonboundary agents. To find correspondences for the
boundary agents, it converts the positions of all the agents in the initial distribution
into formation coordinates and subtracts the formation orientation from each agent’s
relative direction to yield the relation agent direction. It stores this direction along
with the relative agent distance in a KD-tree data structure. This approach performs
the same operations for each point on the target formation template boundaries. Thus
it can efficiently compute the agent corresponding to each boundary point by finding
the nearest neighbor in the KD-tree. To find correspondences for the nonboundary
agents, it identifies the corresponding template point for each nonboundary agent
that was not selected in the previous step. Similarly, it uses each agent’s formation
coordinate to find the closest inner template point. The approach further transforms
that point to its world coordinate representation, i.e., the agent’s target position.
In the method presented in Xu et al. (2015), Delaunay Triangulation is employed
to represent the relationship among adjacent agents. Pair assignment can be formu-
lated as the problem of building a one-to-one correspondence between the source
point set and the target point set (see Fig. 2). It can be further formulated as finding
an optimal assignment in a weighted bipartite graph. In the matching process, they
apply a novel match measure to effectively minimize the overall disorder including
the variations of both time synchronization and local structure. That is to minimize
the distance from source to target for each agent, and the average distances to
neighbors are similar between source formation and target formation. Finally, this
method applies the classical Kuhn-Munkres algorithm (Kuhn 1955; Munkres 1957)
to solve the pair assignment problem.
6 J. Ren et al.
Fig. 2 Delaunay triangulation and pair assignment in Xu et al. (2015)
Trajectory Generation
In this section, we address how each agent reaches its destination at every time step
after determining its target position. Six different simulation approaches are
described in the following.
Reciprocal Velocity Obstacles (RVO) (Van den Berg et al. 2008) which extends
the Velocity Obstacle concept (Fiorini and Shiller 1998) is a widely used velocity-
based crowd simulation model. This method introduces a new concept for local
reactive collision avoidance. The only information each agent is required to have
about the other agents is their current position and velocity and their exact shape. The
basic idea is that: instead of choosing a new velocity for each agent that is outside the
other agent’s velocity obstacle, this method chooses a new velocity that is the
average of its current velocity and a velocity that lies outside the other agent’s
velocity obstacle.
Optimal reciprocal collision avoidance (ORCA) (Van Den Berg et al. 2011) is a
revised model derived from the RVO, and it presents a rigorous approach for
reciprocal n-body collision avoidance that provides a sufficient condition for each
agent to be collision-free for at least a fixed amount of time into the future, only
assuming that the other robots use the same collision-avoidance protocol. There are
infinitely many pairs of velocity sets that make two agents avoiding collision, but
among those, it selects the pair maximizing the amount of permitted velocities
“close” to the optimized velocities for two agents. For n-body collision-avoidance,
each agent performs a continuous cycle of sensing the acting within the time step. In
each iteration, the agent acquires the radius, the current position, and the current
optimization velocity of the other robots. Based on this information, the agent infers
the permitted half-plane of velocities that make the agent avoiding collision with
other agents. The set of velocities that are permitted for the agent with respect to all
agents is the intersection of the half-planes of permitted velocities induced by each
other agent. Then the agent chooses a new velocity that is closest to its preferred
velocity among all velocities inside the region of permitted velocities.
The method in Gu and Deng (2011) can automatically compute the desired
position of each agent in the target formation and generate the agent correspondences
between key frames. The force that drives an agent from its original position to its
estimated target position can simply be the direction vector between the two
positions. However, this force only considers the group formation factor. In a
dense group, such a pure formation-driven strategy cannot fully avoid agent colli-
sions. As such, a local collision model is needed to refine within-group collision
avoidance. To this end, the authors employ a force-based model (Pelechano et al.
2007) for the collision avoidance task due to its capability of handling very high-
density crowds. Because this model takes into account the collision avoidance,
forces and repulsion force from neighboring group members and obstacles.
In Gu and Deng (2013), local formation transition is the transition from one
formation to another without considering the whole group’s general locomotion. In
addition to considering to compute a linear interpolation from an agent’s initial
position to its estimated target position, this method also consider an extra repulsion
force to avoid collision. Without user interactions, each agent would go straight to
the target position with minor transition adjustments on the way to avoid local
collisions with other agents.
For an agent, the method (Xu et al. 2015) locally adjusts its trajectory by applying
social forces such as driving and repulse forces to navigate and avoid collisions. A
mutual information-based method is introduced, which is a well-known concept in
the field of information theory, and it is designed to quantify the mutual dependence
between two random variables. Mutual information is somewhat correlated with the
fluency and stability of agent subgroup’s localized movements in a crowd. In their
method, the mutual information between direction and position and the mutual
information between velocity and position are used to adjust the ideal heading and
desired velocity in the basic social force model.
The online real-time motion synthesis method (Han et al. 2016) transforms the
initial motion automatically using the following parameters: target turning angle of
the agent, the target scaling factor for the moving speed with respect to that of the
source formation, and time required to achieve the target formation. This method
builds interpolation functions of turning angle and scaling factor for each agent, thus
contains the velocity and position for each agent in every frame.
Motion Control
As we can generate the trajectory for each agent from one formation to another, in
this section, we want to address the following problem: How to make the transfor-
mation more reliable, more controllable? Here, we describe five different methods to
solve this problem.
The method presented in Gu and Deng (2011) is a two-level hierarchical group
control, that is, breaking the full group dynamics into within-group dynamics and
intergroup dynamics. Xu et al. (2015) extend the adapted social force method with a
subgroup formation constraint (see Fig. 3). This method clusters individual agents in
a crowd into subgroups to maximally maintain the formation of the collective
subgroups during the formation transition. An Affinity Propagation
(AP) clustering algorithm (Frey and Dueck 2007) is used by Xu et al. (2015). The
AP algorithm identifies a set of exemplars to best represent agents’ positions in the
formation. They choose the AP algorithm to cluster agents since an exemplar can be
8 J. Ren et al.
Fig. 3 The adapted social

force method with a subgroup
Mutual Information
formation constraint of Xu Movement
et al. (2015)
Control
Subgroup Clustering
conceptually considered to represent the overall movement of its corresponding

agent-subgroup, and the cluster number is determined adaptively and automatically.
The measure for similarity is the local relative distance variance for the collective
subgroup clustering.
In the method presented in Gu and Deng (2013), the authors construct a second
virtual local grid field to evaluate a flow vector to guide transitions, for the need of
implementing a splitting and merging transition. In order to form a user-defined
formation (see Fig. 4) while moving as a whole to other location, this method
introduce three factors to the group level: the global navigation vector heading to
the target formation’s location, the velocity driven by global collision avoidance
between different groups, and the user-guided velocity computed from the sketching
interface.
Vector field is introduced (Jin et al. 2008) to guide agents’ movements (see
Fig. 5). A vector field can be considered as position-to-vector mapping in the
problem domain.
A physics-based predictive motion control is described in Han et al. (2014). It first
generates a reference motion automatically at run-time based on existing data-driven
motion synthesis methods. Given a reference motion, it repeatedly generates an
optimal control policy for a small time window that spans a couple of footsteps in the
reference motion through model predictive control while shifting the window along
the time axis, which supports an on-line performance.
Evaluation
As there are lots of methods for crowd formation transformation, which one is better
in one definite situation? Here we introduce six different approaches to evaluate
transformation results.
The visual method is the most direct and basic way to judge whether an
animation is aesthetic or not. However, the results by visual methods are subjective
by nature, and quantitative approaches are more reliable and objective for
researchers.
Fig. 4 Formation transitions with trajectory controls of Gu and Deng (2013)
Fig. 5 Agents’ movements guided by a vector field of Jin et al. (2008)
The simulation time consumption is a generally used measure of quantifying the

performance of simulation methods. With the same function, people prefer the
method that has a lower time consumption. Especially in interactive applications,
run-time performance is required for a good user experience.
The mutual information introduced by Xu et al. (2015) can be adapted to
measure the aesthetic aspect of a crowd formation transform, as well as how to
compute the mutual information of a dynamics crowd.
The stability of local structure Xu et al. (2015), just as its name implies, is a
measure of the local stability for transformation. It uses the standard deviation and
the average value to quantify the stability property of local structure during a crowd
formation transformation. The stability of local structure is the standard deviation of
minimum neighbor distance for agents. Clearly, a lower value of the standard
deviation indicates that the agents have more similar distances from their neighbors
and vice versa.
The effort balancing (Xu et al. 2015) is a metric to measure the synchronization
of the transformation and employs the standard deviation and the average value to
estimate the balancing of the agents’ efforts. When a formation transformation is
smooth and visually pleasing, for any agent, we anticipate that the effort from its
source position to its current position is not only the least but also balanced. Effort
balancing is the standard deviation of distance to agents’ source position and their
current position.
10 J. Ren et al.
A data-driven quantitative approach (Ren et al. 2016; Wang et al. 2015) is

presented to evaluate collective dynamics models by using real-world trajectory
datasets. It is possible that two different groups with noisy trajectories may exhibit
similar behaviors even when their trajectory positions are quite different. This
approach uses discrete probability density distribution functions that are generated
from the time-varying metrics and reflects the global characteristics of groups. The
influence of a small amount of data abnormality or noise can be ignored. It intro-
duces seven time-varying metrics: the velocity, the acceleration, the angular velocity,
the angular acceleration, the Cartesion jerk, the shortest distance, and the velocity
difference. The evaluation model is related to the sum of the differences in discrete
PDF between the real-world data and the simulation data for the seven metrics. To
compare different simulation methods, the overall evaluation of this approach is the
iterations of two components: optimizing the dynamics model parameters and
optimizing the weights of seven energy terms. This method uses a genetic algorithm
to compute the optimal parameters by maximizing the evaluation model and intro-
duces entropy to compute the weights of seven time-varying metrics.
Summary
We have discussed the processes involved in collective crowd formation transfor-

mation in this chapter.
In formation sampling level, we describe two formation sampling strategies. The
first strategy samples on the input sketches with a roughly even distribution and then
adopts Roulette Wheel Selection to relate the agents to their certain positions. The
second strategy fills the boundary and considers the inner area. In pair assignment
level, we introduce two strategies to find the correspondence between an agent’s
original position and its new position. One method finds correspondences for the
boundary agents followed by the nonboundary agents. The other method presents a
measure to minimize the overall disorder including the variations of both time
synchronization and local structure and transfers the pair assignment problem into
an optimization problem. In the trajectory generation level, we introduce six
methods to guide the movement of each agent in the formation. The main concern
is about collision-avoidance. In the motion control level, we describe five methods to
make the results more reliable and controllable. In the evaluation level, we introduce
six different approaches to evaluate the simulation methods, the transformation
results, and the execution performance, and most of them are quantitative methods.
Limitation and Future Work Although various methods can be applied into the
crowd formation transformation problem, these approaches still cannot handle
complex scenarios with obstacles and the multiformation’s mixture and separation.
Moreover, current approaches suppose that the collective crowds are moving on the
two dimensions. In the future, we can improve the existing methods by dealing with
complex obstacles, considering interactions among different groups, extending the
2D methods to 3D, and changing target formation dynamically.
Cross-References
▶ Crowd Evacuation Simulation

▶ Functional Crowds
▶ Optimal Control Modeling of Human Movement
▶ Segmental Movements in Cycling
References
van den Berg J, Lin M, Manocha D (2008) Reciprocal velocity obstacles for real-time multi-agent
navigation. In: Robotics and automation, 2008. ICRA 2008. IEEE international conference on
Robotics and Automation, IEEE, pp. 1928–1935
van den Berg J, Guy SJ, Lin M, Manocha D (2011) Reciprocal n-body collision avoidance. In:
Robotics research, Springer, pp 3–19
Fiorini P, Shiller Z (1998) Motion planning in dynamic environments using velocity obstacles. Int J
Rob Res 17(7):760–772
Frey BJ, Dueck D (2007) Clustering by passing messages between data points. Science
315(5814):972–976
Gu Q, Deng Z (2011) Formation sketching: an approach to stylize groups in crowd simulation. In:
Proceedings of graphics interface 2011, Canadian Human-Computer Communications Society,
pp 1–8
Gu Q, Deng Z (2013) Generating freestyle group formations in agent-based crowd simulations.
IEEE Comput Graph Appl 33(1):20–31
Guy SJ, van den Berg J, Liu W, Lau R, Lin MC, Manocha D (2012) A statistical similarity measure
for aggregate crowd dynamics. ACM Trans Graph 31(6):190:1–190:11
Han D, Noh J, Jin X, S Shin J, Y Shin S (2014) On-line real-time physics-based predictive motion
control with balance recovery. Comput Graphics Forum 33:245–254. Wiley Online Library
Han D, Hong S, Noh J, Jin X, Shin JS (2016) Online real-time locomotive motion transformation
based on biomechanical observations. Comput Anim Virtual Worlds 27(3–4):378–384
Helbing D, Molnar P (1995) Social force model for pedestrian dynamics. Phys Rev E 51(5):4282
Helbing D, Farkas I, Vicsek T (2000) Simulating dynamical features of escape panic. Nature
407(6803):487–490
Henry J, Shum HP, Komura T (2012) Environment-aware real-time crowd control. In: Proceedings
of the 11th ACM SIGGRAPH/Eurographics conference on computer animation, Eurographics
Association, pp 193–200
Henry J, Shum HP, Komura T (2014) Interactive formation control in complex environments. IEEE
Trans Vis Comput Graph 20(2):211–222
Jin X, Xu J, Wang CC, Huang S, Zhang J (2008) Interactive control of large-crowd navigation in
virtual environments using vector fields. IEEE Comput Graph Appl 28(6):37–46
Klotsman M, Tal A (2012) Animation of flocks flying in line formations. Artif Life 18(1):91–105
Kuhn HW (1955) The hungarian method for the assignment problem. Naval Res Logist Q
2(1–2):83–97
Kwon T, Lee KH, Lee J, Takahashi S (2008) Group motion editing. ACM Trans Graph 27:80
Lakoba TI, Kaup DJ, Finkelstein NM (2005) Modifications of the helbing-molnar-farkasvicsek
social force model for pedestrian evolution. Simulation 81(5):339–352
Lerner A, Fitusi E, Chrysanthou Y, Cohen-Or D (2009) Fitting behaviors to pedestrian simulations.
In: Proceedings of the 2009 ACM SIGGRAPH/Eurographics symposium on computer anima-
tion, ACM, pp 199–208
Munkres J (1957) Algorithms for the assignment and transportation problems. J Soc Ind Appl Math
5(1):32–38
12 J. Ren et al.
Pelechano N, Allbeck JM, Badler NI (2007) Controlling individual agents in high-density crowd
simulation. In: Proceedings of the 2007 ACM SIGGRAPH/Eurographics symposium on com-
puter animation, Eurographics Association, pp 99–108
Pettré J, Ondřej J, Olivier AH, Cretual A, Donikian S (2009) Experiment-based modeling, simu-
lation and validation of interactions between virtual walkers. In: Proceedings of the 2009 ACM
SIGGRAPH/Eurographics symposium on computer animation, ACM, pp 189–198
Ren J, Wang X, Jin X, Manocha D (2016) Simulating flying insects using dynamics and data-driven
noise modeling to generate diverse collective behaviors. PLoS One 11(5):e0155698
Reynolds CW (1987) Flocks, herds and schools: a distributed behavioral model. ACM SIGGRAPH
Takahashi S, Yoshida K, Kwon T, Lee KH, Lee J, Shin SY (2009) Spectral-based group formation
control. Comput Graphics Forum 28: 639–648. Wiley Online Library
Wang X, Ren J, Jin X, Manocha D (2015) Bswarm: biologically-plausible dynamics model of insect
swarms. In: Proceedings of the 14th ACM SIGGRAPH/Eurographics symposium on computer
animation, ACM, pp 111–118
Xu M, Wu Y, Ye Y, Farkas I, Jiang H, Deng Z (2015) Collective crowd formation transform with
mutual information–based runtime feedback. Comput Graphics Forum 34:60–73. Wiley Online
Library
Functional Crowds
Jan M. Allbeck
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Animation to AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
HCI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Abstract
Most crowd simulation research either focuses on navigating characters through
an environment while avoiding collisions or on simulating very large crowds.
Functional crowds research focuses on creating populations that inhabit a space
as opposed to passing through it. Characters exhibit behaviors that are typical for
their setting, including interactions with objects in the environment and each
other. A key element of this work is ensuring that these large-scale simulations are
easy to create and modify. Automating the inclusion of action and object seman-
tics can increase the level at which instructions are given. To scale to large
populations, behavior selection mechanisms must be kept relatively simple and,
to demonstrate typical human behavior, must be based on sound psychological
models. The creation of roles, groups, and demographics can also facilitate
behavior selection. The simulation of functional crowds necessitates research in
animation, artificial intelligence, psychology, and human-computer interaction
(HCI). This chapter provides a brief introduction to each of these elements and
their application to functional crowds.
J.M. Allbeck (*)

George Mason University, Fairfax, VA, USA
e-mail: jallbeck@gmu.edu

DOI 10.1007/978-3-319-30808-1_16-1
2 J.M. Allbeck
Keywords
Crowd simulation • Virtual humans • Patterns of life • Computer animation • AI
Introduction
Virtual humans can be used as stand-ins when using real humans would be too
dangerous and cost-prohibitive or precise control is required. Virtual humans are
often used as extras or background characters in movies and games (see Fig. 1). They
are similarly used in virtual training scenarios for military personnel and first
responders. They can also be used to analyze urban and architectural design as
well as various policies and procedures. For many of these applications and others,
the virtual humans must both reflect typical or normal human behavior and also be
controllable or directable. Furthermore, in order to create sizeable crowds of virtual
humans functioning in rich virtual environments, they must have relatively simple
behavior selection mechanisms.
Functional crowds, in contrast to more typical crowd simulations, depict ani-
mated characters interacting with the environment in meaningful ways. They do not
simply walk from one location to another avoiding obstacles. They perform the same
behaviors we see from real humans every day, as well as not so typical behaviors that
might be required for the application.
The first element needed to achieve functional crowds is animation. Traditional
crowd simulations focus on walking animation clips, perhaps with a few idle
behaviors or depending on the application some battle moves. There is little or no
interaction with objects in the environment. Animating virtual humans manipulating
objects can be quite challenging. It involves detection of collisions and fine motor
movements. We will give an overview of some of these challenges and approaches
for solving them in this chapter.
Fig. 1 Virtual characters in a scene in the unreal game engine

Functional Crowds 3
Another required element relates to providing the virtual humans with the knowl-
edge of what actions can be performed and what objects are required to perform
them. If we are going to eat, we need an object corresponding to food to eat. We may
optionally need instruments such as utensils. A lot of this needed information could
be considered commonsense, but unless explicitly supplied to the virtual humans,
they lack it. This information is also needed as input to higher-level artificial
intelligence mechanisms such as behavior selectors, planners, and narrative
generators.
Functional crowds should also depict a heterogeneous population. In real life not
everyone does the same thing, they do not have the same priorities, and they do not
all perform tasks in exactly the same way. Some of these variations stem from prior
observations and experiences. They are learned. Some stem from psychological
states and traits, such as personalities and emotions.
Finally, many, if not all, applications of functional crowds require some human-
computer interaction (HCI). This interaction may come during the authoring of the
crowd behavior. The application or scenario may require some of the behaviors to be
more tightly controlled or even scripted. The application may also require users (e.g.,
players, trainees, evaluators, etc.) to interact with the crowd during the simulation.
These interactions may simply require the virtual humans to avoid collisions with the
real human’s avatar, or they may require communication and perhaps even cooper-
ation between the real and virtual humans.
This chapter will present these various elements of functional crowds and discuss
challenges and approaches to address them. We will start by providing a snapshot of
the current state of the art in related research fields. Then we will in turn discuss
issues related to AI and animation, psychological models, and HCI. Finally, we will
conclude with a brief summary and potential future direction.
State of the Art
In the past decade or so, crowd simulations have made enormous progress in the
number of characters that can be simulated and in creating more natural behaviors.
More detailed analysis of crowd simulation research can be found in a number of
published volumes (Kapadia et al. 2015; Pelechano et al. 2008, 2016; Thalmann
et al. 2007). It is now possible to simulate over one million characters in real time in
high-density crowds.
Crowd simulations can also be more heterogeneous. Not every character looks or
behaves exactly the same. Certainly some variations stem from differences in
appearance and motion clips (Feng et al. 2015; McDonnell et al. 2008). Others
come from psychological models such as emotion and personality (Balint and
Allbeck 2014; Durupinar et al. 2016; Li et al. 2012). Most crowd simulations assign,
fairly randomly, starting positions and ending destinations for the characters in the
simulation. While this appears fine for short durations at a distance, if a player
follows a character for a period of time, it quickly appears false. Sunshine-Hill and
4 J.M. Allbeck
Badler have created a framework for generated plausible destinations for characters
on the fly to provide reasonable “alibis” for them (Sunshine-Hill and Badler 2010).
Simulating functional crowds also requires other advanced computer graphics
techniques. Commercial game engines, such as Unity ® and Unreal ®, provide much
of the technology necessary. In the past couple of years, they have both changed their
licensing structure in ways that enable researchers to take advantage of and add to
their capabilities. Other needed advancements come from the animation research
community. A key feature of functional crowds is the ability of characters to interact
with objects in their environment in meaningful ways. We require animations of
characters sitting and eating food, getting in and out of vehicles, conversing with one
another, displaying emotions, getting dressed, etc. (Bai et al. 2012; Clegg et al. 2015;
Hyde et al. 2016; Shapiro 2011).
Animation to AI
To simulate a functional crowd, we need the characters to interact with their object-
rich environments and with each other. While great work has been done in pathfind-
ing, navigation, and path following, additional advancements are still needed
(Kapadia et al. 2015; Pelechano et al. 2016). Characters still struggle to get through
cluttered environments with narrow walkways. We need to give characters enhanced
abilities to turn sideways, sidestep, and even back up in seamless natural motions.
Furthermore, characters need to be able to grab, carry, place, and use objects of
different shapes and sizes and do so when the objects are placed at various locations
in the world and approached from any direction. The core of motions for characters
is generally generated in one of three ways: key framed, motion capture, or proce-
dural. Artist created key-framed and motion-captured motions that tend to look
natural and expressive, but lack the flexibility needed for most object interactions.
Procedurally generated motions use algorithms such as inverse kinematics that work
well to target object locations (e.g., for a reach and grab), but often lack a natural
look and feel and require objects to be labeled with sites, regions, and directions
referenced in the code. While progress continues in virtual human animation
research, natural-looking functional crowds will require even more advancement
to make the authoring and animating large populations of characters more feasible.
Once the characters can be animated interacting with objects in the environment,
they need to possess an understanding of what can be done with objects and what
objects are needed in order to perform various actions. In other words, they need to
understand object and action semantics. This includes knowing what world state
must exist prior to the start of an action (i.e., applicability conditions and preparatory
specifications), what state must hold during the action, what the execution steps of
the action are, and finally what the new world state is after the successful execution
of the action. As indicated previously, there also needs to be information about the
parts and various locations of the objects (e.g., grasp locations, regions to sit on (see
Fig. 2), etc.) so that animation components can be effective. Representations, such as
the Parameterized Action Representation (PAR), are designed to hold this
Functional Crowds 5
Fig. 2 Regions indicating

places where characters
could sit
information, but authoring them is time-consuming and error prone (Bindiganavale

et al. 2000).
In order to scale to the level needed to simulate functional crowds in large,
complex environments, the creation of action and object semantics needs to be
automated. Automating action and object semantics would also help to ensure
some consistency within and between scenarios, whereas ad hoc, handing authoring
tends to be sloppy and error prone. Online lexical databases, such as WordNet,
VerbNet, and FrameNet, have been shown to provide a viable foundation for action
and object semantics for virtual worlds (Balint and Allbeck 2015; Pelkey and
Allbeck 2014). Additional work is needed to ensure the information represented is
what is needed for the applications in virtual worlds and to ensure that mechanisms
for searching and retrieval are fast enough.
Given that characters have some basic understanding of the virtual world they are
inhabiting, the next question is at any given time, how should characters select their
behaviors? Planning and other sophisticated AI techniques can be computationally
intensive and difficult to control. For functional crowds, it would be better to start
with simple techniques both in authoring and execution (J. M. Allbeck 2009, 2010).
Human behaviors stem from a variety of different impetuses. Some behaviors, such
as going to work or school or attending a meeting, are scheduled. These actions
provide some structure to our lives and the lives of our virtual counterparts. They are
selected based on the simulated time of day. Reactive actions are responses to the
world around us. They add life and variation to the behaviors of virtual characters.
They are selected based on the objects, people, and events around the character.
Aleatoric or stochastic actions include sub-actions with different distributions. For
example, we may want a character to appear to be working in her office, but are not
very concerned with the details. Our WorkInOffice action would include sub-actions
like talking on the phone, filing papers, and using the computer. The character would
switch between these actions for the specified period of time at the specified
distribution, but what exact sub-action is being performed at any point in time
would not need to be specified. Need-based actions add humanity to the virtual
characters. Needs grow over time and are satisfied by performing certain actions
with the necessary object participants (e.g., eat food). As a need grows, the priority
6 J.M. Allbeck
of selecting a behavior that would fulfill it also grows. These needs could correspond
to established psychological models, such as Maslow’s hierarchy of needs, or they
could be specific to the scenario (e.g., drug addiction).
These are just a few examples of simple behavior selection mechanisms. Cer-
tainly, others are possible and may be more applicable to some scenarios. Practically
speaking, it may be best to completely script the behaviors of some key characters in
a scenario. Background characters could then have variations in their schedules,
reactions, needs and distributions. More sophisticated AI techniques could be
included when and where needed, as long as the overall framework remains fast
enough for human interaction.
Heterogeneity
In real human populations, not everyone is doing the same thing at the same time.
There are variations in behaviors that stem from different factors. The psychological
research committee has spent decades positing numerous models of personality,
emotions, roles, status, culture, and more. The virtual human research community
has taken these models as inspiration for computational models for virtual human
behaviors (Allbeck and Badler 2001; IVA 1998; Li and Allbeck 2011). Variations in
behavior and behavior selection can also evolve as the characters learn about and
from their environment and each other (Li and Allbeck 2012).
All of this research needs additional attention and revision. In particular, how
these different traits are manifest in expressive animation needs continuing work, as
does the interplay of psychological models. How does personality effect emotion
and the display of emotion? How does a character’s roles and changing role effect
emotional displays? Certainly culture and its impacts are not well modeled in virtual
humans. How do all of these psychological models influence a character’s priorities?
At any point in time, a character’s behavior selection should reflect what is most
important for them to achieve at that time. Their priorities can be influenced by any
number of factors. For functional crowds, it is important that priorities be weighed
quickly and behavior selection is not delayed by an overly complex psychological
framework. An open question for most scenarios is what parts of human behavior are
really important to model and what can be left out? It is possible that a fair amount of
just random choices would suffice for the majority of the characters a lot of the time,
but this depends on the duration of the simulation and how often the same character
or characters are encountered by the viewer.
HCI
Most applications of functional crowds require them to have some interaction with
real humans either during the authoring process and/or while the simulation unfolds.
Authoring the behavior of an entire population of characters from the ground up
would be infeasible. Providing a layer of automatically generated common
Functional Crowds 7
understanding (i.e., action and object semantics) does help. Simple, yet robust,
behavior selection mechanisms are also helpful. Furthermore, the action types
described earlier can be linked to even higher-level constructs, such as groups and
roles (Li and Allbeck 2011).
When authoring behaviors, it is important to balance autonomy and control. To
accomplish the objectives of the scenario, authors need to have control over the
characters and their behaviors. However, authoring every element of every behavior
of every character would be overwhelming even for short-duration simulations of
forty or fifty characters. The characters need to have some level of autonomy. They
need to be able to decide what to do and how to do it on their own. Then, when and if
they receive instruction from the simulation author, they need to suspend or preempt
their current behaviors to follow those instructions. There may also be times when
authors have an overall narrative in mind for the simulations, but are less concerned
about some of the details of the characters’ behaviors. This is one place where more
sophisticated AI methods like partial planners may play a role (Kapadia et al. 2015).
HCI also comes into play as one or more humans interact with the functional
crowds during the simulation. They may be using a standard keyboard, mouse, and
monitor. They may be using a mobile device. They may be using a gaming console.
Or they may be using more advanced virtual reality (VR) devices. VR devices can
provide a higher fidelity and therefore enable the subjects to see the virtual world in
more detail. Using head-mounted displays (HMD) or CAVE systems allows the
subject to view the virtual characters in a life-size format. The movements of
subjects can also be motion captured in real time and displayed on their avatar,
providing more realistic interaction with the virtual characters. Hardware interfaces
can impact the level of a subject’s immersion into a virtual world and potentially
their level of presence in the virtual world.
Another aspect of HCI with virtual characters and functional crowds is a kind of
history. If a subject spends longer durations in the virtual world and/or has repeated
exposure to it, he or she may become familiar with some of the characters and form
expectations about them. Subjects may learn their personality and behavioral quarks.
Subject will expect some consistency in these behaviors. They may also expect the
virtual characters to have some level of memory of past interactions. While these
expectations can be met, it is still a challenge to provide the virtual characters with
techniques that make these memories compact, efficient, and plausibly imperfect
(Li et al. 2013). More research is needed.
Conclusions
Functional crowds can increase the number of applications of crowd simulations and
increase their accuracy, but as this chapter has discussed, there is additional research
needed from character animation to AI to psychological models to HCI. Increased
computing power will help, but is not an overall solution. Attempting to simulate
realistic human behaviors is difficult. It is even more challenging at a large scale.
When attempting to simulate realistic human behavior, we can end up losing focus.
8 J.M. Allbeck
One model or technique leads us to another and another until we have lost sight of
our original goal. Too often researchers also design and implement a method and
then go in search of a problem it might address. We might be better served to keep
focused on an application or scenario and then determine what is and is not most
critical to achieving its goals. Does the application really require a sophisticated
planner or emotion model? How closely and for how long is the subject going to be
observing the characters’ behaviors? Also, do we really need to simulate 500,000
characters at a time? At ground level in the center of a village or even large city, how
many people can be seen at one time? Are there existing tools, open source or
commercial, that can be used or modified? Too often researchers feel they have to
construct their own models from scratch, ignoring years of effort done by others. In
terms of both human effort and computation, use available resources wisely and do
not put a large amount of effort into areas that will have little impact on the
application.
In this area of research, another question that is often asked is how do you
validate your model. How can one validate human behavior? We could show videos
of functional crowds to hundreds of people and ask them a variety of questions to try
to determine if they think the character behaviors are realistic, reasonable, or even
plausible, but we all have rather different ideas of what is reasonable behavior.
Instead we choose to framework work in this area as the construction of a toolset to
be used by subject matter experts to achieve their own goals. For example, an urban
planner may wish to use functional crowds to analyze a proposed transportation
system. Evaluate then becomes about whether or not the urban planner can use the
functional crowds toolset to do the desired analysis. Does it have the parameters
required? Is it usable by nonprogrammers? Can they increase or decrease fidelity
relative to the input effort?
As a research area, functional crowds is a young, but promising direction. It sits at
the overlap of several other research communities, namely, computer graphics and
animation, artificial intelligence, human-computer interaction, and psychology. As
advances are made in each one of these disciplines, functional crowds can benefit.
Cross-References
▶ Biped Controller for Character Animation

▶ Comparative Evalution of Crowd Animation
▶ Crowd Formation Generation and Control
▶ Data-Driven Hand Animation Synthesis
▶ Depth Sensor Based facial and Body Animation Control
▶ Example-Based Skinning Animation
▶ Eye Animation
Functional Crowds 9
▶ Head Motion Generation

▶ Laughter Animation Generation
▶ Perceptual Evaluation of Human Animation
▶ Perceptual Study on Facial Expressions
▶ Perceptual Understanding of Virtual Patient Appearance and Motion
▶ Physically-Based Character Animation Synthesis
▶ Real-time Full Body Motion Control
▶ Real-Time Full Body Pose Synthesis and Editing
▶ Visual Speech Animation
References
Allbeck JM (2009) Creating 3D animated human behaviors for virtual worlds. University of
Pennsylvania, Philadelphia, PA
Allbeck JM (2010) CAROSA: a tool for authoring NPCs. In: Presented at the international
conference on motion in games. Springer, pp 182–193.
Allbeck J, Badler NI (2001) Consistent Communication with Control. In: Workshop on non-verbal
and Verbal Communicative Acts to achieve contextual embodied agents at autonomous agents.
Bai Y, Siu K, Liu CK (2012) Synthesis of concurrent object manipulation tasks. ACM Trans Graph
31(6):156
Balint T, Allbeck JM (2014) Is that how everyone really feels? emotional contagion with masking
for virtual crowds. In: Presented at the international conference on intelligent virtual agents.
Springer, pp 26–35
Balint T, Allbeck JM (2015) Automated generation of plausible agent object interactions. In:
Presented at the international conference on intelligent virtual agents. Springer, pp 295–309
Bindiganavale R, Schuler W, Allbeck JM, Badler NI, Joshi AK, Palmer M (2000) Dynamically
altering agent behaviors using natural language instructions. In: Proceedings of the fourth
international conference on autonomous agents. ACM, New York, pp 293–300. doi:10.1145/
336595.337503
Clegg A, Tan J, Turk G, Liu CK (2015) Animating human dressing. ACM Trans Graph 34(4):116
Durupinar F, Gudukbay U, Aman A, Badler N (2016) Psychological parameters for crowd
simulation: from audiences to Mobs. IEEE Trans Vis Comput Graph 22(9):2145–2159
Feng A, Casas D, Shapiro A (2015) Avatar reshaping and automatic rigging using a deformable
model. In: Presented at the proceedings of the 8th ACM SIGGRAPH conference on motion in
games. ACM, pp 57–64
Hyde J, Carter E, Kiesler S, Hodgins J (2016) Evaluating animated characters: facial motion
magnitude influences personality perceptions. ACM Trans Appl Percept 13(2):8:1–8:17
IVA (1998) International conference on intelligent virtual humans. Springer, Berlin
Kapadia M, Pelechano N, Allbeck J, Badler N (2015) Virtual crowds: steps toward behavioral
realism. Morgan & Claypool Publishers, San Rafael, California
Li W, Allbeck JM (2011) Populations with purpose. In: Motion in games. Springer, Berlin/
Li W, Allbeck JM (2012) Virtual humans: evolving with common sense. In: Presented at the
international conference on motion in games. Springer, pp 182–193
Li W, Di Z, Allbeck JM (2012) Crowd distribution and location preference. Comput Anim Virtual
Worlds 23(3–4):343–351
Li WP, Balint T, Allbeck JM (2013). Using a parameterized memory model to modulate NPC
AI. In: Presented at the intelligent virtual agents: 13th international conference, IVA 2013,
Edinburgh, August 29–31, 2013, Proceedings, vol 8108. Springer, p 1
10 J.M. Allbeck
McDonnell R, Larkin M, Dobbyn S, Collins S, O’Sullivan C (2008) Clone attack! perception of

crowd variety. In: Presented at the ACM Transactions on Graphics (TOG), vol 27. ACM, p 26
Pelechano N, Allbeck JM, Badler NI (2008) Virtual crowds: methods, simulation, and control.
Synth Lect Comput Graph Animation 3(1):1–176
Pelechano N, Allbeck JM, Kapadia M, Badler NI (eds) (2016) Simulating heterogeneous crowds
with interactive behaviors. CRC Press, Boca Raton, FL
Pelkey CD, Allbeck JM (2014) Populating semantic virtual environments. Computer Animation
and Virtual Worlds 25(3–4):403–410
Shapiro A (2011) Building a character animation system. In: Presented at the international confer-
ence on motion in games. Springer, pp 98–109
Sunshine-Hill B, Badler NI (2010) Perceptually realistic behavior through alibi generation. In
Proceedings of the Sixth AAAI Conference on Artificial Intelligence and Interactive Digital
Entertainment (AIIDE’10), G. Michael Youngblood and Vadim Bulitko (Eds.). AAAI Press
83–88
Thalmann D, Musse SR, Braun A (2007) Crowd simulation, vol 1. Springer, Berlin
Crowd Evacuation Simulation
Tomoichi Takahashi
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Lessons from the Past and Requirements for Simulation Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Agent-Based Approach to Evacuation Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Crowd Evacuation Using Agent-Based Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Evacuation Scenarios and Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Agent Mental States and their Action Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Pedestrian Dynamics Model and the Mentality of Individuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Guidance to Agents and Communication During Evacuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Abstract
Evacuation simulation systems simulate the evacuation behaviors of people
during emergencies. In an emergency, people are upset and hence do not behave
as they do during evacuation drills. Reports on past disasters reveal various
unusual human behaviors. An agent-based system enables an evacuation simu-
lation to consider these human behaviors, including their mental and social status.
Simulation results that take the human factor into consideration seem to be a good
tool for creating and improving preventions plans. However, it is important to
verify and validate the simulation results for evacuations in unusual scenarios that
have not yet occurred. This chapter shows that the combination of an agent’s
physical and mental status and pedestrian dynamics is the key to replicating
T. Takahashi (*)
Department of Information Engineering, Meijo University, Nagoya, Japan
e-mail: ttaka@meijo-u.ac.jp

DOI 10.1007/978-3-319-30808-1_17-1
2 T. Takahashi
various human behaviors in crowd evacuation simulation. This realistic crowd

evacuation simulation has the potential for practical application in the field.
Keywords
Evacuation behavior • Emergency scenario • Agent-based simulation • Cognitive
map • Psychological factor • Belief-desire-intention • Information transfer and
sharing model • Verification and validation
Introduction
Emergencies such as fires, earthquakes, or terrorist attacks can occur at any time in
any location. Human lives are at risk both from man-made and natural disasters. The
importance of emergency management has been reaffirmed by a number of reports
related to various disasters. The September 11, 2001 World Trade Center (WTC)
attacks and the Great East Japan Earthquake (GEJE) and ensuing tsunami that
occurred on March 11, 2011 took many lives and caused serious injuries. Detailed
reports that focus on occupant behavior during the WTC disaster and evacuation
behavior after the tsunami alarm indicate that safety measures implemented before-
hand and evacuation announcements on site can exert significant influence on
individual evacuation behaviors (de Walle and Murray 2007; Averill et al. 2005;
Cabinet Office Government of Japan).
Many organizations engage in emergency preparation and provide training to
save human lives during emergencies and reduce damage during future disasters
(Cabinet Office of UK 2011; Turoff 2002). The disaster-prevention departments of
governments, buildings, and other organizations develop these training programs.
This training, executed beforehand, is useful to check whether the people are well
prepared for unseen emergencies, can operate according to prevention plans, and
evacuate quickly to safer locations.
It is difficult to replicate emergent situations in the real world and drill for these
situations while involving real humans. It is well known that humans behave
differently when training and during emergencies. Sometimes, a drill can cause
accidents. In fact, in December 2015, a university in Nairobi executed an anti-
terror exercise. The drill included the use of gunshots and this caused students and
staff to panic. A number of people jumped from windows of the university buildings
and were injured (News). Even statutory training in real situations can create risks
for disabled people and some vulnerable groups.
Simulation of the movement of people has been studied in various fields includ-
ing computer graphics, movie special effects, and evacuations (Hawe et al. 2012;
Dridi 2015). This technology allows a prevention center to simulate crowd evacu-
ation behavior in multiple emergency scenarios that cannot be executed in the real
world. Computer simulations help the prevention center to assess their plans for all
emergencies that need to be considered. Crowd evacuation simulation is a key
technology for making safety plans for future emergencies.
Crowd Evacuation Simulation 3
State of the Art
Lessons from the Past and Requirements for Simulation Systems
A number of studies have focused on human behavior during past disasters. The
National Institute of Standards and Technology (NIST) examined occupant behavior
during the attacks on the WTC buildings (Averill et al. 2005; Galea et al. 2008). The
Cabinet Office of Japan also reported on evacuations of individuals during the GEJE
(Cabinet Office Government of Japan). Common types of evacuation behaviors have
been discovered: some individuals evacuated immediately when the disasters
occurred, but others did not evacuate, even though they heard emergency alarms
provided by the authorities. These people consisted of individuals who had family
members located in remote areas, individuals who attempted to contact their families
by phone, and individuals who continued to work because they believed they
were safe.
It is interesting to note that the individuals’ behaviors during these two disasters
were similar to the behaviors of individuals during a flood in Denver, USA, on June
16, 1965, even though communication methods have changed over the past 50 years
(Drabek 2013). Approximately 3,700 families were suddenly told to evacuate from
their homes. The family behaviors that occurred following the warnings were
categorized as follows: those who evacuated immediately, those who attempted to
confirm the threat of disaster, and those who ignored the initial warning and
continued with routine activities.
Other features of human behaviors have been reported in other disasters. (1) In the
2003 fire in the Station nightclub, Rhode Island, most building occupants attempted
to retract their steps to the entrance rather than follow emergency signs, even though
the emergency exit was adequately signposted (Grosshandler et al. 2005). (2) In
emergencies, humans tend to fulfill the roles assigned to them beforehand. For
instance, trained people led the others in their offices promptly to safe places in
the WTC attacks (Ripley 2008). (3) In contrast, a tragedy that occurred at the Okawa
Elementary School during the GEJE demonstrates how untrained leaders may lead to
tragedies (Saijo 2014). The school was located 5 km from the sea and had never
practiced evacuation drills. When the earthquake occurred, an hour elapsed before
teachers decided on an evacuation location. When moving to that location, they were
informed that the tsunami was imminent and that their evacuation location was
unsafe. They tried to evacuate to a higher location, but their efforts were too late.
Most of the students and staff of the Okawa Elementary School were engulfed by the
tsunami and died.
The human behaviors that typically occur during emergencies vary by individual,
and the behaviors may be different from those that are planned. The fluctuations in
these behaviors are the key features that must be simulated in evacuation simula-
tions. The evacuation behaviors depend on the individual who makes decisions and
changes his/her actions according to his/her conditions and information. This infor-
mation includes signs and public announcements (PAs) and is thought to affect
4 T. Takahashi
human behavior and be useful for guiding people quickly to safe places during
dynamically changing situations.
Agent-Based Approach to Evacuation Simulations
NIST simulated some evacuation scenarios to estimate the evacuation time from the
WTC buildings (Kuligowski 2005). The travel times of several cases were simulated
using several evacuation simulation systems, all which assume the following.
People are equal mentally and functionally. In some simulators, sex and age are
taken into consideration as parameters for walking speed in pedestrian dynamics
models. To address roles in society, only human behaviors such as leaders in an
office guiding people to get out of the building immediately were modeled.
All people start their evacuation simultaneously. In fact, some people evacuate after
they finish their jobs. The difference in premovement time of the individuals is
not considered in these simulations.
All people have the same knowledge about the building. They use one route when
they evacuate from the building. Indeed, knowledge about the evacuation route
differs among people, and the evacuation routes can be different.
An agent-based approach provides a platform that corrects these assumptions.

An agent-based simulation system (ABSS) models and computes individuals’
behaviors related to evacuation (Musse and Thalmann 2007). Various types of
human behavior have been studied using the ABSS platform, for example, simu-
lation of human behavior in a hypothetical human-initiated crisis in the center of
Washington DC and a simulation tool incorporating different agent types and three
kinds of interaction: emotion, information, and behavior (Tsai et al. 2011; Parikh
et al. 2013).
An ABSS consists of three parts: the agents, the interaction methods among
agents and environments, and the surrounding environment. Agents perceive data
from the environment and determine their actions according to their goals. An agent
has the properties of physical features, social roles, mentality, and others. The actions
are interactions with other agents and the environment. Information exchanges
among agents and starting to evacuate are examples of actions. The interactions
with the environment are simulated by sub-simulators and affect the status of the
environment. The ABSS repeats these simulation steps: agent perception, agent
decision-making, and environment calculations. The environment involves CAD
models of buildings and scenarios of disaster situations.
The following example demonstrates the ABSS process applied to an evacuation
from a building during a fire. Agents hear alarms and PAs directing them to evacuate
the building. The alarm noise and announcements can increase the anxiety of the
agents, which is calculated using a psychological status model. The mental status
and individual knowledge of the agent determine its actions. When it decides to go to
a safe place, it visualizes the route to that place and moves. One sub-simulator
calculates the agent locations and the status of pedestrian jams inside the building,
and the other sub-simulator calculates the spread of the fire.
Crowd Evacuation Using Agent-Based Simulations
Evacuation Scenarios and Environment
The environment corresponds to the tasks that the ABSS is applied to. The param-
eters in the environment affect the results of the simulations. Table 1 lists the
categories of building evacuation scenarios. Case 1 is a situation in everyday life
and the scenario corresponds to an emergency drill. The other four cases correspond
to emergency situations in which some accident happens, but people do not have
all the information they need. The conditions of each situation worsen from Cases
2 to 5. Providing a real-time evacuation guide for dynamically changing situations is
thought to effectively reduce evacuation time. Case 2 corresponds to a minor
emergency such as a small fire inside a building. The layout of the floor inside the
buildings remains the same during the evacuation, as in Case 1. People also keep
calm in this case. Cases 3, 4, and 5 correspond to situations where some people
become distressed and may have trouble evacuating safely to exits. Case 3 is a
situation where an earthquake causes furniture to fall to the floor that hinders or
prevents evacuation. A case in which fire spreads and humans operate fire shutters to
prevent the fire from spreading further is modeled in Case 4. This operation may
block the evacuation routes and cause differences between the cognitive map of the
evacuees and the real situation. Case 5 is the situation in extreme disasters, where
large earthquakes cause so much destruction to parts of the building that the floor
layout is completely changed.
In Cases 3, 4, and 5, it is necessary to improve prevention plans in terms of
available safe-escape time and required safe-escape time (ISO TR16738 2009).
However, it is difficult to execute evacuation drills for such situations, as the case
in Nairobi demonstrated. Evacuation simulation systems are instead proposed to
simulate the evacuation behaviors of people in such situations.
Table 1 Category of changing situations at evacuations

Map Agent
Mental Interaction Fitness for
Case Situation Map (3D) Layout state mode drills
1 Everyday Static Same Normal Normal Fit (getting
2 Emergency environment (getting (getting more
3 Dynamic Different more more unexpected)
environment anxious) confusing)
4
5 Unknown Distressed Crisis Beyond the
scope of
drill
6 T. Takahashi
Agent Mental States and their Action Selection
People’s state of distress reflects the motions of agents during emergencies. As a

result, the agents take various actions according to the information that they have.
Some people may prefer to trust only information from an authority figure, but others
will trust their neighbors or heed messages sent from their acquaintances. These
individual behaviors form into crowd behavior in emergencies. During the GEJE,
about 34 % of 496 evacuees began their evacuation by taking the advice of acquain-
tances who themselves took the evacuation guidance seriously (Cabinet Office
Government of Japan). The value of 34 % is the average of three prefectures,
Iwate, Miyagi, and Fukushima. Their averages are 44 %, 30 %, and 3 %, respectively.
The question then arises as to where and how people evacuate during emergencies.
Abe, et al. conducted a questionnaire survey with individuals who shopped at a Tokyo
department store (Abe 1986). Three hundred subjects were selected from shoppers in
the department store. The number of male and female participants was equal, and
participants ranged in age from teenagers to adults in their 60s. The questions addressed
the following factors that occur during emergencies: the provision of evacuation
instructions during emergencies, knowledge of emergency exit locations, an individ-
ual’s ability to evacuate safely, and other factors. The results in Table 2 reveal that:
• Individuals’ intentions during emergencies were diverse. Differences were appar-

ent between the sexes and between age groups.
• Half of all surveyed individuals stated they would follow the authorities’ instruc-
tions. The other half stated they would select directions by themselves, and
individuals who chose the fourth and fifth strategies (in Table 2) tended to choose
opposite directions.
Agents act according to their code of conduct or will, and social psychological
factors affect human behavior. The implementation of autonomous agents includes
modeling the process of an individual’s perception, planning, and decision-making.
Modeling the mental state of an agent is key to simulating the evacuation behavior of
people. The psychological factors affect human actions that include selfish
Table 2 Responses to “In which direction would you evacuate?” (Abe 1986)
Sex
Selected actions All (%) Male (%) Female (%)
1 Follow instructions from clerks or announcements 48.7 38 54.7
2 Hide from smoke 26.3 30.7 22
3 Go to the nearest staircase or emergency exit 16.7 20.7 12.7
4 Follow other individuals' movements 3 1.3 4.7
5 Go in the direction that has fewer people 3 2.7 3.3
6 Go to bright windows 2.3 2.7 2
7 Retrace his/her path 1.7 2.7 0.7
8 Other 0.3 0.7 –
movements, altruistic movements, and others. The following cases demonstrate

some properties of human behavior. These actions also change the behavior of
crowd evacuations.
• People swerve when they come close to colliding with each other. When people
see responders approaching, they make way to pass them automatically. The two
behaviors are similar; however, they are different at the conscious level of an
agent. Agents categorize the agents around them into normal or high-priority
groups depending on common beliefs in the agent’s community. For example, the
agent gives consideration to the rescuers and disabled, both of whom are catego-
rized as agents with high priority.
• Families evacuate together. When parents are separated from their children during
emergencies, they become anxious and go to their children at the risk of their own
safety. For instance, the child might be in a toy section in a department store and
have no ability to ask others about his/her parents.
Pedestrian Dynamics Model and the Mentality of Individuals
The belief-desire-intention (BDI) model is one method for representing how agents
select actions according to the situation during the sense-reason-act cycle (Weiss
2000). Belief represents the information that the agent obtains from the environment
and other agents. Desire represents the objectives or situations that the agent would
like to accomplish or bring about, and their actions, which are selected after
deliberation, are represented by intention. In the case of evacuation in emergency
situations, the desires are to move quickly to a safe place, know what happened, or
join families. The associated actions are to move to specific places. These actions are
represented as a sequence of target points. The target points are the places where
people go to satisfy their desires.
Movements, including bidirectional movements in a crowd, can be micro-
simulated in one step using pedestrian dynamics models (Helbing et al. 2000). The
models are composed of geometrical information and a force model that resembles
the behaviors of real people. The behaviors of individuals may block others who are
hurrying to refuges and hence cause pedestrian jams in evacuation (Pelechano
et al. 2008; Okaya and Takahashi 2014).
Guidance to Agents and Communication During Evacuation
The NIST report showed differences in evacuation behaviors between the two
buildings, WTC1 and WTC2. The buildings were similar in size and layout, and
similar numbers of individuals were present in the buildings during the attacks.
Individuals in both buildings began to evacuate when WTC1 was attacked, and
WTC2 was attacked 17 min later. At that time, about 83 % of survivors from WTC1
remained inside the tower, and about 60 % of survivors remained inside WTC2. The
8 T. Takahashi
difference in evacuation rates between two buildings given similar conditions

indicates that there are other interactive and social issues that should be taken into
consideration to simulate crowd evacuation behavior.
A PA gives evacuation guidance to people. According to the GEJE report, only
56 % of evacuees heard the emergency alert warning from a loudspeaker. Of these,
77 % recognized the urgent need for evacuation, and the remaining 23 % did not
understand the announcement because of noisy and confused situations. Nowadays,
people communicate with others in public using cellar phones. This behavior is
assumed to happen during emergencies. Indeed, in GEJE, 2011, it was reported that
people knew and shared information using SNS and personal communications
(Okumura 2014). In a case of family’s evacuation, the following communications
between parents and their children often occurs when they are apart.
Where are you?

I am at location X.
All right, I will be there soon, stay there.
Information regarding the situation and personal circumstances play an important

role when determining actions. The information affects both the premovement and
travel times of evacuation behaviors. With respect to the information or knowledge
of people, whether broadcast or communicated personally, the evacuation process
has the following phases:
When emergencies occur, people either perceive the occurrence themselves or

authorities make announcements. The alarm contains urgent messages conveying
that an emergency situation has occurred and gives evacuation instructions.
People confirm and share the information that they obtain by communicating with
people nearby. After that, people perform actions according to their personal
reasons: some evacuate to a safe place, others hurry to their families, and still
others join rescue operations.
People who are unfamiliar with the building follow guidance from authorities or
employees who act according to prescribed rules or manuals of the buildings. The
information that authorities and employees have may vary with time.
The information transfer and sharing model enables the announcement of proper
guidance to people or information sharing during evacuation (Okaya et al. 2013).
The difference in agents’ information and style of communication causes the diver-
sity of human behavior and affects the behavior of evacuations (Niwa et al. 2015).
Future Directions
An ABSS is expected to simulate the behaviors of agents in unusual scenarios that

are difficult to test in the real world. We learn how people behave and evacuate
during disasters from media stories and reports published by those in authority.
These reports cover evacuation from airplanes, ships, theaters, sport stadiums,
stations, underground transport systems, and others (Wanger and Agrawal 2014;
Peacock et al. 2011; Weidmann et al. 2014). Behavior models have been formulated
to meet the innate human features that were described in the reports and are key
features of these evacuation simulations. Table 3 shows the parameters of the
evacuation models in which human behaviors are taken into account. The parameters
represent the features of the agents, environment, and interactions among agents or
others during the scenarios. In addition, the parameters specify the evacuation
scenarios. Some of the parameters are related to each other; for example, parameters
related to pedestrian dynamics are personal spaces, speed, and avoidance sides, and
others are dependent on countries (Natalie Fridman and Kaminka 2013).
In scientific and engineering fields, the following principle, hypothesis ! com-
pute consequence ! compare results, has been used to make models and to increase
the fidelity of simulations (Feynman 1967). Fundamentally, this principle is applied
to the crowd evacuation simulation. The following points are assumed when model-
ing crowd evacuation behaviors:
Whole-part relations assumption. A crowd evacuation simulation system is com-

posed of subsystems: an agent’s action planning, pedestrian dynamics, and
disaster situations. A model for evacuation behavior is implemented in each
Table 3 Evacuation simulation parameters

Subsystem Parameters
Agent Physical Age
Sex
Impaired/unimpaired
Mental/social State of mind
Human relationships (family, office member, etc.)
Role (teacher, leader, rescue responder, etc.)
Perception Visual data
Auditory data
Action Evacuate (walk/run)
Communicate (hear, talk, share information among agents)
Others (altruistic behavior, rescue operation)
Preference Culture
Nationality
Environment Map/buildings 2D/3D
Elevator
Subsystem Pedestrian dynamics
Disasters effects (fire, smoke, etc.)
Interaction Communication Announcement (guidance from PA)
Information sharing
Human Personal
Relationship Community
10 T. Takahashi
agent, and the pedestrian dynamics models calculate the positions of the agents.
The movements of agents are integrated into crowd behaviors.
Subsystem causality assumption. The agent’s behavior is simulated by formulas or
rules in each agent at every simulation step. In each step, the status of the system
is changed to a new status according to the parameters, models, and formulas.
They may be refined to cover more phenomena or make the results of subsystem
simulations more consistent with experimental data or empirical rules.
Total system validity assumption. The simulation results of the subsystems and the
positions of all agents are integrated into the results of the crowd evacuation
simulation. The results of the simulation are checked with empirical rules or
previous data.
At the second assumption, the model of the subsystem is verified with respect to
real data, and the parameters are tuned to the conditions of the scenarios (Peacock
et al. 2011; Weidmann et al. 2014; Ronchi et al. 2013). The Tokyo fire department
publishes a guide for building fire safety certificates based on simulation results
(Tokyo Fire Department). The results predict the evacuation time at fire under their
specified method and can be used to certify the likelihood of a safe evacuation. The
simulations are in Case 1 of Table 1, which is equivalent to evacuation drills in
everyday conditions. At the third assumption, people evaluate the results of simula-
tion from their personal and organizational perspectives. Using an ABSS with the
functions mentioned in section “Introduction” can simulate more realistic conditions
such as those in Cases 2 to 4.When the integrated simulation results are likely to be
reasonable in unexpected situations, there is no evidence to endorse whether or not
the results can be used in real applications.
In a case in which the results do not fit the empirical rule, even though it may
involve a significant predictor, it is difficult to adopt the simulation results in a
prevention plan according to scientific and engineering principles because we do not
have enough real data and cannot perform experiments in real situations, as in the
case of evacuation simulation. It is important and required to verify the results of
evacuation simulations for emergency situations that have not occurred and affirm
that the planning based on the simulation results will work well in a possible
emergency situation.
Verification and validation (V&V) of the simulation tools and results has been
one of the most important issues in crowd evacuation simulations. V & V problems
are represented using the following questions:
How do we judge if a tool is accurate enough?

How many and which tests should be performed to assess the accuracy of the model
predictions?
Who should perform the test, i.e., the model developers, the model users, or a third
party?
Does the model accurately represent the source system?
Does the model accommodate the experimental frame?
Is the simulator correct?
The questions are essential to ABSS. Questions 1 to 3 are from the test methods
that are suggested from quantitative/qualitative points for behavioral uncertainty
(Ronchi et al. 2013). Questions 4 to 6 are from a study of validation on evacuation
drills from a ten-story building (Isenhour and Löhner 2014).
A method of comparing simulation results to real scenarios as macroscopic
patterns in a quantitative manner has been proposed as a validation method
(Banerjee and Kraemer 2010). Interactions among agents and dynamically changing
environments also affect the behavior of crowd evacuations. A verification test is
suggested in order to check evacuation plans under the dynamic availability of exits
(Ronchi et al. 2013). The following qualitative standards are proposed for applica-
tion in simulations without real-world data that involve real evacuation data and
experimental data (Takahashi 2015):
Consistency with data. The simulation results or its variations after changing param-
eters or modifying subsystems are compatible with past anecdotal reports.
Generation of new findings. The results involve something that was not recognized
as important before the simulations, which is reasonable given empirical rules.
Accountability of results. The cause of the changes can be explainable from the
simulation data systematically.
While we do not have answers to these questions, ABSS has been applied to
more realistic situations. For example, an evacuation from a building with fire
shutters is a realistic case (Takahashi et al. 2015). The fire shutters are installed in
buildings by law to prevent fire and smoke from spreading inside. Some agents
evacuate instantly and others evacuate after finishing their jobs. Operators at the
prevention center close the fire shutters at time t1 to prevent fire spreading. If there
is no announcement regarding the shutter closing, the agents don’t know the
changes of environments. They evacuate according to their own cognitive map,
which might not be updated until they notice the fire shutter closing at t2. As a
result, the evacuation time from t1 to t2 is wasted, even though the agent starts
evacuation instantly. This simulation demonstrates that evacuation times change
for various scenarios in dynamically changing environments corresponding to
Case 3, 4 and 5 and proves the potential of evacuation simulation for future
applications.
In this chapter, we presented some features of crowd evacuation simulations: the
role of human mental conditions during emergencies, the presentation of agent
mental states, and information on evacuations. We also showed that the combination
of an agent’s physical and mental status and pedestrian dynamics is the key to
simulating crowd evacuation and replicating various human behaviors. Simulating
crowd evacuation more realistically introduces additional human-related factors.
This makes it difficult to systematically analyze the simulation results and compare
them with data from the real world. At present, the simulation results are not so much
objectively measured as subjectively interpreted by humans. Future research and
model development will focus on the study of agent interactions, human mental
models, and verification and validation problems.
12 T. Takahashi
References
Abe K (1986) Panic and human science: prevention and safety in disaster management. Buren
Shuppan. in Japanese, Japan
Averill JD, Mileti DS, Peacock RD, Kuligowski ED, Groner NE (2005) Occupant behavior, egress,
and emergency communications (NIST NCSTAR 1–7). Technical report, National Institute of
Standards and Technology, Gaitherburg
B. news Woman dies after ‘terror drill’ at Kenya’s strathmore university. http://www.bbc.com/news/
world-africa-34969266. Date:16 Mar 2016
Banerjee B, Kraemer L (2010) Validation of agent based crowd egress simulation (extended
abstract). In: International conference on autonomous agents and multiAgent systems
(AAMAS’10). pp 1551–1552. http://www.aamas-conference.org/proceeding.html
Cabinet Office Government of Japan. Prevention Disaster Conference, the Great West Japan
Earthquake and Tsunami. Report on evacuation behavior of people (in Japanese). http://www.
bousai.go.jp/kaigirep/chousakai/tohokukyokun/7/index.html. Date: 16 Mar 2016. in Japanese.
Cabinet Office of UK (2011) Understanding crowd behaviours: documents. https://www.gov.uk/
government/publications/understanding-crowd-behaviours-documents. 20 Mar 2016
de Walle BV, Murray T (2007) Emergency response information systems: emerging trends and
technologies. Commun ACM 50(3):28–65
Drabek TE (2013) The human side of disaster, 2nd edn. CRC Press, Boca Raton
Dridi M (2015) Simulation of high density pedestrian flow: a microscopic model. Open J Model
Simul 3(4):81–95
Feynman RP (1967) Seeking new laws. In: The character of physical law. The MIT Press,
Cambridge
Galea ER, Hulse L, Day R, Siddiqui A, Sharp G, Boyce K, Summerfield L, Canter D, Marselle M,
Greenall PV (2008) The uk wtc9/11 evacuation study: an overview of the methodologies
employed and some preliminary analysis. In: Schreckenberg A, Klingsch WWF, Rogsch C,
Schreckenberg M (eds) Pedestrian and evacuation dynamics 2008 (pp. 3–24). Springer,
Heidelberg
Grosshandler WL, Bryner NP, Madrzykowski D, Kuntz K (2005) Report of the technical investi-
gation of the station nightclub fire (NIST NCSTAR 2). Technical report, National Institute of
Standards and Technology, Gaitherburg
Hawe GI, Coates G, Wilson DT, Crouch RS (2012) Agent-based simulation for large-scale
emergency response. ACM Comput Surv 45(1):1–51
Helbing D, Farkas I, Vicsek T (2000) Simulating dynamical features of escape panic. Nature
407:487–490
Isenhour ML, Löhner R (2014) Validation of a pedestrian simulation tool using the {NIST}
stairwell evacuation data. Transp Res Procedia 2:739–744, The Conference on Pedestrian and
Evacuation Dynamics 2014 (PED 2014), 22–24 October 2014, Delft, The Netherlands
ISO:TR16738:2009. Fire-safety engineering – technical information on methods for evaluating
behaviour and movement of people
Kuligowski ED (2005) Review of 28 egress models. In: NIST SP 1032; Workshop on building
occupant movement during fire emergencies.
Musse SR, Thalmann D (2007) Crowd simulation. Springer-Verlag, London
Natalie Fridman AZ, Kaminka GA (2013) The impact of culture on crowd dynamics: an empirical
approach. In: International conference on autonomous agents and multiagent systems,
AAMAS’13, p 143–150
Niwa T, Okaya M, Takahash T (2015) TENDENKO: agent-based evacuation drill and emergency
planning system. Lecture Notes in Computer Science 9002. Springer, Heidelberg
Okaya M, Takahashi T (2014) Effect of guidance information and human relations among agents on
crowd evacuation behavior. In: Kirsch U, Weidmann U, Schreckenberg M (eds) Pedestrian and
evacuation dynamics 2012. Springer, Cham
Okaya M, Southern M, Takahashi T (2013) Dynamic information transfer and sharing model in
agent based evacuation simulations. In: International conference on autonomous agents and
multiagent systems, AAMAS 13. pp 1295–1296
Okumura H (2014) The 3.11 disaster and data. J Inf Process 22(4):566–573
Parikh N, Swarup S, Stretz PE, Rivers CM, Bryan MVM, Lewis L, Eubank SG, Barrett CL, Lum K,
Chungbaek Y (2013) Modeling human behavior in the aftermath of a hypothetical improvised
nuclear detonation. In: International conference on autonomous agents and multiagent systems,
AAMAS’13, pp 949–956
Peacock RD, Kuligowski ED, Averill JD (2011) Pedestrian and evacuaion dynamics. Springer,
Heidelberg
Pelechano N, Allbeck J, Badler N (2008) Virtual crowds: methods, simulation, and control. Morgan
& Claypool Publishers series, California, New York
Ripley A (2008) The unthinkable: who survives when disaster strikes – and why. New York: Three
Rivers Press
Ronchi E, Kuligowski ED, Reneke PA, Peacock RD, Nilsson D (2013) The process of verification
and validation of building fire evacuation models. Technical report. National Institute of
Standards and Technology, Gaitherburg. Technical Note 1822
Saijo T (2014) Be a tsunami survivor. http://wallpaper.fumbaro.org/survivor/tsunami_en_sspj.pdf.
Date:17 Mar 2016
Takahashi T (2015) Qualitative methods of validating evacuation behaviors. In Takayasu H, Ito N,
Noda I, Takayasu M (eds) Proceedings of the international conference on social modeling and
simulation, plus econophysics colloquium 2014. Springer proceedings in complexity. Springer
International Publishing, pp 231–242
Takahashi T, Niwa T, Isono R (2015) Method for simulating the evacuation behaviours of people in
dynamically changing situations. In: Proceedings of TGF2015. Springer. To be published in
2016 Fall
Tokyo Fire Department. Excellence mark -certified fire safety building indication system. http://
www.tfd.metro.tokyo.jp/eng/inf/excellence_mark.html. Date: 25 Jan 2016
Tsai J, Fridman N, Bowring E, Brown M, Epstein S, Kaminka G, Marsella S, Ogden A, Rika I,
Sheel A, Taylor ME, Wang X, Zilka A, Tambe M (2011) Escapes: evacuation simulation with
children, authorities, parents, emotions, and social comparison. In: The 10th international
conference on autonomous agents and multiagent systems, vol 2, AAMAS’11. International
Foundation for Autonomous Agents and Multiagent Systems, Richland, pp 457–464
Turoff M (2002) Past and future emergency response information systems. Commun ACM 45
(4):29–32
Wanger N, Agrawal V (2014) An agent-based simulation system for concert venue crowd evacu-
ation modeling in the presence of a fire disaster. Expert Syst Appl 41:2807–2815
Weidmann U, Kirch U, Schreckenberg M (eds) (2014) Pedestrain and evacuation dynamics 2012.
Springer, Heidelberg
Weiss G (2000) Multiagent systems. The MIT Press, Massachusets
Perceptual Study on Facial Expressions
Eva G. Krumhuber and Lina Skora
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Early Beginnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Dynamic Advantage in Facial Expression Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Temporal Characteristics: Directionality and Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Effects of Facial Motion on Perception and Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Ratings of Authenticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Person Judgments and Behavioral Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Facial Mimicry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Neuroscientific Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Abstract
Facial expressions play a paramount role in character animation since they reveal
much of a person’s emotions and intentions. Although animation techniques have
become more sophisticated over time, there is still need for knowledge in terms of
what behavior appears emotionally convincing and believable. The present chap-
ter examines how motion contributes to the perception and interpretation of facial
expressions. This includes a description of the early beginnings in research on
facial motion and more recent work, pointing toward a dynamic advantage in
facial expression recognition. Attention is further drawn to the potential charac-
teristics (i.e., directionality and speed) that facilitate such dynamic advantage.
This is followed by a review on how facial motion affects perception and
E.G. Krumhuber (*) • L. Skora

University College London, London, UK
e-mail: e.krumhuber@ucl.ac.uk

DOI 10.1007/978-3-319-30808-1_18-1
2 E.G. Krumhuber and L. Skora
behavior more generally, with the neural systems that underlie the processing of
dynamic emotions. The chapter concludes by discussing remaining challenges
and future directions for the animation of natural occurring emotional expressions
in dynamic faces.
Keywords
Motion • Dynamic • Facial expression • Emotion • Perception
Introduction
Among the 12 principles of animation developed in the early 1930s, animators at the
Walt Disney Studios considered motion to be fundamental for creating believable
characters. They were convinced that the type and speed of an action help define a
character’s intentions and personality (Kerlow 2004). Since the early days of
character animation, much has changed in animation techniques and styles. From
hand-drawn cartoon characters to real-time three-dimensional computer animation,
the field has seen a major shift toward near-realistic characters that exhibit humanlike
behavior. Whether those are used for entertainment, therapy, or education, the
original principle of motion continues to be of interest in research and design. This
particularly applies to the topic of facial animation as subtle elements of the
character’s thoughts and emotions are conveyed through the face (Kappas et al.
2013). Facial expressions provide clues and insight about what the character thinks
and feels. They act as a powerful medium in conveying emotions. Although the tools
for facial animation have become more sophisticated over time (i.e., techniques for
capturing and synthesizing facial expressions), there is still need for knowledge
about how humans respond to emotional displays in moving faces. Only if the
character appears emotionally convincing and believable will the user/audience
feel comfortable in interaction. The present chapter aims to help with this task by
providing an overview of the existing literature on the perception of dynamic facial
expressions. Given the predominant focus on static features of the face in past
research, we seek to highlight the beneficial role of facial dynamics in the attribution
of emotional states. This includes a description of the early beginnings in research on
facial motion and more recent work pointing toward a dynamic advantage in facial
expression recognition. The next section draws attention to the potential character-
istics that facilitate such dynamic advantage. This is followed by a review on how
facial motion affects perception and behavior more generally. Neural systems in the
processing of dynamic emotions and their implications for action representation are
also outlined. The final section concludes the paper by discussing remaining chal-
lenges and future directions for the animation of natural occurring emotional expres-
sions in dynamic faces.
Perceptual Study on Facial Expressions 3
State of the Art
Early Beginnings
In everyday settings, human motion and corresponding properties (e.g., shapes,

texture) interact to produce a coherent percept. Yet, motion conveys important
cues for recognition even in isolation from the supportive information. The human
visual system, having evolved in dynamic conditions, is highly attuned to dynamic
signals within the environment (Gibson 1966). It can use this information to identify
an agent or infer its actions purely by the motion patterns inherent to living
organisms, called biological motion (Johansson 1973). Investigations of biological
motion of the face suggest that the perception of faces is aided by the presence of
nonrigid facial movements, such as stretching, bulging, or flexing of the muscles and
the skin. In an early and now seminal point-light paradigm (Bassili 1978), all static
features of actors’ faces, such as texture, shape, and configuration, were obscured
with the use of black makeup. Subsequently, the darkened faces were covered with
approximately 100 luminescent white dots and video recorded in a dark room
displaying a range of nonrigid motion, from grimaces to the basic emotional
expressions (happiness, sadness, fear, anger, surprise, and disgust). The dark setup
resulted in only the bright points being visible to the observer, moving as a result of
facial motion. In a recognition experiment, the moving dots were recognized as faces
significantly better than when the stimulus was shown as a sequence of static frames
or as a static image. Similarly, moving point-light faces enabled above-chance
recognition of the six basic emotional expressions in comparison to motionless
point-light displays (Bassili 1979; Bruce and Valentine 1988). This suggests that
when static information about the face is absent, biological motion alone is distinc-
tive enough to provide important cues for recognition.
Dynamic Advantage in Facial Expression Recognition
Subsequent research has pointed toward a motion advantage especially when static
facial features are compromised. This is of particular relevance for computer-
generated, synthetic faces (e.g., online avatars, game characters). In comparison to
natural human faces, synthetic faces are still inferior in terms of their realistic
representation of the finer-grained features, such as textures, skin stretching, or
skin wrinkling. Such impairment in quality of static information can be remedied
by motion. Numerous studies have shown that expression recognition in dynamic
synthetic faces consistently outperforms recognition in static synthetic faces (Ehrlich
et al. 2000; Wallraven et al. 2008; Wehrle et al. 2000; Weyers et al. 2006). This
suggests that motion is able to add a relevant layer of information when synthetic
features fail to provide sufficient cues for recognition. The effect is found both under
uniform viewing quality and when the featural or textural information is degraded
(e.g., blurred).
For natural human faces, however, the dynamic advantage is weaker or inexistent
when the quality of both static and dynamic displays is comparably good (Fiorentini
and Viviani 2011; Kamachi et al. 2001; Kätsyri and Sams 2008). As such, motion is
likely to provide additional cues for recognition when key static information is
missing (i.e., in degraded and obscured expressions). Its benefits may be redundant
when the observer can draw enough information from the static properties of the
face. This applies to static stimuli that typically portray expressions at the peak of
emotionality. Such stimuli, prominently used in face perception research, are char-
acterized by their stereotypical depiction of a narrow range of basic emotions. They
are often also posed upon instructions by the researcher and follow a set of proto-
typical criteria (e.g., Facial Action Coding System, FACS; Ekman and Friesen
1978). In this light, it is likely that stylized static expressions contain the prototypical
markers of specific emotions, thereby facilitating recognition.
Yet, everyday emotional expressions are spontaneous and often include
non-prototypical emotion blends or patterns. They are normally also of lower
intensity, potentially becoming more difficult to identify without supportive cues
such as motion. For instance, low-intensity expressions, which tend to be more
difficult to identify the less intense they get, are recognized significantly better in a
dynamic than static form (Ambadar et al. 2005; Bould and Morris 2008). In this
context, motion appears to provide additional perceptual cues, making up for
insufficient informative signals.
Temporal Characteristics: Directionality and Speed
How can we explain the motion advantage in expression recognition? Could it

simply derive from an increase in the amount of cues in a dynamic sequence?
Early hypotheses point out that a moving sequence contains a greater number of
static information from which to infer emotion judgments than a single static
portrayal (Ekman and Friesen 1978). Arguably, as a dynamic sequence unfolds, it
provides multiple samples of the developing expression compared to a single sample
in static displays. To test this assumption, Ambadar et al. (2005) compared emotion
recognition performances between dynamic, static, and multi-static expressions. In
the multi-static condition, static frames constituting a video were interspersed with
visual noise masks disrupting the fluidity of motion. Out of these, dynamic expres-
sions were recognized with a significantly greater accuracy than both multi-static
and static portrayals (see also Bould and Morris 2008). This suggests that the
intrinsic temporal quality of the unfolding expression is what helps to disambiguate
its content rather than a mere increase in static frames.
A likely candidate that facilitates the dynamic advantage is the directionality of
change in the expression over time. Research shows that humans are highly sensitive
to the direction in which the expression unfolds. For example, they are able to
accurately detect the directionality in a set of scrambled images and arrange them
into a temporally correct sequence (Edwards 1998). Similarly, disrupting the natural
temporal direction of the expression results in worse recognition accuracy than when
the expressions unfold naturally. In a series of experiments, Cunningham and

Wallraven (2009b) demonstrated this by applying various manipulation techniques
to the direction of unfolding, such as scrambling the frames in a dynamic sequence
or playing them backward. Their results indicate that the identification of emotional
expressions suffers considerably when natural motion is interrupted. Recognition
performance also appears to be better in sequences in which the temporal unfolding
is preserved, thereby allowing the directionality of change to be observed as the
expression emerges (Bould et al. 2008, but see Ambadar et al. 2005 for a contrasting
result). Yet, it is noteworthy that this effect might not affect all emotions equally. For
example, happiness is typically recognized better than other basic emotions regard-
less of condition.
In addition to the movement direction, the velocity of unfolding plays a crucial
role in emotion perception. Changes in viewing speed, such as slowing down or
speeding up of the dynamic sequences, significantly affect expression recognition
accuracy. This effect appears to be different between emotions based on the differ-
ences in their intrinsic optimum velocities. For example, sadness is naturally slow; so
slowed-down viewing conditions do not impact it negatively as much as they impact
recognition accuracy for all other tested emotions (Kamachi et al. 2001). Conversely,
surprise is naturally fast, and it could be its natural velocity that distinguishes it from
the morphologically similar expression of fear which is slower (Sato and Yoshikawa
2004). Importantly, changing the speed throughout an entire expression results in
different effects as changes to the duration of the peak. This suggests that the
beneficial effects of natural movements cannot simply be explained by the mere
exposure time to the expression (Kamachi et al. 2001; Recio et al. 2013). Overall,
altering the speed of expression unfolding appears to influence perception without
affecting the direction of change. As such, the intrinsic velocities of particular
emotional expressions are likely to provide stronger cues for recognition than the
perception of change alone (Bould et al. 2008).
Finally, the perception of dynamic faces is also linked to the quality of motion.
While expressions in real faces unfold in a biologically natural manner (i.e., non-
linearly), facial animations have been often characterized by linear techniques. Such
linearly unfolding facial expressions (e.g., dynamic displays morphed from individ-
ual static displays) yield slower and poorer recognition accuracy in comparison to
natural, nonlinear unfolding, as well as worse naturalness and genuineness ratings
(Cosker et al. 2010; Wallraven et al. 2008). As a result, linear morphs might not
constitute a good representation of the real-life quality of facial motion, which is
particularly relevant to the construction of realistic synthetic faces. However, recent
developments within the field of affective computing identify multiple parameters
linked to naturalistic expression unfolding that can improve the quality of motion in
computer-generated faces and raise their recognition rates, such as appropriate
speeds, action unit (AUs) activations, intensities, asymmetries, and textures
(Krumhuber et al. 2012; Recio et al. 2013; Yang et al. 2013). As such, the benefits
provided by motion appear to be more than the perception of motion itself. Instead, it
is a comprehensive set of information deriving from the temporal characteristics
including the perception of change, intrinsic velocity of an expression, and the

quality of motion.
Effects of Facial Motion on Perception and Behavior
In addition to the supportive role in expression recognition, motion also affects a

number of perceptual and behavioral factors. Those include expression judgments
such as intensity and authenticity, as well as behavioral responses and even mimicry.
Firstly, emotions expressed in a dynamic form are perceived to be more intense
than the same emotions in a static form (Biele and Grabowska 2006; Cunningham
and Wallraven 2009a). Motion appears to enhance intensity estimates because of the
changes in the expression as it develops from neutral to fully emotive. While static
portrayals retain the same intensity level throughout the presentation time, dynamic
changes highlight the contrast between the neutral and fully emotional expression.
As such, the contrast makes the expression seem more intense (Biele and Grabowska
2006). Another explanation for this effect was offered in terms of representational
momentum (RP). RP is a visual perception phenomenon in which the observer
exaggerates the final position of a gradually moving stimulus. It often involves a
forward displacement. For example, when a moving object disappears from the
visual field, observers tend to report its final position as displaced further down its
trajectory than it objectively was. In a study about dynamic facial expressions and
RP, Yoshikawa and Sato (2008) found that participants exaggerated the last – fully
emotive – frame of the dynamic sequence and remembered it as more intense that it
was in reality. The effect also got more pronounced with increasing velocity of
expression unfolding. As such, it seems that the gradual shift from neutral to
emotional in dynamic expressions generates a forward displacement, inducing an
exaggerated and intensified perception of the final frame in the sequence.
Ratings of Authenticity
Motion also appears to help observers assess the authenticity of an expression better
than static portrayals can. Authenticity refers to more than correct identification of
the emotional expression observed. It is a quality telling us whether the emotion is
genuinely experienced or not. Smiles have been prominently used to study this
dimension. Being universal and widespread in everyday interactions, smiles can
indicate a range of feelings, from happiness and amusement to politeness and
embarrassment (Ambadar et al. 2009). However, smiles can also be easily used to
mask real emotions or to deceive others (e.g., Ekman 1985). As such, they constitute
a good stimulus to study the genuineness of the underlying feeling.
Traditionally, the so-called Duchenne marker has been considered as an indicator
of smile authenticity (Ekman et al. 1990), where its presence signals that a smile is
genuine (“felt”) as opposed to false (“unfelt”). The Duchenne marker involves, in
addition to the lip corner puller (zygomaticus major muscle), the activation of the
orbicularis oculi muscle surrounding the eye. This results in wrinkling on the sides
of the eyes, commonly referred to as crow’s feet. While the validity of the Duchenne
marker in the perception of static expressions is well documented, motion properties
are crucial for assessing smile authenticity in dynamic displays (e.g., Korb et al.
2014; Krumhuber and Manstead 2009). For example, genuine smiles differ in lip
corner and eyebrow movements from deliberate, false smiles (Schmidt et al. 2006;
Schmidt et al. 2009). More specifically, Frank et al. (1993) highlighted three
dynamic markers of genuine smiles: expression duration, synchrony in muscle
activation (between zygomaticus major and orbicularis oculi muscles), and smooth-
ness of mouth movements.
Overall, genuine smiles last between 500 and 4000 ms, whereas false smiles tend
to be shorter or longer (Ekman and Friesen 1982). Furthermore, the smoothness and
duration of the expressive components of smiles are meaningful indicators. Bugental
(1986) and Weiss et al. (1987) were first to show that the onset and offset in false
smiles tend to be faster in comparison to felt smiles (see also Hess and Kleck 1990).
To investigate whether these differences affect expression perception, Krumhuber
and Kappas (2005) manipulated onset, apex, and offset timings of computer-
generated smiles. Their results confirmed the proposition that each dynamic element
of a smiling expression has an intrinsic duration range at which it looks genuine. In
particular, expressions are perceived as more authentic the longer their onsets and
offsets, while a long apex is linked to lower genuineness ratings.
Person Judgments and Behavioral Responses
Besides their effects on authenticity ratings, dynamic signals influence trait attribu-
tions and behavioral responses to the target expressing an emotion. For instance,
people displaying dynamic genuine smiles (long onset and offset) are rated as more
trustworthy, more attractive, and less dominant than those who show smile expres-
sions without those characteristics (Krumhuber et al. 2007b). In addition, facial
movement helps to regulate interpersonal relations by shaping someone’s intention
to approach or cooperate with another person. In economic trust games, participants
can receive a financial gain if their counterpart cooperates but incur a loss if the
counter-player fails to cooperate. As such, their performance depends on accurate
assessment of the counterpart’s intentions. Krumhuber and colleagues (Krumhuber
et al. 2007a) showed that people are more likely to trust and engage in an interaction
with a counterpart who displays a dynamic authentic smile than a dynamic false
smile or neutral expression. Participants with genuinely smiling counterparts also
ascribe more positive emotions and are more inclined to meet them again. Further-
more, people showing dynamic genuine smiles are evaluated more favorably and
considered more suitable candidates in a job interview than those who do not smile
or smile falsely (Krumhuber et al. 2009). Notably, this effect applies to real human
faces as well as to computer-generated ones.
When comparing static and dynamic facial features, it appears that they contrib-
ute to different evaluations and social decisions. Static and morphological features,
such as bone structure or width, have been found to affect judgments of ability and
competence. In turn, features that are dynamic and malleable, like muscular patterns
in emotional expressions, affect judgments of intentions (Hehman et al. 2015).
Given that these facial signals are also linked to evaluations of trustworthiness and
likeability, they are likely to drive decision-making in social interactions. In line with
this argument, participants were shown to choose a financial advisor, a role requiring
trust, based on dynamic rather than static facial properties (Hehman et al. 2015).
Facial Mimicry
Existing evidence suggests that dynamic facial displays elicit involuntary and subtle
imitative responses more evidently than do static versions of the same expression
(Rymarczyk et al. 2011; Sato et al. 2008; Weyers et al. 2006). Those responses,
interpretable as mimicry, are a result of activity in facial muscles corresponding to a
given perceived expression (i.e., lowering the eyebrows in anger, pulling the lip
corners in happiness). They occur spontaneously and swiftly (about 800–900 ms)
after detecting a change in the observed face. While involuntary facial mimicry is a
subtle rather than full-blown replication of a witnessed emotion, it is evident enough
to be distinguished in terms of its valence (positive or negative quality) by indepen-
dent observers (Sato and Yoshikawa 2007a).
Crucially, the presence of mimicry has a supporting role in emotion perception.
For example, being able to mimic helps observers to recognize the emotional valence
of expressions (Sato et al. 2013). Happiness and disgust are less well identified when
corresponding muscles are engaged by biting on a pen which effectively blocks
mimicry in the lower part of the face (Oberman et al. 2007; Ponari et al. 2012). In a
similar vein, blocking mimicry in the upper part of the face by drawing together two
stickers placed above the eyebrows impairs the recognition of anger. Mimicry also
appears useful in detecting changes in expressions. Having to identify the point at
which an expression transforms from one emotion into another (e.g., happiness to
sadness) proves more difficult when mimicry is blocked by holding a pen sideways
between the teeth. For this task, participants who are free to mimic are quicker in
spotting changes in the dynamic trajectory of facial expressions (Niedenthal et al.
2001). Furthermore, mimicry aids emotion judgments, particularly in the context of
smile authenticity. Dynamic felt smiles are more easily distinguished from dynamic
false ones when expressions can be freely mimicked compared to when mimicry is
blocked by a mouth-held pen (Maringer et al. 2011; Rychlowska et al. 2014).
Overall, those findings suggest that facial mimicry helps to make inferences about
dynamic emotional faces such as emotion recognition and trajectory changes or
authenticity judgments. As such, it adds to the evidence that facial motion conveys
information that is essential to comprehensive expression perception, while also
driving behavioral responses.
Neuroscientific Evidence
Evidence from neuroscience suggests that differences in the processing of dynamic

and static facial stimuli begin at a neural level. For example, studies of patients with
brain lesions or neurological disorders point toward a dissociation in the neural
routes for processing dynamic and static faces. In the most notable cases, patients
who are unable to recognize emotions from static displays can easily do so from
moving displays (Adolphs et al. 2003; Humphreys et al. 1993).
In healthy people, dynamic facial expressions evoke significantly larger and more
widespread activation patterns in the brain than static expressions (LaBar et al. 2003;
Sato et al. 2004). This enhanced activation is apparent in a range of brain regions,
starting with the visual area V5 which subserves motion perception. It has also been
observed in the fusiform face area (FFA), a number of frontal and parietal regions,
and the superior temporal sulcus (STS), areas implicated in the processing of faces,
emotion, and biological motion, respectively (Kessler et al. 2011; Trautmann et al.
2009). The STS has been given particular consideration due to its involvement in
interpreting social signals, in addition to biological motion. As such, enhanced
activation in the STS in response to dynamic facial stimuli could be related to
extracting socially relevant information (i.e., intentions) from the changeable fea-
tures of the face (Arsalidou et al. 2011; Kilts et al. 2003). Additionally, in an
electroencephalograph (EEG) study, attention-related brain activity was found to
be greater and longer when participants observed dynamic compared to static stimuli
(Recio et al. 2011). This higher activity continued throughout the duration of an
expression, contributing to more elaborate processing of dynamic faces.
Such enhanced and more widespread brain activation in response to facial motion
could be caused by the fact that dynamic expressions are inherently more complex to
process. Equally, it could derive from greater evolutionary experience with moving
faces and the need to extract social meaning from them for effective communication.
In this light, neurological evidence lends support to the behavioral findings.
Improved recognition accuracy, sensitivity to the temporal characteristics, and the
ability to make inferences about genuineness, trustworthiness, or approachability
could be an effect of enhanced processing of dynamic faces.
Besides phenomena of neural adaptation, there is work suggesting that brain
activity while observing facial movements may encompass regions which are linked
to one’s own experience of emotional states, as well as areas reported to contain
mirror neurons (Dapretto et al. 2006). Initially observed in macaque monkeys, mirror
neurons fire both when performing an action and when watching the action in others
(Rizzolatti et al. 1996). Emotion perception may therefore be partially subserved by
the mirror neuron system (i.e., premotor and parietal regions, superior temporal
sulcus; Iacoboni and Dapretto 2006; Rizzolatti and Craighiero 2004) which activates
an internal representation of the observed state almost as if it was felt by oneself.
Supportive evidence comes from research showing that facial mimicry in response to
observed expressions activates similar patterns in the brain of the perceiver (Lee
et al. 2006). Also, observing an emotional experience of someone elicits
corresponding subjective arousal in oneself (Lundqvist and Dimberg 1995) which is

found to be stronger for dynamic than static faces (Sato and Yoshikawa 2007b).
Importantly, it has been proposed that this mirror neuron system has evolved to
produce an implicit internal understanding of others’ mental states and intentions
(Dimberg 1988; Gallese 2001). Following from this assumption, mirroring brain
activity in response to facial expressions could be the driving force behind higher-
order cognitive processes such as empathy or mentalizing (Iacoboni 2009). For
example, witnessing a painful expression on someone’s face and feeling pain oneself
activate largely overlapping neural pathways which are correlated with regions
linked to empathy (Botvinick et al. 2005; Singer et al. 2004). The ability to mimic
expressions was also shown to cause greater prosocial behavior, arguably mediated
by greater empathy derived from mimicry and shared activations (Stel et al. 2008).
Overall, this has been taken to suggest that humans understand, empathize with, and
make inferences about mental states of others because the action-perception overlap
activates internal experiences of the same state (Schulte-Rüther et al. 2007).
Future Directions
From the literature reviewed above, there is conclusive evidence suggesting that
humans have remarkable abilities to perceive and understand the actions of others.
Driven by the universal need for social connection, the efficient detection and
interpretation of social signals appears essential for successful interaction. Given
the rapid advances in technology, these uniquely adaptive skills are likely to be
translated to a new form of social partners in the near future. With the move of
computing into the social domain, nonhuman agents are envisaged to become
integral parts of our daily lives, from the workplace to social and private applications
(Küster et al. 2014). As a result, many interactions will not occur in their traditional
form (i.e., human to human) but instead involve computer-generated avatars and
social robots. In order to build animated systems that emit appropriate social cues
and behavior, it is imperative to understand the factors that influence perception.
Facial expressions prove to play a vital part in this process since they reveal much of
a character’s emotions and intentions. While animation techniques offer more
control than ever over visual elements, subtle imperfections in the timing of facial
expressions could evoke negative reactions from the viewer.
In 1970, Masahiro Mori described a phenomenon called the “uncanny valley”
(UV) in which human-realistic characters are viewed negatively if they are almost
but not quite perfectly human. As such, increased human-likeness may result in
unease when appearance or behavior fall short of emulating those of real human
beings. Classic examples can be found in computer-animated films, such as The
Polar Express and Final Fantasy: The Spirits Within, which many viewers find
disturbing due to their human-realistic but eerie characters (Geller 2008). According
to Mori, this perceived deviation from normal human behavior is further pronounced
when movement is added. Particularly, if the appearance is more advanced than the
behavior, violated perceptual expectations could make the moving character less
acceptable. In line with this argument, Saygin et al. (2012) showed that androids that
look human but don’t move in a humanlike (biological) manner elicit a prediction
error that leads to stronger brain activity in the perceiver. Furthermore, virtual
characters are more likely to be rated as uncanny when their facial expressions
lack movement in the forehead and eyelids (Tinwell et al. 2011).
Although the exact role of motion in the UV remains an issue of debate (see
Kätsyri et al. 2015), there is increasing evidence suggesting that natural human
motion positively influences the acceptability of characters, particularly those that
would fall appearance-wise into the UV (i.e., zombies; Piwek et al. 2014). As such,
high-quality motion has the potential to improve ratings of familiarity and human-
likeness by eliciting higher affinity (McDonnnell et al. 2012, Thompson et al. 2011).
In order for natural motion to become the standard in animation, it is essential to rely
on behavior representative of the real world. At the moment, databases depicting
dynamic emotional expressions are still limited in the range and type of facial
movements being captured. The majority of them contain deliberately posed affec-
tive displays recorded under highly constrained conditions (for a review see
Krumhuber et al. in press). Such acted portrayals may not provide an optimal basis
for the modeling of naturally occurring emotions. For progress to occur in the future,
efforts that target the dynamic analysis and synthesis of spontaneous behavior will
prove fruitful. This also includes the study of how multiple dynamic cues interact to
produce a coherent percept. Only once the dynamic nature of facial expressions is
fully understood will it be possible to successfully incorporate this knowledge into
animation models. The present chapter underscores the importance of this task by
showing that perceivers are highly sensitive to the motion dynamics in the perceptual
study of facial expressions.
Cross-References

▶ Real-Time Full Body (or face) Posing
References
Adolphs R, Tranel D, Damasio AR (2003) Dissociable neural systems for recognizing emotions.
Brain Cogn 52:61–69. doi:10.1016/S0278-2626(03)00009-5
Ambadar Z, Cohn JF, Reed LI (2009) All smiles are not created equal: morphology and timing of
smiles perceived as amused, polite, and embarrassed/nervous. J Nonverbal Behav 33:17–34.
doi:10.1007/s10919-008-0059-5
Ambadar Z, Schooler JW, Cohn JF (2005) Deciphering the enigmatic face. The importance of facial
dynamics in interpreting subtle facial expressions. Psychol Sci 16:403–410. doi:10.1111/j.0956-
7976.2005.01548.x
Arsalidou M, Morris D, Taylor MJ (2011) Converging evidence for the advantage of dynamic facial
expressions. Brain Topogr 24:149–163. doi:10.1007/s10548-011-0171-4
Bassili JN (1978) Facial motion in the perception of faces and of emotional expression. J Exp
Psychol Hum Percept Perform 4:373–379. doi:10.1037/0096-1523.4.3.373
Bassili JN (1979) Emotion recognition: the role of facial movement and the relative importance of
upper and lower face. J Pers Soc Psychol 37:2049–2058. doi:10.1037//0022-3514.37.11.2049
Biele C, Grabowska A (2006) Sex differences in perception of emotion intensity in dynamic and
static facial expressions. Exp Brain Res 171:1–6. doi:10.1007/s00221-005-0254-0
Botvinick M, Jha AP, Bylsma LM, Fabian SA, Solomon PE, Prkachin KM (2005) Viewing facial
expressions of pain engages cortical areas involved in the direct experience of pain. Neuroimage
25:312–319. doi:10.1016/j.neuroimage.2004.11.043
Bould E, Morris N (2008) Role of motion signals in recognizing subtle facial expressions of
emotion. Br J Psychol 99:167–189. doi:10.1348/000712607X206702
Bould E, Morris N, Wink B (2008) Recognising subtle emotional expressions: the role of facial
movements. Cognit Emot 22:1569–1587. doi:10.1080/02699930801921156
Bruce V, Valentine T (1988) When a nod’s as good as a wink: the role of dynamic information in
facial recognition. In: Gruneberg MM, Morris PE, Sykes RN (eds) Practical aspects of memory:
current research and issues, vol 1. John Wiley and Sons, New York, pp 169–174
Bugental DB (1986) Unmasking the “polite smile”. Situational and personal determinants of
managed affect in adult-child interaction. Pers Soc Psychol Bull 12:7–16. doi:10.1177/
0146167286121001
Cosker D, Krumhuber EG, Hilton A (2010) Perception of linear and nonlinear motion properties
using a FACS validated 3D facial model. In: Proceedings of the symposium on applied
Perception in graphics and visualization (APGV), Los Angeles
Cunningham DW, Wallraven C (2009a) The interaction between motion and form in expression
recognition. In: Bodenheimer B, O’Sullivan C (eds) proceedings of the 6th symposium on
applied perception in graphics and visualization (APGV2009), New York
Cunningham DW, Wallraven C (2009b) Dynamic information for the recognition of conversational
expressions. J Vis 9:1–17. doi:10.1167/9.13.7
Dapretto M, Davies MS, Pfeifer JH, Scott AA, Sigman M, Bookheimer SY, Iacoboni M (2006)
Understanding emotions in others: mirror neuron dysfunction in children with autism spectrum
disorders. Nat Neurosci 9:28–30. doi:10.1038/nn1611
Dimberg U (1988) Facial electromyography and the experience of emotion. J Psychophysiol
2:277–282
Edwards K (1998) The face of time: temporal cues in facial expressions of emotion. Psychol Sci
9:270–276. doi:10.1111/1467-9280.00054
Ehrlich SM, Schiano DJ, Sheridan K (2000) Communicating facial affect: it’s not the realism, it’s
the motion. In: Proceedings of ACM CHI 2000 conference on human factors in computing
systems, New York
Ekman P (1985) Telling lies. Norton, New York
Ekman P, Friesen WV (1982) Felt, false, and miserable smiles. J Nonverbal Behav 6:238–252.
doi:10.1007/BF00987191
Ekman P, Friesen WV (1978) Facial action coding system: a technique for the measurement of
facial movement. Consulting Psychologists Press, Palo Alto
Ekman P, Davidson RJ, Friesen WV (1990) The Duchenne smile: emotional expression and brain
physiology: II. J Pers Soc Psychol 58:342–353. doi:10.1037/0022-3514.58.2.342
Fiorentini C, Viviani P (2011) Is there a dynamic advantage for facial expressions. J Vis 11:1–15.
doi:10.1167/11.3.17
Frank MG, Ekman P, Friesen WV (1993) Behavioral markers and recognizability of the smile of
enjoyment. J Pers Soc Psychol 64:83–93. doi:10.1037/0022-3514.64.1.83
Gallese V (2001) The ‘shared manifold’ hypothesis. From mirror neurons to empathy. J Conscious
Stud 8:33–50
Geller T (2008) Overcoming the uncanny valley. IEEE Comput Graph Appl 28:11–17. doi:10.1109/
MCG.2008.79
Gibson JJ (1966) The senses considered as perceptual systems. Houghton Mifflin, Boston
doi:10.1080/00043079.1969.10790296
Hehman E, Flake JK, Freeman JB (2015) Static and dynamic facial cues differentially affect the
consistency of social evaluations. Pers Soc Psychol Bull 41:1123–1134. doi:10.1177/
0146167215591495
Hess U, Kleck RE (1990) Differentiating emotion elicited and deliberate emotional facial expres-
sions. Eur J Soc Psychol 20:369–385. doi:10.1002/ejsp.2420200502
Humphreys GW, Donnely N, Riddoch MJ (1993) Expression is computed separately from facial
identity, and is computed separately for moving and static faces: neuropsychological evidence.
Neuropsychologia 31:173–181. doi:10.1016/0028-3932(93)90045-2
Iacoboni M (2009) Imitation, empathy, and mirror neurons. Annu Rev Psychol 60:653–670.
doi:10.1146/annurev.psych.60.110707.163604
Iacoboni M, Dapretto M (2006) The mirror neuron system and the consequences of its dysfunction.
Nat Rev Neurosci 7:942–951. doi:10.1038/nrn2024
Johansson G (1973) Visual perception of biological motion and a model for its analysis. Percept
Psychophys 14:201–211. doi:10.3758/BF03212378
Kamachi M, Bruce V, Mukaida S, Gyoba J, Yoshikawa S, Akamatsu S (2001) Dynamic properties
influence the perception of facial expressions. Perception 30:875–887. doi:10.1068/p3131
Kappas A, Krumhuber EG, Küster D (2013) Facial behavior. In: Hall JA, Knapp ML (eds)
Nonverbal communication (Handbooks of Communication Science, HOCS 2). Mouton de
Gruyter, Berlin, pp 131–165
Kätsyri J, Sams M (2008) The effect of dynamics on identifying basic emotions from synthetic and
natural faces. Int J Hum Comput Stud 66:233–242. doi:10.1016/j.ijhcs.2007.10.001
Kätsyri J, Förger K, Mäkäräinen M, Takala T (2015) A review of empirical evidence on different
uncanny valley hypotheses: support for perceptual mismatch as one road to the valley of
eeriness. Front Psychol 6:390. doi:10.3389/fpsyg.2015.00390
Kerlow IV (2004) The art of 3D computer animation and effects, 3rd edn. John Wiley and Sons,
Hoboken
Kessler H, Doyen-Waldecker C, Hofer C, Hoffmann H, Traue HC, Abler B (2011) Neural correlates
of the perception of dynamic versus static facial expressions of emotion. GMS Psychosoc Med
8:1–8. doi:10.3205/psm000072
Kilts CD, Egan G, Gideon DA, Ely TD, Hoffmann JM (2003) Dissociable neural pathways are
involved in the recognition of emotion in static and dynamic facial expressions. Neuroimage
18:156–168. doi:10.1006/nimg.2002.1323
Korb S, With S, Niedenthal PM, Kaiser S, Grandjean D (2014) The perception and mimicry of
facial movements predict judgments of smile authenticity. PLoS One 9:e99194. doi:10.1371/
journal.pone.0099194
Krumhuber EG, Tamarit L, Roesch EB, Scherer KR (2012) FACSGen 2.0 animation software:
generating three-dimensional FACS-valid facial expressions for emotion research. Emotion
12:351–363. doi:10.1037/a0026632
Krumhuber EG, Kappas (2005) Moving smiles: the role of dynamic components for the perception
of the genuineness of smiles. J Nonverbal Behav 29:3–24. doi:10.1007/s10919-004-0887-x
Krumhuber EG, Manstead ASR (2009) Can Duchenne smiles be feigned? New evidence on felt and
false smiles. Emotion 9:807–820. doi:10.1037/a0017844
Krumhuber EG, Manstead ASR, Cosker D, Marshall D, Rosin PL (2009) Effects of dynamic
attributes of smiles in human and synthetic faces: a simulated job interview setting. J Nonverbal
Behav 33:1–15. doi:10.1007/s10919-008-0056-8
Krumhuber EG, Manstead ASR, Cosker D, Marshall D, Rosin PL, Kappas A (2007a) Facial
dynamics as indicators of trustworthiness and cooperative behavior. Emotion 7:730–735.
doi:10.1037/1528-3542.7.4.730
Krumhuber EG, Manstead ASR, Kappas A (2007b) Temporal aspects of facial displays in person
and expression perception: the effects of smile dynamics, head-tilt and gender. J Nonverbal
Behav 31:39–56. doi:10.1007/s10919-006-0019-x
Krumhuber EG, Skora P, Küster D, Fou L (in press) A review of dynamic datasets for facial
expression research. Emotion Rev
Küster D, Krumhuber EG, Kappas A (2014) Nonverbal behavior online: a focus on interactions
with and via artificial agents and avatars. In: Kostic A, Chadee D (eds) Social psychology of
nonverbal communications. Palgrave MacMillan, New York, pp 272–302
LaBar KS, Crupain MJ, Vovodic JT, McCarthy G (2003) Dynamic perception of facial affect and
identity in the human brain. Cereb Cortex 13:1023–1033. doi:10.1093/cercor/13.10.1023
Lee TW, Josephs O, Dolan RJ, Critchley HD (2006) Imitating expressions: emotion specific neural
substrates in facial mimicry. Soc Cogn Affect Neurosci 1:122–135. doi:10.1093/scan/nsl012
Lundqvist L, Dimberg U (1995) Facial expressions are contagious. J Psychophysiol 9:203–211
Maringer M, Krumhuber EG, Fischer AH, Niedenthal P (2011) Beyond smile dynamics: mimicry
and beliefs in judgments of smiles. Emotion 11:181–187. doi:10.1037/a0022596
McDonnnell R, Breidt M, Buelthoff HH (2012) Render me real? Investigating the effect of render
style on the perception of animated virtual humans. ACM Trans Graph 31:1–11. doi:10.1145/
2185520.2185587
Mori M (1970) Bukimi No Tani. The Uncanny Valley (MacDorman KF and Minato T, Trans).
Energy 7:33–35
Niedenthal P, Brauer M, Halberstadt JB, Innes-Ker AH (2001) When did her smile drop? Facial
mimicry and the influences of emotional state on the detection of change in emotional expres-
sion. Cognit Emot 15:853–864. doi:10.1080/02699930143000194
Oberman LM, Winkielman P, Ramachandran VS (2007) Face to face: blocking facial mimicry can
selectively impair recognition of emotional expressions. Soc Neurosci 2:167–178. doi:10.1080/
17470910701391943
Piwek L, McKay LS, Pollick FE (2014) Empirical evaluation of the uncanny valley hypothesis fails
to confirm the predicted effect of motion. Cognition 130:271–277. doi:10.1016/j.
cognition.2013.11.001
Ponari M, Conson M, D’Amico NP, Grossi D, Trojano L (2012) Mapping correspondence between
facial mimicry and emotion recognition in healthy subjects. Emotion 12:1398–1403.
doi:10.1037/a0028588
Recio G, Schacht A, Sommer W (2013) Classification of dynamic facial expressions of emotion
presented briefly. Cognit Emot 27:1486–1494. doi:10.1080/02699931.2013.794128
Recio G, Sommer W, Schacht A (2011) Electrophysiological correlates of perceiving and evaluat-
ing static and dynamic facial emotional expressions. Brain Res 1376:66–75. doi:10.1016/j.
brainres.2010.12.041
Rizzolatti G, Craighero L (2004) The mirror-neuron system. Annu Rev Neurosci 27:169–192.
doi:10.1146/annurev.neuro.27.070203.144230
Rizzolatti G, Fadiga L, Gallese V, Fogassi L (1996) Premotor cortex and the recognition of motor
actions. Cogn Brain Res 3:131–141. doi:10.1016/0926-6410(95)00038-0
Rychlowska M, Canadas E, Wood A, Krumhuber EG, Niedenthal P (2014) Blocking mimicry
makes true and false smiles look the same. PLoS One 9:e90876. doi:10.1371/journal.
pone.0090876
Rymarczyk K, Biele C, Grabowska A, Majczynski H (2011) EMG activity in response to static and
dynamic facial expressions. Int J Psychophysiol 79:330–333. doi:10.1016/j.
ijpsycho.2010.11.001
Sato W, Yoshikawa S (2004) Brief report. The dynamic aspects of emotional facial expressions.
Cognit Emot 18:701–710. doi:10.1080/02699930341000176
Sato W, Yoshikawa S (2007a) Spontaneous facial mimicry in response to dynamic facial expres-
sions. Cognition 104:1–18. doi:10.1109/DEVLRN.2005.1490936v
Sato W, Yoshikawa S (2007b) Enhanced experience of emotional arousal in response to dynamic
facial expressions. J Nonverbal Behav 31:119–135. doi:10.1007/s10919-007-0025-7
Sato W, Fujimura T, Suzuki N (2008) Enhanced facial EMG activity in response to dynamic facial
expressions. Int J Psychophysiol 70:70–74. doi:10.1016/j.ijpsycho.2008.06.001
Sato W, Fujimura T, Kochiyama T, Suzuki N (2013) Relationships among facial mimicry, emotional
experience, and emotion recognition. PLoS One 8:e57889. doi:10.1371/journal.pone.0057889
Sato W, Kochiyama T, Yoshikawa S, Naito E, Matsumura M (2004) Enhanced neural activity in
response to dynamic facial expressions of emotion: an fMRI study. Cogn Brain Res 20:81–91.
doi:10.1016/S0926-6410(04)00039-4
Saygin AP, Chaminade T, Ishiguro H, Driver J, Frith C (2012) The thing that should not be:
predictive coding and the uncanny valley in perceiving human and humanoid robot actions. Soc
Cogn Affect Neurosci 7:413–422. doi:10.1093/scan/nsr025
Schmidt KL, Ambadar Z, Cohn J, Reed LI (2006) Movement differences between deliberate and
spontaneous facial expressions: zygomaticus major action in smiling. J Nonverbal Behav
301:37–52. doi:10.1007/s10919-005-0003-x
Schmidt KL, Bhattacharya S, Delinger R (2009) Comparison of deliberate and spontaneous facial
movement in smiles and eyebrow raises. J Nonverbal Behav 33:35–45. doi:10.1007/s10919-
008-0058-6
Schulte-Rüther M, Markowitsch HJ, Fink GR, Piefke M (2007) Mirror neuron and theory of mind
mechanisms involved in face-to-face interactions: a functional magnetic resonance imaging
approach to empathy. J Cogn Neurosci 19:1354–1372. doi:10.1162/jocn.2007.19.8.1354
Singer T, Seymour B, O’Doherty JP, Frith CD (2004) Empathy for pain involves the affective but
not the sensory components of pain. Science 303:1157–1162. doi:10.1126/science.1093535
Stel M, Van Baaren RB, Vonk R (2008) Effects of mimicking: acting prosocially by being
emotionally moved. Eur J Soc Psychol 38:965–976. doi:10.1002/ejsp.472
Thompson JC, Trafton JG, McKnight P (2011) The perception of humanness from the movements
of synthetic agents. Perception 40:695–704. doi:10.1068/p6900
Tinwell A, Grimshaw M, Nabi DA, Williams A (2011) Facial expression of emotion and perception
of the Uncanny Valley in virtual characters. Comput Hum Behav 27:741–749. doi:10.1016/j.
chb.2010.10.018
Trautmann SA, Fehr T, Hermann M (2009) Emotions in motion: dynamic compared to static facial
expressions of disgust and happiness reveal more widespread emotion-specific activations.
Brain Res 1284:100–115. doi:10.1016/j.brainres.2009.05.075
Wallraven C, Breidt M, Cunningham DW, Bülthoff H (2008) Evaluating the perceptual realism of
animated facial expressions. ACM Trans Appl Percept 4:1–20. doi:10.1145/1278760.1278764
Wehrle T, Kaiser S, Schmidt S, Scherer K (2000) Studying the dynamics of emotional expression
using synthetized facial muscle movement. J Pers Soc Psychol 78:105–119. doi:10.1037/0022-
3514.78.1.105
Weiss F, Blum GS, Gleberman L (1987) Anatomically based measurement of facial expressions in
simulated versus hypnotically induced affect. Motiv Emot 11:67–81. doi:10.1007/BF00992214
Weyers P, Mühlberger A, Hefele C, Pauli P (2006) Electromyographic responses to static and
dynamic avatar emotional facial expressions. Psychophysiology 43:45–453. doi:10.1111/
j.1469-8986.2006.00451.x
Yang M, Wang K, Zhang L (2013) Realistic real-time facial expression animation via 3D morphing
target. J Softw 8:418–425. doi:10.4304/jsw.8.2.418-425
Yoshikawa S, Sato W (2008) Dynamic facial expressions of emotion induce representational
momentum. Cogn Affect Behav Neurosci 8:25–31. doi:10.3758/CABN.8.1.25
Utilizing Unsupervised Crowdsourcing
to Develop a Machine Learning Model
for Virtual Human Animation Prediction
Michael Borish and Benjamin Lok
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
CB Framework and VPF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Model Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Application and Real-Time Prediction Adjustments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Crowdsourcing Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Analysis and Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
User Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Crowdsourcing and Expert Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Abstract
One type of experiential learning in the medical domain is chat interactions with a
virtual human. These virtual humans play the role of a patient and allow students
to practice skills such as communication and empathy in a safe, but realistic
sandbox. These interactions last 10–15 min, and the typical virtual human has
approximately 200 responses. Part of the realism of the virtual human’s response
is the associated animation. These animations can be time consuming to create
and associate with each response.
M. Borish (*) • B. Lok

Computer and Information Sciences and Engineering Department, University of Florida,
Gainesville, FL, USA
e-mail: mborish@ufl.edu; lok@cise.ufl.edu

DOI 10.1007/978-3-319-30808-1_21-1
2 M. Borish and B. Lok
We turned to crowdsourcing to assist with this problem. We decomposed the

process of creating basic animations into a simple task that nonexpert workers can
complete. We provided workers with a set of predefined basic animations: six
focused on head animation and nine focused on body animation. These anima-
tions could be mixed and matched for each question/response pair. Then, we used
this unsupervised process to create a machine learning model for animation
prediction: one for head animation and one for body animation. Multiple models
were evaluated and their performance was assessed.
In an experiment, we evaluated participant perception of multiple versions of a
virtual human suffering from dyspepsia (heartburn-like symptoms). For the
version of the virtual human that utilized our machine learning approach, partic-
ipants rated the character’s animation on par with a commercial expert. Head
animation specifically was rated more natural and typically expected than other
versions. Additionally, analysis of time and cost show the machine learning
approach to be quicker and cheaper than an expert alternative.
Keywords
Crowdsourcing • Machine Learning • Virtual Human • Animation Pipeline
Introduction
One style of experiential learning in the medical domain is chat interactions with
virtual humans. Virtual human interactions allow students to learn by interacting with
and observing the response of a virtual human. A typical example of this style of
interaction is shown in Fig. 1. Here, students type in questions to the virtual patient and
receive responses. Students usually interact with multiple versions of a character that
differ in certain details. A single interaction lasts 10–15 min and trains the student on
appropriate questions to ask in order to diagnose a patient. To facilitate these interac-
tions, the virtual human typically contains approximately 200 responses.
Part of the realism of a virtual human response is animation. Reasonable response
animations play a role in believability of a virtual human and fulfill our natural
coherency model with how a human should behave. Animations also play a role in
emotion and personality expression (Cassell and Thorisson 1999). However, devel-
oping animations for such a large number of responses can contribute to significant
logistical, technological, and content requirements necessary to deliver an effective
interaction (Triola et al. 2007, 2012).
There are numerous approaches for creating the necessary animations for a virtual
human. Approaches include automated and procedural algorithms, motion capture
pipelines, and animation experts. All of these approaches have trade-offs when
considering cost, time, and quality. We propose the use of crowdsourcing as an
alternative to create a machine learning model for animation prediction. This pre-
diction model is specifically meant for virtual humans playing the roles of patients
for medical interviews.
Utilizing Unsupervised Crowdsourcing to Develop a Machine Learning Model for. . . 3
Fig. 1 Example interaction page – student interacts with a virtual patient by typing questions and
receiving responses while also tracking their interview progress
We first decomposed the process of adding animations to a virtual human into a

simple task that nonexperts could complete. We then leveraged nonexperts recruited
from Amazon’s Mechanical Turk Marketplace to provide creative input to generate
complex animations from a small, generic set of 15 basic animations. These animations
then provided the basis for a machine learning model to predict future animations.
In this chapter, we describe the machine learning models created via our
crowdsourcing process. We then apply the prediction models to a virtual human
and report on an experiment. In this experiment, participants reported improved
perceptions of the virtual human using these prediction models. We also provide a
detailed analysis of the prediction models separate from the experiment as well as a
time and cost analysis.
State of the Art
The effort and people required to construct a virtual human scenario is significant
(Triola et al. 2007, 2012). One area of significant effort is the realism of virtual
humans afforded through animations. Expert modelers and animators represent a
gold standard in providing realism as the skills these experts utilize can take many
years to acquire. An expert was compared to our machine learning models, and
details of this comparison will be discussed in both the Experiment and Results
section. This expert has over 15 years of modeling and animation experience in the
video game industry with over a dozen games to his name. While the contributions
of this individual are of high quality, the effort and time provided by this expert are
substantial. When faced with limited resources, various attempts to automate the
creation and application of animations have been made.
Relative success with procedural algorithms has been found in facial animation.
Work such as Hoon et al. (2014) has shown automatic generation of facial expres-
sions to be possible and effective. In work by Brand and Hertzmann (2000),
reference motion capture clips were used as the basis of new animation synthesis.
Additionally, Deng et al. (2009) presented a method of example-based animation
selection and creation for virtual characters.
Similarly, Min and Chai (2012) developed a methodology for procedurally
generated animation via short descriptions such as “walk four steps to green desti-
nation.” In this work too, motion capture clips were used as reference. Our work
builds upon a similar structure as reference animation blocks were used to construct
more complex animations. However, unlike this work, our model does not rely
exclusively on motion capture data or expert coding. Rather, our model could be
applied to a variety of animations and all construction is handled via crowdsourcing.
Generation of animation from user cues has also been explored (Cassell and
Thorisson 1999; Sargin et al. 2006). While this work has been successful, typically,
the user is evaluated during the interaction for behavior such as posture and
intonation. While such analysis provides additional features to evaluate, we limit
ourselves to text of a virtual human’s response and basic parameters of the audio.
This allows our model to be used for virtual humans in large classroom settings,
typical of many medical schools. In these settings, individual simulation presentation
is not feasible, and a student is likely to interact via laptop in informal surroundings.
In-depth analysis of audio and visual components has also been used with
success. In both Marsella et al. (2013) and and Levine and Theobolt (Levine and
Theobolt 2009), in-depth analysis of audio cues were evaluated to predict appropri-
ate gestures for a virtual human response. These gestures were created using motion
capture clips as a ground truth to establish timings and constraints on the gestures.
Similarly, in Xu et al. (2014), audio and visual analysis was conducted to structure
the gesture generation into ideational units. Ideational units are conceptual units that
bind verbal and nonverbal behavior together as well as provide constraints on
various attributes of interaction such as transitions and rhythm.
All of this work assumes the existence of databases of motion capture information
or video clips tagged by experts with a variety of potentially complex pieces of
information. While our system does require annotation information, the information
is simpler than in similar systems and can be provided on demand by crowdsourced
workers. Additionally, this simple information can produce the same affect as an
expertly animated character.
CB Framework and VPF
Our animation process builds upon both the Crowdsource Bootstrapping

(CB) Framework (Borish and Lok 2016) and Virtual People Factory (VPF) (Rossen
et al. 2010; Rossen and Lok 2012). VPF is a web-based tool for the creation and
improvement of virtual humans. VPF also facilitates online virtual human interac-
tions and has been integrated into multiple classes at several universities. VPF has
been used by thousands of medical, health, and pharmacy students to practice
interpersonal skills and develop diagnostic reasoning.
The CB Framework is a gateway tool for VPF that allows an educator to rapidly
develop a new virtual experience once a need is identified. The CB Framework
decomposes the process of virtual human creation into several discrete steps that
utilize crowdsourced nonexperts. The completion of these steps results in a basic
virtual human corpus. The corpus is the structured set of text that comprises the
knowledge of the virtual human. These stages can be completed in a matter of hours
with minimal commitment from the author.
With the initial stages of the CB Framework complete, our animation process can
be applied as a subsequent stage in the CB Framework. Alternatively, this process
can be applied to already existing virtual humans as well. However, the
crowdsourcing and machine learning models are framework agnostic, and the
implementation described in the subsequent section can be applied to any creation
pipeline or virtual human.
Implementation
In order to rapidly create animation predictions for the virtual human’s responses,
several discrete steps are necessary. The dataflow outlining these steps is shown in
Fig. 2. The animation predictions for each response occur before our virtual human
begins an interaction. These predictions are stored as part of the virtual human’s
corpus. Any additional crowdsourcing also occurs as part of this process. Once the
interaction begins, real-time adjustments to the predictions are needed.
First, we will describe feature selection and prediction model specifics. We will
then discuss real-time adjustments needed to combat repetition in the application of
the models. Lastly, we will discuss details related to the crowdsourcing task used to
create the prediction models.
Model Metrics
In order to create the animation prediction models, we focused on two sets of related
features: sentiment and lexical similarity. We reasoned that an accurate animation
choice for a specific response should be similar in both sentiment and lexical content.
To facilitate these features, we utilized N-grams. N-grams are often used in NLP
analysis, and bigrams have proven effective when utilized for sentiment analysis
Highly Ranked
Input Response
Input NLP Predicted Animation Shuffled and Sentiment-
Prediction Model
Response comparison appropriate Animation
Real-time Adjustment
Lowly Ranked
Input Response
Crowdsourcing
Fig. 2 Example dataflow for a response that utilizes our animation prediction system and the role
of crowdsourcing within it
(Wang and Cardie 2014). Since multiple sentiment metrics were used as part of the
animation prediction model, we too utilized bigrams as well as unigrams as features
for prediction. All feature and machine learning model analyses were carried out
using Weka (Hall et al. 2009). A list of the features used is as follows:
• Sentence Sentiment – overall sentiment for the entire sentence was provided as
part of the crowdsourcing task. This will be discussed in more detail in section
“Crowdsourcing Task”.
• Bigram Sentiment – automated calculations of sentiment were provided by the
Stanford NLP Sentiment Pipeline (Socher et al. 2013). Sentiment analysis
consisted of five categories including very negative, negative, neutral, positive,
and very positive. These five categories form a distribution of overall sentiment
opinion. Further, five additional metrics were calculated. These metrics included
kurtosis, skewness, minimum, maximum, and range of the sentiment distribution.
These metrics were calculated to describe the overall shape and agreement of the
distribution. Bigram and sentence sentiment have previously been used to create
virtual human animation (Hoque et al. 2013). Many previous systems are gener-
ally concerned with facial animation resulting from specific emotions. Our system
is concerned with body animation; however, like facial animation, body anima-
tion is informed in part by sentiment.
• Bigram Position and Total – the position of a bigram in the sentence as well as the
total number of bigrams in a sentence were included. We found bigrams near the
beginning of sentences to be of relatively higher importance. Typically,
crowdsourced workers would favor the beginning of sentences to assign an
animation even in multi-sentence responses. This makes intuitive sense as ani-
mations would be expected to begin when the speech for a response does.
Sentence and clause boundaries have also been shown to be important locations
for information in a sentence including head motion (Toshinori et al. 2014).
• Bigram Part of Speech (POS) – bigram POS was also included. POS tagging for
individual words was provided by the Stanford CoreNLP Pipeline. Each bigram
POS was an aggregation of the POS tagging of the individual words that comprise
the bigram.
• Bigram and Bag of Words – the actual bigram as well as a “bag of words”
approach to the sentence the bigram was drawn from was included in the
model. We reasoned that beyond sentiment similarity and location. Any predic-
tion should be based on lexically similar bigrams and sentences. Thus, we broke
each sentence into unigrams for a “bag of words” approach commonly used for
lexical similarity.
• Head Animation – this feature was only included in the body animation predic-
tion model. Feature evaluation indicated that the head animation was the single
best predictor. Thus, the body prediction model forms a small predictor chain
whereby head animations are predicted first. Then, all listed features including the
predicted head animation are used to predict body animation.
Application and Real-Time Prediction Adjustments
The animation prediction models previously described can suffer from repetition.
For example, if two subsequent responses that occur during an interaction were “No,
I don’t drink.” and “No, I don’t smoke.” similar predictions of “Shake No Once” and
“Arms Crossed” might be predicted. While both predictions would be correct from a
machine learning perspective, the repetition would hurt user perception during the
actual interaction. So, our system also adds an additional layer of animation selection
logic at interaction time. This selection logic is a simple shuffled deck algorithm. A
shuffled deck algorithm randomly shuffles all items into a list and iterates over that
list. Then, all items are reshuffled and the process repeated. This process creates a
pseudo-random selection. For head animations, the shuffled deck algorithm only
shuffles animations that have the same general sentiment grouping. Body animations
were simply shuffled regardless of sentiment. As will be shown in Tables 3 and 5 in
the Results section, there is a clear grouping for head animations while body
animations do not show the same pattern.
Before an interaction begins, animation prediction is applied to each response in a
virtual human’s corpus. First, each bigram in the input response has head and body
animations predicted. Then, these predictions are culled. Animation predictions for
bigrams at the start of sentences are prioritized due to the importance of sentence
boundaries as previously explained. Once these predictions are selected, remaining
time is greedily filled. Remaining time is calculated based on access to the audio file
associated with the input response and the length of the animations already
predicted. Additionally, no animations that extend past the end of the audio file
will be suggested. This restriction is to prevent the character from performing
gestures after the response to a question is complete.
A similarity evaluation takes place separately and simultaneously from animation
prediction. This evaluation compares the full input response to the responses used in
construction of the animation prediction model. The probability that the input
response is a paraphrase of any of the other responses is calculated. This comparison
is exactly the same as the NLP algorithm that performs paraphrase selection during a
conversation in VPF and is based on the work of McClendon et al. (2014). While
animation predictions will still be returned regardless of score, scores below a certain
threshold are sent to the crowdsourcing task to be evaluated by workers and included
for future predictions. In this way, we increase the size of the prediction model
whenever a lexically dissimilar response is encountered during the prediction
process.
Crowdsourcing Task
As previously mentioned, whenever a lexically dissimilar response is encountered

during the prediction process, animation predictions are still provided. However, that
input response is sent to a crowdsourcing task for future inclusion in the prediction
model. The interface for the crowdsourcing task is shown in Fig. 3. Here, at the top,
workers are shown a simple unity scene with the character. Below the scene, workers
are presented with the current question/response pair for which animation assign-
ment will take place. The words that comprise the response are spread across two
timelines: one for head and one for body animations. These timelines represent the
length of the audio and all animations are sized proportional to this length for each
question/response pair. Proportional sizing of the animations was done so that
animation length would be commensurate with the speech and not be assigned or
playing for a significant period of time afterwards. In this area, there is also a button
to play the entire timeline. This button allows workers to see how their animation
selections appear in context with the audio and facial animations that the virtual
character normally has for the given response. Workers can also click on any of the
individual animations to have the virtual character perform a single, simple anima-
tion in order to better visualize the resulting timeline selection. On the bottom right
are the animation lists and each animation can be dragged to the corresponding
timeline. For example, in Fig. 3, for the question/response pair of “Can you read the
lines?” and “I can’t read any of the lines.” a worker might drag a “Shake No Once”
and “Hand Gesture” animation to the head and body timelines respectively.
By providing a defined set of animations, overall task difficulty can be reduced.
Low task difficulty leads to lower variability in responses by workers and overall
better quality data (Callison-Burch and Dredze 2010). Additionally, workers can be
confident they are suggesting an animation that is worthwhile since a majority of
question/response pairs should be covered by the predefined set of generic animation
building blocks. Further, confidence of the workers also plays a role in the quality of
data that is provided (Madirolas and de Polavieja 2014).
Once animation selections are complete, workers were also asked to provide a
rating for the overall sentiment of the response. Worker choices were limited to
negative, neutral, and positive. The ultimate sentiment of a response was determined
by simple majority vote. This vote occured after at least three workers provided
animation and sentiment selections for a question/response pair.
Fig. 3 Crowdsourced worker task page – workers suggest animations for a question/response pair
from a predefined, base set
Compensation payment for this task was set at $.40 a Human-Intelligence Task
(HIT) on Amazon’s Mechanical Turk service and workers were asked to complete
five question/response pairs. The HIT compensation was selected at this level due to
previous pilot study findings related to the effect of payment on quality and completion
time. Paying workers additional money does not improve the overall quality of data as
there has been no link found between compensation and quality of work (Mason et al.
2009). As a result, keeping compensation reasonable will result in a lower overall cost
for the task without a detriment to quality. In contrast, low compensation reduces the
incentive by mainstream workers to complete your task. Additionally, high compen-
sation as Adda et al. (2013) note can attract malicious workers that game the task in
order to maximize payment and thus reduces data quality.
Analysis and Experiment
We conducted an experiment to evaluate user perception of a virtual human utilizing

these head and body animation prediction models. Four versions of a virtual patient
suffering from dyspepsia (heartburn-like symptoms) were created. The first version
is our control and consisted of a virtual patient whose only animation was lip syncing
and an idle “breathing” animation. This version will be referred to as Dcontrol. The
second version added response animations randomly selected from the set of possi-
ble animations supplied to crowdsourced workers. This version will be Drandom. The
third version had animations suggested by two predictive models: one for head
animation and one for body animation as described in the last section. This version
will be referred to as DML. Finally, DPro was created by an animation expert. This
expert was provided the character and audio files and was allowed to animate the
character as he saw fit. This animator regularly produces animations for commercial
projects and serves as a gold standard comparison.
We also conducted an in-depth analysis of the predictive models in isolation as
well as a comparison of cost/time for the crowdsourcing process and expert anima-
tor. These will be presented separately.
Procedure
Participants were recruited and paid US$1.00 to view a video showing a previous
interaction between a student and one of the virtual humans previously described.
Participants then filled out a survey. Previous interaction logs were used to create the
interaction and the same interaction was used for all four versions of the dyspepsia
virtual human. The video was approximately 5 min in length.
The survey consisted of questions based on assessments for naturalness of virtual
human interactions (Huang et al. 2011; Ho and MacDorman 2010). The questions
were answered on a 7-point scale from 1- Not at all to 7- Absolutely and were as
follows:
Q1 Do you think the virtual agent’s overall behavior is natural?

Q2 Do you think the virtual agent’s overall behavior is sincere?
Q3 Do you think the virtual agent’s overall behavior is typical?
Q4 Do you think the virtual agent’s head behavior is natural?
Q5 Do you think the virtual agent’s head behavior is sincere?
Q6 Do you think the virtual agent’s head behavior is typical?
Q7 Do you think the virtual agent’s body behavior is natural?
Q8 Do you think the virtual agent’s body behavior is sincere?
Q9 Do you think the virtual agent’s body behavior is typical?
The participants also described the virtual human on a number of attributes. Each
attribute was on a 7-point binary scale. Those attributes were:
Q10 Artificial – Natural

Q11 Synthetic – Real
Q12 Human-made – Humanlike
Q13 Mechanical – Biological movement
Q14 Predictable – Thrilling
Q15 Passive – Animated
Q16 Smooth/graceful – Sudden/jerky movement
Results and Discussion
In total, N = 89 participants were recruited and were distributed as follows: N = 16

viewed Dcontrol, N = 23 viewed Drandom, N = 25 viewed DML, and N = 25 viewed
DPro.
Prediction Model
The animations used in the prediction models were taken from previous virtual
humans. A simple description of the animations is provided below. Animations A–F
represent the head animations, while animations G–O represent the body animations.
A NodYesOnce – Single head nod

B NodYesTwice – Multiple head nods
C TiltHeadLeft – Head tilt looking toward the left side
D TiltHeadRight – Head tilt looking toward the right side
E ShakeNoOnce – Single head shake
F ShakeNoTwice – Multiple head shakes
G HandsInLap – Both hands are placed in the lap
H ArmsSweepOut – Arms progress from low to high position sweeping out from the body
I ArmsCrossed – Both arms are crossed in front of body
J HandFlickGestureA – Both hands are raised and in motion in front of the body. The
animation ends with the hands flicking away from the body
K HandFlickGestureB – Similar to the previous gesture, however, the length of arm motion is
shortened and the flick is less pronounced
L ScratchHeadLeft – Left hand is used to scratch the head
M ScratchHeadRight – Right hand is used to scratch the head
N HandGesture – Arms are up in front of the body at alternate times
O Shrug – Simple shoulder shrug
Multiple models were evaluated using both test sets created from 10 % of the data
set and 10-fold cross-validation. The head prediction model contained 992 entries,
while the body prediction model contained 939. Due to the relatively small size of
the test sets, we present 10-fold cross-validation as more representative of perfor-
mance. The overall accuracy of several different models is shown in Table 1.
Ultimately, we settled on the use of a Bayesian network. The Bayesian net
outperformed simpler models and was also on par with more computationally
expensive models.
We also investigated how often specific animations were applied by
crowdsourcing workers according to the sentiment of the response. Tables 2 and 3
show the confusion matrix and sentiment distribution for the head prediction model.
Tables 4 and 5 show the confusion matrix and sentiment distribution for the body
prediction model.
Table 1 Model Head model Body model

accuracies – percentage
J48 42.9 16.4
accuracies for a selection of
different machine learning Naive Bayes 37.8 14.8
models for head and body Bayesian Net 48.4 19.8
animation Multilayer Preceptron 21.7 16.6
Table 2 Head confusion A B C D E F

matrix
131 37 48 1 1 1 A
77 22 19 0 0 0 B
20 7 157 0 16 4 C
10 6 110 0 13 3 D
0 0 10 0 153 15 E
0 0 11 0 103 17 F
Table 3 Sentiment of head Animation Negative Neutral Positive

animations according to
A 1.0 20.4 78.6
crowdsourced workers (all
values are percentages) B 1.0 14.4 84.6
C 8.6 78.0 13.4
D 12.4 76.1 11.5
E 99.4 0.6 0
F 94.2 5.8 0
Table 4 Body confusion G H I J K L M N O

matrix
32 58 0 0 0 0 0 45 38 G
16 63 1 0 0 0 0 25 28 H
27 50 1 0 0 0 0 21 34 I
15 26 0 0 0 0 0 24 26 J
10 14 2 0 0 0 0 14 16 K
12 18 0 0 0 0 0 16 28 L
7 13 1 0 0 0 0 7 10 M
35 28 1 0 0 0 0 42 35 N
12 31 1 0 0 0 0 13 43 O
As can be seen from the confusion matrix in Table 2, the head prediction is
reasonably accurate overall and achieves an accuracy of 48.4 %. This is significantly
better than a random chance of 16.7 % for a classification with six alternatives.
Further, when the prediction model is incorrect, it is generally incorrect in a
reasonable way. Most of the errors in the confusion matrix are with interchangeable
animations. For example, if the participant were to ask “Do you do any drugs?” and
receive a response of “No, I don’t do any illegal drugs.” then the typical response
would include an animation in which the virtual human is shaking its head
no. Whether the model predicts shaking the head no once or multiple times, both
Table 5 Sentiment of Animation Negative Neutral Positive

body animations according
G 36.5 25.8 37.7
to crowdsourced workers
(all values are percentages) H 50.8 22.7 26.6
I 37.5 29.7 32.8
J 26.3 40.0 33.8
K 31.3 29.2 39.6
L 28.0 40.0 32.0
M 36.1 36.1 27.8
N 21.4 40.5 38.2
O 34.4 46.9 18.8
are reasonable. The same holds true for the other head animations. The interchanging
of certain animations is also shown in the sentiment distribution. There are three
clear groupings in the distribution as assigned by crowdsourced workers. The highly
skewed distributions indicate agreement among workers as to when certain head
animations are expected based on the sentiment of the response. Indeed, feature
evaluation shows that worker assigned sentiment is the best predictor of head
animation out of all features.
While the head animation prediction model showed clear patterns and groupings,
the same is not true for the body prediction model. The confusion matrix and
sentiment distributions are shown in Tables 4 and 5, respectively.
As can be seen in Table 4, the overall accuracy is lower than the head prediction
model and is around 20 %. This is still higher than random chance at 11 %, but
several of the animations are simply never predicted. Alternative simpler models
such as naive Bayes and J48 do capture some of these predictions; however, the
overall accuracy of these models is even lower. Again, similar to head prediction,
more complex models did not produce an increase in overall accuracy. The reason
for this performance becomes clearer when looking at the sentiment distribution for
body animations.
As shown in Table 5, the sentiment is roughly evenly distributed for each of the
body animations. The body animations do not show any clear groupings that were
evident for head animations. This relatively equal use of a body animation regardless
of response sentiment indicate crowdsourced workers could not truly arrive at an
agreement on what a “correct” body animation was for a given response.
Our use of a neutral medical interview scenario could be one factor. Many of our
virtual humans such as the dyspepsia scenario are meant to allow students to practice
basic interviewing skills. The students are learning what questions to ask and how to
gather information. A typical example of this interview would be an exchange
similar to the following: “Do you do any drugs?”, “No, I don’t do any illegal
drugs.”, “Do you drink at all?”, “Yeah, I have a couple of beers a week.” For both
of these responses, there are expected head behaviors. The first response would have
some variation of shaking no, while the second response would contain some
variation of nodding yes. However, what is the “correct” body animation?
In such an example, a patient might simply need to be animated to be believable

without any pattern. An alternative scenario such as revealing a diagnosis of cancer
may contain emotional moments where specific body motions would be expected
and represents one avenue of future work.
Another factor might have been the coarse grained nature of control provided to
crowdsourced workers. Workers were allowed only to place animations on the
appropriate timeline. However, there are multiple ways to provide more fine-grained
control. Two examples would be specific gaze targets to accompany the head
animation and speed control to allow fine tuning of an animation. With additional
controls, a more notable pattern to the body animations might present itself. The
expansion of the controls is still another opportunity for future investigation.
User Perception
An ANOVA was calculated for each survey response, Q1–Q16, listed in the previous
section. For each question for which statistical significance was found, a Tukey
analysis was conducted. While the ANOVA was significant for questions Q7, Q8,
Q10, Q11, Q13, Q14, and Q15, the Tukey analysis did not show any interesting
results. For all these questions, statistical significance for a difference in means was
between Dcontrol and one of the other models. This makes sense as any type of
animation would increase user perception over a virtual human with no animation.
Tukey analysis showed results of interest for Q4 and Q6. These questions asked
whether or not the virtual human head behavior is natural or typical, respectively.
The results are summarized in Table 6.
Users found the head animation of DML more natural and more typical than both
DControl and DRandom. Further, DML was found to be comparable to DPro. However,
no difference was found for the overall or body animations. These differences make
sense in context of the model analysis from the previous section.
With clear sentiment groupings and patterns for the head prediction model,
crowdsourced workers generally agree that there exists a “correct” head animation
for specific responses. This is reflected in the improved ratings for the virtual human.
Additionally, the machine learning models produced from the crowdsourced
workers produced the same affect as an expert animator. As we will discuss in the
next section, these models were created cheaper and faster than the expert version.
These savings make this an attractive alternative.
In contrast, the body animation model consisted of animations whose sentiment
was evenly distributed across all categories. There were no clear patterns and
crowdsourced workers could not agree on a “correct” body animation for a response.
This lack of agreement aligns with user perception as body predictions for DML were
not perceived any differently from the other versions.
As participant perception shows, appropriate animation for a virtual human’s
response is important. In the medical domain, realistic virtual human interactions are
an increasingly used educational component. This realism is directly effected by the
appropriate choice of head animation and a reasonable choice for body animation.
Table 6 User perception results with significant Tukey analysis

Tukey
Question FAnova pAnova significance Mean SD
Do you think 7.593 .000 DML- DControl = 3.50 DControl = 1.50
the virtual Dpro = .062 DRandom = 4.360 DRandom = 1.15
agent’s head DML- Dml = 5.38 D
ml = 1.01
behavior is DRandom = .025 DPro = 4.48 DPro = 1.32
natural? DML-
DControl = .000
Do you think 7.194 .000 DML- DControl = 3.44 DControl = 1.75
the virtual DPro = .105 DRandom = 4.40 DRandom = 1.29
agent’s head DML- Dml =5.45 Dml = 0.93
behavior is DRandom = .049 DPro = 4.56 DPro = 1.50
typical? DML-
DControl = .000
Additionally, the animations predicted by our machine learning approach produce

the same affect as an expertly animated character.
Crowdsourcing and Expert Cost
As shown in Table 7, the cost and time per response for the machine learning model
construction is much lower than the use of an expert. The model construction
required 15.75 h of effort and cost $75.00, while the expert animator required
25.75 h of effort and $901.25 to complete their work. While time will always be
variable in a crowdsourced approach, the construction effort occurred over an
approximately 24 h period. We are confident that 24–48 h is plenty of time to
accomplish model construction in the future. The expert required one week to
accomplish animation of the responses.
Additionally, the machine learning models covered a larger breadth of responses
and was constructed from 314 responses, while the expert animator worked on only
34 responses. The animator focused on these 34 responses because those were the
responses necessary for the 5 min video shown to participants. This creates a cost
and time per response as shown in Table 7. Based on these results, if every response
in the 314 responses used in the machine learning model required a unique anima-
tion, an expert animator would require approximately six weeks of full-time work
and $8,000.00 to complete the work.
The machine learning models utilized for DML have several clear advantages. The
models cover a larger breadth of responses and were evaluated much more quickly.
Further, while there will certainly be some reuse of the animations created by the
expert, the animations were not intended to be generic and reusable.
These results suggest a shift in the role of expert animators from covering all
responses to covering only necessary responses. As previously described, our system
performs a similarity comparison. This comparison is a paraphrase identification
algorithm. This algorithm determines if a particular response is a paraphrase than any
Table 7 Time and cost estimates

Time per response (minutes) Cost per response (US dollars)
ML model construction 3 $0.24
Animation expert 45 $26.51
other response previously encountered. For those responses that are scored low, an
expert could create the necessary animation. Alternatively, different approaches that
specifically address unfamiliar content could be used.
For responses that score highly, the expert’s time is better spent elsewhere as the
machine learning models can produce the same affect quicker and cheaper. These
highly scoring responses are likely to make up the bulk of any virtual human corpus
for a chat interaction in the medical domain. As mentioned at the beginning of this
chapter, these virtual humans have several hundred responses but contain numerous
similarities as multiple versions of the same scenario are often created. For example,
most virtual humans would be asked whether or not they smoke, drink, or do drugs.
Expert animators’ time can be utilized more efficiently by targeting information that
requires their attention rather than these common responses.
Our crowdsourcing approach also offers the benefit of continuous improvement.
Whenever an unfamiliar response is encountered, regardless of whether an anima-
tion expert provides a new animation, the crowdsourcing algorithm can quickly
incorporate new information on demand. The crowdsourcing algorithm can send the
response to workers who provide timeline selections and sentiment scores. Our
system also avoids requiring experts to provide complex information tagging as in
related systems previously described.
Limitations
Our prediction models have been applied to virtual humans in the medical domain
for a dialogue type interaction with a relatively short conversation time of
10–15 min. While this process would generalize to similar dialogue domains,
other types of interactions such as instruction or open-ended conversations may
not be applicable. Such interaction styles may not have the same type of information
overlap present here and may suffer from repetition from this model, even with the
mitigation described. However, this does present an opportunity for future improve-
ment as crowdsourced workers could be given a larger set of animations to work
with and additional controls. This does need to be balanced with the increase in task
difficulty and the potential decrease in agreement among workers.
Our prediction models are also bias based on the interaction context and domain.
Virtual human chat interactions in the medical domain are meant to teach students
how to ask questions and retrieve information. A majority of these questions have
muted emotional context as they are usually matter of fact. When emotional context
is involved, the context is skewed negative toward sad emotions as the virtual
humans are suffering from some medical issue. These biases must be accounted
for and the models trained on appropriate data if the domain differs from what is
described here.
Future Directions
Our process has demonstrated the unsupervised creation of an animation prediction

model via crowdsourced workers to be a viable alternative to more resource inten-
sive creation methods. A virtual human whose animations are assigned by our
prediction models has head animation that is regarded as more natural and typical
by participants. Based on crowdsourced workers input, any reasonable body anima-
tion will suffice as no difference in perception occured. This was in alignment with
the data collected for the machine learning model that did not find a clear agreement
on what constitutes a “correct” body animation for a general virtual human response.
The virtual human animated using our prediction model produces similar affect to
a version of the virtual human animated by an expert. Importantly, this crowdsourced
model is cheaper and faster to create, and can also be updated on demand through the
use of additional crowdsourcing. These benefits highlight the need to refocus experts
only on the information that requires their attention while leaving mundane
responses to be animated automatically by our models. By freeing experts from
repetitious work, our system aims to reduce the barrier to creation to allow virtual
humans to be utilized more widely in medical education.
We intend to pursue this research with additional studies. One such study we are
planning is expansion of the crowdsourcing task. With additional controls such as
animation speed and gaze targets, additional patterns may present themselves.
We also intend to continue development of the body prediction model. As
mentioned, specific emotional events or scenarios may play a role in the “correct”--
ness of body animation, and we plan to investigate such cases with models specif-
ically tuned to those situations. Additionally, we plan to further investigate whether
other features may be required to identify an already existing pattern in body
animations.
References
Adda G, Mariani J, Besacier L, Gelas H (2013) Economic and ethical background of crowdsourcing
for speech. In: Crowdsourcing for speech processing: applications to data collection, pp
303–334
Borish M, Lok B (2016) Rapid low-cost virtual human bootstrapping via the crowd. Trans Intell
Syst Technol 7(4):47
Brand M, Hertzmann A (2000) Style machines. In: 27th SIGGRAPH, pp 183–192
Callison-Burch C, Dredze M (2010) Creating speech and language data with Amazons Mechanical
Turk. In: Proceedings of the NAACL HLT 2010 workshop on creating speech and language data
with Amazon’s Mechanical Turk, number June, pp 1–12
Cassell J, Thorisson KR (1999) The power of a nod and a glance: envelope vs. emotional feedback
in animated conversational agents. Appl Artif Intell 13(4–5):519–538
Deng Z, Gu Q, Li Q (2009) Perceptually consistent example-based human motion retrieval. In:

Interactive 3D graphics and games, vol 1, pp 191–198
Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH (2009) The WEKA data mining
software. ACM SIGKDD Explor Newsl 11(1):10
Ho C-C, MacDorman KF (2010) Revisiting the uncanny valley theory: developing and validating
an alternative to the Godspeed indices. Comput Hum Behav 26(6):1508–1518
Hoon LN, Chai WY, Aidil K, Abd A (2014) Development of real-time lip sync animation
framework based on viseme human speech. Arch Des Res 27(4):19–29
Hoque ME, Courgeon M, Mutlu B, Picard RW, Link C, Martin JC (2013) MACH: My Automated
Conversation coacH. In: Pervasive and ubiquitous computing, pp 697–706
Huang L, Morency LP, Gratch J (2011) Virtual rapport 2.0. In: Intelligent virtual agents, pp 68–79
Levine S, Theobalt C (2009) Real-time prosody-driven synthesis of body language. ACM Trans
Graph 28(5):17
Madirolas G, de Polavieja G (2014) Wisdom of the confident: using social interactions to eliminate
the bias in wisdom of the crowds. In: Collective intelligence, pp 2012–2015
Marsella S, Lhommet M, Feng A (2013) Virtual character performance from speech. In: 12th
SIGGRAPH/Eurographics symposium on computer animation, pp 25–35
Mason W, Street W, Watts DJ (2009) Financial incentives and the performance of crowds. SIGKDD
11(2):100–108
Mcclendon JL, Mack NA, Hodges LF (2014) The use of paraphrase identification in the retrieval of
appropriate responses for script based conversational agents. In: Twenty-seventh international
flairs conference, pp 19–201
Min J, Chai J (2012) Motion graphs++. ACM Trans Graph 31(6):153
Rossen B, Lok B (2012) A crowdsourcing method to develop virtual human conversational agents.
IJHCS 70(4):301–319
Rossen B, Cendan J, Lok B (2010) Using virtual humans to bootstrap the creation of other virtual
humans. In: Intelligent virtual agents, pp 392–398
Sargin ME, Aran O, Karpov A, Ofli F, Yasinnik Y, Wilson S, Erzin E, Yemez Y, Tekalp AM (2006)
Combined gesture-speech analysis and speech driven gesture synthesis. In: Multimedia and
Expo, number Jan 2016, pp 893–896
Socher R, Perelygin A, Wu JY, Chuang J, Manning CD, Ng AY, Potts C (2013) Recursive deep
models for semantic compositionality over a sentiment treebank. In: EMNLP, p 1642
Toshinori C, Ishiguro H, Hagita N (2014) Analysis of relationship between head motion events and
speech in dialogue conversations. Speech Comm 57:233–243
Triola MM, Campion N, Mcgee JB, Albright S, Greene P, Smothers V, Ellaway R (2007) An XML
standard for virtual patients: exchanging case-based simulations in medical education. In:
AMIA, pp 741–745
Triola MM, Huwendiek S, Levinson AJ, Cook DA (2012) New directions in e-learning research in
health professions education: report of two symposia. Med Teach 34(1):15–20
Wang L, Cardie C (2014) Improving agreement and disagreement identification in online discus-
sions with a socially-tuned sentiment lexicon. In: ACL, vol 97, p 97
Xu Y, Pelachaud C, Marsella S (2014) Compound gesture generation: a model based on ideational
units. In: IVA, pp 477–491
Clinical Gait Assessment by Video
Observation and 2D Techniques
Andreas Kranzl
Abstract
Observational gait analysis, in particular video-based gait analysis, is extremely
valuable in the daily clinical routine. Certain requirements are necessary in order
to be able to perform a high-quality analysis. The walking distance must be
sufficiently long enough (dependent on the type of patient), the utilized equip-
ment should meet the requirements, and there should be a recording log. The
quality of the videos for evaluation is dependent on the recording conditions of
the video cameras. Exposure time, additional lighting, and camera position all
need to be adjusted properly for sagittal and frontal imaging. Filming the video in
a room designated for this purpose will help to ensure constant recording condi-
tions and quality. The recordings should always be carried out based on
a recording log. The test form can act as a guide for the evaluation of the video.
This provides an objective description of the gait. It is important to always keep in
mind that the evaluation must remain subjective to a certain degree. Based on the
gait parameter, the reproducibility of this value (intra- and inter-reliability) is
moderate to good. In addition to a database function, current video recording
software is able to measure angles and distances. It should also be possible to play
back two videos in parallel, in order to, for example, play back both the pre-
surgical and postsurgical gait simultaneously. Despite the implementation of
three-dimensional measurement systems for gait analysis, observation or video-
supported gait analysis is justified in daily clinical operations.
Keywords
Database • Observational gait • Reliability • Room size • Video camera
A. Kranzl (*)
Laboratory for Gait and Human Motion Analysis, Orthopedic Hospital Speising, Vienna, Austria
e-mail: andreas.kranzl@oss.at

DOI 10.1007/978-3-319-30808-1_24-1
2 A. Kranzl
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Observational Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Treadmill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Patient Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Structured Analysis of the Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Reliability of the Evaluation of Observational Video-Based Gait Analysis . . . . . . . . . . . . . . . . . . . . 10
Video Recording Log . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Clinical Use of 2D Motion Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Database Software and Capturing Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Conclusion/Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
Apart from pure observational gait analysis (optical), recording of gait with a video
camera is the most common type of analysis. Modern video cameras for the
consumer market are readily available and fairly inexpensive. Recording is mostly
performed directly in the camera or stored on a PC using suitable software. Advan-
tages of video recording are being able to replay the video imaging several times, the
pause option and slow motion option. Video recording is also easy to use. Video
recordings are often also used additionally to three-dimensional motion capture
systems, on the one hand, to visually document how the patient walks and, on the
other hand, to document movements that the implemented biomechanical model
does not image in the analysis. Size of the room, position of the camera, lighting in
the room, marking of the walking distance, and the recording log are all important
for producing high-quality video for gait evaluation.
State of the Art
Video recordings are used in the clinical movement analysis laboratory as well as in
other institutions such as medical offices and physiotherapy offices, to evaluate
movement processes (in particular in walking). The use of gait analysis as an
additional component in treatment plans is uncontested. Recordings are usually
carried out using two video cameras (sagittal and frontal planes). The advantage of
video recordings without additional measurement instruments is that the patient is
not burdened with additional material (markers, measurement equipment, etc.).
Video recordings are stored in a database and are therefore easy to find. This allows
simple comparisons to be carried out before and after therapy.
Clinical Gait Assessment by Video Observation and 2D Techniques 3
Fig. 1 Room for video recording Optimal use of a room 6 m wide and 10 m long is shown in Fig. 1.
The walking distance is color coded (gray) and cross lines help determine stride length
Observational Gait Analysis
Observation of gait analysis has an invaluable role to play in clinical routine.

However, human vision can only observe a frequency of 12–18 Hz. Due to this
low resolution, not all movement details can be recognized with the human eye.
A further problem is that gait disturbances not only occur in one plane of the joint,
but in numerous planes and numerous joints simultaneously. The recording of gait
with a video camera significantly simplifies analysis. With the possibility of
watching the video several times and at various speeds, a detailed analysis of
individual joints and planes is possible. A video analysis can be carried out quickly;
however, preparations are necessary in order to perform a high-quality video anal-
ysis. In order to obtain a high-quality recording, a suitable recording room is required
and certain technical requirements need to be fulfilled for the camera.
It is not always easy to find a room of suitable size. For recording the sagittal
plane, the room should be at least 10 m long to ensure sufficient room for walking,
enabling the patient to walk at their normal speed with enough steps recorded for
evaluation of the gait (Fig. 1). This walking distance is sufficient for patients with
mild gait disturbances; the distance can be shorter for patients with more severe gait
disturbances or for children, due to the shorter step length.
For better guidance of the patient, it is useful to mark the walking distance in
a color that contrasts from the rest of the floor (Fig. 1). The start and end of the
walking distance should be marked for the patient. Subtle lines or predefined
distances along the walking distance help determine step length in the subsequent
4 A. Kranzl
Fig. 2 Treadmill wide angle Due to the size of the room, a video camera cannot be positioned far
enough from the treadmill to be able to see the entire patient in the image (image within the red
section), whereas a wide angled lens allows imaging of the entire patient in the same position. The
camera position in the left image is oriented at exactly 90 to the axis of movement, the central
image is positioned at an angle of 30 to the patient, and the right image at 45 . The measured knee
angles are 30 , 28 , and 25 , although the recordings were performed simultaneously with three
cameras. These different angle measurements are caused by parallax error. It is necessary to be
aware of this error if the patient is not centered in the image
evaluation by video. The spatial depth of the room is important for the lateral
recording. The spatial depth of the room is defined by the number of double steps
as well as by the camera optics. In the sagittal recording, in order to perform an
adequate evaluation, the entire patient needs to be seen, from head to toe. The camera
should be positioned at a right angle to the walking distance. It is possible to have
the camera follow the patient; however, this can create difficulties in the measure-
ment of joint positions and distances. Parallax error can lead to one or more errors in
the determination of the joint angles. Therefore, care is to be taken to ensure that the
planes of movement are at a right angle to the camera.
For evaluation of joint positions, the view of the camera should be positioned
centrally on the patient and not at an angle. The evaluation of joint angles in the
border areas of the video image is only possible to a limited extent due to parallax
error.
A wide-angle lens can be of assistance in a room with insufficient depth (Fig. 2).
A wide-angle lens allows video recordings to be carried out in rooms with limited
room depth. It should be noted though that distortion can occur in the border areas of
the video. Depending on the quality of the wide-angle lens, these can be more or less
pronounced.
For frontal recordings, it is necessary to ensure that the patient is in the center of
the image. For all recordings, the patient should be imaged as large as possible. This
can be performed with the zoom function of the video camera. The video format 16:9
Fig. 3 Patient recorded with three cameras simultaneously
Fig. 4 Accompanying
camera
also has the advantage that the patient can be recorded in portrait layout. It is also
possible to record numerous gait sequences in the sagittal recording with the wider
video image.
In some centers, the side camera is mounted on a motorized or manually operated
track system (Fig. 4). This has the advantage that the camera can follow the patient at
a 90 angle and thereby record numerous gait sequences. Numerous gait sequences
can also be recorded by moving the camera, but these extra gait sequences are more
difficult to evaluate due to parallax error (Fig. 3).
A camera that accompanies the patient is shown in Fig. 4. This camera position
allows a right-angled view of the patient. The camera is mounted on a track system
and is controlled with a motion capture system.
It is useful to record the frontal image of the patient in full size, from the pelvis
downward and from the knees downward (focus ankle joint). The detailed view
allows a more precise evaluation of movements. This can be performed either via the
positioning of the camera on a height adjustable tripod or on a height adjustable wall
6 A. Kranzl
Fig. 5 Optimized view of the relevant segments
Fig. 6 Camera position for

frontal recordings With a
height-adjustable frontal
camera, an optimized position
can be achieved to be able to
image the relevant body
segments as large as possible.
From left to right: entire body,
pelvis downward, and foot
area (Fig. 5)
mount (manual or electric) (Fig. 6). In general, it is useful to attach the video camera
to a tripod, which allows optimal horizontal positioning. A fixed position on the wall
is even better, so that the camera position is the same for all recordings and does not
need to be adjusted or checked for each recording.
In the left of the image (Fig. 6), a manual height-adjustable camera can be seen,
which is focused on the walking distance of the instrumented 3D gait analysis. On
the right, a camera can be seen which is automatically height adjusted with the PC
and focused on the walking distance for the video recording.
Apart from the room measurements, lighting conditions and the color of the floor
and walls are also important. There should be no interfering light from windows,
Fig. 7 Influence of illumination on the imaging quality
since this can lead to lighting problems. Furthermore, the background of the video
recording should have calm colors. Equipment that is standing around should also
not be visible in the video. The room lighting should be suitable for evening lighting
from above. The room should be evenly illuminated. The light source should not
flicker, as especially in high-resolution video recording, a flickering light can be very
disturbing. Conventional video cameras have an automatic feature that adjusts
exposure time and shutter, depending on available light. For high-quality recordings,
however, a manual adjustment of exposure time is useful. Exposure time should be
as short as possible without the video image being too dark (Fig. 7). Illumination of
4000 lx or more is useful. Additional lighting, which is appropriate for the camera,
8 A. Kranzl
can be helpful to improve recording conditions. It is necessary to ensure, however,

that no reflections on the skin are created with the additional lighting.
For a patient with moderate walking speed on a treadmill, exposure time and
additional lighting will vary. Top left: automatic setting of the camera without
additional lighting, top right: Automatic setting of the camera without additional
lighting, left center: exposure time 1/100 with additional lighting, right center:
exposure time 1/500 with additional lighting, bottom left: exposure time 1/1000
with additional lighting. It can be seen that the exposure time of 1/500 with
additional lighting provides the clearest image; a shorter exposure time creates an
image that is too dark.
Some centers also perform video recordings from above and/or from below in
order to better record transversal movement in the gait. The transversal view,
however, does not provide imaging of all body parts while walking. From below,
the feet block the pelvis; from above, the upper body is well recognized; but from the
pelvis downward, there is almost no visibility. In order to evaluate femur rotation,
marking the patella is useful. The foot angle is evaluated from behind in the frontal
video image.
The sagittal and frontal video recordings are performed either sequentially or
simultaneously. The advantage of a simultaneous video is that gait events can be
viewed in the frontal and sagittal video recordings at the same time. Videos can also
be compiled to a single video image (split screen) or videos can be recorded in
parallel, synchronized on a computer screen.
Most video cameras have a built-in storage option for recording the video. This
has the disadvantage that it can be difficult to find the recording of the video. It is
better to record directly to the computer via suitable recording software. This should
include a database function so that patient videos can be found quickly and without
difficulty. The connection of the video camera to the computer depends on the
requirements of the camera. Common connections to the computer are USB or
HDMI or the component connection. The FireWire/IEEE 1394 connection is seldom
found in video cameras today, if at all. If possible, when using two or more video
cameras, care should be taken that the hardware can be synchronized. If the
hardware cannot be synchronized, at least the software should allow subsequent
synchronization.
Technical requirements for a video camera:
• Lens size
• Manual adjustment of shutter and illumination
• Automatic and manual focus
• 16:9 image sensor ratio
• Optical zoom
• USB or HDMI or component connection
• Recording frequency of 50 Hz or more depending on speed of motion
The costs for a video camera range from € 200 to € 1500, depending on the
quality. A higher purchase price for a high-quality camera is quickly reimbursed by
the evaluation of the video recordings. High-quality cameras usually have a large
lens and sensors, which means better imaging quality with the same light conditions.
The recording frequency should be at least 25 images per second for gait evaluation.
For running analyses, the recording speed should be at least 100 Hz.
Most operating systems have simple software for video recording on the com-
puter. There is, however, usually a limitation in playback options. It is better to
purchase suitable software for video recordings. There are numerous manufacturers
that have developed such software. The requirements for software include the
possibility of recording one or more video sources simultaneously, as well as
a database function. The following points are important for playback: Real time,
slow motion, exact focusing of frozen images, as well as moving forward and
backward in the frozen images. To playback numerous videos for comparison of
various conditions and examination times is extrem helpful.
Treadmill
The use of a treadmill in combination with video cameras allows optimal positioning
of the camera as well as the ability to record a larger number of gait sequences in
a short time. With the exact positioning of the camera parallel to or at a right angle to
the axis of motion, the measurement of angles and distances is simplified. However,
not all patients are used to walking on a treadmill. Examinations show that healthy
patients require a period of 6 min for walking and running, whereas older patients
require at least 14 min (Wass et al. 2005). Even then there are differences in the
kinematics and kinetics of the gait (Alton et al. 1998; Wass et al. 2005). Therefore,
the use of a treadmill for recording gait is only sensible if the patient is used to
it. Usually, there is insufficient time and staff to provide the patient with enough time
for adaptation.
Patient Preparation
It is important for the quality of the recording that the patient is adequately adjusted.
For this, it is best if the patient walks in his/her undergarments during the recording.
Shorts and a t-shirt are also possible; however, it must be ensured that the pelvis can
be well observed and that the t-shirt does not cover the pelvis. Marking of the joint
points (ankle, knee, hip) as well as marking of the patella is helpful in video
recordings in order to be able to better evaluate and/or measure the joint angle
during walking. The markings should be performed during standing (not while lying
down ➔ due to skin movement). Otherwise, the marked spots may no longer
represent the desired skeletal reference points.
10 A. Kranzl
Structured Analysis of the Gait
It is important for a high-quality evaluation that every plane and every joint is
observed in a structured manner. This should ensure that gait disturbances or gait
patterns are also recognized, even if this is not the main focus of the pathology.
Numerous gait cycles should be observed: Experts tend to concentrate on certain
parameters in the gait (Toro et al. 2003). Prefabricated examination forms which
guide the examiner through the analysis are useful. One of the most well-known
examination forms is the form from Jaqueline Perry (1992). Values can be entered
for each gait phase and for each joint. Other structured examination forms go a step
further. In addition, a further evaluation system is introduced (Visual Gait Assess-
ment Scale, Edinburgh Visual Gait Scale, Observational Gait Scale, Physician’s
Rating Scale). This score system allows determination of the degree of a gait
disturbance and whether the gait is improved in the second analysis following the
therapeutic measures (Viehweger et al. 2010). In particular, in the therapeutic
environment, observational gait analysis is suitable for documenting therapeutic
advances (Coutts 1999). The score compilation describes the gait with one or
more values. However, it needs to be taken into account that a reduction of data
may no longer allow an exact description of the disturbance.
Reliability of the Evaluation of Observational Video-Based Gait

Analysis
With respect to study data for gait evaluation (validity and reliability) via observa-
tional gait analysis (including implementation of videos), different images are
presented dependent on the parameters of the analysis. Determination of the initial
floor contact, for example, shows good interobserver reliability (Mackey et al. 2003).
The reproducibility of the results depends on the experience of the reviewer.
Experienced persons tend to demonstrate a higher reproducibility of results. These
evaluations, however, remain subjective and mostly only show moderate reproduc-
ibility (Borel et al. 2011; Brunnekreef et al. 2005; Eastlack et al. 1991; Hillman et al.
2010; Krebs et al. 1985; Rathinam et al. 2014).
The recording of force is not possible with video recordings only, although force
plates can be included via video recording software products. Following calibration
of the position of the force plates with the video image, an overlay function is
provided which adds a force vector to the video image. In this way, force and
leverage can be visualized. This is especially helpful in the fine adjustment of
lower leg orthotics and prosthetics.
Video Recording Log
It is useful to be guided by a standardized recording log so that a basic recording is

present for each patient. This log (a and b) is carried out for each video recording at
Speising Orthopaedic Hospital in Vienna. These recordings are usually carried out
barefoot, if it is possible for the patient to perform the examination barefoot. In video
recordings with numerous auxiliary means (shoes, orthoses), it needs to be ensured
that the patient does not become tired during the recording. The gait speed should be
determined by the patient. Prior to the actual recording, the patient should have time
to become familiar with the laboratory surroundings and the requested performance.
If possible, the patient should walk the walking distance a few times prior to the
actual recording.
(a) While walking:

1. sagittal and frontal recording
2. entire body in the frontal recording
3. from the pelvis downward in the frontal recording
4. lower legs and feet in the frontal recording
The video image is zoomed in the frontal recording in order to continuously
have the entire person or the relevant body segments in the image (Fig. 5).
The recording during walking is performed at normal speed, and addition-
ally at a quicker speed and running. Stopping and turning of the patient is
also recorded in the frontal recording.
If the patient requires an aid or various aids, the recording is performed
with each of these aids individually. It is to be noted that recording with
additional aids also requires more recording and analysis time.
(b) While standing:
1. standing on one leg, both left and right
2. standing on toes
3. standing on heels
4. knee bending with both legs
A further recording possibility is the recording of standing up from a chair and

sitting down again.
The 2D analysis can be used relatively easily to get an overview of the gait
pattern. As already mentioned, 2D video analyses show only moderate reproduc-
ibility. If the motion occurs strictly perpendicular to the plane, we do not have any
parallax error. Data presented by Davis et al. (1991) at the International Symposium
on 3D Analysis of Human Movement in 1991 compared 62 normal subjects with
124 sides and 5 patients with 10 sides with 2D- and 3D-captured data. For normal
subjects, they found a good accordance for the hip (sagittal, mean relative % differ-
ence 1%1% and frontal, mean relative % difference 9%7%) and knee joint
angles (sagittal, mean relative % difference 4%2%). For the ankle joint angle, it
seems that the accordance is less good (sagittal, mean relative % difference 13%
5%). For the impaired gait pattern, these values increase in all joints. Hip joint angle
12 A. Kranzl
(sagittal, mean relative % difference 8%8% and frontal, mean relative % difference
28%17%), knee joint angles (sagittal, mean relative % difference 8%5%), and
ankle joint angle (sagittal, mean relative % difference 54%120%).
The authors conclude the following: “Moreover, the utilization of 2D gait anal-
ysis strategies in clinical settings where the pathology can result in significant
‘out-of-plane’ motion is not appropriate and ill-advised. For gait analysis, there is
no substitute for 3D motion analysis if we have ‘out-of-plane’ motion even though
the 2D method is simpler and less expensive, it may produce results which are
wrong.”
Also Clarke and Murphy (2014) were able to show for healthy participants that
for the sagittal knee joint motion, a revealed excellent agreement between 2D and 3D
measurement exists. This is supported by Nielsen and Daugaard (2008) and Fatone
and Stine (2015).
Clinical Use of 2D Motion Capture
Looking at the literature in which the observational gait analysis is used, it is shown
that this is used in a wide variety of diseases to analysis gait disorders. A large area of
use is found in the range of the assessment of musculoskeletal abnormalities or gait
disorders with cerebral palsy (Chan et al. 2014; Chantraine et al. 2016; Deltombe
et al. 2015; Esposito and Venuti 2008; Maathuis et al. 2005; Satila et al. 2008). In
addition to the quantification of gait disorders (Moseley et al. 2008), it is also used
for the examination of therapeutic programs and therapeutic devices (Taylor et al.
2014). In patients with amputation of the lower extremity, it is used for checking and
determining the function of prostheses. Through the standardized use of the video
analysis, an objective documentation of the advantages and disadvantages of pros-
thesis parts is achieved (Lura et al. 2015; Vrieling et al. 2007). In the area of
neurology, such as Parkinson’s patients (Guzik et al. 2017; Johnson et al. 2013;
Obembe et al. 2014) or traumatic brain injuries (Williams et al. 2009), the video
analysis serves as a tool to describe gait pattern. Button et al. (2008) used
two-dimensional gait analysis to analyze patients with anterior cruciate ligament
rupture after rehabilitation.
Database Software and Capturing Software
The use of a database for video recordings is useful since it is easier to find previous
recordings for gait comparison. Specialized software products for recording usually
include a database. Depending on the software product, the search function may not
be present or be only rudimentary, for searching for already recorded videos. If the
software allows a keyword for the video, this should be carried out. The keyword
makes finding the video with a special aid significantly easier.
There are many programs on the market that allow you to record your videos on
a PC and use a database function at the same time (e.g., Opensource KINOVEA or
Fig. 8 Knee joint angle

measurement at initial contact,
manual digitization
Tracker) or commercial products like TEMPLO (Contemplate), SIMI Motion

(SIMI), myDartfish (Dartfish), MyoMotion (Noraxon), and others. The requirements
for video recording software are relatively low. However, additional functions are
essential for the quality of the gait analysis itself. Besides the playback function in
real-time speed, the possibility of a slow motion should be available. In addition to
the exact selection of a still image, a frame-by-frame function should also be
available. For a more accurate analysis, it is helpful to measure distances and/or
joint angles (Fig. 8) from the video. This is supported by most analysis programs and
the results can be taken relatively easily into a report. When it comes to comparing a
gait with two conditions (e.g., walking barefoot and walking with the shoe, or
preoperative and postoperative), the software should allow you to play two videos
or more in parallel (Fig. 9). Thus, a direct comparison between the conditions is
possible. Another good option for comparing conditions is the overlay function;
here two videos are superimposed and thus changes in the gait are easier to
recognize. Newer systems also use a tracking function (Fig. 10). Thus, it is possible
to output joint angles over the full gait cycle. For this purpose, use of markers is
useful (white/black circular stickers).
Conclusion/Summary
The use of the observational gait analysis is an important part of clinical practice.
The use of video cameras for the documentation and evaluation of gait distur-
bances supports diagnosis and selection of treatment. Therefore, optimal recording
14 A. Kranzl
Fig. 9 Follow-up control, comparison of changes over years (screenshot TEMPLO software from
Contemplas)
Fig. 10 Automatic tracking option to calculate the knee joint angle sagittal (screenshot MyoVideo
software(MR3)) from Noraxon (Screen capture, myoVIDEO™ software, © Noraxon USA.
Reprinted with permission)
conditions as well as structured processes in the analysis of the videos are important.
In addition, optimal room size as well as optimum adjustment of equipment is a key
component in obtaining high-quality video recordings.
Cross-References
▶ Activity Monitoring in Orthopaedic Patients

▶ Ankle Foot Orthoses and Their Influence on Gait
▶ Assessing Club Foot and Cerebral Palsy by Pedobarography
▶ Assessing Pediatric Foot Deformities by Pedobarography
▶ Gait Scores – Interpretations and Limitations
▶ Motion Analysis Through Video – How Does Dance Changes with the Visual
Feedback
▶ The Use of Low Resolution Pedoboragraphs
▶ Upper Extremity Activities of Daily Living
References
Alton F, Baldey L, Caplan S, Morrissey MC (1998) A kinematic comparison of overground and
treadmill walking. Clin Biomech 13:434–440
Borel S, Schneider P, Newman CJ (2011) Video analysis software increases the interrater reliability
of video gait assessments in children with cerebral palsy. Gait Posture 33:727–729
Brunnekreef JJ, van Uden CJ, van Moorsel S, Kooloos JG (2005) Reliability of videotaped
observational gait analysis in patients with orthopedic impairments. BMC Musculoskelet
Disord 6:17
Button K, van Deursen R, Price P (2008) Recovery in functional non-copers following anterior
cruciate ligament rupture as detected by gait kinematics. Phys Ther Sport 9:97–104
Chan MO, Sen ES, Hardy E, Hensman P, Wraith E, Jones S, Rapley T, Foster HE (2014)
Assessment of musculoskeletal abnormalities in children with mucopolysaccharidoses using
pGALS. Pediatr Rheumatol Online J 12:32
Chantraine F, Filipetti P, Schreiber C, Remacle A, Kolanowski E, Moissenet F (2016) Proposition of
a classification of adult patients with hemiparesis in chronic phase. PLoS One 11:e0156726
Clarke L, Murphy A (2014) Validation of a novel 2D motion analysis system to the gold standard
in 3D motion analysis for calculation of sagittal plane kinematics. Gait Posture 39(Suppl 1):
S44–S45
Coutts F (1999) Gait analysis in the therapeutic environment. Man Ther 4:2–10
Davis R, Ounpuu S, Tyburski D & Deluca P (1991) A comparison of two dimensional and three
dimensional techniques for the determination of joint rotation angles. Proceedings of interna-
tional symposium on 3D analysis of human movement. p 67–70.
Deltombe T, Bleyenheuft C, Gustin T (2015) Comparison between tibial nerve block with anaes-
thetics and neurotomy in hemiplegic adults with spastic equinovarus foot. Ann Phys Rehabil
Med 58:54–59
Eastlack ME, Arvidson J, Snyder-Mackler L, Danoff JV, McGarvey CL (1991) Interrater reliability
of videotaped observational gait-analysis assessments. Phys Ther 71:465–472
Esposito G, Venuti P (2008) Analysis of toddlers’ gait after six months of independent walking to
identify autism: a preliminary study. Percept Mot Skills 106:259–269
Fatone S, Stine R (2015) Capturing quality clinical videos for two-dimensional motion analysis.
J Prosthet Orthot 27:27–32
Guzik A, Druzbicki M, Przysada G, Kwolek A, Brzozowska-Magon A, Wolan-Nieroda A (2017)
Analysis of consistency between temporospatial gait parameters and gait assessment with the
use of Wisconsin gait scale in post-stroke patients. Neurol Neurochir Pol 51:60–65
Hillman SJ, Donald SC, Herman J, McCurrach E, McGarry A, Richardson AM, Robb JE (2010)
Repeatability of a new observational gait score for unilateral lower limb amputees. Gait Posture
32:39–45
16 A. Kranzl
Johnson L, Burridge JH, Demain SH (2013) Internal and external focus of attention during gait
re-education: an observational study of physical therapist practice in stroke rehabilitation.
Phys Ther 93:957–966
Krebs DE, Edelstein JE, Fishman S (1985) Reliability of observational kinematic gait analysis.
Phys Ther 65:1027–1033
Lura DJ, Wernke MM, Carey SL, Kahle JT, Miro RM, Highsmith MJ (2015) Differences in knee
flexion between the Genium and C-Leg microprocessor knees while walking on level ground
and ramps. Clin Biomech (Bristol, Avon) 30:175–181
Maathuis KG, van der Schans CP, van Iperen A, Rietman HS, Geertzen JH (2005) Gait in children
with cerebral palsy: observer reliability of physician rating scale and Edinburgh visual gait
analysis interval testing scale. J Pediatr Orthop 25:268–272
Mackey AH, Lobb GL, Walt SE, Stott NS (2003) Reliability and validity of the observational gait
scale in children with spastic diplegia. Dev Med Child Neurol 45:4–11
Moseley AM, Descatoire A, Adams RD (2008) Observation of high and low passive ankle
flexibility in stair descent. Percept Mot Skills 106:328–340
Nielsen D, Daugaard M (2008) Comparison of angular measurements by 2D and 3D gait analysis.
Dissertation, Jonkoping University.
Obembe AO, Olaogun MO, Adedoyin R (2014) Gait and balance performance of stroke survivors
in south-western Nigeria – a cross-sectional study. Pan Afr Med J 17(Suppl 1):6
Perry J (1992) Gait analysis, normal and pathological function. SLACK, Thorofare
Rathinam C, Bateman A, Peirson J, Skinner J (2014) Observational gait assessment tools in
paediatrics – a systematic review. Gait Posture 40:279–285
Satila H, Pietikainen T, Iisalo T, Lehtonen-Raty P, Salo M, Haataja R, Koivikko M, Autti-Ramo I
(2008) Botulinum toxin type A injections into the calf muscles for treatment of spastic equinus
in cerebral palsy: a randomized trial comparing single and multiple injection sites. Am J Phys
Med Rehabil 87:386–394
Taylor P, Barrett C, Mann G, Wareham W, Swain I (2014) A feasibility study to investigate the
effect of functional electrical stimulation and physiotherapy exercise on the quality of gait of
people with multiple sclerosis. Neuromodulation 17:75–84. Discussion 84
Toro B, Nester CJ, Farren PC (2003) The status of gait assessment among physiotherapists in the
United Kingdom. Arch Phys Med Rehabil 84:1878–1884
Viehweger E, Zurcher Pfund L, Helix M, Rohon MA, Jacquemier M, Scavarda D, Jouve JL,
Bollini G, Loundou A, Simeoni MC (2010) Influence of clinical and gait analysis experience on
reliability of observational gait analysis (Edinburgh gait score reliability). Ann Phys Rehabil
Med 53:535–546
Vrieling AH, van Keeken HG, Schoppen T, Otten E, Halbertsma JP, Hof AL, Postema K (2007)
Obstacle crossing in lower limb amputees. Gait Posture 26:587–594
Wass E, Taylor NF, Matsas A (2005) Familiarisation to treadmill walking in unimpaired older
people. Gait Posture 21:72–79
Williams G, Morris ME, Schache A, Mccrory P (2009) Observational gait analysis in traumatic
brain injury: accuracy of clinical judgment. Gait Posture 29:454–459
The Conventional Gait Model - Success and
Limitations
Richard Baker, Fabien Leboeuf, Julie Reay, and Morgan Sangeux
Abstract
The Conventional Gait Model (CGM) is a generic name for a family of closely
related and very widely used biomechanical models for gait analysis. After
describing its history, the core attributes of the model are described followed by
evaluation of its strengths and weaknesses. An analysis of the current and future
requirements for practical biomechanical models for clinical and other gait
analysis purposes which have been rigorously calibrated suggests that the CGM
is better suited for this purpose than any other currently available model. Mod-
ifications are required, however, and a number are proposed.
Keywords
Clinical Gait Analysis • Biomechanical Modeling
R. Baker (*)
University of Salford, Salford, UK
e-mail: R.J.Baker@salford.ac.uk
F. Leboeuf • J. Reay
School of Health Sciences, University of Salford, Salford, UK
e-mail: fabien.leboeuf@gmail.com; J.Reay@salford.ac.uk
M. Sangeux
Hugh Williamson Gait Analysis Laboratory, The Royal Children’s Hospital, Parkville/Melbourne,
VIC, Australia
Gait laboratory and Orthopaedics, The Murdoch Childrens Research Institute, Parkville/Melbourne,
VIC, Australia
e-mail: morgan.sangeux@gmail.com

DOI 10.1007/978-3-319-30808-1_25-2
2 R. Baker et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Model Structure and Anatomical Segment Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Marker Placement to Estimate Anatomical Segment Position . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Kinematic Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Kinetic Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Weaknesses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Alternatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Introduction
The Conventional Gait Model (CGM) is a generic name for a family of biomechan-
ical models which emerged in the 1980s based on very similar principles and giving
very similar results. It has a rather complex history (outlined below) and as a
consequence has been referred to by a range of different names. The use of the
name Conventional Gait Model is an attempt to emphasize the essential similarity of
these models despite those different names.
For a number of reasons, the CGM became the de facto standard for gait analysis
in the 1990s, particularly in clinical and clinical research applications. Despite
considerable strengths, technological advances have left aspects of the CGM looking
quite outdated. The model, as originally formulated, also has a number of intrinsic
limitations and, as these have become more widely appreciated, a variety of mod-
ifications and alternatives have been developed. Although the model can no longer
be regarded as an industry-wide standard as was once the case, many of the more
established and respected clinical centers still prefer to use the model considering its
strengths to outweigh its limitations.
After a brief summary of the historical development of the CGM, this chapter will
describe its characteristics and then assess its strengths and limitations, concluding
with some suggestions as to how the model could be developed in future in order to
address those limitations while preserving its strengths.
History
(Italicized words in this section are names that are sometimes used to refer to the
CGM.)
The origins of the model can be traced to the work of John Hagy in the laboratory
established by David Sutherland (Sutherland and Hagy 1972) who digitized the
The Conventional Gait Model - Success and Limitations 3
positions of skin markings indicating anatomical landmarks from bi-planar movie

stills. The coordinates were then used to compute a number of joint angles. Patrick
Shoemaker extended this approach (Shoemaker 1978) to incorporate Ed Chao’s
ideas on representing three-dimensional joint motion as Euler angles (Chao 1980).
Jim Gage on a visit to San Diego prior to developing his own gait analysis laboratory
at the Newington Hospital in Connecticut and a succession of engineers including
Scott Tashman, Dennis Tyburski, and Roy Davis(Davis et al. 1991) further devel-
oped the ideas in a number of ways. Perhaps the most important of these were the
calculation of joint angles on the basis of estimated joint centers (rather than directly
from marker locations) and the incorporation of three-dimensional inverse dynamics
to estimate joint moments (Ounpuu et al. 1991) based on the approach of David
Winter (Winter and Robertson 1978). At about this time Murali Kadaba developed a
very similar model at the Helen Hayes Hospital (Kadaba et al. 1990; Kadaba et al.
1989). There was communication between the two groups over this period, but there
are now different memories as to the extent of this collaboration and the precise role
of the different individuals involved.
Although some minor modifications have been proposed since, the subsequent
history is largely about how the model was distributed. The Helen Hayes Model was
developed as a package and distributed across seven American hospitals. A little
later, Oxford Metrics (now Vicon), the manufacturers of Vicon movement analysis
systems, chose to develop their own version of the model (with support from
individuals at both Newington and Helen Hayes). This was embedded within a
package known as the Vicon Clinical Manager (VCM) and later developed as the
Plug-in Gait (PiG) model for Workstation software. Most manufacturers of gait
analysis systems produce some version of the model which goes under a variety of
names. Perhaps because of commercial sensitivities it is generally rather unclear
what level of agreement there is between data processed with these alternative
models.
Perhaps the most important factor leading to the widespread adoption of the CGM
was the prominence of Vicon measurement systems in clinical and academic gait
analysis at this time with VCM and PiG being delivered alongside with their
hardware. Many of the more established clinical services were founded at this time
and most adopted VCM and continued to use PiG. Jim Gage became a strong
advocate for clinical gait analysis and with Roy Davis and Silvia Ounpuu established
extremely well-regarded teaching courses first at Newington, then Gillette Chil-
dren’s Hospital which were based on what they regarded as the Newington Model.
The model was also explained and validated in a number of key papers (Kadaba et al.
1989, 1990; Davis et al. 1991; Ounpuu et al. 1991,1996) in considerably more detail
than any other model at the time. Thus by the early 2000s, the CGM had become
established as the predominant gait model for clinical and clinical research purposes,
and a large community of users had developed embodying a solid understanding of
its strengths and limitations.
Since that time, this status has diminished somewhat. A larger number of
suppliers to the gait analysis market and the increasing ease of integrating different
software have widened the options for data processing. There have been
4 R. Baker et al.
considerable and often justified criticisms of the limitations of the CGM and a
general failure of the CGM community to develop the model to address these issues.
Despite this, the model is still almost certainly the most widely used and understood
single model within the clinical and clinical research community.
State of the Art
As stated above, the CGM is actually a family of closely related models but for
simplicity this section will be limited to a description of that embodied in the VCM
and PiG which are identical and the most commonly used versions. It is arguable
whether the CGM is a model at all as the word is now understood in biomechanics
and it was originally described as “an algorithm for computing lower extremity joint
motion” (Kadaba et al. 1990) and “a data collection and reduction technique” (Davis
et al. 1991) when first described. In the sections below, however, a modern under-
standing of biomechanical modeling will be used to describe the underlying
concepts.
Model Structure and Anatomical Segment Definitions
The model has seven segments linked in a chain by ball joints (three rotational
degrees of freedom) in the sequence left foot, left tibia, left femur, pelvis, right
femur, right tibia, right foot. An orthogonal coordinate system is associated with
each segment. While the three segment axes are mathematically equivalent, clinical
convention is to define the segment alignment in terms of the alignment of a primary
axis and the rotation about this as defined by some off-axis reference point. The
primary axes for each segment is taken to be that linking the joints which attach it to
the two neighboring segments in the kinematic chain. Conceptually the segment axis
systems are thus defined by specifying a primary axis and reference point for each.
These are defined in Table 1.
Marker Placement to Estimate Anatomical Segment Position
Markers are placed in such a way that the segment orientations can be estimated.
When the model was developed, optoelectronic measurements systems were limited
to resolving a small number of markers and thus the minimum number of markers
possible is used. This is based on the assumption that the proximal joint of any leg
segment (all those other than the pelvis) is known from the position and orientation
of the joint to which it is linked proximally. More distal segment orientations are
dependent on the orientation of the more proximal segments and the model is thus
often described as being hierarchical. Because of the difficulty in resolving more
than two markers on the foot at the time when the model was developed, it defined
Table 1 Anatomical segment definition for the CGM

Pelvis
The primary axis is the mediolateral axis running from one hip
joint center to the other. In most clinical applications, it is
assumed that the pelvis is symmetrical and that this axis is thus
parallel to the line running from one anterior superior iliac spine
(ASIS) to the other.
The reference point for rotation about this axis is the mid-point of
the posterior superior iliac spines (PSIS).
Femur
The primary axis is that running from the hip joint center to knee
joint center.
The reference point is the lateral epicondyle.
For validation purposes:
• The hip joint center will be taken as the geometrical center of a
sphere fitted to the articular surface of the femoral head.
• The knee joint center will be taken as the mid-point of the
medial and lateral epicondyles. These are often difficult to
palpate, however, and for some purposes the line between these
landmarks will be assumed to be parallel to that linking the most
posterior aspects of the femoral condyles.
Tibia
The primary axis is that running from the knee joint center to
ankle joint center.
The reference point is the lateral malleolus.
For validation purposes:
• The ankle joint center will be assumed to be the mid-point of the
medial and lateral epicondyles.
Foot
The primary axis is that running from the most posterior axis of
the calcaneus along the second ray and parallel to the plantar
surface of the foot.
Rotation about this axis is not defined.
the orientation of its primary axis but not any rotation about this. The locations of
markers are given in Table 2.
The hierarchical process requires a method for determining the location of the
joints within each segment. The hip joint location within the pelvis coordinate
system is specified by three equations (Davis et al. 1991) which are functions of
leg length and ASIS to ASIS distance. These are measured during physical exam-
ination (although ASIS to ASIS distance can also be calculated from the marker
positions during a static trial). The knee joint center in the femur coordinate system is
assumed to lie in the coronal plane at the point at which the lines from it to the hip
joint center and lateral femoral epicondyle are perpendicular and the distance
between joint center and epicondyle is half the measured knee width. The ankle
6 R. Baker et al.
Table 2 Marker placement for the CGM

Pelvis
Markers are placed over both ASIS and PSIS in order that they lie
in the plane containing the anatomical landmarks.
A set of equations are used to estimate the location of the hip joint
within the pelvic coordinate system.
Femur
The hip joint center within the femur is coincident with that within
the pelvis.
A marker is placed over the lateral femoral epicondyle and another
on a wand on the lateral thigh in such a way that the two markers
and the hip joint center lie within the coronal plane of the femur.
The knee joint center is to be defined such that it, the hip joint
center and the epicondyle marker form a right angle triangle within
the coronal plane of the femur with a base of half the measured
knee width.
Tibia
The knee joint center within the tibia is coincident with that within
the femur.
A marker is placed over the lateral malleolus and another on a
wand on the lateral leg in such a way that the two markers and the
knee joint center lie within the coronal plane of the tibia.
The ankle joint center is to be defined such that the knee joint
center and the malleolar marker form a right angle triangle within
the coronal plane of the tibia with a base of half the measured ankle
width.
Foot
The ankle joint center in the foot is defined to be coincident with
that with the tibia.
A marker is placed on the forefoot.
Another marker is placed on the posterior aspect of the heel for the
static trial such that the line between the two makers is parallel to
the long axis of the foot. The angles between this and the line from
the ankle joint center to the forefoot marker in the sagittal and
horizontal planes are calculated.
The heel marker is not used in walking trials but the offsets are
used to estimate the alignment of the long axis of the foot based on
the line between ankle joint center and forefoot marker.
joint center within the tibia is specified analogously with respect to the lateral
malleolus.
The wand markers (on both femur and tibia) are thus important to define the
segmental coronal plane. Use of the wand (rather than a surface mounted marker) has
two main purposes. The first is that wands (particularly those with a moveable ball
and socket joint at the base) can be adjusted easily to define the correct plane. At least
as important, however, is that by moving the marker away from the primary axis of
the segment they make definition of the coronal plane much less sensitive to marker
placement error or soft tissue artifact. Concerns have been expressed that the markers
wobble but there is little evidence of this in gait data (it would appear as fluctuation
in the hip rotation graph) if they are taped or strapped securely to the thigh.
The foot segment uses the ankle joint center (which has already been defined in
the tibia coordinate system) and one forefoot or toe marker. The placement of this
marker varies considerably with some centers placing quite distally (typically at the
level of the metatarsophalangeal joint) in which case it indicates overall foot
alignment. Other centers, particularly those dealing with clinical populations who
often have foot deformities, choose a more proximal placement (typically at the level
of the cuneiforms) in order to give a better indication of hind foot alignment.
Placement of a heel marker during the static trial also allows for offsets to ensure
that ankle measurements were aligned with the long axis of the foot rather than
simply by the line from the ankle joint center to the toe marker. A common variant is
to calculate the plantar flexion offset on the assumption that the foot is flat and thus
that the long axis of the foot is in the horizontal plane, during the static trial.
Kinematic Outputs
Kinematic outputs are mainly joint angles describing the orientation of the distal
segment with respect to that of the proximal segment. The orientation of the pelvis is
output as segment angles (with respect to the laboratory-based axis system) as is the
transverse plane alignment of the foot (called foot progression). In three dimensions,
the orientation of one segment with respect to another must be represented by three
numbers. The CGM uses Cardan angles which represent the set of sequential
rotations about three different and mutually perpendicular axes that would rotate
the distal segment from being aligned with the proximal segment (or the laboratory-
based coordinate system) to its actual position.
In the original model, the rotation sequence was about the medial-lateral, then the
anterior-posterior and finally the proximal-distal axis for all joints (and segments).
Although this sequence maps onto the conventional clinical understanding of the
angles for most joints, it does not for the pelvis (Baker 2001). This is because with
this rotation sequence, pelvic tilt is calculated as the rotation around the medial-
lateral axis of the laboratory coordinate system, rather than the medial-lateral axis of
the pelvis segment, as per conventional understanding. Baker (Baker 2001) proposed
to reverse the rotation sequence which results in pelvic angles that more closely map
onto the conventional clinical understanding of these terms (confirmed by Foti et al.
2001). Following Baker’s recommendation to use globographic angles (Baker
2011), these can be interpreted exactly as listed in Table 3.
While not formally a part of the model, the CGM is closely associated with a
particular format of gait graph (see Fig.1). All data is time normalized to one gait
cycle and the left side data plotted in one color (often red) and the right side data in
another (often green, but blue reduces the risk of confusion by those who are color
blind). The time of toe off is denoted by a vertical line across the full height of the
graph and opposite foot off and contact by tick marks at either the top or bottom of
the graphs (in the appropriate color). Normative data is often plotted as a grey band
8 R. Baker et al.
Table 3 Definition of joint angles as commonly used with the CGM

Pelvis (with respect to global coordinate system)
Internal/external rotation: rotation of the mediolateral axis about the vertical axis
Obliquity (up/down): rotation of the mediolateral axis out of the horizontal plane
Anterior/posterior tilt: rotation around the mediolateral axis
Hip (femur with respect to pelvis coordinate system)
Flexion/extension: rotation of the proximal-distal axis about the medio-lateral axis
Ad/abduction: rotation of the proximal-distal axis out of the sagittal plane
Internal/external rotation: rotation around the proximal-distal axis
Knee (tibia with respect to femur coordinate system)
Flexion/extension: rotation of the proximal distal axis about the medio-lateral axis
Ad/abduction: rotation of the proximal-distal axis out of the sagittal plane
Internal/external rotation: rotation around the proximal-distal axis
Ankle (foot with respect to tibia coordinate system)
Dorsiflexion/plantarflexion: rotation of the proximal distal axis about the medio-lateral axis
Internal/external rotation: rotation of the proximal-distal axis out of the sagittal plane
Foot (with respect to global coordinate system)
Foot progression (in/out): rotation of the proximal-distal axis out of the “sagittal” plane
in the background (typical one standard deviation about the mean). The graphs are
then commonly displayed as arrays with the columns representing the different
anatomical planes and the rows representing the different joints.
Kinetic Outputs
The CGM is commonly used to calculate kinetic as well as kinematic outputs (Davis
et al. 1991; Kadaba et al. 1989). Both the Newington and Helen Hayes approaches
used inverse dynamics to estimate joint moments from force plate measurements of
the ground reaction, an estimate of segment accelerations from kinematic data and
estimates of segment inertial parameters. The main difference was that the Newing-
ton group took segment inertial parameters from the work of Dempster (1955)
whereas the Helen Hayes group (Kadaba et al. 1989) took them from Hindrichs
(1985) based on Clauser et al. (1969). Joint moments are fairly insensitive to these
parameters (Rao et al. 2006; Pearsall and Costigan 1999), and it is unlikely that this
would have led to noticeable differences in output. VCM and PiG used values from
Dempster (1955). Joint moments were presented in the segment coordinate systems.
The early papers do not specify whether the proximal or distal segment was used for
this. PiG and VCM allowed the user to select which (or to use the global coordinate
system) and the default setting of the distal system is probably most widely used.
Joint power is also calculated as the vector dot product of the joint moment and
angular velocity (note that this should be the true angular velocity vector and not that
of the time derivatives of the Cardan angles). Power is a scalar quantity and there is
thus no biomechanical justification for presenting “components” of power.
75
60
Knee flexion (degrees)
45
30
15
0
Left
Right
-15
0% 20% 40% 60% 80% 100%
% gait cycle
Fig.1 A standard gait graph. The curves represent how a single gait variable varies over the gait
cycle. The vertical lines across the full height of the graph represents foot-off and the tick marks
represent opposite foot off (to the left of graph) and opposite foot contact (to the right). Line in red is
for the left side and in blue is for the left side. The grey areas represent the range of variability in
some reference population as 1 standard deviation either side of the mean value
Variants
Over the years a number of variants to the CGM have been implemented by
particular groups. Most of these have not been formally described in the academic
literature.
• The original papers describing the model assumed that the femur and tibia wand
markers could be placed accurately. Early experience was that this was challeng-
ing and an alternative technique was developed in which the markers were only
positioned approximately and a Knee Alignment Device (KAD) was used during
static trials to indicate the orientation of the knee flexion axis and hence the
coronal plane of the femur. This allowed rotational offsets to be calculated to
correct for any misalignment of the wand markers (with the tibial offset requiring
an estimate of tibial torsion from the physical examination).
• A development within PiG allowed a medial malleolar marker to indicate the
position of the transmalleolar axis during the static trial and hence to calculate a
value of tibial torsion rather than requiring this to be measured separately.
10 R. Baker et al.
• A method of allowing for the thickness of soft tissues over the ASIS was provided
by allowing the measurement of the ASIS to greater trochanter distance which is
an estimate of the distance by which the hip joint center was posterior to the base
plate of the ASIS marker.
• A technique called DynaKAD has been proposed (Baker et al. 1999) to define the
thigh rotation offset by minimizing the varus-valgus movement during the walk-
ing trial. Other techniques have been used suggested to define this from functional
calibration trials (Schwartz and Rozumalski2005; Sauret et al. 2016; Passmore
and Sangeux2016).
• VCM and PiG introduced an angular offset along the tibia such that knee rotation
is defined as being zero during a static trial when the KAD is used and the
orientation of the ankle joint axis is defined by a measurement of tibial torsion
made during the physical exam (rather than the tibial wand marker).
• Another development of PiG allowed the heel marker to be used to give an
indication of inversion/eversion of the foot (rotation about the long axis) if it
was left in place during the walking trial.
• A further development allowed an angular offset to be applied allowing for the
foot being pitched forward by a known amount during a static trial (to take
account of the pitch of a shoe, for example).
• An upper body model was developed by Vicon which, though widely used, has
never been rigorously validated.
Strengths
Recent opinion has tended to emphasize the weaknesses of the CGM, but it is also
important to acknowledge its many strengths. In a world in which clinical gover-
nance is increasingly important, the CGM has been more extensively validated than
any other model in routine clinical use. The early papers of Kadaba et al. were
considerably ahead of their time in their approach to validation. The basic descrip-
tion of the model (Kadaba et al. 1990) includes presentation of normative data, a
comparison of this against normative data from a range of previous papers and a
sensitivity analysis of the most common measurement artifact arising from the
difficulty in placing thigh wands accurately. The follow up paper (Kadaba et al.
1989, which was actually published first!) is also a definitive repeatability study.
15 out of the 23 papers identified in the classic systematic review of repeatability
studies of kinematic models of McGinley et al. (McGinley et al. 2009) used a variant
of the CGM, and a more recent study (Pinzone et al. 2014) has demonstrated the
essential similarity of normative kinematic data collection from gait analysis services
on different sides of the world but captured by the CGM. This body of formal
validation literature is strongly reinforced by a large number of papers reporting use
of the CGM in a very wide range of clinical and research applications. The CGM is
thus particularly appropriate as a standardized and validated model for users who are
more interested in interpreting what the results mean than in further model devel-
opment and validation.
Although the implementation of the model is not trivial, the basic concepts are
about as simple as possible for a clinically useful model. It uses a minimal marker set
which can be applied efficiently in routine clinical practice. The model is determin-
istic (does not require any optimized fitting process) and thus the effects of marker
misplacement and or soft tissue artifact are entirely predictable (Table 4 illustrates
the effect that a given movement in each marker will affect outputs). It is thus
possible to develop a comprehensive understanding of how the model behaves
without being an expert in biomechanics. This can be logically extended to give
clear indications of how marker placement can be best adapted in order to obtain
clinically meaningful outputs in the presence of bone and joint deformities or
devices such as orthoses and prostheses.
It is unfortunate, therefore, that in the early years the model developed a reputa-
tion for behaving as a “black box.” This probably arose because the most commonly
available implementation, in the VCM, incorporated some refinements to the previ-
ously published versions (e.g., the thigh and shank offsets) which were only
described conceptually in the accompanying product documentation. Many people
assumed that there was insufficient information to fully understand the model; an
assumption proved false by a number of exact clones emerging (Baker et al. 1999 is
an example).
Weaknesses
Accuracy While the CGM has been subjected to several studies to investigate its
repeatability, there have been very few studies of its accuracy and those have focused
on very specific issues such as the location of the hip joint center ( Sangeux et al.
2011,2014; Peters et al. 2012) and orientation of the knee flexion axis (Sauret et al.
2016; Passmore and Sangeux 2016) in standing. The model is intended to track the
movements of the bones and there have been no studies performed to establish how
accurately it can do this. This is principally because gold standard methods for
tracking bone movement during walking are challenging (although a range of
techniques are available – see section on “Future Directions” below). It should be
emphasized, however, that this is a weakness of all commonly used biomechanical
models for gait analysis and not just the CGM.
Hip Joint Center Position A considerable body of knowledge now suggests that
there are better methods for specifying the location of the hip joint center within the
pelvic coordinate system than those used within the CGM (Leardini et al. 1999;
Sangeux et al. 2011, 2014; Harrington et al. 2007; Peters et al. 2012). While the first
of these (Leardini et al. 1999) suggested that functional calibration methods were
superior to equations, more recent studies suggest that alternative equations can give
results at least as good as functional methods in healthy adults (Sangeux et al. 2011,
2014; Harrington et al. 2007) and better in children with disabilities (Peters et al.
2012).
12
Table 4 Effects of moving a marker 5 mm in the specified direction on the outputs of the CGM. Note that because of the hierarchical basis of the
model, movements can only affect segments on or below that to which a given marker is attached. Changes in angle of less than 0.1 are left blank
Pelvis Hip Knee Ankle Foot
Marker Internal Internal Internal Internal
moved Tilt Obliquity Rotation Flexion Adduction rotation Flexion Varus rotation Dorsiflexion rotation progression
RASI up -0.9 1.4 0.1 -1.2 1.8 -0.5 -0.2 -0.4
RASI out 0.4 0.2 -0.1 -0.3 -0.2 -0.1
SACR up 1.8 2 0.2 0.1 -0.3 -0.1 -0.1
SACR
out
RTHI up
RKAD -0.5 -0.1 2.8 -0.9 1 -0.1 -0.1 -2.7
int
RKNE up -0.2 -0.1 0.4 -0.2 -0.4 0
RKNE 1.3 0.1 -1.8 2.2 -0.9 0.8 1.9 0.1
ant
RTIB up1
RTIB ant1
RANK 0.1 0.1 0.1
up
RANK -0.9 -0.1 -1 -1.1 0.1
ant
RTOE out 0.1 -4.6 4.7
RTOE ant
Notes:
a
Data is unaffected by the location of the tibial wand marker as a KAD was used for the static trial
b
Moving the toe marker anteriorly or posteriorly has no effect on outputs as a “foot flat” option was used for the static trial
R. Baker et al.
Defining the Coronal Plane of the Femur The first of the papers of Kadaba et al.
(1990) highlighted the sensitivity of the CGM to misplacement of the thigh markers
leading to erroneous definition of the coronal plane of the femur. This leads to a well-
known artifact in which the coronal plane knee kinematics show cross-talk from
knee flexion-extension which is generally of little clinical significance but highlights
uncertainty in hip rotation which is a major limitation of the model. Use of the KAD
(which is very poorly documented in the literature) led to some improvements but
this is still generally regarded as one of the most significant limitations of the model.
Over-Simplistic Foot Modeling Modeling the foot as single axis rather than three-
dimensional segment arose from the difficulty early models had in detecting more
than one marker placed on a small foot. While reliable detection of many markers on
the foot has been possible for many years now, a formal extension of the model has
never been proposed to model the foot more comprehensively. The Oxford Foot
Model (Carson et al. 2001), which is probably now the most widely used in clinical
and research practice, differs markedly from the CGM in that it allows translations
between the forefoot, hind foot, and tibia (rather than the spherical joints that are a
characteristic of the CGM).
Unconstrained Segment Dimensions The CGM does not require the segments to
be of a fixed length and soft-tissue artifact generally acts in such a way that the
distance between the hip and knee joint centers can vary by as much as 2 cm over the
gait cycle during walking. While this probably has a small effect on kinematic and
kinetic outputs, it does prevent the use of the model with more advanced modeling
techniques such as muscle length modeling and forward dynamics for which a rigid
linked segment model is required. Modern inverse kinematic techniques (Lu and
O’Connor 1999) which depend on rigid linked segment models also offer the
potential to incorporate modeling of soft-tissue artifact (Leardini et al. 2005) based
on data such as fluoroscopy studies (Tsai et al. 2009; Akbarshahi et al. 2010) in a
manner that is not possible within the CGM.
Inadequate Compensation for Soft Tissues over Pelvic Landmarks While

methods have been proposed for measuring and taking into account the soft tissues
over pelvic landmarks, none are particularly convincing or validated. As
populations, particularly those with limited walking abilities, become increasingly
overweight, this becomes a more important problem.
Poorly Validated Upper Body Model While Davis et al. (Davis et al. 1991) did
suggest placement of markers on the shoulders to give an indication of trunk
alignment, this has not been widely implemented. Vicon developed an upper body
model for PiG but, despite this being quite widely used, there have been no
published validations of its outputs. It is still not clear how important upper limb
movements are in relation to clinical gait analysis, but knowledge of trunk alignment
and dynamics is clearly important to understand the mechanics of the gait patterns of
many people with a range of conditions.
14 R. Baker et al.
Alternatives
Perhaps the most commonly used alternatives to the CGM are 6 degree of freedom
(6DoF) models. These can be traced back to the work of Cappozzo et al. (1995) and
have been popularized through Visual 3D software (C-motion, Kingston, Canada).
They track the segments independently (without constraining the joints) and can be
based on skin mounted markers (as implied by the illustration in the original paper)
or rigid marker clusters (as is more common nowadays). Perhaps the most important
limitation of this approach is that it refers to a modeling technique rather than any
specific model (CAST is an abbreviation for calibrated anatomical landmark tech-
nique) and no specific model has been widely used and rigorously validated. The
Cleveland Clinic Marker Set was an early example which achieved popularity when
it was implemented in the Orthotrack Software (Motion Analysis Corporation, Santa
Rosa, USA) but has never been validated (or even fully described) in the peer-review
literature. More recently Leardini et al. (2007) published and validated the IOR
model but there are only limited reports of use outside Bologna in the literature (and
it is worth noting that the IOR model, in using skin mounted markers, differs quite
markedly from most contemporary 6DoF modeling which uses rigid clusters).
6DoF models are sometimes presented as addressing the known limitations of the
CGM. Sometimes there is justification in these claims (e.g., the segments are fixed
length) but often corresponding issues are overlooked (e.g., nonphysiological trans-
lations between the proximal and distal bones at some joints). Soft tissue artifact
between markers is certainly eliminated by using rigid clusters but a different form of
soft tissue artifact will affect the orientation and position of the whole cluster in
relation to the bones (Barre et al. 2013). Other issues such as the difficulty in
estimating the hip joint center or knee axis alignment affects all models. One
advantage of most 6DoF models is that they use medial and lateral epicondyle
markers during a static trial to define the knee joint axis. This may be more
repeatable than precise alignment of thigh wands or KADs. It is also worth noting
that this is only a difference of knee calibration technique which could easily be
incorporated into the CGM.
Inverse kinematic (often referred to as kinematic fitting or global optimization)
models have also been reported (Lu and O’Connor 1999; Reinbolt et al. 2007;
Charlton et al. 2004), and this approach has become more popular since it was
incorporated within OpenSim (Seth et al. 2011) as the default technique for tracking
marker data. In this, a linked segment rigid body model is defined and an optimiza-
tion technique is used to fit the model to the measured marker positions, generally
using some weighted least-squares fit cost function. As with 6DoF models, this
approach has advantages and disadvantages with respect to the CGM. It is also
similar to the 6DoF approach in that no single model has received widespread use or
been subject to rigorous validation. The approach is inherently compatible with
advanced modeling techniques (e.g., muscle length modeling and forward dynam-
ics) and is well suited to either stochastic or predictive approaches to modeling soft
tissue artifact. Its most notable weakness is that it is nondeterministic and on
occasions artifacts can arise in the data from soft-tissue artifact, marker
misplacement, or erroneous model definition that can be extremely difficult to

source. On balance, however, it is likely that future developments will be based on
an inverse kinematic approach.
Future Directions
Over the lifetime of the CGM, the nature of gait analysis has changed considerably in
at least two important ways. The first is the growing importance of clinical gover-
nance (Scally and Donaldson 1998) and evidence-based practice within healthcare
organizations. This requires increasing standardization of all operations based upon
well-validated procedures. The emergence of accreditation schemes such as those
now operated by the Clinical Movement Analysis Society (CMAS, UK, and Ireland)
or the Committee for Motion Laboratory Accreditation (USA) is a consequence of
this. At present the focus is on whether written protocols exist at all, but it is
inevitable, as this minimal standard becomes universally implemented, that more
attention will be paid to ensuring that any procedures are appropriately validated.
This may be reinforced by more rigorous implementation of medical device legis-
lation to gait analysis software which should require manufacturers to ensure that
clinically relevant outputs (such as joint angles from a specific biomechanical
model) are reproducible (rather than just the technical outputs such as marker
locations).
The other change, which has implications beyond gait analysis for purely clinical
purposes, is that gait analysis systems are getting much cheaper and more user
friendly. It can no longer be assumed that laboratories will have a staff member
suitably qualified in biomechanics to create and adapt their own models. People
using current technology generally want to implement standardized techniques
allowing them to focus on the interpretation of data rather than on developing
individualized solutions and being distracted by the challenge of their validation.
Such users will require a model that is simple enough to be understood conceptually
in sufficient detail to guide quality assurance and interpretation of the data produced.
In scientific research, it would also be useful to have a widely accepted standardized
approach to capturing data to ensure that results from different centers are as
comparable as possible.
For clinical users and those in other fields who want to focus on the interpretation
of data rather than the mechanics of data capture, therefore, there is a real need for a
widely accepted, standardized, and validated approach to data capture (including
biomechanical modeling) which is efficient and robust in application and sufficiently
simple to be understood by the users themselves (rather than relying on biomechan-
ical experts). To be useful in this context, it needs to be widely applicable to all
people who are old enough to walk and who have a range of different health
conditions (or none). There needs to be a strong evidence base for the reproducibility
of measurements, specific training for staff involved in the capture and processing of
data, and appropriate metrics to assure the quality of measurements in routine
practice.
16 R. Baker et al.
The CGM satisfies all of these requirements at least, as well, and in most cases
considerably better than alternatives. Despite this, many users are frustrated by its
limitations while potential users are often put off by its commonly perceived
weaknesses (some justified, some not). It is clear that if the CGM is to have a future,
it will require modifications to address these.
A particular issue for the CGM is that many older laboratories have databases
stretching back over considerable periods of time (several decades in many cases)
and backward compatibility is perceived as extremely important. Ensuring rigorous
backwards is incompatible with improving the modeling of course, so a compromise
is required. The most obvious is to ensure that any new model uses the same
anatomical segment definitions (see Table 1) as the original. It may be that modifi-
cations lead to systematic differences with the original CGM, but it will be clear that
these are consequences of improvements in the modeling rather than redefinition of
what is being measured. It will also be important to quantify any such systematic
changes in order that they can be accounted for if data processed using different
versions of the model can be compared.
Another specific issue with the CGM is the perception of it as a “black box”
processing technique which cannot be properly understood. This has persisted
despite increasingly good documentation being produced but will be best addressed
by publishing the actual computer code through which the model is implemented.
Implementing the code in an open source language (such as Python) which is
available to all users will also be important. Training and education packages will
also be required for those less technically minded.
The specific modifications that are indicated would be:
• Adoption of a robust inverse kinematic fitting approach based around a linked

rigid segment model that is compatible with advanced musculoskeletal modeling
techniques.
• Replacement of wand markers with a limited number of skin mounted tracking
markers on the femur and tibia positioned to minimize sensitivity to soft tissue
artifact (Peters et al. 2009) or marker misplacement.
• Incorporation of more accurate equations for estimating the hip joint center and
techniques for accounting for the depth of soft tissues anatomical landmarks on
the pelvis.
• Improved methods for determining the orientation of the coronal plane of the
femur. Basing this upon the position of medial and lateral femoral epicondyle
markers during a calibration trial may be an improvement and functional calibra-
tion of the knee should be implemented as a quality assurance measure.
• Improvement of foot modeling by formalizing the PiG approach to using the heel
marker to give an indication of inversion and eversion of about the long axis of
the foot. There is a lack of standardization in where the forefoot (toe) marker is
placed. Opting for a more proximal placement (at about the level of the
tarsometatarsal joints) would lead to the foot representing movement of the
hind foot and open the possibility for some indication of forefoot alignment in
relation to this using markers placed on the metarsophalangeal joints.
• Validation of an appropriate trunk model should be regarded as essential. Doing

so on the basis of force plate measurements of center of mass displacement during
walking (Eames et al. 1999) would be useful to establish just how important
measuring upper limb movement is in gait analysis.
Future versions should be adequately validated in line with a modern understand-

ing of clinical best practice. At a minimum, this should include evidence of repro-
ducibility of results, but it would also be useful to have accuracy established with
reference to a variety of static and dynamic imaging techniques such as three-
dimensional ultrasound (Peters et al. 2010; Hicks and Richards 2005; Passmore
and Sangeux2016), low intensity biplanar x-rays (Pillet et al. 2014; Sangeux et al.
2014; Sauret et al. 2016), or fluoroscopy (Tsai et al. 2009; Akbarshahi et al. 2010).
There should also be publication of benchmark data with which services can
compare their own to ensure consistency (Pinzone et al. 2014) and streamlined
processed for conducting in-house repeatability studies would also be extremely
useful.
Cross-References

Sensors
▶ 3D Dynamic Probablistic Pose Estimation From Data Collected Using Cameras
and Reflective Markers
▶ 3D Kinematics of Human Motion
▶ Next Generation Models Using Optimized Joint Center Location
▶ Observing and revealing the hidden structure of the human form in motion
throughout the centuries
▶ Physics-based Models for Human Gait Analysis
▶ Rigid Body Models of the Musculoskeletal System
▶ Variations of Marker-sets and Models for Standard Gait Analysis
References
Akbarshahi M, Schache AG, Fernandez JW, Baker R, Banks S, Pandy MG (2010) Non-invasive
assessment of soft-tissue artifact and its effect on knee joint kinematics during functional
activity. J Biomech 43(7):1292–1301. doi:10.1016/j.jbiomech.2010.01.002
Baker R (2001) Pelvic angles: a mathematically rigorous definition which is consistent with a
conventional clinical understanding of the terms. Gait Posture 13(1):1–6. doi:10.1016/S0966-
6362(00)00083-7
Baker R (2011) Globographic visualisation of three dimensional joint angles. J Biomech 44
(10):1885–1891. doi:10.1016/j.jbiomech.2011.04.031
Baker R, Finney L, Orr J (1999) A new approach to determine the hip rotations profile from clinical
gait analysis data. Hum Mov Sci 18:655–667. doi:10.1016/S0167-9457(99)00027-5
18 R. Baker et al.
Barre A, Thiran JP, Jolles BM, Theumann N, Aminian K (2013) Soft tissue artifact assessment
during treadmill walking in subjects with total knee arthroplasty. IEEE Trans Biomed Eng 60
(11):3131–3140. doi:10.1109/TBME.2013.2268938
Cappozzo A, Catani F, Croce UD, Leardini A (1995) Position and orientation in space of bones
during movement: anatomical frame definition and determination. Clin Biomech 10
(4):171–178. doi:10.1016/0268-0033(95)91394-T
Carson MC, Harrington ME, Thompson N, O’Connor JJ, Theologis TN (2001) Kinematic analysis
of a multi-segment foot model for research and clinical applications: a repeatability analysis. J
Biomech 34(10):1299–1307. doi:10.1016/S0021-9290(01)00101-4
Chao EY (1980) Justification of triaxial goniometer for the measurement of joint rotation. J
Biomech 13:989–1006. doi:10.1016/0021-9290(80)90044-5
Charlton IW, Tate P, Smyth P, Roren L (2004) Repeatability of an optimised lower body model. Gait
Posture 20(2):213–221. doi:10.1016/j.gaitpost.2003.09.004
Clauser C, McConville J, Young J (1969) Weight volume and centre of mass of segments of the
human body (AMRL Technical Report). Wright-Patterson Air Force Base, Ohio
Davis RB, Ounpuu S, Tyburski D, Gage J (1991) A gait analysis data collection and reduction
technique. Hum Mov Sci 10:575–587. doi:10.1016/0167-9457(91)90046-Z
Dempster W (1955) Space requirements of the seated operator (WADC Technical Report :55–159).
Wright-Patterson Airforce Base, Ohio
Eames M, Cosgrove A, Baker R (1999) Comparing methods of estimating the total body centre of
mass in three-dimensions in normal and pathological gait. Hum Mov Sci 18:637–646.
doi:10.1016/S0167-9457(99)00022-6
Foti T, Davis RB, Davids JR, Farrell ME (2001) Assessment of methods to describe the angular
position of the pelvis during gait in children with hemiplegic cerebral palsy. Gait Posture 13:270
Harrington ME, Zavatsky AB, Lawson SE, Yuan Z, Theologis TN (2007) Prediction of the hip joint
centre in adults, children, and patients with cerebral palsy based on magnetic resonance
imaging. J Biomech 40(3):595–602. doi:10.1016/j.jbiomech.2006.02.003
Hicks JL, Richards JG (2005) Clinical applicability of using spherical fitting to find hip joint
centers. Gait Posture 22(2):138–145. doi:10.1016/j.gaitpost.2004.08.004
Hinrichs RN (1985) Regression equations to predict segmental moments of inertia from anthropo-
metric measurements: an extension of the data of Chandler et al. (1975). J Biomech 18
(8):621–624. doi:10.1016/0021-9290(85)90016-8
Kadaba MP, Ramakrishnan HK, Wootten ME, Gainey J, Gorton G, Cochran GV (1989) Repeat-
ability of kinematic, kinetic, and electromyographic data in normal adult gait. J Orthop Res 7
(6):849–860. doi:10.1002/jor.1100070611
Kadaba MP, Ramakrishnan HK, Wootten ME (1990) Measurement of lower extremity kinematics
during level walking. J Orthop Res 8(3):383–392. doi:10.1002/jor.1100080310
Leardini A, Cappozzo A, Catani F, Toksvig-Larsen S, Petitto A, Sforza V, Cassanelli G, Giannini S
(1999) Validation of a functional method for the estimation of hip joint centre location. J
Biomech 32(1):99–103. doi:10.1016/S0021-9290(98)00148-1
Leardini A, Chiari L, Della Croce U, Cappozzo A (2005) Human movement analysis using
stereophotogrammetry. Part 3. Soft tissue artifact assessment and compensation. Gait Posture
21(2):212–225. doi:10.1016/j.gaitpost.2004.05.002
Leardini A, Sawacha Z, Paolini G, Ingrosso S, Nativo R, Benedetti MG (2007) A new anatomically
based protocol for gait analysis in children. Gait Posture 26(4):560–571. doi:10.1016/j.
gaitpost.2006.12.018
Lu TW, O’Connor JJ (1999) Bone position estimation from skin marker co-ordinates using global
optimisation with joint constraints. J Biomech 32(2):129–134. doi:10.1016/S0021-9290(98)
00158-4
McGinley JL, Baker R, Wolfe R, Morris ME (2009) The reliability of three-dimensional kinematic
gait measurements: a systematic review. Gait Posture 29(3):360–369. doi:10.1016/j.
Ounpuu S, Gage J, Davis R (1991) Three-dimensional lower extremity joint kinetics in normal
pediatric gait. J Pediatr Orthop 11:341–349
Ounpuu O, Davis R, Deluca P (1996) Joint kinetics: methods, interpretation and treatment decision-
making in children with cerebral palsy and myelomeningocele. Gait Posture 4:62–78.
doi:10.1016/0966-6362(95)01044-0
Passmore E, Sangeux M (2016) Defining the medial-lateral axis of an anatomical femur coordinate
system using freehand 3D ultrasound imaging. Gait Posture 45:211–216. doi:10.1016/j.
Pearsall DJ, Costigan PA (1999) The effect of segment parameter error on gait analysis results. Gait
Posture 9(3):173–183
Peters A, Sangeux M, Morris ME, Baker R (2009) Determination of the optimal locations of
surface-mounted markers on the tibial segment. Gait Posture 29(1):42–48. doi:10.1016/j.
Peters A, Baker R, Sangeux M (2010) Validation of 3-D freehand ultrasound for the determination
of the hip joint centre. Gait Posture 31:530–532. doi:10.1016/j.gaitpost.2010.01.014
Peters A, Baker R, Morris ME, Sangeux M (2012) A comparison of hip joint centre localisation
techniques with 3-DUS for clinical gait analysis in children with cerebral palsy. Gait Posture 36
(2):282–286. doi:10.1016/j.gaitpost.2012.03.011
Pillet H, Sangeux M, Hausselle J, El Rachkidi R, Skalli W (2014) A reference method for the
evaluation of femoral head joint center location technique based on external markers. Gait
Pinzone O, Schwartz MH, Thomason P, Baker R (2014) The comparison of normative reference
data from different gait analysis services. Gait Posture 40(2):286–290. doi:10.1016/j.
Rao G, Amarantini D, Berton E, Favier D (2006) Influence of body segments’ parameters
estimation models on inverse dynamics solutions during gait. J Biomech 39(8):1531–1536.
doi:10.1016/j.jbiomech.2005.04.014
Reinbolt JA, Haftka RT, Chmielewski TL, Fregly BJ (2007) Are patient-specific joint and inertial
parameters necessary for accurate inverse dynamics analyses of gait? IEEE Trans Biomed Eng
54(5):782–793. doi:10.1109/TBME.2006.889187
Sangeux M, Peters A, Baker R (2011) Hip joint centre localization: Evaluation on normal subjects
in the context of gait analysis. Gait Posture 34(3):324–328. doi:10.1016/j.gaitpost.2011.05.019
Sangeux M, Pillet H, Skalli W (2014) Which method of hip joint centre localisation should be used
in gait analysis? Gait Posture 40(1):20–25. doi:10.1016/j.gaitpost.2014.01.024
Sauret C, Pillet H, Skalli W, Sangeux M (2016) On the use of knee functional calibration to
determine the medio-lateral axis of the femur in gait analysis: Comparison with EOS biplanar
radiographs as reference. Gait Posture 50:180–184. doi:10.1016/j.gaitpost.2016.09.008
Scally G, Donaldson L (1998) Clinical governance and the drive for quality improvement in the
new NHS in England. Br Med J 317:61–65. doi:10.1136/bmj.317.7150.61
Schwartz MH, Rozumalski A (2005) A new method for estimating joint parameters from motion
data. J Biomech 38(1):107–116. doi:10.1016/j.jbiomech.2004.03.009
Seth A, Sherman M, Reinbolt JA, Delp SL (2011) OpenSim: a musculoskeletal modeling and
simulation framework for in silico investigations and exchange. Procedia IUTAM 2:212–232.
doi:10.1016/j.piutam.2011.04.021
Shoemaker P (1978) Measurements of relative lower body segment positions in gait analysis.
University of California, San Diego
Sutherland D, Hagy J (1972) Measurement of gait movements from motion picture film. J Bone
Joint Surg 54A(4):787–797
Tsai T-Y, Lu T-W, Kuo M-Y, Hsu H-C (2009) Quantification of three-dimensional movement of
skin markers realtive to the underlying bones during functional activities. Biomed Eng: Appl
Basis Commun 21(3):223–232. doi:10.4015/S1016237209001283
Winter D, Robertson D (1978) Joint torque and energy patterns in normal gait. Biol Cybern
29:137–142. doi:10.1007/BF00337349
Variations of Marker Sets and Models
for Standard Gait Analysis
Felix Stief
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Anatomical and Technical Markers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Marker Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
The Definition of a Segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Prediction Approach or the Conventional Gait Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Functional Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Impact of Marker Set and Joint Angle Calculation on Gait Analysis Results . . . . . . . . . . . . . . . . . . 7
Errors Involved with Marker Placement and Soft-Tissue Artifacts . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Errors Associated with the Regression Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
How to Address the Measurement Error and What is the Extent of This Error? . . . . . . . . . . . 9
Accuracy for Marker-Based Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Comparison of Marker Sets and Models for Standard Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . 10
The Importance of Repeatability Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Removing the Effects of Marker Misplacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Abstract
A variety of different approaches is used in 3D clinical gait analysis. This chapter
provides an overview of common terms, different marker sets, underlying ana-
tomical models, as well as a fundamental understanding of measurement tech-
niques commonly used in clinical gait analysis and the consideration of possible
errors associated with these different techniques. Besides the different marker
F. Stief (*)
Movement Analysis Lab, Orthopedic University Hospital Friedrichsheim gGmbH, Frankfurt/Main,
Germany
e-mail: felix.stief@gmx.de; f.stief@friedrichsheim.de

DOI 10.1007/978-3-319-30808-1_26-1
2 F. Stief
sets, two main approaches can be used to quantify marker-based joint angles: a
prediction approach based on regression equations and a functional approach.
The prediction approach uses anatomical assumptions and anthropometric refer-
ence data to define the locations of joint centers/axes relative to specific anatom-
ical landmarks. In the functional approach, joint centers are determined via
optimization of marker movement. The accuracy of determining skeletal kine-
matics is limited by ambiguity in landmark identification and soft-tissue artifacts.
When the intersubject variability of control data becomes greater than the
expected change due to pathology, the clinical usefulness of the data becomes
doubtful. To allow a practical interpretation of a comparison of approaches,
differences and the measurement error should be quantified in the unit of interest
(i.e., degree or percent). The highest reliability indices occurred in the hip and
knee in the sagittal plane, with lowest reliability and highest errors for hip and
knee rotation in the transverse plane. In addition, knowledge about sources of
errors should be known before the approach is applied in practice.
Keywords
Marker sets • Anatomical markers • Technical markers • Clusters • Modeling •
Segment definition • Prediction approach • Functional approach • Regression
equations • Conventional Gait Model • Measurement error • Soft-tissue artifacts •
Reliability • Accuracy
Introduction
This chapter provides an overview of common terms, different marker sets, under-
lying anatomical models, as well as a fundamental understanding of measurement
techniques commonly used in clinical gait analysis and the consideration of possible
errors associated with these different techniques.
It is possible for a clinician or physician to subjectively study gait; however, the
value and repeatability of this type of assessment is questionable due to poor inter-
and intra-tester reliability. For instance, it is impossible for one individual to study,
by observation alone, the movement pattern of all the main joints involved during an
activity like walking simultaneously. Therefore, skeletal movements in three dimen-
sions during gait are typically recorded using markers placed on the surface of the
skin on various anatomical landmarks to represent body segments. The marker-
based analysis of human movement helps to better understand normal and patho-
logical function and results in a detailed and objective clinical assessment of
therapeutic and surgical interventions.
A variety of different anatomical models and marker sets were used for clinical
gait analysis. While a certain amount of standardization could be established in
recent years for the marker placement on anatomical points and the definition for
most of the rigid body segments (pelvic, thigh, shank, foot), protocols differ in the
underlying biomechanical model, the definition of joint centers and axes, and the
Variations of Marker Sets and Models for Standard Gait Analysis 3
number of markers used. These differences have an effect on the outcome measures
(e.g., joint angles and moments). The main focus of this chapter is to demonstrate the
impact of marker sets and joint angle calculations on gait analysis results.
State of the Art
Markers are either described as passive or active (“▶ Estimating Hierarchical Rigid
Body Models of the Musculoskeletal System”). Passive markers for camera-based
systems are generally made of a retroreflective material. This material is used to
reflect light emitted from around the camera back to the camera lens. Some camera-
based systems use a stroboscopic light, while others use light from synchronized
infrared light-emitting diodes mounted around the camera lens.
In contrast, active markers produce light at a given frequency, so these systems do
not require illumination, and, as such, the markers are more easily identified and
tracked (Chiari et al. 2005). These light-emitting diodes (LED) are attached to a body
segment in the same way as passive markers, but with the addition of a power source
and a control unit for each LED. Active markers can have their own specific
frequency which allows them to be automatically detected. This leads to very stable
real-time three-dimensional motion tracking as no markers can be misidentified as
adjacent markers. Regardless of whether passive or active, the use of markers should
not significantly modify the movement pattern being measured.
Anatomical and Technical Markers
Anatomical markers are used to set up the segment reference frame. This is generally
done during a static trial with the subject standing still. Anatomical markers may be
attached on bony landmarks directly to the skin or fixed to a pointer. These markers
are not required for the dynamic trials as long as at least three fixed points are
available on each segment.
Technical markers have no specific location and are chosen purely to meet the
other requirements above. Additional technical markers can be used to create a
technical coordinate system from data collected in a static calibration trial during
which both anatomical and technical markers are present. In subsequent dynamic
trials, absent anatomical markers can be expressed in relation to the technical
coordinate system. Technical markers can also be used to avoid areas of adipose
tissue in obese patients, to accommodate walking aids, or to replace markers that are
obscured dynamically. Two approaches are commonly used. Technical markers may
be used to replace only those anatomical markers that cannot be used dynamically. In
this case, the majority of anatomical markers remain in place for the walking trials.
Alternatively, clusters of technical markers attached to a plate (see “Marker Clusters”
below) may be used to provide all the dynamic information needed. Anatomical
markers are then only used for the static trial to allow segment reconstruction.
4 F. Stief
Fig. 1 Rigid marker cluster

with four retroreflective
markers
Marker Clusters
Other techniques for minimizing soft-tissue artifacts and in order to reduce

intersubject variability are marker clusters (an array of markers) (Cappozzo et al.
1995). They must be in place during the static anatomical calibration. The exact
placement of the clusters is less reliant as this technique uses the relative positions to
the anatomical landmarks used in the static calibration. The purpose is to define the
plane of each segment with 3–5 markers and then track its movement through the
basic reference planes. Clusters can be directly attached to the skin or mounted on
rigid fixtures (Fig. 1), which are dependent upon the anatomy, the activity, and the
nature of the analysis.
In a rigid body or cluster, the distance between any two points within the body or
cluster does not change. In general, tracking of marker clusters helps to reduce noise
within the motion signal and improve accuracy of kinematic data. When the markers
are fixed to rigid plates, the markers never move independently with deformation of
the skin. It has been shown that the absolute and relative variance in out-of-sagittal
plane rotations tended to be higher for the Conventional Gait Model (“▶ The
Conventional Gait Model: Success and Limitations”) compared with a cluster
technique (Duffell et al. 2014) and that a cluster marker set overcomes a number
of theoretical limitations compared to the conventional set (Collins et al. 2009) when
both models were compared simultaneously. Much work has been carried out
determining the optimal configuration of marker clusters, and it is now widely
accepted that a rigid shell with a cluster of four markers is a good practical solution
(Cappozzo et al. 1997; Manal et al. 2000). However, when the cluster markers were
fixed to a rigid plate, these methods were not able to address absolute errors and can
still result in inaccurate identification of joint centers (Holden and Stanhope 1998).
Although an extended version of this method has reported improvements in estima-
tion of the position of the underlying bones (Alexander and Andriacchi 2001), it can
only model skin deformations and has limited use in clinical applications due to the
number of additional markers required.
The Definition of a Segment
In general, three markers are needed to fix a rigid body in space. When using motion
capture to define the pelvic segment (“▶ Estimating Hierarchical Rigid Body Models
of the Musculoskeletal System”) and measure pelvic motion, the International
Society of Biomechanics (ISB) recommends the pelvic anatomical coordinate sys-
tem be defined by surface markers placed on the right and left anterior superior iliac
spines (ASISs) and on the right and left posterior superior iliac spines (PSISs). The
pelvic anatomical coordinate system can be described as the origin at the midpoint
between the right ASIS and the left ASIS; the Z-axis points from the origin to the
right ASIS; the X-axis lies in the plane defined by the right ASIS, left ASIS, and the
midpoint of the right PSIS and left PSIS markers and points ventrally orthogonal to
the Z-axis; and the Y-axis is orthogonal to these two axes (Wu et al. 2002). These
markers would ideally be used to track the pelvis during gait or clinical assessment
protocols that involve movement. However, situations in which the ASIS or PSIS
markers are obscured from view require that alternative technical marker sets are
used. Occlusion of the ASIS markers could be as a result of soft tissue around the
anterior abdomen (a common issue in overweight and obese subjects), arm move-
ment, or activities that require high degrees of hip and trunk flexion, such as running,
stair climbing, or level walking. It has been shown that pelvic models that include
markers placed on the ASISs and the iliac crests (ICs), and PSISs and ICs, are
suitable alternatives to the standard pelvic model (ASISs and PSISs) for tracking
pelvic motion during gait (Bruno and Barden 2015). Alternatively, the use of a rigid
cluster of three orthogonal markers as technical markers attached to the sacrum can
be used (Borhani et al. 2013). Using the calibrated anatomical system technique
(Benedetti et al. 1998; Cappello et al. 2005) allows the position of ASIS defined
relative to the cluster in a static trial, and then during dynamic trial, the position of
the ASIS is linked to the cluster and thus affected by the same skin movement artifact
that affects the cluster. Another alternative to solve skin artifacts is to use the right
and left hip joint centers described in the technical coordinate system of the right and
left thighs, together with the right PSIS and left PSIS markers, as technical markers
for tracking the pelvis movement (Kisho Fukuchi et al. 2010).
6 F. Stief
Prediction Approach or the Conventional Gait Model
Besides the different marker sets, two main approaches can be used to quantify joint
angles: a prediction approach based on regression equations and a functional
approach. The prediction approach uses anatomical assumptions and anthropomet-
ric reference data to define the locations of joint centers/axes relative to specific
anatomical landmarks (Isman and Inman 1969; Weidow et al. 2006). In the func-
tional approach, joint centers are determined via optimization of marker movement.
The advantages and disadvantages of both approaches were described below in
detail.
Most biomechanical analysis systems use regression equations based on predic-
tive methods to calculate joint centers. Kadaba et al. (1989), Davis III et al. (1991),
and Vaughan et al. (1992) provided detailed descriptions of a marker-based system to
calculate joint centers in the lower extremities. This marker setup has become one of
the most commonly used models in gait analysis. It is referred to as Helen Hayes
Hospital marker setup, and the regression equations are referred to as the Plug-in-
Gait (PiG) model or the Conventional Gait Model (“▶ The Conventional Gait
Model: Success and Limitations”).
Functional Approach
In general, technical marker sets require data capture in a static standing trial to
determine rotation values (offsets) to place these markers into the anatomical
coordinate system. If a marker does, for instance, not accurately represent the
position of the hip during standing data capture, the technical markers will not be
placed into the correct anatomical plane for the dynamic trial. This is particularly
problematic if the static and dynamic positions of the hip vary from one another. It
has been shown that static standing posture greatly affected the dynamic hip rotation
kinematics when using a thigh wand in the typical clinical gait analysis process for
the Conventional Gait Model (McMulkin and Gordon 2009). Therefore, if a thigh
wand is to be used in clinical practice, it is necessary that patients stand in a hip
rotation posture that is equivalent to hip rotation position used in gait. This can be
very difficult because it requires clinicians to have a priori knowledge of the gait hip
rotation before testing. Also, patients may use different strategies in static standing
than with walking posture. One way of addressing this issue is to use functional joint
center techniques (Ehrig et al. 2006; Leardini et al. 1999; Schwartz and Rozumalski
2005). This functional approach is considered functional due to the calculation of
subject-specific joint centers/axes by using specific movement data of adjacent
segments derived from basic motion tasks. With a focus on assessing motion patterns
in a subject-specific manner, functional methods rely on the relative motion between
the marker clusters of neighboring segments to identify joint centers and axes
(Cappozzo et al. 1997; Ehrig et al. 2006). Previously developed functional methods
have been demonstrated to be precise (Ehrig et al. 2006; Kornaropoulos et al. 2010;
Kratzenstein et al. 2012) as well as rapid and robust (Schwartz and Rozumalski
2005) in estimating joint centers. Nevertheless, in many patient groups, functional

calibration has been reported to be difficult (Sangeux et al. 2011) due to the fact that
the range of motion (ROM) of affected joints is restricted. In addition, functional
methods have not been able to demonstrate consistent advantages over more tradi-
tional regression-based approaches (Assi et al. 2016; Bell et al. 1990; Davis III et al.
1991; Harrington et al. 2007), possibly due to issues of marker placement and the
nonlinear distribution of soft-tissue artifacts across a segment (Gao and Zheng 2008;
Stagni et al. 2005). Kratzenstein et al. (2012) present an approach for understanding
the contribution of different regions of marker attachment on the thigh toward the
precise determination of the hip joint center. This working group used a combination
of established approaches (Taylor et al. 2010) to reduce skin marker artifacts (Taylor
et al. 2005), determine joint centers of rotation (Ehrig et al. 2006), and quantify the
weighting of each of a large number of markers (Heller et al. 2011) attached to the
thigh. Consequently, markers that are suboptimally located and therefore strongly
affected by soft-tissue artifacts are assigned a lower weighting compared to markers
that follow spherical trajectories around the joint. Based on these methods, six
regions of high importance were determined that produced a symmetrical center of
rotation estimation (Ehrig et al. 2011) almost as low as using a marker set that
covered the entire thigh. Such approaches could be used to optimize marker sets for
targeting more accurate and robust motion capture for aiding in clinical diagnosis
and improving the reliability of longitudinal studies.
Impact of Marker Set and Joint Angle Calculation on Gait Analysis

Results
Errors Involved with Marker Placement and Soft-Tissue Artifacts
The accuracy of determining skeletal kinematics is limited by ambiguity in landmark

identification and soft-tissue artifacts that is the motion of markers over the under-
lying bones due to skin elasticity, muscle contraction, or synchronous shifting of the
soft tissues (Leardini et al. 2005; Taylor et al. 2005). Generally, two types of errors
are referred to soft-tissue artifacts. Relative errors are defined as the relative move-
ment between two or more markers that define a rigid segment. Absolute errors are
defined as the movement of a marker with respect to the bony landmark it is
representing (Richards 2008). Relative and absolute errors are often caused by
movement of the soft tissue on which the markers are placed (Cappozzo et al.
1996). The magnitude of these errors has been studied by using pins secured directly
into the bone and comparing the data collected from skin-mounted markers to
markers attached to bone pins. These data give a direct measure of soft-tissue
movement with respect to the skeletal system (Cappozzo 1991; Cappozzo et al.
1996; Reinschmidt et al. 1997a; b). However, the applicability of this method is
limited due to their invasive nature. The amount and the effects of soft-tissue artifacts
from skin markers are discussed controversial with relative skin to bone marker
movements in the range of 3 mm up to 40 mm, dependent upon the specific body
8 F. Stief
segment and soft-tissue coverage (Cappozzo et al. 1996; Holden et al. 1997; Manal
et al. 2000, 2003; Reinschmidt et al. 1997b). Differences can be accounted for by
variation in marker placement and configuration, differences in techniques,
intersubject differences, and differences in the task performed (Leardini et al. 2005).
Inaccuracies in lower limb motion and in particular knee kinematics are present
mainly because of soft-tissue artifacts at the thigh segment (Alexander and
Andriacchi 2001; Cappello et al. 1997; Fuller et al. 1997; Leardini et al. 2005;
Lucchetti et al. 1998). Conversely, soft-tissue movement on the shank has only a
small effect on three-dimensional kinematics and moments at the knee (Holden et al.
1997; Manal et al. 2002). In addition, mainly in the frontal and transverse planes,
substantial angular variabilities were noted (Ferrari et al. 2008; Miana et al. 2009)
due to the small ROM in these planes compared to sagittal plane movements. This
reasoning agrees with the results of Leardini et al. (2005) who assert that angles out-
of-sagittal planes should be regarded with much more caution as the soft-tissue
artifact produces spurious effects with magnitudes comparable to the amount of
motion actually occurring in the joints. In addition, an increase in velocity (for
instance, during running) produces an increased variability of the joint centers’
distances and increases the maximum differences between the joint angles when
using different protocols (Miana et al. 2009).
Errors Associated with the Regression Equations
Besides soft-tissue artifacts and variability of the marker placement, errors associ-
ated with the regression equations used to calculate the joint center locations are also
considerable (Harrington et al. 2007; Leardini et al. 1999; Sangeux et al. 2011).
Clinically, the definition of the joint center is generally achieved by using palpable
anatomical landmarks to define the medial-lateral axis of the joint. From these
anatomical landmarks, the center of rotation is generally calculated in one of two
ways: through the use of regression equations based on standard radiographic
evidence or simply calculated as a percentage offset from the anatomical marker
based on some kind of anatomical landmarks (Bell et al. 1990; Cappozzo et al. 1995;
Davis III et al. 1991; Kadaba et al. 1989).
The issue of hip joint center (HJC) identification is one that has been covered in
much depth, and there are still many debates around this area. The location of this
joint center is one of the most difficult anatomic reference points to define. The
center of the femoral head is the center of the hip joint and located within the
acetabulum on the obliquely aligned and tilted lateral side of the pelvis. Therefore,
common approaches have used landmarks on the pelvis as the anatomical reference
(Perry and Burnfield 2010). The regression equations in the Conventional Gait
Model are based on the HJC regression equations by Davis et al. (1991) and chord
functions to predict the knee and the ankle joint centers. The HJC regression
equation was based on 25 male subjects and has been validated in later studies
(Harrington et al. 2007; Leardini et al. 1999; Sangeux et al. 2011) showing
significant errors, which were corrected with new regression equations (Sandau et al.
2015).
In the chord function, the HJC, thigh wand marker, and the epicondyle marker
were used to define a plane. The knee joint center (KJC) was then found so that the
epicondyle marker was at a half knee diameter distance from the KJC in a direction
perpendicular to the line from the HJC to KJC. The ankle joint center (AJC) was
predicted in the same way as the knee, where the chord function was used to predict
the joint center based on the KJC, the calf wand marker, and the malleolus marker.
The chord functions predict the KJC and the AJC with the assumption that the joint
centers are lying on the transepicondylar axis and the transmalleolar axis in the
frontal plane, respectively. This assumption seems reasonable for the knee (Asano
et al. 2005; Most et al. 2004), but to a lesser extent regarding the ankle joint
(Lundberg et al. 1989). The exact position of the joint centers influences the joint
angles as well as joint angular velocity and acceleration which are part of inverse
dynamics. Likewise, the location of segmental center of mass will influence the
inverse dynamics calculations via the moment arms acting together with both
proximal and distal joint reaction forces.
How to Address the Measurement Error and What is the Extent

of This Error?
In general, when addressing the measurement error in marker-based movement

analysis, it is helpful to provide an absolute measure of reliability, for instance, the
root mean square error or standard error of measurement (SEM). It is thus possible to
express the variability in a manner that can be directly related to the measurement
itself, in the same measurement units (e.g., degrees). Furthermore, with the trans-
formation of the absolute error into relative error, one can obtain the error expressed
as percentage corresponding to the total ROM of the variable to be analyzed. This is
of particular importance for the between-plane comparison of the measurement error
with different amplitude of the kinematic and kinetic parameters (Stief et al. 2013).
In contrast, the commonly reported intraclass correlation coefficient or coefficient
of variation and coefficient of multiple correlations allow limited information, as
high coefficient values can result from a low mean value of the variable of interest
and thus could hide measurement errors of clinical importance (Luiz and Szklo
2005). Furthermore, expressing data variability as a coefficient results in units that
are difficult to interpret clinically (Leardini et al. 2007).
Regarding the literature, kinematic measurement errors of less than 4 and 6
were reported for the intertrial and intersession variability, respectively (Stief et al.
2013). A systematic review from McGinley et al. (2009) identifies that the highest
reliability indices occurred in the hip and knee in the sagittal plane, with lowest
reliability and highest errors for hip and knee rotation in the transverse plane. Most
studies included in this review article providing estimates of data error reported
values of less than 5 , with the exception of hip and knee rotation. Fukaya et al.
(2013) investigated the interrater reliability of knee movement analyses during the
10 F. Stief
stance phase using a rigid marker set with three attached markers affixed to the thigh
and shank. Each of three testers independently attached the infrared reflective
markers to four subjects. The SEM values for reliability ranged from 0.68 to
1.13 for flexion-extension, 0.78 –1.60 for external-internal rotation, and
1.43 –3.33 for abduction-adduction. In general, the measurement errors between
testers are considered to be greater than the measurement errors between sessions
and within testers (Schwartz et al. 2004).
Accuracy for Marker-Based Gait Analysis
The accuracy of body protocols can hardly be assessed in clinical routine since
invasive methods such as radiographic imaging (Garling et al. 2007) or bone pins
(Taylor et al. 2005) are required in order to provide sufficient access to the skeletal
anatomy but are generally not available. Ultrasound assessment of the joint provides
one noninvasive opportunity (Sangeux et al. 2011), but assessment of the images can
be somewhat subjective. According to Schwartz and Rozumalski (2005), the fol-
lowing indirect indicators of accuracy can be computed instead:
1. Knee varus/valgus ROM during gait: An accurate knee flexion axis alignment
minimizes the varus/valgus ROM resulting from cross-talk, that is, one joint
rotation (e.g., flexion) being interpreted as another (e.g., adduction or varus)
due to axis malalignment (Piazza and Cavanagh 2000).
2. Knee flexion/extension ROM during gait: An accurate knee flexion axis align-
ment maximizes knee flexion/extension ROM by reducing cross-talk.
In general, the knee varus/valgus curve can be evaluated for signs of marker
misplacement or Knee Alignment Device misalignment. Moreover, it has been
shown that for the stable knee joint, the physiological ROM of knee varus/valgus
only varies between 5 and 10 (Reinschmidt et al. 1997a). Minimization of the knee
joint angle cross talk can therefore be considered to be a valid criterion to evaluate
the relative merits of different protocols and marker sets.
Comparison of Marker Sets and Models for Standard Gait Analysis
There is still a variety of different approaches being used in clinical gait analysis.
Protocols differ in the underlying biomechanical model, associated marker sets, and
data recording and processing. The former defines properties of the modeled joints,
the number of involved segments, the definitions of joint centers and axes, the used
anatomical and technical reference frames, and the angular decomposition technique
to calculate joint angles. Despite apparent differences of the outcome measures
derived from different gait protocols (Ferrari et al. 2008; Gorton et al. 2009),
specifically for out-of-sagittal plane rotations (Ferrari et al. 2008), data of different
studies are compared and interpreted.
Any protocol for movement analysis will only prove useful if it displays adequate
reliability (Cappozzo 1984). Moreover, and as stated before, the placement of the
markers has considerable influence on the accuracy of gait studies (Gorton et al.
2009). One of the first protocols proposed by Davis et al. (1991), and known as
Conventional Gait Model or PiG model, is still used by a vast majority of gait
laboratories (Schwartz and Rozumalski 2005). Although the protocol is practicable
and has been established over the years, some main disadvantages exist. It has been
shown that intersession and interexaminer reliability are low for this protocol,
especially at the hip and knee joint in the frontal and transverse plane (McGinley
et al. 2009). The errors in the PiG protocol, for example, knee varus/valgus ROM up
to 35 (Ferrari et al. 2008), are very likely caused by inconsistent anatomical
landmark identification and marker positioning by the examiner. This leads to
well-documented errors of skin movement (Leardini et al. 2005) and kinematic
cross talk. Moreover, accurate placement of the wand markers on the shank and
the thigh is difficult (Karlsson and Tranberg 1999). Wands on the lateral aspect of the
thighs and shanks are also likely to enlarge skin motion artifact effects (Manal et al.
2000) and variability of the gait results (Gorton et al. 2009).
One way of addressing these errors is the usage of additional medial malleolus
and medial femoral condyle markers to determine joint centers. This eliminates the
reliance on the difficult, subjective palpation of the thigh and tibia wand markers
necessary for the PiG model, which has been shown to have large variability (Gorton
et al. 2009) between laboratories and to enlarge skin motion artifact effects (Manal
et al. 2002), especially when placed proximally where the greatest soft-tissue artifact
of any lower-limb segment is found (Stagni et al. 2005). Besides that, it has been
shown that thigh wand markers capture approximately half of actual femoral axial
rotation (Schache et al. 2008; Schulz and Kimmel 2010; Wren et al. 2008). The
reason for this may be that substantial proportions of hip external-internal rotations
were being detected as knee motions by the marker sets using thigh markers (Schulz
and Kimmel 2010). Wren et al. (2008) have suggested using a patella marker (placed
in the center of the patella), which was reported to detect 98% of the actual hip
rotation ROM. And indeed, dynamic hip rotation during gait when utilizing a patella
marker in lieu of a thigh wand was not effected by static hip posture (McMulkin and
Gordon 2009).
In a comparative study, the reliability and accuracy of the PiG model and an
advanced protocol (MA) with additional medial malleolus and medial femoral
condyle markers were estimated (Stief et al. 2013) (Fig. 2).
For the MA, neither anthropometric measurements nor joint alignment devices
are necessary. Knowledge of anatomical landmarks spatial location enables auto-
matic calculation of anthropometric measurements necessary for joint center deter-
mination. In both protocols, the center of the hip joint was calculated using a
geometrical prediction method (Davis III et al. 1991). The PiG model derived the
rotational axis of the knee joint from the position of the pelvic, knee, and thigh
markers and the rotational axis of the ankle joint from the position of the knee, ankle,
and tibia markers. In contrast to the PiG model, the centers of the knee and ankle
joints using the MA were statically defined as the midpoint between the medial and
12 F. Stief
SACR
LASI
RASI
RTRO LTRO
RTHI
LTHI
LKNEL
RKNEL LKNEM
RKNEM
RTIB LTIB
LHEE
RHEE LANKL
RANKL
LANKM
RANKM
LTOE
RTOE
Abbreviation Placement Required for protocol

SACR On the skin mid-way between the posterior superior iliac spines PiG / MA
LASI (RASI) On the left (right) anterior superior iliac spine PiG / MA
LTRO (RTRO) On the prominent point of the left (right) trochanter major MA
LTHI (RTHI) Rigid wand marker mounted on the skin over the distal and lateral aspect PiG
of the left (right) thigh aligned in the plane that contains the hip and knee joint centers
and the knee flexion/extension axis
LKNEL (RKNEL) On the left (right) lateral femoral condyle PiG / MA
LKNEM (RKNEM) On the left (right) medial femoral condyle MA
LTIB (RTIB) Rigid wand marker mounted on the skin over the distal and lateral aspect PiG
of the left (right) shank aligned in the plane that contains the knee and ankle joint
centers and the ankle flexion/extension axis
LANKL (RANKL) On the left (right) lateral malleolus aligned with the bimalleolar axis PiG / MA
LANKM (RANKM) On the left (right) medial malleolus aligned with the bimalleolar axis MA
LTOE (RTOE) Ont the left (right) second metatarsal head, on the mid-foot sides PiG / MA
of the equinus break between fore-foot and mid-foot
LHEE (RHEE) On the left (right) aspect of the Achilles tendon insertion, on the calcaneous PiG / MA
at the same height above the plantar surface of the foot as the LTOE (RTOE) marker
Fig. 2 Marker set of both lower body protocols. The markers indicated by circles are part of the
standard Plug-in-Gait (PiG) marker set (Conventional Gait Model); those indicated by triangles are
the additional markers used in the custom made protocol (MA)
lateral femoral condyle and malleolus markers. The anatomical medial malleolus and
femoral condyle markers can then be removed for the dynamic trials.
The results of this comparative study (PiG model vs. MA) show for both pro-
tocols and healthy subjects a good intersession reliability for all ankle, knee, and hip
joint angles in the sagittal plane. Nevertheless, the lower intersession errors for the
MA compared to the PiG model regarding frontal plane knee angles and moments
and transverse plane motion in the knee and hip joint suggest that the error in
repeated palpation of the landmarks is lower using the MA. Moreover, the MA
significantly reduced the knee axis cross talk phenomenon, suggesting improved
accuracy of knee axis alignment compared to the PiG model. These results are
comparable to those reported by Schwartz and Rozumalski (2005) using a functional
approach in comparison with the PiG model. The MA eliminates the reliance on the
subjective palpation of the thigh and tibia wand markers and the application of a
Knee Alignment Device method (Davis and DeLuca 1996), which is difficult to
handle and less reliable within or between therapists than manual palpation, espe-
cially in non-experienced investigators (Serfling et al. 2009). Nevertheless, a correct
marker placement based on the exact identification of the characteristic anthropo-
logical points of the body (bony landmarks) is required. Especially the position of
the knee markers is very important, because it influences not only knee joint
kinematics, but also hip and ankle joints. It has been shown that simultaneous
knee hyperextension, internal hip rotation, and external ankle rotation can be caused
by back lateral knee marker misplacement, and simultaneous knee overflexion,
external hip rotation, and internal ankle rotation may be influenced by forward
knee marker misplacement (Szczerbik and Kalinowska 2011). Therefore, if such
phenomena are represented by kinematic graphs, their presence should be confirmed
by video registration prior to the formulation of clinical conclusions.
Future Directions
There are many anatomical models and marker sets reported in the literature. The
increase in complexity in the models relates not only to the ability of movement
analysis systems to track more and more markers but also in the increase in the
knowledge of modeling human movement.
The Importance of Repeatability Studies
Some of the theoretical aspects of marker placement have been presented in this
chapter. The practical implications are best explored in the gait laboratory by repeat
marker placement. Repeated testing of a single subject will give some insight into
the variability for a single person placing markers (intrasubject reliability) and
between different people placing markers (intersubject reliability). Intersubject var-
iability would additionally be affected by differences in each subject’s walking style
and between-subject differences in marker placement and motion relative to bony
landmarks. When the intersubject variability of control data becomes greater than the
expected change due to pathology, the clinical usefulness of the data becomes
doubtful. To allow a practical interpretation of a comparison of approaches, differ-
ences and their variability should be quantified in the unit of interest (i.e., degree or
percent).
14 F. Stief
Removing the Effects of Marker Misplacement
The placement of markers is not easy, and there is a limit to the accuracy we can
realistically achieve. Even if the markers are in the right place, the effects of skin
movement and oscillation will introduce errors once the subject is walking. One
possibility is that the marker placement is “corrected” as part of the data processing.
Complex algorithms are now becoming available for performing such corrections. A
simpler approach has been used for some time to increase the accuracy in joint center
determination. In addition to the Conventional Gait Model, at least the use of medial
malleolus and medial femoral condyle markers is recommended when analyzing
frontal and transverse plane gait data. This should lead to lower measurement errors
for most of the gait variables and to a more accurate determination of the knee joint
axis. Nevertheless, gait variables in the transverse plane are poorly reproducible
(Ferber et al. 2002; Krauss et al. 2012), and their variability associated with the
underlying biomechanical protocol is substantial (Ferrari et al. 2008; Krauss et al.
2012; Noonan et al. 2003).
In future, approaches that combine key characteristics of proven methods (func-
tional and/or predictive methods) for the assessment of skeletal kinematics could be
used to optimize marker sets for targeting more accurate and robust motion capture
for aiding in clinical diagnosis and improving the reliability of longitudinal studies.
On the other hand, procedural distress should be minimized. Especially children
cannot always stand still for a long time, walk wearing a large number of markers,
and perform additional motion trials. The marker set and possible associated ana-
tomical landmark calibration or anthropometric measurement procedures, therefore,
must be minimized to contain the time taken for subject preparation and data
collection (Leardini et al. 2007).
Conclusion
When comparing movement data, it is worth noting that care must be taken where
different marker sets have been used. Whatever approach is used, the problem is
separating patterns produced by errors from those produced by pathology. Till this
day, it is, for instance, not clear how different marker configurations impact hip
rotation for the typical clinical gait analysis process. For this reason, the “true”
values for rotation often remain unknown. Therefore, gait protocols have to be
described precisely, and comparison with other studies should be done critically.
In addition, knowledge about sources of errors should be known before the approach
is applied in practice. Learning and training of the examiners, which is considered to
be a critical issue (Gorton et al. 2009), is important to ensure exact anatomical
landmark locations which may also reduce intra- and inter-examiner variability.
Moreover, the graphs from instrumented gait analysis should be confirmed by
video registration prior to the formulation of clinical conclusions.
References
Alexander EJ, Andriacchi TP (2001) Correcting for deformation in skin-based marker systems. J
Biomech 34(3):355–361
Asano T, Akagi M, Nakamura T (2005) The functional flexion-extension axis of the knee corre-
sponds to the surgical epicondylar axis: in vivo analysis using a biplanar image-matching
technique. J Arthroplast 20(8):1060–1067. doi:10.1016/j.arth.2004.08.005
Assi A, Sauret C, Massaad A, Bakouny Z, Pillet H, Skalli W, Ghanem I (2016) Validation of hip
joint center localization methods during gait analysis using 3D EOS imaging in typically
developing and cerebral palsy children. Gait Posture 48:30–35. doi:10.1016/j.
Bell AL, Pedersen DR, Brand RA (1990) A comparison of the accuracy of several hip center
location prediction methods. J Biomech 23(6):617–621
Benedetti MG, Catani F, Leardini A, Pignotti E, Giannini S (1998) Data management in gait
analysis for clinical applications. Clin Biomech (Bristol, Avon) 13(3):204–215
Borhani M, McGregor AH, Bull AM (2013) An alternative technical marker set for the pelvis is
more repeatable than the standard pelvic marker set. Gait Posture 38(4):1032–1037.
doi:10.1016/j.gaitpost.2013.05.019
Bruno P, Barden J (2015) Comparison of two alternative technical marker sets for measuring 3D
pelvic motion during gait. J Biomech 48(14):3876–3882. doi:10.1016/j.jbiomech.2015.09.031
Cappello A, Cappozzo A, La Palombara PF, Lucchetti L, Leardini A (1997) Multiple anatomical
landmark calibration for optimal bone pose estimation. Hum Mov Sci 16(2–3):259–274.
doi:10.1016/S0167-9457(96)00055-3
Cappello A, Stagni R, Fantozzi S, Leardini A (2005) Soft tissue artifact compensation in knee
kinematics by double anatomical landmark calibration: performance of a novel method during
selected motor tasks. IEEE Trans Biomed Eng 52(6):992–998. doi:10.1109/tbme.2005.846728
Cappozzo A (1984) Gait analysis methodology. Hum Mov Sci 3(1–2):27–50. doi:10.1016/0167-
9457(84)90004-6
Cappozzo A (1991) Three-dimensional analysis of human walking: experimental methods and
associated artifacts. Hum Mov Sci 10(5):589–602. doi:10.1016/0167-9457(91)90047-2
Cappozzo A, Catani F, Croce UD, Leardini A (1995) Position and orientation in space of bones
during movement: anatomical frame definition and determination. Clin Biomech (Bristol, Avon)
10(4):171–178
Cappozzo A, Catani F, Leardini A, Benedetti MG, Croce UD (1996) Position and orientation
in space of bones during movement: experimental artefacts. Clin Biomech (Bristol, Avon) 11(2):
90–100
Cappozzo A, Cappello A, Della Croce U, Pensalfini F (1997) Surface-marker cluster design criteria
for 3-D bone movement reconstruction. IEEE Trans Biomed Eng 44(12):1165–1174.
doi:10.1109/10.649988
Chiari L, Della Croce U, Leardini A, Cappozzo A (2005) Human movement analysis using
stereophotogrammetry. Part 2: instrumental errors. Gait Posture 21(2):197–211. doi:10.1016/j.
Collins TD, Ghoussayni SN, Ewins DJ, Kent JA (2009) A six degrees-of-freedom marker set for
gait analysis: repeatability and comparison with a modified Helen Hayes set. Gait Posture 30(2):
173–180. doi:10.1016/j.gaitpost.2009.04.004
Davis RB, DeLuca PA (1996) Clinical gait analysis: current methods and future directions. In:
Harris GF, Smith PA (eds) Human motion analysis: current applications and future directions.
The Institute of Electrical and Electronic Engineers Press, New York, pp 17–42
Davis RB III, Õunpuu S, Tyburski D, Gage JR (1991) A gait analysis data collection and reduction
technique. Hum Mov Sci 10(5):575–587. doi:10.1016/0167-9457(91)90046-Z
Duffell LD, Hope N, McGregor AH (2014) Comparison of kinematic and kinetic parameters
calculated using a cluster-based model and Vicon’s plug-in gait. Proc Inst Mech Eng H 228
(2):206–210. doi:10.1177/0954411913518747
16 F. Stief
Ehrig RM, Taylor WR, Duda GN, Heller MO (2006) A survey of formal methods for determining
the centre of rotation of ball joints. J Biomech 39(15):2798–2809. doi:10.1016/j.
jbiomech.2005.10.002
Ehrig RM, Heller MO, Kratzenstein S, Duda GN, Trepczynski A, Taylor WR (2011) The SCoRE
residual: a quality index to assess the accuracy of joint estimations. J Biomech 44
(7):1400–1404. doi:10.1016/j.jbiomech.2010.12.009
Ferber R, McClay Davis I, Williams DS 3rd, Laughton C (2002) A comparison of within- and
between-day reliability of discrete 3D lower extremity variables in runners. J Orthop Res 20
(6):1139–1145. doi:10.1016/s0736-0266(02)00077-3
Ferrari A, Benedetti MG, Pavan E, Frigo C, Bettinelli D, Rabuffetti M, Crenna P, Leardini A (2008)
Quantitative comparison of five current protocols in gait analysis. Gait Posture 28(2):207–216.
doi:10.1016/j.gaitpost.2007.11.009
Fukaya T, Mutsuzaki H, Wadano Y (2013) Interrater reproducibility of knee movement analyses
during the stance phase: use of anatomical landmark calibration with a rigid marker set. Rehabil
Res Pract 2013:692624. doi:10.1155/2013/692624
Fuller J, Liu LJ, Murphy MC, Mann RW (1997) A comparison of lower-extremity skeletal
kinematics measured using skin- and pin-mounted markers. Hum Mov Sci 16(2–3):219–242.
doi:10.1016/S0167-9457(96)00053-X
Gao B, Zheng NN (2008) Investigation of soft tissue movement during level walking: translations
and rotations of skin markers. J Biomech 41(15):3189–3195. doi:10.1016/j.
Garling EH, Kaptein BL, Mertens B, Barendregt W, Veeger HE, Nelissen RG, Valstar ER (2007)
Soft-tissue artefact assessment during step-up using fluoroscopy and skin-mounted markers. J
Biomech 40(Suppl 1):S18–S24. doi:10.1016/j.jbiomech.2007.03.003
Gorton GE 3rd, Hebert DA, Gannotti ME (2009) Assessment of the kinematic variability among
12 motion analysis laboratories. Gait Posture 29(3):398–402. doi:10.1016/j.
imaging. J Biomech 40(3):595–602. doi:10.1016/j.jbiomech.2006.02.003
Heller MO, Kratzenstein S, Ehrig RM, Wassilew G, Duda GN, Taylor WR (2011) The weighted
optimal common shape technique improves identification of the hip joint center of rotation
in vivo. J Orthop Res 29(10):1470–1475. doi:10.1002/jor.21426
Holden JP, Stanhope SJ (1998) The effect of variation in knee center location estimates on net knee
joint moments. Gait Posture 7(1):1–6
Holden JP, Orsini JA, Siegel KL, Kepple TM, Gerber LH, Stanhope SJ (1997) Surface movement
errors in shank kinematics and knee kinetics during gait. Gait & Posture 5(3):217–227.
doi:10.1016/S0966-6362(96)01088-0
Isman RE, Inman VT (1969) Anthropometric studies of the human foot and ankle. Bull Prosthet Res
10(11):97–219
(6):849–860. doi:10.1002/jor.1100070611
Karlsson D, Tranberg R (1999) On skin movement artefact-resonant frequencies of skin markers
attached to the leg. Hum Mov Sci 18(5):627–635. doi:10.1016/S0167-9457(99)00025-1
Kisho Fukuchi R, Arakaki C, Veras Orselli MI, Duarte M (2010) Evaluation of alternative technical
markers for the pelvic coordinate system. J Biomech 43(3):592–594. doi:10.1016/j.
Kornaropoulos EI, Taylor WR, Duda GN, Ehrig RM, Matziolis G, Muller M, Wassilew G,
Asbach P, Perka C, Heller MO (2010) Frontal plane alignment: an imageless method to predict
the mechanical femoral-tibial angle (mFTA) based on functional determination of joint centres
and axes. Gait Posture 31(2):204–208. doi:10.1016/j.gaitpost.2009.10.006
Kratzenstein S, Kornaropoulos EI, Ehrig RM, Heller MO, Popplau BM, Taylor WR (2012)
Effective marker placement for functional identification of the centre of rotation at the hip.
Gait Posture 36(3):482–486. doi:10.1016/j.gaitpost.2012.04.011
Krauss I, List R, Janssen P, Grau S, Horstmann T, Stacoff A (2012) Comparison of distinctive gait
variables using two different biomechanical models for knee joint kinematics in subjects with
knee osteoarthritis and healthy controls. Clin Biomech (Bristol, Avon) 27(3):281–286.
doi:10.1016/j.clinbiomech.2011.09.013
(1999) Validation of a functional method for the estimation of hip joint centre location. J
Biomech 32(1):99–103
21(2):212–225. doi:10.1016/j.gaitpost.2004.05.002
based protocol for gait analysis in children. Gait Posture 26(4):560–571. doi:10.1016/j.
Lucchetti L, Cappozzo A, Cappello A, Della Croce U (1998) Skin movement artefact assessment
and compensation in the estimation of knee-joint kinematics. J Biomech 31(11):977–984
Luiz RR, Szklo M (2005) More than one statistical strategy to assess agreement of quantitative
measurements may usefully be reported. J Clin Epidemiol 58(3):215–216. doi:10.1016/j.
jclinepi.2004.07.007
Lundberg A, Svensson OK, Nemeth G, Selvik G (1989) The axis of rotation of the ankle joint. J
Bone Joint Surg (Br) 71(1):94–99
Manal K, McClay I, Stanhope S, Richards J, Galinat B (2000) Comparison of surface mounted
markers and attachment methods in estimating tibial rotations during walking: an in vivo study.
Gait Posture 11(1):38–45
Manal K, McClay I, Richards J, Galinat B, Stanhope S (2002) Knee moment profiles during
walking: errors due to soft tissue movement of the shank and the influence of the reference
coordinate system. Gait Posture 15(1):10–17
Manal K, McClay Davis I, Galinat B, Stanhope S (2003) The accuracy of estimating proximal tibial
translation during natural cadence walking: bone vs. skin mounted targets. Clin Biomech
(Bristol, Avon) 18(2):126–131
gait measurements: a systematic review. Gait Posture 29(3):360–369. doi:10.1016/j.
McMulkin ML, Gordon AB (2009) The effect of static standing posture on dynamic walking
kinematics: comparison of a thigh wand versus a patella marker. Gait Posture 30(3):375–378.
doi:10.1016/j.gaitpost.2009.06.010
Miana AN, Prudencio MV, Barros RM (2009) Comparison of protocols for walking and running
kinematics based on skin surface markers and rigid clusters of markers. Int J Sports Med 30
(11):827–833. doi:10.1055/s-0029-1234054
Most E, Axe J, Rubash H, Li G (2004) Sensitivity of the knee joint kinematics calculation to
selection of flexion axes. J Biomech 37(11):1743–1748. doi:10.1016/j.jbiomech.2004.01.025
Noonan KJ, Halliday S, Browne R, O’Brien S, Kayes K, Feinberg J (2003) Interobserver variability
of gait analysis in patients with cerebral palsy. J Pediatr Orthop 23(3):279–287 discussion
288-291
Perry J, Burnfield JM (2010) Gait Analysis. Normal and pathological function, 2nd edn. SLACK
Incorporated, Thorofare
Piazza SJ, Cavanagh PR (2000) Measurement of the screw-home motion of the knee is sensitive to
errors in axis alignment. J Biomech 33(8):1029–1034
Reinschmidt C, van den Bogert AJ, Lundberg A, Nigg BM, Murphy N, Stacoff A, Stano A (1997a)
Tibiofemoral and tibiocalcaneal motion during walking: external vs. skeletal markers. Gait
Posture 6(2):98–109. doi:10.1016/S0966-6362(97)01110-7
18 F. Stief
Reinschmidt C, van den Bogert AJ, Nigg BM, Lundberg A, Murphy N (1997b) Effect of skin
movement on the analysis of skeletal knee joint motion during running. J Biomech 30
(7):729–732
Richards J (2008) Biomechanics in clinic and research. Elsevier, Philadelphia
Sandau M, Heimburger RV, Villa C, Jensen KE, Moeslund TB, Aanaes H, Alkjaer T, Simonsen EB
(2015) New equations to calculate 3D joint centres in the lower extremities. Med Eng Phys 37
(10):948–955. doi:10.1016/j.medengphy.2015.07.001
Sangeux M, Peters A, Baker R (2011) Hip joint centre localization: evaluation on normal subjects in
the context of gait analysis. Gait Posture 34(3):324–328. doi:10.1016/j.gaitpost.2011.05.019
Schache AG, Baker R, Lamoreux LW (2008) Influence of thigh cluster configuration on the
estimation of hip axial rotation. Gait Posture 27(1):60–69. doi:10.1016/j.gaitpost.2007.01.002
Schulz BW, Kimmel WL (2010) Can hip and knee kinematics be improved by eliminating thigh
markers? Clin Biomech (Bristol, Avon) 25(7):687–692. doi:10.1016/j.clinbiomech.2010.04.002
Schwartz MH, Trost JP, Wervey RA (2004) Measurement and management of errors in quantitative
gait data. Gait Posture 20(2):196–203. doi:10.1016/j.gaitpost.2003.09.011
Serfling DM, Hooke AW, Bernhardt KA, Kaufman KR, 2009 Comparison of techniques for finding
the knee joint center. In: Proceedings of the gait and clinical movement analysis society. p 43
Stagni R, Fantozzi S, Cappello A, Leardini A (2005) Quantification of soft tissue artefact in motion
analysis by combining 3D fluoroscopy and stereophotogrammetry: a study on two subjects. Clin
Biomech (Bristol, Avon) 20(3):320–329. doi:10.1016/j.clinbiomech.2004.11.012
Stief F, Bohm H, Michel K, Schwirtz A, Doderlein L (2013) Reliability and accuracy in three-
dimensional gait analysis: a comparison of two lower body protocols. J Appl Biomech 29
(1):105–111
Szczerbik E, Kalinowska M (2011) The influence of knee marker placement error on evaluation of
gait kinematic parameters. Acta Bioeng Biomech 13(3):43–46
Taylor WR, Ehrig RM, Duda GN, Schell H, Seebeck P, Heller MO (2005) On the influence of soft
tissue coverage in the determination of bone kinematics using skin markers. J Orthop Res 23
(4):726–734. doi:10.1016/j.orthres.2005.02.006
Taylor WR, Kornaropoulos EI, Duda GN, Kratzenstein S, Ehrig RM, Arampatzis A, Heller MO
(2010) Repeatability and reproducibility of OSSCA, a functional approach for assessing the
kinematics of the lower limb. Gait Posture 32(2):231–236. doi:10.1016/j.gaitpost.2010.05.005
Vaughan CL, Davis BL, O’Conner JC (1992) Dynamics of human gait. Human Kinetics Publishers,
Champaign
Weidow J, Tranberg R, Saari T, Karrholm J (2006) Hip and knee joint rotations differ between
patients with medial and lateral knee osteoarthritis: gait analysis of 30 patients and 15 controls. J
Orthop Res 24(9):1890–1899. doi:10.1002/jor.20194
Wren TA, Do KP, Hara R, Rethlefsen SA (2008) Use of a patella marker to improve tracking of
dynamic hip rotation range of motion. Gait Posture 27(3):530–534. doi:10.1016/j.
Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D, Whittle M, D’Lima DD,
Cristofolini L, Witte H, Schmid O, Stokes I (2002) ISB recommendation on definitions of
joint coordinate system of various joints for the reporting of human joint motion – part I: ankle,
hip, and spine. International Society of Biomechanics. J Biomech 35(4):543–548
Next-Generation Models Using Optimized
Joint Center Location
Ayman Assi, Wafa Skalli, and Ismat Ghanem
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Motion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Motion Capture Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Joint Kinematics and Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Hip Joint Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Predictive Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Functional Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Knee Joint Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Ankle Joint Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Glenohumeral Joint Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Validation of the Joint Center Localization Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
X-Rays and Stereophotogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Magnetic Resonance Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3D Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Low-Dose Biplanar X-Rays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Effect of Errors on JC Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
A. Assi (*)
Laboratory of Biomechanics and Medical Imaging, Faculty of Medicine, University of
Saint-Joseph, Mar Mikhael, Beirut, Lebanon
Institut de Biomécanique Humaine Georges Charpak, Arts et Métiers ParisTech, Paris, France
e-mail: ayman.assi@gmail.com
W. Skalli
Institut de Biomécanique Humaine Georges Charpak, Arts et Métiers ParisTech, Paris, France
e-mail: wafa.skalli@ensam.eu
I. Ghanem
Laboratory of Biomechanics and Medical Imaging, Faculty of Medicine, University of
Saint-Joseph, Mar Mikhael, Beirut, Lebanon
Hôtel-Dieu de France Hospital, University of Saint-Joseph, Beirut, Lebanon
e-mail: ismat.ghanem@gmail.com

DOI 10.1007/978-3-319-30808-1_27-1
2 A. Assi et al.
Errors on Kinematics and Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Errors on Musculoskeletal Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Correction of 3D Positioning of the JC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Registration Techniques for the Use of Exact Joint Center Location . . . . . . . . . . . . . . . . . . . . . . . . 16
Estimation from External Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Abstract
Joint center location is essential in order to define anatomical axes of skeletal
segments and is therefore clinically significant for the calculation of joint kine-
matics during motion analysis. Different methods exist to localize joint centers
using either predictive methods, based on anthropometric measurements, or
functional methods, based on the relative movement of the segments adjacent
to the joint. Validations of these methods using medical imaging have been
extensively studied in the literature on different groups of subjects. Consequently,
methods of correction between the calculated location of the joint center and the
exact one, found by medical imaging, were suggested by several authors. Recent
studies showed that new age-specific predictive methods could be computed in
order to better locate joint coordinate systems. In the future, new techniques could
use the exact locations of joint centers, which would be localized by medical
imaging, in combination with motion capture techniques using registration tech-
niques; thus, exact kinematics and kinetics of the joints could be computed.
Keywords
Joint center • Predictive • Functional • Medical imaging • Validation
State of the Art
Joint center location is essential in order to calculate anatomical 3D joint kinematics,

kinetics, and muscle lever arms during musculoskeletal simulations. Several
methods can be used in order to localize the joint center. These methods can be
either predictive or functional. While the predictive methods are mainly based on
regression equations that use anthropometric measurements, the functional ones
require the performance of a joint movement in order to estimate the center of
rotation between the two adjacent segments. Most of the softwares used in motion
capture systems are based on the predictive methods. These methods have the
advantage of being more rapid and easier to use, especially when used on subjects
with disabilities who require assistance in performing the ranges of motion needed
for the use of functional methods.
Several authors have attempted to compare different methods or to validate these
methods using 3D medical imaging as a gold standard.
The CT scan has been used as a validation method since it is known to be precise
for 3D reconstruction; however, its use is critical because of the high dose of
Next-Generation Models Using Optimized Joint Center Location 3
radiation which it entails. Magnetic resonance imaging could solve the problem of
radiation, but this technique is known to be time-consuming because of acquisition
time and the necessity of image segmentation during the post-processing phase.
More recently, new techniques, using 3D ultrasound to obtain the joint center
location, have been explored and were found to have a precision of 4 mm (Peters
et al. 2010). The major inconvenience of this technique is the calibration needed
prior to acquisition. Moreover, the skeletal segments, joint center, and external
markers needed for motion analysis cannot be captured in the same image.
In the past years, the low-dose biplanar X-ray technique has shown very high
potential in clinical diagnosis of skeletal deformities through accurate 3D recon-
structions of the spine and the lower limbs (Dubousset et al. 2005; Humbert et al.
2009; Chaibi et al. 2012). In addition to this feature, it allows 3D reconstruction of
points of interest, such as joint centers. This technique has been shown to be precise
(2.9 mm) in localizing external markers and joint centers that appear in the same
image along with the skeletal segments. This technique was recently used in the
validation of joint center localization techniques which are commonly used in the
literature (Sangeux et al. 2014; Assi et al. 2016).
The validation studies have shown that the functional methods are more precise
than the predictive methods in localizing the joint center in the adult population.
However, special attention should be given to the range of motion performed by the
subject during the calibration trial (i.e., flexion-extension, abduction-adduction, and
internal-external rotation should be > 30 ). It was surprising to find that this was not
the case when it comes to children, where the predictive methods were found to be
more precise compared to the functional ones (Peters et al. 2012; Assi et al. 2016).
This chapter will first develop the need of joint center localization in motion
analysis and then the current techniques of joint center localization that can be used
during motion analysis processing. The validation process of these techniques will
be reviewed at a later stage along with the effect of errors of misplacement of the
joint center on kinematics, kinetics, and model simulations. The future directions
will be discussed at the end of this chapter.
Motion Analysis
Purpose
Medical imaging is widely used in the diagnosis of musculoskeletal diseases through

the use of images of human body anatomy. While different medical images can be
used in order to visualize the musculoskeletal system, such as X-rays, CT scan, or
MRI, these modalities allow images only in a static position. Apparent dynamic
images can be obtained when the patient or subject is asked to perform a certain
motion (i.e., shoulder) and then hold still in a given position during image acquisi-
tion. Thereafter, images are collected from different joint positions in order to obtain
pseudo-dynamic images. The same images can be obtained using fluoroscopy, but
4 A. Assi et al.
this technique is known to be highly irradiant (since it is an X-ray video) and has
small image dimensions.
The technique of motion analysis has been widely used since the early 1990s in
order to assess the joint motions of the musculoskeletal system, especially in patients
with orthopedic disorders such as cerebral palsy (Gage 1993). This technique is
based on 3D reconstruction by stereophotogrammetry of external markers positioned
on the skin of a subject (Cappozzo et al. 2005).
Motion Capture Techniques
Different motion capture techniques exist (check section on “Methods and Models:
Dynamic Pose Estimation”). In this chapter, we will be focusing on infrared wave-
based systems. The markers that are fixed on the subject’s skin could either be active
or passive. Active markers send waves to the cameras fixed in the acquisition room,
whereas passive markers only reflect waves to the transmissive-receptive cameras.
These cameras send their waves at a high frequency (i.e., 50 Hz and more) in order to
reconstruct the movement.
Joint Kinematics and Kinetics
Since a 3D coordinate system can be obtained using three nonlinear points, marker
placement on the skin respects this rule by placing at least three markers on each
skeletal segment. Then, a local coordinate system is calculated for each skeletal
segment at each frame of movement. These local coordinate systems are expressed
in a global coordinate system defined in the acquisition room during a calibration
process, performed prior to motion trials (Cappozzo et al. 1995, 2005) (check section
on “Methods and Models: Data Analysis”).
In a second step, the angles between adjacent local coordinate systems are
calculated by applying either the Euler method or the Cardan method, which require
the specification of the axes sequence. A consensus of joint angle definitions was set
by the International Society of Biomechanics in 2005 (Wu et al. 2002, 2005). An
illustration of motion capture and kinematic curves is presented in Fig. 1.
Need for a Joint Center

The markers are usually placed on palpable bony landmarks and this is for two main
reasons: (1) to avoid displacement of the marker during motion because of the
underlying soft tissue movement and (2) to ensure repeatability of placement of
markers between operators and between subjects. Moreover, the skeletal coordinate
system obtained from these reflective markers should be anatomically relevant in
order to obtain anatomical angles between adjacent segments, i.e., a local coordinate
system of the humerus should have one axis that represents the diaphysis and a
second axis in relation with the line joining the two epicondyles, and the third axis
derives from the two others.
Fig. 1 Motion analysis: (a) subject equipped with reflective markers, (b) 3D capture of subject’s
movement, (c) example of kinematic curves
6 A. Assi et al.
In some cases, a rigid body on which three or more markers are fixed, called
cluster, is attached to the segment. This method serves to reduce soft tissue artifacts.
The local coordinate system obtained from a cluster is not anatomically representa-
tive of the segment. Thus, the performance of a static calibration trial, in which a
transformation matrix is calculated between the anatomical and cluster coordinate
systems, is recommended prior to the movement trials.
Since the local coordinate system of a skeletal segment should be anatomically
relevant and representative of the geometry of the bone, in some cases the placement
of a marker on the joint center is required, i.e., hip joint center for the femoral
segment, knee joint center for the tibial segment, and glenohumeral joint center for
the humeral segment. Since it is impossible to place a marker in the joint center,
several techniques exist to approximate the location of this point.
Kinematics calculated during gait analysis usually includes nine graphs. These
graphs represent the joint angular waveforms during a gait cycle in the three planes:
sagittal, frontal, and horizontal. Six of these nine graphs, the hip and the knee joints in
the three planes each, are based on the use of the local coordinate system of the femur,
which necessitates the hip joint center localization. Thus, the hip joint center is one of
the most important joint centers to be localized in gait analysis. In the following section,
we discuss the hip joint center in details followed by the methods used in other joints.
Hip Joint Center
Different methods are available for the hip joint center localization. These methods
could be predictive, based on cadaveric or anthropometric measurements, or func-
tional, based on the movement of a segment relatively to the adjacent one.
Predictive Methods
The predictive methods usually use anthropometric regression-based equations

established from cadaveric specimens or in vivo medical imaging of the skeletal
segments. Some examples of predictive methods are exposed in this section (more
information can be found in previous chapters in section “Medical Application:
Assessment of Kinematics”).
The Davis Method

This is the most used model both in the literature and in many laboratories around the
world since it is implemented in most of the motion analysis softwares (Davis et al.
1991). It is based on the examination of 25 hips. The predictive equations use the
inter-ASIS (anterior superior iliac spine) distance, the anteroposterior distance
between the ASIS, marker radius, and leg length. Thus, the location of the hip
joint center is obtained in the pelvic coordinate system.
Fig. 2 Example of predictors

(pelvic depth, pelvic width,
and lower limb length) used in
regression equations to
localize the hip joint center
The Bell Method

In 1982, Tylkowski defined a method of prediction of the hip joint center in the
lateral plane, while Andriacchi predicted the position of the hip joint center in the
frontal plane (Tylkowski et al. 1982; Andriacchi and Strickland 1985). Bell et al.
used these two techniques to develop a new predictive method to localize the hip
joint center (Bell et al. 1989). The method was based on anterior-posterior
(AP) radiographs for children and adults; however, validation was only performed
using AP and lateral radiographs on dry pelvises of adults.
The Harrington Method

In 2007, Harrington suggested an image-based validation of previous predictive
methods (Davis, Bell, and motion analysis software) using MRI and a new predictive
method (Harrington et al. 2007). The new method is derived from 8 adults,
14 healthy children, and 10 children with CP. The hip joint center location in a
pelvic coordinate system was found by fitting a sphere to points identified on the
femoral head. The new predictive method was based on pelvic width, pelvic depth,
and leg length (Fig. 2).
8 A. Assi et al.
Functional Methods
The functional methods derive from the movement of a skeletal segment rela-
tively to the adjacent one. The subject is asked to perform a movement of the
joint, i.e., for the hip joint, a movement of flexion-extension and/or abduction-
adduction is performed. Since the hip is approximated as a ball and socket joint,
each marker attached to the moving segment (i.e., the thigh) moves on a sphere
surface where its center is the center of the hip joint. Thus, the joint center is
assumed to be the center of rotation (CoR) defined by the movement between
two adjacent segments. The first to describe this technique was Cappozzo in 1984
(Cappozzo 1984).
Several algorithms or mathematical techniques of computing the CoR exist; we
will be discussing the most common ones in the following section.
Sphere Fit Methods

This technique assumes that the CoR is stationary; this assumption can be true if one
segment (segment 1) is at rest. The markers on the adjacent segment (segment 2)
move on the surface of a sphere with specific radii around one common CoR. The
frequent approach uses the minimization of the sum of the squared Euclidean
distances between the sphere and the marker positions. A selected cost function
determines whether the optimal solution can be calculated exactly or only approx-
imately by successive iterative steps toward the optimal CoR. Some authors use the
least squares method which gives an exact CoR estimation (Pratt 1987; Gamage and
Lasenby 2002). An example is displayed in Fig. 3. Alternative approaches are
iterative (Halvorsen 2003). In the geometric method, an initial guess of the CoR is
required, whereas in the algebraic methods do not require a starting estimate.
The major disadvantage of these techniques is in the convergence to local minima
of the cost function and in their poor accuracy of estimating the CoR when a reduced
ROM is performed. Chang et al. (2007) proposed a new numerical technique of
sphere fit methods that can be used for reduced ROM.
Center Transformation Technique

The center transformation technique (CTT) assumes that at least three markers on the
moving segment are present; it is then possible to define a rigid-body transformation
(rotations and translations) which transforms a given reference marker at one frame
into another frame (Piazza et al. 2004). The appropriate transformation of these local
systems for all time frames into a common reference system enables the approxi-
mation of the joint center at a fixed position. Another approach, called the two-sided
approach that does not require the assumption of a stationary CoR, can be alterna-
tively used (Schwartz and Rozumalski 2005).
Score Technique
This algorithm is a continuation of the CTT method, with the assumption that the
coordinates of the CoR must remain constant relative to both segments, without
requiring the assumption that one segment remain at rest (Ehrig et al. 2006).
Fig. 3 Sphere fit method: trajectories of four different markers (each marker with a different color)
during calibration trials. The star points represent the center of the spheres. The black star point
represents the middle of the different points
Calibration Trial
While different movements can be performed especially in a ball and socket joint
(i.e., the hip or the glenohumeral joint), such as flexion-extension, abduction-adduc-
tion, internal-external rotation, or circumduction, Camomilla et al. in 2006 showed
that the movement that gives the most accurate results on calculating the hip joint
center using the functional technique is the star arc (Camomilla et al. 2006). This
calibration trial consisted of several flexion-extension/abduction-adduction move-
ments performed on vertical planes of different orientations, followed by a
circumduction movement.
Uncertainties Related to Low ROM

It has been shown that the errors on the localization of a joint center increase when
the range of motion (ROM) performed during the calibration method decreases.
In Ehrig et al.’s study, the approaches of CTT, score, and sphere fit were tested in a
simulation model, while noise was added to the markers (Ehrig et al. 2006). The
RMS between the calculated CoR and the exact one was calculated. RMS errors
decreased exponentially with increasing ROM. Theoretical accuracy on the position
of the CoR was within 1 cm using all approaches when the ROM increased beyond
20 . Accuracy was within 0.3 cm as long as the ROM of the joint was 45 or more.
Some of these simulations have taken into account the skin-marker movement which
is an additional source of uncertainty.
10 A. Assi et al.
Therefore, some authors prefer to assist the patients/subjects when they are
performing the functional calibration movement, in order to make sure that the
ROM is adequate for functional localization of the joint center (Peters et al. 2012).
Piazza in 2001 used a mechanical model to simulate the errors on the localization
of the hip joint center using functional methods (Piazza et al. 2001). Significant
increases in the magnitude of HJC location errors (4–9 mm) were noted when the
range of hip motion was reduced from 30 to 15 . The same result was found by
Camomilla et al. (2006). The accuracy of the HJC estimate improved, with an
increasing rate, as a function of the amplitude of the performed movements in the
hip.
Knee Joint Center
In the particular case of the knee, a center of rotation is usually calculated using the
predictive method, using external markers located on the condyles. The Davis
model, presented above, also predicts the knee joint center which is approximated
as being in the middle of the knee width (measured during clinical examination):
from the external to the internal condyle and defined relatively to the thigh marker
(Davis et al. 1991).
In the functional method, a knee axis is usually calculated that represents the
complexity of the flexion-extension movement. The knee flexion is a combination of
the femoral condyles rolling (rotation) over the tibial plateau and the posterior
gliding (translation) of the condyles along the plateau (Ramsey and Wretenberg
1999).
Usually, two mathematical methods exist to calculate the axis of rotation (AoR) of
the knee. The first method fits cylindrical arcs to the moving segment, while
assuming that the adjacent segment is at rest (Gamage and Lasenby 2002; Halvorsen
2003). The second method is based on the transformation techniques (CTT) exposed
above, where the helical axes technique is used based on the work of Woltring et al.
(1985). More recently, another algorithm for the localization of the AoR was
presented by Ehrig et al. (2007); the symmetrical axis of rotation approach deter-
mines a unique axis of rotation that considers the movement of two dynamic
segments simultaneously.
Ankle Joint Center
The most frequently used methods to localize the ankle joint center are predictive,
using external markers located on the malleoli. The Davis method (Davis et al. 1991)
is the most common. A similar strategy to the knee localization method is applied to
obtain the ankle joint center.
While the Davis model is the most commonly used for clinical gait analysis, the
foot is represented as a single segment, and only ankle joint motion is quantified. In
order to quantify the dynamic adaptability of the different foot segments, several
models have been described. The most commonly used model is the Oxford foot
model, where three segments are defined in the foot (hindfoot, forefoot, hallux) in
addition to the tibial segment (Stebbins et al. 2006). Other models described in the
literature use four segments, such as the Leardini (calcaneus, midfoot, first metatar-
sal, and the hallux) (Leardini et al. 1999b) and Jenken models (hindfoot, midfoot,
medial forefoot, lateral forefoot) (Jenkyn and Nicol 2007). Further information on
this topic can be found in the preceding chapter (▶ Variations of Marker-Sets and
Models for Standard Gait Analysis).
Glenohumeral Joint Center
Different predictive and functional methods exist to localize the glenohumeral joint
center. In the predictive Meskers’ method, a linear regression is used to predict the
glenohumeral joint center based on specific points on the scapula, the
acromioclavicular joint and the processus coracoideus (Meskers et al. 1997). This
method was elaborated by digitizing 36 sets of cadaver scapulae and adjacent
humeri. The functional methods are based on the movement of the humerus rela-
tively to the scapula or the thorax; the same algorithms as the one used for the hip
joint center can be applied. The same result, as for the hip joint center, was also
found regarding the low ROM when performed during the calibration trial;
Lempereur et al. showed that high amplitude of movement should be performed
(>60 ) in order to improve reliability when functional methods are used for the
localization of the glenohumeral joint center (Lempereur et al. 2011). Further
information on this topic can be found in the chapter on “▶ Upper Extremity Models
for Clinical Movement Analysis.”
Validation of the Joint Center Localization Methods
Several authors have attempted to assess the accuracy of both predictive and functional
methods by localizing the joint center obtained by medical imaging as a gold standard.
The technique consists of obtaining the joint center in 3D, while being expressed in the
local coordinate system of the adjacent segment. The latter is built based on the
external markers placed on the skin. The joint centers calculated through predictive
and functional methods would also be expressed in the same local coordinate system
of the adjacent segment. Thus, when all calculated joint centers are expressed in the
same coordinate system, distances from each joint center to the gold standard could be
calculated, and a comparison between methods could be performed.
X-Rays and Stereophotogrammetry
The method developed by Bell in 1989 was based on the use of digitized AP
radiographs with localization of specific bony landmarks on the radiograph as well
12 A. Assi et al.
as digitization of the center of a circle that matches the size of the femoral head (Bell
et al. 1989).
In 1990, Bell et al. were the first to use pairs of orthogonal radiographs (Bell et al.
1990); by knowing the exact distances between X-ray sources and film cassette
locations, it was possible to estimate in 3D the location of the bony landmarks and
the pelvic skin markers. The accuracy of HJC localization methods were thus
assessed for methods such as the functional method described by Cappozzo in
1984 but also the predictive methods described by Tylkowski and Andriacchi
(Tylkowski et al. 1982; Andriacchi and Strickland 1983; Cappozzo 1984).
In another study, Leardini et al. assessed the validity of functional and predictive
methods in calculating the HJC on 11 healthy adults (Leardini et al. 1999a). The
average root mean square (RMS) distance to the gold standard was 25–30 mm for
predictive methods and 13 mm for functional method.
The technique of stereoradiography has the major disadvantage of being irradiant
to the patient.
Magnetic Resonance Imaging
In a study lead by Harrington et al., MRI was used as the gold standard in obtaining
the hip joint center for a population of healthy adults, healthy children, and children
with cerebral palsy (Harrington et al. 2007). The validation of existing predictive
methods was assessed in addition to a new method presented by the authors.
In a study performed by Lempereur et al., the authors used MRI acquisition to
validate several functional methods for the localization of the glenohumeral JC
(Lempereur et al. 2010). This technique required the coverage of the scapula with
120 reflective markers in order to perform the matching between the surface of the
scapula obtained by motion analysis capture and MRI reconstruction. This technique
is time-consuming since it requires manual segmentation.
The major disadvantage of the MRI technique is the time of acquisition and the
time of image processing.
3D Ultrasound
The ultrasound technique was widely used in order to validate the JC localization
methods since it is not irradiant and easier to perform compared to the MRI
technique. However, the US method requires a calibration process in order to obtain
3D reconstructions of the JC. In a study performed by Peters et al., the authors
described the required calibration process in order to obtain 3D US reconstructions
(Peters et al. 2010). The repeatability of the technique was assessed as well as the
accuracy of the localization of a reference object within a water basin. The accuracy
was about 4 2 mm. After the validation of this technique, the same authors
performed different studies on the validation of both predictive and functional
HJC localization techniques in adults (Sangeux et al. 2011) and both typically
developing children and children with cerebral palsy (Peters et al. 2012). In the study
on adults, it was shown that the functional method and more precisely the geometric
sphere fitting method was the most precise in localizing the HJC (mean absolute
distance error of about 15 mm) followed by the Harrington predictive method. In the
study on TD and CP children, the Harrington method was the closest method to the
3D US technique (14 8 mm), whereas the functional techniques performed much
worse (22–33 mm). It should be noted that the functional calibration trials of the hip
had been assisted by an external operator.
The 3D ultrasound technique has also been used for the localization of the
glenohumeral joint (Lempereur et al. 2013).
Low-Dose Biplanar X-Rays
More recently, the low-dose biplanar X-ray technique (Dubousset et al. 2005;
Humbert et al. 2009; Chaibi et al. 2012) was applied in order to validate JC
localization techniques. The EOS system was used as an image-based reference.
The localization of external markers was reliable within 0.15 mm for trained
operators, and the mean accuracy for HJC localization was 2.9 1.3 mm (Pillet
et al. 2014), even less than the values obtained by the 3D US method.
The EOS system allows the acquisition in the same image of external markers,
skeletal segments, and joint centers. Thus, a joint center can be located directly in the
local coordinate system of the adjacent segment, based on the location of the external
markers (Fig. 4).
The EOS system was used to compare the accuracy of several predictive and
functional techniques in localizing the HJC in healthy adults (Sangeux et al. 2014).
Different scenarios were applied when functional methods were assessed: different
algorithms, different ranges of motion of the hips (<30 and >30 ), and self-
performance or assisted performance. The best results were obtained for the comfortable
ROM when they were self-performed by the subjects. The best method was the
functional geometrical sphere fitting method which localized the hips 1.1 cm from the
EOS reference. It was shown that the worst results were obtained for functional methods
when the ROM was reduced. In the latter case, the best method was the Harrington
predictive method which localizes the HJC at 1.7 cm from the EOS reference.
In a more recent study, the EOS system was used to evaluate the accuracy of both
predictive and functional methods in TD and CP children (Assi et al. 2016). Contrarily
to the findings in adults, the functional methods performed much worse (>60 mm)
compared to the predictive methods, where the Harrington method showed the best
results (18 9 mm). The authors explained the differences in results between adults
and children as being due to the shorter length of the thigh segment in children, which
could increase the noise when the algorithms of functional methods are applied to
locate the CoR. It was also shown that children with CP performed significantly lower
ROM of hip movements during calibration compared to TD children. However,
average ROM in both groups was >30 , and the ROM was not a confounding factor
on the errors on the HJC calculated by the functional methods.
14 A. Assi et al.
Fig. 4 Frontal and lateral

X-rays of the lower limbs
obtained by low-dose biplanar
X-rays, with the external
markers fixed on the skin as
well as the 3D reconstruction
of the femur and the hip joint
center, expressed in the local
coordinate system of the
pelvis
In a study performed by Lempereur et al., different functional methods as well as

the 3D ultrasound technique were compared in localizing the glenohumeral joint
center relatively to the one obtained by the 3D EOS reconstruction, considered as the
reference (Lempereur et al. 2013). The 3D ultrasound technique placed the
glenohumeral joint center at 14 mm from the EOS image-based reference, while
functional methods varied from 15.4 mm (the helical axis method) to 34 mm using
iterative methods (Halvorsen 2003).
Effect of Errors on JC Localization
Errors on Kinematics and Kinetics
Misplacement errors of joint centers can distort kinematics and kinetics of the hip
and knee in the case of gait analysis, since the thigh local coordinate system is
affected. The effects of hip JC misplacement on gait analysis were studied by Stagni
et al. (2000). The latter found an error on the joint moment that can reach 22% of
flexion-extension and 15% on abduction-adduction with a delay of 25% of the

flexion to extension timing of the stride duration.
In a study performed by Kiernan et al., the authors assessed the clinical agreement
of the Bell, Davis, and Orthotrak methods in localizing the HJC compared to the
Harrington method as a gold standard (Kiernan et al. 2015). This was applied on
18 healthy children. Kinematics, kinetics, Gait Profile Score, and Gait Deviation
Index were calculated. The authors found that errors, when the Davis or Orthotrak
methods were used, are clinically meaningful especially on kinetics.
The results on the glenohumeral joint were different from those obtained on the
hip. Lempereur et al. showed that misplacement of the glenohumeral joint center will
propagate to the kinematics of the shoulder, but errors do not exceed 4.8 on the
elevation angle during shoulder flexion and 4.3 on the elevation plane during
shoulder abduction (Lempereur et al. 2014). The authors related this difference of
propagated errors between the movements of the arm and the thigh to the difference
in mass between the two segments.
Errors on Musculoskeletal Simulations
Inaccurate localization of JC can also influence the results obtained in musculoskel-

etal simulations. In a study performed by Scheys et al., moment-arm and muscle-
tendon lengths were computed using three kinds of musculoskeletal models: a
personalized model based on MRI data, an isotropic rescaled generic model, and
an anisotropic rescaled generic model (Scheys et al. 2008). Different hip joint center
techniques were used in each of the models. These simulations were applied on the
gait of an asymptomatic adult. The generic model simulations showed large offsets
of moment arm and muscle-tendon lengths when compared to the personalized
model for most of the major muscles of the lower limbs.
Future Directions
Joint center localization is essential in order to obtain anatomically accurate kine-

matics and kinetics. Joint center localization techniques could be either predictive or
functional. While predictive techniques are based on regression equations that use
anthropometric measurements, the functional techniques require the performance by
the subject of ranges of motion in the joint of interest in order to calculate the center
of rotation between the two adjacent segments of the joint. Several authors have
validated these techniques in children and adults by comparing the location of the
joint center obtained by these methods to the joint center obtained by 3D medical
imaging. The most frequently used medical imaging system in the validation
processes were stereoradiography, CT scan, MRI, 3D ultrasound, and more recently
low-dose biplanar X-rays.
It was shown that the functional methods were more accurate in locating the joint
center compared to the predictive ones in the adult population, which was not the
16 A. Assi et al.
case in children. This could be due to the shorter segment in children which renders
the markers in movement closer to the joint center, thus increasing the noise during
the calculation of the joint center. Moreover, the amount of ROM performed during
functional calibration should be within certain limits: a low ROM could not be
sufficient for the calibration process and a high ROM could induce more soft tissue
artifacts.
The errors on the localization of the joint center have been shown to directly
affect both kinematic and kinetic calculation. They also affect the computation of
muscle lever arms when running musculoskeletal simulations.
Correction of 3D Positioning of the JC
Several authors have shown the deviation of the joint center calculated by either
predictive or functional methods compared to the exact joint center obtained by 3D
medical imaging in each direction: anterior-posterior, medial-lateral, and superior-
inferior. A first solution could be the correction of this location prior to the calcu-
lation of kinematics or kinetics or computation of musculoskeletal simulation
(Sangeux et al. 2014; Assi et al. 2016).
Registration Techniques for the Use of Exact Joint Center Location
It was shown in the validation methods that both predictive and functional methods
localize the joint center at 11–18 mm from its exact location. In an ideal setting,
when a medical imaging tool is present in the same laboratory along with the motion
capture equipment, the exact location of the joint center should be used in the
calculation process of kinematics and kinetics, even for musculoskeletal simulations.
Markers would be placed on the patient, and an image acquisition, such as the
EOS biplanar X-rays, will be performed in order to obtain the exact 3D location of
the joint center in the local coordinate system of the adjacent segment (i.e., location
of the hip joint center expressed in the local coordinate system of the pelvis,
glenohumeral joint center expressed in the local coordinate system of the scapula
or thorax). These 3D coordinates would be used after motion analysis acquisition for
either kinematic/kinetic calculations or musculoskeletal model simulations.
Estimation from External Information
Another solution could be to optimize regression equations of joint center localiza-

tion techniques. Since the validation methods showed different results depending on
the population type, new regression equations could be tailored to each population
(i.e., children and adults). The new biplanar low-dose X-ray technique could allow
the acquisition in the same image of the 3D reconstruction of both the external
markers and the joint center. Thus, a large cohort of subjects/patients of different age
Fig. 5 Estimation of the hip joint center using the 3D reconstruction of the skin and the skeleton
and based on morphological and barycentermetric predictors (Nerot et al. 2016)
intervals could allow the attainment of age-specific regression equations based on

anthropometric measurements.
The possibility to get both the external envelope and the internal skeleton (Nérot
et al. 2015a, b) opens the way for a large-scale analysis and improvement on the
regression equations for joint center localization by combining morphological and
barycentermetric predictors (Fig. 5).
References
Andriacchi T, Strickland A (1985) Gait analysis as a tool to assess joint kinetics. In: Berme N, Engin
A, Correia Da Silva K, (eds). Biomechanics of Normal and Pathological Human Articulating
Joints. Martinus Nijhoff, Dordrecht: NATO SI Series. pp. 83–102.
Assi A, Sauret C, Massaad A, Bakouny Z, Pillet H, Skalli W, et al (2016) Validation of
hip joint center localization methods during gait analysis using 3D EOS imaging in
typically developing and cerebral palsy children. Gait Posture [Internet] 42:30–5. Available
18 A. Assi et al.
from http://dx.doi.org/10.1016/j.gaitpost.2016.04.028%5Cn, http://linkinghub.elsevier.com/

retrieve/pii/S0966636216300455%5Cn, http://dx.doi.org/10.1016/j.gaitpost.2015.06.089
Bell AL, Brand RA, Pedersen DR (1989) Prediction of hip joint centre location from external
landmarks. Hum Mov Sci [Internet] 8(1):3–16. Available from: http://www.sciencedirect.com/
science/article/pii/0167945789900201. [cited 2015 Oct 27]
Bell AL, Pedersen DR, Brand RA (1990) A comparison of the accuracy of several hip center. J
Biomech 23:6–8
Camomilla V, Cereatti A, Vannozzi G, Cappozzo A (2006) An optimized protocol for hip joint
centre determination using the functional method. J Biomech [Internet] 39(6):1096–1106.
Available from: http://linkinghub.elsevier.com/retrieve/pii/S0021929005001004
Cappozzo A (1984) Gait analysis methodology. Hum Mov Sci 3:27–50
Cappozzo A, Catani F, Della Croce U, Leardini A (1995) Position and orientation in space
of bones during movement: anatomical frame definition and determination. Clin Biomech
[Internet] 10(4):171–178. Available from: http://www.sciencedirect.com/science/article/pii/
026800339591394T
Cappozzo A, Della Croce U, Leardini A, Chiari L (2005) Human movement analysis using
stereophotogrammetry. Part 1: theoretical background. Gait Posture [Internet] 21(2):186–196.
Available from: http://www.sciencedirect.com/science/article/pii/S0966636204000256. [cited
2015 Nov 4]
Chaibi Y, Cresson T, Aubert B, Hausselle J, Neyret P, Hauger O et al (2012) Fast 3D reconstruction
of the lower limb using a parametric model and statistical inferences and clinical measurements
calculation from biplanar X-rays. Comput Methods Biomech Biomed Eng 15(5):457–466
Davis RB, Ounpuu S, Tyburski D, Gage JR (1991) A gait analysis data collection and reduction
technique. Hum Mov Sci 10(5):575–587
Dubousset J, Charpak G, Dorion I, Skalli W, Lavaste F, Deguise J et al (2005) A new 2D and 3D
imaging approach to musculoskeletal physiology and pathology with low-dose radiation and the
standing position: the EOS system. Bull Acad Natl Med [Internet]. 189(2):287–297. Available
from: http://www.ncbi.nlm.nih.gov/pubmed/16114859
the centre of rotation of ball joints. J Biomech [Internet] 39(15):2798–2809. Available from:
http://linkinghub.elsevier.com/retrieve/pii/S002192900500446X
functional joint axes. J Biomech 40(10):2150–2157
Gage JR (1993) Gait analysis. An essential tool in the treatment of cerebral palsy. Clin Orthop Relat
Res [Internet] (288):126–134. Available from: http://www.ncbi.nlm.nih.gov/pubmed/8458125
Gamage SSHU, Lasenby J (2002) New least squares solutions for estimating the average centre of
rotation and the axis of rotation. J Biomech 35(1):87–93
Halvorsen K (2003) Bias compensated least squares estimate of the center of rotation. J Biomech
S0021929003000708. [cited 2016 Jun 21]
Harrington ME, Zavatsky AB, Lawson SEM, Yuan Z, Theologis TN (2007) Prediction of the hip
joint centre in adults, children, and patients with cerebral palsy based on magnetic resonance
imaging. J Biomech [Internet] 40(3):595–602. Available from: http://linkinghub.elsevier.com/
retrieve/pii/S0021929006000583
Humbert L, De Guise JA, Aubert B, Godbout B, Skalli W (2009) 3D reconstruction of the spine
from biplanar X-rays using parametric models based on transversal and longitudinal inferences.
Med Eng Phys 31(6):681–687
Jenkyn TR, Nicol AC (2007) A multi-segment kinematic model of the foot with a novel definition
of forefoot motion for use in clinical gait analysis during walking. J Biomech 40(14):3271–3278
Kiernan D, Malone A, O’Brien T, Simms CK (2015) The clinical impact of hip joint centre
regression equation error on kinematics and kinetics during paediatric gait. Gait Posture
S0966636214007255
Leardini A, Cappozzo A, Catani F, Toksvig-Larsen S, Petitto A, Sforza V et al (1999a) Validation of

a functional method for the estimation of hip joint centre location. J Biomech 32(1):99–103
Leardini A, O’Connor JJ, Catani F, Giannini S (1999b) Kinematics of the human ankle complex in
passive flexion; a single degree of freedom system. J Biomech 32(2):111–118
Lempereur M, Leboeuf F, Brochard S, Rousset J, Burdin V, Rémy-Néris O (2010) In vivo
estimation of the glenohumeral joint centre by functional methods: accuracy and repeatability
assessment. J Biomech 43(2):370–374
Lempereur M, Brochard S, Rémy-Néris O (2011) Repeatability assessment of functional methods to
estimate the glenohumeral joint centre. Comput Methods Biomech Biomed Eng 5842:1–6
Lempereur M, Kostur L, Leboucher J, Brochard S, Rémy-Néris O (2013) 3D freehand ultrasound
to estimate the glenohumeral rotation centre. Comput Methods Biomech Biomed Eng [Internet]
16(Suppl 1):214–215. Available from: http://www.ncbi.nlm.nih.gov/pubmed/23923914
Lempereur M, Leboeuf F, Brochard S, Rémy-Néris O (2014) Effects of glenohumeral joint centre
mislocation on shoulder kinematics and kinetics. Comput Methods Biomech Biomed Eng [Inter-
net] 17(Suppl 1):130–131. Available from: http://www.ncbi.nlm.nih.gov/pubmed/25074199
Meskers CGM, Van Der Helm FCT, Rozendaal LA, Rozing PM (1997) In vivo estimation of the
glenohumeral joint rotation center from scapular bony landmarks by linear regression. J
Biomech 31(1):93–96
Nérot A, Choisne J, Amabile C, Travert C, Pillet H, Wang X, et al (2015a) A 3D reconstruction
method of the body envelope from biplanar X-rays: evaluation of its accuracy and reliability. J
Biomech [Internet] 48(16):4322–4326. Available from: http://dx.doi.org/10.1016/j.
Nérot A, Wang X, Pillet H, Skalli W (2015b) Estimation of hip joint center from the external body
shape: a preliminary study. Comput Methods Biomech Biomed Eng [Internet] 5842:1–2.
Available from: http://www.tandfonline.com/doi/full/10.1080/10255842.2015.1069603
of the hip joint centre. Gait Posture [Internet] 31(4):530–2. Available from: http://linkinghub.
elsevier.com/retrieve/pii/S0966636210000299
Peters A, Baker R, Morris ME, Sangeux M (2012) A comparison of hip joint centre localisation
techniques with 3-DUS for clinical gait analysis in children with cerebral palsy. Gait Posture
[Internet] 36(2):282–286. Available from: http://linkinghub.elsevier.com/retrieve/pii/
S0966636212000999
Piazza SJ, Okita N, Cavanagh PR (2001) Accuracy of the functional method of hip joint center
location: effects of limited motion and varied implementation. J Biomech 34(7):967–973
Piazza SJ, Erdemir A, Okita N, Cavanagh PR (2004) Assessment of the functional method of hip
joint center location subject to reduced range of hip motion. J Biomech 37:349–356
Pillet H, Sangeux M, Hausselle J, El Rachkidi R, Skalli W (2014) A reference method for the
evaluation of femoral head joint center location technique based on external markers. Gait
Posture [Internet] 39(1):655–658. Available from: http://linkinghub.elsevier.com/retrieve/pii/
S096663621300578X
Pratt V (1987) Direct least-squares fitting of algebraic surfaces. Comput Graph (ACM) 21:145–152
Ramsey DK, Wretenberg PF (1999) Biomechanics of the knee: methodological considerations in
the in vivo kinematic analysis of the tibiofemoral and patellofemoral joint. Clin Biomech 14
(9):595–611
Sangeux M, Peters A, Baker R (2011) Hip joint centre localization: evaluation on normal subjects in
the context of gait analysis. Gait Posture [Internet] 34(3):324–328. Available from: http://dx.doi.
org/10.1016/j.gaitpost.2011.05.019
in gait analysis? Gait Posture [Internet] 40(1):20–25. Available from: http://linkinghub.elsevier.
com/retrieve/pii/S0966636214000642
Scheys L, Spaepen A, Suetens P, Jonkers I (2008) Calculated moment-arm and muscle-tendon
lengths during gait differ substantially using MR based versus rescaled generic lower-limb
musculoskeletal models. Gait Posture 28(4):640–648
20 A. Assi et al.
data. J Biomech [Internet] 38(1):107–116. Available from: http://linkinghub.elsevier.com/
retrieve/pii/S002192900400137X
Stagni R, Leardini A, Cappozzo A, Grazia Benedetti M, Cappello A (2000) Effects of hip joint
centre mislocation on gait analysis results. J Biomech 33(11):1479–1487
Stebbins J, Harrington M, Thompson N, Zavatsky A, Theologis T (2006) Repeatability of a model
for measuring multi-segment foot kinematics in children. Gait Posture. 23(4):401–410
Tylkowski C, Simon S, Mansour J (1982) Internal rotation gait in spastic cerebral palsy. In: Nelson
JP (ed) Proceedings of the 10th Open Scientific Meeting of the Hip Society. C. V. Mosby, St
Louis, pp 89–125
Woltring H, Huiskes R, de Lange A, Veldpaus F (1985) Finite centroid and helical axis estimation
from noisy landmark measurements in the study of human joint kinematics. J Biomech 18
(5):379–389
Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D et al (2002) ISB recommendation
on definitions of joint coordinate system of various joints for the reporting of human joint
motion – Part I: ankle, hip, and spine. J Biomech [Internet] 35(4):543–548. Available from:
http://www.sciencedirect.com/science/article/pii/S0021929001002226
Wu G, Van Der Helm FCT, Veeger HEJ, Makhsous M, Van Roy P, Anglin C et al (2005) ISB
recommendation on definitions of joint coordinate systems of various joints for the reporting of
human joint motion – Part II: shoulder, elbow, wrist and hand. J Biomech 38(5):981–992
Kinematic Foot Models for Instrumented
Gait Analysis
Alberto Leardini and Paolo Caravaggi
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Validation and Application of Foot Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Validation Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Musculoskeletal Multi-Segment Foot Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Kinetic Analysis Including Foot Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Abstract
In many clinical and biomechanical contexts of human motion analysis the model
assumption of the foot as a single rigid segment is no longer acceptable. This has
given rise to a large number of proposals for multi-segment foot models. The
relevant experimental and analytical techniques differ for many aspects: the
number of foot segments; the bony landmarks involved; the type of marker
clusters; the definition of the anatomical frames; and the convention for the
calculation of joint rotations. Different definitions of neutral reference posture
have also been adopted, along with their utilization to offset kinematic data.
Following previous partial review papers, the present chapter aims at introducing
the current methodological studies for in vivo analysis of multi-segment foot
kinematics. The survey has found more than 30 different techniques; however,
only a limited number of these have reported convincing validation activities and
A. Leardini (*) • P. Caravaggi

Movement Analysis Laboratory and Functional-Clinical Evaluation of Prostheses, Istituto
Ortopedico Rizzoli, Bologna, Italy
e-mail: leardini@ior.it; paolo.caravaggi@ior.it; pacarasta@gmail.com

DOI 10.1007/978-3-319-30808-1_28-1
2 A. Leardini and P. Caravaggi
have been exploited in a clinical context. A number of papers have also compared
the experimental performance of different multi-segment foot models and
highlighted the main advantages and disadvantages of each of them. Important
biomechanical applications of musculoskeletal models for reliable estimation of
biomechanical parameters are also discussed. In addition, we report on the feasibil-
ity and limitations of kinetic analysis applied to multi-segment foot models from
ground reaction force data. The chapter ends with recommendations both for the
selection of a most suitable technique from those already available, as well as for the
design of an original one suitable to address the needs of the specific application.
Keywords
Foot joint mobility • 3D joint motion • Multi-segment kinematics • Ankle
complex • Chopart joint • Lisfranc joint • Metatarsophalangeal joint • Foot arches •
Stereophotogrammetry • Marker clusters • Skin motion artifact
Introduction
In standard clinical gait analysis body segments are tracked in three-dimensions

(3D) by a stereophotogrammetric system, while their relative positions are calculated
to assess patterns of joint rotations during the execution of motor activities. Full-body
gait analysis requires several passive reflective markers to be fixated to the trunk,
pelvis, thigh, shank, and foot. Kinematics of segments are assessed with respect to
the laboratory co-ordinate system, i.e., the absolute motion, and with respect to any
adjoining segments for calculation of relative joint rotations. Frequently, motion of
the trunk and pelvis is reported in the laboratory reference frame, in case with respect
to the line of progression, together with hip, knee, and ankle joint motion. The
moment of the external forces can also be calculated at these joints as the product of
the joint center distance and the ground reaction force recorded by the force plate.
Together with the spatiotemporal parameters (e.g., walking speed and stride length),
these are the standard kinematic parameters necessary to assess, and characterize,
most of the pathological conditions investigated in gait analysis. The use of skin-
based reflective markers to track body segments comprised of single bones (e.g., the
femur or the humerus) can result in a fairly accurate representation of their real
motion. From a kinematic prospective, these segments can reasonably be assumed to
move as rigid bodies, thus the position of at least three nonaligned points is required
for their motion to be tracked in the 3D space. In gait analysis, these points are mostly
palpable bony landmarks, the temporal position of which is tracked via reflective
markers attached to the skin. With skin markers, however, the rigid body assumption
is violated and accuracy is lower when multiple small bones are connected in a
reduced volume (Nester et al. 2010). On the other hand, prospects to continuously
track different segments are of particular relevance in the evaluation of pathologies
regarding the shank and foot. Although the tibia and fibula bones present very small
relative motion and thus can be considered reasonably as a single segment for
Kinematic Foot Models for Instrumented Gait Analysis 3
kinematic analysis (Arndt et al. 2007), the foot is made up of 26 bones and several
joints connecting them. Therefore, standard kinematic protocols based on three
markers only appear inadequate in describing the complex foot biomechanics. In
addition, foot bones are rather small and some of them, e.g., the talus, have no clear
palpable landmarks, thus making it very difficult for those to be tracked in 3D space.
The importance of multi-segment foot models (MFMs) rather than single-
segment foot tracking has been largely discussed in the literature. Benedetti et al.
(2011) demonstrated the value of 3D motion analysis of the ankle joint in the clinical
context. De Ridder et al. (2015) compared results from a MFM (“Ghent” in Table 1,
by De Mits et al. 2012) and a single-segment foot model and showed the value of
distal factors in chronic ankle instability, in particular the deviation in kinematics at
the midfoot, which simply cannot be detected with a rigid foot model. Pothrat et al.
(2015) reported significant differences and even opposite results for the same vari-
ables when the multi-segment Oxford Foot Model (OFM, see Table 1) and the Plug-
in-Gait (modeling the foot as a single segment) were used to characterize normal and
flat feet, concluding that the type of foot model strongly affects the measured ankle
joint kinematics. Dixon et al. (2012) performed a similar study, i.e., OFM versus
Plug-in-Gait albeit on kinetic data, and revealed that the latter overestimated ankle
power by about 40% with respect to OFM and neglected the important midfoot
loading. The authors of these papers shared the same recommendation of using
caution when foot and ankle kinematics are measured with a single-segment foot
model. Interestingly, the value of multi-segment foot kinematic analysis has been
praised also in studies related to the analysis of more proximal lower limb joints
(Arnold et al. 2014b). In 2003, a special technical session of the Gait and Clinical
Movement Analysis Society (GCMAS) agreed on and recommended that shank,
rearfoot, forefoot, and hallux are clinically meaningful foot segments to be tracked.
These segments are in fact to be found in most of the multi-segment foot techniques
reported in the literature.
Many basic foot biomechanic studies and clinical investigations employing
various MFMs can be found in the literature. A number of relevant review papers
have also been published, which represent valuable sources for an overview of foot
modeling in kinematic analysis. Rankine et al. (2008) reported first a systematic
analysis of 25 papers on foot kinematic modeling, thoroughly classified in terms of
number of bony segments and joint rotations. All major technical and exploitation
related issues were discussed systematically. Later, Deschamps et al. (2011) reported
and assessed many of these techniques in relation to their exploitation in the clinical
context. It was shown that whereas many foot joint rotations can be tracked in a
consistent and repeatable way, some measures are still critical, and several of these
techniques have yet to be used to address clinical problems. According to Bishop
et al. (2012) this is the consequence of poorly described or flawed methodologies,
preventing the readers from obtaining the same algorithms and programs to replicate
the analysis. A minimum of five reporting standards were proposed in this paper; this
aimed at guaranteeing full access to the most relevant modeling concepts and at
providing a common platform for sharing and comparing foot kinematic data and as
to improve their interpretation and usability. The association between foot posture
Table 1 Papers on multi-segment foot techniques and models (i.e., methodological studies).
The column Number of segments counts all foot and shank segments. The model name is indicated
when it was recognized somehow in the following literature or cited frequently in that way. Some of
these studies were taken from a previous review paper (Rankine et al. 2008). Papers that reported
further assessments of the technique/model are cited in the last column. Models and marker sets
designed for bone pin analysis are excluded from this Table
Following papers, with
Number Number Model name developments and
Authors Year of segments of subjects (best known as) technical assessments
Scott and 1991 3 3
Winter
Kepple et al. 1990 3 5
Moseley et al. 1996 3 14
Kidder et al. 1996 5 1 Milwaukee Myers et al. (2004),
Foot Model Long et al. (2010)
Cornwall and 1999 3 43 Cornwall I Cornwall and McPoil
McPoil (1999b),
Cornwall and McPoil
(2002)
Woodburn 1999 3 10 Woodburn I
et al.
Rattanaprasert 1999 5 10 Rattanaprasert Hetsroni et al. (2011)
et al.
Leardini et al. 1999 6 9 Leardini Foot
Model I
Wu et al. 2000 4 10 Atkinson et al. (2010)
Hunt et al. 2001 4 18
Carson et al. 2001 5 1 Oxford Foot Stebbins et al. (2006),
Model (OFM) Curtis et al. (2009),
Levinger et al. (2010),
Wright et al. (2011),
van Hoeve et al.
(2015), Carty et al.
(2015), Lucareli et al.
(2016), Milner and
Brindle (2016),
Halstead et al. (2016)
Arampatzis 2002 7 6
et al.
MacWilliams 2003 10 18 Kinfoot
et al.
Hwang et al. 2004 10 5
Davis et al. 2006 3 1 Shriners Maurer et al. (2013)
Hospital for Saraswat et al. (2013)
Children
Greenville Foot
Model (SHCG)
Pohl et al. 2006 12 3
Kitaoka et al. 2006 4 20
(continued)
Table 1 (continued)
Following papers, with
Number Number Model name developments and
Authors Year of segments of subjects (best known as) technical assessments
Rao et al. 2006 4 10
Simon et al. 2006 11 10 Heidelberg Kalkum et al. (2016)
Foot
Measurement
Method l
Tome et al. 2006 5 14
Jenkyn and 2007 6 12 Jenkyn et al. 2010
Nicol
Leardini et al. 2007 5 10 Rizzoli Foot Caravaggi et al. (2011),
Model (RFM) Deschamps et al.
(2012a, b), Arnold
et al. (2013), Portinaro
et al. (2014), Van den
Herrewegen et al.
(2014)
Wolf et al. 2008a 4 6
Sawacha et al. 2009 4 10
Cobb et al. 2009 4 11
Hyslop et al. 2010 6 9
Oosterwaal 2011 26 25 Glasgow- Oosterwaal et al.
et al. Maastricht foot (2016)
model
Bruening et al. 2012a 4 10 Bruening et al. (2012b)
De Mits et al. 2012 6 10 Ghent Foot
Model
Saraswat et al. 2012 4 15 Saraswat Saraswat et al. (2013)
Bishop et al. 2013 4 18
Nester et al. 2014 6 100 Salford Foot
Model
Seo et al. 2014 5 20
Souza et al. 2014 3 10
and lower limb kinematics has been the objective of another interesting review
analysis of twelve papers (Buldt et al. 2013). Evidence was found for increased
frontal plane motion of the rearfoot during walking in individuals with pes planus.
The latest review thus far by Novak et al. (2014) has highlighted the strengths and
weaknesses of the most widely used and known MFMs, including an insight on their
kinetic analyses.
While joint rotations have been thoroughly addressed in the literature, joint
translations have been studied and discussed very rarely: generally these are within
2 mm (Bruening et al. 2012a) in any anatomical direction. Because this is in the
order of magnitude of skin motion artifact, this topic would not be further discussed
in this chapter. For foot joint kinematics reconstruction, also the so-called “global
Table 2 Multi-segment foot models most used in clinical context. For each model (first column)
the relevant clinical papers are reported (second column)
Model Clinical papers
Milwaukee Foot Model Khazzam et al. (2007), Ness et al. (2008), Canseco et al. (2008),
(1996) Marks et al. (2009), Brodsky et al. (2009), Canseco et al.
(2009), Graff et al. (2010), Canseco et al. (2012), Krzak et al.
(2015)
Oxford Foot Model (2001) Theologis et al. (2003), Woodburn et al. (2004), Turner et al.
(2006), Turner and Woodburn (2008), Alonso-Vázquez et al.
(2009), Wang et al. (2010), Deschamps et al. (2010), Stebbins
et al. (2010), Bartonet al. (2011a, b, c), Hösl et al. (2014),
Merker et al. (2015)
Heidelberg Foot Houck et al. (2009), Twomey et al. (2010), Dubbeldam et al.
Measurement Method (2006) (2013)
Rao et al. (2006) Nawoczenski et al. (2008), Neville et al. (2009), Rao et al.
(2009)
Rizzoli Foot Model (2007 and Chang et al. (2008), Deschamps et al. (2013), Portinaro et al.
2014) (2014), Chang et al. (2014), Deschamps et al. (2016), Arnold
et al. (2014a, b), Lin et al. (2013), Hsu et al. (2014), Kelly et al.
(2014), Deschamps et al. (2016)
optimization” has been used recently (Arnold et al. 2013; Bishop et al. 2016). This
basically entails with an iterative search of the best estimation of foot segment
position and orientation, all together also called “pose”. This procedure starts from
skin marker trajectories, but the optimal poses must be compatible also with pre-
determined kinematic models for all the joints, i.e. global, this according to an
original technique for the lower limbs (Lu and O’Connor 1999).
The present chapter aims at introducing the current full series of methodological
studies on this topic, in order to provide the basic knowledge for either the selection
or the design of the most appropriate technique, according to the specific populations
and hypotheses of the foot kinematic study to be performed.
State of the Art
An extensive survey of the currently available multi-segment foot techniques and

models is reported in Table 1. Several differences can be found between multi-
segment foot techniques in the following factors:
– Foot segments
– Bony landmarks
– Type of marker clusters
– Definition of the anatomical frames
– Joint convention – including 2D versus 3D measurements
– Neutral reference posture
– Offsets
The major difference between MFMs is found in the number and selection of foot
segments (Fig. 1). While tibia, rearfoot, and forefoot are tracked by most techniques,
the hallux – or the first metatarso-phalangeal joint – is seldom tracked, and the
midfoot is tracked only by few models (MacWilliams et al. 2003; Leardini et al.
2007; Rouhani et al. 2011; Portinaro et al. 2014). Medial and lateral forefoot sub-
divisions have also been proposed (MacWilliams et al. 2003; Hwang et al. 2004;
Buczek et al. 2006; Rouhani et al. 2011). The current models available include up to
12 segments (Table 1); even a 26 segment foot model has been proposed
(Oosterwaal et al. 2011, 2016), but its application is limited to advanced musculo-
skeletal modeling studies. The number and selection of foot segments to be tracked,
somehow the resolution of the model, is usually defined according to the field of
application, the clinical interest, but also to the number, quality, and location of
available cameras of the stereophotogrammetric system. While kinematic analysis of
foot segments has been devised mostly for barefoot gait analysis, a number of
techniques were explicitly designed for the analysis of shod feet (Wolf et al.
2008b; Cobb et al. 2009; Shultz et al. 2011b; Bishop et al. 2013). Moreover, the
effect of foot and ankle orthoses has been investigated by established models (Lin
et al. 2013; Leardini et al. 2014). The overall results, in terms of patterns of foot joint
kinematics, can be confusing and difficult to interpret because of the differences
mentioned above. Also, the varying populations analyzed, as highlighted in Table 1,
in terms also of physical status, size, age, gender, etc., make it difficult to compare
data across different studies.
The process to include and track a segment within a MFM for kinematic analysis
requires a profound knowledge of foot biomechanics and of the limits and accuracy
of the measuring system. For example, the actual joint motion to be recorded should
be much larger than the accuracy of the stereophotogrammetric instrumentation used
for the analysis and than any other source of error (particularly, the skin motion
artifact). In addition to the known large rotations occurring at the tibiotalar and
metatarso-phalangeal joints, and between metatarsus and calcaneus, in vivo and
in vitro studies have demonstrated that significant and consistent rotations are
experienced in normal feet also at the Chopart (talo-calcaneo-navicular and
calcaneo-cuboid joints) and Lisfranc joints (tarso-metatarsal joints). These studies
have confirmed that midfoot motion during gait is significant and its assessment
should be included in relevant foot kinematic studies. The subtalar joint is also
subjected to relatively large motion; however, this is very difficult to track in vivo
with skin-based markers.
Both skin-based and plate-mounted marker clusters (Leardini et al. 1999; Carson
et al. 2001; Houck et al. 2006; Hyslop et al. 2010; Nester et al. 2014; Raychoudhury
et al. 2014; Souza et al. 2014; Buldt et al. 2015) on relevant foot and shank bony
landmarks have been used to track foot segments. The differences between skin-
markers and plate-mounted markers in measured joint motion were found to be small
(Nester et al. 2007a). The markers used for motion analysis are usually passive, i.e.,
reflecting IR (infrared) light emitted by LEDS embedded in the motion cameras, or
active, i.e., emitting IR light. While the latter usually provide a more accurate 3D
location, they also require a wired external power which can result in uncomfortable
Fig. 1 Diagrammatic representation of the foot segment subdivisions (different grey tones) for the
main MFMs
setups also restraining the movement of the subject. The number of markers
used in MFM can be as high as 35, as in Oosterwaal et al. (2011, 2016) and
Raychoudhury et al. (2014). A compromise must always be found between the
required degrees of freedom of the model, which is related also to the number of
segments tracked in 2D or 3D, and the number, quality, and location of the available
cameras. These are arranged usually to collect motion data for other anatomical
districts and motor tasks in the same laboratory, and therefore compromising layouts
must be found, as explicitly discussed for one widely used MFM (Leardini et al.
2007).
As mentioned above, at least three markers need to be fixated to each segment for
a complete 3D representation of its motion. This setup is technically suitable for
establishing a local reference frame on each segment and for calculation of triplanar
joint rotations using the Euler or the Joint-Coordinate-System convention (Grood
and Suntay 1983) (see typical results in Fig. 2). Anatomical landmarks are necessary
to establish anatomical based reference frames. However, the paucity of bony
landmarks and the small size of several foot bones limit the application of the
three-marker tracking for foot segments kinematics. While most techniques for the
kinematic analysis of foot segments use the 3D approach, i.e., three independent
rotations about three different axes, 2D projection angles can also be used to measure
relative rotations of a joint, with respect to anatomical planes. In the latter, line
segments determined by the position of two markers are projected at each time
sample onto an anatomical or other relevant planes, for the planar rotation to be
calculated during motion (Simon et al. 2006; Leardini et al. 2007; Portinaro et al.
2014). 2D planar angles have been largely used to track motion of metatarsal bones,
as well as for motion representations of the arches of the foot, particularly the medial
Shank-Calcaneus Calcaneus-Metatarsus
0 0
Dorsiflexion
–10 –10
–20
angle [deg]
–20
–30
–30
Plantarflexion
y –40 y
–40 –50
x
x
sagittal sagittal
–50 –60
1 20 40 60 80 100 1 20 40 60 80 100
40
Invesion
20
30
10 20
angle [deg]
0 10
0
Eversion
–10
z
x –10 z x
–20 frontal frontal
–20
1 20 40 60 80 100 1 20 40 60 80 100
30 40
Adduction
20 30
20
10
angle [deg]
10
0 y
0 y
Abduction
–10 z –10
transverse z
transverse
–20 –20
1 20 40 60 80 100 1 20 40 60 80 100
load mid-stance late-stance swing load swing

mid-stance late-stance
response response
% gait cycle % gait cycle
Fig. 2 Typical mean ( one standard deviation) temporal profiles of foot joint rotations over the
full gait cycle from a control population of normal subjects. In the left and right columns,
respectively: motion of the calcaneus in the shank reference frame and of the metatarsals in the
calcaneus reference frame. From top to bottom rows: rotations in the sagittal, frontal, and transverse
anatomical planes
longitudinal arch, and the varus/valgus inclination of the calcaneus. With this
approach, however, very erroneous and misleading values can be obtained in
extreme conditions, particularly in case of large ranges of joint motion and in case
of large deviations between the line segment and the projection plane.
Another important question is whether to use a reference neutral position for the
foot and ankle joints. Most frequently, a double leg standing posture is recorded to
provide reference orientations of the foot and lower limb segments. The neutral
orientation can be used as offset and subtracted from the corresponding temporal
profile of joint rotation. The so-called “subtalar neutral” is also sought (Rao et al.
2009) to establish the correct initial alignment of the foot and ankle. Plaster molds
have also been exploited to control the foot resting position (Saraswat et al. 2012,
2013), ensuring foot placement reproducibility and segment neutral orientation. This
procedure is intended to compensate for differing anatomical frame definitions and
foot static deformities, in order to establish a common “zero reference level” for
inter-subject comparisons. The use of a neutral posture has the advantage of remov-
ing the bias associated to the anatomical frame definitions, thus allowing to focus the
analysis and all relevant measurements on the “dynamic” pattern of the joint rota-
tions. Unfortunately, it also removes any joint misalignments due to bone and/or joint
deformity, which are frequently included in the clinical evidence of each patient and
therefore should not be removed from the analysis. The choice of offsetting joint
rotations by using a neutral posture is thus related to the specific study and its
hypotheses and should take into consideration, for example, if there is any ongoing
treatment to correct a foot deformity. Regardless of its application to offset the
kinematic data, the inter-segmental orientations with the subject in the neutral
posture represent extremely valuable information that should always be analyzed
and assessed, in relation to the corresponding temporal profiles of joint rotations.
In order to help final users to identify which MFM is more reliable, repeatable,
and/or best fitting the aims of their investigation, few studies have been published
which compare the performance of the most popular MFM. Mahaffey et al. (2013)
have used intra-class correlation coefficients to analyze the OFM, the Rizzoli Foot
Model (RFM), and the Kinfoot (MacWilliams et al. 2003) in 17 children on two
testing sessions. Although some variability has been found between segments, multi-
segment foot kinematics were shown to be quite repeatable even in pediatric feet. A
standard error of measurement greater than 5 was found in 26%, 15%, and 44% of
the kinematic parameters, respectively, for the OFM, RFM, and the Kinfoot model.
The latter showed the lowest repeatability and the highest errors. The OFM demon-
strated moderate repeatability and reasonable errors in all segments except for the
hindfoot in the transverse plane. The RFM resulted in moderate repeatability and
reasonable test-retest error similar to that of the OFM, but with original additional
data also on midfoot kinematics. In another paper by Powell et al. (2013), the OFM
and RFM were assessed in the context of foot function and alignment as possible
predisposition factors for overuse and traumatic injury in athletes. Both models
helped detect significant differences in frontal plane motion between high- and
low-arched footed athletes. However, the RFM was suggested to be the more
appropriate because it allows to track also midfoot motion. While it was not the
main scope of the study, a comparison between the Shriners Hospital for Children
Greenville Foot Model (Davis et al. 2006) and the OFM can be found also in Maurer
et al. (2013). The former model was shown to be more effective in quantifying the
presence and severity of midfoot break deformity in the sagittal plane and in
monitoring the progression over time. Di Marco et al. (2015a, b) performed the
most comprehensive comparative analysis to date of the OFM, RFM, the Sawacha
et al. (2009), and Saraswat et al. (2012) models. The best coefficient of multiple
correlation between-sessions of the kinematic parameters during ground and tread-
mill walking was observed for the RFM (range 0.83–0.95).
Perhaps an overabundance of multi-segment foot techniques and models has been

proposed to date. Some of these have been made available to the motion analysis
community also via simple-to-use software codes. New users are free to choose the
most appropriate model/technique for their needs according to the experimental
conditions. In particular, the visibility and traceability of the relevant markers must
be considered, both in relation to their dimension and location, together with its
applicability on the clinical population under investigation, and to the motor activ-
ities to be analyzed. Moreover, foot and leg deformities should be carefully assessed
before starting the data collection campaign. The advantages and disadvantages of
existing techniques should be considered and analyzed before developing and
validating a novel MFM suitable to the aims of the investigation.
Validation and Application of Foot Models
Validation Studies
New motion analysis procedures always require proper validation, but this is partic-
ularly challenging for the kinematic analysis of foot segments via skin-markers.
Usually, MFMs are only assessed for repeatability of measurements (see Table 1)
(Mahaffey et al. 2013; Di Marco et al. 2015a, b). Videofluoroscopy has been
employed to estimate the error in the measurements due to the skin motion artifacts
(Leardini et al. 2005; Wrbaskić and Dowling 2007; Shultz et al. 2011a). Skin motion
artifacts were shown to be as large as 16 mm in very strenuous foot conditions. The
largest errors were measured in the hindfoot and midfoot clusters at toe-off, likely
because of the large deformations experienced by the foot bones and skin in this
phase of stance. Still, the skin-to-bone relative motion at the foot was found to be
smaller than that of typical markers on the shank and thigh (Leardini et al. 2005),
thus it has been deemed sufficiently reliable for foot bone tracking. However, the
most convincing evidence of skeletal motion is from in-vitro and in-vivo bone pins
measurements. In vitro, robotic gait simulators are used to replicate the biomechan-
ical conditions of the stance phase of walking on foot cadaver specimens (Whittaker
et al. 2011; Peeters et al. 2013), and kinematics of foot bones can be accurately
tracked via bone pins instrumented with markers. This data helped verify a promis-
ing consistency in foot joint kinematic patterns, for most of the foot joints, between
skin-markers and bone pin measurements. Moreover, it has been possible to detect
motion in a number of joints that are difficult to analyze in-vivo. In-vitro kinematic
data should always be critically evaluated in relation to the fidelity of the replication
of the real in-vivo conditions.
Validation of MFMs has been performed also by tracking real bone motion in-
vivo (Nester et al. 2007a, b, 2010; Arndt et al. 2007; Lundgren et al. 2008; Wolf et al.
2008a; Okita et al. 2009). This required bone pins to be instrumented with marker
clusters and fixated to a number of foot segments in volunteers under a small dose of
local anesthesia. In this condition, the motion pattern of the main foot joints during
walking and running can be established very accurately. It has been shown that the
motion patterns with and without the inserted pins compare well, indicating that the
subjects had little motion restriction due to such invasive intervention. Motion of
major joints was revealed to be very complex, and that of small joints, such as the
talo-navicular, to be larger than what expected – about 10 in the three anatomical
planes – and also larger than that of the talo-calcaneal joint. Motion larger than 3 ,
therefore non negligible, was also measured between tibia and fibula. These studies
also showed the kinematic differences between multibone segments, as measured by
external skin clusters, and single bone pins. These experiments are limited by the
small number of subjects and are hardly replicable for technical and ethical reasons.
The relevant data published so far must serve as reference for other investigations on
normal and pathological feet.
Musculoskeletal Multi-Segment Foot Modeling
MFMs can be used also to develop and validate complex musculoskeletal computer
models for forward and inverse dynamic analysis. Typically, medical imaging is
used to define geometrical models of the anatomical structures and in vivo recorded
kinematics, whereas ground reaction forces provide the data to perform inverse-
dynamics. This allows measurement of bone segment kinematics and estimes of
loading conditions, at the joints, muscle-tendon units, and ligaments. These models
are particularly valuable for an insight into pathological conditions, understanding
disease mechanisms, and simulating the effects of possible treatments, whether
surgical, pharmacological or physical. Saraswat et al. (2010) proposed a generic
musculoskeletal model of an adult foot, including the intrinsic muscles and liga-
ments of the foot and ankle, configured and scaled by skin marker trajectories and an
optimization routine. The predicted muscle activation patterns were assessed against
corresponding EMG measurements from the literature. It was shown that small
marker placement error may result in large differences in model outcomes. Another
large investigation (Oosterwaal et al. 2011, 2016) has proposed a more complex
musculoskeletal model which implies 26 segments and 26 idealized joints, either
cylindrical, universal, or spherical, for a total of 39 degrees of freedom. The model
geometry can be customized using CT and MRI data, and dynamic simulations can
be performed by using bone kinematic data from the 34 skin markers. Both forward
and inverse dynamic modeling were claimed to be exploitable, with the former
integrating a multibody approach with FE analysis and the latter allowing to describe
the interactions of the musculoskeletal structures. Computer models designed to
estimate the mechanics of single structures in the foot, which can hardly be measured
non invasively have also been developed and reported. One of these models has
allowed to estimate the effect of walking speed on the tension in the plantar
aponeurosis (Caravaggi et al. 2010).
A most recent study has exploited the modified OFM to develop a method for
configuring personalized foot models to patients suffering of juvenile idiopathic
arthritis (Prinold et al. 2016). This has highlighted the criticality of patient-specific
definition of the ankle joint axes and the location of the Achilles tendon insertions.
These models have great potential for the analysis of loading conditions in healthy
and pathological feet, and is beneficial to predicting the effects, and thus improving
the efficacy, of surgical treatments and foot and ankle orthoses. However: (a) the
relevant data sets are hardly available or difficult to create, with real personalization
nearly impossible; (b) the mechanical parameters of soft tissues (ligaments, tendons,
retinacula, etc.) are difficult to attain and not fully available in the literature; (c) the
external (so-called “boundary”) conditions depend on many factors not all measur-
able; and (d) the collection of relevant marker trajectories is demanding and requires
a cameras setup with special arrangements. Confidence in interpretation diminishes
because of the many unknowns and the relevant assumptions, optimization criteria,
conventions, calculations, etc. The final concern is about validation, which is
difficult to obtain for these bulky computational models.
Kinetic Analysis Including Foot Models
The moment and power at the ankle joint have been reported extensively in standard
gait analysis with traditional single-segment foot models. However, MFMs offer
specific insight also into muscle performance which would enlighten their function,
particularly in physiological and pathological conditions. This assessment must take
advantage of foot joint kinetics, where ground reaction force and joint center
location are combined to obtain joint moment, work, and power. A number of
studies (Dixon et al. 2012; MacWilliams et al. 2003) have demonstrated that the
contribution of the midfoot to the overall power is important for forward propulsion
during gait; power generation at rearfoot would instead be overestimated by a single-
segment foot model. Detection of abnormal kinetic patterns, even prior to significant
worsening of the prognosis, may help with the formulation of early specific inter-
ventions, aimed at reducing the progression of deformity and disability (DiLiberto
et al. 2015).
In kinetic analysis, in addition to meaningful references associated to the seg-
ments of interest, the complete ground reaction forces, comprising normal and shear
components, must be measured. The inertial properties of segments and the location
of joint centers must also be determined. The latter is particularly critical at the
Chopart and Lisfranc joints, but also at the metatarso-phalangeal joints since these
encompass a number of anatomical articulations. Usually additional virtual points
are defined on these joint lines, however these do not precisely represent the exact
position of anatomical elements. Several attempts have been made to determine an
optimum suitable technique for measuring reaction forces under each segment, i.e.,
subarea ground reaction. In order to estimate these regional forces, a miniature force
platform, requiring the superposition of many targeted trials to create a full analysis
(Scott and Winter 1991), and combined pressure and force plates with assumptions
of proportionality between the two (Giacomozzi and Macellari 1997; MacWilliams
et al. 2003; Giacomozzi et al. 2006) have been proposed.
Early kinetic measures of the foot (Scott and Winter 1991; MacWilliams et al.
2003) have been limited by intricate assumptions and equipment restrictions. The
former was an 2D eight-joint model and relied on the superposition of multiple trials
interacting with a small custom-built force sensor. The latter paper proposed a first
3D model with eight segments but implied many assumptions on joint motion,
pressure, as well as force data for a final estimation of joint moments, which
neglected the mediolateral component of the force. A theoretical extension of the
MacWilliams model, addressing kinetic-based calculations, has been proposed a
little later (Buczek et al. 2006).
An extensive investigation undergone in a clinical and research setting (Bruening
et al. 2012a, b), involving respectively 17 and 10 healthy pediatric subjects,
explicitly aimed at exploiting the feasibility and relevance of a proposed technique
for kinetic analysis applied to a MFM. A three-segment kinetic model was first
characterized and assessed. In the second paper, kinetic parameters, joint moments
and powers during level walking were reported. Three submodels of the shank and
foot complex were created and two adjacent force platforms employed for calcula-
tion of rotation, net moment, and total power at the ankle, midtarsal, and 1st
metatarso-phalangeal joints. Unfortunately, the protocol required visual targeting
of the force platforms, whose confounding contribution, together with that of the
inertial parameters, was assessed separately. The study confirmed that not only
sagittal plane motion but also generated peak power are generally overestimated
at the ankle (35% on average) when using a single-segment, rather than a MFM.
This “split force platform approach” can be an alternative to ad-hoc hardware, but it
is critical when applied to pathological populations for the special targeting of the
force plates.
DiLiberto et al. (2015) have reported, for the first time, multi-joint foot kinetics in
subjects with diabetes mellitus, peripheral neuropathy, and in a healthy adult control
group, by using an electromagnetic sensor motion capture system. The model
consisted of the tibia, rearfoot, and forefoot, with ideal hinge joints connecting the
segments. It has been shown that (i) positive peak power and work of the forefoot
with respect to rearfoot was smaller in the patients group with respect to the control,
(ii) negative peak powers of both the forefoot to rearfoot and rearfoot to tibia was
larger, and (iii) a greater proportion of negative work was present at both these joints.
While, the value of joint kinematics from MFMs has been repeatedly demon-
strated, relevant joint kinetics are still controversial because motion occurs at
multiple articulations between bony segments (Nester et al. 2007a), particularly for
the critical estimation of position and orientation of the axes of rotation, which
would significantly affect kinetic measurements.
It is no doubt that kinetic measurements can increase the knowledge of foot and
ankle function, as well as influence the evolution of MFMs. Though some prelim-
inary attempts in this direction might have been too complex (Wang et al. 2016), the
currently available simple representations here discussed seem a good compromise
between comprehensiveness of foot kinetics and comprehension of the results for
clinical use. These techniques, however, are hardly exploitable by the large scientific
community, due to the modeling and the implementation issues.
Summary
Stereophotogrammetry has made enormous progress in the last few years, with
cheaper motion analysis systems achieving high performance in markers tracking.
Today, multi-segment foot kinematic analysis is easier than what it was years ago. A
large number of multi-segment foot techniques for in vivo gait analysis based on
stereophotogrammetric systems have been published over the last three decades. The
large variety of currently available marker sets and protocols is of great value for
anyone interested in the analysis of foot motion in normal or pathological conditions,
following treatments, in the evaluation of sport performance, etc. Following this
long survey of the current techniques for kinematic analysis of foot segments, the
reader interested in the application of an existing MFM, or in the design of a new one
suitable to the aims of his/her specific investigation, shall find valuable recommen-
dations for either options here below.
In case of selection of an existing MFM, the researcher must be careful in deciding
on the most appropriate model from the literature; this must be done according to the
number of segments and degrees of freedom, the technique used for the calculation of
joint rotations, the position and visibility of the markers, etc. As far as the latter, the
number of cameras and their positions should be checked against the motor tasks
analyzed. Using a camera setup closer to the acquisition volume, and better optics,
normally will result in a better quality of the measurements. However, the cameras are
often arranged for full body gait analysis, and the necessary frequent changes in
the cameras configuration can be a tedious operation requiring repeated calibrations.
The consistency with existing single-segment foot and lower limb models should also
be discussed with the motion analysis team. First of all the user must comprehend
which segments should be tracked for the study, thus excluding unnecessarily
complex marker sets. Another important decision regards the reference neutral
orientation and whether this shall be used as an off-set to compensate for anatomical
frame definitions and foot deformities. As already stated above, the decision should
be taken according to the scope of the study and the population analyzed. Relevant
literature supporting the model should also be sought, in particular in relation to the
repeatability of measurements and the validation against gold standards. Calculation
and analysis of foot joint rotations can be supported by commercial software or
freeware. While this allows users to save time and resources in writing ad-hoc
analysis programs, the consistency of the calculations should always be assessed
against corresponding data published by the authors. Unfortunately, many papers do
not provide complete instructions on exact marker mounting, marker trajectory
smoothing, anatomical frame definition, joint conventions, etc. (Bishop et al. 2012);
even small changes to the original definitions may affect the final measurements. The
involvement of experienced operators is therefore recommended, and a program of
continuous training activities should be implemented. Software tools which have
been developed and/or verified by the original model designers are preferable.
In case it is required to design new specific models, a consistent and careful
identification of specific single or combined bony segments, which are originally
tracked according to the aims of the study, must be performed. From a theoretical
perspective, an extremely large number of models can be devised and formalized

simply by mixing and matching the 26 foot bones. As long as a reference system is
clearly defined and a sufficient number of markers can be fixated to each segment,
the model can be considered original and reproducible. However, many of the papers
reported here have failed to properly describe the relevant details of the model,
including origin and orientation of the reference frames. Moreover, some of these
allegedly novel MFMs are merely a subset or small variations of previously existing
models, whereby a fewer number of segments has been tracked and/or only minor
modifications to the location of the markers have been implemented. In this respect,
we recommend that for a new, in vivo multi-segment technique for kinematic
analysis of the foot to be considered as a novel model, the following criteria should
be followed:
(1) The rationale for proposing a new segmentation of the foot for kinematic
analysis should be explicit and clear, e.g., for biomechanical or clinical purposes.
The authors should be able to address the question: why any of the existing
MFMs is not appropriate to address the aims of my study?
(2) The description of the new model should be exhaustive as far as the following
information: foot segments; relevant anatomical landmarks; position and orien-
tation of the local reference co-ordinate system; and number and location of the
skin markers for motion tracking.
(3) A validation study should be performed – possibly by a different research group
– to demonstrate that the repeatability of the kinematic outcome is comparable or
better than that of the current most common and widely used foot models.
(4) At least one clinical or biomechanical application of the model should be
reported to demonstrate the value and efficacy of the new model.
If the proposed criteria were strictly followed, only a few of the foot segmentation
techniques found in the literature and here reported could be considered as appro-
priate MFMs. In fact, even following these criteria would not necessarily guarantee
the new model to become popular in everyday gait analyses and to be widely used.
Many other factors influence the “popularity” and diffusion of gait analysis protocols
in the scientific and clinical contexts, such as their comprehension (as far as the basic
concepts, its application, the data processing and reporting), their usability and
utility, the availability of relevant processing software, the support provided by the
inventors, the distribution and actions of motion capture systems vendors, etc. The
scientific community of human motion analysis would however be pleased to
welcome any such robust contribution in the discipline.
Future Directions
While foot kinematics can now be described with multiple degrees of freedom foot
models, a number of issues still remain to be addressed in the future, particularly for
in vivo studies. The complex procedures for validation of the models and the
unavoidable bias from the soft tissue artifacts are known to affect these measures
(Leardini et al. 2005). Nevertheless single or plate-mounted skin markers are neces-
sary, the alternatives in routine clinical analyses having several limitations. While
videofluoroscopy and bone pins provide more accurate measures of foot joints motion,
these are also excessively invasive for patients. Inertial sensors (Rouhani et al. 2012,
2014) and marker-less dynamic 3D scanning (Van den Herrewegen et al. 2014) are
definitely less invasive, but accuracy and anatomical based analysis fall short.
Due to the variety of motion analysis systems and protocols for data acquisition, a
normative database for foot kinematic data should be available in every single gait
analysis facility (Deschamps et al. 2012a). Moreover, kinematic analysis of foot
joints is highly sensitive to errors in marker placement due to the small size of foot
bones in comparison to other body segments. Even a small deviation from the
recommended marker location can result in a large error within the position and
orientation of the reference co-ordinate system. In order to decrease the error rate and
improve the protocol repeatability, only experienced operators with extensive
knowledge of foot anatomy and practice in multi-segment foot analysis should be
in charge of mounting the markers (Caravaggi et al. 2011; Deschamps et al. 2012a).
To ensure for an increased and coherent use of these techniques, standardization
of the references shall be sought, in terms of anatomical landmark and frame
definitions. In this respect, another fundamental step forward would imply
establishing a common terminology, which can avoid confusion and inconsistency.
The utilization of these models is still critical in the presence of shoes or foot
orthotics (Lin et al. 2013; Bishop et al. 2013; Leardini et al. 2014; Bishop et al.
2015; Halstead et al. 2016), hence the limitation of available literature currently. All
these issues perhaps can account for the limited number of relevant clinical appli-
cations (Deschamps et al. 2011).
Cross-References

Sensors
▶ Conventional Gait Model: Success and Limitations
▶ EMG Activity in Gait: The Influence of Motor Disorders
▶ Foot and Ankle Motion in Cerebral Palsy
▶ Functional Effects of Ankle Sprain
▶ Functional Effects of Foot Orthoses
▶ Functional Effects of Shoes
▶ Integration of Foot Pressure and Foot Kinematics Measurements for Medical
Applications
▶ Interpreting Joint Moments and Powers in Gait
▶ Interpreting Spatiotemporal Parameters, Symmetry and Variability in Clinical Gait
Analysis

▶ The Effects of Ankle Joint Replacement on Gait
▶ Three-Dimensional Human Kinematics Estimation Using Magneto and Inertial
Measurement Units
▶ Variations of Marker-Sets and Models for Standard Gait Analysis
References
Alonso-Vázquez A, Villarroya MA, Franco MA, Asín J, Calvo B (2009) Kinematic assessment of
paediatric forefoot varus. Gait Posture 29(2):214–219
Arampatzis A, Brüggemann GP, Klapsing GM (2002) A three-dimensional shank-foot model to
determine the foot motion during landings. Med Sci Sports Exerc 34(1):130–138
Arndt A, Wolf P, Liu A, Nester C, Stacoff A, Jones R, Lundgren P, Lundberg A (2007) Intrinsic foot
kinematics measured in vivo during the stance phase of slow running. J Biomech 40
(12):2672–2678
Arnold JB, Mackintosh S, Jones S, Thewlis D (2013) Repeatability of stance phase kinematics from
a multi-segment foot model in people aged 50 years and older. Gait Posture 38(2):349–351
Arnold JB, Mackintosh S, Jones S, Thewlis D (2014a) Differences in foot kinematics between
young and older adults during walking. Gait Posture 39(2):689–694
Arnold J, Mackintosh S, Jones S, Thewlis D (2014b) Altered dynamic foot kinematics in people
with medial knee osteoarthritis during walking: a cross-sectional study. Knee 21(6):1101–1106
Atkinson HD, Daniels TR, Klejman S, Pinsker E, Houck JR, Singer S (2010) Pre- and post-
operative gait analysis following conversion of tibiotalocalcaneal fusion to total ankle
arthroplasty. Foot Ankle Int 31(10):927–932
Barton CJ, Levinger P, Crossley KM, Webster KE, Menz HB (2011a) Relationships between the
Foot Posture Index and foot kinematics during gait in individuals with and without
patellofemoral pain syndrome. J Foot Ankle Res 4:10
Barton CJ, Levinger P, Webster KE, Menz HB (2011b) Walking kinematics in individuals with
patellofemoral pain syndrome: a case–control study. Gait Posture 33(2):286–289
Barton CJ, Menz HB, Levinger P, Webster KE, Crossley KM (2011c) Greater peak rearfoot
eversion predicts foot orthoses efficacy in individuals with patellofemoral pain syndrome. Br
J Sports Med 5(9):697–701
Benedetti MG, Manca M, Ferraresi G, Boschi M, Leardini A (2011) A new protocol for 3D
assessment of foot during gait: application on patients with equinovarus foot. Clin Biomech
26(10):1033–1038
Bishop C, Paul G, Thewlis D (2012) Recommendations for the reporting of foot and ankle models.
J Biomech 45(13):2185–2194
Bishop C, Paul G, Thewlis D (2013) The reliability, accuracy and minimal detectable difference of a
multi-segment kinematic model of the foot-shoe complex. Gait Posture 37(4):552–557
Bishop C, Arnold JB, Fraysse F, Thewlis D (2015) A method to investigate the effect of shoe-hole
size on surface marker movement when describing in-shoe joint kinematics using a multi-
segment foot model. Gait Posture 41(1):295–299
Bishop C, Arnold JB, May T (2016) Effects of taping and orthoses on foot biomechanics in adults
with flat arched feet. Med Sci Sports Exerc 48(4):689–696
Brodsky JW, Charlick DA, Coleman SC, Pollo FE, Royer CT (2009) Hindfoot motion following
reconstruction for posterior tibial tendon dysfunction. Foot Ankle Int 30(7):613–618
Bruening DA, Cooney KM, Buczek FL (2012a) Analysis of a kinetic multi-segment foot model.
Part I: model repeatability and kinematic validity. Gait Posture 35(4):529–534
Bruening DA, Cooney KM, Buczek FL (2012b) Analysis of a kinetic multi-segment foot model.
Part II: kinetics and clinical implications. Gait Posture 35(4):535–540
Buczek FL, Walker MR, Rainbow MJ, Cooney KM, Sanders JO (2006) Impact of mediolateral
segmentation on a multi-segment foot model. Gait Posture 23(4):519–522
Buldt AK, Murley GS, Butterworth P, Levinger P, Menz HB, Landorf KB (2013) The relationship
between foot posture and lower limb kinematics during walking: a systematic review. Gait
Posture 38(3):363–372
Buldt AK, Levinger P, Murley GS, Menz HB, Nester CJ, Landorf KB (2015) Foot posture is
associated with kinematics of the foot during gait: a comparison of normal, planus and cavus
feet. Gait Posture 42(1):42–48
Canseco K, Long J, Marks R, Khazzam M, Harris G (2008) Quantitative characterization of gait
kinematics in patients with hallux rigidus using the Milwaukee foot model. J Orthop Res 26
(4):419–427
Canseco K, Long J, Marks R, Khazzam M, Harris G (2009) Quantitative motion analysis in patients
with hallux rigidus before and after cheilectomy. J Orthop Res 7(1):128–134
Canseco K, Long J, Smedberg T, Tarima S, Marks RM, Harris GF (2012) Multisegmental foot and
ankle motion analysis after hallux valgus surgery. Foot Ankle Int 33(2):141–147
Caravaggi P, Pataky T, Gunther M et al (2010) Dynamics of longitudinal arch support in relation to
walking speed: contribution of the plantar aponeurosis. J Anat 217:254–261
Caravaggi P, Benedetti MG, Berti L, Leardini A (2011) Repeatability of a multi-segment foot
protocol in adult subjects. Gait Posture 33(1):133–135
Carson MC, Harrington ME, Thompson N, O’Connor JJ, Theologis TN (2001) Kinematic analysis
of a multi-segment foot model for research and clinical applications: a repeatability analysis.
J Biomech 34(10):1299–1307
Carty CP, Walsh HP, Gillett JG (2015) Sensitivity of the Oxford foot model to marker misplace-
ment: a systematic single-case investigation. Gait Posture 42(3):398–401
Chang R, Van Emmerik R, Hamill J (2008) Quantifying rearfoot-forefoot coordination in human
walking. J Biomech 41(14):3101–3105
Chang R, Rodrigues PA, Van Emmerik RE, Hamill J (2014) Multi-segment foot kinematics and
ground reaction forces during gait of individuals with plantar fasciitis. J Biomech 47
(11):2571–2577
Cobb SC, Tis LL, Johnson JT, Wang YT, Geil MD, McCarty FA (2009) The effect of low-mobile
foot posture on multi-segment medial foot model gait kinematics. Gait Posture 30(3):
334–339
Cornwall MW, McPoil TG (1999a) Effect of ankle dorsiflexion range of motion on rearfoot motion
during walking. J Am Podiatr Med Assoc 89(6):272–277
Cornwall MW, McPoil TG (1999b) Three-dimensional movement of the foot during the stance
phase of walking. J Am Podiatr Med Assoc 89(2):56–66
Cornwall MW, McPoil TG (2002) Motion of the calcaneus, navicular, and first metatarsal during the
stance phase of walking. J Am Podiatr Med Assoc 92(2):67–76
Curtis DJ, Bencke J, Stebbins JA, Stansfield B (2009) Intra-rater repeatability of the Oxford foot
model in healthy children in different stages of the foot roll over process during gait. Gait
Posture 30(1):118–121
Davis RB, Eugene G, Jameson E, Davids JR, Christopher LM, Benjamin M, Rogozinski B,
Anderson JP (2008) The design, development, and initial evaluation of a multi-segment foot
model for routine clinical gait analysis. In: Harris GF, Smith PA, Marks RM (eds) Foot and ankle
motion analysis: clinical treatment and technology. CRC Press, Taylor and Francis Group, pp
425–444
De Mits S, Segers V, Woodburn J, Elewaut D, De Clercq D, Roosen P (2012) A clinically applicable
six-segmented foot model. J Orthop Res 30(4):655–661
De Ridder R, Willems T, Vanrenterghem J, Robinson MA, Palmans T, Roosen P (2015) Multi-
segment foot landing kinematics in subjects with chronic ankle instability. Clin Biomech 30
(6):585–592
Deschamps K, Birch I, Desloovere K, Matricali GA (2010) The impact of hallux valgus on foot
kinematics: a cross-sectional, comparative study. Gait Posture 32(1):102–106
Deschamps K, Staes F, Roosen P, Nobels F, Desloovere K, Bruyninckx H, Matricali GA (2011)

Body of evidence supporting the clinical use of 3D multi-segment foot models: a systematic
review. Gait Posture 33(3):338–349
Deschamps K, Staes F, Bruyninckx H, Busschots E, Jaspers E, Atre A, Desloovere K (2012a)
Repeatability in the assessment of multi-segment foot kinematics. Gait Posture 35(2):255–260
Deschamps K, Staes F, Bruyninckx H, Busschots E, Matricali GA, Spaepen P, Meyer C, Desloovere
K (2012b) Repeatability of a 3D multi-segment foot model protocol in presence of foot
deformities. Gait Posture 36(3):635–638
Deschamps K, Matricali GA, Roosen P, Nobels F, Tits J, Desloovere K, Bruyninckx H, Flour M,
Deleu PA, Verhoeven W, Staes F (2013) Comparison of foot segmental mobility and coupling
during gait between patients with diabetes mellitus with and without neuropathy and adults
without diabetes. Clin Biomech 28(7):813–819
Deschamps K, Dingenen B, Pans F, Van Bavel I, Matricali GA, Staes F (2016) Effect of taping on
foot kinematics in persons with chronic ankle instability. J Sci Med Sport 19(7):541–546
Di Marco R, Rossi S, Racic V, Cappa P, Mazzà C (2015a) A comparison between four foot model
protocols: the effect of walking on a treadmill. Conference paper 2015: XXV congress of the
international society of biomechanics, Glasgow
Di Marco R, Rossi S, Racic V, Cappa P, Mazzà C (2015b) Concurrent reliability assessment of four
foot models for gait analysis. Conference paper 2016: XVI congress of the Società Italiana di
Analisi del Movimento in Clinica, Padova
DiLiberto FE, Tome J, Baumhauer JF, Quinn JR, Houck J, Nawoczenski DA (2015) Multi-joint foot
kinetics during walking in people with diabetes mellitus and peripheral neuropathy. J Biomech
48(13):3679–3684
Dixon PC, Böhm H, Döderlein L (2012) Ankle and midfoot kinetics during normal gait: a multi-
segment approach. J Biomech 45(6):1011–1016
Dubbeldam R, Nester C, Nene A, Hermens H, Buurke J (2013) Kinematic coupling relationships
exist between non-adjacent segments of the foot and ankle of healthy subjects. Gait Posture
37:159–164
Giacomozzi C, Macellari V (1997) Piezo-dynamometric platform for a more complete analysis of
foot-to-floor interaction. IEEE Trans Rehabil Eng 5(4):322–330
Giacomozzi C, Benedetti MG, Leardini A, Macellari V, Giannini S (2006) Gait analysis with an
integrated system for functional assessment of talocalcaneal coalition. J Am Podiatr Med Assoc
96(2):107–115
Graff A, Hassani S, Krzak J, Long J, Caudill A, Flanagan A, Eastwood D, Kuo KN, Harris G, Smith
P (2010) Long-term outcome evaluation in young adults following clubfoot surgical release.
J Pediatr Orthop 30(4):379–385
Grood ES, Suntay WJ (1983) A joint coordinate system for the clinical description of three-
dimensional motions: application to the knee. J Biomech Eng 105(2):136–144
Halstead J, Keenan AM, Chapman GJ, Redmond AC (2016) The feasibility of a modified shoe for
multi-segment foot motion analysis: a preliminary study. J Foot Ankle Res 9:7
Hetsroni I, Nyska M, Ben-Sira D, Arnson Y, Buksbaum C, Aliev E, Mann G, Massarwe S,
Rozenfeld G, Ayalon M (2011) Analysis of foot and ankle kinematics after operative reduction
of high-grade intra-articular fractures of the calcaneus. J Trauma 70(5):1234–1240
Hösl M, Böhm H, Multerer C, Döderlein L (2014) Does excessive flatfoot deformity affect
function? A comparison between symptomatic and asymptomatic flatfeet using the Oxford
foot model. Gait Posture 39(1):23–28
Houck JR, Tome JM, Nawozensky DA (2006) Subtalar neutral position as an offset for a kinematic
model of the foot during walking. Gait Analy 28:29–37
Houck JR, Neville C, Tome J, Flemister AS (2009) Foot kinematics during a bilateral heel rise test
in participants with stage II posterior tibial tendon dysfunction. J Orthop Sports Phys Ther 39
(8):593–603
Houdijk H, Doets HC, van Middelkoop M, DirkjanVeeger HE (2008) Joint stiffness of the ankle during
walking after successful mobile-bearing total ankle replacement. Gait Posture 27(1):115–119
Hsu WH, Lewis CL, Monaghan GM, Saltzman E, Hamill J, Holt KG (2014) Orthoses posted in
both the forefoot and rearfoot reduce moments and angular impulses on lower extremity joints
during walking. J Biomech 47(11):2618–2625
Hunt AE, Smith RM, Torode M (2001) Extrinsic muscle activity, foot motion and ankle joint
moments during the stance phase of walking. Foot Ankle Int 22(1):31–41
Hwang SJ, Choi HS, Kim YH (2004) Motion analysis based on a multi-segment foot model in
normal walking. Conf Proc IEEE Eng Med Biol Soc 7:5104–5106
Hyslop E, Woodburn J, McInnes IB, Semple R, Newcombe L, Hendry G, Rafferty D, De Mits S,
Turner DE (2010) A reliability study of biomechanical foot function in psoriatic arthritis based
on a novel multi-segmented foot model. Gait Posture 32(4):619–626
Jenkyn TR, Nicol AC (2007) A multi-segment kinematic model of the foot with a novel definition
of forefoot motion for use in clinical gait analysis during walking. J Biomech 40(14):3271–3278
Jenkyn TR, Shultz R, Giffin JR, Birmingham TB (2010) A comparison of subtalar joint motion
during anticipated medial cutting turns and level walking using a multi-segment foot model.
Kalkum E, van Drongelen S, Mussler J, Wolf SI, Kuni B (2016) A marker placement laser device
for improving repeatability in 3D-foot motion analysis. Gait Posture 44:227–230
Kelly LA, Cresswell AG, Racinais S, Whiteley R, Lichtwark G (2014) Intrinsic foot muscles have
the capacity to control deformation of the longitudinal arch. J R Soc Interface 11(93):20131188
Kepple TM, Stanhope SJ, Lohmann KN, Roman NL (1990) A video-based technique for measuring
ankle-subtalar motion during stance. J Biomed Eng 12(4):273–280
Khazzam M, Long JT, Marks RM, Harris GF (2007) Kinematic changes of the foot and ankle
in patients with systemic rheumatoid arthritis and forefoot deformity. J Orthop Res 25
(3):319–329
Kidder SM, Abuzzahab FS Jr, Harris GF, Johnson JE (1996) A system for the analysis of foot and
ankle kinematics during gait. IEEE Trans Rehabil Eng 4(1):25–32
Kitaoka HB, Crevoisier XM, Hansen D, Katajarvi B, Harbst K, Kaufman KR (2006) Foot and ankle
kinematics and ground reaction forces during ambulation. Foot Ankle Int 27(10):808–813
Krzak JJ, Corcos DM, Damiano DL, Graf A, Hedeker D, Smith PA, Harris GF (2015) Kinematic
foot types in youth with equinovarus secondary to hemiplegia. Gait Posture 41(2):402–408
Leardini A, Benedetti MG, Catani F, Simoncini L, Giannini S (1999) An anatomically based
protocol for the description of foot segment kinematics during gait. Clin Biomech 14
(8):528–536
21(2):212–225
Leardini A, Benedetti MG, Berti L, Bettinelli D, Nativo R, Giannini S (2007) Rear-foot, mid-foot
and fore-foot motion during the stance phase of gait. Gait Posture 25(3):453–462
Leardini A, Aquila A, Caravaggi P, Ferraresi C, Giannini S (2014) Multi-segment foot mobility in a
hinged ankle-foot orthosis: the effect of rotation axis position. Gait Posture 40(1):274–277
Levinger P, Murley GS, Barton CJ, Cotchett MP, McSweene SR, Menz HB (2010) A comparison of
foot kinematics in people with normal- and flat-arched feet using the Oxford foot model. Gait
Posture 32:519–523
Lin SC, Chen CP, Tang SF, Wong AM, Hsieh JH, Chen WP (2013) Changes in windlass effect in
response to different shoe and insole designs during walking. Gait Posture 37(2):235–241
Long JT, Eastwood DC, Graf AR, Smith PA, Harris GF (2010) Repeatability and sources of
variability in multi-center assessment of segmental foot kinematics in normal adults. Gait
Posture 31(1):32–36
optimisation with joint constraints. J Biomech 32(2):129–134
Lucareli PR, Contani LB, Lima B, Rabelo ND, Ferreira CL, Lima FP, Correa JC, Politti F (2016)
Repeatability of a 3D multi-segment foot model during anterior and lateral step down tests. Gait
Posture 43:9–16
Lundgren P, Nester C, Liu A, Arndt A, Jones R, Stacoff A, Wolf P, Lundberg A (2008) Invasive in vivo
measurement of rear-, mid- and forefoot motion during walking. Gait Posture 28(1):93–100
MacWilliams BA, Cowley M, Nicholson DE (2003) Foot kinematics and kinetics during adolescent
gait. Gait Posture 17(3):214–224
Mahaffey R, Morrison SC, Drechsler WI, Cramp MCJ (2013) Evaluation of multi-segmental
kinematic modelling in the paediatric foot using three concurrent foot models. J Foot Ankle
Res 6(1):43
Marks RM, Long JT, Ness ME, Khazzam M, Harris GF (2009) Surgical reconstruction of posterior
tibial tendon dysfunction: prospective comparison of flexor digitorumlongus substitution com-
bined with lateral column lengthening or medial displacement calcaneal osteotomy. Gait Posture
29(1):17–22
Maurer JD, Ward V, Mayson TA, Davies KR, Alvarez CM, Beauchamp RD, Black AH (2013)
A kinematic description of dynamic midfoot break in children using a multi-segment foot
model. Gait Posture 38(2):287–292
Merker J, Hartmann M, Kreuzpointner F, Schwirtz A, Haas JP (2015) Pathophysiology of juvenile
idiopathic arthritis induced pesplanovalgus in static and walking condition: a functional view
using 3D gait analysis. J Pediatr Rheumatol Online 13:21
Milner CE, Brindle RA (2016) Reliability and minimal detectable difference in multi-segment foot
kinematics during shod walking and running. Gait Posture 43:192–197
Moseley L, Hunt A, Grant R (1996) Three-dimensional kinematics of the rearfoot during the stance
phase of walking in normal young adult males. Clin Biomech 11(1):39–45
Myers KA, Wang M, Marks RM, Harris GF (2004) Validation of a multi-segment foot and ankle
kinematic model for pediatric gait. IEEE Trans Neural Syst Rehabil Eng 12(1):122–130
Nawoczenski DA, Ketz J, Baumhauer JF (2008) Dynamic kinematic and plantar pressure changes
following cheilectomy for hallux rigidus: a mid-term followup. Foot Ankle Int 29(3):265–272
Ness ME, Long J, Marks R, Harris G (2008) Foot and ankle kinematics in patients with posterior
tibial tendon dysfunction. Gait Posture 27(2):331–339
Nester C, Jones RK, Liu A, Howard D, Lundberg A, Arndt A, Lundgren P, Stacoff A, Wolf P
(2007a) Foot kinematics during walking measured using bone and surface mounted markers.
J Biomech 40(15):3412–3423
Nester CJ, Liu AM, Ward E, Howard D, Cocheba J, Derrick T, Patterson P (2007b) In vitro study of
foot kinematics using a dynamic walking cadaver model. J Biomech 40(9):1927–1937
Nester CJ, Liu AM, Ward E, Howard D, Cocheba J, Derrick T (2010) Error in the description of foot
kinematics due to violation of rigid body assumptions. J Biomech 43(4):666–672
Nester CJ, Jarvis HL, Jones RK, Bowden PD, Liu A (2014) Movement of the human foot in
100 pain free individuals aged 18–45: implications for understanding normal foot function.
J Foot Ankle Res 7(1):51
Neville C, Flemister AS, Houck JR (2009) Effects of the AirLift PTTD brace on foot kinematics in
subjects with stage II posterior tibial tendon dysfunction. J Orthop Sports Phys Ther 39(3):201–209
Novak AC, Mayich DJ, Perry SD, Daniels TR, Brodsky JW (2014) Gait analysis for foot and ankle
surgeons – topical review, part 2: approaches to multi-segment modeling of the foot. Foot Ankle
Int 35(2):178–191
Okita N, Meyers SA, Challis JH, Sharkey NA (2009) An objective evaluation of a segmented foot
Oosterwaal M, Telfer S, Torholm S, Carbes S, van Rhijn LW, Macduff R, Meijer K, Woodburn
J (2011) Generation of subject-specific, dynamic, multi-segment ankle and foot models to
improve orthotic design: a feasibility study. BMC Musculoskelet Disord 12:256
Oosterwaal M, Carbes S, Telfer S, Woodburn J, Tørholm S, Al-Munajjed A, van Rhijn L, Meijer K
(2016) The Glasgow-Maastricht foot model, evaluation of a 26 segment kinematic model of the
foot. J Foot Ankle Res 9:19
Peeters K, Natsakis T, Burg J, Spaepen P, Jonkers I, Dereymaeker G, Vander Sloten J (2013) An
in vitro approach to the evaluation of foot-ankle kinematics: performance evaluation of a
custom-built gait simulator. Proc Inst Mech Eng H 227(9):955–967
Pohl MB, Messenger N, Buckley JG (2006) Changes in foot and lower limb coupling due to
systematic variations in step width. Clin Biomech 21(2):175–183
Portinaro N, Leardini A, Panou A, Monzani V, Caravaggi P (2014) Modifying the Rizzoli foot
model to improve the diagnosis of pes-planus: application to kinematics of feet in teenagers.
J Foot Ankle Res 7(1):754
Pothrat C, Authier G, Viehweger E, Berton E, Rao G (2015) One- and multi-segment foot models
lead to opposite results on ankle joint kinematics during gait: implications for clinical assess-
ment. Clin Biomech 30(5):493–499
Powell DW, Williams DS, Butler RJ (2013) A comparison of two multi-segment foot models in
high-and low-arched athletes. J Am Podiatr Med Assoc 103(2):99–105
Prinold JA, Mazzà C, Di Marco R, Hannah I, Malattia C, Magni-Manzoni S, Petrarca M, Ronchetti
AB, Tanturri de Horatio L, van Dijkhuizen EH, Wesarg S, Viceconti M, MD-PAEDIGREE
Consortium Ann (2016) A patient-specific foot model for the estimate of ankle joint forces in
patients with juvenile idiopathic arthritis. Ann Biomed Eng 44(1):247–257
Rankine L, Long J, Canseco K, Harris GF (2008) Multisegmental foot modeling: a review. Crit Rev
Biomed Eng 36(2–3):127–181
Rao S, Saltzman C, Yack HY (2006) Segmental foot mobility in individuals with and without
diabetes and neuropathy. Clin Biomech 22:464–471
Rao S, Baumhauer JF, Tome J, Nawoczenski DA (2009) Comparison of in vivo segmental foot
motion during walking and step descent in patients with midfoot arthritis and matched asymp-
tomatic control subjects. J Biomech 42(8):1054–1060
Rattanaprasert U, Smith R, Sullivan M, Gilleard W (1999) Three-dimensional kinematics of the
forefoot, rearfoot, and leg without the function of tibialis posterior in comparison with normals
during stance phase of walking. Clin Biomech 14(1):14–23
Raychoudhury S, Hu D, Ren L (2014) Three-dimensional kinematics of the human meta-
tarsophalangeal joint during level walking. Front Bioeng Biotechnol 2:73
Rouhani H, Favre J, Crevoisier X, Jolles BM, Aminian K (2011) Segmentation of foot and ankle
complex based on kinematic criteria. Comput Methods Biomech Biomed Engin 4(9):773–781
Rouhani H, Favre J, Aminian K, Crevoisier X (2012) Multi-segment foot kinematics after total
ankle replacement and ankle arthrodesis during relatively long-distance gait. Gait Posture 36
(3):561–566
Rouhani H, Favre J, Crevoisier X, Aminian K (2014) A wearable system for multi-segment foot
kinetics measurement. J Biomech 47(7):1704–1711
Saraswat P, Andersen MS, MacWilliams BA (2010) A musculoskeletal foot model for clinical gait
analysis. J Biomech 43:1645–1652
Saraswat P, MacWilliams BA, Davis RB (2012) A multi-segment foot model based on anatomically
registered technical coordinate systems: method repeatability in pediatric feet. Gait Posture 35
(4):547–555
Saraswat P, MacWilliams BA, Davis RB, D’Astous JL (2013) A multi-segment foot model based on
anatomically registered technical coordinate systems: method repeatability and sensitivity in
pediatric planovalgus feet. Gait Posture 37(1):121–125
Sawacha Z, Cristoferi G, Guarneri G, Corazza S, Donà G, Denti P, Facchinetti A, Avogaro A,
Cobelli C (2009) Characterizing multi-segment foot kinematics during gait in diabetic foot
patients. J Neuroeng Rehabil 6:37
Scott SH, Winter DA (1991) Talocrural and talocalcaneal joint kinematics and kinetics during the
stance phase of walking. J Biomech 24(8):743–752
Seo SG, Lee DY, Moon HJ, Kim SJ, Kim J, Lee KM, Chung CY, Choi IH (2014) Repeatability of a
multi-segment foot model with a 15-marker set in healthy adults. J Foot Ankle Res 7:24
Shultz R, Kedgley AE, Jenkyn TR (2011a) Quantifying skin motion artifact error of the hindfoot
and forefoot marker clusters with the optical tracking of a multi-segment foot model using
single-plane fluoroscopy. Gait Posture 34(1):44–48
Shultz R, Birmingham TB, Jenkyn TR (2011b) Differences in neutral foot positions when measured
barefoot compared to in shoes with varying stiffnesses. Med Eng Phys 33(10):1309–1313
Simon J, Doederlein L, McIntosh AS, Metaxiotis D, Bock HG, Wolf SI (2006) The Heidelberg foot
measurement method: development, description and assessment. Gait Posture 23(4):411–424
Souza TR, Fonseca HL, Vaz AC, Antero JS, Marinho CS, Fonseca ST (2014) Between-day
reliability of a cluster-based method for multi-segment kinematic analysis of the foot-ankle
complex. J Am Podiatr Med Assoc 104(6):601–609
Stebbins J, Harrington M, Thompson N, Zavatsky A, Theologis T (2006) Repeatability of a model
for measuring multi-segment foot kinematics in children. Gait Posture 23(4):401–410
Stebbins J, Harrington M, Thompson N, Zavatsky A, Theologis T (2010) Gait compensations
caused by foot deformity in cerebral palsy. Gait Posture 32(2):226–230
Theologis TN, Harrington ME, Thompson N, Benson MK (2003) Dynamic foot movement in
children treated for congenital talipesequinovarus. J Bone Joint Surg Br 85(4):572–577
Tome J, Nawoczenski DA, Flemister A, Houck J (2006) Comparison of foot kinematics between
subjects with posterior tibialis tendon dysfunction and healthy controls. J Orthop Sports Phys
Ther 36(9):635–644
Turner DE, Helliwell PS, Emery P, Woodburn J (2006) The impact of rheumatoid arthritis on foot
function in the early stages of disease: a clinical case series. BMC Musculoskelet Disord 7:102
Turner DE, Woodburn J (2008) Characterising the clinical and biomechanical features of severely
deformed feet in rheumatoid arthritis. Gait Posture 28(4):574–580
Twomey D, McIntosh AS, Simon J, Lowe K, Wolf SI (2010) Kinematic differences between normal
and low arched feet in children using the Heidelberg foot measurement method. Gait Posture 32
(1):1–5
Van den Herrewegen I, Cuppens K, Broeckx M, Barisch-Fritz B, Vander Sloten J, Leardini A,
Peeraer L (2014) Dynamic 3D scanning as a markerless method to calculate multi-segment foot
kinematics during stance phase: methodology and first application. J Biomech 47
(11):2531–2539
vanHoeve S, de Vos J, Weijers P, Verbruggen J, Willems P, Poeze M, Meijer K (2015) Repeatability
of the Oxford foot model for kinematic gait analysis of the foot and ankle. Clin Res Foot Ankle
3:171
Wang R, Thur CK, Gutierrez-Farewik EM, Wretenberg P, Broström E (2010) One year follow-up
after operative ankle fractures: a prospective gait analysis study with a multi-segment foot
Wang Y, Wong DW, Zhang M (2016) Computational models of the foot and ankle for
pathomechanics and clinical applications: a review. Ann Biomed Eng 44(1):213–221
Whittaker EC, Aubin PM, Ledoux WR (2011) Foot bone kinematics as measured in a cadaveric
robotic gait simulator. Gait Posture 33(4):645–650
Wolf P, Stacoff A, Liu A, Nester C, Arndt A, Lundberg A, Stuessi E (2008a) Functional units of the
human foot. Gait Posture 28(3):434–441
Wolf S, Simon J, Patikas D, Schuster W, Armbrust P, Döderlein L (2008b) Foot motion in children
shoes: a comparison of barefoot walking with shod walking in conventional and flexible shoes.
Woodburn J, Turner DE, Helliwell PS, Barker S (1999) A preliminary study determining the
feasibility of electromagnetic tracking for kinematics at the ankle joint complex. Rheumatology
(Oxford) 38(12):1260–1268
Woodburn J, Nelson KM, Siegel KL, Kepple TM, Gerber LH (2004) Multisegment foot motion
during gait: proof of concept in rheumatoid arthritis. J Rheumatol 31(10):1918–1927
Wrbaskić N, Dowling JJ (2007) An investigation into the deformable characteristics of the human
foot using fluoroscopic imaging. Clin Biomech 22(2):230–238
Wright CJ, Arnold BL, Coffey TG, Pidcoe PE (2011) Repeatability of the modified Oxford foot
model during gait in healthy adults. Gait Posture 33(1):108–112
Wu WL, Su FC, Cheng YM, Huang PJ, Chou YL, Chou CK (2000) Gait analysis after ankle
arthrodesis. Gait Posture 11(1):54–61
Trunk and Spine Models for Instrumented
Gait Analysis
Robert Needham, Aoife Healy, and Nachiappan Chockalingam
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Optoelectronic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Soft Tissue Artifact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Two-Dimensional Modelling of the Spine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Three-Dimensional Modelling of the Spine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Kinematic Modeling of the Pelvis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Data Analysis Techniques and Clinical Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Abstract
There are several types of motion capture systems which can measure trunk and
spine movement as a part of gait analysis. These range from wearable sensors to
optoelectronic systems. This chapter focuses on models used within optoelec-
tronic systems and covers both two- and three-dimensional models. This chapter
while providing an outline of the current thorax and pelvis models highlights
novel concepts in terms of 3-dimensional clusters. Latest methods on data
analysis techniques using vector coding have been outlined which will facilitate
comprehensive reporting of the movement data.
Keywords
Spine models • Gait Analysis • Thorax model
R. Needham (*) • A. Healy • N. Chockalingam

Life Sciences & Education, Staffordshire University, Stoke On Trent, UK
e-mail: r.needham@staffs.ac.uk; a.healy@staffs.ac.uk; N.Chockalingam@staffs.ac.uk

DOI 10.1007/978-3-319-30808-1_29-1
2 R. Needham et al.
Introduction
Imaging modalities such as X-rays and computer tomography scans are considered
by clinicians as a criterion reference for the assessment of spine posture. While it is
possible to gain a three-dimensional representation of the spine from conventional
scans, two-dimensional imaging is considered a suitable assessment technique for
clinical practice. As both techniques cannot be applied in a dynamic situation such as
gait due to a restriction in the capture volume, these assessment methods are not
suitable for routine clinical use. Also, repeated exposure to radiation means these
techniques could be harmful to a patient. In addition, planning and assessing
treatment outcomes based on postural observations and movement assessments of
static standing posture disregards the coordination between body segments, move-
ment variability, and the motor control strategies chosen to perform activities of daily
living.
Due to strong evidence to suggest that structural alterations of the spine may not
necessarily lead to a condition such as low back pain, recent clinical guidelines
strongly recommend against using imaging techniques (Chou et al. 2007). It is
proposed that unless a serious condition is suspected, clinicians should focus on
assessing and understanding the severity of functional deficits that individuals
experience when performing activities of daily living (Chou et al. 2007). An
alternative noninvasive method to assess function of the spine and back is therefore
highly desired; one that can accurately and reliably track movement allows for the
physiological loading that is experienced during dynamic movements and has an
established practicality to support clinical management strategies.
Several types of motion capture systems exist which can measure trunk and spine
movement over time. These range from wearable sensors (electromagnetic tracking
systems, electro-goniometers, inertial sensors, and ultrasound transmitters) to opto-
electronic systems (please refer to the section “Medical Application: Assessment of
Kinematics” for further information). While there are known limitations for all types
of motion capture systems (Don et al. 2012), advances in optoelectronic technology
and improved calibration protocols offer a measurement accuracy to less than 1 mm
(Levine et al. 2012). Optoelectronic marker-based systems are considered to be the
gold standard for movement analysis (Cappozzo et al. 2005) and are often used as a
criterion measure to validate wearable sensors.
Marker-based systems may not be suitable for standard clinical assessments, due
to the time required for participant preparation (marker application) and for data
collection, processing, and analysis (Lee 2002). Nevertheless, when integrated with
other movement analysis technologies (i.e., force plates, electromyography),
marker-based systems can detail the complexity of human gait (please refer to the
section “Methods and Models: Data Analysis” for further information). Thus, the
research that is conducted in gait analysis laboratories that employ marker-based
systems can inform clinical practice strategies and aid in the development, design,
Trunk and Spine Models for Instrumented Gait Analysis 3
and implementation of rehabilitation interventions (McGinley et al. 2009; Baker

2006).
The design and validation of kinematic models for the assessment of pelvis and
lower limb movement are evident in the research literature and demonstrate pro-
gression toward an advanced understanding of human movement (Manal et al. 2000;
Cappozzo et al. 1995, 1997) On the other hand, a kinematic model/technique to
quantitatively measure movement of the trunk and regions within the trunk has yet to
be agreed upon in the scientific community.
The trunk has been termed the “passenger unit” (Perry 1992; Perry and Burnfield
2010) that moves in response to actions of the lower limbs (Crosbie et al. 1997a;
Thorstensson et al. 1984). Nonetheless, since the trunk, head, and upper limbs can
account for two-thirds of an individual’s body mass, lower limb movement patterns
associated with pathological gait can induce compensatory movements in the trunk
region that can have a major influence on gait dynamics. There is empirical evidence
to suggest that the coordinated interaction between the spine, pelvis, and lower limbs
is essential to maintain balance and to achieve a smooth and efficient gait
(MacWilliams et al. 2013; Cappozzo 1983; Van Emmerik et al. 2005).
State of the Art
Recently, there have been several advances in terms of technologies and techniques
involved in motion capture resulting in sophisticated and accurate gait analysis.
However, most of these have been on lower limb analysis, and currently there is no
specific trunk and spine model which is widely adopted for use within gait and
posture analysis during gait. While there are several reported models, there is little
agreement between them, and their reproducibility is questionable because of the
lack of detail. Recent paper by Needham et al. (2016b) has produced a review which
explains the state of the art and outlines the requirements for a good marker set.
Another paper outlines a marker set which has been shown to be repeatable and valid
(Needham et al. 2016a). In terms of the thorax model, this research provides
evidence and informs that we should consider moving away from using the conven-
tional thorax model. Improvements in motion capture technology should allow for
more complex spine models. This in turn will provide further insight into
intersegmental movement. One should also consider alternative approaches to
models such as the use of 3D clusters.
Optoelectronic Systems
Optoelectronic systems capture human movement using markers that are placed on
anatomical landmarks. In biomechanics, markers are usually attached to the skin and
can be referred to as either passive or active. Passive markers are covered in
4 R. Needham et al.
retroreflective material that is tracked by infrared cameras. Active markers produce

an infrared signal (light-emitting diodes). This chapter will focus on the application
of passive markers.
Soft Tissue Artifact
The primary concern when using marker-based systems is soft tissue artifact (STA),
whereby skin-mounted markers can displace away from identified bony landmarks
during dynamic movements, thus having an impact on segmental angle calculations.
There are two types of error that can have an impact on marker position as a result of
STA. The first is relative error, which observes the relative movement between
markers that describe a rigid segment. The second is absolute error, which examines
the movement of a marker against the anatomical landmark it is overlying (Richards
2008). A comparison between angle data from indwelling bone pins and skin-
mounted markers has provided an estimate of the degree relative, and absolute
error influences segment/joint kinematics of the lower extremities (Peters et al.
2010). Bone pins have been utilized to quantify 3D motion of the lumbar vertebrae
during gait (Rozumalski et al. 2008; MacWilliams et al. 2013). While simultaneous
measurements to skin-mounted markers were not considered, these results can be
used for reference and comparative purposes (Needham et al. 2016a). Advancements
in mathematical modeling techniques could eventually compensate for the degree of
STA experienced during movement and provide an appropriate estimation of bone
pose position and joint/segmental kinematics (Lu and O’Connor 1999; Cappello
et al. 2005; Cappozzo et al. 1997). For detailed information on methods to assess and
compensate for STA, readers are directed elsewhere (Leardini et al. 2005).
Two-Dimensional Modelling of the Spine
Attaching markers on several spinous processes of the vertebrae provides a measure

of trunk inclination in the sagittal and frontal plane. Various marker configurations
have been developed using this approach (Frigo et al. 2003; Chockalingam et al.
2003; Syczewska et al. 1999). The configuration used will depend on the either the
research question or the model application in a clinical setting. For instance, if there
is a requirement to assess the trunk region as a rigid segment, a line between the
seventh cervical spinous process (C7) and sacrum could be used (please refer to the
section “Methods and Models: Rigid Body Modeling” for further information). As
stated previously, this provides movement data in two dimensions. Therefore, a
representation of axial rotation of the entire trunk could be defined by a projection of
a line between the left and right acromion process (outlines shoulder girdle) and a
line between left and right posterior-superior iliac spines (PSIS) onto a horizontal
plane (Frigo et al. 2003).
Fig. 1. Segment angle to global reference frame (a) proposed by Frigo et al. (2003) (b) proposed
by Frigo et al. (2003); Heyrman et al. (2013) (c) proposed by Heyrman et al. (2014). C7, seventh
cervical vertebra; T2, second thoracic vertebra; T6, sixth thoracic vertebra; T10, tenth thoracic
vertebra; L1, first lumbar vertebrae; L3, third lumbar vertebrae; L5, fifth lumbar vertebrae
A direct line between two adjacent markers that are attached to the spinous
processes can be projected onto the principle planes with respect to a global
reference frame, i.e., the vertical or the horizontal, providing a segmental angle
measure (Fig. 1a). This approach has been used to study segmental movements of
the spine during gait. Attaching markers on the spinous processes along the length of
the spine, Syczewska et al. (1999) reported that during gait spine segments display
lateral flexion toward the opposite side at initial contact in comparison to segments
above a spinal level of the seventh thoracic vertebrae (T7). With further evidence that
demonstrates intersegment movement of the thoracic spine (Needham et al. 2016a;
Crosbie et al. 1997b), these findings question the practicality of thorax segment that
is defined below the level of the sixth thoracic vertebrae (T6). This information when
combined with knowledge of the functional workings of the spine, i.e., facet joints,
would support the development of a defined three-dimensional thorax model.
Planar projections between adjacent segments offer an assessment of
intersegmental movement. This approach provides an understanding of the compen-
satory movements in the trunk region as a result of a spinal condition such
as scoliosis or a related lower limb condition that can influence spine movement,
i.e., leg length discrepancy. If the appropriate spinous processes are used, planar
projections can assess static and dynamic spine posture. For example, a planar
projection using three markers can provide lordosis angle in the sagittal plane
(Frigo et al. 2003) (Fig. 1b). Four markers that define two segments that are not
6 R. Needham et al.
Fig. 2. Marker placements for thorax model according to Baker (2013) (a) and Wu et al. (2005)
(ISB) (b) Anatomical landmarks SJN, deepest point of incisura jugularis; SXS, xiphoid process; C7,
seventh cervical vertebra; T2, second thoracic vertebra; T8, eighth thoracic vertebra; T10, tenth
thoracic vertebra
adjacent to each other can be used to represent sagittal plane angle of kyphosis
(Heyrman et al. 2014) (Fig. 1c).
Three-Dimensional Modelling of the Spine
It is the arrangement of markers in accordance with the kinematic model recom-

mendations that define body segments. A minimum of three non-collinear markers
are required to define the position and orientation of a body segment in three-
dimensional space (3D). As mentioned previously, two-dimensional movement
can be analyzed by a direct line between two markers (M1 and M2). To define
axial rotation of a segment, a third non-collinear marker (M3) is required that does
not lie on the line directly between M1 and M2 (Fig. 2a). Readers are directed to a
paper by Wu et al. (2005) for a detailed example on defining a thorax segment and
respective coordinate system.
Markers fixed to the manubrium, spinous process of the second thoracic vertebrae
(T2), and the midpoint between the PSIS is a nonlinear marker set that can represent
a 3D trunk model (Baker 2013). Landmarks on the shoulder girdle can also be
incorporated into a defined 3D trunk model (Rab et al. 2002). However, markers that
define the center link of the trunk and are attached to the thorax segment are favored
due to the movement artifact imposed by the upper limb movement associated with
using acromion landmarks (Nguyen and Baker 2004).
The thorax and abdomen are the two regions associated with the trunk. While
substantial evidence on thorax movement is documented in the research literature,
there is a lack of agreement on which kinematic model best defines the thorax
segment. The conventional gait model typically defines a 3D thorax segment using
anatomical landmarks on the proximal and distal end of the sternum along with the
spinous processes of the spine. This approach is based on the International Society
of Biomechanics (ISB) guidelines (Wu et al. 2005) (Fig. 2b). The Plug-in-Gait
thorax model (Vicon, OMG, UK) uses a similar approach to the ISB thorax model,
although the tenth thoracic vertebrae (T10) defines the posterior-distal aspect of the
thorax instead of the eighth thoracic vertebrae (T8). Since the position of C7 can be
influenced by movement of the head and neck (Armand et al. 2014) and represents
the cervical region of the spine, the spinous process of the second thoracic
vertebrae (T2) should be used as the posterior-superior landmark of a thorax
segment.
Alignment with the distal apex of each scapula (defined as MAI) has been
suggested to be a suitable landmark to define the posterior-distal aspect of a thorax
segment (Leardini et al. 2011). Since the MAI landmark is approximately at a level
of the ninth thoracic vertebrae (T9) (Haneline et al. 2008), and that similar thorax
movements during gait have been noted when either T8 or T10 was included as part
of the model (Armand et al. 2014), the MAI could be a viable landmark. Differences
in scapula position and spinal postures between individuals need to be considered
before the MAI landmark is used as part of the thorax pose estimation.
So far thorax models have been defined based on four markers, two on the
sternum and two on the spine. Defining a thorax segment that requires a marker to
be placed on the distal end of the sternum could have practical concerns for female
participants. A minimal marker set using the T2, T10, and manubrium can define a
3D thorax segment (Baker 2013) (Fig. 2a). Consistent thorax movements have been
demonstrated using this approach (Armand et al. 2014).
Unlike the thorax region, there are no anatomical landmarks on the anterior side of
the abdomen to support a 3D lumbar segment. To define axial rotation in the transverse
plane in the lumbar region, additional markers on either side of the spinous processes
can be applied (Crosbie et al. 1997a; Seay et al. 2008). These lateral markers could be
subjected to greater soft tissue artifact than those attached to the spinous processes.
Elastikon elastic tape wrapped around the lumbar region could be a possible solution to
minimize the influence of soft tissue artifact (Seay et al. 2008; Mason et al. 2016).
Alternatively, a 3D cluster could be used. This cluster consists of at least three markers
in a nonlinear configuration. These are normally attached to a rigid or semirigid base
which can be applied to relevant anatomical landmarks on the back (Fig. 3a, b). When
implemented in the same gait laboratory, consistent and reliable measurements of
lumbar spine movement have been reported using the 3D cluster technique (Schache
et al. 2002; Needham et al. 2016a,b; Taylor et al. 1996; Thurston 1982). With applica-
tion to track thorax motion (Mason et al. 2016; Seay et al. 2011), the 3D cluster has
recently been shown to produce similar patterns of movement and range of motion in
comparison to the conventional thorax model (Fig. 3a, b).
8 R. Needham et al.
Fig. 3. (a) Anatomical landmarks used in the IOR thorax model (T2, second thoracic vertebra; T8,
eighth thoracic vertebra; SJN, deepest point of incisura jugularis; SXS, xiphoid process) and 3D
cluster on T1 (first thoracic vertebra); (b) 3D cluster structural dimensions. Unless stated measure-
ments are in millimeters
Kinematic Modeling of the Pelvis
According to the International Society of Biomechanics (ISB) guidelines (Wu et al.

2002), the coordinate system of a pelvis segment is defined by skin-mounted
markers placed on the anterior and posterior landmarks of the ilium. This marker
configuration is an approach commonly employed in clinical gait analysis (Baker
2013). It has been shown that markers attached to the anterior superior iliac spines
(ASISs) are influenced more by STA during gait than markers which are attached to
the posterior-superior iliac spines (Hara et al. 2014). The effect of STA on pelvis
kinematics would be more of an issue for those individuals with a higher body mass
index and excessive soft tissue in the abdominal region. The arms could occlude the
optoelectronic camera’s view of ASIS markers while walking. In light of the
concerns of using ASIS landmarks in pelvis kinematic modelling, alternative tech-
niques have been proposed. These include the use of individual markers placed on
the posterior side of the ilium and sacrum (Frigo et al. 1998) or by attaching a rigid
cluster of markers onto the sacrum region (Borhani et al. 2013). In addition, with
advancements in biomechanical software, this has presented an opportunity to use
virtual markers to support the creation of a segment and aid in the tracking of
segmental movement (Kisho Fukuchi et al. 2010; McClelland et al. 2010). As
there are a number of techniques available, an appreciation of the differences
between kinematic modeling approaches is essential when interpreting relative

movement between the trunk region and a pelvis segment (Needham et al. 2016b).
Data Analysis Techniques and Clinical Outcomes
Normally, trunk and spine movement will be analyzed with an established pelvis and
lower limb marker set. Once a reliable model is used to collect the data, there are
several reported methods for data analysis, and reporting. Depending on the type of
clinical condition and use of a dataset, an appropriate method could be adopted. It
might range from measuring and reporting a simple lateral flexion or forward
bending to the assessment of rotation in specific clinical conditions or sporting
situations/physical activity.
Movement estimated from a 3D thorax model is often reported relative to a pelvis
segment or to a laboratory location, although the former is the most common
approach used by the scientific community. Kinematic waveforms and range of
motion values are interpreted by a comparison against a reported normative data-
base. Since the 3D thorax model (Wu et al. 2005) is associated with the conventional
gait model, a substantial database on relative movement between the thorax and pelvis
is available. This includes the use of subgroup classifications allowing for comparisons
by age, by gender, and in some instances by clinical condition (e.g. cerebral palsy and
low back pain (LBP)). In contrast to the 3D thorax model, there is scarce data on
alternative models/methods outlined previously in this chapter.
Reporting discrete kinematic values such as range of motion and peak angle
values alone can be misleading when unaccompanied by kinematic waveforms.
During gait for instance, while no difference in lumbar spine range of motion has
been noted between individuals with chronic LBP and those considered healthy,
differences in the kinematic waveforms were evident between groups, and this was
due to the variability in the stride-to-stride movement (Vogt et al. 2001). A similar
observation was reported in a study that examined lumbar kinematic variability
during gait in participants with chronic LBP before and after a 12-week isolated
lumbar extension exercise intervention (Steele et al. 2015).
Subtle differences in range of motion and the timing of rotations in different
directions of two defined segments influence the reporting of relative movement. To
support the interpretation of a kinematic waveform that details relative movement, it
is strongly advised that global waveforms (defined as movement of a segment
relative to a laboratory location) of each respective segment is also documented.
Examples of this type of interpretation are provided in a recent systematic review
(Needham et al. 2016b). Alternatively, one should consider using novel data analysis
techniques which has the capability to quantify the coordination pattern between two
segments and can assess the variability of this interaction over time (Hamill et al.
2012). A modified vector coding technique is one approach, and while it is beyond
10 R. Needham et al.
the scope of this chapter to explain the calculations and coordination classification
system of the vector coding technique, there are published papers which outlines this
data analysis approach (Needham et al. 2014, 2015).
Future Directions
There has been a lot of activity in the recent past with various papers reporting
several types of marker sets for spine, trunk, and pelvis analysis. Whilst these data
sets provide a good debate, there has to be a consensus not only on the model and
methodology but also the reporting procedures, if any of these approaches are to be
adopted for clinical use. In addtion, these papers have to provide clear detail on
kinematic data reporting. For example details on whether it is intersegmental data or
the range of movement of a specific segment in a specific plane, which will help in
designing effective clinical management. The intersegmental data on the thorax is
still scarce, and further work needs to be completed to understand the movement in
normative gait and in various clinical conditions. In addition, new data analyses
techniques outlined within this chapter could be employed to help understand these
conditions in detail. Furthermore, these types of data will fuel innovation in the
development of orthotics or prosthetics.
References
Armand S, Sangeux M, Baker R (2014) Optimal markers’ placement on the thorax for clinical gait
analysis. Gait Posture 39(1):147–153
Baker R (2006) Gait analysis methods in rehabilitation. J Neuroeng Rehabil 3(1):4
Baker R (2013) Measuring Walking: A Handbook of Clinical Gait Analysis. Mac Keith Press,
Cambridge
Borhani M, McGregor AH, Bull AMJ (2013) An alternative technical marker set for the pelvis is
more repeatable than the standard pelvic marker set. Gait Posture 38(4):1032–1037
Cappello A et al (2005) Soft tissue artifact compensation in knee kinematics by double anatomical
landmark calibration: performance of a novel method during selected motor tasks. IEEE Trans
Biomed Eng 52(6):992–998
Cappozzo A (1983) The forces and couples in the human trunk during level walking. J Biomech 16
(4):265–277
Cappozzo A et al (1995) Position and orientation in space of bones during movement: anatomical
frame definition and determination. Clin Biomech (Bristol, Avon) 10(4):171–178
Cappozzo A et al (1997) Surface-marker cluster design criteria for 3-D bone movement reconstruc-
tion. IEEE Trans Biomed Eng 44(12):1165–1174
Cappozzo A et al (2005) Human movement analysis using stereophotogrammetry. Part 1: theoret-
ical background. Gait Posture 21(2):186–196
Chockalingam N et al (2003) A comparison of three kinematic systems for assessing spinal range of
movement. Int J Ther Rehabil 10(9):402–407
Chou R et al (2007) Diagnosis and treatment of low back pain: a joint clinical practice guideline
from the American College of Physicians and the American Pain Society. Ann Intern Med 147
(7):478–491
Crosbie J, Vachalathiti R, Smith R (1997a) Age, gender and speed effects on spinal kinematics
during walking. Gait Posture 5(1):13–20
Crosbie J, Vachalathiti R, Smith R (1997b) Patterns of spinal motion during walking. Gait Posture 5
(1):6–12
Don R et al (2012) Instrumental measures of spinal function: is it worth? A state-of-the art from a
clinical perspective. Eur J Phys Rehabil Med 48(2):255–273
Frigo C et al (1998) Functionally oriented and clinically feasible quantitative gait analysis method.
Med Biol Eng Comput 36(2):179–185
Frigo C et al (2003) The upper body segmental movements during walking by young females. Clin
Biomech (Bristol, Avon) 18(5):419–425
Hamill J, Palmer C, Van Emmerik REA (2012) Coordinative variability and overuse injury.
SMARTT 4(1):45
Haneline MT et al (2008) Determining spinal level using the inferior angle of the scapula as a
reference landmark: a retrospective analysis of 50 radiographs. J Can Chiropr Assoc 52
(1):24–29
Hara R et al (2014) Quantification of pelvic soft tissue artifact in multiple static positions. Gait
Posture 39(2):712–717
Heyrman L, Feys H, Molenaers G, Jaspers E, Van de Walle P, Monari D, Aertbeliën E, Desloovere
K (2013) Reliability of head and trunk kinematics during gait in children with spastic diplegia.
Heyrman L et al (2014) Altered trunk movements during gait in children with spastic diplegia:
compensatory or underlying trunk control deficit? Res Dev Disabil 35(9):2044–2052
Kisho Fukuchi R et al (2010) Evaluation of alternative technical markers for the pelvic coordinate
system. J Biomech 43(3):592–594
Leardini A et al (2005) Human movement analysis using stereophotogrammetry. Part 3. Soft tissue
artifact assessment and compensation. Gait Posture 21(2):212–225
Leardini A et al (2011) Multi-segment trunk kinematics during locomotion and elementary exer-
cises. Clin Biomech (Bristol, Avon) 26(6):562–571
Lee R (2002) Measurement of movements of the lumbar spine. Physiother Theory Pract 18
(4):159–164
Levine D, Richards J, Whittle MW (2012) Whittle’s Gait Analysis, fifth ed. Churchill Livingstone,
London
optimisation with joint constraints. J Biomech 32(2):129–134
MacWilliams BA et al (2013) Assessment of three-dimensional lumbar spine vertebral motion
during gait with use of indwelling bone pins. J Bone Joint Surg (Am Vol) 95(23):e1841–e1848
Manal K et al (2000) Comparison of surface mounted markers and attachment methods in
estimating tibial rotations during walking: an in vivo study. Gait Posture 11(1):38–45
Mason DL et al (2016) Reproducibility of kinematic measures of the thoracic spine, lumbar spine
and pelvis during fast running. Gait Posture 43:96–100
McClelland JA et al (2010) Alternative modelling procedures for pelvic marker occlusion during
motion analysis. Gait Posture 31(4):415–419
McGinley JL et al (2009) The reliability of three-dimensional kinematic gait measurements: a
systematic review. Gait Posture 29(3):360–369
Needham R, Naemi R, Chockalingam N (2014) Quantifying lumbar-pelvis coordination during gait
using a modified vector coding technique. J Biomech 47(5):1020–1026
Needham RA, Naemi R, Chockalingam N (2015) A new coordination pattern classification to
assess gait kinematics when utilising a modified vector coding technique. J Biomech 48
(12):3506–3511
Needham R, Naemi R, Healy A, Chockalingam N (2016a) Multi-segment kinematic model to
assess three-dimensional movement of the spine and back during gait. Prosthet Orthot Int 40
(5):624–635
Needham R, Stebbins J, Chockalingam N (2016b) Three-dimensional kinematics of the lumbar
spine during gait using marker-based systems: a systematic review. J Med Eng Technol 40
(4):172–185
12 R. Needham et al.
Nguyen TC, Baker R (2004) Two methods of calculating thorax kinematics in children with
myelomeningocele. Clin Biomech 19(10):1060–1065
Perry J (1992) Gait Analysis: Normal and Pathological Function. SLACK Incorporated
Perry J, Burnfield J (2010) Gait analysis: normal and pathological function, 2nd ed. Thorofare, NJ:
SLACK Incorporated
Peters A et al (2010) Quantification of soft tissue artifact in lower limb human motion analysis: A
systematic review. Gait Posture 31(1):1–8
Rab G, Petuskey K, Bagley A (2002) A method for determination of upper extremity kinematics.
Richards J (2008) Biomechanics in Clinic and Research. Churchill Livingstone, London
Rozumalski A et al (2008) The in vivo three-dimensional motion of the human lumbar spine during
Schache AG et al (2002) Intra-subject repeatability of the three dimensional angular kinematics
within the lumbo-pelvic-hip complex during running. Gait Posture 15(2):136–145
Seay J, Selbie WS, Hamill J (2008) In vivo lumbo-sacral forces and moments during constant speed
running at different stride lengths. J Sports Sci 26(14):1519–1529
Seay JF, Van Emmerik REA, Hamill J (2011) Influence of low back pain status on pelvis-trunk
coordination during walking and running. Spine 36(16):E1070–E1079
Steele J et al (2015) A randomized controlled trial of the effects of isolated lumbar extension
exercise on lumbar kinematic pattern variability during gait in chronic low back pain. PM R 8
(2):105–114
Syczewska M, Oberg T, Karlsson D (1999) Segmental movements of the spine during treadmill
walking with normal speed. Clin Biomech (Bristol, Avon) 14(6):384–388
Taylor NF, Evans OM, Goldie PA (1996) Angular movements of the lumbar spine and pelvis can be
reliably measured after 4 minutes of treadmill walking. Clin Biomech (Bristol, Avon) 11
(8):484–486
Thorstensson A et al (1984) Trunk movements in human locomotion. Acta Physiol Scand 121
(1):9–22
Thurston AJ (1982) Repeatability studies of a television/computer system for measuring spinal and
pelvic movements. J Biomed Eng 4(2):129–132
Van Emmerik REA et al (2005) Age-related changes in upper body adaptation to walking speed in
human locomotion. Gait Posture 22(3):233–239
Vogt L et al (2001) Influences of nonspecific low back pain on three-dimensional lumbar spine
kinematics in locomotion. Spine 26(17):1910–1919
Wu G et al (2002) ISB recommendation on definitions of joint coordinate system of various joints
for the reporting of human joint motion – part I: ankle, hip, and spine. J Biomech 35(4):543–548
Wu G et al (2005) ISB recommendation on definitions of joint coordinate systems of various joints
for the reporting of human joint motion – part II: shoulder, elbow, wrist and hand. J Biomech 38
(5):981–992
Upper Extremity Models for Clinical
Movement Analysis
Andrea Giovanni Cutti, Ilaria Parel, and Andrea Kotanxis
Abstract
The quantitative analysis of upper-extremity motion is a challenging task. A
single, universally accepted methodology does not exist, but it is possible to
define a standardized way to report a measurement protocol and to formulate
recommendations on the most important aspect. The aim of this chapter is to
provide such guidelines, addressing common issues such as joint modeling,
scapula tracking, soft-tissue artifact compensation, and summary of results.
Keywords
Upper-extremity • Shoulder • Scapula • Elbow • Functional axis • Motion
analysis • Kinematics • Optoelectronic system • Inertial sensors • Protocol •
Measurement methods • Scapulo-humeral rhythm • Marker-set • Coordinate
system • Anatomical landmarks • Functional frames • Cluster • Skin artifacts
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Guidelines for Upper-Limb Motion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
A Framework for the Definition of Standardized Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Joints/Segments of Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Mechanical Model of Joints/Degrees of Freedom of Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Definition of Segment or Joint CSs and Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
A.G. Cutti (*)

Applied Research, INAIL Prosthetic Center, Vigorso di Budrio, BO, Italy
e-mail: ag.cutti@inail.it
I. Parel
Unit of Shoulder and Elbow Surgery, Cervesi Hospital, Cattolica, RN, Italy
e-mail: ilaria.parel@auslromagna.it
A. Kotanxis
Leon Root Motion Analysis Laboratory, Hospital for Special Surgery, New York, NY, USA
e-mail: kontaxisa@hss.edu

DOI 10.1007/978-3-319-30808-1_30-1
2 A.G. Cutti et al.
Marker/Sensor Set-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Activities to Be Measured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Extraction of Summarizing Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Upper-Extremity Models for Inertial and Magnetic Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Future Direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
State of the Art
The upper limb is composed of six body segments (thorax, clavicle, scapula,
humerus, forearm, and hand) and three main joints (shoulder, elbow, and wrist).
Upper-extremity models attempt to reconstruct segments’ attitude (position and
orientation) and joints’ kinematics during arm movements. This chapter will focus
on shoulder and elbow, and occasionally discuss about wrist and hand.
Quantitative motion analysis of the upper-extremity is not new, as described by
Anglin and Wyss (2000), and received a great impulse in the late 1980–early 1990,
in particular by the works produced in The Netherlands and UK, using the “palpator”
and electromagnetic sensors (Anglin and Wyss 2000; Johnson and Anderson 1990;
Van der Helm et al. 1992; Van der Helm ad Pronk 1995; Pronk and van der Helm
1991). Given the variety of tasks performed by the upper-extremity and the overall
complexity of the system, researchers adopted a broad range of instrumentation and
methods to address their specific needs.
Over the last 10 years, however, instrumentations tended to stabilize, with a
predominance of electromagnetic and stereophotogrammetric systems, thanks to
the availability of well-established commercial products (“3D Dynamic Pose Esti-
mation Using Reflective Markers or Electromagnetic Sensors”; “3D Dynamic Prob-
abilistic Pose Estimation From Data Collected Using Cameras and Reflective
Markers.”).
The situation for the models used to extract biomechanically meaningful param-
eters from the instrumentation (“protocols” in short) is partially different. As
opposed to clinical gait analysis, in which the “Plug-in gait” protocol (or its slight
derivations) has reached enough popularity to be named “conventional”, this is not
completely the case for the upper-extremity.
Establishing a standardized protocol is a delicate matter and requires the com-
pletion of a detailed process. The application of motion analysis cannot be reduced to
the “where do I apply the markers?” type of questions, but it starts from the clinical
problem, passes through the definition of a mathematical model of the body, the
construction of coordinate systems (CSs), and then markers start playing a role on
how to track the CSs.
In the remaining of the chapter we will explain these steps in details, by recalling
the outline proposed by Kontaxis et al. (2009). This will give the framework to
summarize which are the available recommendations. Whenever possible, we will
Upper Extremity Models for Clinical Movement Analysis 3
refer to the proposals formulated by the International Society of Biomechanics

(ISB), or by the International Shoulder Group (ISG), one of ISB’s technical groups.
Following this outline, sections “Definition of Segment or Joint CSs and Angles”
and “Marker/Sensor Set-Up” will report about models easily applicable with any
electromagnetic or stereophotogrammetric system. However, these might be difficult
or cumbersome to apply to the (now popular) inertial & magnetic measurement
systems (IMMS or IMUs). IMMS are wearable system containing accelerometers,
gyroscopes, and possibly magnetometers and barometers that, through sensor fusion
techniques, provide information about the orientation of the sensors themselves
(“Three-dimensional Human Kinematics Estimation Using Magneto-Inertial Mea-
surement Units”). The ISB/ISG standard for the CSs is hardly applicable to these
sensors, and a recent meeting during the ISB Conference in Glasgow (2015) clearly
showed the interest of the scientific community to formulate recommendations for
this emerging technology. A few protocols have been proposed so far on this topic,
but at present the most clinically validated is the ISEO protocol described in Cutti
et al. (2008). To introduce the reader to these new topics, section “Upper-Extremity
Models for Inertial and Magnetic Sensor” will provide a brief overview of ISEO
(Parel et al. 2016).
Guidelines for Upper-Limb Motion Analysis
A Framework for the Definition of Standardized Protocols
Kontaxis et al. (2009) proposed a framework to support the definition of standard-

ized protocols, their description, and the formulation of general recommendations.
The framework is not only useful for developers, but it also represents a useful
checklist for the young biomechanist or the clinician who wants to verify what is
implemented in the commercial system available in the laboratory and make sure
that the system is really suitable for the (clinical) needs.
The steps for a proper definition of a standardized protocol are summarized in
Fig. 1. The framework is composed of two nested flowcharts. The first flowchart (left
panel) defines what is meant by a motion analysis study and the role of a motion
analysis protocol. The second flowchart (central panel) specifies the steps required to
build a motion analysis protocol appropriate to the study design. On the right hand
panel, a summary of basic recommendations and current standards is reported,
together with a list of open issues.
The first flowchart is important because it reminds the basics of any scientific
approach: the first step is a clear definition of the research question formulated in
terms of hypotheses, simple and objective. This is a challenging task, because
clinical problems can be multifactorial and curiosity can push us to formulate too
many questions and ultimately ask too much to the patients/subjects. Good hypoth-
eses will naturally indicate the parameters to measure and this will be the goal of our
protocol. The protocol will be applied on subjects and patients to test the hypotheses
4 A.G. Cutti et al.
Fig. 1 Flowchart for the definition of a motion analysis protocol, based on Kontaxis et al. 2009
and possibly confirm them. The “protocol” is therefore a tool that should be simple
and specific enough to provide the data to solve our questions.
Following this line of reasoning, it might be objected that each clinical question
could deserve a special, different protocol. While we think that if a single marker
solves the problem that is the way to pursue, the researcher should keep in mind that
replication of experiments and confirmation of results by independent centers is the
foundation of science. This is especially true in the motion analysis field, in which
the number of subjects typically recruited is incomparably smaller than in the

pharmaceutical field: no standardization, limited comparability, no common data-
bases, and difficult meta-analysis. Moreover, patients and clinicians called to make
clinical decisions on the reports provided by motion analysis laboratories should be
confident in the quality of the data supplied. Therefore, we strongly believe that a
basic protocol for the upper-extremity should be looked for.
The second flowchart examines the construction of a motion analysis protocol
and next to it, suggestions are made. These points will be commented separately in
sections “Joints/Segments of Interest,” “Mechanical Model of Joints/Degrees of
Freedom of Segments,” “Definition of Segment or Joint CSs and Angles,”
“Marker/Sensor Set-Up,” “Activities to be Measured,” and “Extraction of Summa-
rizing Parameters.”
Joints/Segments of Interest
It is important to define if segments and/or joints are of interest. As described in

Kontaxis et al., segment kinematics is defined as the attitude (position and orienta-
tion), of one bony segment with respect to either a global CS or a nonadjacent bone.
Joint kinematics is defined as the description of the relative attitude of two adjacent
bony segments.
Typical segments of interest are the thorax, clavicle, scapula, humerus, forearm,
and carpus of the hand. Axial rotations of the clavicle are almost impossible to track
noninvasively (Ludewig et al. 2004; Marchese and Johnson 2000; Teece et al. 2008).
For this reason, it was proposed to consider a fictitious segment linking sternum and
glenohumeral head and refer that as “shoulder girdle” (Garofalo et al. 2009). Typical
joints of interest are the sternoclavicular/girdle, scapulothoracic, glenohumeral,
elbow, and wrist.
By looking at the literature, it is common to describe elbow joint kinematics and
to have a mixed segment/joint description of the shoulder; both scapula and humerus
are usually referred to the thorax to describe the so-called scapulohumeral rhythm
(SHR), which is the coordinated movement between scapula and humerus while the
arm is elevated. Similarly, for the clavicle/girdle-humeral rhythm, with both seg-
ments referred to the thorax.
Mechanical Model of Joints/Degrees of Freedom of Segments
For the segments and joints of interest, it is essential to define the degrees of freedom
(DoFs), i.e., how they can move. A simple and commonly applied open-chain model
is reported in Fig. 2, focusing on joint kinematics. Segments are depicted as bars
connected by hinges, each representing a rotational degree of freedom. Only rota-
tional DoFs are assumed between segments because noninvasive motion analysis
does not allow a reliable description of translations within joints, given the relative
6 A.G. Cutti et al.
Fig. 2 A simple biomechanical model of the upper-extremity
movement of the sensors positioned on the skin with respect to the underlying bones
(the so-called skin artifact or soft-tissue artifact).
The order in which the hinges are connected and named is very important.
Different sequences will bring different descriptions of the joint/segment orientation
for the same movement, as reported in Karduna et al. (2000). The description
provided in this step will need to be consistent with the Euler decomposition chosen
in the next step (section “Definition of Segment or Joint CSs and Angles”). None of
the sequences is wrong or correct per se, but some sequences are closer to the typical
“clinical language” or easier to interpret, and for this reason, they are most frequently
used or recommended. Specifically, the sequence in Fig. 2 is the one recommended
in Kontaxis et al. (2009) for shoulder movements in (mostly) the sagittal plane. It
changes for frontal plane movements into abduction–adduction, flexion–extension,
and internal–external rotation. The one preferred by the ISB is different: selection of
the plane of elevation, elevation, and internal–external rotation. Kontaxis and
coworkers made a different suggestion compared to ISB to avoid the gimbal-lock
position (undefined decomposition) when the humerus is vertical (down or up). This
comes to the expense of a different decomposition based on the predominant plane
of movement, which is not present in the ISB recommended sequence. This topic is
further discussed in the section “Definition of Segment or Joint CSs and Angles.”
Interestingly, motion analysis data can be used as input to complex upper-
extremity models, such as those implemented in OpenSim. OpenSim is a freely
available and open-source software (Delp et al. 2007). Users can develop models of
musculoskeletal structures and create dynamic simulations of a wide variety of
movements. In the last years, the OpenSim community has published and tested
several upper-extremity models, and recently, a new OpenSim Shoulder Model has
been developed with an improved description of the scapulothoracic and
glenohumeral joints (Seth et al. 2016).
Preferential axes of rotations of joints must be clearly incorporated and marked.
While scapulathoracic and glenohumeral joints are commonly considered with three
rotational DoF, this simplification is not supported by the literature considering the
elbow.
The elbow can be reasonably described as a double-hinge joint with non-
intersecting axes of rotation. The first hinge describes the rotation of ulna and radius
with respect to the humerus (flexion–extension), and the second describes the
movement of the radius around the ulna (pronation–supination). There is sometimes
confusion about the so-called carrying angle, and it is very important to distinguish
between the anatomical and biomechanical definition. The first depicts a purely
exterior situation, i.e., it states that the carrying angle is the angle between the long
axis of the humerus and the ulna, when the elbow is completely extended and
supinated, and decreases with elbow flexion; in the reference position, the carrying
angle ranges between 10 to 15 (Goto et al. 2004). The second definition states
that the carrying angle measures the relative orientation of the flexion–extension and
pronation–supination axes of the joint and therefore it is an (almost) constant value
which is subject specific (Stokdijk et al. 2000; Cutti et al. 2006a, b, 2008). Therefore,
the carrying angle is not affected by the amount of elbow flexion–extension
and pronation–supination. This second definition should always be preferred
in biomechanical studies focusing on the elbow joint.
The same holds for the wrist, in which an axial rotation of the hand relative to the
forearm is not reported.
Definition of Segment or Joint CSs and Angles
Local CSs are required to define the attitude of one segment relative to another or to
the laboratory CS. CSs must be defined consistently with the mechanical model.
As described by Kontaxis et al. (2009), when dealing with segment kinematics,
CSs should be representative of the anatomy of the bone. The ISB, in conjunction
with the ISG, proposed a set of such CSs that are commonly referred to as “ana-
tomical frames” (AFs). AFs are typically constructed using the position of bony
anatomical landmarks (ALs). A list of ALs and AFs is reported in Tables 1 and 2,
respectively. The centre of the glenohumeral head (GH) is an internal landmark
which cannot be directly identified by palpation. How to estimate it will be the topic
of section “Activities to be Measured.” The CS reported for the shoulder-girdle is not
part of the ISB proposal but comes from Garofalo et al. (2009).
It is important for the reader to know that the planes of an AF approximate the
frontal, transverse, and sagittal anatomical planes of the bone. Therefore, the AF
8 A.G. Cutti et al.
Table 1 Anatomical landmarks and functional axes used for the definition of the
anatomical/functional CSs. The last column reports the segment/cluster to which the anatomical
landmark/functional axis is typically associated to. Please refer to section “Marker/Sensor Set-Up”
for the definition of “cluster” and the alternative use with the ulnar cluster
Abbreviation Name Segment of reference/cluster
IJ Incisura jugularis Thorax
PX Xiphoid process Thorax
C7 7th Cervical vertebra Thorax
T8 8th Thorac vertebra Thorax
AA Angulus acromialis Scapula
TS Trigonum spinae Scapula
AI Angulus inferior Scapula
GH Centre of glenohumeral head Scapula
EL Lateral epicondyle Usual: Humerus
Alterative: H4 (see Table 2 and
section “Marker/Sensor Set-Up”)
EM Medial epicondyle Usual: Humerus
Alterative: H4 (see Table 2 and
section “Marker/Sensor Set-Up”)
RS Radial styloid Forearm
US Ulnar styloid Forearm
M3 3rd metacarpus Hand
Vflex Direction of the elbow flexion- Usual: Humerus
extension axis (pointing laterally) Alternative: Ulna
pflex Pivot point of the elbow flexion- Ulna
extension axis
Vflex Direction of the forearm prono- Forearm
supination axis (pointing proximally)
“anatomical axes” are only rough approximations of the real axes of rotation of the
joint(s) that the segment forms. As such, when anatomical axes are assumed to be the
axes about which a joint rotates, the resultant joint kinematics can be substantially
affected by kinematic cross-talk (Piazza and Cavanagh 2000; Cutti et al. 2006a, b),
i.e., part of the angular variation of a DoF is read on another DoF. This is why the
ISB standard (Wu et al. 2005) can be misleading when talking about Joint CSs (JCS)
and using anatomical axes to calculate the rotations of adjacent bones. While the
situation is relatively acceptable when joints are modeled as ball and sockets, the
matter becomes critical for joints such as the elbow, in which axes of rotations are not
aligned with the anatomical axes of the humerus and forearm. Similarly, this is
dangerous for the description of joints (girdle-thoraxic or wrist) in which the
kinematic model assumes a restricted number of DoF, i.e., not the full three rotational
DoF, as in Fig. 2.
To avoid the limitations above, the use of functional frames (FF) is recommended
for joints of interest. As defined in Kontaxis et al. (2009), a FF is a CS associated
with a segment and is specifically intended to describe the kinematics of a joint
formed by the segment. The FF is based on at least one functional axis of rotation of
Table 2 Definition of the anatomical and functional frames. Please refer to section “Marker/Sensor
Set-Up” for details on how to track CSs over time
Segment Axes
Thorax (THX) YTHX = ((IJ + C7) / 2 (PX + T8) / 2)) / k (IJ + C7) /
2 (PX + T8) / 2 k : longitudinal
XTHX = YTHX ^ (T8 PX) / k YTHX ^ (T8 PX) k : medio - lateral
ZTHX = XTHX ^ YTHX : antero - posterior
Origin: IJ
Shoulder girdle XGRD = (GH (IJ + C7) / 2) / k (GH (IJ + C7) /
(GRD) 2) k : medio - lateral
ZGRD = (XGRD ^ YTHX) / k k : medio - lateral
YGRD = (ZGRD ^ XGRD) / k k : upward
Origin: IJ
Scapula (SC) XSC = (AA TS)/ k AA TS k : medio - lateral
ZSC = (XSC ^ (AA AI) / k XSC ^ (AA AI) k : antero - posterior
YSC = (ZSC ^ XSC) / k k : longitudinal
Origin: AA
Proximal humerus YH1 = (GH E) / k GH E k : longitudinal
(H1) ZH1 = (YH1 ^ (EM EL)) / k YH1 ^ (EM EL) k : antero - posterior
XH1 = YH1 ^ ZH1 / k k : medio - lateral
E = (EL + EM)/2
Origin: GH
Proximal humerus YH2 = (GH E) / k GH E k : longitudinal
(H2) XH2 = (YH1 ^ YPS) / k YH1 ^ YPS k : antero - posterior
ZH2 = XH2 ^ YH2 / k k : medio - lateral
Origin: GH
Distal humerus XH3 = VFLEX / k VFLEX k : medio - lateral
(H3) ZH3 = XH3 ^ (GH REF) /
k XH3 ^ (GH REF) k : antero - posterior
YH3 = (ZH3 ^ XH3) / k ZH3 ^ XH3 k : longitudinal
REF = E (usual) or pflex(alternative when using ulnar cluster)
Origin = REF
Forearm F1 YF1 = (E – US) / ||(E – US)|| : longitudinal
ZF1 = (RS – US) ^ YF1 / ||(RS – US) ^ YF1|| : antero posterior
XF1 = YF1 ^ ZF1 : medio lateral
Origin = E
Proximal forearm YF = VPS / k VPS k : longitudinal
(F) ZF = ((RS US) ^ YF / k (RS SU) ^ YF k : antero - posterior
XF = (YF ^ ZF) / k k : medio - lateral
S = (US + RS) / 2
Origin: S
Hand YHN = (S M3) / k S M3 k : longitudinal
ZHN = (YHN ^ (US RS)) /
k YHN ^ (US RS) k : anterior - posterior
XHN = (YHN ^ ZHN) / k k : medio - lateral
Origin: M3
10 A.G. Cutti et al.
the joint, best expressed in an AF associated with the segment. A functional axis of a
joint is the axis of rotation of the distal and proximal segments that are forming the
joint, when these are actively or passively rotated relative to each other. For a pure
hinge joint, the functional axis coincides with the rotational axis of the hinge. For an
“almost hinge” joint (e.g., the elbow and the knee), the functional axis is assumed as
the mean axis of rotation, computed from the joint instantaneous axes of rotation. For
a ball and socket joint, there are no preferential axes of rotation. However, if the
distal segment is rotated in a constant plane, the functional axis is defined as the axis
perpendicular to the plane of rotation. Functional axes can be computed through a
number of algorithms. One of the most commonly used is based on the estimation of
Instantaneous Helical Axes (Woltring 1990) which expresses the mean axis of
rotation of the joint of interest in the AF of the corresponding segment of interest.
Other algorithms exist which also proved to be effective (Halvorsen et al. 1999;
Gamage and Lasenby 2002). With the functional axis expressed in the AF of the
segment of interest, FF can then be constructed (1) by using a combination of more
functional axes of rotation or (2) by using a combination of the functional and
anatomical axes. When computing the kinematics of the joint, the functional axis is
assumed to be the axis around which joint rotations occur. This minimizes the
kinematic cross-talk.
Table 2 reports a recommended set of FFs to track the elbow: one for the distal
humerus and one for the proximal forearm. It is therefore recommended to use two
systems associated to humerus and forearm (the use of H3 is further discussed in
section “Activities to be Measured”). The proximal humerus CS is intended to
describe the glenohumeral joint or the humerus relative to the thorax, and the distal
one to describe the elbow (Fig. 3). Similarly, a proximal forearm for the elbow and a
distal forearm for the wrist. Also, it is important to notice that the CSs for the distal
forearm and hand, as well as girdle and thorax, implement a constraint on the DoFs,
which is consistent with the model of Fig. 2.
Two coordinate systems are reported for the proximal humerus, H1 and H2. H1 is
most commonly applied, but H2 has advantages for tracking the humerus
internal–external rotation because it is insensitive to the soft-tissue artifact when
the elbow is kept at a constant position, as described in section “Marker/Sensor
Set-Up.”
The typical Euler decompositions used to calculate joint angles from the relative
orientation of CSs is reported in Table 3.
Marker/Sensor Set-Up
The application of noninvasive methods to identify (van Sint Jan 2007) and track
with sufficient accuracy the position of the ALs and of the anatomical/functional CSs
during a motion is critical and often a challenging task.
Before discussing recommendations about markers/sensors placement, it is
important to define the purpose of the Technical Frame (TF). Kontaxis et al.
(2009) report the following definition; a TF is a CS associated with a body segment.
Fig. 3 Description of the

anatomical and functional
frames for the humerus. In
particular, a proximal
anatomical frame can be
defined (H1 or H2) to describe
the glenohumeral kinematics.
H3 can be used to optimally
describe the elbow kinematics
(Reprint from Kontaxis et al.
2009)
Table 3 Sequence of Euler angles for each segment/joint kinematics of interest

Segment/Joint Euler Sequence (positive sign)
Thorax relative to global CS XZ0 Y00
(flexion-abduction-internal rotation)
Scapula relative to thorax YZ0 X00
(protraction-lateral rotation-posterior tilt)
Shoulder-girdle relative to thorax YZ0 X00
(protraction-lateral rotation-dummy)
Humerus relative to thorax Mostly sagittal plane movements: XZ0 Y00
(flexion-abduction-internal rotation)
Mostly frontal plane movements: ZX0 Y00
(abduction-flexion-internal rotation)
Forearm relative to humerus XZ0 Y00
(flexion-carrying angle-pronation)
Hand relative to forearm Y0 Z0 X0
(dummy – radial deviation-flexion)
It has normally no repeatable reference to the morphology of the segment and as

such has an arbitrary position and orientation with respect to the bone. The place-
ment of markers or sensors that define a TF is usually such as to help with some
technical difficulties that can occur during tracking a motion (e.g., minimize soft
tissue artifact, enhance visibility and comfort, adapt to the muscular constitution of
the subject, etc.). For optoelectronic systems, a TF is usually constructed by using
the instantaneous position of at least three nonaligned superficial markers positioned
on the segment (altogether referred to as “cluster”), based on an arbitrary geometric
rule. Clusters with a redundant number of reflective markers (4 or more) are very
common when visibility or marker occlusion is a concern. For other measurement
systems, e.g., electromagnetic or inertial and magnetic, a TF is the embedded local
CS in the sensor that is attached on the segment. Hereinafter, we will adopt the term
“cluster” to refer to both clusters of markers and to a sensor with an embedded CS.
There are not universally accepted recommendations for cluster placement, size,
and shape, since it can vary significantly based on the segment to track and the
requirements of the motion to be recorded. However, there are some common
practices that have been adopted in current studies. These will be firstly reviewed.
Then, a more technical description of the options for the different segments will be
provided.
The use of clusters prepositioned on a rigid plastic plate (e.g., 3D printed) or
low-temperature thermoplastic material (e.g., Thermolyn Pedilon, Ottobock, D) is
very popular, and it is equally applicable with electromagnetic/inertial sensors. Once
the cluster is prepared, it is positioned on the body segment. Markers can also be
glued on elastic bandages wrapped around the segment to form a cluster. The general
goal is to ensure an easy tracker placement on the subjects, without causing any
movement restrictions. One of the basic practices is to shape the cluster with an
asymmetric design and as big as possible (to maximize marker visibility and
identification). More specific indications for clusters are provided in Cappozzo
et al. (1997). The use of clusters is preferred compared to placement of markers
directly on ALs since it has been shown to reduce skin artifacts (Cappozzo et al.
2005). However, the process of recognizing and correctly palpating the landmarks is
critical and is recommended to practice between operators within the same labora-
tory to increase repeatability (McGinley et al. 2009).
With clusters in place, ALs are calibrated with a pointer which has an embedded
cluster, by applying what is known in gait analysis as the “CAST approach” but that
was described for upper-extremity by Johnson et al., and van der Helm et al. in the
early 1990 (Johnson and Anderson 1990; Van der Helm et al. 1992). The pointer is
built with a known geometry so that the tip of the pointer has known coordinates in
its embedded technical frame. The tip of the pointer is gently positioned on the AL,
while the clusters on the segment and on the pointer are both visible to the
measurement system. It is easy, at that stage, to calculate the coordinate of the AL
in the TF of the segment (Cappozzo et al. 2005). Once all coordinates of the ALs are
known in the segment TF, it is equally easy to build the AF and link it to the TF that
is tracked by the measurement system over time.
Sometimes, ALs are located within the body and are impossible to palpate (e.g.,
GH centre). Estimation of the position of those landmarks can be based either on
regression equations or functional movements that are actively or passively executed
by the subject that is under analysis. This is discussed in details in section “Activities
to be Measured.” The GH centre of rotation is commonly reported with respect to the
anatomical CS of the scapula (Veeger et al. 1997; Nikooyan et al. 2011; Lempereur
et al. 2010; Lempereur et al. 2013). Campebell et al. (2009) recommended estimat-
ing the GH as the midpoint between the GH that is tracked by the acromion/scapula
tracker and the proximal humerus. However, results were based on static measures
Fig. 4 Marker-set used at the INAIL Prosthesis Center, Italy. For the marker-set used at the
Hospital for Special Surgery see Fig. 7
and noise and tracking issues were not considered. We are currently tracking GH
with the acromion cluster.
Pictures describing the cluster set applied by the INAIL Prosthesis Centre is
shown in Fig. 4 and by the Hospital for Special Surgery in Fig. 7.
Considering the thorax, the cluster is usually positioned on the upper portion of
the sternum (close to IJ), centered with respect to the midline of the thorax. An
alternative popular tracking method for the thorax is to place four individual
markers; two of them located between the manubrium and the sternum and the
other two on the spine and between the landmarks of C7 and T8. Both solutions are
equally possible, but the former might be easier to apply on females and on
overweight subjects.
Considering the scapula, different tracking methods are possible. One of the first
methods that was applied to record scapula motion was the scapula locator (Johnson
et al. 1993; Barnett et al. 1999), which has been previously defined as “the silver
standard” (Cutti and Veeger 2009) for scapula tracking (Fig. 5). The scapula locator
is a rigid configurable structure that is designed to locate the three most palpable
landmarks of the scapula (acromion tip, root of scapula spine, inferior angle) and
collect a sequence of static positions. A regression equation that describes scapula
motion is built based on those discrete positions.
More recently, noninvasive dynamic scapula tracking methods are preferred,
namely, the scapula tracker, the acromion tracker, and the scapula spine tracker
(Fig. 6).
The scapula tracker is designed to follow the scapula motion by tracking the
scapula spine and the acromion, by means of a base that has a hinge joint that
conforms to the scapula spine and a footpad that rests on the acromion (hinge joint
and footpad are connected and are adjustable in length and height in order to fit
different scapula sizes, as recommended by Karduna et al. 2001).
Fig. 5 The scapula locator is

adjustable and specially
designed to locate the most
palpable landmarks of scapula
(Johnson et al. 1993; Barnett
et al. 1999). Data can be
recorded in static humeral
positions and after palpating
and locating the position of
the scapula ALs. Here in
combination with the MTx
Xsens sensors (IMMS) when
applying the ISEO protocol
(Cutti et al. 2008)
The acromion tracker consists of a cluster placed on the skin over the acromial
plate of the scapula. This is one of the most popular tracking methods because the
design is simpler compared to the scapula tracker and it is easily applicable. It has
shown to record reliable scapula kinematics data for humerus elevation up to
100 degrees (Van Andel et al. 2009; Prinold et al. 2011; Lempereur et al. 2014). In
order to minimize soft tissue artifacts, the cluster should be positioned at the meeting
point of the acromion and the scapular spine (Shaheen et al. 2011; Lempereur et al.
2014), and multiple calibration of ALs should be adopted for the full range of motion
(Brochard et al. 2011). The actual shape of the acromion cluster varies from
laboratory to laboratory. A possible design, which can be easily replicated through
a 3D printer, was designed at Centro Protesi INAIL by one of the authors (AGC) and
is available for download through the ISG website.
Finally, the spine tracker collects scapula kinematics by means of a cluster glued
on a box 47 mm 30 mm 13 mm in size (the cluster can be replaced by an IMU
embedded into the box itself). The box is positioned over the central third of the
scapula spine, between AA and TS, aligned with the upper edge of the spine (Cutti
et al. 2008; Parel et al. 2014).
Humerus H1 and H3 CSs are typically tracked with a cluster on the distal part of
the humerus itself. This position has shown to have the optimal tracking of the
humeral long-axis by Hamming et al. (2012). However, the reader should be aware
of the soft-tissue artifact affecting the humerus axial rotation. Cutti et al. (2005)
reported that humeral clusters can significantly underestimate the internal/external
Fig. 6 Images depicting the

scapula tracker (a) (model in
use at the Hospital for Special
Surgery, USA), the acromion
tracker (b) (model in use at the
INAIL Prosthetic Center,
Italy), and the spine tracker (c)
(applied with the ISEO
protocol and MTx Xsens
sensors)
humerothoracic rotation as much as 48% of the effective humeral axial rotation

performed. Accurate estimation of humeral internal/external rotation can be of high
importance depending on the clinical question, especially in the area of sports
biomechanics. As an example, in overhead pitching, a sport where shoulder injuries
occur in 57% of the athletes during a season, it is important to have accurate
measurements of humeral axial rotation in order to fully understand the mechanics
of throwing. To tackle this problem, some authors proposed alternative methods to
Fig. 7 The size of the ulnar cluster is small and has an oval shape (b), it is positioned in the
proximal ulnar and distal from the olecranon (a)
compensate for the artifact, based on the humerus CS H2 (Table 2) possibly associ-
ated with specific algorithms (Cutti et al. 2006c) or through the adoption of specific
marker-sets, e.g., using two markers on the ulna Rettig et al. (2009). A further
alternative has been developed at the Hospital for Special Surgery (Kontaxis et al.
2014), who used an additional marker cluster on the ulna (together with the humeral
cluster) in order to increase accuracy on the estimation of the internal/external
rotation during pitching. The design of this cluster, referred herein as “ulnar cluster,”
has an oval shape and is small in dimensions (<3 cm length and <2 cm width) but
can accommodate four markers that are placed in asymmetric positions. The cluster
is attached to the proximal ulnar, with its long axis following the long ulnar axis
(Fig. 7). It is found that in this position, the cluster is minimally affected by skin
artifact.
The purpose of the ulnar cluster is to track the elbow flexion/extension axis VFlex
which can be calculated with a functional motion, by calculating the instantaneous
helical axis between the ulnar and humeral clusters during an elbow flexion/exten-
sion motion (section “Definition of Segment or Joint CSs and Angles”). The distal
humeral frame H3 (Table 2) can then be defined by the VFlex and the GH that is
tracked by the scapula/acromion tracker. In order to define the proximal (anatomical)
humeral frame H1 (Table 2), the medial and lateral epicondyles (EM and EL) can be
referred to H3 by a simple pointer palpation/calibration (e.g., through the CAST
approach). By knowing the coordinates of the EL, EM, and GH (in the H3 frame), the
anatomical humeral frame H1 can be defined and the transformation between H3 and
H1 can be calculated. This relationship is not changing during the motion and as
such, tracking of the ulnar cluster and scapula/acromion tracker can result in accurate
tracking of the humerus. Comparing results of concurrent tracking of the humerus
with the two methods (humeral vs. ulnar cluster) showed that the former can
underestimate humeral axial rotation by 49 degree during overhead pitching. How-
ever, it needs to be noted that there are some limitations when the ulnar cluster
method is used: (i) the position and size of the cluster creates substantial problems on
marker occlusion and visibility which may require a complex and multi-camera
set-up, (ii) the ulnar-cluster is not fully free from skin artifacts with forearm
pronation/supination (at the end range of motion) may affect the cluster.
Regarding the forearm, as reported by Anglin et al. (2000), a distal cluster tracks
most of the movement and this is a position commonly applied in clinics (Stokdijk
et al. 1999; Cutti et al. 2005; van Andel et al. 2008).
Activities to Be Measured
The protocol should include a set of activities for calibration and activities aimed to
investigate the function. The number of repetitions should be also fixed. Specifically,
the activities for calibration should include:
1. Static measures to calibrate the ALs. All ALs should be calibrated with the
subject in the same position, otherwise the soft-tissue artifact can change the
position of the cluster relative to the bone and lead to wrong estimations of the
ALs relative to each other. As reference position, the subject might be asked to
stand upright, with the humerus aligned with the vertical axis of the trunk, elbow
flexed at 90 with the thumb pointing up);
2. Functional movement for the estimation of the centre of the humeral head (GH).
The subject is typically asked to perform small circumduction movements with
the arm, e.g., three consecutive with spiral movements from 0 to 40 of arm
elevation. Alternatively, three cycles of three different movements
(flexion–extension, abduction–adduction, and circumduction) (Lempereur et al.
2013). The algorithm of Gamage and Lasenby (2002) is one of the most popular
method adopted to estimate GH;
3. Functional movements for the estimation of the elbow flexion–extension and
pronation–supination axes. The subject is asked to perform “pure movements” of
elbow flexion–extension and pronation–supination separately. The algorithm
proposed by Woltring (1990) or by Halvorsen et al. (1999) or by Gamage and
Lasenby (2002) can be used to calculate both functional axes.
Once the calibration procedure is completed, the investigator can ask the subject
to perform a wide range of activities, depending on the topic of the survey/analysis,
e.g., assessment of the scapulohumeral coordination, activities of daily living, sport
gestures, or work gestures. To ensure comparability among studies about the
scapulohumeral coordination, it is strongly recommended to always collect the
kinematics of the subjects during “pure humerus elevation” in the sagittal and in
the scapular and/or frontal plane. Usually, a minimum of three “good” repetitions for
each movement should be completed, in order to compute and analyze the mean
movement among the repetitions.
Extraction of Summarizing Parameters
Quantitative motion analysis can produce a considerable amount of data. The

definition of the hypotheses will guide in the definition of the specific parameters
to calculate and report.
As an indication, the description of the scapulohumeral rhythm (also named

“scapulahumeral coordination) can be effective when dealing with disorders involv-
ing a scapula dyskinesis (Kibler et al. 2013). This is typically analyzed through
angle–angle plots, where the humerus elevation angle (flexion–extension, abductio-
n–adduction, or elevation) is reported on the x-axis, and one of the scapula rotation
angle (protraction–retraction, mediolateral rotation, and tilting) is reported on the
y-axis (Cutti et al. 2014; Parel et al. 2016).
Another field of investigation is the assessment of Activities of Daily Living
(ADL). Recommendations are difficult in this case, but the reporting provided by
Magermans et al. (2005) might be used as example.
A different approach to summarize motion analysis data is to integrate them with
a clinical outcome measure, as reported in Cutti et al. (2016). In this paper, the
authors proposed the “Scapula Weighted Constant-Murley score,” i.e., a new for-
mulation of the popular “Constant-Murley Score” (Constant et al. 2008): alterations
of the scapulothoracic joint are detected and used to weight the points assigned to the
patient for the humerothoracic elevation. Alterations of the SHR in a patient are
detected comparing the sound-to-affected side differences for the SHR, with the
typical difference between sides of asymptomatic control subjects.
Another summary score of increasing popularity is the “Arm Profile Score”
(Jaspers et al. 2011a). This score is based on kinematic data and is calculated from
the root mean square difference between kinematics of an individual child with
upper-limb movement deficits and average data from typically developing children.
The APS summarizes the overall severity of upper-limb movement pathology and
can be decomposed in 13 Arm Variable Scores, representing the different joint
angles. It was tested on a group of children with cerebral palsy.
Upper-Extremity Models for Inertial and Magnetic Sensor
Systems based on inertial and magnetic sensors (IMMS) are becoming increasingly
popular. The INAIL Shoulder and Elbow Outpatient protocol (ISEO) is an example
of system for recording and representation of upper-extremity kinematic data, based
on this new technology. Since it is, at present, the most validated system, with an
available set of prediction bands from asymptomatic subjects, and integration in a
clinical outcome measure (Cutti et al. 2008, 2014, 2016; Parel et al. 2014, 2016),
ISEO will be used to introduce the reader to this emerging topic.
Instrumentation
ISEO uses four MTw sensor units (Xsens Technologies, NL). Each MTw is a small
(47 mm 30 mm 13 mm) and lightweight (16 g) box, containing a 3D-gyroscope,
accelerometer, and magnetometer, which together provide the orientation of the
technical CS of the MTw relative to a global, earth-based CS. MTw sensors are
wirelessly connected to the Awinda Station which ensures highly accurate time
synchronized data sampling in all connected MTw’s.
Body Segments and Joint Kinematic Description

ISEO assumes thorax, scapula, humerus, and forearm as rigid segments forming the
upper limb (Cutti et al. 2008). The orientations of the scapula and humerus are
computed with respect to the thorax, while the orientation of the forearm is com-
puted with respect to the humerus. Scapulothoracic kinematics is described by three
independent angles: protraction–retraction (PR-RE), mediolateral rotation (ME-LA),
and posterior–anterior (P-A) tilting. Humerothoracic kinematics is described by
three independent angles: flexion–extension (shFL-EX), abduction–adduction
(AB-AD), and internal–external rotation (IN-EX) for sagittal plane movements
(or AB-AD, shFL-EX, and IN-EX for frontal plane movements). The elbow is
modeled with two hinge joints with nonintersecting axes: flexion–extension (elFL-
EX) and pronation–supination (PR-SU).
Sensors Position
The four sensor units are positioned over the thorax, the scapula, the humerus, and
the forearm (Fig. 5). For the thorax, the sensor is positioned over the flat portion of
the sternum, with the Z axis of the sensor pointing away from the body. For the
scapula, the X axis of the sensor is aligned with the cranial edge of the scapular
spine, over the central third of the scapula (Fig. 6). For the humerus, the sensor is
positioned and oriented to minimize the soft tissue artifacts: usually over the central
third of the humerus, slightly posterior. For the forearm, the base of the sensor is
positioned over the distal, flat surface of radius and ulna, with the local Z axis
pointing away from the wrist.
Definition of Anatomical/Functional Frames

The anatomical CSs of thorax (TH) and scapula (SC) have a constant orientation
with respect to the SoR of the sensors on the corresponding segment, as reported in
Table 4.
For the humerus, two sets of CSs are defined (Table 4), similarly to Table 2: one
set is related to the proximal humerus (H1 and H2, from Wu et al. (2005)) and
describes humerothoracic kinematics; the other is related to the distal humerus and
defined by applying the Denavit–Hartenberg method (HD, from Cutti et al. (2008)) at
the elbow FL-EX hinge joint. The anatomical CSs of H1 have a constant orientation
with respect to the CS of the sensors on the corresponding segment, while for H2, the
orientation with respect to the CS of the humerus sensor is not constant but is
updated sample-by-sample during every task, based on the orientation of the long
axis of the forearm. The anatomical CSs of HD is computed and expressed in the
forearm sensor’s CS using the equations reported in Table 4.
The forearm anatomical CS (F) is computed and expressed in the segment
sensor’s CS using the equations reported in Table 4.
To define the anatomical CSs of the thorax, scapula, and proximal humerus, and
to express the orientation of these anatomical CSs in the sensor CS of the
corresponding segment, the orientation of the sensors’ CSs is measured during a
static trial.
Table 4 Definition of the anatomical/fictional frames as specified in ISEO, which exploits inertial
and magnetic sensors. SU sensor
Segment Axes
Thorax (TH) YTH = SU THZG / || SU THZG || : cranial
XTH = YTH ^ [0 0 1] / || YTH ^ [0 0 1] || : lateral
ZTH = XTH ^ YTH / || XTH ^ YTH || : posterior
Scapula (SC) XSC = [1 0 0] : lateral
ZSC = XSC ^ ZG / || XSC ^ ZG || : posterior
YSC = ZSC ^ XSC / || ZSC ^ XSC || : cranial
SU H
Proximal humerus (H1) RH1 = SU HRTH
Proximal humerus (H2) YH2 = YH1 : cranial
XH2 = YH2 ^ SU HYF / || YH2 ^ SU HYF || : lateral
ZH2 = XH2 ^ YH2 / || XH2 ^ YH2|| : posterior
Distal humerus (HD) XHD = VFLEX / || VFLEX || : lateral
ZHD = XHD ^ YH1 / || XHD ^ YH1|| : posterior
YHD = ZHD ^ XHD / || ZHD ^ XHD || : cranial
Forearm (F) YF = VPS / || VPS || : cranial
XF = YF ^ [0 0 1] / || YF ^ [0 0 1] || : lateral
ZF = XF ^ YF / || XF ^ YF || : posterior
To define the anatomical CSs for distal humerus and forearm, the orientation of
sensors over the humerus and forearm is measured during two tasks: firstly, a “pure”
movement of elbow flexion–extension, keeping a constant PR-SU, and the humerus
alongside the body; secondly, a “pure” movement of elbow PR-SU, keeping the
elbow flexed 90 and the humerus alongside the body. In both cases, the functional
method described in Woltring (1990) and Stokdijk et al. (1990) is applied, and elbow
FL-EX axis is expressed in the CS of the sensor over the humerus, while the PR-SU
axis is expressed in the CS of the sensor over the forearm.
Computation of Joint Angles

During data acquisition, the orientation of the anatomical CSs is updated sample-by-
sample based on the orientation of sensors’ CS. Joint angles are obtained, sample-
by-sample, by decomposing the relative orientation of the anatomical CS using the
following Euler sequences: for the scapulothoracic movement, YZ0 X00 (PR-RE,
ME-LA, and P-A); for the humerothoracic movement almost on the sagittal plane,
XZ0 Y00 (shFL-EX, AB-AD, and IN-EX); for the humerothoracic movement almost
on the frontal plane ZX0 Y00 (AB-AD, shFL-EX, and IN-EX); for the elbow, XZ0 Y00
(elFL-EX, carrying angle and PR-SU).
Future Direction
With this chapter we tried to provide a guideline to readers interested in applying

quantitative motion analysis techniques to the upper-extremity, with special attention
to the shoulder and elbow.
The description of a single protocol is not possible at present. However, it is

possible to define a common language to describe protocols, and also to indicate
recommendations, mostly based on documents from the International Society of
Biomechanics and the International Shoulder Group. Among the technical details
provided, we think that a few are particularly important:
1. Protocols cannot be reduced to the application of markers. Markers come into

play almost at the end of the process, well after the definition of the clinical
problem, the formulation of hypothesis, the schematization of joints and seg-
ments, the definition of anatomical/function CSs, and angle calculation.
2. The description of joint kinematics must be based on functional axis. It can be
very misleading to calculate joint angles using anatomical axes. This is particu-
larly important for the elbow, and the use of two coordinate systems associated to
a single segment to describe its two joints should be considered in practice.
3. The soft tissue artifact is troublesome. However, the proposal formulated at the
Hospital for Special Surgery of using an ulnar cluster is very promising and could
solve the problem to a large extent. Future works must be directed to have a full
validation of this approach;
4. To ensure comparability of studies, it is proposed to always report the
scapulohumeral rhythm in terms of angle–angle plots.
5. The application of inertial sensors is very promising because angle calculations
take place in real time and measurements can be completed in rehabilitation
rooms or outdoor, and not just in dedicated laboratories. ISEO has been largely
validated and its application to clinical problems is now underway. In particular,
we aim at proving the effectiveness of biofeedback on scapula control during the
rehabilitation of patients surgically treated for rotator-cuff tear.
Cross-References

▶ 3D Dynamic Probablistic Pose Estimation From Data Collected Using Cameras
▶ Three-dimensional Human Kinematics Estimation Using Magneto-Inertial Mea-
surement Units
References
Anglin C, Wyss UP (2000) Review of arm motion analyses. Proc Inst Mech Eng H 214(5):541–555.
Review
Barnett ND, Duncan RD, Johnson GR (1999) The measurement of three dimensional
scapulohumeral kinematics–a study of reliability. Clin Biomech (Bristol, Avon) 14
(4):287–290. PubMed PMID: 10619117
Brochard S, Lempereur M, Rémy-Néris O (2011) Double calibration: an accurate, reliable and easy-
to-use method for 3D scapular motion analysis. J Biomech 44(4):751–754
Campbell AC, Lloyd DG, Alderson JA, Elliott BC (2009) MRI development and validation of two
new predictive methods of glenohumeral joint centre location identification and comparison
with established techniques. J Biomech 42(10):1527–1532
Cappozzo A, Cappello A, Della Croce U, Pensalfini F (1997) Surface-marker cluster design criteria
for 3-D bone movement reconstruction. IEEE Trans Biomed Eng 44(12):1165–1174. PubMed
PMID: 9401217
stereophotogrammetry. Part 1: theoretical background. Gait Posture 21(2):186–196
Constant CR, Gerber C, Emery RJ, Sojbjerg JO, Gohlke F, Boileau P (2008) A review of the
Constant score: modifications and guidelines for its use. J Shoulder Elbow Surg 17:355–361
Cutti AG, Veeger HE (2009) Shoulder biomechanics: today’s consensus and tomorrow’s perspec-
tives. Med Biol Eng Comput 47(5):463–466. doi:10.1007/s11517-009-0487-3. PubMed PMID:
19396486
Cutti AG, Paolini G, Troncossi M, Cappello A, Davalli A (2005) Soft tissue artefact assessment in
humeral axial rotation. Gait Posture 21(3):341–349
Cutti AG, Raggi M, Davalli A, Cappello A (2006a) Definition of two reference elbow models from
cadaver data. Gait Posture 24S:S224–S225
Cutti AG, Garofalo P, Davalli A, Cappello A (2006b) How accurate is the estimation of elbow
kinematics using ISB recommended joint coordinate systems? Gait Posture 24S:S36–S37
Cutti AG, Cappello A, Davalli A (2006c) In vivo validation of a new technique that compensates
for soft tissue artefact in the upper-arm: preliminary results. Clin Biomech (Bristol, Avon) 21
(Suppl 1):S13–S19
Cutti AG, Giovanardi A, Rocchi L, Davalli A, Sacchetti R (2008) Ambulatory measurement
of shoulder and elbow kinematics through inertial and magnetic sensors. Med Biol Eng Comput
46(2):169–178
Cutti AG, Parel I, Raggi M, Petracci E, Pellegrini A, Accardo AP, Sacchetti R, Porcellini G (2014)
Prediction bands and intervals for the scapulo-humeral coordination based on the Bootstrap and
two Gaussian methods. J Biomech. 47(5):1035–1044
Cutti AG, Parel I, Pellegrini A, Paladini P, Sacchetti R, Porcellini G, Merolla G (2016) The Constant
score and the assessment of scapula dyskinesis: proposal and assessment of an integrated
outcome measure. J Electromyogr Kinesiol 29:81–89
Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, Guendelman E, Thelen DG (2007)
OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE
Trans Biomed Eng 54(11):1940–1950
Gamage SSHU, Lasenby J (2002) New least squares solutions for estimating the average centre of
Garofalo P, Cutti AG, Filippi MV, Cavazza S, Ferrari A, Cappello A et al (2009) Inter-operator
reliability and prediction bands of a novel protocol to measure the coordinated movements of
shoulder-girdle and humerus in clinical settings. Med Biol Eng Comput 47(5):475–486
Goto A, Moritomo H, Murase T, Oka K, Sugamoto K, Arimura T, Nakajima Y, Yamazaki T, Sato Y,
Tamura S, Yoshikawa H, Ochi T (2004) In vivo elbow biomechanical analysis during flexion:
three-dimensional motion analysis using magnetic resonance imaging. J Shoulder Elbow Surg
13(4):441–7. PubMed PMID: 15220886
Halvorsen K, Lesser M, Lundberg A (1999) A new method for estimating the axis of rotation and
the center of rotation. J Biomech 32:1221–1227
Hamming D, Braman JP, Phadke V, LaPrade RF, Ludewig PM (2012) The accuracy of measuring
glenohumeral motion with a surface humeral cuff. J Biomech 45(7):1161–1168
Jaspers E, Feys H, Bruyninckx H, Klingels K, Molenaers G, Desloovere K (2011a) The Arm Profile
Score: a new summary index to assess upper limb movement pathology. Gait Posture
34(2):227–233
Johnson GR, Anderson JM (1990) Measurement of three-dimensional shoulder movement by an
electromagnetic sensor. Clin Biomech 5(3):131–136
Johnson GR, Stuart PR, Mitchell S (1993) A method for the measurement of three-dimensional
scapular movement. Clin Biomech (Bristol, Avon) 8(5):269–273. doi: 10.1016/0268-0033(93)
90037-I. PubMed PMID: 23915988
Karduna AR, McClure PW, Michener LA (2000) Scapular kinematics: effects of altering the Euler
angle sequence of rotations. J Biomech 33(9):1063–1068
Karduna AR, McClure PW, Michener LA, Sennett B (2001) Dynamic measurement of three-
dimensional scapular kinematics: a validation study. J Biomech Eng 123(2):184–190
Kibler WB, Ludewig PM, McClure PW, Michener LA, Bak K, Sciascia AD (2013) Clinical
implications of scapular dyskinesis in shoulder injury: the 2013 consensus statement from the
‘Scapular Summit’. Br J Sports Med 47(14):877–885
Kontaxis A, Cutti AG, Johnson GR, Veeger HEJ (2009) A framework for the definition of
standardized protocols for measuring upper-extremity kinematics. Clin Biomech 24:246–253
Kontaxis A, Kraszewski A, Gibbsons M, Graziano J, Fealy S, Hillstrom H (2014). Effect of fatigue
on high school basebal pitchers, Proceedings of the International Shoulder Group, Waterloo
estimation of the glenohumeral joint centre by functional methods: accuracy and repeatability
assessment. J Biomech 43(2):370–374
Lempereur M, Brochard S, Rémy-Néris O (2013) Repeatability assessment of functional methods
to estimate the glenohumeral joint centre. Comput Methods Biomech Biomed Engin 16(1):6–11
Lempereur M, Brochard S, Leboeuf F, Rémy-Néris O (2014) Validity and reliability of 3D marker
based scapular motion analysis: a systematic review. J Biomech 47(10):2219–2230
Ludewig PM, Behrens SA, Meyer SM, Spoden SM, Wilson LA (2004) Three dimensional
clavicular motion during arm elevation: reliability and descriptive data. J Orthop Sports Phys
Ther 34(3):140–149
Magermans DJ, Chadwick EK, Veeger HE, van der Helm FC (2005) Requirements for upper
extremity motions during activities of daily living. Clin Biomech 20(6):591–599
Marchese SS, Johnson GR (2000) Measuring the kinematics of the clavicle. 6th international
symposium of the 3D analysis of human motion, Cape Town, pp 37–40
gait measurements: a systematic review. Gait Posture 29(3):360–369
Nikooyan AA, van der Helm FCT, Westerhoff P, Graichen F, Bergmann G, Veeger HEJ (2011)
Comparison of two methods for in vivo estimation of the Glenohumeral Joint Rotation Center
(GH-JRC) of the patients with shoulder hemiarthroplasty. PLoS One 6(3):e18488
Parel I, Cutti AG, Fiumana G, Porcellini G, Verni G, Accardo AP (2012) Ambulator measurement
of the scapulohumeral rhythm: intra- and inter-operator agreement of a protocol based on inertial
and magnetic sensors. Gait Posture 35(4):636–640
Parel I, Cutti AG, Kraszewski A, Verni G, Hillstrom H, Kontaxis A (2014) Intra-protocol repeat-
ability and inter-protocol agreement for the analysis of scapulo-humeral coordination. Med Biol
Eng Comput 52(3):271–282
Parel I, Jaspers E, DE Baets L, Amoresano A, Cutti AG (2016) Motion analysis of the shoulder in
adults: kinematics and electromyography for the clinical practice. Eur J Phys Rehabil Med 52
(4):575–582. PubMed PMID: 27434612
Piazza SJ, Cavanagh PR (2000) Measurement of the screw-home motion of the knee is sensitive to
errors in axis alignment. J Biomech 33(8):1029–1034
Prinold JAI, Shaheen AF, Bull AMJ (2011) Skin-fixed scapula trackers: a comparison of two
dynamic. J Biomech 44:2004–2007
Pronk GM, van der Helm FC (1991) The palpator: an instrument for measuring the positions of
bones in three dimensions. J Med Eng Technol 15(1):15–20
Rettig O, Fradet L, Kasten P, Raiss P, Wolf SI (2009) A new kinematic model of the upper extremity
based on functional joint parameter determination for shoulder and elbow. Gait Posture 30
(4):469–476
Seth A, Matias R, Veloso AP, Delp SL (2016) A biomechanical model of the scapulothoracic joint
to accurately capture scapular kinematics during shoulder movements. PLoS One 11(1):
e0141028
Shaheen AF, Alexander CM, Bull AM (2011) Effects of attachment position and shoulder orien-
tation during calibration on the accuracy of the acromial tracker. J Biomech 44(7):1410–1413
Stokdijk M, Meskers CG, Veeger HE, de Boer YA, Rozing PM (1999) Determination of the optimal
elbow axis for evaluation of placement of prostheses. Clin Biomech 14(3):177–184
Stokdijk M, Biegstraaten M, Ormel W, de Boer YA, Veeger HE, Rozing PM (2000) Determining the
optimal flexion-extension axis of the elbow in vivo - a study of interobserver and intraobserver
reliability. J Biomech 33(9):1139–1145. PubMed PMID: 10854888
Teece RM, Lunden JB, Lloyd AS, Kaiser AP, Cieminski CJ, Ludewig PM (2008) Three-
dimensional acromioclavicular joint motions during elevation of the arm. J Orthop Sports
Phys Ther 38(4):181–190
van Andel CJ, Wolterbeek N, Doorenbosch CA, Veeger DH, Harlaar J (2008) Complete 3D
kinematics of upper extremity functional tasks. Gait Posture 27(1):120–127
van Andel C, van Hutten K, Eversdijk M, Veeger D, Harlaar J (2009) Recording scapular motion
using an acromion marker cluster. Gait Posture 29(1):123–128
Van der Helm FC, Pronk GM (1995) Three-dimensional recording and description of motions of the
shoulder mechanism. J Biomech Eng 117(1):27–40
Van der Helm FC, Veeger HE, Pronk GM, Van der Woude LH, Rozendal RH (1992) Geometry
parameters for musculoskeletal modelling of the shoulder system. J Biomech 25(2):129–144
van Sint Jan S (2007) Color atlas of skeletal landmark definitions: guidelines for reproducible
manual and virtual palpation. Churchill Livingstone, Belgium
Veeger HE, Yu B, An KN, Rozendal RH (1997) Parameters for modeling the upper extremity. J
Biomech 30(6):647–652
Woltring HJ (1990) Data processing and error analysis. In: Cappozzo A, Berme N (eds) Biome-
chanics of human movement. Bertec Corporation, Worthington, pp 203–237
Wu G, van der Helm FC, Veeger HE, Makhsous M, Van Roy P, Anglin C, Nagels J, Karduna AR,
McQuade K, Wang X, Werner FW, Buchholz B (2005) ISB recommendation on definitions of
joint CSs of various joints for the reporting of human joint motion – part II: shoulder, elbow,
wrist and hand. J Biomech 38(5):981–992
Interpreting Ground Reaction Forces in Gait
Nachiappan Chockalingam, Aoife Healy, and Robert Needham
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Technology for Ground Reaction Force Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Ground Reaction Force and Its Measurements During Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Analysis of Ground Reaction Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Time Domain Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Frequency Domain Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Normalization Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Ground Reaction Force in the Understanding of Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Ground Reaction Force in the Assessment of Clinical Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Scoliosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Stroke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Multiple Sclerosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Abstract
This chapter provides a description of ground reaction forces (GRFs) in gait,
detailing the technology used to measure them and their use in gait analysis.
Representative ground reaction force data is provided, and information on how
data is analyzed and interpreted is discussed. In addition, examples of how GRF
data has been used to gain an understanding of healthy and pathological gait
(scoliosis, cerebral palsy, stroke, multiple sclerosis) are provided and discussed.
N. Chockalingam (*) • A. Healy • R. Needham

Faculty of Health Sciences, Staffordshire University, Stoke On Trent, UK
e-mail: N.Chockalingam@staffs.ac.uk; a.healy@staffs.ac.uk; r.needham@staffs.ac.uk

DOI 10.1007/978-3-319-30808-1_31-1
2 N. Chockalingam et al.
This chapter highlights that the GRF is an essential part of any clinical gait
analysis and contributes to both surgical and conservative clinical management
involving gait.
Keywords
Ground reaction force • Center of pressure • Shear force • Gait analysis • Human
kinetics • Walking
Introduction
The examination of forces as the cause of movement is termed kinetics. Force

describes the interaction of an object and its surroundings. Major forces involved
in human locomotion are the ground reaction force (GRF), joint reaction force (JRF),
and muscle force (MF). Force provided by a supporting horizontal surface is known
as the ground reaction force (GRF). This in turn can be resolved into three compo-
nents, namely, vertical, mediolateral, and anterior-posterior. The center of pressure
(COP) represents the point of application of the resultant GRF. It is simply the central
point of pressure exerted on the foot.
Mechanical interactions between a muscle and its surroundings result in human
movement. A force exerted by a muscle is transmitted through its tendon to bone and
leads to a movement of the body segments. Muscle can only generate a tensile force
and not a compressive force; simply put – muscles cannot push but can only pull.
Ligament and bony contact forces exerted across a joint are transmitted from one
body segment to another. These forces are essential in some aspects of musculo-
skeletal biomechanics and are normally called a joint reaction force (JRF). JRF, for
example, is essential in the design of prosthesis and prosthetic joints. Detailed
explanation of JRF is provided elsewhere (Winter 1990), and it is beyond the
scope of this chapter. Since the JRF does not represent the absolute bone-to-bone
force, it may have significant magnitude (Whiting and Zernicke 1998). Another
important parameter within human kinetics which involves both force and displace-
ment is the moment. This can be defined as the product of the magnitude of the force
and the perpendicular distance to its line of action.
Recording and understanding of kinematic and kinetic parameters is required in
order to analyze the function of the musculoskeletal system operating to generate
and transmit forces and counteract external forces to perform smooth movement.
Gait analysis plays an important role in orthopedics, physical medicine, and reha-
bilitation and has been applied to the assessment of clinical conditions such as
cerebral palsy (Gage and Koop 1995), Parkinson’s disease (Koozekanani et al.
1987; Defebvre et al. 1996), scoliosis (Giakas et al. 1996; Schizas et al. 1998),
amputation (Hill et al. 1997), Paget’s disease (Gainey et al. 1989), Alzheimer’s
disease (Alexander et al. 1995), and multiple sclerosis (Gehlsen et al. 1986) and in
the development of orthoses and prostheses (Andriacchi and Hurwitz 1997;
Sanderson and Martin 1996). While a description of gait, its phases, and its detailed
analysis can be found elsewhere, this chapter focuses on GRF.
Interpreting Ground Reaction Forces in Gait 3
State of the Art
Technology for Ground Reaction Force Measurement
There are a number of types of force platforms which are used to record ground
reaction forces. The majority either consist of strain gauge transducers or piezoelec-
tric crystal transducers. The application of stress (force) to a structure changes its
geometry, in turn changing its resistance. This change of resistance is calibrated to
indicate the applied force in strain gauge devices. On the other hand, a non-
conducting crystal which generates an electrical charge when subjected to mechan-
ical strain is used in piezoelectric devices. Both these types of platform measure
forces indirectly, based on a calibration matrix. Technical characteristics and advan-
tages are described elsewhere (Vaughan et al. 1992), and these are beyond the scope
of this chapter. In addition to force platforms, there are several pressure measurement
systems. These have been descried in detail elsewhere (Razak et al. 2012) and are
available both as in-shoe systems and platforms. These can give an indication of the
vertical component of the GRF. Given the technical nature and the setup of the
pressure measurement systems, current commercially available systems cannot
measure shear (or horizontal) components. (For further information on pressure
distribution patterns, please refer to the section titled: “▶ Medical Applications:
Pedobarography” within this book.)
Ground Reaction Force and Its Measurements During Gait
The GRF acting on the foot during upright movements is a major external force,
normally recorded and measured in three dimensions using a force platform. It
comprises of a single vertical and a pair of horizontal shear force components acting
in the anterior-posterior and mediolateral directions. Force platform data also pro-
vide coordinates of the point of application of the resultant vector relative to the
platform origin.
Force platforms are widely used in the biomechanical assessment of human
movement, especially in gait analysis. There is a plethora of studies that used
GRFs and the COP to reflect the acceleration of the center of mass (COM)
(MacKinnon and Winter 1993), to assess balance (Winter et al. 1990), or combined
with anthropometric and kinematic data to determine the joint dynamics (Koopman
et al. 1995). Accurate estimation of these parameters is affected by the calibration.
COP is an important derivative of the GRF. Its misposition by 1 cm would result in
approximately 7% and 14% errors in the calculation of peak and average joint
moments of the lower segments during running (McCaw and Devita 1995). In gait
analysis, it was shown that high errors in the net joint dynamics could result mainly
from uncertainties due to marker position, anthropometric data, and COP (Davis
et al. 1997). These errors are amplified when joint dynamics of proximal segments
are calculated due to the cumulative input of errors from distal segments. While there
are studies which consider various options for calibration (Chockalingam et al.
2002), current commercially available systems have improved options for

calibration.
Vertical Component of the Ground Reaction Force

At initial contact (IC) and throughout the loading response (LR) phase, the body
decelerates downward, and the vertical GRF progressively increases as the weight of
the individual is transferred from the back foot to the front foot1. Peak loading is
reached during the initial stage of mid-stance (MS) due to weight acceptance and an
increase in muscular forces as an individual transits from double- to single-limb
support (Fig. 1a – F1). As the knee extends through MS, the COM displaces in an
upward direction. A deceleration of the COM is experienced near its highest position
and is the reason for the reduction in the vertical GRF below an individual’s actual
body weight (F2). From this MS trough (F2), the vertical GRF increases until a
second peak is reached in late TS (F3). The second peak relates to the foot pushing
against the floor as a result of an increase in the activity of the ankle plantar flexors
and from the deceleration of the COM as an individual’s body weight falls forward.
Anterior-Posterior Component of the Ground Reaction Force

At IC the foot applies a force in an anterior direction causing the body to decelerate.
This results in a posterior shear force that peaks during early stance (Fig. 1b – F4). As
the body moves over the stance limb, the posterior shear force reduces until the end
of MS (F5). From this crossover point, the COM moves ahead of the foot, and the
foot applies a force in a posterior direction causing an anterior shear force that
propels an individual forward. Peak anterior GRF occurs during late stance (F6). The
reduction that follows relates to the transference of force to the front foot.
Mediolateral Component of the Ground Reaction Force

At IC a shear force is applied by the foot to the ground that is directed medially. This
creates a lateral shear force that peaks during LR (Fig. 1c – F7). A lateral directed
force applied by the foot to the ground from mid LR onward is applied that results in
a medial shear force that is characterized by two peaks approximately at the
beginning of MS (F8) and toward the end of TS (F9).
Walking speed can influence the three GRF profiles over time. Regarding the
vertical GRF, for example, a reduced acceleration of the body’s COM while walking
at a slower speed results in a decrease in the height of the peaks and depth of the MS
trough. A reduced first peak could also be attributed to the presence of pain. For a
comparison on the effects of walking speed on the components of the GRF, readers
are directed elsewhere (Schwartz et al. 2008; Nilsson and Thorstensson 1989).
1
Gait events as defined by Perry and Burnfield (2010) are included in the following descriptive
account to support the interpretation of the three components of the GRF across the stance phase
(IC, initial contact; LR, loading response; MS, mid-stance; TS, terminal stance; PS, pre-swing;
swing phase).
Stride
Step
Stance Swing
DLS SLS DLS SLS
IC TO IC
a
1.6
LR MS TS PS SWING
1.4 F3
F1
Force / Body weight
1.2
1
0.8
F2
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100
b 0.3
F6
0.2
Force / Body weight
0.1 Anterior
F5
0
-0.1 Posterior
-0.2 F4
-0.3
0 10 20 30 40 50 60 70 80 90 100
c 0.1
0.08 F8 F9
Force / Body weight
0.06
0.04
0.02 Medial
0
-0.02 Lateral
F7
-0.04
0 10 20 30 40 50 60 70 80 90 100
% of Gait cycle
Fig. 1 Vertical (a), Anterior-Posterior (b) and Mediolateral (c) components of the Ground
Reaction Force
Center of Pressure (COP)

COP is the location of the GRF, and during normal gait, the COP progresses from the
heel to toe during the stance phase. Initially the COP is in a medial position at the
heel and rapidly moves to the lateral side of the mid-foot. It remains there during the
mid-support period and then rapidly transfers to the medial side of the fore foot. To
maintain balance while walking, the support force has to be on the same line from the
point of ground support (COP) to the COM.
Analysis of Ground Reaction Forces
Recorded GRFs can be analyzed in two ways. Conventional time domain analysis
examines specific force and time values, such as the magnitude of local peaks
(highest point in a known gait phase) as well as their time of occurrence (Fig. 1a –
F1 and F3). Examples of normal (Cavanagh and Lafortune 1980; Williams et al.
1987; Buczek and Cavanagh 1990) and pathological (Koozekanani et al. 1987;
Giakas et al. 1996; Sanderson and Martin 1996; Schizas et al. 1998) GRF data can
be found in the literature. In contrast, frequency domain analysis allows the study of
harmonic coefficients.
Several isolated force and temporal parameters have been established in order to
analyze the GRF in the time domain. A normal GRF pattern with various temporal
parameters (stride or step/min) is illustrated in Fig. 1a. This type of analysis is useful
when specific variables have different magnitudes in different pathological condi-
tions. For example, individuals with a neuromuscular dysfunction show evidence of
a significant difference in GRF pattern when compared with a healthy individual
(Fig. 2 – F10). As indicated in the study by Williams et al. (2011), one should be
aware that some of the patients with cerebral palsy have difficulty in supporting their
body weight in late stance and will be employing compensatory mechanisms to
prevent collapse of the affected limb. As shown in Fig. 2, the second peak might be
much smaller than the first.
In other instances, individuals may have instability problems, as in Parkinson’s
disease or other neurological disorders. Here the first priority is to reduce the tremor,
so the most appropriate approach is the examination of the entire force-time pattern
and not just isolated time domain parameters. This measurement is achieved by
frequency domain analysis. Using this technique, higher frequencies in the total
signal can be evaluated, leading to the quantification of signal oscillations.
While it is relatively simple to measure external forces, the measurement of
internal forces acting on body segments remains difficult. Direct measurement of
forces acting on a joint or within a body segment would require instrumentation
attached to each tendon. Hence, direct measurement is not always possible. How-
ever, with the help of known anthropometric and kinematic data, it is possible to
calculate and predict these forces using inverse dynamics. The dynamics within a
joint indicate the mechanical cause of movement and contribute to the understanding
of compensatory mechanisms adopted by the body to any disturbances within the
500
450
400
F10
350
Force (N)
300
250
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100
% of Stance
Fig. 2 A representative example of Ground Reaction Force during walking from a patient with
Cerbral Palsy
central nervous system (Winter 1990). Inverse dynamics calculation requires the
measurement of mass, location of COM and moment of inertia of the segments
examined, and location of joint centers. This investigation also requires kinetic
information such as the gravitational force and the GRF. Several studies have been
conducted to demonstrate the usefulness of inverse dynamics (Koopman et al. 1995;
Mommersteeg et al. 1997; Risher et al. 1997). Most commercial gait analysis
software available today uses inverse dynamics, requiring minimum input from the
operator. However, an understanding of convention reference frames on which
moments are calculated requires careful consideration (Manal et al. 2002; Schache
and Baker 2007). There has been much work on the standardization of data collec-
tion and management in clinical gait analysis which has led to the reliability and
repeatability (Wu and Cavanagh 1995; Benedetti et al. 1998).
Time Domain Analysis
The features of the GRF which are typically analyzed include the peak forces, initial
slopes, and symmetry between limbs. For example, for the vertical GRF, commonly
assessed variables include the maximum vertical forces (first peak usually termed
F1, second peak termed F3), the minimal vertical force (the valley termed F2), and
loading/unloading rates (Hesse et al. 1994).
Kesar et al. (2011a) reported the minimal detectable change for gait variables,
including peak anterior GRF and mean vertical GRF, providing references to help
interpret the magnitudes of changes in poststroke gait variables. Also statistical
methods such as signal correlation coefficient matrix can be applied to time domain
GRF data, to assess poststroke gait during gait retraining (Szczerbik et al. 2014).
(For further information please refer to the section titled: “▶ Methods and Models:
Signal Processing” within this book.)
Frequency Domain Analysis
As opposed to time domain analysis which is limited to selected points on force-time

graphs, frequency domain analysis allows examination of the entire waveform. This
type of analysis has been shown useful in the identification of gait abnormalities in
clinical condition such as cerebral palsy using harmonic analysis (White et al. 2005)
and multiple sclerosis using fast Fourier transformation (Wundeman et al. 2011). In a
study comparing the GRF between patients with scoliosis and healthy participants,
Giakas et al. (1996) demonstrated the usefulness of frequency domain analysis as
opposed to simple time domain analysis. The time domain data of various compo-
nents did not show any changes, but the frequency content in the mediolateral
component of patients was higher than the healthy participants indicating that
there is a change balance parameter.
Normalization Procedures
Traditionally kinetic and kinematic gait data is normalized to percentage of the gait
cycle. In addition, normalization techniques are used to compare gait data between
individuals of significantly different heights and/or bodyweights; GRF data is often
normalized to body weight.
There are numerous methods available to normalize gait data (Lulić et al. 2008;
Chockalingam et al. 2008, Dixon et al. 2014a). Moisio et al. (2003) demonstrated
that the application of two normalization techniques (body mass and bodyweight
times height) was highly effective in reducing height and weight differences when
examining joint moments. Research by Stansfield et al. (2003) examined normali-
zation of gait data in children and recommended the use of nondimensional normal-
ization as opposed to semi-dimensional normalization. (For further information
please refer to the section titled: “▶ Methods and Models: Data analytics” within
this book.)
Ground Reaction Force in the Understanding of Walking
Researchers have utilized measurement of the GRF during walking gait to under-
stand the normal functioning of the lower limbs. Hunt et al. (2001) used motion
capture and force plates to examine walking gait in participants. This study showed
that most of the intersegmental range of motion (rear-foot segment relative to the leg
and forefoot segment relative to the rear-foot) occurred at the beginning and end of
the stance phase when support was only on the rear-foot or forefoot and the GRF was
maximal. These findings confirm the significance of midfoot joints in foot function.
Using GRF data from the second force peak (Fig. 1) of the stance phase, Jacob
(2001) estimated forces acting across the joints of the first and second rays of the
forefoot in order to understand the physiological function of the foot. Results
showed that the resultant force on the first metatarsal head amounts to about 119%
body weight and the second metatarsal bone is also significantly loaded, however
more in bending.
The analysis of GRF has also been used to examine the effect of footwear and
orthoses. Nester et al. (2003) examined the effect of foot orthoses (medially and
laterally wedged) on the kinematics and kinetics of walking gait (for further infor-
mation please refer to the chapter titled: “▶ Functional Effects of Foot Orthoses”
within the Prosthetics and Orthotics section of this book). Examination of the GRFs
found that medially wedged orthoses increased the laterally directed GRF during the
stance phase, suggesting reduced shock attenuation, while laterally wedged orthoses
were found to decrease the laterally directed GRF, suggesting increased shock
attenuation. A comparison of the kinetics of high-heeled walking gait (Esenyel
et al. 2003) compared walking in low-heeled sports shoes to high-heeled dress
shoes, showing that when walking in high heels muscle moments and work are
reduced at the ankle which resulted in compensatory increases in these variables at
the hip and knee.
Ground Reaction Force in the Assessment of Clinical Conditions
Scoliosis
Adolescent idiopathic scoliosis (AIS) is a spinal deformity that regularly develops

during the period of rapid growth. While the causes and progression of AIS are still
not understood, biomechanical factors require careful consideration (Burwell et al.
2000; Stokes 1997). In gait analysis the majority of investigations have focused on
kinematic measurements. Therefore, establishing a relationship between kinematic
and kinetic data could lead to a better understanding of the etiology of scoliosis.
Both angular and translational asymmetry of the vertebrae of the rib cage and
back surface are associated with AIS (Stokes and Gardner-Morse, Stokes and
Gardner-Morse 1991). As a result of this structural deformity, the position of the
COM and the weight distribution about the lower limb are altered. This would have
implications on the three components of the GRF. An analysis of GRF measure-
ments have indicated significant differences between scoliotic and healthy children
that was mostly in the mediolateral direction (Giakas et al. 1996). In contrast, an
examination of the GRFs revealed no relationship between a gait asymmetry and a
scoliotic curve (Schizas et al. 1998).
Another study considered the vertical component of the ground reaction force to
report differences in the symmetry indices (SIs) within various reported parameters
(Chockalingam et al. 2004). Although the SIs did not exceed the range reported in
previous studies, there was a marked difference between the left and right side
impulse.
The reported results indicated that the patients with a left compensation curve had
a greater SI for a left side impulse and subjects with very little or no compensation
had a greater right side impulse. This was considered to be due to the compensation
in gait where the subjects compensate on the opposite pelvis/lower limb. Although a
previous investigation (Giakas et al. 1996) had indicated that there were no differ-
ences in the peak force values in the time domain, the results showed differences in
the frequency domain. Chockalingam et al. 2004 were the first group to consider
impulse. Since impulse is the rate of change of momentum and represented by
product of force and time, if either of the values is higher, the estimated
impulse will be higher. While results indicated that the average peak force values
are not that different between left and right sides, the interpretation was that the
subjects had a higher stance time on the side of major curve in the lumbar or lower
thoracic region.
One of the previous investigations (Kim and Eng 2003) looking into the temporal
parameters of gait in stroke patients showed asymmetries with swing phase time.
However, this investigation did not consider gait speed or the timing of various
phases. From the descriptive statistics, it appears that the standard deviations for the
vertical force peaks and the peaks of shear forces sum were small, indicating the
variability of the gait parameters was relatively low. Parameters showing highest
deviation were the loading and unloading rates. Previous studies indicate higher
symmetry for vertical forces and lower symmetry for mediolateral forces (Herzog
et al. 1989; Kim and Eng 2003). Taking this factor of higher asymmetry for vertical
forces into consideration and since the results indicate differences between left and
right side impulse, this method could be extended to detect the severity of the curve
and gait compensation in scoliotic subjects.
Chockalingam et al. (2004) while highlighting the relationship between the side
of the scoliotic curve and impulse showcased the value of using kinetic parameters in
developing further understanding of the pathogenesis and etiology of scoliosis and
similar conditions.
Recent research (Park et al. 2016) identified that asymmetry indices of GRF
magnitudes in walking positively correlated with adjusted Cobb’s angle and maxi-
mum Cobb’s angle, while asymmetry indices of GRF time variables positively
correlated with pelvic tilt. (For further information please refer to the chapter titled:
“▶ Impact of Scoliosis on Gait” within the Other Pathologic Gait Disorders section
of this book.)
Cerebral Palsy
People with cerebral palsy (CP) often experience difficulty in walking. These
individuals can display difficulty in supporting their bodyweight during the late
stance phase of gait and utilize compensatory mechanisms to prevent collapse of the
affected limb (Williams et al. 2011). While it has been shown that spatiotemporal
parameters in this population are reproducible measures, only selected GRF param-
eters (vertical peak force in terminal stance and terminal stance time) have shown
acceptable stability and reproducibility (White et al. 1999).
Traditionally clinical gait analysis solely consists of the assessment of straight-
line walking; however recent research has examined the differences in kinematic and
kinetics between typically developing children and children with CP during turning
tasks (Dixon et al. 2014b, 2016). Significant differences were evident between the
typically developing children and those with CP for a number of spatiotemporal
parameters, and in many individuals, the intra-subject differences between straight-
line walking and turning gait were larger in the children with CP than the typically
developing children. The authors propose that turning gait may be a better discrim-
inant of pathology than straight-line walking and could be used to improve the
management of gait abnormalities. (For further information please refer to the
section titled “▶ Medical Applications: Cerebral Palsy” within this book.)
Stroke
Multiple lower extremity joints can be affected by stroke resulting in individuals

experiencing gait abnormalities poststroke. Anterior GRFs in the nonparetic and
paretic limbs of individuals’ poststroke are lower than those of healthy individuals,
while medial GRFs remain unaffected by stroke and lateral GRF shows some
differences which are dependent on the side of lesion (Sharma et al. 2015). Exam-
ination of muscle activity during gait suggests the reduction in anterior GRF is
related to exaggerated flexor muscle activity which may counteract the effects of the
plantar flexors by offloading the leg (Turns et al. 2007). Gait retraining in this
population has been shown to increase the GRFs, but asymmetry between the
limbs may remain (Hesse et al. 1994). In this population more symmetrical weight
bearing is related to more symmetrical temporal but not distance measures during
gait (Kim and Eng 2003).
Poststroke individuals are found to increase nonparetic propulsive forces when
changing walking speed; however, with gait retraining increases in paretic propul-
sive forces can be observed which contribute to increased self-selected walking
speeds (Hsiao et al. (2016). Recent research has shown that fast treadmill walking in
combination with function electrical stimulation (FES) showed improvement in peak
anterior ground reaction force when compared to self-selected walking speed with
FES (Kesar et al. 2011b), suggesting its potential in gait retraining. (For further
information please refer to the chapter titled: “▶ Movement Disorders in Stroke”
within the Other Pathologic Gait Disorders section of this book.)
Multiple Sclerosis
The combined effects of skeletal muscle weakness, sensory disturbances, spasticity,

gait ataxia, and reduction in aerobic capacity lead to gait abnormalities in this
population (Rodgers et al. 1999). Research into gait classifications in this population
is limited; recent research found a number of variables (taken from 2D gait analysis,
EMG, and GRFs) that suggest insufficient ankle push-off in this population
(Kempen et al. 2016). Frequency content analysis has identified a lower frequency
content in this population when compared to healthy controls, suggesting lesser
vertical oscillation of the center of gravity (Wundeman et al. 2011).
Various treatment interventions are used which aim to improve gait in these
individuals. A 6-month aerobic training program was found to have minimal effect
on gait abnormalities (Rodgers et al. 1999). While the use of textured insoles has
been shown to alter gait patterns, their contribution to improving the gait pattern with
sufficient propulsion from the ankle plantarflexors is uncertain (Kelleher et al. 2010).
(For further information please refer to the chapter titled: “▶ Gait and Multiple
Sclerosis” within the Other Pathologic Gait Disorders section of this book.)
Summary
The majority of scientific and clinical studies using force plates and GRF are of gait
analysis. Force plates are an integral part of any gait analysis protocol, and the data is
used for analyzing the gait pattern and the development of effective clinical
management.
One also needs to appreciate that the force plates and GRF have also been used in
a wide range of medical and clinical applications, other than gait analysis. These
include the examination of postural sway during quiet standing, for example, in
elderly patients (Patla et al. 1992), the initiation of stepping (Patla et al. 1993), the
study of standing up in the elderly (Ikeda et al. 1991), and the analysis of stair
climbing (Andriacchi et al. 1980). Force plates as a technology have also been found
to be useful for nonmedical applications. For example, they are used in ergonomics,
in car impact tests, and in studies of the biomechanics of equine gait. While GRFs are
commonly used in sports to help optimize performance, force plates and reaction
forces have even been used on the space shuttle to measure astronauts’ ability to
apply forces and moments in a weightless environment.
As such, as indicated within this chapter, GRF is an essential part of any clinical
gait analysis and contributes not only to the description of gait in the time domain but
also to the assessment of joint moments. Overall, GRF is a crucial part of both
surgical and conservative clinical management involving gait.
References
Alexander NB, Mollo JM, Giordani B, Ashton-Miller JA, Schultz AB, Grunawalt JA, Foster NL
(1995) Maintenance of balance, gait patterns and obstacle clearance in Alzheimer’s disease.
Neurology 45(5):908–914
Andriacchi TP, Andersson GB, Fermier RW, Stern D, Galante JO (1980) A study of lower-limb
mechanics during stair-climbing. J Bone Joint Surg Am 62(5):749–757
Andriacchi TP, Hurwitz DE (1997) Gait biomechanics and the evolution of total joint replacement.
Benedetti MG, Catani F, Leardini A, Pignotti E, Giannini S (1998) Data management in gait
analysis for clinical applications. Clin Biomech 13(3):204–215
Burwell RG, Dangerfield PH, Lowe TG, Margulies JY (2000) Etiology of adolescent idiopathic
scoliosis, State of the art reviews. Hanley & Belfus Inc., Philadelphia
Buczek FL, Cavanagh PR (1990) Stance phase knee and ankle kinematics and kinetics during level
and downhill running. Med Sci Sports Exerc 22(5):669–677
Cavanagh PR, Lafortune MA (1980) Ground reaction forces in distance running. J Biomech
13(5):397–406
Chockalingam N, Giakas G, Iossifidou A (2002) Do strain gauge force platforms need in situ
correction? Gait Posture 16(3):233–237
Chockalingam N, Dangerfield PH, Rahmatalla A, Ahmed E-N, Cochrane T (2004) Assessment of
ground reaction force during scoliotic gait. Eur Spine J 13(8):750–754
Chockalingam N, BandCi S, Rahmatalla A, Dangerfield PH, Ahmed E-N (2008) Assessment of the
centre of pressure pattern and moments about S2 in scoliotic subjects during normal walking.
Scoliosis 3:10. doi:10.1186/1748-7161-3-10
Davis BL, de Vasconcellos AS, Lundin TM (1997) Uncertainty in 3-D joint moments associated
with human gait. In: Proceedings of the XVI congress of the international society of biome-
chanics. University of Tokyo, Tokyo, p 136
Defebvre L, Blatt JL, Blond S, Bourriez JL, Gieu JD, Destee A (1996) Effect of thalamic
stimulation on gait in parkinson disease. Arch Neurol 53(9):898–903
Dixon PC, Bowtell MV, Stebbins J (2014a) The use of regression and normalisation for the
comparison of spatio-temporal gait data in children. Gait Posture 40(4):521–525
Dixon PC, Stebbins J, Theologis T, Zavatsky AB (2014b) Ground reaction forces and lower-limb
joint kinetics of turning gait in typically developing children. J Biomech 47(15):3726–3733
Dixon PC, Stebbins J, Theologis T, Zavatsky AB (2016) The use of turning tasks in clinical gait
analysis for children with cerebral palsy. Clin Biomech 32:286–294
Gage JR, Koop SE (1995) Clinical gait analysis: application to management of cerebral palsy. In:
Allard P, Stokes IAF, Blanchi JP (eds) Three dimensional analysis of human movement. Human
Kinetics. Champaign, IL, pp 349–362
Gainey JC, Kadaba MP, Wootten ME, Ramakrishnan HK, Siris ES, Lindsay R, Canfield R, Cochran
GV (1989) Gait analysis of patients who have Paget disease. J Bone Joint Surg 71(4):568–579
Gehlsen G, Beekman K, Assmann N, Winant D, Seidle M, Carter A (1986) Gait characteristics in
multiple sclerosis: progressive changes and effects on parameters. Arch Phys Med Rehabil
67(8):536–539
Giakas G, Baltzopoulos V, Dangerfield PH, Dorgan JC, Dalmira S (1996) Comparison of gait
patterns between healthy and scoliotic patients using time and frequency domain analysis of
ground reaction forces. Spine 21(19):2235–2234
Herzog W, Nigg BM, Read LJ, Olsson E (1989) Asymmetries in ground reaction force patterns in
normal human gait. Med Sci Sports Exerc 21:10–114
Hesse SA, Jahnke MT, Bertelt CM, Schreiner C, Lücke D, Mauritz KH (1994) Gait outcome in
ambulatory hemiparetic patients after a 4-week comprehensive rehabilitation program and
prognostic factors. Stroke 25(10):1999–2004
Hill SW, Patla AE, Ishac MG, Adkin AL, Supan TJ, Barth DG (1997) Kinematic patterns of
participants with a below knee prosthesis stepping over obstacles of various heights during
locomotion. Gait Posture 6(3):186–192
Hsiao H, Awad LN, Palmer JA, Higginson JS, Binder-Macleod SA (2016) Contribution of paretic
and nonparetic limb peak propulsive forces to changes in walking speed in individuals post-
stroke. Neurorehabil Neural Repair 30(8):743–52
Hunt AE, Smith RM, Torode M, Keenan AM (2001) Inter-segment foot motion and ground reaction
forces over the stance phase of walking. Clin Biomech 16(7):592–600
Ikeda ER, Schenkman ML, Riley PO, Hodge WA (1991) Influence of age on dynamics of rising
from a chair. Phys Ther 71(6):473–481
Jacob HA (2001) Forces acting in the forefoot during normal gait – an estimate. Clin Biomech 16
(9):783–792
Kelleher KJ, Spence WD, Solomonidis S, Apatsidis D (2010) The effect of textured insoles on gait
patterns of people with multiple sclerosis. Gait Posture 32(1):67–71
Kempen JC, Doorenbosch CA, Knol DL, de Groot V, Beckerman H (2016) Newly identified gait
patterns in patient with multiple sclerosis may be related to push-off quality. Phys Ther 96
(11):1744–1752
Kesar TM, Binder-Macleod SA, Hicks GE, Reisman DS (2011a) Minimal detectable change for gait
variables collected during treadmill walking in individuals post-stroke. Gait Posture 33
(2):314–317
Kesar TM, Reisman DS, Perumal R, Jancosko AM, Higginson JS, Rudolph KS, Binder-Macleod
SA (2011b) Combined effects of fast treadmill walking and functional electrical stimulation on
post-stroke gait. Gait Posture 33(2):309–313
Kim CM, Eng JJ (2003) Symmetry in vertical ground reaction force is accompanied by symmetry in
temporal but not distance variables of gait in persons with stroke. Gait Posture 18(1):23–28
Koopman B, Grootenboer HJ, de Jongh HJ (1995) An inverse dynamics model for the analysis
reconstruction and prediction of bipedal walking. J Biomech 28(11):1369–1376
Koozekanani SH, Balmaseda MT Jr, Fatehi MT, Lowney ED (1987) Ground reaction forces during
ambulation in parkinsonism: pilot study. Arch Phys Med Rehabil 68(1):28–30
Lulić TJ, Susić A, Kodvanj J (2008) Effects of arm swing on mechanical parameters of human gait.
Coll Antropol 32(3):869–873
MacKinnon CD, Winter DA (1993) Control of whole body balance in the frontal plane during
human walking. J Biomech 26:633–644
Manal K, McClay I, Richards J, Galinat B, Stanhope S (2002) Knee moment profiles during
walking: errors due to soft tissue movement of the shank and the influence of the reference
coordinate system. Gait Posture 15(1):10–17
McCaw ST, Devita P (1995) Errors in alignment of center of pressure and foot coordinates affect
predicted lower extremity torques. J Biomech 28:985–988
Moisio KC, Sumner DR, Shott S, Hurwitz DE (2003) Normalization of joint moments during gait: a
comparison of two techniques. J Biomech 36(4):599–603
Mommersteeg TJ, Huiskes R, Blankevoort L, Kooloos JG, Kauer JM (1997) An inverse dynamics
modelling approach to determine the restraining function of human knee ligament bundles. J
Biomech 30(2):139–146
Nester CJ, van der Linden ML, Bowker P (2003) Effect of foot orthoses on the kinematics and
kinetics of normal walking gait. Gait Posture 2:180–187
Nilsson J, Thorstensson A (1989) Ground reaction forces at different speeds of human walking and
running. Acta Physiol Scand 136(2):217–227
Park YS, Lim YT, Koh K, Kim JM, Kwon HJ, Yang JS, Shim JK (2016) Association of spinal
deformity and pelvic tilt with gait asymmetry in adolescent idiopathic scoliosis patients:
investigation of ground reaction force. Clin Biomech 36:52–57
Patla AE, Frank JS, Winter DA (1992) Balance control in the elderly: Implications for clinical
assessment and rehabilitation. Can J Public Health 83(Suppl 2):S29–S33
Patla AE, Frank JS, Winter DA, Rietdyk S, Prentice S, Prasad S (1993) Age-related changes in
balance control system: initiation of stepping. Clin Biomech 8(4):179–184
Perry J, Burnfield J (2010) Gait analysis: normal and pathological function, 2nd edn. SLACK Inc.,
Thorofare
Razak AHA, Zayegh A, Begg RK, Wahab Y (2012) Foot plantar pressure measurement system: a
review. Sensors 12:9884–9912
Risher DW, Schutte LM, Runge CF (1997) The use of inverse dynamics solutions in direct
dynamics simulations. J Biomech Eng 119(4):417–422
Rodgers MM, Mulcare JA, King DL, Mathews T, Gupta SC, Glaser RM (1999) Gait characteristics
of individuals with multiple sclerosis before and after a 6-month aerobic training program. J
Rehabil Res Dev 36(2):183–188
Sanderson DJ, Martin PE (1996) Joint kinetics in unilateral below-knee amputee patients during
running. Arch Phys Med Rehabil 77(12):1279–1285
Schache AG, Baker R (2007) On expression of joint moments during gait. Gait Posture 25
(3):440–452
Schizas CG, Kramers-de Quervain IA, Stussi E, Grob D (1998) Gait asymmetries in patients with
idiopathic scoliosis using vertical force measurement only. Eur Spine J 7:95–98
Schwartz MH, Rozumalski A, Trost JP (2008) The effect of walking speed on the gait of typically
developing children. J Biomech 41(8):1639–1650
Sharma S, McMorland AJ, Stinear JW (2015) Stance limb ground reaction forces in high function-
ing stroke and healthy subjects during gait initiation. Clin Biomech 30(7):689–695
Stansfield BW, Hillman SJ, Hazlewood ME, Lawson AM, Mann AM, Loudon IR, Robb JE (2003)
Normalisation of gait data in children. Gait Posture 17(1):81–87
Szczerbik E, Krawczyk M, Syczewska M (2014) Ground reaction force analysed with correlation
coefficient matrix in group of stroke patients. Acta Bioeng Biomech 16(2):3–9
Stokes IAF (1997) Analysis of symmetry of vertebral body loading consequent of lateral spinal
curvature. Spine 22:2495–2503
Stokes IAF, Gardner-Morse M (1991) Analysis of the interaction between vertebral lateral deviation
and axial rotation in scoliosis. J Biomech 24:753–759
Turns LJ, Neptune RR, Kautz SA (2007) Relationships between muscle activity and anteroposterior
ground reaction forces in hemiparetic walking. Arch Phys Med Rehabil 88(9):1127–1135
Vaughan CL, Davis BL, O’Connor J (1992) Gait analysis laboratory. Human Kinetics, Champaign
White R, Agouris I, Selbie RD, Kirkpatrick M (1999) The variability of force platform data in
normal and cerebral palsy gait. Clin Biomech 14(3):185–192
White R, Agouris I, Fletcher E (2005) Harmonic analysis of force platform data in normal and
cerebral palsy gait. Clin Biomech 20(5):508–516
Whiting WC, Zernicke RF (1998) Biomechanics of musculoskeletal injury. Human Kinet:41–85
Williams KR, Cavanagh PR, Ziff JL (1987) Biomechanical studies of elite female distance runners.
Int J Sports Med 8(S2):107–118
Williams SE, Gibbs S, Meadows CB, Abboud RJ (2011) Classification of the reduced vertical
component of the ground reaction force in late stance in cerebral palsy gait. Gait Posture 34
(3):370–373
Winter DA (1990) Biomechanics and motor control of human movement. Wiley, New York
Winter DA, Patla AE, Frank JS, Walt SE (1990) Biomechanical walking pattern changes in the fit
and healthy elderly. Phys Ther 70:340–347
Wu G, Cavanagh PR (1995) ISB recommendation for standardization in the reporting of kinematic
data. J Biomech 28(10):1257–1261
Wundeman SR, Huisinga JM, Filipi M, Stergiou N (2011) Multiple sclerosis affects the frequency
content in vertical ground reaction forces during walking. Clin Biomech 26(2):207–212
Interpreting Joint Moments and Powers
in Gait
L. H. Sloot and M. M. van der Krogt
Abstract
Gait analysis is becoming an increasingly important tool to provide a quantitative
description of a patient’s gait deviations. It is not only used to diagnose walking
disorders but also for treatment selection and evaluation. While spatiotemporal,
kinematic, and EMG parameters are commonly used to describe movement and
muscle activity, kinetic measures are less often evaluated, even though they give
insight into the moments and powers that drive human walking. As such, kinetic
parameters are able to connect abnormal movement to underlying muscle mal-
function and bony malalignment. This chapter focuses on the role of joint
moments and powers of the lower extremities in clinical gait analysis. After a
brief introduction of normal kinetic patterns, the clinical interpretation of abnor-
mal joint moments and powers is described. Next, typical deviations in lower
limb kinetics are illustrated for several patient populations, including stroke,
cerebral palsy, Duchenne muscular dystrophy, anterior cruciate ligament (ACL)
injury, and osteoarthritis (OA), and for patients walking with prostheses or
orthotics. This section also illustrates the clinical usefulness of specific kinetic
parameters in these patient populations, including their sensitivity to treatment
and ability to predict treatment outcome. The chapter illustrates that the role of
kinetics within clinical gait analysis deserves more attention, and potential appli-
cations should be further pursued.
Keywords
Clinical gait analysis • Kinetics • Joint moments • Joint powers
L.H. Sloot (*) • M.M. van der Krogt

Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, VU University
Medical Center, Amsterdam, The Netherlands
e-mail: l.sloot@vumc.nl; m.vanderKrogt@vumc.nl

DOI 10.1007/978-3-319-30808-1_32-1
2 L.H. Sloot and M.M. van der Krogt
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Interpretation of Joint Kinetics in Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Kinetic Patterns of Physiological Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Kinetic Patterns of Pathological Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Kinetic Deviations in Patient Populations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Precautions for Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Introduction
Clinical gait analysis is often performed in orthopedic and rehabilitation practice, to

assess walking disorders in patients with complex neurological, neuromuscular, or
orthopedic disorders. Together with physical examination, gait analysis is generally
considered an essential tool to diagnose impairments that limit gait and to evaluate
the effects of treatment. You can read more about it in section “Clinical Gait
Analysis” in chapter “▶ Clinical Gait Assessment by Video Observation and
2D-Techniques.” Often, kinematic data (joint and segment angles) are considered
the main outcome of gait analysis, as they quantitatively describe the abnormalities
of the movement pattern as covered in section “General Kinematics” chapter
“▶ 3D Kinematics of Human Motion.” In addition, they are easy to interpret and
directly related to how we “see” the abnormal gait, through video images or with the
naked eye. This chapter, however, describes the role of kinetic data in various
clinical contexts and why these data give essential additional information related to
the causes of abnormal movement patterns and the underlying muscle malfunction.
By definition, kinetics (synonymous with “dynamics”) describe the relation between
the motion of a body and its causes, namely, the forces acting on the body. Following
typical clinical gait analysis, this chapter focuses on lower body joint moments and
powers. A brief introduction into the main kinetic parameters is given below, since
the basics of kinetics in human gait is covered in section “General Kinetics” chapter
“▶ The Conventional Gait Model – Success and Limitations.”
Joint moments describe the net sum of all internal moments delivered by all
internal structures around a joint. Typically, joint moments are delivered by muscles
and, toward the end range of motion, by ligamentous or bony tissue. Each of these
structures generates an internal force across the joint, the effect of which is given by
the multiplication of this force with its moment arm. Where internal muscle,
ligament, and bone-generated forces are difficult to measure in vivo, their summed
effect can be calculated by the so-called inverse dynamics. This calculation takes
into account the measured segment motions, ground reaction forces, and the (esti-
mated) segment masses and inertia. The net joint moments during gait give an
indication of the minimum force level that muscles (or ligaments) produce at any
Interpreting Joint Moments and Powers in Gait 3
instant during the gait cycle in that they do not take into account for instance any
cocontraction of the muscles. It is important to note that while joint moments are in
this chapter defined as the internal moments (as delivered by internal structures),
literature has not always been consistent and depending on the time period and
discipline might report external moments (as delivered by the ground reaction force
over a joint).
Joint powers describe the net rate, amount, and timing of energy generation and
dissipation of all muscles and ligaments around a joint. If a muscle is delivering force
while shortening at the same time, this means it is generating mechanical energy.
This is calculated as the internal muscle force times the muscle velocity. Similarly,
when a muscle is lengthening while generating force, mechanical energy is dissi-
pated. When integrating power over time, the net joint work is obtained, which is a
measure for the total amount of generated (or dissipated) energy. It is important to
note that ligaments and tendons can store energy and return this energy at a later
point in the gait cycle. Energy can also be transferred between joints, with an
isometric biarticular muscle generating power at one joint while dissipating power
at another – without doing any actual work.
This chapter first describes the history and state of the art of measuring kinetics
for clinical applications, followed by a description of joint moment and power
patterns that are typically measured during normal gait. The clinical interpretation
section starts with the presentation of a clinical case, followed by a general discus-
sion of the clinical relevance of kinetic measures and finally its application to various
patient groups. The chapter ends with a discussion on future directions to improve
clinical interpretation of kinetic data.
State of the Art
Historically, the clinical measurement and interpretation of kinetic data followed the
introduction of kinematics and muscle activation. For the calculation of kinetics, the
objective measurement of ground reaction forces is important, the search for which
began as early as the late nineteenth century by Carlet and Marey, using air reservoirs
to measure applied force (Sutherland 2005). Ground reaction forces were initially
deduced from kinematic data and later measured using the first developed force
plates in the 1920s by Marey and Demeny (Sutherland 2005). It wasn’t until the
1970s that the first commercial force plates became available, enabling a wider
introduction in clinical settings by lowering the technical complexity and improving
their reliability and accuracy. Not long after that, methods were developed to
superimpose the ground reaction force on video images to qualitatively estimate
the ground reaction moment over a joint (see section “Clinical GRF” chapter
“▶ Interpreting Ground Reaction Forces in Gait” for more details into the clinical
interpretation of ground reaction forces in gait). The availability and increasing
power of computers in the 1980s and 1990s allowed for automatic calculation of
joint moments and powers in individual patients.
There are two ways to quantify joint moments (Rose et al. 1991). The first method
provides a quick approximation by relating the resultant ground reaction force vector
to the perpendicular distance from the rotation center of a joint. This quasi-static
method does not take into account the influence of inertial forces on the segments. In
addition, the method is limited to the stance phase of walking, and errors increase at
the more proximal joints. The more complex alternative, called inverse dynamics, is
typically used for clinical gait analysis. It calculates kinetics using a link-segment
model, with the input consisting of the magnitude, orientation, and point of appli-
cation of the ground reaction force vector, as well as the location of the joint centers
and segments, and the linear and angular acceleration of the segments derived from
kinematic data. In addition, this method is based on assumptions about the inertial
characteristics of the segments. A more detailed explanation of inverse dynamics can
be found in section “Inverse Dynamics” chapter “▶ The Conventional Gait Model –
Success and Limitations.” Standard models include the lower limb segments in the
calculations; however, new models are also proposed that calculate, for instance,
midfoot kinetics.
Today, several types of equipment are available to measure kinetics. Standard
clinical gait labs typically include one or a few ground-embedded force plates that
are synchronized with motion capture systems. More recently, high-tech gait labs
have been developed that include instrumented treadmills, i.e., with two or more
force plates incorporated, that allow for measurement of force data for left and right
feet separately over many consecutive strides while controlling walking speed. In
addition, instrumented treadmills open up new possibilities, such as measurement of
force data during mechanical perturbations to estimate, for instance, stiffness prop-
erties or to give real-time visual feedback on kinetic parameters for training pur-
poses. Another clinically important advancement is the ability to perform kinetic
measurements outside of the gait lab, using force shoes or pressure insoles in
combination with (smartphone-based) inertial sensors, digital goniometers, or com-
mercial motion controllers used in popular games (Muro-de-la-Herran et al. 2014;
Sprager and Juric 2015). These ambulatory measurements could potentially provide
more representative and long-term measurements of every day walking in patients
while eliminating the need for an expensive gait laboratory.
Interpretation of Joint Kinetics in Gait
Kinetic Patterns of Physiological Gait
The discussion of normal kinetic patterns focuses on the internal sagittal ankle, knee,
and hip moments as well as the knee and hip moments in the frontal plane, since
transversal joint moment curves vary considerably without showing a clearly inter-
pretable pattern (Schwartz et al. 2008; Sadeghi et al. 2001; Bovi et al. 2011; Woolley
2001; Rose et al. 1991). In addition to these joint moments, typical ankle, knee, and
hip power patterns are presented in Fig. 1. Trunk and arm moments and powers can
also have a large effect on gait, and perhaps an even larger effect in pathological gait,
Hip extension moment (Nm/kg) Hip abduction moment (Nm/kg) Hip power (W/kg)
← absorption − generation →
1.5 1.5 4
←adduction − abduction →
← flexion− extension →
1 1 3
2
0.5 0.5
1
0 0
0
−0.5 −0.5 −1
−1 −2
0 50 100 0 50 100 0 50 100
Knee extension moment (Nm/kg) Knee abduction moment (Nm/kg) Knee power (W/kg)
← absorption− generation →
1.5 1.5 4
← adduction− abduction→
← flexion −extension →
1 3
1
2
0.5 0.5
1
0 0
0
−0.5 −0.5 −1
−1 −1 −2
0 50 100 50 100 0 50 100
←dorsiflexion − plantarflexion→
Ankle extension moment (Nm/kg) Ankle power (W/kg)
1.5 4
1 3
2
0.5
1
0
0
−0.5 −1
−1 −2
0 50 100 0 50 100
% Gait Cycle % Gait Cycle
Fig. 1 Joint moments and powers from a normative dataset of 20 healthy adults (Bovi et al. 2011)
but are outside the scope of this chapter. A more in depth discussion of normal
kinetics can be found in section “General Kinetics” chapter “▶ The Conventional
Gait Model – Success and Limitations.”
At the ankle, three phases can be distinguished in the moment curve. In the first
10% of the gait cycle, the eccentric contraction of the ankle dorsiflexor muscles
exerts a dorsiflexion moment to control plantarflexion after heel strike, resulting in
negative power or energy absorption. This is followed by a plantarflexion moment
building up to about 50% of the gait cycle to control the progression of the shank
over the foot, which coincides with energy absorption up to around 40% of the gait
cycle. The plantarflexion moment continues during a concentric contraction of the
plantarflexors toward toe-off. This phase, characterized by ankle power generation,
is often called push-off and is used to propel the body upward and forward, as well as
to initiate the leg into swing. In normal gait, it is the most important energy
generation source during the gait cycle, contributing to about half of the total
positive work during a stride (Umberger and Martin 2007; Schwartz et al. 2008;
Bovi et al. 2011). In swing, moments and powers in the ankle are small; only a low
dorsiflexion moment is needed to lift up the foot from the ground.
The knee starts with a short flexion moment and energy absorption, associated
with eccentric contraction of the knee extensors to control weight acceptance. This is
followed by an energy generation phase with concentric contraction of the knee
extensors applying an extension moment to stretch the knee during 15–40% of the
gait cycle and an energy absorption phase with eccentric contraction of the extensors
to keep the knee from collapsing around push-off. Finally, eccentric contraction of
the knee flexors during terminal swing, to slow down the knee for initial contact,
results in a flexion moment and energy absorption. The energy absorbing role of the
knee has been related to the smoothing of the gait movement. This is facilitated by
the biarticulate muscles spanning the knee that transfer energy from the knee to the
ankle and hip (Sadeghi et al. 2001). In the frontal plane, the knee muscles generate an
abduction moment during most of the stance phase, which together with the
moments generated by the ligamentous tissue and bone provides stability around
the knee joint.
The hip joint starts with an extension moment, during which the concentric hip
extensor contraction results in energy generation during the loading response, to
control the forward acceleration of the trunk and hip. This extension moment
gradually continues into a flexion moment with the peak prior to push-off due to
eccentric hip flexor activity, resulting in energy absorption. The next phase consists
of energy generation by concentric contraction of the hip flexors during pre-swing
and initial swing. This is also referred to as the “pull-off phase”, because the leg is
pulled up and forward, and it is the second largest contribution to the propulsive
power during the gait cycle. At the end of swing, an extension moment is generated
to decelerate the forward-swinging leg. In the frontal plane, the hip muscles generate
an abduction moment throughout the stance phase to stabilize the pelvis.
It should be noted that the magnitude of all moment and power patterns highly
depends on walking speed, gender, and age. Increasing walking speed will increase
peak values and can shift the patterns (Stansfield et al. 2001; Schwartz et al. 2008).
Females have been found to walk with an increased knee flexion moment and
increased energy absorption in pre-swing (Kerrigan et al. 1998). Younger children
show immature kinetic patterns with a reduced peak ankle plantar flexion moment
and increased peak hip flexion moment (Sutherland 1997). These effects should be
taken into account when comparing the data of a patient to normative data.
Kinetic Patterns of Pathological Gait
Case Example
As an illustration of how kinetic data can complement clinical gait analysis, sagittal
plane kinetics and kinematics are shown for a 12-year-old patient with diplegic
spastic paresis in Fig. 2, both before and after treatment (see chapter “▶ Diagnostic
Gait Analysis Use in the Treatment Protocol for Cerebral Palsy” for more details
over diagnostic gait analysis in cerebral palsy). Before treatment, the boy walked
with excessive hip and knee flexion especially around initial contact. These devia-
tions were accompanied by increases in hip and knee moments. The ankle was only
Hip flexion angle (°) Hip extension moment (Nm/kg) Hip power (W/kg)
80 2
← extension − flexion →
← flexion − extension →
Norm
4 Pre SDR
60
1 Post SDR
40 2
20
0
0
0
−20 −1 −2
0 50 100 0 50 100 0 50 100
Knee flexion angle(°) Knee extension moment (Nm/kg) Knee power (W/kg)
80 2
← extension − flexion →
← flexion − extension →
4
60
1
40 2
20
0
0
0
−20 −1 −2
0 50 100 0 50 100 0 50 100
Ankle flexion angle(°) Ankle extension moment (Nm/kg) Ankle power (W/kg)
← plantarflexion − dorsiflexion →
← dorsiflexion − plantarflexion →
40 2
4
20
1
2
0
0
−20 0
−40 −1 −2
0 50 100 0 50 100 0 50 100
% Gait Cycle % Gait Cycle % Gait Cycle
Fig. 2 Case example of clinical gait analysis including kinetic data. One representative stride
before and after treatment is shown. Gray areas show normative data for typically developing
children, as derived from Schwartz et al. (2008). SDR selective dorsal rhizotomy treatment, TO
toe-off
in slight dorsiflexion during the second half of stance and showed an early peak in
ankle moment, as is typical for patients with spasticity in the calf muscles (see
section “Clinical EMG” chapter “▶ EMG Activity in Gait: The Influence of Motor
Disorders” for the clinical interpretation of muscle activity). The hip, knee, and ankle
power curves all show large peaks in the first half of stance. This early excessive
generation of energy is likely ineffective and not contributing to propulsion, but
rather to accelerating the body upward in early stance. After selective dorsal
rhizotomy treatment (in which some of the sensory nerve fibers that enter the spinal
cord are cut in order to reduce the effects of spasticity), the joint angles show some
improvement, but the excessive flexion remains. The joint moments and powers
however improve considerably, with the knee moment and power almost normaliz-
ing. The ankle push-off is reduced compared to normal both before and after
treatment, which is compensated by increased hip power in early stance.
Clinical Interpretation of Joint Moments

Clinically, net joint moments are used to obtain an estimate of the total load on
muscles, ligaments, or bones around a joint of an individual subject. Generally
speaking, the more these loads are increased compared to normal, the more reason
there is to intervene and to search for treatment options that would reduce the net
moments and hence the muscle or joint loading. Joint moments can then be com-
pared before and after treatment, to quantify the improvement, as illustrated for the
above clinical case (Fig. 2).
If joint moments are generated by muscles, they are typically interpreted in
parallel with muscle activity derived from electromyography (EMG). For instance,
if a high knee extension moment is measured, this moment has to be generated
internally by the knee extensor muscles, and hence activity of the quadriceps is
expected. If antagonist muscles (e.g., hamstrings) are active at the same time, this
indicates an increased level of cocontraction and hence an even higher burden on
muscles and joints. When joint moments are seen without significant muscle acti-
vation, this is indicative of passive force generation by muscle tissue or ligaments.
Joint moments that are caused by the ligaments or bone, such as knee
ab-adduction moments or knee flexion moments toward hyperextension, give an
indication of loading on the bone or ligaments and can be indicative of further
damage or joint degeneration. In case of back, hip, knee, or ankle pain, joint
moments are used to assess whether or not this pain is caused by abnormal loading
in the joint and hence should be the focus of treatment. Another example is the
relevance of the knee adduction moment in knee osteoarthritis patients, as discussed
in more detail below.
Next to abnormally high moments that indicate potential muscle or joint loading
problems, patients can also exhibit decreased joint moments. Decreased joint
moments can be an indication of muscle weakness but also of ineffective muscle
function. For instance, muscles may generate forces outside their optimal working
area due to shortened muscles and tendons. Another cause can be lever-arm dys-
function, referring to a reduction in the moment arm due to bony deformities,
inefficient flexible moment arms, or an abnormal point of rotation or muscle
attachment. Abnormal moment arms can greatly interfere with a person’s ability to
walk. For example, some patients with spastic diplegia have a flexible and
mal-directed ankle moment arm due to valgus feet, midfoot break, or other defor-
mities. During normal gait, the soleus muscle slows the forward progression of the
shank and brings the ground reaction force in front of the knee, resulting in a knee
extension moment. The ineffective moment arm, however, results in a decreased
extension moment for the same soleus muscle force, and thus increased muscle effort
is required to extend the leg. Since both lever-arm dysfunction and a shortened
muscle-tendon complex can usually be corrected, decreased joint moments might be
indicative of a need for orthopedic surgery or bracing.
Clinical Interpretation of Joint Powers

Abnormal joint powers and joint work can show us how a patient manages to propel
his body forward during gait but also where energy is dissipated without effective
propulsion. As stated above, ankle push-off is the main “driver” of normal gait, but it
is often affected in many patient populations. Reduced ankle push-off can be due to
weakness or spasticity of calf muscles, issues with voluntary control, or the restric-
tions imposed by ankle-foot orthoses (AFOs) or prostheses. It has been estimated
that weakness of the calf muscles limits gait more than weakness in any of the other
muscles (van der Krogt et al. 2012), stressing the importance of adequate ankle
power generation.
Ankle push-off is important to assess during clinical gait analysis as a measure for
gait efficacy. If ankle push-off is reduced, the most typical compensation is the use of
additional hip joint work, either by the hip flexors around toe-off or hip extensors
during loading response and early stance (Fig. 2). Using a conceptual double-
inverted pendulum model, it has been shown that such a hip strategy could be up
to four times more energetically costly compared with an ankle strategy (Donelan
et al. 2002). This is due to the fact that an adequate and timely ankle push-off
accelerates the body’s center of mass upward before the leading leg hits the ground.
As a result, the impact of the leading leg is reduced and less energy is dissipated.
Although this model has several simplifications as compared to human (pathologi-
cal) gait, the relative contribution between ankle and hip work may still be a relevant
measure to determine the efficacy of gait. This ratio was found to be 0.9 in healthy
children (indicating slightly less ankle than hip work) and greatly diminished in
patients with cerebral palsy (Riad et al. 2008).
Another clinically important aspect of ankle push-off is its role in swing initiation.
Several deviations during swing may result from a weak push-off, including limited
knee flexion in early swing, foot clearance, and knee extension in late swing.
However, these kinematic deviations could also have other potential causes, such
as rectus femoris spasticity for limited knee flexion and toe clearance in early swing.
Accurate measurements of ankle push-off power as well as of moments and powers
around other joints are therefore essential to determine the cause of gait abnormal-
ities. Other clinically relevant aspects of joint powers related to specific patient
populations are described in more detail below.
Kinetic Deviations in Patient Populations
Patients with different pathologies walk with varying deviations that are often typical
for their impairments. Therefore, the following sections highlight the alterations in
lower limb joint moments and powers for different patient groups. These are related
primarily to altered neuromuscular control (stroke and CP), muscle weakness
(Duchenne muscular dystrophy), ligament rupture (anterior cruciate ligament
injury), bone degeneration (osteoarthritis), and mechanical properties of assistive
devices (prostheses and orthotics). In addition, the sensitivity of specific kinetic
parameters to different interventions and their ability to predict treatment outcomes

are discussed. Although this is not an extensive overview of current literature, it
highlights the clinical importance of the measurement of kinetic parameters for
different patient groups.
Stroke
Stroke, also known as cerebrovascular accident (CVA), is caused by hampered blood
flow in the brain, often resulting in a reduced ability to move one side of the body
(see chapter “▶ Movement Disorders in Stroke” on movement disorders in stroke).
Even though the changes in gait patterns are highly variable between patients,
common kinetic gait deviations at the paretic side include reduced peak knee
power absorption, associated with knee hyperextension, and reduced push-off
ankle power during late stance (Kim and Eng 2004; Olney and Richards 1996;
Riley and Kerrigan 2005). In addition, the ankle dorsiflexion moment at initial
contact is often reduced due to limited ankle dorsiflexion during swing (Kim and
Eng 2004). Coping strategies at the paretic side often include larger hip extension
moments during early stance as well as increased hip flexion moments and hip power
generation during pre-swing (Jonkers et al. 2009; Olney and Richards 1996). The
non-paretic side also adapts to compensate for reductions in moments and powers of
the paretic side, but peak moments remain within normal ranges (Kerrigan et al.
1999; Woolley 2001).
Treatments that are focused on improving gait performance and push-off in stroke
patients through gait training often report the effect on kinetic outcome measures.
Generally, it is reported that gait training, augmented with body-weight support and
visual or biofeedback, results in an increase in ankle power generation at push-off,
hip flexion power, and a reduction in knee extension moment at the affected side, as
well as an increase in hip flexion moment at the unaffected side (Aiello et al. 2005;
Lennon et al. 2006; Mirelman et al. 2010; Mulroy et al. 2010). Ankle power
generation at push-off has been shown to be sensitive to differences in training,
with a larger increase after virtual reality-based training compared with normal gait
training (Mirelman et al. 2010). Next to gait training, kinetic parameters have been
found to be sensitive to other interventions. For instance, walking aids such as canes
or crutches can lead to increasing hip flexion power generation; tendon surgery in
equinovarus feet resulted in increasing ankle power generation during stance at the
unaffected side and in decreasing ankle power generation and absorption at the
paretic side; pharmacological spasticity treatment improved knee power absorption
and increased ankle push-off, and functional electrical stimulation has been shown to
improve peak hip and ankle power around push-off (Sheffler et al. 2016; Stoquart
et al. 2008; Carda et al. 2009; Polese et al. 2012; Novak et al. 2015). These results
justify more focus on kinetic parameters, especially when an intervention aims to
improve walking economy and thus ankle push-off or hip pull-off.
Cerebral Palsy
Cerebral palsy (CP) describes a group of permanent movement disorders that are
attributed to nonprogressive disturbances in the developing fetal or infant brain
(Rosenbaum et al. 2007). Due to the heterogeneity of the condition, it is associated

with a range of common gait patterns (Lin et al. 2000) (see chapters “▶ Spasticity
Effect in Cerebral Palsy Gait,” “▶ Swing Phase Problems in Cerebral Palsy,”
“▶ Stance Phase Problems in Cerebral Palsy (Strength),” and “▶ Foot And Ankle
Motion in Cerebral Palsy” for more details on movement disorders in CP). The
excessive hip, knee, and ankle flexion angles in crouch gait are related to abnormally
high extension moments in these joints during stance. The knee hyperextension in
recurvatum gait is associated with a large knee flexion moment during stance and
plantar flexion moment after initial contact. Jump knee gait is illustrated by a large
knee extension moment and abnormal knee power generation during the loading
response, resulting in the typical upward body movement. Jump knee gait also
includes toe walking, which is related to abnormal ankle power dissipation at initial
contact and premature generation in early and mid-stance as well as a “double-
bump” ankle moment pattern, with a plantar flexion moment peak during the loading
response and toward terminal stance. The ratio between these peaks has been shown
to provide objective means to quantify the severity of toe walking (Alvarez et al.
2007). Finally, ankle power during push-off is typically diminished in all gait
patterns (Riad et al. 2008), often resulting in a shift toward a hip strategy that
consists of an increasing hip extension moment during early stance and flexion
moment during pre-swing.
Although treatment effects seem primarily evaluated in terms of spasticity scales,
functional measures as well as spatiotemporal and kinematic parameters, the nor-
malization of kinetic gait patterns is getting increased attention. For instance,
pharmacological spasticity treatment has been shown to result in normalization of
the ankle double-bump pattern (Boyd et al. 2016; Desloovere et al. 2001), and
strength training led to an increase in for instance hip flexion moment (Lee et al.
2008). Orthotics are often prescribed in CP, and their effect on knee and ankle
moments will be discussed in more detail later. Soft tissue surgery normalized the
knee flexion moments in recurvatum gait and knee extension moments in crouch gait
(Park et al. 2006; Adolfsen et al. 2007). Overall, the recognition of kinetic measures
to provide essential information for treatment selection as well as evaluation is
slowly growing.
Duchenne Muscular Dystrophy

Duchenne muscular dystrophy (DMD) results in progressive muscle weakness of the
legs and pelvis and eventually the arms and other areas. To compensate for the
weakness of, for instance, the lower leg extensors, the demand is decreased by
positioning the ground reaction vector behind the hip joint and in front of the knee
joint during stance (Heberer et al. 2016). Due to these compensation strategies,
kinetic patterns of children with DMD deviate from normal. This includes reduced
hip extension moments and powers and reduced hip abductor moments (Gaudreault
et al. 2010; D’Angelo et al. 2009; Heberer et al. 2016). In addition, reduced knee
extension moments have been reported (Doglio et al. 2011; Gaudreault et al. 2010),
although this finding is not consistent (Heberer et al. 2016). Generally, children with
DMD walk without an ankle dorsiflexion moment around initial contact and with
reduced ankle plantar flexion moments and power generation around push-off
(Gaudreault et al. 2010; D’Angelo et al. 2009; Heberer et al. 2016; Doglio et al.
2011).
These kinetic parameters seem of extra clinical importance, because of their
sensitivity to treatment. While kinematic parameters could not detect a difference
between treated and nontreated groups, peak and duration of hip extension moment
as well as peak hip and ankle power generation were sensitive for steroid treatment in
DMD patients (Heberer et al. 2016; D’Angelo et al. 2009). In addition, it is
suggested that hip kinetics might reveal muscle weakness before kinematic abnor-
malities can be detected (Heberer et al. 2016). These findings urge for the measure-
ment of lower leg joint kinetics as early markers of muscle weakness in children
with DMD.
Anterior Cruciate Ligament Injury

Injury of the anterior cruciate ligament (ACL) inhibits its normal function, which is
to provide internal stability in the knee joint and limit knee hyperextension (see
chapter “▶ ACL Injuries and Gait” on ACL injuries and gait). The injury usually
occurs in young and active individuals, predominantly woman. Kinetic parameters
are of particular interest to study the increased loading of the knee joint after ACL
injury, which is associated with additional injuries such as knee joint degeneration.
Patients with an ACL injury walk with increased hip extension moments and power
generation during early and mid-stance as well as decreased hip flexion moments
and reduced knee power absorption during late stance (Ferber et al. 2002). The
uninjured knee has also been reported to have altered kinetics (Hart et al. 2010).
The knee abduction moment during gait has been found to predict ACL injury
status, while the hip rotation moment could predict the recurrence of an ACL injury
(Hewett et al. 2005; Paterno et al. 2010). The kinetic patterns become more normal
after reconstructive surgery, although they remain considerably different and show
increased knee extension moments during early stance (Hart et al. 2010; Ferber et al.
2002; Paterno et al. 2010). Thus, monitoring knee and hip joint kinetics helps to
understand neuromuscular control of the knee and identify individuals at risk for
further injury.
Osteoarthritis
Osteoarthritis (OA) is a degenerative joint disease, mostly affecting the medial
tibiofemoral compartment of the knee, which causes chronic pain, stiffness, and
disability in primarily the older generation (Chu and Andriacchi 2015). Patients with
knee OA typically demonstrate increased knee abduction moments during
mid-stance as well as decreased knee extensor moments, hip abduction moments,
and hip flexion moments. The external knee adduction moment (KAM) serves as a
surrogate measure of the compressive forces on the medial compartment of the knee,
leading to an equal internal knee abduction moment. Increased KAM has indeed
been linked to varus malalignment of the knee joint in medial OA patients, to
increased pain, and to some extent to decreased function (Tate et al. 2012; Foroughi
et al. 2009). In addition, this measure and its integral (the KAM impulse) have been
found to correlate with OA severity and progression and differentiate between

medial and lateral compartment knee OA (Reeves and Bowling 2011; Foroughi
et al. 2009). Other kinetic measures as hip moment and ankle dorsiflexion moment
have also been related to OA disease progression (Astephen et al. 2008).
The KAM has been used to monitor treatment outcome. The measure has been
reported to reduce after surgical interventions and can be used to predict subsequent
reemerging of varus deformity (Andriacchi and Mündermann 2006; Tate et al.
2012). The effect of conservative treatments, such as walking with a cane, lateral
wedge insoles, load-modifying shoes, or knee braces, on KAM seems to depend on
the patient and disease severity, but reductions have been found (Simic et al. 2011;
Reeves and Bowling 2011). Another emerging intervention is gait training with the
use of feedback to stimulate toe-out or toe-in walking, lateral trunk lean, or wider
step width. These kinds of gait training have been found to reduce the KAM
(Richards et al. 2016). Thus, the KAM has been shown to be a functional marker
for risk of OA progression and recurrence of medial compartment knee OA.
Prostheses and Orthotics

Lower limb amputations can occur below the knee (transtibial), above the knee
(transfemoral), or through the knee or ankle and are usually necessitated due to
vascular disease, trauma, or infection (see chapters “▶ Prosthetic Foot Principles and
their Influence on Gait” and “▶ The Influence of Prosthetic Knee Joints on Gait” for
more details on the effect of ankle and knee prostheses on gait). As a result of these
amputations, the ankle power generation during push-off to propel the body forward
is lost. Many prostheses are able to store energy during early stance and return
energy during push-off, although this only partially compensates for the lost ankle
power (Winter and Sienko 1988; Ventura et al. 2011). A common compensation
strategy is to increase hip extension moments during early stance and hip flexion
moments and power generation during pre-swing (Whittle 1996; Silverman et al.
2008; Winter and Sienko 1988). The non-amputated side typically increases the
ankle plantar flexion moment and hip flexion moment during mid-stance to enable
the swing phase of the prosthesis (Rietman et al. 2002; Seroussi et al. 1996). In
transfemoral amputees the knee moments and powers are abnormal, because the
knee is kept in an extended position during the stance phase and, after allowing some
knee flexion during swing, slowed down mechanically for initial contact (Ingraham
et al. 2016; Whittle 1996; Seroussi et al. 1996). Kinetic parameters have been shown
to be able to differentiate between different (controlled or activated) prostheses and
fallers versus non-fallers (Vanicek et al. 2009; Sawers and Hafner 2013; Versluys
et al. 2009). Kinetic gait parameters are valuable to optimize alignment and adjust-
ment of prostheses, as well as to examine the energy absorption and return charac-
teristics of different passive and active designs.
Lower leg orthotics, such as ankle-foot orthoses (AFOs), are often prescribed for
different patient groups, such as CP and stroke. AFOs are generally used to improve
gait economy and joint stability as well as to correct deformities and prevent
contractures (see chapters “▶ Ankle Foot Orthoses and their Influence on Gait”
and “▶ Functional Effects of Foot Orthoses” for more details). The effect of an
orthosis depends on its specific characteristics, such as design (anterior or posterior

brace in an AFO), material stiffness, and inclusion of springs or hinges (Whittle
1996). An orthosis designed to reduce knee flexion will generally result in a
decreased knee extension moment. Spring-like orthoses can support the ankle
power generation during push-off by returning some stored energy during loading
response. Rigid orthoses on the other hand often hamper the ankle power generation
around push-off. Since contradictory results have been found on the effectiveness of
orthoses, individual alignment and verification using gait analysis is increasingly
advocated. The current recommended parameter for alignment is based on the
shank-to-vertical angle of the orthosis and footwear combination (i.e., a kinematic
parameter), although kinetic parameters might better reflect whether the optimal
alignment of the ground reaction force vector is achieved (Ridgewell et al. 2010).
The development from premade orthoses to the possibility of rapid prototyping and
3D printing of customized orthoses only further increases the need for kinetic gait
analysis (Mavroidis et al. 2011).
Precautions for Interpretation
While joint moment and powers can provide essential information for treatment
selection in certain patient groups, there are some factors that should be taken into
account when clinically interpreting the measured data. First and foremost, it is
important to realize that the presented joint moments and powers are net values, thus
describing the net exerted forces and associated energy conversion and ignoring, for
instance, muscle cocontraction. Since activation of different muscles can result in the
same net force, the role of individual muscles cannot be derived from these kinetic
measures alone. This problem is exacerbated for neurological patients, with more
unpredictable and abnormal muscle activation patterns.
When evaluating kinetic data, it is important to consider the limitations of the
measurement equipment used. For instance, the representativeness and repeatability
of kinetic data collected during common overground walking might be affected by
the limited number of available strides, restricted walking distance, the necessity to
hit on the force plate, and lack of standardization of walking speed. Instrumented
treadmills on the other hand have been shown to alter the walking pattern, especially
in patients (e.g., Brouwer et al. 2009; van der Krogt et al. 2015). In general, it is
important to realize that most lab-based set-ups measure gait in an artificial situation
and that results can highly depend on the patient’s motivation, fatigue, and the
instructions of the lab personnel.
Another important aspect of the measurement instrument is the resulting force
data accuracy. Typically, the large force plates in instrumented treadmills are slightly
less accurate and have more (movement) noise and system resonance compared with
ground-embedded force plates. Small changes in the force data, and thus the derived
center of pressure, can considerably alter the magnitude of joint kinetics. A shift of,
for instance, 10 mm in the fore-aft center of pressure can result in changes up to 7%
and 16% in the ankle and knee flexion moment, respectively, while a similar shift in
the mediolateral direction can reverse the inversion-eversion ankle joint moment
(Kim et al. 2007; McCaw and DeVita 1995). The effect of these errors is especially
important to keep in mind when studying transverse or frontal plane kinetic param-
eters, such as the knee abduction moment in medial knee osteoarthritis patients.
Besides errors in force data, the accuracy of the motion data as well as the synchro-
nization between these two can largely influence the timing and peak values of
kinetic measures.
Next to the measurement systems, the choices made in data analysis can also have
an impact on kinetic outcomes. Filtering of kinetic data and the application of
different filter characteristics to kinematic versus kinetic data can affect the timing
and magnitude of peak moments and powers. In addition, the method to calculate
kinetic measures, i.e., simple versus inverse dynamics, and the underlying assump-
tions of the latter, such as inferring healthy properties on pathological limbs,
prostheses, and overweight subjects, should be considered. Finally, factors such as
walking speed, age, and gender have been shown to affect the gait pattern (see the
“▶ Kinetic Patterns of Physiological Gait” section). Thus, for clinical interpretation
of kinetic data, it is important to have a good understanding of the effect of the used
measurement instrument, the force data quality, and the chosen analysis and to use
suitable normative datasets to compare patient data.
Future Directions
To aid further implementation of kinetic data in clinical gait analysis, several

developments are warranted. First, technological developments should focus on
reducing patient contact time, reducing complexity of measurement and analysis,
and almost instantly providing a gait report to the physician. In addition, the shift
toward portable, easy-to-use, and less expensive equipment will increase the acces-
sibility to kinetic parameters as a tool to diagnose and monitor patients for more
clinical sites.
Next to equipment development, considerable effort should be dedicated to
establish the role of kinetic parameters in the clinical decision-making process for
different patient populations. While the general contribution of clinical gait analysis
to treatment selection has been subject of examination, the specific added value of
kinetics remains largely unknown. To provide the context to pursue this question, the
relation between kinetic parameters versus gait deviations, other physiological
measures, and patient needs should be established in different patient populations.
In addition, the sensitivity of kinetic parameters to different interventions, their
efficacy to tune devices in individual patients, and ability to predict treatment
success should be subject of investigation. These studies will all contribute to
guidelines as to which cases require kinetic measurements as part of a clinical gait
analysis and as such direct the allocation of resources and time.
A better understanding of the relevance of kinetic outcomes could also contribute
to more focused gait reports and as such aid the interpretation of kinetics in relation
with kinematics and muscle activity. Essential prerequisites for accurate
interpretation of kinetic outcomes are the general availability of appropriate norma-

tive datasets, the establishment of the reliability of different measurement tech-
niques, and consensus on the clinically relevant difference and minimal detectable
change for different kinetic parameters.
This chapter has shown several clinical examples in which kinetic parameters are
relevant or even required to identify the underlying cause of gait deviations or select
the proper treatment in different patient populations. Therefore, the wider application
of kinetic parameters in clinical gait analysis deserves more attention, and its
potential contribution should be further established.
Cross-References

▶ ACL Injuries and Gait
▶ Clinical Gait Assessment by Video Observation and 2D-Techniques
▶ Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy
▶ Interpreting Ground Reaction Forces in Gait
▶ Movement Disorders in Stroke
▶ Prosthetic Foot Principles and Their Influence on Gait
▶ Spasticity Effect in Cerebral Palsy Gait
▶ Stance Phase Problems in Cerebral Palsy (Strength)
▶ Swing Phase Problems in Cerebral Palsy
▶ The Conventional Gait Model – Success and Limitations
▶ The Influence of Prosthetic Knee Joints on Gait
References
Adolfsen SE et al (2007) Kinematic and kinetic outcomes after identical multilevel soft tissue
surgery in children with cerebral palsy. J Pediatr Orthop 27(6):658–667
Aiello E et al (2005) Visual EMG biofeedback to improve ankle function in hemiparetic gait.
Conference proceedings: ... Annual international conference of the IEEE engineering in med-
icine and biology society. IEEE Engineering in Medicine and Biology Society. Conference, 7
(August 2016), pp 7703–7706
Alvarez C et al (2007) Classification of idiopathic toe walking based on gait analysis: development
and application of the ITW severity classification. Gait Posture 26(3):428–435
Andriacchi TP, Mündermann A (2006) The role of ambulatory mechanics in the initiation and
progression of knee osteoarthritis. Curr Opin Rheumatol 18(5):514–518
Astephen JL et al (2008) Biomechanical changes at the hip, knee, and ankle joints during gait are
associated with knee osteoarthritis severity. J Orthop Res 26(3):332–341
Bovi G et al (2011) A multiple-task gait analysis approach: kinematic, kinetic and EMG reference
data for healthy young and adult subjects. Gait Posture 33(1):6–13
Boyd RN et al (2016) Biomechanical transformation of the gastroc – soleus muscle with botulinum
toxin A in children with cerebral palsy. Dev Med Child Neurol 42:32–41
Brouwer B, Parvataneni K, Olney SJ (2009) A comparison of gait biomechanics and metabolic
requirements of overground and treadmill walking in people with stroke. Clin Biomech 24
(9):729–734. Available at: http://dx.doi.org/10.1016/j.clinbiomech.2009.07.004
Carda S et al (2009) Gait changes after tendon functional surgery for equinovarus foot in patients
with stroke. Am J Phys Med Rehabil 88(4):292–301. Available at: http://content.wkhealth.com/
linkback/openurl?sid=WKPTLP:landingpage&an=00002060-200904000-00008
Chu CR, Andriacchi TP (2015) Dance between biology, mechanics, and structure: a systems-based
approach to developing osteoarthritis prevention strategies. J Orthop Res 33(7):939–947
D’Angelo MG et al (2009) Gait pattern in Duchenne muscular dystrophy. Gait Posture 29(1):36–41
Desloovere K et al (2001) A randomized study of combined botulinum toxin type A and casting in
the ambulant child with cerebral palsy using objective outcome measures. Eur J Neurol 8
(5):75–87
Doglio L et al (2011) Early signs of gait deviation in Duchenne muscular dystrophy. Eur J Phys
Rehabil Med 47(4):587–594
Donelan JM, Kram R, Kuo AD (2002) Mechanical work for step-to-step transitions is a major
determinant of the metabolic cost of human walking. J Exp Biol 205(23):3717–3727. Available
at: http://jeb.biologists.org/content/205/23/3717.short
Ferber R et al (2002) Gait mechanics in chronic ACL deficiency and subsequent repair. Clin
Biomech 17(4):274–285
Foroughi N, Smith R, Vanwanseele B (2009) The association of external knee adduction moment
with biomechanical variables in osteoarthritis: a systematic review. Knee 16(5):303–309.
Available at: http://dx.doi.org/10.1016/j.knee.2008.12.007
Gaudreault N et al (2010) Gait patterns comparison of children with Duchenne muscular dystrophy
to those of control subjects considering the effect of gait velocity. Gait Posture 32(3):342–347.
Available at: http://dx.doi.org/10.1016/j.gaitpost.2010.06.003
Hart JM et al (2010) Sagittal plane knee joint moments following anterior cruciate ligament injury and
reconstruction: a systematic review. Clin Biomech 25(4):277–283. Available at: http://www.scopus.
com/inward/record.url?eid=2-s2.0-77949571691&partnerID=40&md5=ec1b2ea2ae68b2391abc48
9db4768a6d
Heberer K et al (2016) Hip kinetics during gait are clinically meaningful outcomes in young boys
with Duchenne muscular dystrophy. Gait Posture 48:159–164. Available at: http://linkinghub.
elsevier.com/retrieve/pii/S0966636216300753
Hewett TE et al (2005) Biomechanical measures of neuromuscular control and valgus loading of the
knee predict anterior cruciate ligament injury risk in female athletes: a prospective study. Am J
Sports Med 33(4):492–501. Available at: http://ajs.sagepub.com/content/33/4/492
Ingraham KM et al (2016) Assessing the relative contributions of active ankle and knee assistance
to the walking mechanics of transfemoral amputees using a powered prosthesis. PLoS One 11
(1):e0147661
Jonkers I et al (2009) Capacity to increase walking speed is limited by impaired hip and ankle power
generation in lower functioning persons post-stroke. Gait Posture 29(1):129–137. Available at:
http://linkinghub.elsevier.com/retrieve/pii/S0966636208002075. Accessed 30 Aug 2016
Kerrigan DC, Todd MK, Della Croce U (1998) Gender differences in joint biomechanics during
walking: normative study in young adults. Am J Phys Med Rehabil Assoc Acad Physiatrists 77
(1):2–7
Kerrigan DC et al (1999) Spastic paretic stiff-legged gait: biomechanics of the unaffected limb. Am
J Phys Med Rehabil 78(4):354–360. Available at: http://www.ncbi.nlm.nih.gov/pubmed/
10418842
Kim CM, Eng JJ (2004) Magnitude and pattern of 3D kinematic and kinetic gait profiles in persons
with stroke: relationship to walking speed. Gait Posture 20(2):140–146
Kim H, Sakurai S, Ahn J (2007) CoP Errors and 3D lower limb joint moment of gait. Int J Sport
Health Sci 5:71–82. Available at: http://www.soc.nii.ac.jp/jspe3/index.htm. Accessed 6 Oct
2016
Lee JH, Sung IY, Yoo JY (2008) Therapeutic effects of strengthening exercise on gait function of
cerebral palsy. Disabil Rehabil 30(19):1439–1444
Lennon S, Ashburn A, Baxter D (2006) Gait outcome following outpatient physiotherapy based on
the Bobath concept in people post stroke. Disabil Rehabil 28:13–14. Available at: http://www.
ncbi.nlm.nih.gov/pubmed/16777775
Lin C et al (2000) Common abnormal kinetic patterns of the knee in gait in spastic diplegia of
cerebral palsy. Gait and Posture 11(3):224–232
Mavroidis C et al (2011) Patient specific ankle-foot orthoses using rapid prototyping.
J Neuroengineering Rehabil 8(1):1. Available at: http://www.jneuroengrehab.com/content/8/1/1
McCaw ST, DeVita P (1995) Errors in alignment of center of pressure and foot coordinates affect
predicted lower extremity torques. J Biomech 28(8):985–988
Mirelman A et al (2010) Effects of virtual reality training on gait biomechanics of individuals post-
stroke. Gait Posture 31(4):433–437
Mulroy SJ et al (2010) Gait parameters associated with responsiveness to treadmill training with
body-weight support after stroke: an exploratory study. Phys Ther 90(2):209–223
Muro-de-la-Herran A, García-Zapirain B, Méndez-Zorrilla A (2014) Gait analysis methods: an
overview of wearable and non-wearable systems, highlighting clinical applications. Sensors 14
(2):3362–3394. Available at: http://www.mdpi.com/1424-8220/14/2/3362/. Accessed 30 Aug
2016
Novak AC et al (2015) Gait changes following botulinum toxin A treatment in stroke. Top Stroke
Rehabil 16(5):367–376
Olney SJ, Richards C (1996) Hemiparetic gait following stroke. Part I: characteristics. Gait Posture
4(2):136–148
Park CI et al (2006) Soft tissue surgery for equinus deformity in spastic hemiplegic cerebral palsy:
effects on kinematic and kinetic parameters. Yonsei Med J 47(5):657–666
Paterno MV et al (2010) Biomechanical measures during landing and postural stability predict second
anterior cruciate ligament injury after anterior cruciate ligament reconstruction and return to sport.
Am J Sports Med 38(10):1968–1978. Available at: http://eutils.ncbi.nlm.nih.gov/entrez/eutils/elink.
fcgi?dbfrom=pubmed&id=20702858&retmode=ref&cmd=prlinks\npapers3://publication/doi/10.
1177/0363546510376053
Polese JC et al (2012) The effects of walking sticks on gait kinematics and kinetics with chronic
stroke survivors. Clin Biomech 27(2):131–137
Reeves ND, Bowling FL (2011) Conservative biomechanical strategies for knee osteoarthritis. Nat
Rev Rheumatol 7(2):113–122. Available at: http://www.nature.com/doifinder/10.1038/nrrheum.
2010.212\ http://dx.doi.org/10.1038/nrrheum.2010.212
Riad J, Haglund-Akerlind Y, Miller F (2008) Power generation in children with spastic hemiplegic
cerebral palsy. Gait Posture 27(4):641–647
Richards R et al (2016) Gait retraining with real-time biofeedback to reduce knee adduction
moment: systematic review of effects and methods used. Arch Phys Med Rehabil
98:137–150. Available at: http://linkinghub.elsevier.com/retrieve/pii/S0003999316303471
Ridgewell E et al (2010) A systematic review to determine best practice reporting guidelines for
AFO interventions in studies involving children with cerebral palsy. Prosthetics Orthot Int 34
(2):129–145
Rietman JS, Postema K, Geertzen JHB (2002) Gait analysis in prosthetics: opinions, ideas and
conclusions. Prosthetics Orthot Int 26(1):50–57
Riley PO, Kerrigan DC (2005) Laboratory-based evaluation of gait disorders: high-tech. In:
Hausdorff JM, Alexander NB (eds) Gait disorders: evaluation and management. T&F, Boca
Raton. Available at: http://books.google.ch/books/about/Gait_Disorders.html?id=
m5Akhi0iJ2YC&pgis=1
Rose SA, Ounpuu S, DeLuca PA (1991) Strategies for the assessment of pediatric gait in the clinical
setting. Phys Ther 71(12):961–980
Rosenbaum P et al (2007) A report: the definition and classification of cerebral palsy April 2006.
Dev Med Child Neurol 49(109):8–14
Sadeghi H et al (2001) Functional roles of ankle and hip sagittal muscle moments in able-bodied
gait. Clin Biomech 16(8):688–695
Sawers AB, Hafner BJ (2013) Outcomes associated with the use of microprocessor-controlled
prosthetic knees among individuals with unilateral transfemoral limb loss: a systematic review.
J Prosthetics Orthot 50(3):273–314
Schwartz MH, Rozumalski A, Trost JP (2008) The effect of walking speed on the gait of typically
developing children. J Biomech 41(8):1639–1650
Seroussi RE et al (1996) Mechanical work adaptations of above-knee amputee ambulation. Arch
Phys Med Rehabil 77(11):1209–1214
Sheffler LR et al (2016) HHS Public Access 94(5):341–357
Silverman AK et al (2008) Compensatory mechanisms in below-knee amputee gait in response to
increasing steady-state walking speeds. Gait Posture 28(4):602–609
Simic M et al (2011) Gait modification strategies for altering medial knee joint load: a systematic
review. Arthritis Care Res 63(3):405–426
Sprager S, Juric MB (2015) Inertial sensor-based gait recognition: a review. Sensors 15
(9):22089–22127
Stansfield BW et al (2001) Sagittal joint kinematics, moments, and powers are predominantly
characterized by speed of progression, not age, in normal children. J Pediatr Orthop 21
(3):403–411. Available at: http://www.ncbi.nlm.nih.gov/pubmed/11371829
Stoquart GG et al (2008) Effect of botulinum toxin injection in the rectus femoris on stiff-knee gait
in people with stroke: a prospective observational study. Arch Phys Med Rehabil 89(1):56–61
Sutherland D (1997) The development of mature gait. Gait Posture 6(2):163–170
Sutherland DH (2005) The evolution of clinical gait analysis part III – kinetics and energy
assessment. Gait Posture 21(4):447–461
Tate J et al (2012) Gait analysis in musculoskeletal conditions, prosthetics and orthotics. In:
Levine D, Richards J, Whittle MW (eds) Gait Analysis. Churchill Livingstone Elsevier, London
Umberger BR, Martin PE (2007) Mechanical power and efficiency of level walking with different
stride rates. J Exp Biol 210(18):3255–3265. Available at: http://www.ncbi.nlm.nih.gov/pubmed/
17766303
van der Krogt MM, Delp SL, Schwartz MH (2012) How robust is human gait to muscle weakness?
Gait Posture 36(1):113–119. Available at: http://dx.doi.org/10.1016/j.gaitpost.2012.01.017
van der Krogt MM et al (2015) Kinetic comparison of walking on a treadmill versus over ground in
children with cerebral palsy. J Biomech 48(13):3577–3583
Vanicek N et al (2009) Gait patterns in transtibial amputee fallers vs. non-fallers: biomechanical
differences during level walking. Gait Posture 29(3):415–420
Ventura JD, Klute GK, Neptune RR (2011) The effects of prosthetic ankle dorsiflexion and energy
return on below-knee amputee leg loading. Clin Biomech 26:298–303
Versluys R et al (2009) Prosthetic feet: state-of-the-art review and the importance of mimicking
human ankle-foot biomechanics. Disabil Rehabil Assist Technol 4(2):65–75
Whittle M (1996) Clinical gait analysis: a review. Hum Mov Sci 15(3):369–387. Available at: http://
linkinghub.elsevier.com/retrieve/pii/0167945796000061
Winter DA, Sienko SE (1988) Biomechanics of below-knee amputee gait. J Biomech 21
(5):361–367
Woolley SM (2001) Characteristics of gait in hemiplegia. Top Stroke Rehabil 7(4):1–18
EMG Activity in Gait: The Influence of Motor
Disorders
Dimitrios A. Patikas
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
EMG Patterns and Activation Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Interpretation of EMG Findings Linked to the Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Power Generation Capacity and EMG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Spasticity and the Role of Stretch Reflex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Agonist/Antagonist Co-activation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Abstract
EMG techniques have been widely used in research and clinical settings, reveal-
ing underlying mechanisms that are involved in pathology and assisting clinicians
to decision-making. Current findings based on EMG, combined with kinematic
and kinetic data during gait, describe very specific activation patterns and muscle
synergies in people with and without motor disorders, which contribute to the
understanding of the disease’s causes and to build more accurate and detailed
simulation models. Furthermore, the EMG findings during stance and swing
phase are discussed and more particularly the co-activation of the agonist and
antagonist muscles. This helps to comprehend the contribution of certain mus-
cles/muscle groups when power generation or stability is required and the effect
of spasticity and stretch reflex on the movement quality. This chapter also
supplies the reader with several examples that reveal, by means of EMG
D.A. Patikas (*)

School of Physical Education and Sport Science, Aristotle University of Thessaloniki, Thessaloniki,
Greece
e-mail: dpatikas@auth.gr

DOI 10.1007/978-3-319-30808-1_33-1
2 D.A. Patikas
assessments, mechanisms of the neuromuscular system to compensate primary or

secondary causes of movement disorders. Finally, the limitations and difficulties
in EMG interpretation in pathological cases are discussed.
Keywords
Electromyography • Gait • Cerebral palsy • Stroke • Spinal cord • Brain • Injury •
Parkinson’s disease • Co-activation • Compensation • Spasticity
List of Abbreviations
CP Cerebral palsy
PD Parkinson’s disease
TD Typically developed
Introduction
Gait is a fundamental motor skill that humans learn during the early age of their life.
A person’s ability to progress his/her body to any direction with no limitations is
very important and determines to a great extent the individual’s quality of life.
During gait, the objective of the neuromuscular system is to achieve body-weight
support, dynamic stability, forward body progression, and foot clearance. This is
accomplished by a complex system of multiple degrees of freedom that involve
numerous muscles, which act on a nonlinear fashion, when adapting to the environ-
mental demands. Hence, understanding human gait is a complex issue that requires
detailed investigation of all involved systems individually and all together, regarding
their interaction under different circumstances.
Motor disorders are responsible for several deviations in movement and more
particularly in gait. Spasticity, hypertonia, tremor, muscular weakness, ataxia,
apraxia, and deficits in coordination are some of the most common symptoms that
may affect the quality of the movement. Research regarding gait in patients with
motor disorders is essential not only to understand the nature, causes, and mecha-
nisms of the disease but also to develop improved rehabilitation strategies. Identifi-
cation of relationships between abnormal neuromuscular activation and deficits in
motor responses is an important step to develop new therapies aiming to “fix” this
abnormal activation. Research concerning the neuromuscular response of patients
with motor disabilities during gait may help health professionals identify the causes
of the disability and better plan their treatment accordingly.
During gait, the central nervous system activates a series of muscles which
produce forces that exert moments at the joints. The net amount of moments at a
joint produces the necessary joint movement to support the body against gravity and
move the body or body limbs to the required direction. Furthermore, depending on
the direction of the movement, joint moments can produce or absorb energy (power).
In general, movement can be described by the joint kinematics and the net moments
and power by the joint kinetics. The former focuses on the final result of the
movement, whereas the latter is closer to the cause that generates movement,
EMG Activity in Gait: The Influence of Motor Disorders 3
namely, the joint moments and power. Both kinematics and kinetics are essential for
gait analysis. However, since kinetic analysis comprises the net moment that is
exerted at a joint, it is not possible to derive conclusions about which muscles or
muscle groups actually contribute to the movement. For example, increase in knee
extension moment could be achieved by an increased force exerted from the knee
extensors or by a reduced force exerted from the knee flexors or both. Being able to
identify which of these possibilities holds true is very important, especially when
analyzing pathologic gait.
Electromyography (EMG) is a very useful method giving an insight into the
activation patterns of the muscles which helps to understand how the movement is
actually generated. This is surely important when trying to identify the causes of a
motor disease. With the use of surface or intramuscular electrodes, it is possible to
record the action potentials that the muscle produces during activation. More
specifically, EMG gives qualitative and quantitative information about muscle
activation and offers an objective measure about whether a muscle is active or not
and about when or how much a muscle is active under different circumstances. In
contrast with the kinematics (referring to joint’s angle) and kinetics (referring to
joint’s moment/power) which supply all necessary information about the net out-
come of the movement, EMG gives an insight into the activation of the muscles or
the muscle groups that are one step closer to the source of the problem, provided the
disease has neural origin. This may help us understand the reasons for the abnormal
movement. For instance, when recording a flexion joint moment with no
(or reduced) concomitant EMG activity of the flexor muscles, we may attribute
this flexion moment to passive component such as ligaments, joint capsules, and
tendons (Õunpuu et al. 1996). On the other hand, EMG by its own does not provide
any kind of information regarding the type of muscle contraction (concentric,
isometric, or eccentric), or the amount of force that the muscle exerts, or the torque
at the joint that the muscle produces. Therefore, its implementation is valuable
when applied with simultaneous recordings of kinematics and kinetics (Whittle
1996).
Since the early 1990s, EMG is frequently used as a research tool to understand the
underlying mechanisms of pathologic gait and, as a diagnostic means used by
clinicians, to document the presence, status, and progression of a disease. Assess-
ment of the abnormal muscle activation can also help to predict and decide, for
instance, about which muscle(s) may benefit from a botulinum toxin injection (Corry
et al. 1999; Manganotti et al. 2007; Metaxiotis et al. 2002; Novak et al. 2013;
Sutherland et al. 1999) or if a muscle transfer surgery will improve gait (Perry
et al. 1974; Miller et al. 1997). Furthermore, EMG can be used in the field of
rehabilitation to improve the clinical status of individual patients, by offering
feedback to them or to external assisting devices (motors, muscle/nerve stimulators,
etc.), which, in turn, move/activate the human body accordingly. For example, it has
been previously demonstrated that processing multiple EMG bursts from two upper
limb muscles (anterior and posterior deltoid muscles) during gait could successfully
reproduce the lower limb movement, in order to produce a more regular gait pattern
(Cheron et al. 2007). This occurs due to the existence of propriospinal neurons
4 D.A. Patikas
coupling the activation of cervical and lumbar motoneurons (Nathan et al. 1996).
During walking, motor commands give reciprocal motor input to the anterior and
posterior deltoid muscles and synchronous activation to the soleus and tibialis
anterior muscles, respectively (Cheron et al. 2007).
Cerebral palsy (CP), stroke, and traumatic brain or spinal cord injury and
Parkinson’s disease (PD) are the most common motor disorders that may influence
gait. Other medical conditions that may impede gait are multiple sclerosis, rheu-
matoid arthritis, osteoarthritis, muscular dystrophy, and lower limb amputation.
The purpose of this chapter is to help the reader understand the implementation of
the EMG technique on pathologies comprising motor disabilities that affect gait,
focusing mainly on the most thoroughly investigated disorders during gait, CP and
stroke, and their cardinal symptom, spasticity. More specifically, the objectives of
this chapter are (a) to focus on characteristic gait patterns and activation modules
during gait; (b) to describe the adaptations that occur in pathologic gait, taking into
account neurological and biomechanical restrictions that may occur due to the
disorder; and (c) to comment with a series of examples and experimental setups on
the importance of EMG measurements to identify compensatory mechanisms.
State of the Art
EMG is a useful tool that with careful and cautious interpretation can give important
information to understand the nature of a motor disease. Taking for granted that each
patient has a unique motor disease with some commonalities, careful clinical
approaches using – among others – EMG measurements may assist in individually
tailored treatment. Nowadays, thanks to EMG techniques, we are able to identify
specific gait patterns and understand better how movement is organized at spinal and
supraspinal level in healthy and diseased populations. More specifically, EMG
assessment gives an insight on deviations in muscle activation that might occur
and the ability of the neuromuscular system to adapt under different environmental
and pathological circumstances. Moreover, EMG can document objectively the
effectiveness of a treatment during a functional task such as gait (Blanchette
et al. 2014). Additionally, EMG techniques combined with electrical or magnetic
stimulation on peripheral, spinal, or supraspinal level supply valuable information,
noninvasively, about the function of the neuromuscular system in healthy and
diseased population.
In rehabilitation, EMG can be used as an auditory or visual biofeedback tool to
achieve a more relaxed or excited state of a specific muscle or muscle group (Dursun
et al. 2004). EMG activity can also drive external devices to produce a requested
movement, setting the proper timing and amplitude (Koller et al. 2015). Further-
more, EMG output of functionally intact muscles can be used to stimulate other
muscles or muscles groups that are not possible to be sufficiently enough recruited
by the central nervous system (Cheron et al. 2007).
EMG Patterns and Activation Modules
Despite the fact that the clinical features of CP are highly variable among patients,
many researchers attempted to assign specific gait patters to group of patients, in
order to facilitate decision-making and help for more efficient communication
among researchers. There are many gait classifications available in the literature
that assigned patients to different numbers of groups (from two to ten) using different
methods (e.g., visual inspection, cluster analysis, principal components analysis,
etc.) and data input sources originating mainly from kinematics, kinetics, ground
reaction forces, spatiotemporal parameters, and, less frequently from EMG, energy
cost variables (Dobson et al. 2007).
Patients with stroke, who were classified by visual inspection according to their
kinematics, revealed no significant difference between the classification groups in
the EMG parameters (De Quervain et al. 1996). However, using more objective
classification techniques, such as cluster analysis, and a larger number of patients
(n = 47 instead of n = 18 in the former study), significant differences in EMG
between the groups were found after classifying the patients according to their
spatiotemporal and kinematic parameters (Mulroy et al. 2003). In general, the
most impaired patients demonstrated decreased EMG for the soleus, tibialis anterior,
hamstrings, and quadriceps muscle in terms of amplitude and duration (Mulroy
et al. 2003).
Knutsson and Richards (1979) were the first who attempted to classify with visual
inspection gait patterns taking into consideration the EMG from the lower limb
muscles. They identified three types of activation patterns, and 4 out 26 patients
could not be classified. Type I (9/26) demonstrated premature plantar flexor EMG
and low dorsiflexor EMG, type II (9/26) showed low EMG in two or more muscles,
and type III (4/26) presented increased level of co-activation.
In order to decrypt the muscle activation patterns during walking in a simplistic
way, we could describe this as a series of organized bursts at specific moments of the
gait cycle (Ivanenko et al. 2013; Winter 1989; Lacquaniti et al. 2012; Zajac
et al. 2003). Hof et al. (2002) identified six basic EMG patterns to describe the
EMG activity of 14 muscles during gait, and Lacquaniti et al. (2012) used one EMG
burst at three different instances of the gait cycle to create five basic muscle
activation patterns that are observed during gait. Considering the above, although
many muscles participate in gait, ultimately only a few patterns are detected. This
advocates the existence of muscle synergies that take place during gait. However, as
described in the following paragraphs, this redundancy, i.e., the existence of many
muscles with less EMG patterns, might be an essential element for gait adaptability
under different circumstances and, more interestingly, in the presence of a motor
disorder (Winter 1989; Cai et al. 2006; Duysens et al. 2013).
According to findings regarding EMG measurements on the lower limbs during
gait, the following four major activation patterns have been identified in typically
developed (TD) individuals (Ivanenko et al. 2013; Lacquaniti et al. 2012): (a) Hip
and knee extensors are active after heel contact and contribute to body-weight
acceptance, (b) plantar flexors provide body support during the stance phase and
6 D.A. Patikas
contribute to forward propulsion during the late stance, (c) during swing phase the
activation of the hip flexors and dorsiflexors contributes to foot clearance, and
(d) during late swing the foot is prepared for placement on the ground by activating
the hamstrings, which decelerate the leg and stabilize the pelvis. It is important to
note though that these activation patterns do not change in a linear manner when
external (or internal) environmental requirements change. For example, the tibialis
anterior and rectus femoris do not increase their EMG amplitude linearly when
walking gradually faster, as other muscles do (Hof et al. 2002), and the soleus,
gastrocnemius, and tibialis anterior do not change in a similar level their activation
level at different body loadings (Dietz and Colombo 1998; McGowan et al. 2010).
The above studies suggest, on the one hand, the existence of a central pattern
generator during gait, but on the other hand, the manipulation of the final outcome,
when environmental requirements change, is not linear, which makes the prediction
of the outcome a rather complex issue. These facts make very difficult the compar-
isons between pathologic populations and healthy controls, in order to deduce an
abnormal activation pattern, because any such comparison should be assessed when
walking under the same conditions. Only then it will be possible to reveal the
potential existence of the pathological mechanisms that affect the internal neuro-
muscular and musculoskeletal properties of the patient during gait.
A partial recovery over the course of the first 6 months after a stroke is well
established. During this recovery period, it is possible to examine with EMG
measurements if new gait patterns develop or if the original ones are basically
retained and partially improved. In other words, EMG data let us understand if
neuromuscular control remains unchanged during recovery and/or if some neuro-
muscular mechanisms are susceptible to change after a treatment. According to Den
Otter et al. (2006), patients with stroke that showed a significant improvements in
body mobility, ambulatory independence, and maximum gait speed had no signifi-
cant changes in EMG, indicating that the restoration of gait was not attributed to a
reorganization of the motor control, as assessed by means of EMG.
The identification of activation modules in pathological cases gives an insight
into how the nervous system is reorganized due to the motor disease. For instance,
during gait, patients with stroke undergo muscle weakness and changes in the
temporal organization of their muscle activation. Despite of this, Gizzi
et al. (2011) identified four EMG patterns in patients with stroke and in TD subjects
and observed that they were very similar between the groups. However, the
weighting of each pattern on each of the examined muscles (n = 16) was different
between the groups, suggesting the existence of a pattern generator that, under
central and sensory control, delivers these activation patterns to the specific muscles
depending on the task’s requirements. Nevertheless, other studies that examined the
activation patterns of smaller number of muscles (n = 7, lower limb only) support
that some patients with stroke use fewer activation modules than the four modules
detected in healthy individuals (Clark et al. 2010). This discrepancy could be due to
the lower number of the examined muscles in the later study.
In general, these findings from patients with stroke and healthy subjects suggest
that there are neural brainstem and/or spinal networks that are regulated by
descending cortical commands, which determine which muscle synergies will take
place. Therefore, if we accept that there is a modular organization in the EMG
patterns of both patients and healthy individuals, the rehabilitation process should be
targeted on delivering the correct activation pattern to the right muscle, i.e.,
correcting the abnormal muscle synergies. For example, using functional electrical
stimulation and more specifically delivering trains of electrical pulses, at the appro-
priate instances during the gait cycle, will not only lead to advanced muscle strength,
joint range of motion, and gait pattern (Popovic et al. 2009) but may also deliver
additional sensory feedback that facilitates motor learning (Yan et al. 2005; Tong
et al. 2006; Popovic et al. 2009; Solopova et al. 2011).
Another interesting issue regarding muscle synergies that take place in different
pathologies regards the synchronous activation of proximal and distal muscles
during the late stance. In patients with stroke, gait is often characterized by decreased
knee flexion of the paretic side during swing (Kerrigan et al. 1991; Mulroy
et al. 2003), which is accompanied by prolonged quadriceps EMG (Kautz and
Brown 1998; Mulroy et al. 2003). In general, patients with hemiplegia (Knutsson
and Richards 1979), unilateral sciatic nerve compression (Ivanenko et al. 2013),
below-knee amputation (Winter and Sienko 1988), and incomplete spinal cord injury
(Beres-Jones and Harkema 2004) demonstrate increased activity in proximal mus-
cles (rectus femoris, vastus medialis, vastus lateralis, gluteus medius, sartorius,
tensor fasciae latae) during late stance. This behavior has been also observed TD
subjects with unilateral ischemic block of distal leg muscles (Dickey and Winter
1992). In TD subjects, the proximal muscles are not active during late stance when
walking at normal speed. Instead, activation of more distal muscles (i.e., plantar
flexors) is typical, which contributes to the power generation during push-off
(Cappellini et al. 2006; Nilsson et al. 1985; Winter 1989; Nene et al. 2004; Prilutsky
and Gregor 2001). Interestingly, a similar behavior with the one observed in patients,
i.e., synchronous burst of activation of the proximal and distal lower limb muscles
during the late stance, has been demonstrated in TD individuals as well, when
increasing walking speed (Ivanenko et al. 2008). As explanation, it has been
suggested that when increasing walking speed, the higher external demand in
force production from the distal muscles is not fully attained, because, at higher
contraction velocities, muscles have lower force capacity (Neptune et al. 2009).
Therefore, additional (and more proximal) extensors are recruited to compensate this
deficit. Similarly, this type of co-activation of distal and proximal lower limb
muscles that is observed in patients could be a compensatory mechanism for the
lower force/power capacity of the distal extensors, independently of the pathology
causing it (Perry et al. 1995; Sutherland et al. 1980; Murray et al. 1978). However,
this extra activation of the proximal extensor muscles is not necessarily depicted in
increased extensor moment possibly due to the existence of a more complex
dynamic coupling across joints (Zajac et al. 2003) and the potential influence of
other muscles (gluteus medius, sartorius, tensor fasciae latae) coacting synergisti-
cally on other planes (Ivanenko et al. 2013). Furthermore, patients with stroke that
demonstrated increased quadriceps EMG during late stance retained their EMG
magnitude even when walking with body-weight support up to 40 % of body weight
8 D.A. Patikas
(Murray et al. 2014). These last findings challenge the hypothesis that the activation
of proximal muscles compensate the weakness of the distal ones, and therefore,
further research is required to clarify this issue.
Interpretation of EMG Findings Linked to the Pathology
The following paragraphs describe some EMG findings that help us understand the
way that the neuromuscular system adapts in the presence of a motor disorder. More
specifically, three basic components of gait are analyzed based on EMG findings:
(a) the muscle recruitment especially during the push-off phase which is important
for the forward progression of the body; (b) the activation of muscles especially
during the swing phase, underpinning the role of stretch reflex in patients with
spasticity; and (c) the level of agonist/antagonist co-activation and its functional
significance.
Power Generation Capacity and EMG
One of the most critical issues, which contributes at most to the forward progression
of the body during the gait cycle, is the efficiency during the push-off phase and
more specifically the power generation at the ankle joint. The push-off phase has two
major components: the passive, attributed to the energy storage/release from the
elastic components, and the active, attributed to the muscle activation (Hof
et al. 1983). Although the contribution of the passive musculotendinous components
in the production of the plantar flexion moment during push-off may be three to four
times higher in children with CP (Tardieu et al. 1989) and hemiparetic adults (Siegler
et al. 1984; Lamontagne et al. 2000a), the resulted plantar flexor moment is signif-
icantly lower in patients with CP (Olney et al. 1990) and stroke (Olney et al. 1991;
Olney and Richards 1996).
Plantar flexors are crucial in regulating walking speed (Brunner and Romkes
2008; Nadeau et al. 1999) and play an important role in body-weight support and
forward propulsion by applying vertical and shear (anterior–posterior) forces on the
ground during push-off. Weakness in plantar flexor muscles results in activation of
other antigravity muscles in order to sustain body weight (Murray et al. 1978) and to
maintain postural control. Moreover, at higher walking speed, hip flexors are more
actively recruited to assist the limb to move forward during the swing phase (Nadeau
et al. 1999).
The weakness during push-off observed in CP and stroke patients has been
attributed to the lower plantar flexor muscle activation (Lamontagne et al. 2002;
Mulroy et al. 2003) and/or higher level of plantar/dorsiflexion co-activation (Shiavi
et al. 1987b). More particularly, it has been shown that the activation of the paretic
medial gastrocnemius is correlated to the walking speed of patients with stroke
(Lamontagne et al. 2002). Additionally, patients with motor disorders often exhibit
symptoms of spasticity and thus hyperactive stretch reflexes. This could be an
advantage for patients with triceps surae spasticity especially during mid-stance
when the tibia progresses over the foot and the triceps surae is stretched. However,
although 30–60 % of the plantar flexor activation in the early stance is attributed to
the stretch reflex (Yang et al. 1991b), concurrent symptoms, such as clonus (Whittle
1996), interfere with the possibility to lengthen the plantar flexor muscles and to
initiate the stretch reflex (Knutsson and Richards 1979).
Another factor that may contribute to the reduced force generation during push-
off is the sensory feedback from foot receptors, because the adaptive activation of the
soleus muscle during perturbations when walking is dependent on the quality of
sensory feedback (Mazzaro et al. 2005). As a consequence, patients with lower
capacity in sensory feedback (e.g., peripheral neuropathy) do not only have weaker
plantar flexor muscles but also demonstrate deficits in static balance (Nardone and
Schieppati 2006), which could result in a decreased gait speed. Reding and Potes
(1988) showed in patients with stroke that sensory deficits are linked to lower
walking scores. In addition to this, when visual deficits were added to the sensory
deficits, walking scores became even lower, suggesting that the probability for a
patient to walk independently is related to the occurrence of visual and sensory
disabilities on top of the motor deficit.
Finally, the reduced EMG could be also linked to biomechanical restrictions and,
possibly, to the feedback received from muscle, tendon, and joint receptors. In
patients with CP and knee recurvatum, a condition that the knee is hyperextended
during mid- and late stance, knee flexors demonstrate decreased EMG, despite the
presence of an internal knee flexion moment (Lin et al. 2000). This indicates that the
flexion moment possibly originates from passive components of soft tissue, liga-
ments, and the joint capsule while the knee is hyperextended, and overuse of these
structures in the absence of (or with reduced) muscular support could increase the
risk of injury (Õunpuu et al. 1996).
Spasticity and the Role of Stretch Reflex
During swing phase, dorsiflexion occurs for foot clearance, and during late swing,
the knee is extended. This movement lengthens the plantar flexor muscles and
hamstrings, respectively. In patients with stroke, limited dorsiflexion is observed
during swing phase (Lamontagne et al. 2002). It has been suggested that spasticity of
the lateral gastrocnemius and medial hamstrings in children with CP contributes to
the limited ankle dorsiflexion and knee extension, respectively, during terminal
swing (Bar-On et al. 2014), which, in turn, results in decreased walking speed and
step length (Delp et al. 1996). Moreover, lower lengthening velocity, reflex threshold
for the hamstrings, is related to a lower knee angular velocity during gait (Bar-On
et al. 2014; Damiano et al. 2006; Tuzson et al. 2003). Especially at higher gait
speeds, although the stretch reflex response of the medial hamstrings and lateral
gastrocnemius during passive stretch is correlated with their activity during swing
phase (Bar-On et al. 2014), the contribution of the stretch reflex to the reduction in
knee and ankle joint angular velocity, respectively, is limited (Bar-On et al. 2014).
10 D.A. Patikas
Furthermore, the increased hamstring activation during swing could have a func-
tional importance, by flexing the knee and shortening the limb length for foot
clearance, when dorsiflexion is limited (Patikas et al. 2005).
On the other hand, the reduced dorsiflexion during swing phase can be attributed
to a lower activation of the dorsiflexors (Richards et al. 1992), an increased passive
stiffness of the plantar flexors (Dietz et al. 1981), and a higher EMG activation of the
plantar flexors when increasing their lengthening velocity (Crenna 1998). However,
it is well recognized that the level of spasticity is different when measured at rest or
during a dynamic movement, such as gait (Fung and Barbeau 1994; Knutsson and
Richards 1979; Lamontagne et al. 2001). Although the stretch reflex of patients is
increased at rest, many authors conclude that during gait its contribution to an
increased EMG activity is limited (Willerslev-Olsen et al. 2014; Bar-On
et al. 2014; Marsden et al. 2012; Bohannon and Andrews 1995; Lamontagne
et al. 2002). Similarly, in patients with stroke, the exaggerated reflex gain recorded
in resting conditions was much lower during an active movement simulating walk-
ing, revealing that the reflex gain is not upregulated during active movements as
expected, but instead, it is actually not (or less) regulated (Ada et al. 1998). In
general, children with CP (Crenna 1998) and patients with stroke (Yang et al. 1991a)
retain their ability to depress the stretch and H-reflex (the electrically induced stretch
reflex) during the swing phase, as it occurs in healthy subjects (Simonsen 2014).
Therefore, it is more likely that the main contributors for the decreased dorsiflexion
during the swing phase could rather be the passive components and the decreased
central drive to the dorsiflexors and not an increased stretch reflex response from the
plantar flexors (Willerslev-Olsen et al. 2014).
Agonist/Antagonist Co-activation
Agonist/antagonist co-activation is defined as the simultaneous activation of the

agonists and antagonist muscles that cross the same joint (Busse et al. 2005) and can
be assessed by recording the EMG activity of the respective muscle groups (Ikeda
et al. 1998). It reduces the net moment that is exerted at the joint (Richards 1985) and
increases joint stiffness (Baratta et al. 1988; Osternig et al. 1995; Solomonow
et al. 1988). Although agonist/antagonist co-activation seems counterproductive,
its functional importance is associated to an improved joint stability (Levin and
Dimov 1997), better pressure distribution over the joint surface, and increased
accuracy in precision tasks (Damiano 1993; Johansson and Westling 1988; Valero-
Cuevas 2005; Higginson et al. 2006b). Depending on the task, there is an optimal
level of agonist/antagonist co-activation (Levin and Dimov 1997; Lacquaniti and
Maioli 1987, 1989), and beyond or below this level, the movement is impaired or
there is a higher risk for injury. In general agonist/antagonist co-activation during
walking is an adaptive property of the neuromuscular system, depending on the age,
pathology, and external requirements of the movement (speed, cadence, walking
surface, etc., Parvataneni et al. 2009; Knarr et al. 2012), in order to provide greater
stability at the joint (Schmitz et al. 2009; Peterson and Martin 2010).
In TD population during gait, the agonist/antagonist co-activation is higher

during loading response (Falconer and Winter 1985; Winter and Yack 1987) and
during the transition from stance to swing but is lower during mid-stance phase
(Nadeau et al. 1996). Furthermore, TD individuals increase the agonist/antagonist
co-activation during walking when gait speed increases (Frost et al. 1997; Unnithan
et al. 1996a; Hortobágyi et al. 2009; Schmitz et al. 2009) and when stability is
challenged (Layne et al. 1997).
According to a recent systematic review (Rosa et al. 2014), there is no concrete
consensus about the role of the agonist/antagonist co-activation in motor disorders. It
has been reported that people with motor disorders have abnormal level of
co-activation (Berger et al. 1982; Unnithan et al. 1996a; Romkes et al. 2006;
Lamontagne et al. 2000b). More specifically, this has been observed in patients
with stroke (Den Otter et al. 2006), especially in postural challenging conditions
(Lamontagne et al. 2000b) and in children with CP, especially for the soleus/tibialis
anterior pair (Berger et al. 1982; Unnithan et al. 1996a; Romkes et al. 2006).
Furthermore, increased level of co-activation has been reported in patients with
PD when compared to age-matched healthy individuals (Dietz et al. 1995). How-
ever, in contrast to young adults, elderly people with and without PD demonstrate
decreased agonist/antagonist co-activation, when walking speed increases (Dietz
et al. 1995).
In hemiplegic stroke, only some patients (4 out of 26 patients) may experience
increased level of co-activation among multiple muscles during gait on the paretic
side (Knutsson and Richards 1979). This reduces the net torque outcome and
probably has a negative functional impact. However, later studies that examined
the medial gastrocnemius/tibialis anterior co-activation during gait showed that the
paretic side has lower co-activation than the non-paretic side or the sides of
age-matched control counterparts, who walked at a similar gait speed (Lamontagne
et al. 2000b, 2002). Moreover, despite this lower level of co-activation on the
involved side, there was a positive correlation between the level of co-activation
during the single support phase of the involved side and the gait speed, the peak
plantar flexion moment, and the postural stability assessed by the duration of the
single support phase as a percentage of the gait cycle (Lamontagne et al. 2000b).
However, since correlation does not imply causation, these findings could either
advocate that the increased co-activation might have beneficial effects in gait
performance and stability or that the most severely affected patients (as reflected
by gait performance and stability) demonstrate less co-activation.
Interestingly, the EMG activity of lower limb muscles in general (Hirschberg and
Nathanson 1952; Marks and Hirschberg 1958; Wortis et al. 1951; Shiavi
et al. 1987b) and the medial gastrocnemius/tibialis anterior co-activation in partic-
ular (Lamontagne et al. 2000b, 2002) is higher on the sound than the involved side of
hemiparetic patients with stroke. Furthermore, this increased co-activation has been
related to the reduced plantar flexor moment during stance phase (Lamontagne
et al. 2002). The excessive level of co-activation reduces the efficiency of the
movement and may explain, at least partially, the increased energy cost during
walking in patients with stroke (Corcoran et al. 1970; Gersten and Orr 1971;
12 D.A. Patikas
Zamparo et al. 1995) and CP (Unnithan et al. 1996b). The higher co-activation of the
sound side that has been observed during double support phase probably compen-
sates the inability of the involved side to stabilize the body (Lamontagne
et al. 2000b), if we assume that the sound side is neurologically intact. The latter
assumption may not always hold true, since a lesion on the one side of the brain may
have impact on the function of the ipsilateral upper (Jones et al. 1989; Desrosiers
et al. 1996) or lower (Adams et al. 1990) limb as well. However, it is quite unlikely
that the observed increment in co-activation on the sound side could be attributed
solely to the effects of the ipsilateral lesion.
Co-activation and Gait Speed

The higher activation of the antagonist muscles may contribute to a lower torque
production, limitation of the range of motion, gait speed reduction, and increase in
energy expenditure. In fact, the level of the biceps femoris/rectus femoris
co-activation has shown a significant correlation with the energy expenditure during
walking in children with CP (Damiano et al. 2000). Despite this negative effect of
agonist/antagonist co-activation, its functional significance might be compensatory,
to increase joint stability, anticipate external perturbations, and limit the number of
degrees of freedom. This is supported by the increased co-activation of the soleus
and tibialis anterior in children with no neurological impairment that walked with an
orthosis at ten or more degrees of plantar flexor position (Houx et al. 2014). Fur-
thermore, it is worth noting that the level of agonist/antagonist co-activation is not
correlated with the Gross Motor Function Measure (GMFM) score in children with
CP (Damiano et al. 2000) indicating that the level of co-activation has no direct link
to functional tasks.
In TD children, the activation time of the shank muscles is longer when walking
speed increases, and therefore the overlap of activation between agonists and
antagonists (co-activation) increases (Detrembleur et al. 1997). This increase in the
co-activation contributes to a greater joint stability that is required at higher speeds
(Frost et al. 1997; Unnithan et al. 1996a), when the exerted forces from the
environment to the body are higher.
A similar behavior has been observed in children with CP (Damiano et al. 2000;
Unnithan et al. 1996a), which supports the protective role of co-activation at higher
gait speeds. In a more recent study involving patients with CP, the involved side
showed actually an even greater sensitivity to co-activation increase in faster gait
speeds (Gross et al. 2013). The authors attributed this increase to the possibly
increased stretch reflex response at higher muscle-lengthening velocities induced
by the higher speed. This indicates that increased co-activation (and thus decreased
ability to exert forces at the respective joint) might be a contributor to the decreased
gait speed that is observed in children with CP (Fonseca et al. 2001). Furthermore, in
TD children, the co-activation of the rectus femoris/semitendinosus is influenced less
by gait speed than the co-activation of the soleus/tibialis anterior and the vastus
medialis/semitendinosus muscles (Gross et al. 2013). Typically, rectus femoris and
semitendinosus muscle activation have different timing (Nene et al. 2004), and
probably this is the reason for the lower increase in co-activation when increasing
gait speed in TD subjects. The fact that children with CP do increase the
co-activation level of this agonist/antagonist muscle pair suggests that the upper
motor lesion causes limited selective control for the activation of some lower limb
muscles.
Regarding the patients with PD, the level of agonist/antagonist co-activation is
increased compared to age-matched healthy subjects walking on the same speed
(Dietz et al. 1995). This could offer greater stabilization at the joint to the patients
with PD, whose postural stability is typically impaired. However, when increasing
gait speed, the level of co-activation does not increase as it does in healthy young
people (Schmitz et al. 2009), but decreases (Dietz et al. 1995). This inability to adapt
to increasing gait speed may increase the risk of falling and could be attributed to the
defective function of proprioceptive reflexes that may contribute to an impaired
activation of the extensor leg muscles (Dietz et al. 1988). However, a more recent
study has underlined that especially for people with PD, controlling for gait speed
only increases variability between and within groups, since it is possible to achieve
the same gait speed with a wide range of step lengths and cadences (Arias
et al. 2012). Therefore, further studies are required to examine in more detail the
level of co-activation in patients with CP.
Description of Compensatory Mechanisms

Differences in the EMG activity between patients and TD subjects during gait could
be attributed to primary or secondary causes, and it is often hard to distinguish
between these two (Gage and Schwartz 2009). The former regard the abnormal
neuromuscular activation due to lesions in brain structures, whereas the latter regards
compensatory activation that contributes to a potentially better motor outcome under
the restrictions set by the disorder. For example, the early activation of the triceps
surae during stance phase (Whittle 1996) could be a result of an abnormal central
activation due to the cerebral lesion (primary cause) or could be an adaptation to a
fixed equinus (secondary cause) (Massion 1992). In other words, EMG recordings
may give evidence about when a muscle is active, but further investigation is
required to understand why this muscle is active and if this activation serves a
purpose or if it is a result of the lesion. Failure to distinguish whether a cause is
primary or secondary may lead to a wrong decision in treatment, with adverse
complications. Ideally, primary causes should be identified, and treatment should
be focused on them solely (Davids et al. 1999).
One approach that helps in making this distinction is to mimic pathologic gait
patterns with TD individuals and to record the kinematic, kinetic, and EMG
responses. In a recent study, Houx et al. (2014) examined TD children walking
with an orthosis fixed at various ankle angles and evaluated the lower limb muscle
activity and the co-activation index of agonist/antagonist muscle pairs. They
reported that at 10 plantar flexion soleus was prematurely activated, tibialis anterior
EMG was decreased, tibialis anterior/soleus co-activation increased, and hamstrings,
vastus lateralis, and rectus femoris EMG increased. Similar findings have been
reported in healthy subjects that simulated unilateral toe walking (Romkes and
Brunner 2007) and bilateral toe walking (Perry et al. 2003) as well. Interestingly,
14 D.A. Patikas
such features of EMG activation patterns during gait are also evident in patients with
CP and unilateral equinus (Romkes et al. 2006; Winters et al. 1987; Davids
et al. 1999; Romkes and Brunner 2007; Perry et al. 1974; Brunt and Scarborough
1988; Higginson et al. 2006a) or bilateral equinus (Romkes et al. 2006). It is also
very characteristic that in earlier studies, it was not possible for Davids et al. (1999)
to distinguish between TD children walking on their toes and children with CP and
equinus when comparing their kinematic, kinetic, and EMG data. Furthermore,
idiopathic toe walking is characterized by a premature activation of the gastrocne-
mius muscle at terminal swing phase, an abnormal EMG pattern for the tibialis
anterior, and an increased co-activation of the gastrocnemius and tibialis anterior
muscles (Griffin et al. 1977; Kalen et al. 1986). However, it is hard to distinguish
between idiopathic toe walking and equinus in CP when comparing EMG data
(Kalen et al. 1986; Rose et al. 1999). Only minor differences in the timing of the
gastrocnemius EMG activity have been detected between CP and idiopathic toe
walking (Rose et al. 1999), although the presence of some clinical signs (e.g.,
clonus) that can be captured with an EMG assessment (Papariello and Skinner
1985), or manual tests (Rose et al. 1999), might be necessary to distinguish between
the two possible diagnoses.
These commonalities between patients and TD people mimicking pathological
conditions or patients with idiopathic toe walking provide some indications that
these common symptoms might have secondary – musculoskeletal (due to fixed
equinus) and not primary – neural (due to brain lesion) origin. More particularly, the
early activation of the plantar flexors observed in patients might be an action
anticipating the contact of the toe with the ground and not necessarily a sign of
spasticity (Massion 1992). This, however, might have further consequences. The
early activation of the medial gastrocnemius muscle during terminal swing, in TD
toe walkers and in children with CP and equinus (Romkes and Brunner 2007), could
increase the knee flexion angle during foot contact, because this muscle is biarticular
and its activation induces not only plantar flexion but also knee flexion (Hullin
et al. 1996).
Another finding that reveals a compensatory mechanism by means of EMG
assessments regards the activation of the tibialis anterior muscle. In normal gait,
the tibialis anterior is active during the loading response to control foot placement
and during swing phase to prohibit drop foot (Perry 1992). Decreased activity of the
tibialis anterior is frequently shown in children with CP and is referred as dorsiflexor
“weakness,” which may result in equinus during swing phase, dragging of the foot,
tripping, and compensatory flexion of the hip and/or the knee for better foot
clearance (Aiona and Sussman 2004). This is also a typical symptom in 60–80 %
of ambulatory stroke survivors (Jorgensen et al. 1995).
However, TD children mimicking unilateral toe walking and TD children with
fixed orthosis at plantar-flexed position also demonstrate reduction in tibialis anterior
EMG during loading response (Romkes and Brunner 2007) and swing (Houx
et al. 2014) phase, respectively. Therefore, this decrease in tibialis anterior EMG
might not be due to reduced central drive, but due to the fixed plantar-flexed position
of the foot which dictates that any activation of the tibialis anterior during swing
would have no effect, because the foot is fixed (Houx et al. 2014), or would be
counterproductive to the eccentric activation of the plantar flexors during toe contact
(Romkes and Brunner 2007). Interestingly, in TD children, the increased activity of
the tibialis anterior during swing phase is retained when the foot is fixed at plantar-
flexed position with orthosis (Houx et al. 2014), but it remains unchanged when the
subject mimics toe walking voluntarily (Romkes and Brunner 2007), showing the
effect that a physical/external constraint (limitation of the range of motion) might
have. For these reasons, the activation of the dorsi- and plantar flexors should be
interpreted with caution prior decision-making, because weakening the plantar
flexors by any means (surgical lengthening, selective rhizotomy, botulinum toxin
injection, etc.) may seem corrective and intuitive, targeted to converge EMG closer
to normative data, but it could actually have an adverse effect in functionality.
On the other side, unilateral toe walking in TD subjects results in no increase in
the rectus femoris EMG during mid-swing (Houx et al. 2014; Romkes and Brunner
2007), as it typically occurs in children with CP (Patikas et al. 2005). This indicates
that this activity which is observed in children with CP could be attributed to
abnormal descending commands (primary cause). However, it cannot be ruled out
that this activation could be related to muscle weakness (or higher force require-
ments due to the position of the body) of the distal lower limb muscles in children
with CP that is compensated by this activation pattern from more proximal muscles
(Ivanenko et al. 2008). Moreover, the rectus femoris, as a biarticular muscle, induces
not only knee extension but also hip flexion. Therefore, it has been suggested for
hemiplegic children with CP that during mid-swing, the increased EMG in the
biarticular rectus femoris could (a) counterbalance the excessive activation of the
hamstrings (acting as knee extensor) and/or (b) contribute to the shortening of the
limp to assist foot clearance, when dorsiflexion is limited (acting as hip flexor)
(Patikas et al. 2005). For this reason, weakening the rectus femoris could limit knee
extension during the initial contact (Hullin et al. 1996; Patikas et al. 2005), which
might subsequently cause crouch gait or stiff-knee gait (Damiano et al. 1995; Beals
2001), or could limit hip flexion, which might obstruct foot clearance (Chambers
et al. 1998; Goldberg et al. 2004).
It is worth noting that the functional significance of the activity of the quadriceps
during the stance phase might be different between equinus and crouch gait. Crouch
gait is a clinical condition observed in patients with CP which is characterized by an
excessive hip and knee flexion during the stance phase. In contrast to the equinus
gait, this position increases the patellofemoral forces, because the ground reaction
force vector passes far behind the knee joint, generating a large internal knee
extension moment. Furthermore, crouch gait obstructs foot clearance and increases
energy consumption (Perry et al. 1975; Arnold et al. 2005). TD individuals demon-
strate higher quadriceps EMG during the stance phase when the knee is more flexed,
in order to counterbalance the increased extensor moment (Simonsen and Alkjær
2012). Therefore, in the presence of increased knee flexion as observed in crouch
gait, it is possible that the overactive quadriceps compensates the increased extensor
moments. Indeed crouch gait is accompanied by a relatively high rectus femoris
activation (Lin et al. 2000). Simulation studies have shown that the activation of the
16 D.A. Patikas
gluteus maximus, vasti (vastus lateralis, intermedius, and medialis), and soleus
muscles contribute to produce the extensor-required moments during crouch gait
(Arnold et al. 2005). There are many studies reporting that weakness on these
muscles may contribute to crouch gait (Rodda and Graham 2001; Beals 2001;
Damiano et al. 1995; Wiley and Damiano 1998), and therefore it is suggested that
strength training – especially for the hip extensors – may limit hip and knee flexion
during stance (Arnold et al. 2005). Hence, by identifying the functional significance
of the quadriceps activation during stance phase in crouch gait, it is recommended
that the seemingly abnormally increased quadriceps EMG activity, should be further
enhanced (e.g. by means of strength training) and not diminished, in order to
improve gait.
Another piece of evidence regarding the exposure of possible compensatory
mechanisms with the help of EMG assessment is by comparing EMG data before
and after an orthopedic surgery. Earlier studies supported that the EMG patterns in
children are some kind of “fingerprint” (Gueth et al. 1985) and are not expected to
change after an orthopedic surgery, since this intervention aims to correct bony and
soft tissue deformities which do not belong to the primary causes that could
influence EMG (Perry and Hoffer 1977; Gueth et al. 1985; Brunt and Scarborough
1988; DeLuca 1991; Lee et al. 1992). However, there are several indications that the
EMG of children with CP can adapt when using orthosis (Romkes et al. 2006). This
implies that when changing the environmental constraints after an orthopedic sur-
gery, it is possible to detect differences in EMG. Indeed, Patikas et al. (2007)
reported that after multilevel orthopedic surgery, the EMG pattern of the soleus,
lateral gastrocnemius, and tibialis anterior approached closer to the means of TD
children. Furthermore, patients who demonstrated abnormal EMG bursts for the
medial hamstrings during pre-swing before surgery had no such bursts after the
surgery. These findings imply that – at least partly – the abnormal EMG before the
surgery was compensating the biomechanical constraints, and after the surgery,
children with CP were able to reorganize their muscle recruitment patterns accord-
ingly (Sporns and Edelman 1993). More specifically, it has been suggested that the
reduction in equinus induced more normal soleus, lateral gastrocnemius, and tibialis
anterior EMG patterns and that the activation of the medial hamstrings during
pre-swing before the surgery might have contributed to the knee flexion.
Considering the above, it is obvious that an abnormal EMG pattern of a muscle
does not necessarily imply that there is a deficit of neural nature. This has been also
shown after examining patients with orthopedic problems but free from any neuro-
logical symptoms, who demonstrated that from the 77 examined legs, 66 % had
abnormal EMG activity, 17 % demonstrated increased knee flexors/extensors
co-activation in stance phase, and 47 % showed abnormal EMG activity in more
than one muscles (Brunner and Romkes 2008). This indicates that the changes in
EMG could be attributed to muscle weakness, skeletomuscular deformities, and/or
pain. Furthermore, the most common atypical muscle activation that was
documented in this study was the earlier activation of the triceps surae muscle during
terminal swing, which is apparent in patients with CP as well (Romkes and Brunner
2007). Therefore, it seems that patients with CP and neurologically intact patients
compensate similarly to a common symptom, which could be muscle weakness

(Brunner and Romkes 2008; Damiano et al. 2002; Elder 2003; Stackhouse
et al. 2005), designating the importance of maintaining or improving muscle strength
in these patients (Damiano and Abel 1998; Dodd et al. 2002).
It should be underlined that the above comparisons are based more on the EMG
activation patterns than on the EMG activation itself, as a magnitude. This interpre-
tation reduces the true biological intersubject variability. Therefore, no change in the
activation pattern implies that the ability of the central nervous system to create a
specific EMG pattern still persists, but it cannot be excluded that the central drive
might have been reduced or enhanced. For instance, it is known that children with
CP, compared to TD children, have a significantly lower capacity to fully recruit their
motor units during a maximal voluntary contraction (Stackhouse et al. 2005; Rose
and McGill 2005). Nevertheless, despite structural differences due to disease
(in muscle length, stiffness, etc.), it seems that children with CP and equinus are
able to create very similar EMG patterns to the one observed in TD children when
walking on their toes.
The equinus deformity in ambulatory children with CP impairs ankle function
due to either a fixed contracture of the triceps surae or a dynamic tightness due to
excessive activation of the ankle plantar flexors (Davids et al. 1999). The distinction
between these two cases is crucial for the treatment, since dynamic muscle tightness
is usually treated nonsurgically (Flett et al. 1999; Rosenthal 1984). The best way to
distinguish between these two conditions with the most certainty is when the patient
is under general anesthesia, but this is not always practically feasible. Classification
using generalized dynamic neural networks after processing kinematic and kinetic
data could distinguish between the two conditions with 91 % accuracy (Zwick
et al. 2004). On the other hand, EMG patterns of the medial gastrocnemius, tibialis
anterior, rectus femoris, and semitendinosus for these two conditions seem rather
similar (Romkes and Brunner 2007).
Limitations
EMG measurements, as all assessments, are subjected to measurement error which

affects the reliability and validity of the results and undermine our confidence for the
conclusions. It has been previously reported that EMG patterns may vary, not only
within patients but also within TD individuals, especially in young populations
(Shiavi et al. 1987a; Hof et al. 2002; Granata et al. 2005; Wootten et al. 1990). In
some cases, it is also difficult to compare EMG data between studies due to
differences in the methodological approaches. The signal quality is influenced by
multiple factors such as thickness and composition of the subcutaneous tissue layer,
fiber diameter, length, type, orientation and position, electrode size and shape, and
interelectrode distance (Farina et al. 2002). Despite the existence of specific guide-
lines regarding the measurement and analysis procedures (Hermens et al. 1999;
Merletti 1999), there is no standardization in the analysis of the signal, leading many
times to inconsistencies between the studies (Burden et al. 2003). Furthermore, it
18 D.A. Patikas
should be underlined that the EMG–force relationship should not be considered as

linear and should be individually analyzed especially in motor disorders that influ-
ence the effect of passive components on the final moment output that is actually
measured. This should be taken into account when the goal is to interpret EMG
patterns and translate them to movement and force quantities.
There are also several issues regarding the level of agonist/antagonist
co-activation and the way it is calculated. One way is to calculate the activation of
the antagonist muscles in absolute values, in terms of amplitude or activation
duration. More often, the level of co-activation is represented as a ratio between
the antagonist and agonistic or between the antagonist and total (antagonist + ago-
nist) EMG (Teixeira da Fonseca et al. 2004). This is useful and necessary when
comparing the co-activation level of different subjects or groups of subjects. How-
ever, with this approach, it is sometimes unclear whether an increased ratio is due to
an increased antagonist EMG or decreased agonist EMG or both. This is of particular
importance when the dominator is relatively small and variable, which may often
occur in pathological cases and younger (under 10 years old) children (Tirosh
et al. 2013; Granata et al. 2005). As an alternative, EMG amplitude of the antagonist
muscles can be normalized to its maximum EMG during an isometric contraction.
This allows in some extent comparisons of normalized EMG of antagonist muscles
which have intrinsic differences in several properties that affect the EMG magnitude
(i.e., anatomy, skin impendence, muscle fiber composition, and electrode position-
ing). Nonetheless, this could still raise some issues of validity when maximum
strength and activation levels are different between subject groups (Damiano
et al. 2000). Finally, the co-activation index is a measure of synergies of two muscles
(agonist and antagonist in this case) which is often used in the literature (Damiano
et al. 2000; Unnithan et al. 1996a; Gross et al. 2013; Ikeda et al. 1998; Falconer and
Winter 1985). It actually calculates the overlap area between the linear envelopes of
the normalized EMG of the two muscles and is most commonly used in the
literature.
Future Directions
EMG has been a very useful diagnostic tool for patients with motor disorders during
the past decades and has given a significant piece of information to understand the
nature of the disorder and the underlying mechanisms causing the disease. Further-
more, feeding kinematic, kinetic, and EMG data to simulation models gives a great
opportunity to create more detailed, sophisticated, and realistic models and to better
understand normal and pathological gait. Nevertheless, there are still unsolved issues
regarding the standardization of measurement and methods to make data more
comparable among studies. For instance, it would be helpful if EMG measurement
procedures (standardized gait speed and cadence) as well as processing and normal-
ization methods would follow a common, widely accepted protocol. Furthermore,
taking into account the great heterogeneity of patients, multicenter studies could give
the possibility to assess a larger number of patients and create more uniform clusters
with common causes and symptoms. Additionally, capturing a wider range of

muscles with surface or intramuscular electrodes could contribute to a more precise
description of the clinical picture and provide more accurate input to simulation gait
models. All these could help health professionals create more individualized and
precise treatment strategies with a potentially better outcome for the patient.
References
Ada L, Vattanasilp W, O’Dwyer NJ, Crosbie J (1998) Does spasticity contribute to walking
dysfunction after stroke? J Neurol Neurosurg Psychiatry 64:628–635
Adams RW, Gandevia SC, Skuse NF (1990) The distribution of muscle weakness in upper
motoneuron lesions affecting the lower limb. Brain 113:1459–1476
Aiona MD, Sussman MD (2004) Treatment of spastic diplegia in patients with cerebral palsy: part
II. J Pediatr Orthop B 13:S13–S38
Arias P, Espinosa N, Robles-García V, Cao R, Cudeiro J (2012) Antagonist muscle co-activation
during straight walking and its relation to kinematics: insight from young, elderly and
Parkinson’s disease. Brain Res 1455:124–131. doi:10.1016/j.brainres.2012.03.033
Arnold AS, Anderson FC, Pandy MG, Delp SL (2005) Muscular contributions to hip and knee
extension during the single limb stance phase of normal gait: a framework for investigating the
causes of crouch gait. J Biomech 38:2181–2189
Baratta RV, Solomonow MR, Zhou BH, Letson D, Chuinard R, D’Ambrosia RD (1988) The role of
antagonistic musculature in maintaining knee stability. Am J Sports Med 16:113–122
Bar-On L, Molenaers G, Aertbelien E, Monari D, Feys H, Desloovere K (2014) The relation
between spasticity and muscle behavior during the swing phase of gait in children with cerebral
palsy. Gait Posture 35:3354–3364
Beals RK (2001) Treatment of knee contracture in cerebral palsy by hamstring lengthening,
posterior capsulotomy, and quadriceps mechanism shortening. Dev Med Child Neurol
43:802–805
Beres-Jones JA, Harkema SJ (2004) The human spinal cord interprets velocity-dependent afferent
input during stepping. Brain 127:2232–2246. doi:10.1093/brain/awh252
Berger W, Quintern J, Dietz V (1982) Pathophysiology of gait in children with cerebral palsy.
Electroencephalogr Clin Neurophysiol 53:538–548
Blanchette AK, Noël M, Richards CL, Nadeau S, Bouyer LJ (2014) Modifications in ankle
dorsiflexor activation by applying a torque perturbation during walking in persons post-stroke:
a case series. J Neuroeng Rehabil 11:98
Bohannon RW, Andrews AW (1995) Relationship between impairments and gait performance after
stroke. A summary of relevant research. Gait Posture 3:236–240
Brunner R, Romkes J (2008) Abnormal EMG muscle activity during gait in patients without
neurological disorders. Gait Posture 27(3):399–407
Brunt D, Scarborough N (1988) Ankle muscle activity during gait in children with cerebral palsy
and equinovarus deformity. Arch Phys Med Rehabil 69:115–117
Burden AM, Trew M, Baltzopoulos V (2003) Normalisation of gait EMGs: a re-examination. J
Electromyogr Kinesiol 13:519–532
Busse ME, Wiles CM, van Deursen WM (2005) Muscle co-activation in neurological conditions.
Phys Ther Rev 7(4):247–253
Cai LL, Courtine G, Fong AJ, Burdick JW, Roy RR, Edgerton VR (2006) Plasticity of functional
connectivity in the adult spinal cord. Philos Trans R Soc Lond B 361:1635–1646
Cappellini G, Ivanenko YP, Poppele RE, Lacquaniti F (2006) Motor patterns in human walking and
running. J Neurophysiol 95:3426–3437. doi:10.1152/jn.00081.2006
20 D.A. Patikas
Chambers H, Lauer A, Kaufman K, Cardelia JM, Sutherland D (1998) Prediction of outcome after
rectus femoris surgery in cerebral palsy: the role of cocontraction of the rectus femoris and
vastus lateralis. J Pediatr Orthop 18:703–711
Cheron G, Cebolla AM, Bengoetxea A, Leurs F, Dan B (2007) Recognition of the physiological
actions of the triphasic EMG pattern by a dynamic recurrent neural network. Neurosci Lett 414
(2):192–196
Clark DJ, Ting LH, Zajac FE, Neptune RR, Kautz SA (2010) Merging of healthy motor modules
predicts reduced locomotor performance and muscle coordination complexity post-stroke. J
Neurophysiol 103:844–857
Corcoran PJ, Jebsen RH, Brengelmann GL, Simons BC (1970) Effects of plastic and metal leg
braces on speed and energy cost of hemiparetic ambulation. Arch Phys Med Rehabil 51:69–77
Corry I, Cosgrove A, Duffy C, Taylor T, Graham H (1999) Botulinum toxin A in hamstring
spasticity. Gait Posture 10(3):206–210
Crenna P (1998) Spasticity and ‘Spastic’ gait in children with cerebral palsy. Neurosci Biobehav
Rev 22:571–578
Damiano DL (1993) Reviewing muscle cocontraction: is it a developmental, pathological, or motor
control issue. Phys Occup Ther Pediatr 12:3–20
Damiano DL, Abel MF (1998) Functional outcomes of strength training in spastic cerebral palsy.
Arch Phys Med Rehabil 79:119–125
Damiano DL, Vaughan CL, Abel MF (1995) Muscle response to heavy resistance exercise in
children with spastic cerebral palsy. Dev Med Child Neurol 37:731–739
Damiano DL, Martellotta TL, Sullivan DJ, Granata KP, Abel MF (2000) Muscle force production
and functional performance in spastic cerebral palsy: relationship of cocontraction. Arch Phys
Med Rehabil 81:895–900
Damiano DL, Dodd KJ, Taylor NF (2002) Should we be testing and training muscle strength in
cerebral palsy? Dev Med Child Neurol 44:68–72
Damiano DL, Laws E, Carmines DV, Abel MF (2006) Relationship of spasticity to knee angular
velocity and motion during gait in cerebral palsy. Gait Posture 23:1–8
Davids J, Foti T, Dabelstein J, Bagley A (1999) Voluntary (normal) versus obligatory (cerebral
palsy) toe-walking in children: a kinematic, kinetic, and electromyographic analysis. J Pediatr
Orthop 19(4):461–469
De Quervain IA, Simon SR, Leurgans S, Pease WS, McAllister D (1996) Gait pattern in the early
recovery period after stroke. J Bone Joint Surg Am 78A(10):1506–1514
Delp SL, Arnold AS, Speers RA, Moore CA (1996) Hamstrings and psoas lengths during normal
and crouch gait: Implications for muscle-tendon surgery. J Orthop Res 14(1):144–151
DeLuca PA (1991) The use of gait analysis and dynamic EMG in the assessment of the child with
cerebral palsy. Hum Mov Sci 10:543–554
Den Otter AR, Geurts ACH, Mulder T, Duysens J (2006) Gait recovery is not associated with
changes in the temporal patterning of muscle activity during treadmill walking in patients with
post-stroke hemiparesis. Clin Neurophysiol 117:4–15
Desrosiers J, Bourbonnais D, Bravo G, Roy PM, Guay M (1996) Performance of the “unaffected”
upper extremity of elderly stroke patients. Stroke 27:1564–1570
Detrembleur C, Willems P, Plaghki L (1997) Does walking speed influence the time pattern of
muscle activation in normal children? Dev Med Child Neurol 39:803–807
Dickey JP, Winter DA (1992) Adaptations in gait resulting from unilateral ischaemic block of the
leg. Clin Biomech (Bristol, Avon) 7:215–225. doi:10.1016/S0268-0033(92)90004-N
Dietz V, Colombo G (1998) Influence of body load on the gait pattern in Parkinson’s disease. Mov
Disord 13(2):255–261
Dietz V, Quintern J, Berger W (1981) Electrophysiological studies of gait in spasticity and rigidity.
Brain 104:431–449
Dietz V, Berger W, Horstrnann GA (1988) Posture in Parkinson’s disease: impairment of reflexes
and programming. Ann Neurol 1988:660–669
Dietz V, Ziiistrab W, Prokopc T, Bergerc W (1995) Leg muscle activation during gait in Parkinson’s
disease: adaptation and interlimb coordination. Electroencephalogr Clin Neurophysiol 97
(6):408–415
Dobson F, Morris ME, Baker RJ, Graham HK (2007) Gait classification in children with cerebral
palsy: a systematic review. Gait Posture 25:140–152
Dodd KJ, Taylor NF, Damiano DL (2002) A systematic review of the effectiveness of strength-
training programs for people with cerebral palsy. Arch Phys Med Rehabil 83:1157–1164
Dursun E, Dursun N, Alican D (2004) Effects of biofeedback treatment on gait in children with
cerebral palsy. Disabil Rehabil 26(2):116–120
Duysens J, De Groote F, Jonkers I (2013) The flexion synergy, mother of all synergies and father of
new models of gait. Front Comput Neurosci 7:14. doi:10.3389/fncom.2013.00014
Elder GBC (2003) Contributing factors to muscle weakness in children with cerebral palsy. Dev
Med Child Neurol 45:542–550
Falconer K, Winter DA (1985) Quantitative assessment of co-contraction at the ankle joint in
walking. Electromyogr Clin Neurophysiol 25:135–149
Farina D, Cescon C, Merletti R (2002) Influence of anatomical, physical, and detection-system
parameters on surface EMG. Biol Cybern 86:445–456
Flett PJ, Stern LM, Waddy H, Connell TM, Seeger JD, Gibson SK (1999) Botulinum toxin A versus
fixed cast stretching for dynamic calf tightness in cerebral palsy. J Paediatr Child Health 35:71–77
Fonseca ST, Holt KG, Saltzman E, Fetters L (2001) A dynamical model of locomotion in spastic
hemiplegic cerebral palsy: influence of walking speed. Clin Biomech (Bristol, Avon) 16:793–805
Frost G, Dowling JJ, Dyson K, Bar-Or O (1997) Cocontraction in three age groups of children
during treadmill locomotion. J Electromyogr Kinesiol 7(3):179–186
Fung J, Barbeau H (1994) Effects of conditioning cutaneomuscular stimulation on the soleus
H-reflex in normal and spastic paretic subjects during walking and standing. J Neurophysiol
72(5):2090–2104
Gage JR, Schwartz MH (2009) Consequences of brain injury on musculoskeletal development. The
identification and treatment of gait problems in cerebral palsy. Mac Keith Press, London
Gersten JW, Orr W (1971) External work of walking in hemiparetic patients. Scand J Rehabil Med
3:85–88
Gizzi L, Nielsen JF, Felici F, Ivanenko YP, Farina D (2011) Impulses of activation but not motor
modules are preserved in the locomotion of subacute stroke patients. J Neurophysiol (Bethesda)
106:202–210
Goldberg SR, Anderson FC, Pandy MG, Delp SL (2004) Muscles that influence knee flexion
velocity in double support: implications for stiff-knee gait. J Biomech 37:1189–1196
Granata KP, Padua DA, Abel MF (2005) Repeatability of surface EMG during gait in children. Gait
Posture 22:346–350
Griffin PP, Wheelhouse WW, Shiavi R, Bass W (1977) Habitual toe-walkers. J Bone Joint Surg Am
59A:97–101
Gross R, Leboeuf F, Hardouin JB, Lempereur M, Perrouin-Verbe B, Remy-Neris O, Brochard S
(2013) The influence of gait speed on co-activation in unilateral spastic cerebral palsy children.
Clin Biomech (Bristol, Avon) 28(3):312–317. doi:10.1016/j.clinbiomech.2013.01.002
Gueth V, Abbink F, Reuken R (1985) Comparison of pre and post-operative electromyographs in
children with cerebral palsy. Electromyogr Clin Neurophysiol 25:233–243
Hermens HJ, Freriks B, Merletti R, Stegeman DF, Blok JH, Raw G, Disselhorst-Klug C, Hägg GM
(1999) European recommendations for surface electromyography. Results of the SENIAM
project. Roessingh Research and Development, Enshede
Higginson JS, Zajac FE, Neptune RR, Kautz SA, Burgar CG, Delp SL (2006a) Effect of equinus
foot placement and intrinsic muscle response on knee extension during stance. Gait Posture
23:32–36
Higginson JS, Zajac FE, Neptune RR, Kautz SA, Delp SL (2006b) Muscle contributions to support
during gait in an individual with post-stroke hemiparesis. J Biomech 39:1769–1777
22 D.A. Patikas
Hirschberg GG, Nathanson M (1952) Electromyographic recording of muscular activity in normal

and spastic gaits. Arch Phys Med Rehabil 33:217–224
Hof AL, Geelen BA, Van den Berg J (1983) Calf muscle moment, work and efficiency in level
walking; role of series elasticity. J Biomech 16:523–537
Hof AL, Elzinga H, Grimmius W, Halbertsma JPK (2002) Speed dependence of averaged EMG
profiles in walking. Gait Posture 16:78–86
Hortobágyi T, Solnik S, Gruber A, Rider P, Steinweg K, Helseth J, DeVita P (2009) Interaction
between age and gait velocity in the amplitude and timing of antagonist muscle coactivation.
Gait Posture 29:558–564
Houx L, Lempereur M, Rémy-Néris O, Gross R, Brochard S (2014) Changes in muscle activity in
typically developing children walking with unilaterally induced equinus. Clin Biomech (Bristol,
Avon) 29(10):1116–1124
Hullin MG, Robb JE, Loudon IR (1996) Gait patterns in children with hemiplegic spastic cerebral
palsy. J Pediatr Orthop 5(B):247–251
Ikeda AJ, Abel MF, Granata KP, Damiano DL (1998) Quantification of cocontraction in spastic
cerebral palsy. Electromyogr Clin Neurophysiol 38:497–504
Ivanenko YP, Cappellini G, Poppele RE, Lacquaniti F (2008) Spatiotemporal organization of
a-motoneuron activity in the human spinal cord during different gaits and gait transitions. Eur
J Neurol 27:3351–3368. doi:10.1111/j.1460-9568.2008.06289.x
Ivanenko YP, Cappellini G, Solopova IA, Grishin AA, MacLellan MJ, Poppele RE, Lacquaniti F
(2013) Plasticity and modular control of locomotor patterns in neurological disorders with motor
deficits. Front Comput Neurosci 7:123. doi:10.3389/fncom.2013.00123
Johansson RS, Westling G (1988) Coordinated isometric muscle commands adequately and
erroneously programmed for the weight during lifting task with precision grip. Exp Brain Res
71:59–71
Jones RD, Donaldson IM, Parkin PJ (1989) Impairment and recovery of ipsilateral sensory-motor
function following unilateral cerebral infarction. Brain 112:113–132
Jorgensen HS, Nakayama H, Raaschou HO, Olsen TS (1995) Recovery of walking function in
stroke patients: the Copenhagen Stroke Study. Arch Phys Med Rehabil 76(1):27–32
Kalen V, Adler N, Bleck EE (1986) Electromyography of idiopathic toe walking. J Pediatr Orthop
6:31–33
Kautz SA, Brown DA (1998) Relationships between timing of muscle excitation and impaired
motor performance during cyclical lower extremity movement in post-stroke hemiplegia. Brain
121:515–526
Kerrigan DC, Gronley J, Perry J (1991) Stiff-legged gait in spastic paresis. A study of quadriceps
and hamstrings muscle activity. Am J Phys Med Rehabil 70:294–300
Knarr BA, Zeni JA Jr, Higginson JS (2012) Comparison of electromyography and joint moment as
indicators of co-contraction. J Electromyogr Kinesiol 22(4):607–611
Knutsson E, Richards C (1979) Different types of disturbed motor control in gait of hemiparetic
patients. Brain 102:405–430
Koller JR, Jacobs DA, Ferris DP, Remy CD (2015) Learning to walk with an adaptive gain
proportional myoelectric controller for a robotic ankle exoskeleton. J Neuroeng Rehabil
12:97. doi:10.1186/s12984-015-0086-5
Lacquaniti F, Maioli C (1987) Anticipatory and reflex coactivation of antagonist muscles in
catching. Brain Res 406:373–378
Lacquaniti F, Maioli C (1989) The role of preparation in tuning anticipatory and reflex responses
during catching. J Neurosci 9:134–148
Lacquaniti F, Ivanenko YP, Zago M (2012) Patterned control of human locomotion. J Physiol 590
(10):2189–2199
Lamontagne A, Malouin F, Richards CL (2000a) Contribution of passive stiffness to ankle
plantarflexor moment during gait after stroke. Arch Phys Med Rehabil 81:351–358
Lamontagne A, Richards CL, Malouin F (2000b) Coactivation during gait as an adaptive behavior
after stroke. J Electromyogr Kinesiol 10:407–415
Lamontagne A, Malouin F, Richards CL (2001) Locomotor-specific measure of spasticity of

plantarflexor muscles after stroke. Arch Phys Med Rehabil 82(12):1696–1704
Lamontagne A, Malouin F, Richards CL, Dumas F (2002) Mechanisms of disturbed motor control
in ankle weakness during gait after stroke. Gait Posture 15:244–255
Layne CS, McDonald PV, Bloomberg JJ (1997) Neuromuscular activation patterns during treadmill
walking after space flight. Exp Brain Res 113:104–116
Lee EH, Goh JCH, Bose K (1992) Value of gait analysis in assessment of surgery in cerebral palsy.
Levin MF, Dimov M (1997) Spatial zones for muscle coactivation and the control of postural
stability. Brain Res 757:43–59
Lin C-J, Guo L-Y, Su F-C, Chou Y-L, Cherng R-J (2000) Common abnormal kinetic patterns of the
knee in gait in spastic diplegia of cerebral palsy. Gait Posture 11:224–232
Manganotti P, Zaina F, Falso M, Milanese F, Fiaschi A (2007) Evaluation of botulinum toxin
therapy of spastic equinus in paediatric patients with cerebral palsy. J Rehabil Med 39
(2):115–120
Marks M, Hirschberg GG (1958) Analysis of the hemiplegic gait. Ann N Y Acad Sci 74:59–77
Marsden J, Ramdharry G, Stevenson V, Thompson A (2012) Muscle paresis and passive stiffness:
key determinants in limiting function in hereditary and sporadic spastic paraparesis. Gait Posture
35:266–271
Massion J (1992) Movement, posture and equilibrium: interaction and coordination. Prog
Neurobiol 38(1):35–56
Mazzaro N, Grey MJ, Sinkjaer T, Andersen JB, Pareyson D, Schieppati M (2005) Lack of on-going
adaptations in the soleus muscle activity during walking in patients affected by large-fiber
neuropathy. J Neurophysiol (Bethesda) 93(6):3075–3085
McGowan CP, Neptune RR, Clark DJ, Kautz SA (2010) Modular control of human walking:
adaptations to altered mechanical demands. J Biomech 43(3):412–419
Merletti R (1999) Standards for reporting EMG data. J Electromyogr Kinesiol 9(1):III–IV
Metaxiotis D, Siebel A, Döderlein L (2002) Repeated botulinum toxin A injections in the treatment
of spastic equinus foot. Clin Orthop Relat Res 394:177–185
Miller F, Cardoso DR, Lipton GE, Albarracin JP, Dabney KW, Castagno P (1997) The effect of
rectus EMG patterns on the outcome of rectus femoris transfers. J Pediatr Orthop 17:603–607
Mulroy SJ, Gronley JK, Weiss W, Newsam CJ, Perry JP (2003) Use of cluster analysis for gait
pattern classification of patients in the early and late recovery phases following stroke. Gait
Posture 18:114–125
Murray MP, Guten GN, Sepic SB, Gardner GM, Baldwin JM (1978) Function of the triceps surae
during gait. Compensatory mechanisms for unilateral loss. J Bone Joint Surg Am 60A:354–363
Murray M, Hardee A, Goldberg RL, Lewek MD (2014) Loading and knee flexion after stroke: less
does not equal more. J Electromyogr Kinesiol 24:172–177
Nadeau S, Gravel D, Arsenault AB, Bourbonnais D (1996) A mechanical model to study the
relationship between gait speed and muscular strength. IEEE Trans Rehabil Eng 4:386–394
Nadeau S, Gravel D, Arsenault AB, Bourdonnais D (1999) Plantarflexor weakness as a limiting
factor of gait speed in stroke subjects and the compensating role of hip flexors. Clin Biomech
(Bristol, Avon) 14(2):125–135. doi:10.1016/S0268-0033(98)00062-X
Nardone A, Schieppati M (2006) Balance control in peripheral neuropathy: are patients equally
unstable under static and dynamic conditions? Gait Posture 23(3):364–373
Nathan PW, Smith M, Deacon P (1996) Vestibulospinal, reticulospinal and descending
propriospinal nerve fibres in man. Brain 119(6):1809–1833
Nene AV, Byrne C, Hermens HJ (2004) Is rectus femoris really a part of quadriceps? Assessment of
rectus femoris function during gait in able-bodied adults. Gait Posture 20:1–13
Neptune RR, McGowan CP, Fiandt JM (2009) The influence of muscle physiology and advanced
technology on sports performance. Annu Rev Biomed Eng 11:81–107
24 D.A. Patikas
Nilsson J, Thorstensson A, Halbertsma J (1985) Changes in leg movements and muscle activity
with speed of locomotion and mode of progression in humans. Acta Physiol Scand
123:457–475. doi:10.1111/j.1748-1716.1985.tb07612.x
Novak I, McIntyre S, Morgan C, Campbell L, Dark L, Morton N, Stumbles E, Wilson SA,
Goldsmith S (2013) A systematic review of interventions for children with cerebral palsy:
state of the evidence. Dev Med Child Neurol 55(10):885–910
4:136–148
Olney SJ, MacPhail HEA, Hedden DM, Boyce WF (1990) Work and power in hemiplegic cerebral
palsy gait. Phys Ther 70:431–438
Olney SJ, Griffin MP, Monga TN, McBride ID (1991) Work and power in gait of stroke patients.
Osternig LR, Caster BL, James CR (1995) Contralateral hamstring (biceps femoris) coactivation
patterns and anterior cruciate ligament dysfunction. Med Sci Sports Exerc 27(6):805–808
Õunpuu S, Davis RB, DeLuca PA (1996) Joint kinetics: methods, interpretation and treatment
decision-making in children with cerebral palsy and myelomeningocele. Gait Posture 4:62–78
Papariello S, Skinner S (1985) Dynamic electromyographic analysis of habitual toewalkers. J
Pediatr Neurol 5:171–175
Parvataneni K, Ploeg L, Olney SJ, Brouwer B (2009) Kinematic, kinetic and metabolic parameters
of treadmill versus overground walking in healthy older adults. Clin Biomech (Bristol, Avon) 24
(1):95–100
Patikas D, Wolf SI, Döderlein L (2005) Electromyographic evaluation of the sound and involved
side during gait of spastic hemiplegic children with cerebral palsy. Eur J Neurol 12(9):691–699.
doi:10.1111/j.1468-1331.2005.01047.x
Patikas D, Wolf SI, Schuster W, Armbrust P, Dreher T, Döderlein L (2007) Electromyographic
patterns in children with cerebral palsy: do they change after surgery? Gait Posture 26
(3):362–371. doi:10.1016/j.gaitpost.2006.10.012
Perry JP (1992) Gait analysis: normal and pathological function. Slack, Thorofare
Perry JP, Hoffer MM (1977) Preoperative and postoperative dynamic electromyography as an
aid in planning tendon transfers in children with cerebral palsy. J Bone Joint Surg Am
59A:531–537
Perry J, Hoffer M, Giovan P, Antonelli D, Greenberg R (1974) Gait analysis of the triceps surae in
cerebral palsy. A preoperative and postoperative clinical and electromyographic study. J Bone
Joint Surg Am 56A(3):511–520
Perry J, Antonelli D, Ford W (1975) Analysis of knee-joint forces during flexed-knee stance. J Bone
Joint Surg Am 57A:961–967
Perry J, Fontaine JD, Mulroy SJ (1995) Findings in post-poliomyelitis syndrome. Weakness of
muscles of the calf as a source of late pain and fatigue of muscles of the thigh after poliomyelitis.
J Bone Joint Surg Am 77A:1148–1153
Perry J, Burnfield JM, Gronley JK, Mulroy SJ (2003) Toe walking: muscular demands at the ankle
and knee. Arch Phys Med Rehabil 84:7–16
Peterson DS, Martin PE (2010) Effects of age and walking speed on coactivation and cost of
walking in healthy adults. Gait Posture 31:355–359
Popovic DB, Sinkaer T, Popovic MB (2009) Electrical stimulation as a means for achieving
recovery of function in stroke patients. Neurorehabilitation 25:45–58. doi:10.3233/NRE-
2009-0498
Prilutsky BI, Gregor RJ (2001) Swing- and support-related muscle actions differentially trigger
human walk-run and run-walk transitions. J Exp Biol 204:2277–2287
Reding MJ, Potes E (1988) Rehabilitation outcome following initial unilateral hemispheric stroke.
Life table analysis approach. Stroke 19:1354–1358
Richards L (1985) EMG activity level comparison in quadriceps and hamstrings in five dynamic
activities. In: Winter DA, Norman RW, Wells RP (eds) Biomechanics, vol IX-A. Human
Kinetics Publishers, Champaign, pp 313–317
Richards CL, Malouin F, Dumas F, Wood-Dauphinee S (1992) The relationship of gait speed to
clinical measures of function and muscle activations during recovery post-stroke. In: North
American Congress on Biomechanics, pp 299–302
Rodda JM, Graham HK (2001) Classification of gait patterns in spastic hemiplegia and spastic
diplegia: a basis for a management algorithm. Eur J Neurol 8(5):S98–S108
Romkes J, Brunner R (2007) An electromyographic analysis of obligatory (hemiplegic cerebral
palsy) and voluntary (normal) unilateral toe-walking. Gait Posture 26(4):577–586
Romkes J, Hell AK, Brunner R (2006) Changes in muscle activity in children with hemiplegic
cerebral palsy while walking with and without ankle-foot orthoses. Gait Posture 24(4):467–474
Rosa MC, Marques A, Demain S, Metcalf CD, Rodrigues J (2014) Methodologies to assess muscle
co-contraction during gait in people with neurological impairment – a systematic literature
review. J Electromyogr Kinesiol 24(2):179–191. doi:10.1016/j.jelekin.2013.11.003
Rose J, McGill KC (2005) Neuromuscular activation and motor-unit firing characteristics in
cerebral palsy. Dev Med Child Neurol 47(5):329–336
Rose J, Martin J, Torburn L, Rinsky LA, Gamble JG (1999) Electromyographic differentiation of
diplegic cerebral palsy from idiopathic toe walking: involuntary coactivation of the quadriceps
and gastrocnemius. J Pediatr Orthop 19(5):677
Rosenthal RK (1984) The use of orthotics in foot and ankle problems in cerebral palsy. Foot Ankle
4:195–200
Schmitz A, Silder A, Heiderscheit B, Mahoney J, Thelen DG (2009) Differences in lower-extremity
muscular activation during walking between healthy older and young adults. J Electromyogr
Kinesiol 19:1085–1091
Shiavi R, Bugle HJ, Limbird T (1987a) Electromyographic gait assessment, part 1: adult EMG
profiles and walking speed. J Rehabil Res Dev 24(2):13–23
Shiavi R, Bugle HJ, Limbird T (1987b) Electromyographic gait assessment, part 2: preliminary
assessment of hemiparetic synergy patterns. J Rehabil Res Dev 24:24–30
Siegler S, Moskowitz GD, Freedman W (1984) Passive and active components of the internal
moment developed about the ankle joint during human ambulation. J Biomech 17:647–652
Simonsen EB (2014) Contributions to the understanding of gait control. Danish Med J 61(4):B4823
Simonsen EB, Alkjær T (2012) The variability problem of normal human walking. Med Eng Phys
34(2):219–224
Solomonow MR, Baratta RV, Zhou B-H, D’Ambrosia RD (1988) Electromyogram coactivation
patterns of the elbow antagonist muscles during slow isokinetic movement. Exp Neurol
100:470–477
Solopova IA, Tihonova DY, Grishin AA, Ivanenko YP (2011) Assisted leg displacements and
progressive loading by a tilt table combined with FES promote gait recovery in acute stroke.
Neurorehabilitation 29:67–77
Sporns O, Edelman GM (1993) Solving Bernstein’s problem: a proposal for the development of
coordinated movement by selection. Child Dev 64(4):960–981
Stackhouse SK, Binder-Macleod SA, Lee SCK (2005) Voluntary muscle activation, contractile
properties, and fatigability in children with and without cerebral palsy. Muscle Nerve
31:594–601
Sutherland DH, Cooper L, Daniel D (1980) The role of the ankle plantar flexors in normal walking.
J Bone Joint Surg Am 62(3):354–363
Sutherland D, Kaufman K, Wyatt M, Chambers H, Mubarak S (1999) Double-blind study of
botulinum A toxin injections into the gastrocnemius muscle in patients with cerebral palsy.
Tardieu C, Lespargot A, Tabary C, Bret MD (1989) Toe-walking in children with cerebral palsy:
contributions of contracture and excessive contraction of triceps surae muscle. Phys Ther
69:656–662
Teixeira da Fonseca S, Silva PL, Ocarino JM, Guimaràes RB, Oliveira MT, Lage CA (2004)
Analyses of dynamic co-contraction level in individuals with anterior cruciate ligament injury.
J Electromyogr Kinesiol 14:239–247
26 D.A. Patikas
Tirosh O, Sangeux M, Wong M, Thomason P, Graham HK (2013) Walking speed effects on the
lower limb electromyographic variability of healthy children aged 7–16 years. J Electromyogr
Kinesiol 23(6):1451–1459
Tong RK, Ng MF, Li LS, So EF (2006) Gait training of patients after stroke using an electrome-
chanical gait trainer combined with simultaneous functional electrical stimulation. Phys Ther
86:1282–1294. doi:10.2522/ptj.20050183
Tuzson AE, Granata KP, Abel MF (2003) Spastic velocity threshold constrains functional perfor-
mance in cerebral palsy. Arch Phys Med Rehabil 84:1363–1368
Unnithan VB, Dowling JJ, Frost G, Ayub B, Bar-Or O (1996a) Cocontraction and phasic activity
during gait in children with cerebral palsy. Electromyogr Clin Neurophysiol 46:487–494
Unnithan VB, Dowling JJ, Frost G, Bar-Or O (1996b) Role of cocontraction in the O2 cost of
walking in children with cerebral palsy. Med Sci Sports Exerc 28:1498–1504
Valero-Cuevas FJ (2005) An integrative approach to the biomechanical function and neuromuscular
control of the fingers. J Biomech 38:673–684
Whittle MW (1996) Clinical gait analysis: a review. Hum Mov Sci 15:369–387
Wiley ME, Damiano DL (1998) Lower-extremity strength profiles in spastic cerebral palsy. Dev
Willerslev-Olsen M, Andersen JB, Sinkjaer T, Nielsen JB (2014) Sensory feedback to ankle plantar
flexors is not exaggerated during gait in spastic hemiplegic children with cerebral palsy. J
Winter DA (1989) Biomechanics of normal and pathological gait: implications for understanding
human locomotor control. J Mot Behav 21:337–355. doi:10.1080/00222895.1989.10735488
(5):361–367
Winter DA, Yack HJ (1987) EMG profiles during normal human walking: stride-to-stride and inter-
subject variability. Electroencephalogr Clin Neurophysiol 67(5):402–411
Winters TF, Gage JR, Hicks RR (1987) Gait patterns in spastic hemiplegia in children and young
adults. J Bone Joint Surg Am 69(3):437–441
Wootten ME, Kadaba MP, Cochran GVB (1990) Dynamic electromyography. II. Normal patterns
during gait. J Orthop Res 8:259–265
Wortis SB, Marks M, Hirschberg GG, Nathanson M (1951) Gait analysis in hemiplegia. Trans Am
Neurol Assoc 76:181–183
Yan T, Hui-Chan CWY, Li LSW (2005) Functional electrical stimulation improves motor recovery
of the lower extremity and walking ability of subjects with first acute stroke. Stroke 36:80–85.
doi:10.1161/01.STR.0000149623.24906.63
Yang JF, Fung J, Edamura M, Blunt R, Stein RB, Barbeau H (1991a) H-reflex modulation during
walking in spastic paretic subjects. Can J Neurol Sci 18:443–452
Yang JF, Stein RB, James KB (1991b) Contribution of peripheral afferents to the activation of the
soleus muscle during walking in humans. Exp Brain Res 87:679–687
Zajac FE, Neptune RR, Kautz SA (2003) Biomechanics and muscle coordination of human
walking. Part II: lessons from dynamical simulations and clinical implications. Gait Posture
17:1–17. doi:10.1016/S0966-6362(02)00069-3
Zamparo P, Francescato MP, De Luca G, Lovati L, di Prampero PE (1995) The energy cost of level
walking in patients with hemiplegia. Scand J Med Sci Sports 5:348–352
Zwick EB, Leistrintz L, Milleit B, Saraph V, Zwick G, Galicki M, Witte H, Steinwender G (2004)
Classification of equinus in ambulatory children with cerebral palsy – discrimination between
dynamic tightness and fixed contracture. Gait Posture 20:273–279
Gait Scores – Interpretations
and Limitations
Veronica Cimolin and Manuela Galli
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Summary Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Summary Measures for Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Summary Measures for Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Abstract
Gait analysis (GA) represents a fundamental method by which quantitative
information is collected to aid in understanding the level of functional limitation
due to pathology, observing its evolution over time and evaluating rehabilitative
intervention effects. Essential to the method is the interpretation of the data
provided by gait analysis by an experienced, interdisciplinary team with substan-
tial knowledge in normal and pathologic gait. A typical gait analysis evaluation
procedure could include the computing of both specific parameters (values of the
plots in specific instant of the gait cycle, maximum values, range of motions, etc.)
and summary measures (one of few scores which quantify the deviation of a
patient’s gait pattern from normality). The use of specific parameters produces a
vast amount of data and graphs; for this reason, despite its objectivity, its use
could be complicated, and the data interpretation is difficult. Thus, there is a
V. Cimolin (*)
Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milano, Italy
e-mail: veronica.cimolin@polimi.it
M. Galli (*)
Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milano, Italy
IRCCS “San Raffaele Pisana” Tosinvest Sanità, Roma, Italy
e-mail: manuela.galli@polimi.it

DOI 10.1007/978-3-319-30808-1_34-1
2 V. Cimolin and M. Galli
growing awareness of the need for a summary index, specifically, a single

measure of the “quality” of a particular gait pattern.
Keywords
Gait analysis • Kinematics • Kinetics • Locomotion • Parameters • Summary
measures
Introduction
Quantitative gait analysis (GA) fully describes in a quantitative way a person’s

walking performance. In particular, clinical evidence and literature support the
efficacy of GA to quantify the level of functional limitation due to pathology and
its evolution over time, to help in the clinical decision-making process and to
quantify the effect of a rehabilitative program. The primary output of any GA is a
report where a multitude of measurements, including time-distance variables (spatio-
temporal data), joint motions (kinematics), forces, joint moments, and powers
(kinetics) as well as signals related to muscle activity (EMG), are provided. In
order to make an interpretation of a GA report, it is important to define a feature,
which represents a specific aspect of the GA data that is considered clinically
significant or, in other words, something that can be seen on one of the graphs.
Examples could be increased anterior pelvic tilt through the gait cycle, reduced knee
range of motion, too much ankle plantar flexion at initial contact, too low ankle
power at terminal stance, and so on. In order to describe a feature completely, the
side (right, left, or bilateral), the timing (the instant or the phase of the gait cycle), the
gait variable (pelvic tilt, hip rotation, ankle power, etc.), and the nature of the feature
(too much, too little, too early, etc.) should be all specified (Baker 2013). To well
identify and quantify the features significant to describe the individual’s gait pattern
and its deviations from physiological walking performance, some parameters could
be used and computed on the GA graphs. In particular, specific parameters could be
identified, starting from basographic, kinematic, and kinetic data: spatio-temporal
parameters (velocity, step length, step width, etc.), joint angle values at a specific
instant of the gait cycle (hip range of motion, value of knee angle in midstance, value
of ankle angle at initial contact, etc.), and peaks in moment and power graphs (the
absolute maximum of the curve representing the dorsi-plantarflexion moment in the
terminal stance and the maximum value of ankle power at push-off) (Fig. 1).
It is easy to understand that from this computation, a GA report produces a vast
amount of data, which sometimes could be seen as an obstacle for its clinical use. If
we consider that to verify the repeatability of the data several reports should be
analyzed for each patient, we can understand that reading and evaluating a GA trial
requires a lot of time. In clinical settings, there is sometimes an issue of having at
disposal a simple and immediate method to evidence, for instance, the gap between
the patient’s gait and the normality of reference, or to evidence the effects of a
treatment. To answer this issue, some summary measures have been developed in
order to provide simple measure of gait function and so to quantify with one (or few)
Gait Scores – Interpretations and Limitations 3
Fig. 1 Examples of specific parameters identified on kinematic graph (a) (knee flexion-extension
plot) and on kinetic graph (b) (ankle power plot)
number the deviation of a patient’s gait pattern from normality. In particular, these
indexes revealed to be useful to objectify clinical impression, to quantify the degree
of gait deviation from normal pattern, to stratify the severity of pathology, to
document changes in gait over time, and to quantify the effects of rehabilitative
interventions.
The aim of this chapter is to evaluate and summarize the current evidence base
related to the most important and widespread summary parameters proposed by the
literature, focusing on studies proposed for clinical applications and making an
interpretation on their advantages and limitations.
Summary Measures
The literature has proposed several summary measures that can be computed starting
from the data of the GA trials and that have a clinical application. Most parameters
were computed starting from kinematic GA data (Normalcy Index, Gait Deviation
Index, Gait Profile Score, Movement Deviation Profile); some attempts have been
conducted to proposed summary measures related partially (Hip Flexor Index (HFI))
or totally (GDI-Kinetic) to kinetics.
Here following, a brief description of each parameter, the data reduction tech-
nique used for the computation, the potential weaknesses/strengths, and the main
clinical/scientific experiences are given.
Summary Measures for Kinematics
Normalcy Index (NI) or Gillette Gait Index (GGI)

The Normalcy Index (NI) or Gillette Gait Index (GGI) (Schutte et al. 2000) repre-
sents the first attempt to define a single parameter to describe the gait quality of the
gait pattern in a global sense. It is a measure of how different a patient’s gait is to the
mean gait pattern for healthy individuals computed from 16 univariate gait param-
eters that were considered significant by the developers to capture the important
features of the gait pattern. The NI is computed using standard multivariate statistical
techniques (principal component analysis) applied to 16 3D-GA variables and, in
particular, three spatio-temporal parameters (percentage of stance phase, normalized
velocity, and cadence) and 13 kinematic parameters (mean pelvic tilt, range of pelvic
tilt, mean pelvic rotation, minimum hip flexion, range of hip flexion, peak abduction
in swing, mean hip rotation in stance, knee flexion at initial contact, time of peak
knee flexion, range of knee flexion, peak of dorsiflexion in stance, peak of
dorsiflexion in swing, and mean foot progression angle). The sum of the square of
these 16 independent variables is interpreted as the deviation of the subject’s gait
from normal. Using this statistical method, it is possible to measure and represent as
a single number the deviation of a pathological gait pattern from a normal average
profile. It has a mean value of 15.7, with high values reflecting more abnormal gait
pattern.
The NI has been the most widely cited and validated summary measure. In
particular, its use has been widely validated in cerebral palsy (CP) and idiopathic
toe walker populations. It has been shown to be effective when used to evaluate the
range of pathology present in specific diagnoses, to compare a subject’s gait to that
of others with the same diagnosis, to track a subject’s gait pathology over time, or to
examine the effectiveness of an intervention (Schutte et al. 2000; Trost et al. 2008;
Domagalska et al. 2013; Szopa et al. 2014).
The NI has been extensively used in CP, not only in children but also in adults, to
quantify the effects of specific treatments that have a global effect on gait pattern,
similar to multilevel orthopedic surgery (Thomason et al. 2011; Dreher et al. 2012;
Putz et al. 2016) and selective dorsal rhizotomy (Trost et al. 2008). However, the NI
did not exhibit the appropriate specificity and sensitivity when evaluating the effects
of targeted interventions, such as ankle-foot orthosis (AFO) (Brehm et al. 2008).
It has limitations though. First of all, the choice of the 16 kinematic parameters
used to calculate the index appears arbitrary and mixes spatio-temporal parameters
with gait variables (Baker 2013). Its computation was based in fact on the subjective
judgment of clinicians and experts in CP and so may not be the best possible set of
parameters universally. The use of the NI in other pathological states, such as in
children and adolescents with tumors in the central nervous system (CNS)
(Syczewska et al. 2010), adults with a diagnosis of central nervous system pathology
(Cretual et al. 2010), and adult lower limb amputees (Kark et al. 2012), has shown in
fact that NI could present some limitations mainly due to the parameter choice
(Cretual et al. 2010). In addition, the selected parameters included only kinematic
variables; it is well known that the inclusion of kinetic variables is useful for a
complete gait pattern assessment and in planning intervention. Third, only the
characteristic points of the curves are included. These all make up a strong limita-
tion. Then, there is no physical meaning to the multivariate components, and results
appear sensitive to the choice of control dataset. According to the computation
method, the NI requires in fact an able-bodied gait dataset to establish the means
and variance values of the control in each of the variables: it was found to be strongly
sensitive to lab-specific control data and the minimum sample size required in the set
of control subjects in order to have a reliable NI tool (McMulkin and MacWilliams
2008). Then, the choice of mean square (rather than root mean square) exaggerates
values for more abnormal gait pattern (Baker 2013).
Gait Deviation Index (GDI)

In order to address the limitations of NI, the Gait Deviation Index (GDI) was
developed (Schwartz and Rozumalski 2008). It uses data from across the gait
cycle for the nine joint angles, which generally regarded as most clinically signifi-
cant in GA (the pelvis and hip in three planes, the knee and ankle on the sagittal
plane, and the foot progression). From a large dataset of people who had been
evaluated with GA, the extraction of 15 gait features using the singular value
decomposition from the GA kinematics was performed. Applied to a control
group, these “gait features” define an averaged, non-pathological gait. The root
mean square distance of the 15 gait features between a subject exhibiting gait
pathology and the control group is computed. As the resulting measure was not
normally distributed, its logarithm was thus taken and scaled such that the mean
value for the healthy population was 100; 10 points below 100 corresponds to one
standard deviation away from the healthy group mean (Schwartz and Rozumalski
2008; Baker et al. 2009). The GDI was moderately correlated with the NI
(r2 = 0.56), suggesting that these two parameters are both measures of the same
underlying construct, although a large spread at any given level indicates that they
measure different aspects of the gait pathology (Schwartz and Rozumalski 2008).
The GDI was found to be normally distributed across people with different levels
of Functional Assessment Questionnaire (FAQ), and mean values for the different
levels were similar increments apart. The validity of the GDI were firstly assessed in
healthy and in individuals with CP, both in children (Schwartz and Rozumalski
2008; Cimolin et al. 2011; Massaad et al. 2014; Lee et al. 2014, McMulkin
et al. 2016) and in adults (Andersson and Mattsson 2001). The GDI in adults
demonstrated similar results in distributional properties as those reported in studies
on healthy and children with CP (Schwartz and Rozumalski 2008). The GDI has
now been applied successfully to patients with different pathological conditions, not
only in CP, such as Batten disease (Galli et al. 2012b), muscular dystrophy (Sienko
Thomas et al. 2010), lower limb amputees (Kark et al. 2012; Eshraghi et al. 2014),
Parkinson’s disease (Galli et al. 2012a; Speciali et al. 2014a), hip osteoarthritis
(Rosenlund et al. 2016), and rheumatoid arthritis (Brostrom et al. 2013; Esbjornsson
et al. 2014), and patients with slipped capital femoral epiphysis (Caskey et al. 2014)
and lumbar spinal stenosis (Garbelotti et al. 2014).
If compared to the NI, the GDI has several advantages. The entire variability in
kinematic variables across the gait cycle is used rather than a small number of
discrete parameters specific for children with CP, thereby removing much of the
subjectivity in the choice of the parameters. In addition, the values of the GDI
seemed to be much less sensitive to differences in the reference data, in contrast
with the NI which requires a reasonably large number of people in the reference
dataset and that values can vary significantly between different reference datasets
(McMulkin and MacWilliams 2008). The GDI proceeds naturally from the analysis
of gait features, which provides considerable data compression and a framework for
other analytical techniques, such as cluster analysis for gait classification (Schwartz
and Rozumalski 2008). In addition, it was shown in a previous study that the
uncertainty on the GGI increases when the GGI value increases with the pathology
(Assi et al. 2009). This could be related to the fact that the GGI increases indefinitely
with the pathology, which participates in the increase of the uncertainty, while the
variation of the GDI is limited by a finite interval from 0 to 100. It is relevant to note
that the calculation of the GDI, based on the whole gait cycle, for each curve, could
limit the propagation of errors, contrary to the GGI which uses well-defined param-
eters, on each curve, during the gait cycle (Massaad et al. 2014).
Gait Profile Score (GPS) and Movement Analysis Profile (MAP)

More recently, the Gait Profile Score (GPS) was proposed as the direct root mean
square distance between an individual’s data and the mean normal data computed
across the gait cycle for the same nine joint angles of one side. Thus, it represents a
simpler interpretation of the distance measures underlying the GDI: the GPS results
in a modified measure that can be calculated independently of the feature analysis. In
addition, to a global measure of the overall gait quality, i.e., the GPS, it can be
deconstructed to provide the Gait Variable Score (GVS) for nine kinematic variables
(the pelvis and hip in three planes, the knee and ankle on the sagittal plane, and the
foot progression) (Baker et al. 2009). The GPS is generally presented with the GVSs
in a bar chart, generating the Movement Analysis Profile (MAP). The MAP
describes the magnitude of deviation of the nine individual variables averaged
over the gait cycle, providing insight into which variables contribute to the increase
in GPS.
The GPS is normally distributed for the control population (mean 5.3 ). In patient
data, the distribution categorized either by GMFCS (for CP) or by FAQ (all patients)
is not normal but its log transform is (Baker et al. 2009). The GPS was validated
against established index measures of gait abnormality and general measures of
mobility in children with CP (Baker et al. 2009). The authors (Baker et al. 2012)
proposed a rationale for defining a minimal clinically important difference (MCID)
for the Gait Profile Score (GPS), which was found to be 1.68.
Strong positive correlations were found between the GPS and MAP component
scores, and the clinicians’ ratings of kinematic gait deviation (Beynon et al. 2010),
thus providing evidence that these indices have criterion-related validity relative to
clinician judgments. The authors proposed that the GPS, particularly MAP, might
be useful in clinical practice and education as an adjunct to the traditional presen-
tation of complex kinematic data. It could also be useful as a measure of both the
group and individual outcomes following an intervention or over time. Thus, the
availability of MAP and GVS components together with GPS represents an
advantage of this summary measure with respect to the other measures (Beynon
et al. 2010).
The GPS has strong face validity because it is based on the RMS difference
between gait data for an individual child and the average data from children with no
gait pathology. Analysis of the intra-session variability suggests that it is also a
reliable measure, and the moderate correlation with the NI and a strong relationship
between the GPS and the GDI are based on essentially similar measures of differ-
ence. The GPS shows a very strong nonlinear correlation with the GDI (r = 0.995),
and analysis showed a close mathematical relationship between the GPS and the
unscaled GDI. The GDI and GPS are in fact different ways of scaling the same
underlying construct, and therefore there is little point in using both outcome
measures (Baker et al. 2009). However, there is debate about the use of the GDI
and GPS in clinical practice and research. At present, there are pros and cons for both
indices and choosing one over the other is often based on personal preference. The
choice between the two parameters is related primarily on whether a scaled or
unscaled score is preferred and whether reference to the MAP enhances interpreta-
tion of the results (Baker 2013). Recently, Rasmussen et al. (2015) investigated the
intra-rater reliability and agreement of GDI and GPS (with its GVSs) in a group of
children with CP. They found that GDI and GPS demonstrated excellent reliability
and acceptable agreement proving that they can both be used in research and clinical
practice. However, the observed large variability for some of the GVS requires
cautious consideration when selecting outcome measures. However, the same
authors highlight that the sample size of 18 children considered in their research
might be considered relatively low, and this may impact the results; in addition, the
relatively narrow inclusion criteria for their study might have limited the results of
reliability and the external validity of the study.
Several researchers conducted with the GPS were present in literature, not only in
CP. The outcomes from orthopedic surgery for children (Thomason et al. 2011; Rutz
et al. 2011, 2013; Firth et al. 2013; Ferreira et al. 2014; Tinney et al. 2015; Terjesen
et al. 2015) and adults (Putz et al. 2016) with CP and the effects of ankle-foot
orthosis (Galli et al. 2016) were assessed using GPS. The suitability of GPS and of
other gait summary measures (NI and GDI) for use with lower limb amputees (Kark
et al. 2012), in young hemiplegics (Danino et al. 2013) and in Parkinson’s disease
(Speciali et al. 2014a), was assessed. Kark et al. (2012) found that the GPS, similar to
the NI, detected significant differences between the levels of amputation on the intact
side, while the GDI did not. The differences of the results between GDI and GPS
could be a result of the calculation methods. The GDI is calculated against a matrix
of able-bodied data, whereas the GPS is calculated against a single column of able-
bodied data. This method of calculation may have afforded the GDI a greater
variability and may have been responsible for its failure to detect significant differ-
ences between the levels of amputation on the intact side. In addition, the MAP was
shown to be useful for the elucidation of the underlying causes of gait pathology,
which could not be achieved via the other overall gait summary measures (Kark
et al. 2012). Another element that might justify these different results was obtained
using the GPS and GDI and may be due to the GPS being defined as a raw score,
whereas the GDI is transformed and scaled. Danino et al. (2013) used the GPS, the
NI, and the GDI to evaluate functional electrical stimulation neuroprosthesis as a
method to improve gait in hemiplegic patients. They found that GPS and GDI
provided similar results; however, the decomposition of the GPS score into the
nine Movement Analysis Profile (MAP) scores was found a helpful tool for evalu-
ating gait studies, as it emphasizes the main components that are responsible for the
overall change displayed in their study. On the contrary, the GGI did not detect
significant improvement, although the trend was positive. This result was because
this index has been found to have some shortcomings, including the arbitrary and
unbalanced nature of the 16 parameters that comprise the index as well as difficulties
in its implementation (Schwartz and Rozumalski 2008; Danino et al. 2013). Speciali
et al. (2014a) assessed the effects of subthalamic deep brain stimulation and levo-
dopa on gait in people with Parkinson’s disease using GPS and GDI. They found that
the GPS and GDI showed similar treatment effects, as did GVS for hip and knee
flexion/extension. The overall conclusion of this study is that the GDI, GPS, and
GVS follow the known responses to medication and stimulation of people with
PD. This strongly suggests that the measures are valid for use on people with this
condition and further suggests a general validity of using the measures on
populations other than those for whom they was designed, which was children
with CP.
Applications of GPS were found also in other pathologies, such as multiple
sclerosis (Pau et al. 2014, 2015; Coghe et al. 2015), Ehlers-Danlos syndrome
(Celletti et al. 2013), hereditary spastic paraparesis (Adair et al. 2016), Parkinson’s
disease (Speciali et al. 2014b), and Down syndrome (Galli et al. 2015), showing that
the GPS appears suitable to represent gait deviations from physiological patterns in
these pathological states.
With respect to the other summary measures, the GPS has some advantages.
Previous indices derived from the conventional gait model imposed a considerable
barrier to the extension of similar techniques to data derived from different gait
models or different activities (running, stair climbing, etc.); in contrast, the GPS is
independent of the feature analysis and can be calculated directly from the data of an
individual and the averaged data of people with no gait pathology. Another potential
advantage of the GPS is the deconstruction, which is referred to as the MAP. The
MAP provides useful insights into which variables contribute to the elevated GPS.
The lack of strong correlations of individual GVSs with the GPS and with each other
suggests that there is considerably more information contained within the MAP
compared to the GPS alone. In addition, because the score is derived from the control
database alone, it is easier to compute equivalent scores for different gait models,
with respect to GDI and NI which are based on the outputs of the conventional gait
model.
One of the limits of the GPS is that, similar to the GDI, no spatio-temporal
parameters and kinetics were included in its computation. For the spatio-temporal
parameters, it is important to stress that, because the gait speed is not correlated with
the GPS (Baker et al. 2009), it is recommended that self-selected walking speed
should be reported in addition to the GPS for clinical studies. However, the first
attempt to compute the GVS for kinetics was performed by some authors for the
ankle dorsi-plantar flexion moment and ankle power (Firth et al. 2013).
Movement Deviation Profile (MDP)

As an alternative to conventional analysis of complex data, artificial neural net-
works, specifically the self-organizing map (SOM) described by Kohonen (1988,
2001), have been used to detect deviation from normality in a wide range of
applications. The use of the SOM’s quantization error to quantify the deviation of
gait from normality has been described by Barton et al. (2007) and was illustrated by
examples from patients with CP. Barton et al. (2006) used the power of self-
organizing artificial neural networks or self-organizing map (SOM) in order to
visualize complex gait patterns in the form of single curves. The SOM operates by
converging gait data to stem patterns, which are arranged on a relational map in the
context of the total data space presented to the SOM during training. This method
enables the identification of existing gait patterns and opens up the possibility of
defining new gait patterns, which are otherwise difficult to identify in the
multidimensional data space. This method provides repeatable dimensionality
reduction with a resolution that can be controlled by careful selection of the input
data. The multidimensional ranking of subjects is possible both cross-sectionally and
longitudinally. This method was used to identify differences in lower extremity
coordination between different types of foot orthoses and to assess the gait quality
in a group of patients with various gait problems (Barton et al. 2007).
In a more recent paper, a refined procedure, which generates the Movement
Deviation Profile (MDP) from information provided by the SOM, was proposed
(Barton et al. 2012). The MDP was validated by comparing it to the GDI, FAQ
scores, and clinical diagnoses of patients from a large dataset. In particular, the MDP
represents a single curve showing the deviation of an individual’s movement from
normality. Joint angles, recorded from typically developing children over one gait
cycle, were used to train a SOM, which then generated MDP curves for patients with
gait problems. The mean MDP over the gait cycle showed a high correlation with the
GDI, a statistically significant difference between groups of patients with a range of
functional levels (Gillette Functional Assessment Questionnaire Walking Scale
7–10) and a trend of increasing values for patients with CP through hemiplegia
I–IV, diplegia, triplegia, and quadriplegia. The small difference between the MDP
and GDI can be explained by the SOM’s method of operation comparing biome-
chanical patterns to the nearest abstract reference pattern and its flexibility to
compensate for temporal shifts in movement data. Contrary to the GDI, and similarly
to the GPS, the MDP does not depend on a condensed representation of a large
database containing control and patient data. In addition, while simple averaging of
the MDP can provide a single score, the MDP is similar to the GPS in that it can
provide more detail than a single number expressing deviation from normality in the
same units as the data used (degrees). The MDP could be considered an alternative
method of processing complex biomechanical data, potentially supporting clinical
interpretation. The ability of the MDP and GDI to detect gait changes was compared
in a child with CP (Barton et al. 2015). The GDI showed a symmetric response on
the two opposite sides of normality, but the neural network-based MDP gave an
asymmetric response reflecting faithfully the unequal biomechanical consequences
of joint angle changes. In conclusion, the MDP can detect altered gait even if the
changes are missed by the GDI, and the authors suggested that to complement the
advantages of the GDI, additional use of other gait indices (e.g., MDP) is
recommended.
Summary Measures for Kinetics
Hip Flexor Index (HFI)

Starting from the limitation of the previous summary measures, which included no
kinetic GA data in their computation, the Hip Flexor Index (HFI) was proposed. The
HFI was developed considering the principal component analysis applied to five
kinematic and kinetic variables: maximum pelvic tilt, pelvic tilt range, maximum hip
extension in stance, percentage of stance phase in which the final crossover of the hip
flexor moment curve from extension to flexion occurs, and peak late stance hip flexor
power. The HFI is focused on a single joint/anatomic level, i.e., the hip joint; it
accurately describes, in fact, the overall hip function during gait (Schwartz
et al. 2000). Since the HFI is used to objectify a change in hip function, the main
limit of this index is that it is too joint specific; a change in hip flexor function may or
may not signify a global change in function of the patient. Thus, the HFI is suitable
only for evaluating an intervention at the hip (Novacheck et al. 2002); no other
applications were possible due to the nature of this parameter. For these reasons, the
researchers conducted in literature using this index are limited and mainly focused
on CP (Schwartz et al. 2000; Choi et al. 2011; Lee et al. 2011); to our knowledge
only one experience in another pathological state, the Legg-Calvé-Perthes disease,
was found (Westhoff et al. 2012), without any considerations about the validity of
the index to quantify the gait deviation in this pathology.
GDI-Kinetic
To overcome the limitation of the HFI, the GDI-Kinetic was developed (Rozumalski
and Schwartz 2011). It uses essentially the same methods of GDI to compute an
index based on the kinetic variables. The method identified 20 gait features of the
raw gait kinetic data using singular value decomposition, whose linear combinations
of the first 20 gait features produced a 91 % faithful reconstruction of the data.
Concurrent and face validity for the GDI-Kinetic are presented via comparisons with
the GDI, Gillette Functional Assessment Questionnaire (FAQ) Walking Scale, and
topographic classifications within the diagnosis of CP. The GDI-Kinetic and GDI are
linearly related, but are not strongly correlated, indicating that for any given level of
GDI-Kinetic, there can be a wide variety of kinematic patterns and vice versa,
suggesting that each index measures a different aspect of gait pathology. Similar to
the GDI, the GDI-Kinetic scales with FAQ level distinguish levels from one another.
The GDI-Kinetic also scales with respect to the clinical involvement based on
topographic CP classification in hemiplegia types I–IV, diplegia, triplegia, and
quadriplegia. Interestingly, in hemiplegia, the unaffected limb exhibits lower
GDI-Kinetic scores than the affected side, indicating that compensations in the
unaffected limb result in greater deviations from normal gait than those observed
in the affected limb (Rozumalski and Schwartz 2011). The GDI-Kinetic was able to
complement the GDI with a more comprehensive measure of gait pathology, includ-
ing not only kinematics but also kinetics. However, to the best of our knowledge,
only few applications of this index are available in the literature. Brostrom
et al. (2013) used GDI and GDI-Kinetic to quantify the impact of anti-tumor necrosis
factor-alpha inhibitor treatment on gait dynamics in patients with rheumatoid arthri-
tis; they showed that GDI and GDI-Kinetic scores appeared to be useful outcome
measures to quantify changes in gait deviations after this intervention. Kiernan
et al. (2015, 2016) used the GPS and GDI-Kinetic to investigate the clinical
agreement of different regression equations based on pelvic anatomy routinely
used to estimate the hip joint center during gait analysis. In another study, the effect
of using different anthropometric sets on predicted sagittal plane moments during
normal and diplegic cerebral palsy gait was investigated using GDI-Kinetic (Kiernan
et al. 2014).
Conclusions
In this chapter, an overview of the most used summary measures to express the
deviation of an individual’s gait from normal walking pattern is provided. According
to the literature found on this topic, to their clinical applications, and to personal
experience, the summary measures could represent a useful tool mainly in clinical
settings, to objectify clinical impression, quantify the degree of gait deviation from
normal, stratify the severity of pathology, document the changes in gait patterns over
time, and evaluate interventions. However, it is common opinion that they should
always be used in conjunction with all of the other specific information represented
inside the GA report, i.e., spatio-temporal parameters, kinematics, kinetics, and
EMG data. The simple use of summary measures alone provides in fact no indica-
tions about the interaction among joints and the interaction among the movement
planes; information about the interaction among kinematics, kinetics, and EMG; and
the assessment of copying response. Then, their use does not extend to identifying
the cause of gait deviations, as the single score characterizes the patient’s gait as a
whole and cannot be related to certain phases of the gait cycle or to anatomical
locations (Barton et al. 2007). Another limitation of the summary measures to
quantify the gait patterns of a subject or the effects of a treatment is that favorable
and adverse changes might be masked when using a single number; only an in-depth
analysis of GA plots is able to give the correct overview of the patient’s gait pattern
(Cimolin and Galli 2014).
References
Adair B, Rodda J, McGinley JL et al (2016) Kinematic gait deficits at the trunk and pelvis:
characteristic features in children with hereditary spastic paraplegia. Dev Med Child Neurol
58:829–835
Andersson C, Mattsson E (2001) Adults with cerebral palsy: a survey describing problems, needs,
and resources, with special emphasis on locomotion. Dev Med Child Neurol 43:76–82
Assi A, Ghanem I, Lavaste F et al (2009) Gait analysis in children and uncertainty assessment for
Davis protocol and Gillette Gait Index. Gait Posture 30:22–26
Baker R (2013) Measuring walking: a handbook of clinical gait analysis. Mac Keith Press, London
Baker R, McGinley JL, Schwartz MH et al (2009) The gait profile score and movement analysis
profile. Gait Posture 30:265–269
Baker R, McGinley JL, Schwartz M et al (2012) The minimal clinically important difference for the
Gait Profile Score. Gait Posture 35:612–615
Barton G, Lees A, Lisboa P et al (2006) Visualisation of gait data with Kohonen self-organising
neural maps. Gait Posture 24:46–53
Barton G, Lisboa P, Lees A et al (2007) Gait quality assessment using self-organising artificial
neural networks. Gait Posture 25:374–379
Barton GJ, Hawken MB, Scott MA et al (2012) Movement deviation profile: a measure of distance
from normality using a self-organizing neural network. Hum Mov Sci 31:284–294
Barton GJ, Hawken MB, Holmes G et al (2015) A gait index may underestimate changes of gait: a
comparison of the Movement Deviation Profile and the Gait Deviation Index. Comput Methods
Biomech Biomed Engin 18:57–63
Beynon S, McGinley JL, Dobson F et al (2010) Correlations of the gait profile score and the
movement analysis profile relative to clinical judgments. Gait Posture 32:129–132
Brehm MA, Harlaar J, Schwartz M (2008) Effect of ankle-foot orthoses on walking efficiency and
gait in children with cerebral palsy. J Rehabil Med 40:529–534
Broström EW, Esbjörnsson AC, von Heideken J et al (2013) Change in Gait Deviation Index after
anti-tumour necrosis factor-α treatment in individuals with rheumatoid arthritis: a pilot study.
Scand J Rheumatol 42:356–361
Caskey PM, McMulkin ML, Gordon AB et al (2014) Gait outcomes of patients with severe slipped
capital femoral epiphysis after treatment by flexion-rotation osteotomy. J Pediatr Orthop
34:668–673
Celletti C, Galli M, Cimolin V et al (2013) Use of the gait profile score for the evaluation of patients
with joint hypermobility syndrome/Ehlers-Danlos syndrome hypermobility type. Res Dev
Disabil 34:4280–4285
Choi SJ, Chung CY, Lee KM et al (2011) Validity of gait parameters for hip flexor contracture in
patients with cerebral palsy. J Neuroeng Rehabil 8:4
Cimolin V, Galli M (2014) Summary measures for clinical gait analysis: a literature review. Gait
Posture 39:1005–1010
Cimolin V, Galli M, Vimercati SL et al (2011) Use of the Gait Deviation Index for the assessment of
gastrocnemius fascia lengthening in children with Cerebral Palsy. Res Dev Disabil 32:377–381
Coghe G, Pau M, Corona F et al (2015) Walking improvements with nabiximols in patients with
multiple sclerosis. J Neurol 262:2472–2477
Cretual A, Bervet K, Ballaz L (2010) Gillette Gait Index in adults. Gait Posture 32:307–310
Danino B, Khamis S, Hemo Y et al (2013) The efficacy of neuroprosthesis in young
hemiplegic patients, measured by three different gait indices: early results. J Child Orthop
7:537–542
Domagalska M, Szopa A, Syczewska M et al (2013) The relationship between clinical measure-
ments and gait analysis data in children with cerebral palsy. Gait Posture 38:1038–1043
Dreher T, Vegvari D, Wolf SI et al (2012) Development of knee function after hamstring length-
ening as a part of multilevel surgery in children with spastic diplegia: a long-term outcome
study. J Bone Joint Surg Am 18:121–130
Esbjörnsson AC, Rozumalski A, Iversen MD et al (2014) Quantifying gait deviations in individuals
with rheumatoid arthritis using the Gait Deviation Index. Scand J Rheumatol 43:124–131
Eshraghi A, Abu Osman NA, Karimi M et al (2014) Gait biomechanics of individuals with
transtibial amputation: effect of suspension system. PLoS One 9:e96988
Ferreira LA, Cimolin V, Costici PF et al (2014) Effects of gastrocnemius fascia lengthening on gait
pattern in children with cerebral palsy using the gait profile score. Res Dev Disabil
35:1137–1143
Firth GB, Passmore E, Sangeux M et al (2013) Multilevel surgery for equinus gait in children with
spastic diplegic cerebral palsy: medium-term follow-up with gait analysis. J Bone Joint Surg Am
95:931–938
Galli M, Cimolin V, De Pandis MF et al (2012a) Use of the Gait Deviation index for the evaluation
of patients with Parkinson’s disease. J Mot Behav 44:161–167
Galli M, Ferrario D, Patti P et al (2012b) The use of 3d motion analysis in a patient with an atypical
juvenile neuronal ceroid lipofuscinoses phenotype with CLN1 mutation and deficient PPT
activity. J Dev Phys Disabil 24(2):155–165
Galli M, Cimolin V, Rigoldi C et al (2015) Use of the gait profile score for the quantification of gait
pattern in down syndrome. J Dev Phys Disabil 27:609–615
Galli M, Cimolin V, Rigoldi C et al (2016) Quantitative evaluation of the effects of ankle foot
orthosis on gait in children with cerebral palsy using the gait profile score and gait variable
scores. J Dev Phys Disabil 28:367–379
Garbelotti SA Jr, Lucareli PR, Ramalho A Jr et al (2014) An investigation of the value of
tridimensional kinematic analysis in functional diagnosis of lumbar spinal stenosis. Gait Posture
40:150–153
Kark L, Vickers D, McIntosh A et al (2012) Use of gait summary measures with lower limb
amputees. Gait Posture 35:238–243
Kiernan D, Walsh M, O’Sullivan R et al (2014) The influence of estimated body segment
parameters on predicted joint kinetics during diplegic cerebral palsy gait. J Biomech
47:284–288
Kiernan D, Malone A, O’Brien T et al (2015) The clinical impact of hip joint centre regression
equation error on kinematics and kinetics during paediatric gait. Gait Posture 41:175–179
Kiernan D, Hosking J, O’Brien T (2016) Is adult gait less susceptible than paediatric gait to hip joint
centre regression equation error? Gait Posture 45:133–136
Kohonen T (1988) Self-organisation and associative memory. Springer, Berlin
Kohonen T (2001) Self-organizing maps. Springer, Berlin
Lee KM, Chung CY, Kwon DG et al (2011) Reliability of physical examination in the measurement
of hip flexion contracture and correlation with gait parameters in cerebral palsy. J Bone Joint
Surg Am 93:150–158
Lee SY, Kwon SS, Chung CY et al (2014) Rectus femoris transfer in cerebral palsy patients with
stiff knee gait. Gait Posture 40:76–81
Massaad A, Assi A, Skalli W et al (2014) Repeatability and validation of gait deviation index in
children: typically developing and cerebral palsy. Gait Posture 39:354–358
McMulkin ML, MacWilliams BA (2008) Intersite variations of the Gillette Gait Index. Gait Posture
28:483–487
McMulkin ML, Gordon AB, Caskey PM et al (2016) Outcomes of orthopaedic surgery with and
without an external femoral derotational osteotomy in children with cerebral palsy. J Pediatr
Orthop 36:382–386
Novacheck TF, Trost J, Schwartz MH (2002) Intramuscular psoas lengthening improves dynamic
hip function in children with cerebral palsy. J Pediatr Orthop 22:158–164
Pau M, Coghe G, Atzeni C et al (2014) Novel characterization of gait impairments in people with
multiple sclerosis by means of the gait profile score. J Neurol Sci 345:159–163
Pau M, Coghe G, Corona F et al (2015) Effect of spasticity on kinematics of gait and muscular
activation in people with Multiple Sclerosis. J Neurol Sci 358:339–344
Putz C, Döderlein L, Mertens EM et al (2016) Multilevel surgery in adults with cerebral palsy. Bone
Joint J 98-B:282–288
Rasmussen HM, Nielsen DB, Pedersen NW, Overgaard S, Holsgaard-Larsen A (2015) Gait
Deviation Index, Gait Profile Score and Gait Variable Score in children with spastic cerebral
palsy: Intra-rater reliability and agreement across two repeated sessions. Gait Posture
42:133–137
Rosenlund S, Holsgaard-Larsen A, Overgaard S et al (2016) The Gait Deviation Index is associated
with hip muscle strength and patient-reported outcome in patients with severe hip osteoarthritis-
a cross-sectional study. PLoS One 11:e0153177
Rozumalski A, Schwartz MH (2011) The GDI-Kinetic: a new index for quantifying kinetic
deviations from normal gait. Gait Posture 33:730–732
Rutz E, Baker R, Tirosh O et al (2011) Tibialis anterior tendon shortening in combination with
Achilles tendon lengthening in spastic equinus in cerebral palsy. Gait Posture 33:152–157
Rutz E, Donath S, Tirosh O et al (2013) Explaining the variability improvements in gait quality as a
result of single event multi-level surgery in cerebral palsy. Gait Posture 38:455–460
Schutte LM, Narayanan U, Stout JL, Selber P, Gage JR, Schwartz MH et al (2000) An index for
quantifying deviations from normal gait. Gait Posture 11:25–31
Schwartz MH, Rozumalski A (2008) The Gait Deviation Index: a new comprehensive index of gait
pathology. Gait Posture 28:351–357
Schwartz MH, Novacheck TF, Trost J (2000) A tool for quantifying hip flexor function during gait.
Sienko Thomas S, Buckon CE, Nicorici A et al (2010) Classification of the gait patterns of boys
with Duchenne muscular dystrophy and their relationship to function. J Child Neurol
25:1103–1109
Speciali DS, Oliveira EM, Cardoso JR et al (2014a) Gait profile score and movement analysis
profile in patients with Parkinson’s disease during concurrent cognitive load. Braz J Phys Ther
18:315–322
Speciali DS, Corrêa JC, Luna NM et al (2014b) Validation of GDI, GPS and GVS for use in
Parkinson’s disease through evaluation of effects of subthalamic deep brain stimulation and
levodopa. Gait Posture 39:1142–1145
Syczewska M, Dembowska-Bagińska B, Perek-Polnik M et al (2010) Gait pathology assessed with

Gillette Gait Index in patients after CNS tumour treatment. Gait Posture 32:358–362
Szopa A, Domagalska-Szopa M, Czamara A (2014) Gait pattern differences in children with
unilateral cerebral palsy. Res Dev Disabil 35:2261–2266
Terjesen T, Lofterød B, Skaaret I (2015) Gait improvement surgery in ambulatory children with
diplegic cerebral palsy. Acta Orthop 86:511–517
Thomason P, Baker R, Dodd K et al (2011) Single-Event multilevel surgery in children with spastic
diplegia: a pilot randomized controlled trial. J Bone Joint Surg Am 93:451–460
Tinney A, Thomason P, Sangeux M et al (2015) The transverse Vulpius gastrocsoleus recession for
equinus gait in children with cerebral palsy. Bone Joint J 97-B:564–571
Trost JP, Schwartz MH, Krach LE et al (2008) Comprehensive short-term outcome assessment of
selective dorsal rhizotomy. Dev Med Child Neurol 50:765–771
Westhoff B, Martiny F, Reith A et al (2012) Computerized gait analysis in Legg-Calvé-Perthes
disease – analysis of the sagittal plane. Gait Posture 35:541–546
Interpreting Spatiotemporal Parameters,
Symmetry, and Variability in Clinical Gait
Analysis
Arnaud Gouelle and Fabrice Mégrot
Abstract
Spatiotemporal parameters (STP) are widely studied variables in clinical gait
analysis. Yet they often remain underutilized despite the rich information they
provide about organization and control of the patient’s progress. Building on
them requires a broad knowledge of the “normal” gait, before to being able to
understand the impact of pathological disorders. We hope to provide information
to better grasp and understand the STP while highlighting important points.
Through this chapter, we will introduce basics of the gait cycle, before
considering the components for which the STP may be informative: rhythm,
pace, phases, postural control, asymmetry, and variability. We will define main
parameters for each component and discuss their use regarding state of the art.
Then factors influencing STP will be addressed to understand how these param-
eters change during life, when a child learns to walk or when the advance in
age-affected gait in the elderly, as well as the influence of diseases. Indeed,
various pathologies affect the walk, and the most relevant STP are not always
the same. We will consider Friedreich ataxia, which is a neurodegenerative
disease, in which combination of cerebellar, pyramidal syndromes, and axonal
neuropathy cause a rapid degeneration of the walking ability and therefore lead to
various observable gait patterns. We will also illustrate how PST can be useful to
A. Gouelle (*)
Gait and Balance Academy, ProtoKinetics, Havertown, PA, USA
e-mail: arnaud.gouelle@gmail.com
F. Mégrot
Unité Clinique d’Analyse de la Marche et du Mouvement, Centre de Médecine Physique et de
Réadaptation pour Enfants de Bois-Larris – Croix-Rouge Française, Lamorlaye, France
UMR CNRS 7338: Biomécanique et Bioingénierie, Sorbonne Universités, Université de
Technologie de Compiègne, Compiègne, France
e-mail: fabrice.megrot@croix-rouge.fr

DOI 10.1007/978-3-319-30808-1_35-1
2 A. Gouelle and F. Mégrot
document the most appropriate time for a patient to change from one assistive
device to another.
The final portion will aim to give paths for clinical interpretation while
thinking about the concepts of limitation and adaptation.
Keywords
Clinical gait • Spatiotemporal • Interpretation • Variability • Symmetry
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Basics for Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Rhythm: Variables Reflecting Gait Rhythm and Absolute Timing . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Pace: Parameters Related to Speed and/or Measures of Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Phases: Duration of Gait Phases Relative to Gait Cycle Time (% GC) . . . . . . . . . . . . . . . . . . . . . 7
Postural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Asymmetry: Differences Between Limbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Variability: Fluctuations of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Factors Influencing the Spatiotemporal Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Natural Evolution During the Life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Influence of the Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Introduction
Spatiotemporal parameters (STP) are one of the most studied variables in clinical
gait analysis. Yet they often remain underutilized despite the rich information they
can provide about the organization and control of the progress of the patient. At first
sight, gait parameters are simple to formulate and easy to calculate. Building on them
requires a broad knowledge of the “normal” gait, before to being able to understand
the impact of pathological disorders. Indeed, behind their apparent simplicity is
hidden all of the complexity to read parameters that are intrinsically linked to each
other that depend on many personal (e.g., age, sex) and environmental (e.g., record-
ing protocol, computational algorithms) factors. For example, taking into account
only the walking speed to assess the results of a rehabilitation protocol will have little
meaning because the speed is dependent on both cadence and stride length. Follow-
ing the evolution of the gait of a child without considering the growth will inevitably
lead to misinterpretation. Similarly, assessing only mean values will obscure an
important part of the information on disturbances and regulations, visible through
the variability.
Analyzing STP requires experience and training, as well as knowledge of the
factors that influence them. We hope this chapter provides information to better grasp
and understand the spatiotemporal parameters while highlighting important points.
Interpreting Spatiotemporal Parameters, Symmetry, and Variability in. . . 3
0% 10% 50% 60% 100%
Initial Terminal
Single support Swing
double double
support support
Stance phase Swing phase
GAIT CYCLE FOR THE RIGHT LIMB
Fig. 1 Different phases of gait cycle illustrated for the right lower limb
We first discuss the basics of the gait cycle, before considering each of the compo-
nents for which the STP may be informative. Then we will see how these parameters
change during life or are affected by disease. The final portion will aim to give the
reader paths for clinical interpretation.
Basics for Gait Analysis
Gait Cycle (See Kirtley 2006; Perry and Burnfield 2010; Whittle 2007)
The gait cycle (Fig. 1), defined as the time from the occurrence of a particular event –
by convention the initial foot contact with the ground – until the next occurrence of
the same event for the same lower limb, is the fundamental unit of gait. While this
term may be an important parameter itself, the gait cycle is usually standardized so
that the initial contact of the foot is at 0% (beginning of cycle), and the next contact
of the same foot is 100% (end of cycle). This process of time normalization makes it
easier to compare between individuals, within individuals, and between cycles of the
two limbs. The detailed description of events and phases that takes place within a
gait cycle depends on the approach to the work.
The simplest approach subdivides the cycle according to the periods of contact of
the feet with the ground, distinguishing two phases, the stance phase and the swing
phase, which alternate for each limb during walking. This involves adding an
additional event to the gait cycle, when the foot leaves the ground. A gait cycle is
composed of a support phase (0–60% of the cycle) followed by a swing phase
(60–100% of the cycle) for each left and right lower limbs.
The observation of both spatial and temporal characteristics of lower limbs allows
the introduction of additional phases. When only one limb is supported, it is called
single support (10–50%), while the second limb is in swing phase. When both
members are in stance phase, we speak about double support. The double support
is often analyzed as an overall phase (20%), but in reality, it splits in a first double
support at the time of loading (from 0% to 10% of the cycle) and a second at the end
of support (50–60% of the cycle, often called pre-swing phase).
Reference Gait
The results of this decomposition of the gait cycle are spatial and temporal param-
eters. Sensitive by nature to walking speed and of course to the experimental
settings, it is usual to study the STP in a framework as simple as possible. The
reference gait consists of walking in healthy adults at spontaneous and stable state,
that is to say, not taking into account either the acceleration phase or deceleration
phase. It is preferably registered in a straight line and is not obstructed by obstacles.
Obviously, this is not totally representative of a walk in everyday life, made up of
adaptations and changes of pace, but is needed for the clinical use of STP. Other
protocols can then be set up to identify other issues (walking at fast or slow rate,
adding obstacles, turn around, double task, etc.) but we will not discuss them in this
chapter.
State of the Art
While many STP can be found in the literature, they are all derived from a few basic
parameters, reflecting the spatiality and the temporality of ground footwork.
Depending on the methods of calculation, the systems used and on the object of
study, other variables are derived from the raw data, including the temporal and
stature standardization, ratio computation, and measures of dispersion. To simplify
the interpretation, it is more convenient to group STP which reflect a same aspect of
the gait and to focus on the individual analysis of each component before linking all.
Six domains can be considered based on the covariance between parameters:
rhythm, pace, phases, postural control, asymmetry, and variability (see Verghese
et al. 2007; Verlinden et al. 2013; Lord et al. 2013, 2014).
Rhythm: Variables Reflecting Gait Rhythm and Absolute Timing
Except for cadence and walk ratio, these parameters are the raw time values that
reflect the walk of the patient. They are not directly used in this raw form, in seconds
or milliseconds, which provides little information and of course is directly related to
the frequency of locomotion. Analysis of the relative durations of each phase as a
percentage of the total duration of the gait cycle is common and will be addressed in
the component (see Phases: Duration of Gait Phases Relative to Gait Cycle Time (%
GC)).
Rh1. Single support (s): period of time when only the current foot is in contact with
the ground.
Rh2. Swing time (s): period of time while the foot is not in contact with the ground;
is exactly equivalent as single support of the contralateral limb for the consid-
ered cycle.
Rh3. Stance time (s): period of time when the foot is in contact with the ground.
Rh4. Double support (s): initial and final parts of the stance phase when two feet are
in contact with the ground; can be considered as a global time or differentiated
as initial and terminal double support.
Rh5. Step time (s): period of time taken for one step measured from first contact of
one foot to the first contact of following other foot.
Rh6. Stride time (s): total amount of time for the cycle; equivalent to two successive
step time; this duration is used to express gait phases in a relative form (%).
Rh7. Cadence (steps/min): number of steps taken in a given time; the usual clinical
units being steps per minute.
Rh8. Walk ratio: step length divided by cadence.
The cadence is given in steps per minute or in cycles per minute. Mathematically,
cadence, which is a frequency (number of events per second), is calculated as the
inverse of the cycle. It is then multiplied by a factor 60 to obtain a number of cycles
per minute, or a factor 120 for the number of steps per minute.
The spontaneous cadence is usually between 98–138 steps/min for women and
91–135 steps/min for men 18–49 years old (Whittle 2007). Women offset a smaller
step length with a higher cadence. Generally, an increased cadence is found,
regardless of the gender, in adults smaller than the average population.
To normalize the cadence, it is advised to use the formula proposed by Hof
(1996), taking into account the leg length and the acceleration g of gravity.
cadence ðsteps=minuteÞ
120
Normalized cadence φ ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g
lower limb length ðmetersÞ
The walk ratio represents the relationship between the amplitude and the fre-
quency of movement of the legs and is calculated as the mean step length divided by
the cadence. In adults, it is relatively invariant through a speed range from very slow
to very fast, that is, independent of the speed (Sekiya et al. 1996). Walking with an
invariant walk ratio would be optimal in terms of energy expenditure, temporal
variability, spatial variability, and attentional demand. It is, therefore, a particularly
interesting parameter for the longitudinal monitoring of a patient, especially during
rehabilitation, which provides information on the rhythmic organization of the gait.
The walk ratio, obtained from the step length in centimeters and the cadence in
steps/minute, has a mean of 0.58 (0.06) in adults and decreases when the person
tends to walk with fast small steps. Attention must be paid because these reference
values change depending on the parameters (step length, cycle length), the units used
(cm, m), and the possible standardization variables (leg length, height).
normalized step length λ

Normalized walk ratio ¼
normalized cadence φ
Pace: Parameters Related to Speed and/or Measures of Length
The pace domain includes the parameters related to walking speed and displacement
in the sagittal plane.
Pa1. Stride length (m or cm): the distance travelled by a person during one stride
(or cycle); can be measured as the length between the rearmost point of the
footprint (often heel) from one (heel) strike to the next (heel) strike on the
same side.
Pa2. Step length (m or cm): distance between corresponding successive heel points
of opposite feet, measured parallel to the direction of progression for the
ipsilateral stride.
Pa3. Speed (m/s or cm/s): covered by the whole body in a given time
The stride length is the distance related to the gait cycle, but, in practice, the step
length proves more informative because it assesses the symmetry between the two
lower limbs. Obviously, the length is directly related to the lower limb length, and
therefore it is particularly important to standardize the length to monitor the progress
of a child/teenager. The step length for a side is partly dependent on the contralateral
support because if this one is deficient, the foot will land to the ground faster and
with less distance covered. Thus, it is often reduced in pathologies affecting the gait,
while the cadence is increased to maintain a certain speed.
step length ðmetersÞ

Normalized step length λ ¼
lower limb length ðmetersÞ
The step length is commonly computed according to the anteroposterior axis of

progression. Hence, it is possible to observe a length of zero or negative if the rear
foot is not brought beyond the front foot. It is also possible to focus on the overall
travel length of the foot in space, often called raw step length, which is the real
distance between two successive contacts (takes into account both anteroposterior
and lateral distance).
The speed represents the overall performance of walking and is regarded as the
sixth vital sign. It may simply be calculated as the distance traveled divided by the
time required; however, it is also the product of step length and cadence.
In other words, it is possible to produce the same speed via multiple configura-
tions ranging from fast small steps to slow long steps. Moreover, while the cadence
increases linearly, the step length, which is more constrained by the physical aspect,
increases logarithmically, changing greatly at low speed, but tending to stabilize at
higher speeds. Only considering walking speed is not enough to properly analyze the
progress of a subject over time.
A normal walking speed can thus result from adequate cadence and step length
(healthy gait), as well as from small steps offset by increased cadence. If the
compensation is insufficient, the speed will also be reduced.
Everyone has a preferential, spontaneous walking speed or rather a comfort speed
zone, determined within plus or minus 1 km/h, in which there is no significant
difference in energy cost. We can agree to a speed range of between 1.3 m/s and
1.6 m/s for adults. We must also keep in mind that being able to significantly change
its walking speed, when necessary, affects the ability of a subject to adapt to different
situations that may have to be managed during the daily life. The speed difference
between the spontaneous pace and the rapid pace can thus be seen as a marker of
functional reserves of a person and adaptability.
The normalized speed can be obtained simply as the product of the normalized
step length by the normalized cadence.
Normalized speed β ¼ normalized step length λ 2 normalized cadence φ
The factor 2 is due to the initial use of the raw cadence in steps per minute. If
cadence in cycles per second is used, no factor intervenes (β = λ ϕ).
Phases: Duration of Gait Phases Relative to Gait Cycle Time (% GC)
Once expressed as a percentage of the cycle, the proportions of the various phases
become easier to exploit; however, they vary again with walking speed (Fig. 2).
Ph1. Stance phase (%)

Ph2. Single support (%)
Ph3. Swing time (%)
Ph4. Double support (%): can be differentiated in initial and terminal double support
The stance phase includes the initial double support, single support, and terminal
double support. It represents an overall component of the support of a foot on the
ground. Its duration at spontaneous speed is about 60% of the gait cycle. It decreases
with increasing speed, increasing the duration of the swing phase. Swing phase
represents 40% of the gait cycle when the leg is no longer in contact with the ground
and is brought forward.
The stance phase is often extended when the subject has a balance problem. When
trouble in walking specifically affects one side (hemiplegia, prosthesis, pain), stance
is shortened on the (most) affected foot and increased on the other limb. It should
also be noted that the stance phase is shorter during walking with shoes compared to
walking barefoot. Shoes create an area of support slightly larger and improve
balance (Eisenhardt et al. 1996).
Fig. 2 Relative proportion of the gait phases depending on the walking pace. These data come
from a single healthy adult (personal data) who walked at various paces on a gait mat and are given
to illustrate the influence of the walking speed on the duration of the phases. In running, there are no
periods when both feet are in contact with the ground. The time spent in stance depends on runner’s
level (for more details, see Novacheck 1998)
The most relevant information is given by the ratio between the single support
phase (40%) and the two double support phases (twice 10%). First because these two
parameters are sufficient to calculate all phases, single support phase of a limb being
obviously equivalent to the swing phase of the contralateral limb. When walking is
symmetrical, reducing the single support phase increases the proportion of double
support. This is the case when a person has to walk more slowly than its spontaneous
speed and/or when there is a problem with dynamic balance. Increased time spent on
two limbs allows more time to control the center of mass. On the contrary, if asked to
walk faster, the double support will be reduced in favor of the single support phase.
On the other hand, although the double support phase is often considered as a
whole, distinguishing the initial double and terminal double support allows you to
specify a problem during the transfer from one side to another.
Postural Control
This domain is mainly related to lateral movements.
Po1. Step width or base of support (cm): space between the feet during walking.
(Note that calculation methods can differ).
Po2. Step width variability (cm or %): standard deviation or coefficient of variation
of the base of support.
Po3. Step length asymmetry (cm): step length difference between the two lower
limbs expressed in centimeters or under the form of a ratio. See also Asym-
metry domain.
Po4. Foot angle ( ): angle of rotation during stance.
The base of support is usually between 8 and 12 cm in children and adults and
wider in the toddlers (when normalized by the pelvis width) and in the elderly. In the
presence of problems of dynamic balance, expanding the base of support is a strategy
to better control the walking and reduce the risk of falls. This can be effective, or not,
but always indicates the presence of a balance disorder. A negative base of support
can also be found in the presence of crossing steps as in cerebellar or ataxic gait, for
example. On the other hand, in a healthy subject, the step width decreases when
asked to walk faster than the spontaneous speed.
The variability of the base of support provides information about the dispersion of
this parameter during walking. In other words, how the width between the feet varies
from one step to another, which is measured through the standard deviation or the
coefficient of variation. Extreme variability (i.e., too high or too low) was associated
with history of falls in the elderly walking at a normal or near normal speed (>1 m/s)
(Brach et al. 2005). Individuals unable to change their stride width (i.e., low
variability) would experience more difficulties adapting to maintain balance, while
excessive variability (often associated with crossing feet) could indicate a lack of
compensation for instability. In the literature, most authors use the standard devia-
tion to quantify the variability of the stride width, but we must be careful in reading
the results when the coefficient of variation is preferred. Indeed, this relative
parameter has mathematical disadvantage to tend toward infinity while, its mean is
close to 0. In other words, with the same standard deviation variance, coefficient of
variation will be greater if the base of support is narrow and smaller if the support
base is wide (Fig. 3). Hence, it is recommended that coefficient of variation should
only be calculated for ratio data. Since step width is interval data, standard deviation
has to be used (Atkinson et al. 1998; Paterson et al. 2009).
The asymmetry of step length plays a role in postural balance since this difference
causes an imbalance in the management of lateral imbalances, especially because the
step length and stride width are geometrically linked. For example, Bril and Brénière
(1992) have shown that two-thirds of the increase in step length, observed during the
first months of walking of a child, could be explained by the decrease in the stride
width.
The foot angle (about 0–15 in control adult) reflects the position of the foot
(in abduction, adduction, or neutral position) during stance. This is a very personal
characteristic, which is related primarily to the motor habits and to the bone
architecture of the lower limb. Differences less than or equal to 5 between the
two feet can be considered as normal. Excessive internal and external rotations are
commonly seen in cerebral palsy subjects, where they derive from architectural
defects that appear gradually during growth (rotational abnormalities of the femoral
and tibialis segment, fixed deviations or irreducible foot in varus adductus or valgus
abductus).
260
Coefficient of Variation for base of support (%)

240
220
200
180
160
140
120
100
80
60
40
20
0
0 5 10 15 20 25 30
Mean for base of support (cm)
Fig. 3 Relationship between the mean and the coefficient of variation of the base of support for a
healthy subject walking at the same speed through several trials, but modifying the space between
his feet (personal data)
Beware, however, because the validity and reproducibility of this parameter is

dependent on the full contact of the foot to the ground. For example, if walking in
equinovarus, with exclusive forefoot pose, the angle given by a system based on
pressure sensors will be incorrect, often with an overestimation of internal rotation.
By indicating the measures of the foot (length and width), some software can give an
indication of the effective area in contact with the ground, which then gives an idea
of the validity of the measure of foot rotation.
Asymmetry: Differences Between Limbs
Asymmetry is the domain linked to differences between the right and left lower
limbs parameters.
As1. Temporal asymmetry

As2. Spatial asymmetry
The spatiotemporal asymmetry is a measure of the quality of the walking pattern

and can be seen as an important aspect in the gait analysis. Especially because it
increases energy costs and is in line with the dynamic balance deficits. The asym-
metry can affect gait in a spatial point of view (e.g., unequal step lengths between
right and left) and temporally (e.g., difference in time spent in swing/stance phase
between the two feet).
The main challenge is to choose the adequate calculation of asymmetry

depending on what one wants to assess, which often depends on the type of disease.
Generally, the easiest way is to use the difference between the two sides (e.g.,
left–right), either with raw or absolute values. In both cases, a difference of 0 repre-
sents a perfect symmetry. With raw difference, the value indicates the direction of the
asymmetry, while the absolute difference gives only the magnitude of asymmetry. To
compare successive evaluations or several subjects, it is better to use parameters
already standardized.
The second possibility is to calculate a ratio (e.g., left/right). A ratio of 1 repre-
sents perfect symmetry. Different equations can be used in this case (symmetry ratio,
symmetry index, logarithmic transformation of the ratio, angle of symmetry).
Patterson et al. (2010) found strong correlations between these equations and
concluded that none had more advantages over others. Thus, they proposed to use
a symmetry ratio that is easier to interpret (no paretic limb/paretic limb, or the
reverse).
Notice, if the symmetry is often observed in the asymptomatic population, it is
possible to observe small asymmetries of parameters in controls. It is difficult to give
a single asymmetry value beyond which it would exceed what can be observed
naturally. Therefore, Patterson et al. (2010) proposed to consider asymmetric indi-
viduals whose asymmetry is beyond the 95% confidence interval in the healthy
population. In other words, asymmetry thresholds for ratio (high value/small value)
were determined for the step length (1.08), the duration of the stance phase (1.05),
and the duration of the double support (1.04). Beyond, gait can be considered
asymmetric for the considered parameter.
Most often, the subject seeks to reduce support on the most affected lower limb,
and we observed:
– Increased swing time for the paretic side and/or decreased for the non-paretic
– Decreased stance time for the paretic side and/or increased for the non-paretic
– Shortened step length on non-paretic side
For more thinking about this, an excellent review of literature on the gait
asymmetry in stroke patients has been published by Lauzière et al. (2014).
Variability: Fluctuations of Parameters
Va1. Temporal variability (standard deviation or coefficient of variation)

Va2. Spatial variability (standard deviation or coefficient of variation)
Va3. Gait variability Index (composite measure)
Gait variability, defined as the fluctuation in spatiotemporal characteristics between

steps, is a sensitive indicator of mobility deficits. For example, variability in spatio-
temporal parameters is reported to predict mobility deficits and future falls better
than the mean of spatiotemporal parameters in older adult.
Fluctuation magnitude, as defined by Hausdorff (2005) in comparison to fluctu-

ation dynamics, is computed through measure of dispersion, as standard deviation or
coefficient of variation. With exception of only a few percent of healthy adults,
variation increased with age, disabilities, and fall risk, as well as during walking in
unstable condition or when subject has to perform a dual task.
One problem is that the gait variability measurement raises a question about
methodological challenges. First, it is unclear which spatiotemporal measures are of
greatest importance when assessing gait variability. Variability has been reported for
at least 11 spatiotemporal parameters, but it is unclear which are most relevant to
mobility function and the deficits that they reflect. How to deal with interdependence
of the parameters? With spatial versus temporal variability? Has the variability to be
considered as a global amount or should two limbs be considered separately? How to
consider the step width variability? All these questions have to be considered as well
as the way to quantify variability, but once again there is a lack of consensus
regarding how best to do it. Standard deviation is sensitive to the scale and coeffi-
cient of variation goes toward high values when the mean is around 0.
Other questions involve how to consider levels of variability. It could be argued
that variability represents noise and so just high levels of variability are needed to be
observed. However, a minimum level is required to ensure ability to regulate step-to-
step variations.
A composite score has been proposed to be used as a unique value, the gait
variability index (Gouelle et al. 2013). Based on nine spatiotemporal parameters
weighted by PCA, it quantifies the distance between the amount of variability
observed for a reference group and the amount of variability observed for an
individual. To enhance applicability, GVI is transformed into a score with 100 and
10 representing, respectively, the mean score and the standard deviation for a
reference group. GVI above 100 indicates that the individual has a similar level of
variability as the reference group. For GVI <100, each 10-point difference corre-
sponds to a separation of 1 SD from the reference group score.
Variability is an area of importance for the assessment of walking because two
walking pattern can be identical in terms of mean spatiotemporal parameters and
asymmetry but present completely different levels of variability. In this case, the
variability allows to decide, for example, on the interest of a strategy to minimize the
risk of falls or interest for an assistive device.
Factors Influencing the Spatiotemporal Parameters
Natural Evolution During the Life
Before assessing the progress of a patient and the influence of pathology, it is

necessary to know the standard values of spatiotemporal parameters and how they
evolve naturally in life.
Several years are needed between the first steps of a child and the achievement of
a fully mature gait. While in the first years of independent walking, a number of
parameters related to the gait cycle progress toward adult profile (Sutherland 1997).
The constraints for dynamic balance during walking will require a longer time before
being fully controlled. The development of dynamic balance strategies involves the
mastery of anticipation functions and coupling between the different mobilized
joints. It, therefore, continues into relatively advanced ages of childhood. With
practice and maturation of systems, a stable and efficient walking pattern will
gradually emerge.
The first months are marked by a rapid evolution of spatiotemporal parameters,
including gait speed which goes from 0.20 to 0.80 m/s, and step length that is
growing while the base of support and relative double support decline. The range of
speeds used during a recording session also expands significantly, indicating greater
capacity for modulation of the movement. Subsequently, a period of refinement of
gait control will last to about 5–6 years of independent walking, that is to say around
6–7 years of age (Bril and Brénière 1992). At this age, all kinematic, kinetic, and
electromyographic characteristics of adult gait will be present in children. Changes
in spatiotemporal parameters, subsequent to that age, would be due mainly to growth
(Vaughan 2003), as these parameters do not evolve more once standardized by size
or lower limb length.
However, the study of gait variability reveals significant changes, independent of
growth, among children of 6–7 years and children aged of 11–14 (Hausdorff et al.
1999). For example, the coefficient of variation of the cycle time is halved between
4 and 7 years (8.4–4.3%); it remains significantly higher than in 11-year-old children
(1.9%). It’s the same for the composite score of variability, the gait variability index,
which is not yet comparable for 7-year-old children to the adults as variability
continues to decline after that age (Gouelle et al. 2016).
Note that normative data are available in the articles of Dusing and Thorpe (2007)
and Gouelle et al. (2016), respectively, on 438 children aged 1–10 years (1-year
interval groups) and 140 children aged 1–17 years (2-year interval groups).
Advanced age is associated with senescence, that is to say changes in anatomical,
physiological, and cognitive systems, in absence of any pathology. Slow degradation
of musculoskeletal and neurological systems that contribute to balance and postural
control is progressive and can lead to vestibular deficit, decrease in visual acuity,
worse sensitivity and proprioception, loss of muscle strength, or increased reaction
time. This aging process is neither linear for an individual nor uniform through
individuals, but the changes that take place can more or less affect the dynamic
stability of subjects and ultimately increase the risk of falls.
It seems that the first signs can appear from the age of 50–55 years
(Balasubramanian et al. 2015), and it accelerates after 60–65 years, but it is only
in older subjects (75 years and more) that the changes are always visible. Thus, the
first effects of aging on the gait stability are not always discernible in spontaneous
walking conditions and are revealed only in the most destabilizing circumstances
(e.g., uneven surface) or during dual task, the control of gait requiring increasingly
cognitive control and attentional resources. In fact, all the characteristics of an
unsteady walk are particularly exacerbated in the older subjects, in fallers and in
people with a fear of falling (Maki 1997).
Studies which have focused on the identification of walking changes associated

with age are numerous. Generally, it has been demonstrated that the organization of
the walk is moving toward lower preferred and maximal gait speed, longer double
support phase, shorter steps, a wider base of support, fewer vertical displacement of
the center of mass, and poorer synchronization between the leg movement and
swinging arms. These changes are often interpreted as the adoption of a steadier
and more secure walking pattern to compensate for the reduced physical abilities.
Moreover, the variability of the spatiotemporal parameters is increased with age,
portraying more concrete dynamic instability.
Note that Hollman et al. (2011) published data on the walk of 294 healthy old
persons, over 70 years, providing reference values for 23 spatiotemporal parameters.
Influence of the Pathology
Various pathologies affect walking and the resulting spatiotemporal parameters.

These are not always the same parameters that are the most relevant to observe.
Some walking patterns are obvious and easily identifiable (festination and freezing
in Parkinsonian), while other diseases can affect different parameters depending on
the type of injury and involved systems (balance control, muscle weakness, joint
limitation, etc.), as possible compensations and strategies. We will give here two
examples, one on the evolution of spatiotemporal parameters in patients with
Friedreich ataxia, the other on the modifications due to the change of walking aids.
Friedreich ataxia is a neurodegenerative disease in which there is a combination
of cerebellar, pyramidal syndromes, and axonal neuropathy causing coordination
deficits, loss of proprioception, and balance difficulties in static conditions and
during gait. The rapid degenerative nature causes instability and falls with increasing
frequency over short periods. In this context, it is important to monitor progress over
time or to gauge the effects of a therapeutic intervention.
The first signs of ataxia are balance difficulties with eyes closed or in low light
condition. Spatiotemporal parameters are still not really altered if one considers only
the mean values, while the variability is already increased. Almost all parameters are
within the ranges of values observed in peer controls. Only the variability compo-
nent (orange) shows coefficients of variation slightly increased at spontaneous gait
speed, already portraying the presence of balance disorders (Fig. 4a).
Subsequently, parameters such as walking speed or step width highlight the
disorder. In Fig. 4b, the gait speed is slightly decreased due to a reduction of the
cadence and the base of support is slightly wider. The variability of the parameters is
even greater.
With the progression of the disease, particularly marked by the worsening of
ataxia and deep sensitivity disorders, it is common to see two types of evolution. One
is based on the adoption of a safer walking pattern, less destabilizing, where the
preservation of the balance is prioritized over the speed of progression: the speed and
step length are reduced, the base of support is widened, and double support is very
long (Fig. 4c). The other, in contrast, is based on an acceleration of the walking speed
a b
Cadence Cadence
Walk ratio Walk ratio
CV of single CV of single
support support
CV of step Step length CV of step Step length

length length
Single support Speed Single support Speed

asymmetry asymmetry
Step length Step length

asymmetry Single support asymmetry Single support
CV of base of support Double support CV of base of support Double support

Base of support Base of support
c d
Cadence Cadence
CV of single Walk ratio CV of single Walk ratio
support support
CV of step CV of step
Step length length Step length
length
Single support Single support

Speed Speed
asymmetry asymmetry
Step length
Step length Single support
Single support asymmetry
asymmetry
CV of base of support Double support CV of base of support Double support

Base of support Base of support
Fig. 4 Spatiotemporal parameters observed during disease progression in Friedreich ataxia. The
yellow band represents the normal range for healthy controls at spontaneous walking speed. The
more the patient parameter departs from this range, the more the value is close to the radar center.
The lightning forms are added here to help the reader to identify where the main visible troubles are
(Notice that this radar is a personal way of the author to present clinical data and is part of a current
project by Gouelle and Poirier, not already published)
with a forward projection, but without efficient control. In Fig. 4d, the speed seems
correct but is in fact the product of a high cadence and too short steps. The double
support is far from the normative data, because it is highly reduced. It is the same for
the base of support that is close to zero due to numerous crossing steps.
Finally, when the patient reached an advanced stage of disease and when disor-
ders become major, it is impossible to walk without support (walk with assistive
device) and all of these parameters are clearly affected (Fig. 5).
It is useful to use both analysis of the mean parameters and variability during
rehabilitation. We will now try to illustrate this through the example of the most
appropriate time for a patient to change from one assistive device to another. To
simplify the connection, we will consider only two scores, one on the mean
Cadence
Cv of single support Walk ratio
CV of step length Step length
Single support
Speed
asymmetry
Step length asymmetry Single support
CV of base of support Double support
Base of support
Fig. 5 Temporal and spatial parameters observed during the use of an assistive device in a patient
with advanced stage of Friedreich ataxia
parameters (Functional Ambulation Performance Score), the other on the variability

(gait variability index).
The Functional Ambulation Performance Score (FAPS) is a quantitative repre-
sentation of the walk of a person based on a selection of spatiotemporal data obtained
at spontaneous speed (Gretz et al. 1998). These parameters are speed normalized by
the length of lower limb, normalized step length, step time, left–right asymmetry of
the step length, and base of support. From a maximum score of 100, points are
subtracted according to the deviation of the patient’s parameters compared to a
normative adult base. A normal walk is reflected by a score between 95 and
100 (Gouelle et al. 2011). The FAPS is commonly used for clinical evaluations
and has been used in an increasing number of publications over the past few years.
However, its use is sometimes distorted by misunderstandings of its composition and
calculation, practical, and/or conceptual limits. We recommend you also read the
review by Gouelle (2014) before using the FAPS. Regarding gait variability index
(Gouelle et al. 2013), it has already been addressed in the definition of variability
parameters (Va3). As a reminder, it is a score of 100 which decreases when the
overall variability of the subject deviates from the variability observed in a control
population. Thus, in both cases (FAPS and GVI), scores moving closer to 100 can be
considered as an overall improvement in walking of the patient.
Along the natural evolution of an optimal rehabilitation for a patient using two
crutches at the onset of rehabilitation and ending up not using technical assistance,
we usually see that the functional improvement is continuous. As the FAPS is very
sensitive to gait speed, it tends to increase gradually as the patient improve his/her
step length and velocity, even if one can observe stabilization phases. The FAPS, by
design, allows also to take into account in its calculation the use of assistive device;
therefore, the score improves as assistive device become lighter. The GVI, mean-
while, reflects the quality of the control and stability. This is naturally degraded when
there is a change of technical assistance, to the extent that each change frees a
number of degrees of freedom that the patient has to master. While the patient
progresses in the control of the assistive device, it becomes more reproducible and
less variable from one cycle to another. In case the change of assistive device is
premature, then the instability is such that it directly affects the functional parameters
and FAPS then deteriorates. When changing from an assistive device to another
(e.g., from two canes to only one), it is normal to observe a slight deterioration in gait
variability, but this should not affect the functional aspect if the timing to switch
walking aids is appropriate.
Conclusion
According to what has been mentioned above, it is complex to give a clinical

interpretation to spatiotemporal parameters due to the large quantity and variety of
information available. This includes the deviation from the mean normal parameters,
which provides information about the functionality of the gait. We also discussed the
notion of temporal and/or spatial asymmetry that shows mainly pathological dom-
inance but also functional strategies to minimize the impact of the disease on the
functional aspects of gait. Finally, taking into account the variability gives primarily
information about quality of the motor control, reflecting the sustained disturbance
and/or temporal and spatial regulation made by the patient.
This variety of information perfectly complements how we organize the reading
of these parameters. Whatever the population is analyzed, regardless of pathology,
the organization of the gait meets two key criteria: to minimize pain and to expend as
little energy as possible. With these two criteria in mind, it becomes particularly
interesting to analyze the set of spatiotemporal parameters. The patient adaptability,
extended to respect these criteria, will be proportional to the degrees of freedom
available. When we talk about the concept of degree of freedom, we are not only
concerned with mechanical degrees of freedom (joint). The temporal and spatial
regulation is also part of these degrees of freedom and, until all these degrees of
freedom are mobilized, the patient can continue to walk. Walking is lost when no
further adjustment is possible under the constraints posed by the disease and its
consequences.
To follow these two immutable rules, minimize pain and reduce energy costs at
the maximum possibilities in the moment, the patient will have only a short window
of time to adapt: the swing phase. During this phase, the patient will be able to
control the time and/or the space to minimize the single support on the side posing
the most problems. Indeed, everyone will try to minimize the duration of the single
support on the most problematic side. The mere mention of a pebble in a shoe allows
us to understand that we will organize to minimize the support time of painful side
(single support). At that time, on the contralateral side during the swing (swing
phase), we will be able to choose between increasing the speed to land sooner the
non-affected side without decreasing the step length, or else reduce the step length to
land earlier without increasing the walking speed. Both solutions are possible and
the choice depends largely on available adaptive capacity. For a patient who has knee
extension limitations on one side, the minimization of the step length is de facto a
strain. For this patient, the adaptive capacities are located mainly at the speed in
order to minimize the duration spent on the affected side.
Interpreting variability is difficult, given that no single number or measure gives
us its nature. Indeed, how does one differentiate variability due to pain during
support and variability due to varying the temporal and/or spatial parameters from
one cycle to another, to adapt to such pain. In fact, within a coefficient of variation
hides the two aspects: perturbation and regulation. However, we can determine by
deduction the disturbance and the control parts. Take the example of a right
hemiparetic patient, in which we will observe a right single support deficit. In the
specific case of the step length, it is possible to observe an increase in variability on
the right and/or left. If it is on the right, I can hypothesize a painful gait on right side.
If it is on the left, I can assume an anteroposterior regulation on healthy side,
compensating for instabilities generated by the affected side. Of course, this reason-
ing is simplistic, and it can have meaning only by matching all of the clinical
elements. However, it helps to understand the logic helping the interpretation of
spatiotemporal parameters.
Cross-References

▶ Detecting and Measuring Ataxia in Gait
▶ Gait Parameters Estimated Using Inertial Measurement Units
▶ Gait scores – Interpretations and Limitations
▶ Measures to Determine Dynamic Balance
▶ Normalization Techniques
▶ Optimal Control Strategies for Human Movement
References
Atkinson G, Nevill AM (1998) Statistical methods for assessing measurement error (reliability) in
variables relevant to sports medicine. Sports Med 26:217–238
Balasubramanian CK, Clark DJ, Gouelle A (2015) Validity of the gait variability index in older
adults: effect of aging and mobility impairments. Gait Posture 41(4):941–946
Brach JS, Berlin JE, VanSwearingen JM, Newman AB, Studenski SA (2005) Too much or too little
step width variability is associated with a fall history in older persons who walk at or near
normal gait speed. J Neuro Eng Rehab 2:21
Bril B, Brénière Y (1992) Postural requirements and progression velocity in young walkers. J Mot
Behav 24:105–116
Dusing SC, Thorpe DE (2007) A normative sample of temporal and spatial gait parameters in
children using the GAITRite electronic walkway. Gait Posture 25:135–139
Eisenhardt JR, Cook D, Pregler I, Foehl HC (1996) Changes in temporal gait characteristics and
pressure distribution for bare feet versus various heel heights. Gait Posture 4(4):280–286
Gouelle A (2014) Use of functional ambulation performance score as measurement of gait ability:
review. JRRD 51(5):665–674
Gouelle A, Mégrot F, Presedo A, Penneçot GF, Yelnik A (2011) Validity of functional ambulation
performance score for the evaluation of spatiotemporal parameters of children’s gait. J Mot
Behav 43(2):95–100
Gouelle A, Mégrot F, Presedo A, Husson I, Yelnik A, Penneçot GF (2013) The gait variability
index: a new way to quantify fluctuation magnitude of spatiotemporal parameters during gait.
Gouelle A, Leroux J, Bredin J, Mégrot F (2016) Changes in gait variability from first steps to
adulthood: normative data for the gait variability index. J Mot Behav 48(3):249–255
Gretz HR, Doering LL, Quinn J, Raftopoulos M, Nelson AJ, Zwick DE (1998) Functional
ambulation performance testing of adults with down syndrome. Neurorehabilitation 11
(3):211–225
Hausdorff JM (2005) Gait variability: methods, modeling and meaning. J Neuro Eng Rehab 2:19
Hausdorff JM, Zemany L, Peng C-K, Goldberger AL (1999) Maturation of gait dynamics: stride-to-
stride variability and its temporal organization in children. J Appl Physiol 86:1040–1047
Hof AL (1996) Scaling gait data to body size. Gait Posture 4:222–223
Hollman JH, McDade EM, Petersen RC (2011) Normative spatiotemporal gait parameters in older
adults. Gait Posture 34(1):111–118
Kirtley C (2006) Clinical gait analysis: theory and practice. Churchill-Livingstone, New York
Lauzière S, Betschart M, Aissaoui R, Nadeau S (2014) Understanding spatial and temporal gait
asymmetries in individuals post stroke. Int J Phys Med Rehabil 2:3
Lord S, Galna B, Verghese J, Coleman S, Burn D, Rochester L (2013) Independent domains of gait
in older adults and associated motor and nonmotor attributes: validation of a factor analysis
approach. J Gerontol A Biol Sci Med Sci 68(7):820–827
Lord S, Galna B, Coleman S, Yarnall S, Burn D, Verghese J (2014) Cognition and gait show a
selective pattern of association dominated by phenotype in incident Parkinson’s disease. Front
Aging Neurosci 6:249
Maki BE (1997) Gait changes in older adults: predictors of falls or indicators of fear. J Am Geriatr
Soc 45:313–320
Novacheck TF (1998) The biomechanics of running. Gait Posture 7:77–95
Paterson KL, Lythgo ND, Hill KD (2009) Gait variability in younger and older adult women is
altered by overground walking protocol. Age Ageing 38(6):745–748
Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE (2010) Evaluation of gait symmetry
after stroke: a comparison of current methods and recommendations for standardization. Gait
Posture 31:241–246
Perry J, Burnfield J (2010) Gait analysis: normal and pathological function, 2nd edn. Slack
Incorporated, Thorofare
Sekiya N, Nagasaki H, Ito H, Furuna T (1996) The invariant relationship between step length and
step rate during free walking. J Hum Mov Stud 30:241–257
Sutherland D (1997) The development of mature gait. Gait Posture 6:163–170
Vaughan CL (2003) Theories of bipedal walking: an odyssey. J Biomech 36(4):513–523.
Verghese J, Wang C, Lipton RB, Holtzer R, Xue X (2007) Quantitative gait dysfunction and risk of
cognitive decline and dementia. J Neurol Neurosurg Psychiatry 78(9):929–935
Verlinden VJA, van der Geest JN, Hoogendam Y, Hofman A, Breteler MMB, Ikram MA (2013)
Gait patterns in a community-dwelling population aged 50 years and older. Gait Posture 37
(4):500–505
Whittle MW (2007) Gait analysis: an introduction. Butterworth-Heinemann, Oxford
Assessing Pediatric Foot Deformities by
Pedobarography
Dieter Rosenbaum
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
General Considerations for the Use of Pedobarography in Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Clinical Applications and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
The majority of children’s feet are considered as normal at birth but still may
develop certain characteristics where a distinction between normal shape and foot
deformity appears necessary, not only for addressing parents’ concerns for their
child’s health and well-being but also to decide whether an intervention – be it
conservative or surgical – is warranted. Therefore the natural development of
children’s feet should be monitored not only with visual and/or radiographical
assessment but preferably also with functional assessment. There are several
methods to assess foot function under dynamic loading conditions, one regularly
applied method being pedobarography, i.e., the measurement of the plantar
pressure distribution during gait.
This method can be and has been used from infancy all the way to adolescence
as it provides a detailed description and analysis of the local load distribution
under the plantar aspect of the foot. The spatial resolution of the sensor area
should be sufficient to identify the main loading areas even in the smaller feet of
children, and the temporal resolution should be fast enough to provide sufficient
D. Rosenbaum (*)
Funktionsbereich Bewegungsanalytik, Institut für Experimentelle Muskuloskelettale Medizin,
Zentrum für Muskuloskelettale Medizin, Universitätsklinikum Münster, Münster, Germany
e-mail: diro@uni-muenster.de

DOI 10.1007/978-3-319-30808-1_36-1
2 D. Rosenbaum
detail for the various phases of the ground contact in dynamic loading situations.
With a range of parameters, the local loading characteristics can be described for
various regions of the foot. This allows for a detailed assessment of the individ-
ual’s foot loading characteristics and for comparisons across the life-span.
The present contribution aims to provide insights into the application of
pedobarography as an assessment tool for normal children’s feet as well as for
pediatric foot deformities such as clubfeet, idiopathic flatfeet, and CMT feet,
supported by demonstrations with appropriate examples.
Keywords
Children’s feet • Foot deformities • Pediatric orthopedics • Foot loading • Plantar
pressure measurements • Pedobarography • Clubfoot • Flatfoot • Charcot-Marie-
Tooth disease
Introduction
“Fat, flat, and floppy” – these are the terms being used to succinctly describe the
appearance of the toddler’s foot after birth. These terms illustrate that it is typically a
stubby, chubby, flexible structure with a wide midfoot and no clearly discernible
longitudinal arch. However, the foot usually develops rapidly as soon as the child
gets up on its feet to use them for upright stance and bipedal gait.
While most children’s feet can be expected to develop normally, up to 30 % may
reveal congenital deformities as a consequence of some fetal maldevelopments or
idiopathic deformities that may appear at birth or in later developmental stages
(Tsaknakis and Hell 2013; Velasco 2012). The most common foot disorders in the
child’s foot are the flatfoot, a.k.a. pes planus or pes planovalgus, and the clubfoot,
a.k.a. pes equinovarus.
However, there are many more and more severe disorders that may, for example,
affect the number and separation of toes (syndactyly with a fusion between two or
more digits, split or cleft foot, polydactyly with one or more extra toes, oligodactyly
with one or more missing digits, brachydactylic with a shorter toe and/or metatarsal
development). These congenital deformities may be identified early on, at birth
already, so that the appropriate therapy can be initiated as soon as possible. On the
other hand, idiopathic deformities may be revealed either at birth or may develop
over time so that the treatment decisions need to be based on the developmental
status of the feet.
The appearance of their child’s feet is a regular concern for caring or careful
parents, and it is not merely the cosmetic aspect that needs to be evaluated. The main
focus should lie on the functional implications of the identified foot deformity, and
the feet should be functionally evaluated in order to determine the degree of a
potential impairment in daily life or at older age.
Assessing Pediatric Foot Deformities by Pedobarography 3
One commonly used diagnostic tool is the pedobarography, i.e., plantar pressure
measurements. When they are assessed during barefoot walking, they allow a
detailed description of the foot loading characteristics during the ground contact
phase of the gait cycle. In-shoe measurements, on the other hand, can be used to
investigate the interaction between the foot and the shoe or an orthotic, if it has been
prescribed for a certain treatment.
Why do we need to assess plantar pressure measurements? The interaction
between the human body and the environment during locomotor activities causes a
force transfer between the foot and the ground. Measurements of these ground
reaction forces can be used to assess the external loads on the human body in normal
locomotor activities as well as specific sports- or work-related situations.
In order to distinguish between normal and abnormal loading, the value of using
objective evaluation tools has been acknowledged already in 1932 (Schwartz and
Heath 1932). The importance of assessing foot mechanics during gait has also been
recognized by Katoh and colleagues (1983): “Clearly, the foot is critical to an
understanding of the mechanics of gait, as the foot often affects the normal motion
pattern of the entire lower extremity. Alteration of normal foot mechanics can
adversely influence the normal functions of the ankle, knee, hip and even the back.”
The additional information gained from dynamic foot loading measurements can
help to determine more objectively:
1. Whether the plantar pressures reveal a pattern of inconspicuous (i.e., normal) or

conspicuous (i.e., pathological) loading.
2. What the extent of deformation or deviation from a normal loading pattern
might be.
3. Whether the loading characteristics change over time, e.g., in the course of
disease or healing processes.
4. Whether treatment (either conservative or surgical) has led to the desired
improvement of the loading situation and foot function
State of the Art
Early investigations of foot pressure patterns have used the imprint that the feet left
in soft materials like plaster of Paris (Beely 1882) which is comparable to the
footprints seen in the sand of the beach. However, these early approaches were
only qualitative capturing only the shape of the foot and the location of the deepest
impressions in the surface but not the actual plantar forces or pressures. Already in
the nineteenth century, the first quantitative measurements were performed by Marey
(1873) and Carlet (1872) who used an air-filled chamber embedded in the sole of the
shoe. Further developments used arrays of deformable materials that left an ink print
or were recorded by optical methods.
4 D. Rosenbaum
Even though there are still some simple mechanical devices like the Harris foot
mat in use, today’s pressure measurement systems rely on specialized electrome-
chanical sensors. In essence, pressure sensors are force transducers that measure the
force acting on a known surface. Thus they provide the necessary information to
determine the pressure by dividing the measured force by the sensor/contact area.
Pressure is usually reported in units of kilo- or megapascal (kPa, mPa). In general,
the electromechanical transducers convert a mechanical event into an electrical
signal that is recorded and stored for further data analysis. Under load application,
the transducers deform and change their electrical properties. A wide variety of
measurement systems are available on the market (e.g., capacitive, resistive, piezo-
electric, optoelectronic).
In general, one should be aware of different sensor principles/devices and their
advantages and limitations:
• Pressure distribution platforms can be used in normal and pathological

populations for static and dynamic measurements (i.e., standing on or walking
across the sensor area), but their use is usually confined to the application in
laboratory settings. They are embedded in a walkway and the subject has to
contact the platform after several steps of approach. Their main advantage is that
the total foot contact area is evaluated without interference from footwear.
• In-shoe systems detect plantar pressures directly at the shoe-foot interface and can
therefore be used to evaluate the effect of different shoe constructions or ortho-
pedic interventions like orthotics. Their main advantage is that repeated steps can
be recorded in one measurement. Some of the available systems use portable data
loggers or bluetooth data transmission that allows field measurements outside the
confines of the laboratory. This enables monitoring of loading situations in daily
life, workspace, or sports environments that may be more realistic and relevant for
causing foot-related problems.
• Single-sensor systems have pressure-sensitive elements only in certain locations
of the insole and most often correspond to the location of anatomical structures of
a normal foot. Thus the amount of data is reduced so that faster sampling rates are
possible. However, critical information may be lost when high-pressure areas are
not in contact with one of the sensors in severe foot deformities that do not
conform to the sensor arrangement and deviate too much from normal feet.
• Discrete sensors may overcome these limitations as they can be placed under
defined anatomic locations after palpation by the investigator. These systems
usually use small and thin sensors, but one has to be aware that the protruding
sensors may act as a foreign body in the shoe causing discomfort and unconscious
changes in the natural walking pattern.
General Considerations for the Use of Pedobarography

in Children
The general feasibility and specific value of plantar pressure measurements in

children has been confirmed by a number of investigations on the foot loading
characteristics of normally developing children in all ages – most of them cross-
sectional studies and only a few longitudinal studies (Aharonson et al. 1980; Alvarez
et al. 2008; Bertsch et al. 2004; Bosch et al. 2007, 2010; Hallemans et al. 2003;
Hennig and Rosenbaum 1991; Hennig et al. 1994; Muller et al. 2012; Orlin et al.
1990; Phethean et al. 2014).
In our own Kidfoot Münster project, we longitudinally followed a group of about
100 children over the course of 9 years from the first independent steps to the end of
their elementary school age. The children were assessed with plantar pressure
measurements in 17 sessions: every 3 months during the first year, twice per year
from age 2 to 5, and once a year from age 6 to 10. With this enormous amount of data
(~17 sessions 100 children 10 trials = 17,000 data files), we were able to
describe the range of normal foot development from a toddler’s age until late
infancy. (A subgroup is currently being followed further, in order to observe the
next developmental steps up to the end of adolescence.) The data analysis revealed
some generally applicable findings with respect to the development of normal
children’s feet:
• The child’s foot develops rapidly as soon as it is loaded in bipedal stance and gait.
However, in the first years of independent walking, pronounced interindividual
differences also demonstrate an individually different speed of the growth and
maturation processes (Bertsch et al. 2004).
• A normal rollover process with initial heel contact usually develops within the
first year of independent walking (Bertsch et al. 2004).
• The “growth velocity” or speed of development is gradually reduced after 4–5
years. The characteristic foot shape of an individual appears to be manifested up
to an age of about 6 years. The foot dimensions increase further but only minor
morphometric changes can be expected (Bosch et al. 2010).
• Therefore, the expectation that slight foot deformities develop into a normal foot
shape occurs naturally, cannot be relied on, and appears unlikely beyond this age.
Hence, the wait-and-see approach should be considered critically when conspic-
uous foot shapes have been diagnosed. This elementary school age would be the
period where foot strengthening exercises should be recommended in an
approach to positively influence the further development and actively improve
the muscular stabilization of the feet.
The specific clinical use of pedobarography has been acknowledged already more
than 20 years ago (Hughes 1993). Since then, the range of clinical applications of
foot pressure measurements has led to a constantly growing interest as can be seen in
the ever-increasing numbers of publications (Fig. 1) that started to rise in the early
1980s and reached over 300 new contributions last year.
6 D. Rosenbaum
Documents by year
350
300
250
Documents
200
150
100
50
0
1923 1931 1939 1947 1955 1963 1971 1979 1987 1995 2003 2011 2019
Fig. 1 A Scopus search for the combination of the terms “plantar pressure,” “foot pressure,”
pedograph*, pedobarograph*, or baropodo* resulted in 3672 hits (3.5.2016) illustrating the
increasing rise of the numbers of yearly publications starting in the 1980s
While the most important topic appears to be the diabetic foot and its related
problems, around 270 papers on children’s foot-related problems illustrate the
widespread pediatric applications, ranging from (neuro-)musculoskeletal problems
(e.g., flatfoot, clubfoot, cerebral palsy, Charcot-Marie-Tooth disease) to general or
systemic health problems (e.g., overweight and obesity, juvenile idiopathic arthritis
(JIA)) and psychosocial disorders (e.g., attention deficit hyperactivity disorder,
autism).
For clinical applications in adults as well as in children, it is recommended to
perform repeated measurements (at least three, preferably five steps with each foot)
in a single session in order to ensure a sufficient reliability that allows for a detailed
description of the child’s foot loading characteristics (Cousins et al. 2012; Gurney
et al. 2008, 2013; Hughes et al. 1991; Putti et al. 2008; Riad et al. 2007; Tong and
Kong 2013; Van Der Leeden et al. 2004).
The main advantage of barefoot pressure measurements in children can be seen in
the “noninvasive” nature of the application since it does not require the attachment of
any markers, sensors, or cables that might irritate young children and impair their
compliance and cooperation. With the use of pressure distribution platforms, which
are embedded openly or may also be covered/camouflaged in a walkway, even the
youngest children at the onset of walking can already be measured (Bertsch et al.
2003; Hallemans et al. 2003, 2006). The lab personnel, supported by parents, should
try to generate a playful atmosphere so that the child’s cooperation just needs to be
directed toward or across the measurement platform. In this case, the use of toys or
toy animals might already be helpful to distract the child’s attention from the actual
measurements.
With respect to the spatial resolution, previous research has shown that “greater
than 6.36 mm 6.18 mm (mediolateral and anteroposterior directions) would result
in sub-optimal sampling of PPPs [plantar pressure profiles]” (Davis et al. 1996).

Therefore, four sensors per cm2 with a sensor size of 5 5 mm appear sufficient in
adult feet as well as for children. A higher resolution of nine sensors per cm2 did not
reveal more detailed information on children’s feet as compared to four sensors per
cm2 (Rosenbaum and Lorei 2003).
Several parameters are available for a detailed description of the load distribution
of the whole foot or a segmental analysis of selected foot regions, and there is no
standard procedure so that the choice should be adapted to the specific clinical
problem and the research question. Peak pressure values can be of interest to
understand the occurrence of pain or to prevent localized overloading in insensate
feet, e.g., of neuropathic patients. Regional forces are of interest when a load
redistribution is being observed in foot deformities, or shoe/orthotic interventions
aim to off-load a painful foot region. Finally, force-time integrals take into account
the amplitude and duration of loading. For certain foot problems, the gait line can be
a further indicator of normal or disturbed foot function, e.g., in equinus feet with
initial forefoot contact, in flatfeet with a more medial course, or in clubfeet with a
more or less pronounced lateralization. These mechanisms may also be described by
medial-to-lateral or rearfoot-to-forefoot loading ratios. In essence, the analytical
choices should be geared toward the specifics of the investigated foot pathology.
Clinical Applications and Examples
As mentioned above, pedobarographic measurements have been applied in a wide

variety of pediatric foot disorders. Therefore, the present clinical applications will
describe a selection, focusing on more often encountered examples of pediatric
flatfeet, clubfeet, neurogenic cavus feet, and juvenile idiopathic arthritis.
In pediatric flatfeet, the treatment regime is less straightforward since a certain
degree of this deformity is generally accepted as a transitional stage of the child’s
foot (that may be combined with a valgus knee alignment) on the way to becoming
an adult foot (Forriol and Pascual 1990; Staheli et al. 1987). There are no generally
accepted scores or thresholds that help to draw the line between normal and
pathological. It is a continuum and the judgment often requires expert opinion.
When a flatfoot is seen in younger (preschool-age) children, the strategy is usually
to wait and see whether the longitudinal arch just develops at a slightly later stage
and the deformity “grows out.” The advice may be to encourage the child to walk
and run barefoot more often in order to stimulate the arch development. This may be
accompanied by foot strengthening exercises that support the development of
extrinsic and intrinsic foot muscles for a better active support of the foot during
static and dynamic loading. Even though there is a clear lack of clinical evidence for
supporting this approach, anecdotal evidence from our lab appears promising in
selected examples where children were self-motivated enough to take up and
maintain a regular exercise program. This presented example shows a reduction of
the midfoot contact area during dynamic loading as well as a better hindfoot
alignment in bipedal stance (Fig. 2).
8 D. Rosenbaum
Fig. 2 Clinical and pedobarographic example of a girl (age 12 years) with an asymmetric pes
planovalgus before and after an intensive regular home exercise program: (a) initial presentation,
(b) 1 year, and (c) 2 years after recommending the foot exercises. The plantar pressure patterns
(shown here as the maximum pressure pictures = MPP that summarize the highest load during the
whole stance phase in one picture) as well as the photograph document a good corrective response
of the foot not only during stance but also during dynamic loading in barefoot gait
It has to be realized, however, that most features of the individual foot shape seem
to be largely established at an age of 5–6 years (Onodera et al. 2008), so that the
wait-and-see strategy beyond this age may not suffice. The value of conservative
treatment with supportive or proprioceptive orthotics is still a matter of debate, and
mostly observational data rather than strong evidence support the effectiveness of
the latter (Bernius 2010). While orthotics may help to prevent excessive hindfoot
valgus malalignment during extended loading, it has not yet been shown in how far a
structural change in foot morphology might be achieved and how long such an
approach should be employed (Evans and Rome 2011). In more severe cases, where
all conservative options did not lead to a clearly positive result, surgical treatment
may be an option to change the bony alignment with an arthroereisis using a
temporary implant in the sinus tarsi or an extra-articular calcaneus stop screw
(Fig. 3) (Hamel 2010; Kellermann et al. 2011; Richter and Zech 2013). More
severely affected children require extensive surgical correction (Westberry et al.
2013). The decisions for or against surgical treatment are usually based on clinical
(and radiographic) assessments but could be backed up by dynamic pedobarography
as it provides a further aspect of foot function and limitations in these patients.
Fig. 3 Clinical and pedobarographic example of a boy (age 11 years) with a pronounced, bilateral
pes planovalgus before and after correction with a calcaneus stop screw: (a) Before surgery, (b)
2 months after implantation, and (c) 1 year after implantation. The plantar pressure patterns as well
as the photograph demonstrate a good corrective response of the foot
However, a generally accepted score based on pedobarography is not available so

that an interdisciplinary exchange between clinicians and movement scientists might
be a beneficial approach.
For pediatric clubfeet, the value of the Ponseti treatment has been most impres-
sively demonstrated in long-term follow-ups of 30 and 45 years revealing the lasting
success in these patients (Cooper and Dietz 1995; Dobbs et al. 2003). These and
other positive results (Sinclair et al. 2009) have surely helped to convert the state-of-
the-art treatment from a surgical regime to the conservative approach with serial
casting that apparently has been adopted worldwide in the meantime. Recent reviews
confirm that it should be the method of first choice and preferred over surgery
(Church et al. 2012; Lykissas et al. 2013; Zhao et al. 2014) even though residual
deformities may persist (Jeans and Karol 2010; Lampe et al. 2016). In these cases,
pedobarographic measurements can still be helpful to follow up on the children’s
further foot development in order to identify early on whether the clubfoot shows a
recurrence or whether the correction effect is maintained. This is illustrated in
selected examples from our lab (Figs. 4 and 5). With respect to comparisons, it
appears advisable not only to compare between affected and contralateral feet as they
are not necessarily considered as “normal” (Favre et al. 2007), probably in the
human body’s attempt to achieve gait symmetry. The authors therefore advised to
use a control group of normal feet for comparison.
10 D. Rosenbaum
Fig. 4 Clinical and pedobarographic example of a boy (age 15 years) with a bilateral clubfoot. The
cavus foot component can clearly be seen in the plantar pressure pattern (left) as well as in the
podometer picture (center, taken from underneath through a glass plate) and the anterior view
(right). This leads to the pronounced loading of the lateral forefoot and is accompanied with the
typically reduced heel loading in clubfeet. However, due to the initial heel contact, no equinus foot
component can be observed
Fig. 5 Clinical and pedobarographic follow-up of a boy (age 9–13 years) with a recurrent clubfoot
on the right: (a) before surgery, (b) after right foot surgery, (c) after left foot surgery, and (d) follow-
up. A clearly improved foot loading pattern can be seen in the right foot that was only loaded in a
small contact area under the lateral forefoot and midfoot leading to high local pressures. After
surgery, the contact area has been increased and has shifted the loads toward the first and second ray
In the course of the Charcot-Marie-Tooth disease (CMT), the patients tend to

develop a neurogenic cavovarus foot deformity with a typical load shift to the lateral
midfoot and forefoot and loss of toe function due to claw toes (e.g., Fig. 6).
Pedobarography has been used to assess the treatment outcome in the cavovarus
feet of nine children suffering from CMT (Chan et al. 2007). Operative treatment
consisted of a combination of osteotomies and muscle transfers. The comparison of
pre- and postoperative measurements revealed decreased forefoot and lateral
midfoot pressures that were not correlated to the amount of radiographically
Fig. 6 A typical neurogenic cavovarus foot of a patient with Charcot-Marie-Tooth disease

(12-year-old girl). A high arch has developed, and the loads are dramatically increased in the lateral
midfoot causing a local pressure peak under the base of the fifth metatarsal. The toes are clawing
and are only slightly loaded during the rollover process
achieved correction. The authors conclude that pedobarography was able to demon-
strate changes in plantar pressure patterns after operative treatment. This is con-
firmed by an exemplary CMT patient from our lab showing plantar pressure patterns
before and after corrective surgery (Fig. 7).
In juvenile idiopathic arthritis (JIA), the first plantar pressure measurements were
carried out already in 1980. Given the limited technical possibilities back then, the
authors were able to describe only some basic features of foot function in these
patients such as reduced loading under the toes and medial metatarsal heads and a
longer heel contact time (Dhanendran et al. 1980). More recently, in-shoe pressure
measurements were used in an attempt to objectively describe and classify gait
abnormalities in JIA patients with a pattern recognition approach that was not always
able to predict the resultant gait pattern (Fairburn et al. 2002). Lately, a combined
approach of three-dimensional, computerized gait analysis and pedobarography in
JIA patients with pes planovalgus allowed for a detailed description of the patients’
pathophysiology in comparison to a control group with excessive hindfoot eversion
that was related to a reduced arch height (Merker et al. 2015).
In normal children, plantar pressure data does not require normalization with
body mass within a range of normal body weight due to a weak association between
12 D. Rosenbaum
Fig. 7 Clinical and pedobarographic example of a cavovarus foot in a boy (age 14 years) with
Charcot-Marie-Tooth disease before and after surgery in the left foot. The pressure patterns reveal a
correction of the hindfoot varus malalignment and less pronounced lateral foot loading after surgery
body weight or BMI with plantar pressure data (Phethean and Nester 2012). How-
ever, obese children at the age of 7 years already revealed higher midfoot and lateral
forefoot loading that were seen also in obese children so that the authors suggested
using pedobarography for early assessment and decisions on necessary interventions
(Cousins et al. 2013). Similar results were also reported for obese prepubescent
Chinese children who revealed flatter feet and higher local peak pressures as
compared to nonobese children (Yan et al. 2013). Furthermore, it was confirmed
that lower activity levels are related to higher plantar pressures in overweight
children, especially in the forefoot (Riddiford-Harland et al. 2015). However, a
decreased body weight after a physical activity program in overweight and obese
children did not positively influence foot loading (Riddiford-Harland et al. 2016).
Future Directions
How should we know what the future will bring? The current situation has demon-
strated a continuous and even growing acceptance of pedobarography for the
functional assessment of children’s feet in research as well as clinical applications
so that I would expect this development to carry on even further.
The instrumentation is well-established and the market offers a range of technical
solutions. The ease of use makes plantar pressure measurements applicable in a wide
range of clinical settings, i.e., not only in specialized centers but also in smaller
institutions and private practices so that more children and their clinicians should be
able to benefit from the additional information that is provided beyond standard
clinical and radiographic data.
However, the interpretation of the results still requires a certain degree of expe-
rience and expert knowledge that can partly be obtained from the large body of
literature even though it needs some regular in-depth involvement in order to
develop sufficient competence and confidence with the application of the
technology.
Research should help to further identify, justify, and validate parameters of
interest so that novice users do not have to make their own choices from the wide
array of data processing methods but can rely on established protocols for guidance.
References
Aharonson Z, Voloshin A, Steinbach TV et al (1980) Normal foot-ground pressure patterns in
children. Clin Orthop 150:220–223
Alvarez C, De Vera M, Chhina H et al (2008) Normative data for the dynamic pedobarographic
profiles of children. Gait Posture 28:309–315
Beely F (1882) Zur Mechanik des Stehens. Langenbecks Arch Chir 27:457–471
Bernius P (2010) Sensomotoric shoe inserts – what’s new, what’s known? Fuß Sprunggelenk
8:16–27
Bertsch C, Rosenbaum D, Winkelmann W (2003) Evaluation of early walking patterns by plantar
pressure distribution. Clin Biomech 18:S3–S4
Bertsch C, Unger H, Winkelmann W et al (2004) Evaluation of early walking patterns from plantar
pressure distribution measurements. First year results of 42 children. Gait Posture 19:235–242
Bosch K, Gerss J, Rosenbaum D (2007) Preliminary normative values for foot loading parameters
of the developing child. Gait Posture 26:238–247
Bosch K, Gerss J, Rosenbaum D (2010) Development of healthy children’s feet – nine-year results
of a longitudinal investigation of plantar loading patterns. Gait Posture 32:564–571
Carlet G (1872) Sur la locomotion humaine. Ann Sci Nat 5:1–92
Chan G, Sampath J, Miller F et al (2007) The role of the dynamic pedobarograph in assessing
treatment of cavovarus feet in children with Charcot-Marie-Tooth disease. J Pediatr Orthop
27:510–516
Church C, Coplan JA, Poljak D et al (2012) A comprehensive outcome comparison of surgical and
Ponseti clubfoot treatments with reference to pediatric norms. J Child Orthop 6:51–59
Cooper DM, Dietz FR (1995) Treatment of idiopathic clubfoot. A thirty-year follow-up note. J
Bone Joint Surg Am 77:1477–1489
Cousins SD, Morrison SC, Drechsler WI (2012) The reliability of plantar pressure assessment
during barefoot level walking in children aged 7–11 years. J Foot Ankle Res 5:8
Cousins SD, Morrison SC, Drechsler WI (2013) Foot loading patterns in normal weight, overweight
and obese children aged 7 to 11 years. J Foot Ankle Res 6:36
Davis BL, Cothren RM, Quesada P et al (1996) Frequency content of normal and diabetic plantar
pressure profiles: implications for the selection of transducer sizes. J Biomech 29:979–983
Dhanendran M, Hutton WC, Klenerman L et al (1980) Foot function in juvenile chronic arthritis.
Rheumatol Rehabil 19:20–24
Dobbs MB, Corley CL, Morcuende JA, Ponseti IV. Late recurrence of clubfoot deformity: a 45-year
followup. Clin Orthop Relat Res. 2003;(411):188–92
Evans AM, Rome K (2011) A Cochrane review of the evidence for non-surgical interventions for
flexible pediatric flat feet. Eur J Phys Rehabil Med 47:69–89
14 D. Rosenbaum
Fairburn PS, Panagamuwa B, Falkonakis A et al (2002) The use of multidisciplinary assessment

and scientific measurement in advanced juvenile idiopathic arthritis can categorise gait devia-
tions to guide treatment. Arch Dis Child 87:160–165
Favre P, Exner GU, Drerup B et al (2007) The contralateral foot in children with unilateral clubfoot:
a study of pressures and forces involved in gait. J Pediatr Orthop 27:54–59
Forriol F, Pascual J (1990) Footprint analysis between three and seventeen years of age. Foot Ankle
11:101–104
Gurney JK, Marshall PWM, Rosenbaum D, Kersting UG (2013) Test–retest reliability of dynamic
plantar loading and foot geometry measures in diabetics with peripheral neuropathy. Gait
posture 37(1):135–137
Hallemans A, D’aout K, De Clercq D et al (2003) Pressure distribution patterns under the feet of
new walkers: the first two month of independent walking. Foot Ankle Int 24:444–453
Hallemans A, De Clercq D, Dongen SV et al (2006) Changes in foot-function parameters during the
first 5 months after the onset of independent walking: a longitudinal follow-up study. Gait
Posture 23:142–148
Hamel J (2010) Die calcaneostop-arthrorise–eine retrospektive klinische Studie mit Komplikations-
Analyse. Fuß Sprunggelenk 8(1):35–41
Hennig EM, Rosenbaum D (1991) Pressure distribution patterns under the feet of children in
comparison with adults. Foot Ankle 11:306–311
Hennig EM, Staats A, Rosenbaum D (1994) Plantar pressure distribution patterns of young school
children in comparison to adults. Foot Ankle Int 15:35–40
Hughes J (1993) The clinical use of pedobarography. Acta Orthop Belg 59:10–16
Hughes J, Clark P, Jagoe JR et al (1991) The pattern of pressure distribution under the
weightbearing forefoot. Foot 1:117–124
Jeans KA, Karol LA (2010) Plantar pressures following Ponseti and French physiotherapy methods
for clubfoot. J Pediatr Orthop 30:82–89
Katoh Y, Chao EYS, Laughman RK et al (1983) Biomechanical analysis of foot function during gait
and clinical applications. Clin Orthop 177:23–33
Kellermann P, Roth S, Gion K et al (2011) Calcaneo-stop procedure for paediatric flexible flatfoot.
Arch Orthop Trauma Surg 131:1363–1367
Lampe R, Mitternacht J, von Pfister L, Turova V, Blumenstein T, Alves-Pinto A (2016) Develop-
ment of congenital clubfoot during growth: a long-term study on the basis of pedobarography,
biomechanics, and magnetic resonance imaging measurements of muscle volumes. J Pediatr
Orthop B [Epub ahead of print]
Lykissas MG, Crawford AH, Eismann EA et al (2013) Ponseti method compared with soft-tissue
release for the management of clubfoot: a meta-analysis study. World J Orthop 4:144–153
Marey M (1873) De la locomotion terrestre chez les bipedes et les quadrupedes. J Anat Physiol 9:42
Merker J, Hartmann M, Kreuzpointner F et al (2015) Pathophysiology of juvenile idiopathic
arthritis induced pes planovalgus in static and walking condition: a functional view using 3D
gait analysis. Pediatr Rheumatol Online J 13:21
Muller S, Carlsohn A, Muller J et al (2012) Static and dynamic foot characteristics in children aged
1–13 years: a cross-sectional study. Gait Posture 35:389–394
Onodera AN, Sacco IC, Morioka EH et al (2008) What is the best method for child longitudinal
plantar arch assessment and when does arch maturation occur? Foot (Edinb) 18:142–149
Orlin MN, Pierrynowski MR, Robertson W (1990) Normative static foot pressure distribution
patterns in children. In: 6th annual east coast clinical gait conference, East Lansing, pp 123–125
Phethean J, Nester C (2012) The influence of body weight, body mass index and gender on plantar
pressures: results of a cross-sectional study of healthy children’s feet. Gait Posture 36:287–290
Phethean J, Pataky TC, Nester CJ et al (2014) A cross-sectional study of age-related changes in
plantar pressure distribution between 4 and 7 years: a comparison of regional and pixel-level
analyses. Gait Posture 39:154–160
Putti AB, Arnold GP, Cochrane LA et al (2008) Normal pressure values and repeatability of the
Emed ST4 system. Gait Posture 27:501–505
Riad J, Coleman S, Henley J et al (2007) Reliability of pediobarographs for paediatric foot

deformity. J Child Orthop 1:307–312
Richter M, Zech S (2013) Arthrorisis with calcaneostop screw in children corrects Talo-1st
Metatarsal-Index (TMT-Index). Foot Ankle Surg 19:91–95
Riddiford-Harland DL, Steele JR, Cliff DP et al (2015) Lower activity levels are related to higher
plantar pressures in overweight children. Med Sci Sports Exerc 47:357–362
Riddiford-Harland DL, Steele JR, Cliff DP et al (2016) Does participation in a physical activity
program impact upon the feet of overweight and obese children? J Sci Med Sport 19:51–55
Rosenbaum D, Lorei T (2003) Influence of density of pressure sensors on parameters of plantar foot
load – a pedographic comparison of two pressure distribution recording platforms. Biomed Tech
(Berl) 48:166–169
Schwartz RP, Heath AL (1932) The pneumographic method of recording gait. J Bone Joint Surg
4:783–794
Sinclair MF, Bosch K, Rosenbaum D et al (2009) Pedobarographic analysis following Ponseti
treatment for congenital clubfoot. Clin Orthop Relat Res 467:1223–1230
Staheli LT, Chew DE, Corbett M (1987) The longitudinal arch. A survey of eight hundred and
eighty-two feet in normal children and adults. J Bone Joint Surg Am 69:426–428
Tong JW, Kong PW (2013) Reliability of footprint geometric and plantar loading measurements in
children using the Emed((R)) M system. Gait Posture 38:281–286
Tsaknakis K, Hell AK (2013) Fußdeformitäten im Kindesalter. Kinderarztl Prax 84:35–40
Van Der Leeden M, Dekker JH, Siemonsma PC et al (2004) Reproducibility of plantar pressure
measurements in patients with chronic arthritis: a comparison of one-step, two-step, and three-
step protocols and an estimate of the number of measurements required. Foot Ankle Int
25:739–744
Velasco R (2012) Fußdeformitäten im Kindesalter. Pädiatrie 9(17):2
Westberry DE, Davids JR, Anderson JP et al (2013) The operative correction of symptomatic flat
foot deformities in children: the relationship between static alignment and dynamic loading.
Bone Joint J 95-B:706–713
Yan SH, Zhang K, Tan GQ et al (2013) Effects of obesity on dynamic plantar pressure distribution
in Chinese prepubescent children during walking. Gait Posture 37:37–42
Zhao D, Li H, Zhao L et al (2014) Results of clubfoot management using the Ponseti method: do the
details matter? A systematic review. Clin Orthop Relat Res 472:1329–1336
Assessing Clubfoot and Cerebral Palsy by
Pedobarography
Julie Stebbins
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Data Collection Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Footprint Masking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Pressure Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Current Uses of Pedobarography in Children with Cerebral Palsy and Clubfoot . . . . . . . . . . . . . . 7
To Assess the Efficacy of Specific Interventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
To Classify Feet into Specific Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
To Correlate Findings with Other Measurement Modalities and Clinical Assessments . . . . 8
To Characterize a Specific Group of Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Reliability of Pedobarography in Children with Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Current Challenges and Limitations of Pedobarography in Children with Foot
Deformity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Abstract
The use of pedobarography to assess foot deformity in children with cerebral
palsy or clubfoot is becoming increasingly prevalent. Given the unique chal-
lenges posed by these populations, it is critical that a standardized protocol is used
when collecting pressure data. A number of factors need to be considered when
developing such a protocol, including the physical setup of the system, as well as
J. Stebbins (*)
Oxford Gait Laboratory, Oxford University Hospitals NHS Foundation Trust, Oxford, UK
Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of
Oxford, Oxford, UK
e-mail: Julie.Stebbins@ouh.nhs.uk

DOI 10.1007/978-3-319-30808-1_37-1
2 J. Stebbins
the process for acquiring the data. Currently, pedobarography is used in four
broad areas for children with foot deformity, including assessing the efficacy of
specific interventions, classification of feet into specific categories, correlation of
findings with other measurement modalities, and characterization of different
groups of patients. In order to reliably use pressure data to aid treatment planning,
further research is required to link pressure findings to specific clinical and
functional limitations in children with foot deformity. In addition, the predictive
value of pedobarography in terms of treatment outcomes is yet to be established.
It is recommended that future work focus on these key areas.
Keywords
Cerebral palsy • Clubfoot • Pedobarography • Plantar pressure • Foot deformity •
Children
Introduction
There are many conditions in childhood that lead to deformity of the foot. Foot
deformity in childhood is a nontrivial problem and has been linked to many other
physiological and psychosocial deficits including impaired overall quality of life
(Kothari et al. 2014, 2015). This is hardly surprising, since foot deformity impacts on
mobility and participation, and these aspects of daily living are closely related to
overall quality of life. Given the potentially significant impact that deformity of the
foot can have on a child’s well-being, it is important to be able to assess the foot
appropriately and therefore prescribe optimal treatment.
There are a number of techniques available that can provide objective assessment
of the degree and type of the deformity of the foot, as well as giving an indication as
to the efficacy of any intervention undertaken. However, the majority of these
assessments are static, or at best, nonfunctional, in nature. Advances in motion
capture technology and modeling have recently promoted the use of multi-segment
foot models to characterize dynamic function of the foot (for example, Stebbins et al.
2006). However, this methodology requires expensive equipment, a dedicated
laboratory, and staffing with specific expertise. Pedobarography, on the other hand,
can be relatively inexpensive, does not require a dedicated room, and is generally
easy to use. It is therefore becoming a popular choice for dynamic assessment of foot
function. Pedobarography is an objective measure which characterizes the loading
patterns beneath the sole of the foot through the use of pressure-sensing devices
located either inside the shoe, or embedded within the floor. It has been used for
several decades to assess foot function in adult pathology, such as diabetes (Bus and
Valk 2008) and rheumatoid arthritis (Van Der Leeden et al. 2006). More recently, the
use of pedobarography to assess pediatric foot deformity has become more wide-
spread. Despite the proliferation of this technology, there is little consensus on
protocols for data collection and analysis, interpretation of the findings, or how
pressure data can be used to aid clinical decision-making.
Assessing Clubfoot and Cerebral Palsy by Pedobarography 3
Two of the most common conditions leading to foot deformity in childhood are
idiopathic clubfoot and cerebral palsy. This chapter will therefore focus on assess-
ment in these two pathologies. In particular, protocols for data collection, data
reduction and analysis techniques, interpretation of data, and impact on clinical
decision-making will be covered. The chapter will be limited to assessment of
pressure during walking gait and will focus primarily on barefoot measurement
through the use of pressure plates or mats.
State of the Art
Pedobarography is currently used in four broad areas in relation to pediatric foot

deformity: (1) to assess efficacy of specific interventions, (2) to classify feet into
specific categories, (3) to correlate findings with other measurement modalities (e.g.,
radiographic measures) and clinical assessments, and (4) to characterize specific
groups of patients (e.g., following an intervention, such as a specific surgical
technique). The evidence in the literature suggests that pedobarography can be
accurate and effective when used for any of these purposes. Conversely, there is
limited evidence relating pressure data to clinical and functional deficits in children
with cerebral palsy or clubfoot, or to prediction of treatment outcomes in these
populations.
Methodology
Data Collection Protocols
There are two main types of pedobarographic data that may be collected: in-shoe and
through the use of a pressure plate or mat. The advantage of in-shoe data collection is
that multiple steps may be obtained relatively easily. However, in cases of foot
deformity, barefoot data is generally considered most relevant. Collecting
pedobarograhic data from children using a pressure mat or plate can be a challenging
proposition. Ensuring that the foot contacts with the plate, while avoiding targeting,
can be difficult to achieve.
Very few authors describe the protocol employed to obtain pressure data when
reporting on studies involving pediatric foot deformity. This lack of information
inhibits interpretation of findings and exacerbates the issue of non-standardization.
MacWilliams and Armstrong (2000) suggested the following recommendations for
recording pedobarographic data in children:
– The walkway should be at least 10 m in length with the pressure mat in the middle
and offset to one side.
– The subject should be able to walk a minimum of five steps prior to stepping on
the mat.
– The subject should be instructed to look straight ahead.
4 J. Stebbins
– An observer should be present to judge if the gait is natural, the subject has
achieved steady-state walking, and the pressure mat has not been targeted.
– A minimum of three successful trials should be recorded.
The authors also state that there is no requirement to conceal the pressure mat
from the participant, provided an observer ensures that the pressure mat is not
targeted during walking.
Various protocols have been described which dictate the number of steps required
prior to reaching the pressure plate. The “mid-gait” or “steady-state” method
assumes that data should not be obtained during gait initiation or termination.
Generally, four to five steps must be taken prior to arriving at the pressure plate,
along with several steps following stepping on the plate, in order to fulfill this
requirement. While this is frequently successful in the adult population, the more
variable gait of children often renders this method impractical. In addition, children
with foot deformity commonly exhibit limited walking endurance and so may find
the mid-gait method too taxing, particularly if multiple walking trials are required.
Other methods such as the one-step, two-step, and three-step techniques have also
proven to be reliable (Wearing et al. 1999; Van Der Leeden et al. 2016) and are
generally preferable when assessing children with foot deformity, particularly in the
presence of learning difficulties.
There is some disagreement in the literature as to the number of trials required to
record representative walking. Sinclair et al. (2009) found that it was necessary to
collect up to 20 walking trials in order obtain five trials that were representative of
the subject’s typical walking pattern. There is little evidence available which con-
tributes to the decision about the required number of footprints for this population.
Where this is reported, the consensus appears to be that a minimum of four footprints
should be collected per foot.
It should be noted that pressure data is highly dependent on walking speed (Herd
et al. 2004), so this should either be taken into account when interpreting the data or
standardized during data collection. Since pressure data is also influenced by age,
weight, and height, this should also be recorded and the data normalized or at least
stratified according to this information (Herd et al. 2004).
Pedobarographic data is known to exhibit a high degree of intra-subject variabil-
ity, compared to other sources of gait data. Standardizing the data collection protocol
is key to removing extraneous sources of variability. The protocol should specify the
number of steps to be taken prior to reaching the pressure plate, the number of trials
required, the definition of what constitutes a “successful” trial, and any instructions
to be given to the subject.
In addition, local institutions should carry out repeatability assessment on a
regular basis to determine the error in measurement due to intersession variance.
This is particularly important following changes in software, hardware, physical
environment, or local staffing. Confidence bands for each pressure parameter can
then be reported, which are vital for interpreting data collected prior to and following
intervention, or collected over a period of time.
In summary, collection of pedobarographic data in children with foot deformity

poses unique challenges. Standardizing the data collection protocol is critical to
obtaining meaningful information. When publishing studies where pedobarographic
data were collected, authors should ensure the specific protocol adopted is included
in the report.
Footprint Masking
Once the pressure data has been collected, the most common next step is to
subdivide the footprint into separate areas using a template called a “mask.” This
can be performed using different techniques including manual masking, geometric
masking, and anatomical masking. Manual masking requires the user to view each
footprint and manually divide the foot up into different anatomical regions. It is
extremely labor intensive and is now rarely used since most software programs
provide automated solutions. Geometric masking involves dividing the footprint into
pre-defined regions based on criteria such as percentages of the width and length of
the foot. It is by far the most common method employed and is effective when the
foot shape is near normal. In cases where deformity of the foot is present, such as
clubfoot and cerebral palsy, automated geometric masking does not always produce
valid results (Stebbins and Giacomozzi 2012). In such cases, anatomy-based
masking may be preferable. This entails synchronization with a motion capture
system and masking the footprint through the use of markers attached to the foot
(Stebbins et al. 2005). Details of this methodology are described in a separate
chapter.
Selection of the most appropriate mask to be used should be dictated by the type
of foot pathology present and the research or clinical question to be answered.
Definitions of subareas differ widely within the literature. For children with cerebral
palsy, reported pressure masks contain between five and ten different areas (Chang
et al. 2002; Femery et al. 2002; Kadhim et al. 2012b; Park et al. 2006; Riad et al.
2007; Son et al. 2015; Stebbins et al. 2005). Most divide the heel and midfoot into
medial and lateral halves and then variously partition the forefoot into between two
and five areas, with the hallux and lesser toes either grouped together or considered
separately. Given the frontal plane nature of foot deformity in this population, it is
logical to ensure medial and lateral halves of the foot are considered separately. In
addition, the foot should at least be divided longitudinally into heel, mid-, and
forefoot regions, to capture the effects of alterations in sagittal plane motion at the
ankle.
Children with clubfoot present with a range of foot pathologies, dependant on the
type of intervention received as an infant and in subsequent years, as well as a range
of other factors. Pressure masks reported in the literature for this population range
from three to ten different areas (Cooper et al. 2014; El-Shamy et al. 2013; Favre
et al. 2007; Hee et al. 2001; Henn et al. 2008; Holt et al. 2015; Huber and Dutoit
2004; Jeans and Karol 2010; Salazar-Torres et al. 2014; Sinclair et al. 2009; Thometz
et al. 2005). Similar to cerebral palsy, pressure masks for children with clubfeet
6 J. Stebbins
commonly divide the heel and midfoot into medial and lateral components. The
forefoot is often divided into three or more areas, with the hallux and lesser toes
considered separately. Given the complexity of the clubfoot deformity, it is reason-
able to consider dividing the pressure print into a larger number of subareas
compared to other types of foot deformity. There are commonly different (even
opposite) deformities present in the rear, mid-, and forefoot regions in this popula-
tion, and so as a minimum, these areas should be considered separately.
In summary, pressure masking is a key component in obtaining meaningful
pedobarographic data in children with clubfoot and cerebral palsy. The number
and definition of subareas to be assessed should reflect the type of foot deformity
present, as well as the clinical or research question to be answered. In particular,
careful attention should be paid to how the mask is generated, as automated,
geometric masking may produce invalid results where foot deformity is present.
Pressure Variables
Once the pressure footprint has been appropriately masked, the next stage of the
process is to generate variables of interest. This can include peak or average
pressure, peak or average force, pressure and force time integrals, impulse, and
contact area. Each variable can reflect the entire foot as well as each local region
within the mask. Additionally, ratios of any of the above variables between different
mask areas can be calculated. Finally, the path of the center of pressure during the
entire stance phase can also be presented.
For children with cerebral palsy, a commonly reported pressure variable is the
coronal plane pressure index (CPPI) which reports the ratio of medial to lateral peak
pressure across the mid- and forefoot regions of the foot (Kadhim et al. 2012b). The
ratio of forefoot to rearfoot pressure is another typical measure (Femery et al. 2002),
along with impulse in specific areas of the foot (Kadhim and Miller 2014; Riad et al.
2007). In children with clubfoot deformity, peak and average pressure in specific
regions of the foot are more commonly reported (El-Shamy et al. 2013; Favre et al.
2007; Hee et al. 2001; Henn et al. 2008; Holt et al. 2015; Liu et al. 1999). In addition,
ratios between peak pressure in medial and lateral regions (Herd et al. 2008; Jeans
and Karol 2010; Salazar-Torres et al. 2014) and the forefoot compared to the heel
(Ramanathan et al. 2009) are also described. Contact area and the path of the center
of pressure have also been used to assess children with clubfeet (Holt et al. 2015;
Huber and Dutoit 2004; Jeans and Karol 2010; Liu et al. 1999).
As with deciding on appropriate pressure masks, choosing optimal pressure
variables is highly dependent on the clinical or research question being asked.
Many pressure variables are inter-related and so more is not necessarily better. In
fact, having too many pressure variables is likely to be counterproductive as
statistical tests mostly assume independence of different variables, and so this may
lead to a type I error. Peak pressure has proven to be a reliable variable that has the
ability to predict sites of ulcer development in adults with diabetes (Bus and Valk
2008) and therefore objectively aids treatment planning. No such predictive
relationship has thus far been discovered for any pressure variable in children with
cerebral palsy or clubfoot. Pedobarography in these populations has only been
shown to discriminate pathological from typically developing feet and to quantify
the effects of intervention. Variables should therefore be selected that have proven to
be reliable in one or both of these areas, as well as in the specific population of
interest.
In summary, many different pressure variables have been reported in the literature
for characterizing feet in children with cerebral palsy or clubfoot. When deciding on
the most appropriate variables to select, care should be taken to match (and limit) the
variables to the research or clinical question. Evidence linking specific pressure
variables to development of pathology or treatment outcomes is currently lacking, so
the choice should be made based on ability of the variable to distinguish pathological
feet from typically developing feet and/or to quantify treatment outcomes. Prospec-
tive research linking specific pedobarographic variables to treatment outcomes is
needed in these populations.
Current Uses of Pedobarography in Children with Cerebral Palsy

and Clubfoot
Currently, pedobarography is utilized in four key areas for children with cerebral
palsy and clubfoot. These are outlined below.
To Assess the Efficacy of Specific Interventions
In children with cerebral palsy, pedobarography has been used to assess the outcome
of specific surgical interventions at the foot and ankle, including correction of
equinus deformity (Park et al. 2006; Abousamra et al. 2015) as well as planovalgus
feet (Kadhim et al. 2012a; Park et al. 2008). The effect of botulinum toxin has also
been determined using pedobarography (Bennett et al. 2007). Assessment of treat-
ment outcomes in clubfoot is less common due to the young age at which correction
of the foot deformity generally occurs, meaning that baseline data is unobtainable.
However, the outcome following secondary interventions has been assessed by some
authors (El-Shamy et al. 2013; Jeans et al. 2014).
Overall, pedobarography has been shown to be capable of distinguishing between
feet prior to and following surgical intervention in both children with cerebral palsy
and those with clubfoot. It therefore appears to be an appropriate method for
objectively quantifying the outcome of surgery to the foot and/or ankle in these
populations. Further research is required to determine if it can also detect changes
following other interventions such as casting or physiotherapy.
8 J. Stebbins
To Classify Feet into Specific Categories
Another common usage for pedobarography is to classify children into different

categories, based on their pressure footprint. Chang et al. (2002) successfully
distinguished between different levels of severity of foot deformity in children
with cerebral palsy. In children with clubfoot, pedobarography has been employed
to categorize children into different foot types (Herd et al. 2008), as well as the level
of severity of foot deformity present (Ramanathan et al. 2009).
While it seems feasible to classify the level and type of foot deformity in these
populations using pedobarography, the link to treatment planning and outcomes is
yet to be established. Further work is needed to connect the different foot types to
optimal treatment recommendations.
To Correlate Findings with Other Measurement Modalities

and Clinical Assessments
Pedobarographic data have also been correlated with other measures of foot defor-
mity including radiographs. In children with cerebral palsy and planovalgus feet,
Kadhim et al. (2012a) found that some pedobarographic measures correlated with
specific radiographic findings. A similar result has been reported in children with
clubfoot deformity (Oto et al. 2011; Thometz et al. 2005). However, a significant
percentage of the variance in pressure data was unaccounted for by radiographic
findings.
Correlations have also been assessed between pedobarography and clinical
assessments of foot deformity, including spasticity (Femery et al. 2002) and ankle
power (Riad et al. 2009) in cerebral palsy, and the degree of passive foot pronation
available (Huber and Dutoit 2004) in clubfoot. In each case, significant correlations
were found between the clinical measure described and specific pressure variables.
It is clear that a degree of relationship exists between dynamic function of the foot
(as measured by pedobarography) and static alignment (as measured by radio-
graphs). However, many authors have also demonstrated that the two are not
equivalent and that the foot functions differently in standing when compared to
walking. This also holds true when comparing walking (a dynamic weight-bearing
activity) to passive clinical assessments. Establishing relationships in how the foot
behaves during walking, standing, and passive motion may help to elucidate funda-
mental biomechanics of the foot. Using one modality to predict function of the foot
in an entirely different modality, however, should be viewed with caution, as the
current evidence suggests that the relationships are far from simple.
To Characterize a Specific Group of Patients
The final key area where pedobarography has proven useful in children with foot
deformity is to characterize specific groups of patients. This is particularly the case in
clubfoot, where obtaining baseline data is impractical. Bensahel et al. (1987) were
the first to use pedobarography to characterize the feet of children with clubfoot
following surgery which consisted of a posteromedial release. Others have subse-
quently measured a similar population and found that there is invariably a degree of
residual deformity present (Hee et al. 2001; Holt et al. 2015; Liu et al. 1999).
Similarly, patients who were treated conservatively as infants also exhibited a degree
of under or overcorrection when measured in later childhood (Holt et al. 2015; Jeans
and Karol 2010; Sinclair et al. 2009).
Other authors have found that pedobarography is useful to describe the contra-
lateral foot in unilateral clubfoot (Cooper et al. 2014; Favre et al. 2007). The data
revealed that contrary to expectation, the contralateral foot did not show normal
loading patterns.
It has been demonstrated by several authors that pedobarography is able to
characterize children following specific interventions. The treatment groups identi-
fied are distinguishable from age-matched, typically developing cohorts, suggesting
that current treatment regimens may be less than optimal. Pedobarography may
therefore be useful in identifying where treatment outcomes are below expectation
and thereby improve treatment planning.
Reliability of Pedobarography in Children with Cerebral Palsy
Relatively little research has been invested into establishing the reliability of
pedobarography data in children with cerebral palsy or clubfoot. Research conducted
at our institution suggested that peak force was a more reliable measure than peak
pressure in children with cerebral palsy (Stebbins et al. 2005). One other study
investigated reliability of pedobarographs in children with cerebral palsy (five
children aged 5–13 years) (Riad et al. 2007). They reported significantly more
variability at the medial midfoot in children with cerebral palsy compared to
typically developing peers, with no difference in the other foot regions. Inter- and
intra-rater reliability was found to be good to excellent. They concluded that
pedobarography is reliable in this population.
Given the very limited evidence available indicating the reliability of
pedobarography in these populations, individual institutions and departments are
urged to assess and be aware of the repeatability of their own local hardware,
software, and measurement techniques. This information should be taken into
account when interpreting pedobarographic findings.
Current Challenges and Limitations of Pedobarography

in Children with Foot Deformity
Collecting, processing, analyzing, and interpreting pressure data from children with
foot deformity can be a challenging task. Lack of walking endurance, variable
walking patterns, and ability to comprehend verbal instructions are all potential,
10 J. Stebbins
inherent obstacles to obtaining reliable data. Choosing an appropriate mask, and the
method to generate it, poses a further problem when deformity of the foot is present.
Finally, selecting the right variables from the vast array available can also cause
difficulties. But perhaps the greatest challenge in this process is to interpret the
findings into something meaningful and then translate this into an appropriate
treatment plan. There are three main hurdles that need to be overcome in this process.
The first is to define what constitutes “normal” pressure data. The range of values
reported in the literature obtained from typically developing subjects varies greatly.
This is especially the case when the data is collected from children. This discrepancy
is due to a range of factors, including differences in hardware, data collection
protocols, masking techniques, and reporting procedures. There is a scarcity of
evidence in the literature supporting the use of age-matched, typically developing
reference data when interpreting pressure findings, particularly in the context of
clinical decision-making. While this is a common practice for other types of gait
data, such as kinematics and kinetics, it is yet to be established whether this method
can be employed with pedobarography. One study has investigated this in children
with clubfeet (Stebbins et al. 2014) and found that these children could be reliably
distinguished from age-matched, typically developing peers. While this is promis-
ing, more research is needed in this area.
The second major hurdle involves establishing the link between specific
pedobarographic findings and clinical problems. Several studies have identified
abnormal features in the pressure patterns of children with clubfoot, as well as
those with cerebral palsy. However, the significance of these abnormal features in
terms of clinical relevance is less apparent.
The final hurdle is to connect pressure findings to treatment planning in these
populations. Prospective studies linking specific treatment to changes in
pedobarography for these populations are currently lacking. Without this informa-
tion, it is difficult to know how to interpret findings and then what intervention to
suggest based on the pressure data. Further research in this area is required.
Conclusion
The use of pedobarography for assessing foot deformity in children is becoming

increasingly common. Despite its prevalence, there continues to be a lack of
standardization, which makes interpretation of findings difficult.
Standardizing local protocols for data collection is key to obtaining accurate and
meaningful data. These protocols should specify requirements such as the physical
setup of the walkway and pressure plate, the number of steps to be taken, constraints
on walking speed, definitions of “successful” walking trials, and the number or trials
required. In addition, knowledge of local repeatability of specific pressure parame-
ters is vital for interpreting outcomes of interventions. When reporting
pedobarographic studies in the literature, it is important to include details of the
specific methodology employed for collecting the data. Generating appropriate
pressure masks is another important aspect of pedobarography. The number and
definition of subareas to be assessed should reflect the type of foot deformity present,
as well as the clinical or research question to be answered. In particular, careful
attention should be paid to how the mask is generated, as automated, geometric
masking may produce invalid results where foot deformity is present. The choice of
pressure variables selected for analysis should be limited to those which answer the
research or clinical question of interest and should be based on the ability of the
variable to distinguish pathological feet from typically developing feet and/or to
quantify treatment outcomes.
Currently, in children with cerebral palsy or clubfoot, pedobarography is used in
four broad areas: to assess efficacy of specific interventions, to classify feet into
specific categories, to correlate findings with other measurement modalities and
clinical assessments, and to characterize specific groups of patients. Evidence in
the literature suggests that pedobarography can reliably distinguish pathological feet
from typically developing feet, as well detect changes in pressure distribution
following specific surgical interventions.
There is currently limited evidence available in the literature linking specific
pressure parameters to relevant clinical findings in children with clubfoot or cerebral
palsy. This inhibits interpretation of the data and makes treatment planning based on
pedobarography challenging. Further work is required in this area, to establish the
relationship between abnormal pressure data and prediction of treatment outcomes.
Cross-References

▶ Fusion of Foot Pressure and Foot Kinematics Measurements for Medical
Applications
▶ Kinematic Foot Models for Instrumented Gait Analysis
▶ Plantar Pressure
▶ Pressure Platforms
▶ Talipes Equinovarus and Hereditary Motor Sensory Neuropathy
▶ The Importance of Foot Pressure in Diabetes
▶ The Use of Low Resolution Pedobarographs
References
Abousamra O et al (2015) The effect of plantar flexor lengthening on foot pressure in children with
cerebral palsy. Gait Posture 42:S15
Bennett D et al (2007) Use of a dynamic foot pressure index to monitor the effects of treatment for
equinus gait in children with cerebral palsy. J Pediatr Orthop 27:288–294
Bensahel H et al (1987) Surgery in residual clubfoot: one-stage medioposterior release. J Pediatr
Orthop 7:145–148
12 J. Stebbins
Bus SA, Valk GD (2008) The effectiveness of footwear and offloading interventions to prevent and
heal foot ulcers and reduce plantar pressure in diabetes: a systematic review. Diabetes Metab Res
Rev 24:162–180
Chang HC, Miller F, Schuyler J (2002) Dynamic pedobarograph in evaluation of varus and valgus
foot deformities. J Pediatr Orthop 22:813–818
Cooper A et al (2014) The contralateral foot in children with unilateral clubfoot, is the unaffected
side normal? Gait Posture 40:375–380
El-Shamy SM et al (2013) Effect of neuromuscular electrical stimulation on foot pressure distribu-
tion in congenital clubfoot. J Am Sci 699(6):178–183, Available at: http://www.jofameri
canscience.org
Favre P et al (2007) The contralateral foot in children with unilateral clubfoot a study of pressures
and forces involved in gait. J Pediatr Orthop 27:54–59
Femery V et al (2002) Measurement of plantar pressure distribution in hemiplegic children: changes
to adaptive gait patterns in accordance with deficiency. Clin Biomech 17:406–413
Hee TH, Hin L, Lee GS (2001) Gait and pedobarographic patterns of surgically treated clubfeet.
J Foot Ankle Surg 40(5):287–294
Henn FR et al (2008) Subtalar release in clubfeet: a retrospective study of 10-year outcomes. Foot
Ankle Int 29(4):390–395
Herd F, Macnicol M, Abboud RJ (2004) The need for biomechanical evaluation in the assessment of
clubfoot. Foot 14:72–76
Herd F et al (2008) Foot pressure in clubfoot-the development of an objective assessment tool. Foot
18:99–105
Holt JB et al (2015) Long-term results of tibialis anterior tendon transfer for relapsed idiopathic
clubfoot treated with the ponseti method a follow-up of thirty-seven to fifty-five years. J Bone
Joint Surg (Am) 97:47–55
Huber H, Dutoit M (2004) Dynamic foot-pressure measurement in the assessment of operatively
treated clubfeet. J Bone Joint Surg (Am) 86-A(6):1203–1210
Jeans KA, Karol LA (2010) Plantar pressures following ponseti and French physiotherapy methods
for clubfoot. J Pediatr Orthop 30:82–89
Jeans KA et al (2014) Plantar pressures following anterior tibialis tendon transfers in children with
clubfoot from the texas scottish rite hospital for children. J Pediatr Orthop 34:552–558
Kadhim M, Miller F (2014) Crouch gait changes after planovalgus foot deformity correction in
ambulatory children with cerebral palsy. Gait Posture 39:793–798
Kadhim M, Holmes L, Church C et al (2012a) Pes planovalgus deformity surgical correction in
ambulatory children with cerebral palsy. J Child Orthop 6:217–227
Kadhim M, Holmes L, Miller F (2012b) Correlation of radiographic and pedobarograph measure-
ments in planovalgus foot deformity. Gait Posture 36:177–181
Kothari A et al (2014) Health-related quality of life in children with flexible flatfeet: a cross-
sectional study. J Child Orthop 8(6):489–496
Kothari A et al (2015) Gait & posture the relationship between quality of life and foot function in
children with flexible flatfeet. Gait Posture 41(3):786–790, Available at: http://dx.doi.org/
10.1016/j.gaitpost.2015.02.012
Liu X et al (1999) Dynamic plantar pressure in the treated clubfoot: an orthopaedic engineering 460
study. In: First joint BMES/EMBS conference, IEEE Xplore. Engineering in Medicine and
Biology - Proceedings of t meeting of Biomedical Engineering Society, p 527. doi: 10.2209/
IEMBS.1999.802605.
MacWilliams BA, Armstrong PF (2000) Clinical applications of plantar pressure measurement in
pediatric orthopedics. In: Pediatric gait, 2000. A new millennium in clinical care and motion
analysis technology. Chicago, pp 143–150. doi: 10.1109/PG.2000.858886
Oto M et al (2011) Correlation between selective pedobarographic and radiographic measures in the
assessment of surgically treated CTEV patients. Joint Dis Relat Res 22(3):145–148
Park ES et al (2006) Dynamic foot pressure measurements for assessing foot deformity in persons
with spastic cerebral palsy. Arch Phys Med Rehabil 87:703–708
Park KB et al (2008) Changes in dynamic foot pressure after surgical treatment of valgus deformity
of the hindfoot in cerebral palsy. J Bone Joint Surg (Am) 90:1712–1721
Ramanathan AK et al (2009) A new scoring system for the evaluation of clubfoot: the IMAR-
Clubfoot scale. Foot 19:156–160
Riad J et al (2007) Reliability of pediobarographs for paediatric foot deformity. J Child Orthop
1:307–312
Riad J, Henley J, Miller F (2009) Does footprint and foot progression matter for ankle power
generation in spastic hemiplegic cerebral palsy? Acta Orthop Traumatol Turc 43(2):128–134
Salazar-Torres JJ et al (2014) Plantar pressures in children with congenital talipes equino varus-A
comparison between surgical management and the Ponseti technique. Gait Posture 39:321–327
Sinclair MF et al (2009) Pedobarographic analysis following ponseti treatment for congenital
clubfoot. Clin Orthop Relat Res 467:1223–1230
Son SM, Park IS, Yoo JS (2015) Short-term effect of botulinum toxin a injection on spastic
equinovarus foot in cerebral palsy patients: a study using the foot pressure measurement system.
Ann Rehabil Med 39(1):1–9
Stebbins J, Giacomozzi C (2012) Clinical relevance of anatomy-based plantar pressure analysis in
clubfoot. In: Proceedings of expert scientific meeting. http://novel.de/novelcontent/images/files/
ESM-2012_proceedings-abstracts.pdf.
Stebbins JA et al (2005) Assessment of sub-division of plantar pressure measurement in children.
Stebbins J et al (2006) Repeatability of a model for measuring multi-segment foot kinematics in
children. Gait Posture 23:401–410
Stebbins J, Way L, Giacomozzi C (2014) Integrating pressure and motion capture to assess
deviations from normal in the clubfoot population. Gait and Posture 42 (Suppl 3):S28–S29
Thometz JG et al (2005) Correlation of foot radiographs with foot function as analyzed by plantar
pressure distribution. J Pediatr Orthop 25(2):249–252
Van Der Leeden M et al (2006) Forefoot joint damage, pain and disability in rheumatoid arthritis
patients with foot complaints: the role of plantar pressure and gait characteristics. Rheumatology
45:465–469
Van Der Leeden M et al (2016) Reproducibility of plantar pressure measurements in patients with
chronic arthritis: a comparison of one-step, two-step, and three-step protocols and an estimate of
the number of measurements required. Foot Ankle Int 25(10):739–744
Wearing SC et al (1999) Technical note a comparison of gait initiation and termination methods for
obtaining plantar foot pressures. Gait Posture 10:255–263
Low Density Pedoboragraphy as a Gait
Analysis Tool
Ruopeng Sun, Tyler A. Wood, and Jacob J. Sosnoff
Abstract
Precise and objective evaluation of gait provides important information about an
individual’s overall health and can be used to detect symptoms of motor impair-
ment, determine appropriate therapeutic procedures, and monitor rehabilitation
progress. There are various gait analysis techniques currently available, including
the three-dimensional motion tracking and the low density pedobarography, etc.
The relative low cost, high accuracy and consistency, and automated data analysis
features make the low density pedobarography technique an ideal platform for
quantifying the spatial and temporal gait characteristics in various populations.
The following chapter will review the use the low density pedobarography in
conducting gait and balance assessment. Topics covered in this chapter will
include the primary outcome measures of gait analysis using low density
pedobarography, the validity and reliability of the equipment, the applications
of the technique in clinical populations, guidelines for use, and the potential
applications in future research and clinical environment.
Keywords
Low density pedobarograph • Gait analysis • Spatiotemporal gait parameters
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Gait Assessment Using Pressure-Sensing Walkways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Validity and Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Current State of Clinical and Research Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
R. Sun (*) • T.A. Wood • J.J. Sosnoff

Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign,
Urbana, IL, USA
e-mail: rusun@illinois.edu; tawood2@illinois.edu; jsosnoff@illinois.edu

DOI 10.1007/978-3-319-30808-1_38-1
2 R. Sun et al.
Assessments of Other Dynamic Locomotion and Postural Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Cross References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
Walking is a fundamental form of movement that is central to human existence. It

depends on the satisfactory functioning of the neuromuscular and skeletal system.
Impairments in these systems are likely to lead to abnormal gait, decreased stability,
and reduced mobility (Whittle 1996). The assessment of gait performance provides
critical information about individual’s overall health and has been referred to as a
sixth “vital sign” (Fritz and Lusardi 2009). Changes in gait characteristics are often
too small to be detected by traditional clinical observation. Yet, the clinical conse-
quences of such subtle change can be immense (Bridenbaugh and Kressig 2010),
making the precise and objective measurement of gait a necessity. More specifically,
measurements of the spatial and temporal parameters of walking (e.g., step length,
step width, timing of gait events, etc.) provide a comprehensive profile on individ-
ual’s walking characteristics. Such information can be used to detect symptoms of
motor impairment, determine appropriate therapeutic procedures, and monitor reha-
bilitation progress (Bilney et al. 2003). Abnormalities in gait patterns, such as
reduced gait speed and step length, have been linked with musculoskeletal (Keefe
and Hill 1985) and neurological disorders (Morris et al. 2001; Sosnoff et al. 2012) as
well as increased risk of falling in general aging population (Verghese et al. 2009).
Particularly, gait speed has been demonstrated to be a reliable, valid, sensitive, and
specific measure that predicts future health status and functional decline including
hospitalization, discharge disposition, and mortality (Fritz and Lusardi 2009).
State of the Art
A variety of techniques have been used for conducting gait analysis in clinical and
research settings. Typical low cost methods for clinical gait evaluation include simple
visual observation (Miyazaki and Kubota 1984), video recording (Eastlack et al. 1991),
timed performance tests and paper walkways (Sekiya et al. 1997). For example, simple
visual observation requires clinicians to provide qualitative observation on partici-
pant’s gait performance. Although widely used, these visual observations lack preci-
sion and depend greatly on each clinician’s clinical experience. Although timed
performance tests such as timed up and go (Podsiadlo and Richardson 1991) are a
cornerstone of community based research, they lack precision and provide no infor-
mation about spatiotemporal markers of gait. The paper-and-pencil method requires
participants to have ink applied to the bottom of their feet and to walk along a paper
walkway. The spacing of the individual footprints is then measured and spatial
Low Density Pedoboragraphy as a Gait Analysis Tool 3
parameters of gait are then calculated. This approach is somewhat dated and time
intensive and lacks temporal resolution. The video-recorded observation method
requires experimenters to analyze the video recordings of participants walking and
identify individual gait events to determine timing. This approach provides limited
information concerning spatial information and has limited inter-rater reliability. Col-
lectively, these methods often lack accuracy and reliability in gait assessment.
In contrast, marker-based three-dimensional motion-tracking technology has
been proven to be the golden standard in conducting research on human gait analysis
(Baker 2006). The 3D motion-tracking system provides precise quantitative mea-
surement of gait including the temporal and spatial dynamic of foot placements, the
range of motion of joints, and the temporal and spatial dynamic of the center of mass
(COM). If coupled with kinetic measures, the muscular and joint forces can be
calculated with inverse modelling. It yields a high degree of precision and consis-
tency in measurement outcomes. Despite the precision of this approach, it has
numerous drawbacks including high cost, time consuming, technically difficult to
use, and labor intensive. Therefore marker-based tracking is not applicable in most
clinical settings (Cutlip et al. 2000). Another commonly used technique for gait
analysis is using body-mounted inertial measurement unit (IMU) for movement
tracking. Its wearable features enable longitudinal gait assessment in a free-living
environment. However, there are major trade-offs on the tracking accuracy, espe-
cially on the calculation of position and orientation from raw acceleration and
angular velocity data, due to the integration noise and drift error (Shull et al. 2014).
Low density pedobarograph walkways combine the precision of marker-based
motion capture with the ease of use of clinic measures. The embedded pressure-
sensing instrumented walkway, or low density pedobarograph walkway, was
designed to measure temporal and spatial gait parameters with accuracy comparable
to sophisticated motion analysis systems, with the additional advantages of marker-
less tracking, automated analysis, and enhanced portability. An example of an
instrumented walkway and its sample data outputs are depicted in Fig. 1. It is
important to note that the pressure-sensing walkway system discussed in this chapter
will be limited to the low resolution pedobarograph walkway (usually less than one
sensor per cm2) for the analysis of spatial and temporal gait parameters.
The primary outcome measures for gait assessment conducted with a low density
pedobarograph system are the spatial and temporal characteristics of gait, which
include stride speed, cadence (number of steps per min), step length, stride length,
stride width, step angle, swing time, stance time, double support time, and base of
support. Figure 2 illustrated the definitions of spatial and temporal parameters in a
typical gait cycle. Beyond standard parameters of gait, data on within-person
variability in gait kinematics, which is also an important marker of neuromuscular
function as well as a potential target for rehabilitation, can also be collected and
analyzed. These systems also provide symmetry measures, as derived from the
bilateral spatial and temporal gait parameters, which are important gait characteris-
tics, especially for stroke rehabilitation (Patterson et al. 2010).
It is worth noting that the embedded pressure sensors, unlike traditional force
platforms (e.g., force plates), cannot quantify horizontal or shear components of the
4 R. Sun et al.
Temporal 1 25.0
Duration: 4 sec.
Y Range: 25.00
[COP-COMe] Dist.
18.8
[COP-COMe] X Dist.
[COP-COMe] Y Dist.
12.5
6.3
0.0
9.00 10.00 11.00 12.00
Fig. 1 Example of an instrumented pressure-sensitive walkway system (Zeno™ Walkway) and its
sample data output
applied forces; thus, the center of pressure (COP) calculation algorithm by a

pedobarograph walkway is not the same as using a force plate system (Chesnin
et al. 2000). However, the benefit of registered pressure distribution from a matrix of
sensors is the ability to measure the spatial and temporal dynamics of COP under-
neath each foot fall (Chisholm et al. 2011).
There are currently several brands of low density pedobarograph walkways
®
commercially available including GAITRite by CIR systems, Inc. (Franklin, NJ),
®
Zeno™ Walkway by ProtoKinetics LLC. (Havertown, PA), FDM system by Zebris
Medical GmbH (Isny im Allgäu, Germany), Walkway™ system by Tekscan, Inc.
Fig. 2 Illustration of typical spatial and temporal gait parameters extracted from a gait cycle
®
(Boston, MA), and BTS P-WALK by BTS Bioengineering Corp. (Brooklyn, NY).
While different brands’ products have various configuration and software interfaces,
the underlined technique to extract the gait parameters are relatively similar. A
typical pedobarograph walkway system is a portable carpet-like walkway (size
ranging from 3 m by 0.5 m to 12 m by 1.2 m) that has pressure sensors embedded
in a grid-like pattern throughout its length. Each pressure-sensing unit measures the
vertical force applied on the sensor and registers the time and geometry of the
activation pattern to derive the gait event as well as the foot placement patterns.
The walkway also senses the relative arrangement between each object contact and
identifies footprint as well as assistive device contact (e.g., walker, cane, etc.) using
built-in algorithms. Subjects can walk over the carpet without the encumbrance of
markers, and data can be obtained and displayed automatically for every step
registered within the length of the walkway. Subjects can be tested with and without
shoes and even using assistance devices with minimal issues.
6 R. Sun et al.
Gait Assessment Using Pressure-Sensing Walkways
Validity and Reliability
When selecting a measurement tool, validity and reliability should be a key concern.
The concurrent validity of the pressure sensing walkway system was investigated
®
from the 1990s to early 2000s, by comparing the data recorded through GAITRite
walkway with other commonly available methods used in gait analysis (For detailed
description about the participant population, validation techniques and key obser-
vations from each study, see Table 1.). Earlier work compared the output of the
®
GAITRite system with the paper-and-pencil analysis and video-based analysis
(Selby-Silverstein and Besser 1999; McDonough et al. 2001). These investigations
demonstrated excellent agreement on spatial parameters between the paper-based
®
analysis and the GAITRite walkway, and on temporal parameters between the
®
video-based analysis and the GAITRite walkway (intra-class correlation coefficient
(ICC) > 0.95). The authors suggested that the lack of timing accuracy of the
traditional methods was the reason for the poor association in some temporal gait
parameters (ICC < 0.7). Bilney et al. (2003) reported high ICC between the
®
GAITRite and the Clinical Stride Analyzer (an insole footswitch-based gait track-
ing system, B&L Engineering) for gait speed, stride length, and cadence among
®
healthy adults (ICC above 0.99). Cutlip et al. (2000) compared the GAITRite
walkway to a video-based motion-tracking system (Peak performance technologies)
and reported strong associations for all gait parameters (Pearson product moment
coefficients above 0.94), albeit some systematic bias in step length and stride
velocity measures. Later work done by Webster et al. (2005) compared the averaged
and individual step parameters (recorded from older adults who had undergone knee
®
replacement surgery) between the GAITRite system and the Vicon motion analysis
system and found high degree of similarity across all measurements (ICC above 0.92
in both individual step and averaged gait parameters) between systems without any
systematic bias. Collectively, this body of validation research have led to the
®
conclusion that the low density pedobarograph walkway, specifically the GAITRite
system which is a valid tool for gait analysis compared with other accepted methods.
The test-retest reliability of the low density pedobarograph walkway has also
been examined. High level of inter-trial reliability for spatial temporal measurements
®
from GAITRite system has been reported in research conducted on healthy subjects
within the same day (Bilney et al. 2003) and in a week interval (Van Uden and Besser
2004). To evaluate whether the low density pedobarograph walkway was suitable for
conducting gait assessment on aging population, Menz et al. (2004) examined the
®
test-retest reliability of the GAITRite walkway among aging adults over a 2-week
period and found the system provided highly reliable measurements of gait param-
®
eters. The reliability of the GAITRite system has also been examined within other
populations, such as in typical developed children aged between 1 year old and
11 years old (Thorpe et al. 2005), in children with motor impairments (Wondra et al.
2007) and cerebral palsy (Sorsdahl et al. 2008), in patients with Huntington disease
(Rao et al. 2005), multiple sclerosis (Sosnoff et al. 2015), Parkinson’s disease
Table 1 Summary of validation studies using the pressure-sensing walkway

Author Subject/
(year) population Goal Gait parameters Results
McDonough A single Test the Gait speed, High agreement in
et al. (2001) healthy subject concurrent cadence, step spatial measures
(27 years old) validity between length, step time (ICC > 0.97)
® ®
GAITRite and between GAITRite
paper-and- and paper-and-
pencil method pencil method; high
and video-based agreement in
method temporal measures
(ICC > 0.95)
®
between GAITRite
and video-based
method
Cutlip et al. Healthy adults Test the Gait speed, step High correlations
(2000) (age range concurrent period, step among all five
21–26 years validity between length, stance kinematic
®
old) GAITRite and duration, and parameters (Pearson
video-based swing duration correlation
motion-tracking coefficient > 0.94).
(Peak Systematic
performance difference in step
technologies length (~3 cm) and
motus 3.1) gait speed (~0.1 m/s)
Bilney et al. Healthy adults Test the Gait speed, High agreement
(2003) (age range concurrent cadence, stride between systems for
21–71 years validity between length, single- all spatial measures
®
old) GAITRite and limb support (ICC > 0.9). Fair to
Clinical Stride time (SLS) and moderate
Analyzer double support association for the
(CSA). Test the percentage of SLS and DS%.
inter-trial gait cycle (DS%) Good inter-trial
reliability of the reliability among
®
GAITRite most gait
measures within parameters
same day period (ICC > 0.8)
Webster et al. Older adults Test the Gait speed, High degree of
(2005) who had concurrent cadence, step similarity between
undergone validity between length, step time systems on all gait
®
knee GAITRite and parameters
replacement a three- (ICC > 0.92). No
surgery (age dimensional systematic bias was
range motion analysis observed
54–83 years (Vicon-512,
old) Oxford Metrics)
(Nelson et al. 2002; Stover 2005; Chien et al. 2006), Alzheimer’s disease (Wittwer
et al. 2008), and stroke (Kuys et al. 2011; Lewek and Randall 2011; Wong et al.
2014; Cho et al. 2015). Detailed description about the participant population,
8 R. Sun et al.
validation techniques, and key observations from aforementioned studies are listed
in Table 2. Across all studies and populations, base of support and foot angle
measures were the least reliable (ICC range 0.2–0.8), possibly due to the spatial
resolution of system and the software algorithm for data extraction (Menz et al.
2004) and thus should be treated with caution.
Even though most of the aforementioned studies focusing on validity and reli-
®
ability on low density pedobarograph systems were conducted with the GAITRite
system, it is assumed that other systems are similarly valid and reliable as the
underlining technology is similar. The appropriateness of this assumption is not
clear.
Current State of Clinical and Research Application
The ease of use and automated analysis features make the low density
pedobarograph walkway systems a standard method for gait assessment across
various populations. Since the early validation studies, over 400 research publica-
tions have been conducted with the low density pressure-sensing instrumented
walkway systems. Gait assessment using these systems have been performed on
various populations including but not limited to infants (Garciaguirre et al. 2007),
children (Dusing and Thorpe 2007), young adults, middle-aged adults, and older
adults (Verghese et al. 2009), as well as individuals with various pathological
conditions (down syndrome (Wu et al. 2007), cerebral palsy (Rinehart et al. 2006),
attention-deficit/hyperactivity disorders (Papadopoulos et al. 2014), Tourette syn-
drome (Liu et al. 2014), traumatic brain injury (Katz-Leurer et al. 2008), leg amputee
and prosthetic gait (Highsmith et al. 2010), diabetes (Paul et al. 2009), multiple
sclerosis (Sosnoff et al. 2012), Parkinson’s disease (Chien et al. 2006), stroke
(Patterson et al. 2010), mild cognitive impairment (Verghese et al. 2007),
Alzheimer’s disease (Webster et al. 2006), and cerebellar ataxia (Schniepp et al.
2012). Investigations either quantified the existing gait deficits in the pathological
populations that may be used for detecting disease onset and progression or evalu-
ated the improvements of gait due to various therapeutic interventions. Pressure-
sensing walkway system has also been used as the standardized comparison for the
accuracy of gait parameters measured from other techniques, such as body-worn
sensors (Hartmann et al. 2009; Kim et al. 2015; González et al. 2016) and marker-
less motion tracking (Clark et al. 2013).
Guidelines
Although the validity and reliability of the pressure-sensing walkway have been well
documented, it is also important to implement a standard procedure for gait assess-
ments to ensure outcome measurements are reliable and comparable across different
studies. In order to enhance reproducibility of clinical gait measures and for better
®
comparability of outcomes using the GAITRite system, in 2006, the European
Table 2 Summary of the test-retest reliability studies using the pressure-sensing walkway
Author Subject/
Menz Young Test the inter-trial Gait speed, Good to excellent
et al. (22–40 years reliability of the cadence, step reliability for gait
®
(2004) old) and older GAITRite length, base of parameters
(76–87 years measurements support, and toe (ICC > 0.82), with
old) healthy over a two-week in/out angle the exception of
adults period base of support
and toe in toe out
angle in older
subjects
Van Healthy adults Test the inter-trial Gait speed, step Good to excellent
Uden (age range reliability of the length, stride reliability among
®
and 19–59 years old) GAITRite length, step time, spatial-temporal
Besser measurements swing time, stance gait measurements
(2004) over a one-week time, double (ICC > 0.89),
period support time, base except the base of
of support, and toe support
in/out angle (ICC = 0.79)
Thorpe Typical To determine the Gait speed, Moderate to good
et al. developed repeatability of the cadence, step reliability
®
(2005) children (age GAITRite length, base of (ICC > 0.6) for
range 1–11 years measurements in support, swing most of the spatial
old) health children time, double parameters. Poor
within same day support time, toe to fair reliability
period in/out angle on the base of
support and toe
in/out angle
Rao et al. Adults with To determine the Gait speed, stride High reliability of
(2005) Huntington’s reliability of the time, stride length, all parameters
®
disease (age GAITRite cadence, and base (ICC > 0.8)
range measurements in of support
35–55 years old) patients with
Huntington’s
disease within
same day period
Stover Adults with To determine the Gait speed, Good reliability
(2005) Parkinson’s reliability of the cadence, base of across all gait
®
disease (age GAITRite support, step parameters
range measurements in length, stride (ICC > 0.8)
49–85 years old) patients with length, and single/
Parkinson’s double support
disease within percentage of gait
same day period cycle
(continued)
10 R. Sun et al.
Table 2 (continued)
Author Subject/
Sorsdahl Children with To determine the Cadence, step Good reliability of
et al. cerebral palsy reliability of the length, stride most parameters
®
(2008) (age range GAITRite length, step with, (ICC > 0.7)
3–13 years old) measurements in single support except the step
children with time width
cerebral palsy
within same day
period
Wittwer Adults with To determine the Gait speed, High test-retest
et al. Alzheimer’s reliability of the cadence, step reliability across
®
(2008) disease (age GAITRite length, stride all gait parameters
range measurements in length, swing (ICC > 0.86)
70–91 years old) in patients with time, stance time,
Alzheimer’s base of support,
disease over a and toe in/out
week period angle
Kuys Adults admitted To determine the Gait speed, Good reliability
et al. for rehabilitation reliability of the cadence, step time, across all gait
®
(2011) following stroke GAITRite step length, and parameters
(mean age of measurements in stance phase (ICC > 0.72)
64 years old) stroke survivors duration
over a 2-day
period
Lewek Adults with To determine the Gait speed, step Excellent
and chronic reliability of the length asymmetry, reliability for all
®
Randall hemiparesis GAITRite stance time symmetry
(2011) resulting from symmetry asymmetry, swing measurements
stroke (mean age measurements in time asymmetry (ICC > 0.91)
of 56 years old) post-stroke
patients over a
10-day period
Wong Adults admitted To determine the Gait speed, step High intra- and
et al. for rehabilitation intra- and inter- time, step length, inter-rater
(2014) following stroke rater reliability of step width reliability
®
(mean age of the GAITRite (ICC > 0.90)
68 years old) measurements in across all
stroke survivors parameters except
within same day step width
period
Cho et al. Adults admitted To determine the Gait speed, Excellent
(2015) for rehabilitation reliability of the cadence, step reliability for all
®
following stroke GAITRite length, stride gait parameters in
measurements length single task
(continued)
Table 2 (continued)
Author Subject/
(mean age of during condition
52.5 years old) performance of (ICC > 0.98) but
single and dual not in dual task
task walking in condition (ICC
post-stroke range 0.69–0.90)
patients over a
2-day period
Sosnoff Adults with To determine the Gait speed, step High reliability for
et al. multiple reliability of gait time, step length, all gait parameters
(2015) sclerosis (age parameters cadence, base of (ICC > 0.90), with
range (measured by support, double the exception of
®
18–64 years old) GAITRite ) in support base of support
patients with percentage of gait (ICC = 0.56)
multiple sclerosis cycle
over a 6-month
period
®
GAITRite Network Group published the “guidelines for clinical applications of
spatio-temporal gait analysis in older adults” (Kressig and Beauchet 2006). Key
guidelines from this report include:
1. Measurements should be performed in a reproducible, well-lit environment.

2. Data collection should exclude any auditory or visual interference for
participants.
3. Participants should be allowed to walk in their own footwear that are not slipper
type or with heel height exceeding 3 cm. For follow-up gait analysis, subjects
should wear the same footwear as was worn at baseline test.
4. Safety measures should be provided in case of an imminent fall.
5. Steady state gait should be tested at different gait speeds (e.g., slow, normal, fast),
preferably in randomized order.
6. In order to achieve steady state walking, it is recommended to instruct participants
to start walking at least 2 m prior to reach the electronic walkway and stopping at
least 2 m beyond it.
7. Assistive devices used by participants, if necessary, should be documented by
its type.
8. In order to evaluate stride-to-stride variability, a minimum of three consecutive
gait cycles for both left and right side (i.e., a total of 6 foot falls) should be
registered in a single walk over trial.
Although such guidelines were specifically developed for gait assessments in

older adults, it is reasonable to apply them to measurements of gait in other clinical
populations.
12 R. Sun et al.
Assessments of Other Dynamic Locomotion and Postural Tasks
Although steady state linear walking is the most commonly used paradigm for gait
assessment, it is only one aspect of functional gait. Therefore, gait assessments in
other dynamic locomotion tasks, such as gait initiation, gait termination, and turning
have also be investigated. Traditionally, such dynamic tasks were often evaluated
with the three-dimensional motion-tracking system in a laboratory setting. Recently,
pressure-sensing walkways have also been used to conduct research in evaluation of
subtasks of gait. For instance, Wajda et al. (2015) examined step initiation in
multiple sclerosis utilizing a Zeno™ walkway. Specifically, the time from stimulus
onset to toe off of the swing foot (step initiation timing) was quantified. It was found
that the initiation timing was positively associated with the physiological fall risk
score in the sample. In a related investigation on planned gait termination in multiple
sclerosis patients (Roeing et al.), the time needed for the estimated center of mass
(COMe) to stabilize during stopping phase of gait termination was extracted using a
Zeno™ walkway and accompanying software. It was found that MS patients
displayed elevated stabilization times for gait termination. In an investigation exam-
ining effect of dual task on turning ability in stroke survivors and older adults
®
(Hollands et al. 2014), participants walked on a GAITRite walkway and made
90 degree turns under single and dual task condition. It was found that both groups
exhibited dual task decrements in turning ability (measured by longer time to turn,
higher variability in time to turn, and increased single support time during turning).
®
The recent introduction of the GAITRite CIRFace system, which utilizes a series of
wireless based square pressure mats, offers customizable walkway pattern that can
be used to test the gait characteristics of turning, obstacle crossing/avoidance, stair
ascending/descending, and other real-world tasks that have functional significance in
the quality of life. At time of publication there was no published work available
using the CIRFace system.
Postural balance, which is also a major index for the evaluation of functional
mobility and risk of falling among aging and pathological populations (Maki et al.
1994), can also be tested using the pedobarograph technique. The linear and
nonlinear measures of COP sway during quiet standing, (sway area, sway amplitude,
and velocity in anterior-posterior (AP)/medio-lateral (ML) direction, and the sample
entropy, etc.) have been shown to be reflective of the mechanism of the balance
control (Horak 1997; Cavanaugh et al. 2005). Recent studies have examined
pedobarography systems as an alternative method to measure balance, Nomura
®
et al. (2009) compared the simultaneous postural sway measures from a Tekscan
pressure-mapping system to a force plate (e.g., gold standard). It was concluded that
the pressure mat may attenuate the sway amplitude slightly but the overall correla-
tion between systems was very high (concordance correlation coefficient > 0.93).
Brenton-Rule et al. (2012) also test the reliability of the postural measures in older
®
adults with rheumatoid arthritis using a Tekscan system and found good to excellent
reliability (ICC above 0.84) in all sway measures in AP and ML direction. Recent
work done by our group used the Zeno™ walkway to assess standing balance in
individuals with multiple sclerosis and healthy controls, and compared the COP
Sway RangeAP Mean Sway Velocity
100
r = 0.8433 60 r = 0.9498
Mean Sway Velocity (mm/s)Zeno

Sway RangeAP(mm)Zeno
40
50
20
0 0
0 50 100 0 20 40 60
Sway RangeAP(mm)FP Mean Sway Velocity (mm/s)FP
Control_EO MS_EO Control_EC MS_EC
Fig. 3 Scatter plot of the sway range in AP direction and the mean sway velocity measured by
force plate and the pressure-sensing walkway (EO eyes open, EC eyes closed, Control healthy
control participants, MS patients with multiple sclerosis)
sway measures from the pressure-sensitive walkway with a force plate. In this study,
subjects were required to complete 30 s standing balance trials (eyes open and closed)
on a Zeno™ walkway and a Bertec force plate (Bertec Corporation, Columbus, OH).
The investigation revealed that the force platform and pressure-sensitive walkway
have high agreement (Pearson’s correlation coefficient > 0.80) in multiple sway
measurements (Fig. 3).
Latest development of the pressure sensor-embedded instrumented treadmill
®
(FDM-T by Zebris Medical GmbH) overcomes the length restrictions of other
instrumented walkways and offers the possibility to investigate long-distance loco-
motion. Gait characteristics in long distance walking can provide more precise
measures of stride-to-stride variability, as there has been suggestion about using at
least 400 steps to provide reliable estimation of gait variability (Owings and
Grabiner 2003). Moreover, long distance locomotion may also reflect the impact
of fatigue on gait performance.
Limitations
Although pressure-sensing walkway systems have been widely used for gait assess-
ment in research and clinical setting, there is at least one major limitation that should
be taken into consideration. Its use is restricted to a given physical location so it
cannot monitor patient’s gait during daily activities outside the lab or clinic nor
provide gait assessments on a continuous basis. Longitudinal measurement of gait,
rather than a single collection of lab-based test, may provide unique and important
information about the progress of disease and symptom fluctuation pattern in
14 R. Sun et al.
response to certain environmental factors (Maetzler et al. 2013; González et al. 2016)
and could be used as specific predictive markers of frailty syndrome (Fontecha et al.
2013), the onset of cognitive decline (Camicioli et al. 1998) and neurodegenerative
diseases (Salarian et al. 2004), and potential risk of falling (Shany et al. 2012).
It is worth noting that in order to better assess gait performance and stability
control in locomotion, in addition to the spatial and temporal gait parameters, the
dynamic interaction between the COP and COM should also be investigated.
Although the estimated center of mass (COMe) measurement has been reported by
some pressure-sensing walkway products, its validity has not been established with
the three-dimensional motion-tracking system; thus, the COMe measurement should
be treated with caution in the use of clinical assessments. The potential on integrating
the pressure-sensing walkway with virtual environments should also be explored to
assess gait in a simulated free-living locomotion tasks.
Summary
Gait assessment conducted through a pressure-sensing walkway system have been

widely utilized since the early 2000s. The relative low cost, high accuracy and
consistency, and automated data analysis features make it an ideal platform for
quantifying the spatial and temporal gait characteristics in various populations.
Although steady state linear walking is the most commonly used paradigm for gait
assessment, recent technological developments allow the pressure-sensing walkway
system to evaluate gait performance in other daily locomotion tasks (e.g., gait
initiation/termination, turning, sit to stand, obstacle crossing, stair ascending/
descending, etc.) as well as the postural balance performance.
Cross References

Analysis
Applications
▶ Slip and Fall Risk Assessment
▶ Gait During Real World Challenge: Gait Initiation, Gait Termination, Accelera-
tion, Deceleration, Turning, Slopes and Stairs
References
Baker R (2006) Gait analysis methods in rehabilitation. J Neuroeng Rehabil 3:1
Bilney B, Morris M, Webster K (2003) Concurrent related validity of the GAITRite ® walkway
system for quantification of the spatial and temporal parameters of gait. Gait Posture 17:68–74
Brenton-Rule A, Mattock J, Carroll M, Dalbeth N, Bassett S, Menz HB, Rome K (2012) Reliability
of the TekScan MatScan ® system for the measurement of postural stability in older people with
rheumatoid arthritis. J Foot Ankle Res 5:1
Bridenbaugh SA, Kressig RW (2010) Laboratory review: the role of gait analysis in seniors’
mobility and fall prevention. Gerontology 57:256–264
Camicioli R, Howieson D, Oken B, Sexton G, Kaye J (1998) Motor slowing precedes cognitive
impairment in the oldest old. Neurology 50:1496–1498
Cavanaugh JT, Guskiewicz KM, Stergiou N (2005) A nonlinear dynamic approach for evaluating
postural control. Sports Med 35:935–950
Chesnin KJ, Selby-Silverstein L, Besser MP (2000) Comparison of an in-shoe pressure measure-
ment device to a force plate: concurrent validity of center of pressure measurements. Gait
Posture 12:128–133
Chien S-L, Lin S-Z, Liang C-C et al (2006) The efficacy of quantitative gait analysis by the
GAITRite system in evaluation of parkinsonian bradykinesia. Parkinsonism Relat Disord
12:438–442
Chisholm AE, Perry SD, McIlroy WE (2011) Inter-limb Centre of pressure symmetry during gait
among stroke survivors. Gait Posture 33:238–243
Cho KH, Lee HJ, Lee WH (2015) Test–retest reliability of the GAITRite walkway system for the
spatio-temporal gait parameters while dual-tasking in post-stroke patients. Disabil Rehabil
37:512–516
Clark RA, Bower KJ, Mentiplay BF, Paterson K, Pua Y-H (2013) Concurrent validity of the
Microsoft Kinect for assessment of spatiotemporal gait variables. J Biomech 46:2722–2725
Cutlip RG, Mancinelli C, Huber F, DiPasquale J (2000) Evaluation of an instrumented walkway for
measurement of the kinematic parameters of gait. Gait Posture 12:134–138
Dusing SC, Thorpe DE (2007) A normative sample of temporal and spatial gait parameters in
children using the GAITRite ® electronic walkway. Gait Posture 25:135–139
Eastlack ME, Arvidson J, Snyder-Mackler L, Danoff JV, McGarvey CL (1991) Interrater reliability
of videotaped observational gait-analysis assessments. Phys Ther 71:465–472
Fontecha J, Navarro FJ, Hervás R, Bravo J (2013) Elderly frailty detection by using accelerometer-
enabled smartphones and clinical information records. Pers Ubiquit Comput 17:1073–1083
Fritz S, Lusardi M (2009) White paper: “walking speed: the sixth vital sign”. J Geriatr Phys Ther
32:2–5
Garciaguirre JS, Adolph KE, Shrout PE (2007) Baby carriage: infants walking with loads. Child
Dev 78:664–680
González I, López-Nava IH, Fontecha J, Muñoz-Meléndez A, Pérez-SanPablo AI, Quiñones-
Urióstegui I (2016) Comparison between passive vision-based system and a wearable inertial-
based system for estimating temporal gait parameters related to the GAITRite electronic
walkway. J Biomed Inform 62:210–223
Hartmann A, Luzi S, Murer K, de Bie RA, de Bruin ED (2009) Concurrent validity of a trunk
tri-axial accelerometer system for gait analysis in older adults. Gait Posture 29:444–448
Highsmith MJ, Schulz BW, Hart-Hughes S, Latlief GA, Phillips SL (2010) Differences in the spatio-
temporal parameters of transtibial and transfemoral amputee gait. J Prosthetics Orthot 22:26–30
Hollands K, Agnihotri D, Tyson S (2014) Effects of dual task on turning ability in stroke survivors
and older adults. Gait Posture 40:564–569
Horak FB (1997) Clinical assessment of balance disorders. Gait Posture 6:76–84
Katz-Leurer M, Rotem H, Lewitus H, Keren O, Meyer S (2008) Relationship between balance
abilities and gait characteristics in children with post-traumatic brain injury. Brain Inj
22:153–159
16 R. Sun et al.
Keefe FJ, Hill RW (1985) An objective approach to quantifying pain behavior and gait patterns in
low back pain patients. Pain 21:153–161
Kim A, Kim J, Rietdyk S, Ziaie B (2015) A wearable smartphone-enabled camera-based system for
gait assessment. Gait Posture 42:138–144
Kressig RW, Beauchet O (2006) Guidelines for clinical applications of spatio-temporal gait analysis
in older adults. Aging Clin Exp Res 18:174–176
Kuys SS, Brauer SG, Ada L (2011) Test-retest reliability of the GAITRite system in people with
stroke undergoing rehabilitation. Disabil Rehabil 33:1848–1853
Lewek MD, Randall EP (2011) Reliability of spatiotemporal asymmetry during overground
walking for individuals following chronic stroke. J Neurol Phys Ther 35:116–121
Liu W-Y, Lin P-H, Lien H-Y, Wang H-S, Wong AM-K, Tang SF-T (2014) Spatio-temporal gait
characteristics in children with Tourette syndrome: a preliminary study. Res Dev Disabil
35:2008–2014
Maetzler W, Domingos J, Srulijes K, Ferreira JJ, Bloem BR (2013) Quantitative wearable sensors
for objective assessment of Parkinson’s disease. Mov Disord 28:1628–1637
Maki BE, Holliday PJ, Topper AK (1994) A prospective study of postural balance and risk of falling
in an ambulatory and independent elderly population. J Gerontol 49:M72–M84
McDonough AL, Batavia M, Chen FC, Kwon S, Ziai J (2001) The validity and reliability of the
GAITRite system’s measurements: a preliminary evaluation. Arch Phys Med Rehabil
82:419–425
Menz HB, Latt MD, Tiedemann A, San Kwan MM, Lord SR (2004) Reliability of the GAITRite ®
walkway system for the quantification of temporo-spatial parameters of gait in young and older
people. Gait Posture 20:20–25
Miyazaki S, Kubota T (1984) Quantification of gait abnormalities on the basis of continuous foot-
force measurement: correlation between quantitative indices and visual rating. Med Biol Eng
Comput 22:70–76
Morris ME, Huxham F, McGinley J, Dodd K, Iansek R (2001) The biomechanics and motor control
of gait in Parkinson disease. Clin Biomech 16:459–470
Nelson AJ, Zwick D, Brody S et al (2002) The validity of the GaitRite and the functional
ambulation performance scoring system in the analysis of Parkinson gait. NeuroRehabilitation
17:255–262
Nomura K, Fukada K, Azuma T, Hamasaki T, Sakoda S, Nomura T (2009) A quantitative
characterization of postural sway during human quiet standing using a thin pressure distribution
measurement system. Gait Posture 29:654–657
Owings TM, Grabiner MD (2003) Measuring step kinematic variability on an instrumented
treadmill: how many steps are enough? J Biomech 36:1215–1218
Papadopoulos N, McGinley JL, Bradshaw JL, Rinehart NJ (2014) An investigation of gait in
children with attention deficit hyperactivity disorder: a case controlled study. Psychiatry Res
218:319–323
Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE (2010) Evaluation of gait symmetry
after stroke: a comparison of current methods and recommendations for standardization. Gait
Posture 31:241–246
Paul L, Ellis B, Leese G, McFadyen A, McMurray B (2009) The effect of a cognitive or motor task
on gait parameters of diabetic patients, with and without neuropathy. Diabet Med 26:234–239
Podsiadlo D, Richardson S (1991) The timed “up & go”: a test of basic functional mobility for frail
elderly persons. J Am Geriatr Soc 39:142–148
Rao AK, Quinn L, Marder KS (2005) Reliability of spatiotemporal gait outcome measures in
Huntington’s disease. Mov Disord 20:1033–1037
Rinehart NJ, Tonge BJ, Iansek R, McGinley J, Brereton AV, Enticott PG, Bradshaw JL (2006) Gait
function in newly diagnosed children with autism: cerebellar and basal ganglia related motor
disorder. Dev Med Child Neurol 48:819–824
Salarian A, Russmann H, Vingerhoets FJ, Dehollain C, Blanc Y, Burkhard PR, Aminian K (2004)
Gait assessment in Parkinson’s disease: toward an ambulatory system for long-term monitoring.
IEEE Trans Biomed Eng 51:1434–1443
Schniepp R, Wuehr M, Neuhaeusser M et al (2012) Locomotion speed determines gait variability in
cerebellar ataxia and vestibular failure. Mov Disord 27:125–131
Sekiya N, Nagasaki H, Ito H, Furuna T (1997) Optimal walking in terms of variability in step
length. J Orthop Sports Phys Ther 26:266–272
Selby-Silverstein L, Besser M (1999) Accuracy of the GAITRite ® system for measuring temporal-
spatial parameters of gait. Phys Ther 79:S59
Shany T, Redmond S, Marschollek M, Lovell N (2012) Assessing fall risk using wearable sensors: a
practical discussion. Z Gerontol Geriatr 45:694–706
Shull PB, Jirattigalachote W, Hunt MA, Cutkosky MR, Delp SL (2014) Quantified self and human
movement: a review on the clinical impact of wearable sensing and feedback for gait analysis
and intervention. Gait Posture 40:11–19
Sorsdahl AB, Moe-Nilssen R, Strand LI (2008) Test–retest reliability of spatial and temporal gait
parameters in children with cerebral palsy as measured by an electronic walkway. Gait Posture
27:43–50
Sosnoff JJ, Sandroff BM, Motl RW (2012) Quantifying gait abnormalities in persons with multiple
sclerosis with minimal disability. Gait Posture 36:154–156
Sosnoff JJ, Klaren RE, Pilutti LA, Dlugonski D, Motl RW (2015) Reliability of gait in multiple
sclerosis over 6 months. Gait Posture 41:860–862
Stover AM (2005) Reliability of the GAITRite (R) Walking System for the assessment of gait in
individuals with Parkinson’s disease. Master’s and Doctoral Project, The University of Toledo
Digital Repository
Thorpe DE, Dusing SC, Moore CG (2005) Repeatability of temporospatial gait measures in
children using the GAITRite electronic walkway. Arch Phys Med Rehabil 86:2342–2346
Van Uden CJ, Besser MP (2004) Test-retest reliability of temporal and spatial gait characteristics
measured with an instrumented walkway system (GAITRite ®). BMC Musculoskelet Disord
5:13
cognitive decline and dementia. J Neurol Neurosurg Psychiatry 78:929–935
Verghese J, Holtzer R, Lipton RB, Wang C (2009) Quantitative gait markers and incident fall risk in
older adults. J Gerontol Ser A Biol Med Sci 64:896–901
Wajda DA, Moon Y, Motl RW, Sosnoff JJ (2015) Preliminary investigation of gait initiation and
falls in multiple sclerosis. Arch Phys Med Rehabil 96(6):1098–1102
Webster KE, Wittwer JE, Feller JA (2005) Validity of the GAITRite ® walkway system for the
measurement of averaged and individual step parameters of gait. Gait Posture 22:317–321
Webster KE, Merory JR, Wittwer JE (2006) Gait variability in community dwelling adults with
Alzheimer disease. Alzheimer Dis Assoc Disord 20:37–40
Whittle MW (1996) Clinical gait analysis: a review. Hum Mov Sci 15:369–387
Wittwer JE, Webster KE, Andrews PT, Menz HB (2008) Test–retest reliability of spatial and
temporal gait parameters of people with Alzheimer’s disease. Gait Posture 28:392–396
Wondra VC, Pitetti KH, Beets MW (2007) Gait parameters in children with motor disabilities using
an electronic walkway system: assessment of reliability. Pediatr Phys Ther 19:326–331
Wong JS, Jasani H, Poon V, Inness EL, McIlroy WE, Mansfield A (2014) Inter-and intra-rater
reliability of the GAITRite system among individuals with sub-acute stroke. Gait Posture
40:259–261
Wu J, Looper J, Ulrich BD, Ulrich DA, Angulo-Barroso RM (2007) Exploring effects of different
treadmill interventions on walking onset and gait patterns in infants with down syndrome. Dev
The Importance of Foot Pressure
in Diabetes
Malindu E. Fernando, Robert G. Crowther, and Scott Wearing
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Foot Pressure Sensors, Types, Spatial Resolutions, and Device Specifications . . . . . . . . . . . . . 6
Types of Foot Pressure Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
The Procedures Used to Acquire Foot Pressure Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Reliability and Validity of Foot Pressures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Factors that Influence Foot Pressure Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Foot Pressure in Individuals with Diabetes Mellitus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Foot Pressure in Individuals with High-Risk Feet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Ultimately, Could a Threshold Be Found for Identifying Individuals at Risk of
Ulceration? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
What Research Is Required in order to Make Foot Pressures More Effective and
Beneficial? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
M.E. Fernando (*)

Podiatry Service, Kirwan Community Health Campus, Townsville, QLD, Australia
College of Medicine, James Cook University, Townsville, QLD, Australia
e-mail: malindu.fernando@my.jcu.edu.au
R.G. Crowther
Sport and Exercise, School of Health and Wellbeing, University of Southern Queensland, Ipswich,
QLD, Australia
Smart Movement, Brisbane, QLD, Australia
e-mail: Robert.Crowther@usq.edu.au
S. Wearing
Institute of Health and Biomedical Innovation, Queensland University of Technology, Brisbane,
QLD, Australia
Faculty for Sport and Health, Technische Universität München, Munich, Bavaria, Germany
e-mail: s.wearing@qut.edu.au

DOI 10.1007/978-3-319-30808-1_39-1
2 M.E. Fernando et al.
Abstract
Foot pressure assessment refers to the study of pressure fields acting between the
plantar surface of the foot and a supporting surface using typically electronic
sensors. Individuals with both type 1 and type 2 diabetes are at risk of developing
diabetic peripheral neuropathy (loss of peripheral sensory and motor function),
which predisposes them to the development of plantar foot ulcers (open wounds).
Foot pressure assessments have been extensively utilized to investigate the out-
comes of biomechanical features in individuals with diabetes related foot com-
plications termed the “high-risk” foot. Thus, the application of foot pressure
measurements in individuals with diabetes includes monitoring patients for risk
of ulceration, determining pressure off-loading capacities, and investigating the
mechanical factors responsible for foot ulceration and ulcer healing.
The ideal application of foot pressure would be to utilize measurements to
predict sites of potential ulceration, prior to ulcer occurrence, and to effectively
guide pressure off-loading of ulcerated sites to progress wound healing. Although
these two applications represent the overall importance of foot pressure assess-
ments within the field of diabetes, such applications have limited use due to
various reasons. The aim of this chapter is to provide the reader with an overview
of foot pressure assessment in relation to diabetes mellitus and describe the
factors which influence foot pressure assessments. In doing this, we hope to
provide a focused discussion of the relevance of foot pressures in diabetes
mellitus, utilizing the most up-to-date literature on the topic.
Keywords
Plantar pressure • Foot ulcers • Pressure–time integral • Peak plantar pressure •
Shear pressure • Diabetic peripheral neuropathy • Diabetes mellitus • Foot
pressure • Pedobarography • Reproducibility • Sensors • Validity • Footwear •
Orthoses
Introduction
Pedobarography (also known as plantar or foot pressure assessment) is the study of

pressure fields acting between the plantar surface of the foot and a supporting surface
using a wide range of techniques that encompass electronic and nonelectronic
methods (see Fig. 1 for an example). Foot pressures, simply stated, are the result
of the vertical force (also known as ground reaction force) acting on a specific site of
the plantar aspect of the foot divided by the contact area (the amount of surface
contact between the plantar surface of the foot and the sensor). The assessment of
foot pressure is employed in a wide range of applications including sports biome-
chanics and clinical assessments but has specific implications in diabetes mellitus.
The clinical application of foot pressures has both direct and indirect uses (Hughes
1993). The direct uses can be to assess the effectiveness of a treatment by examina-
tion before and after a surgical procedure, to monitor treatment progress with
The Importance of Foot Pressure in Diabetes 3
Fig. 1 Foot pressure assessment of an individual demonstrating the various areas of foot pressure.
The diagram to top indicates the different masks (or sites of foot pressure in various colors); the
diagram to the bottom indicates an example of a pressure measurement in an individual. Note the
area of red and yellow under the fifth metatarsal of the right foot (bottom diagram), indicating a site
of peak pressure
repeated measurements, and to design and assess the effectiveness of foot orthoses in
relation to pressure off-loading (Kato et al. 1996; Bus et al. 2004, 2008b, 2013;
Armstrong et al. 1999; Mueller et al. 2003b; Fernando et al. 2015). Indirect uses
come from the growing body of knowledge emerging from laboratories using this
equipment for research which provide scientists with fundamental information
regarding the control of gait and the influence of foot pressures (Hughes 1993;
Fernando et al. 2016a; Hamatani et al. 2016; Barn et al. 2015; Qiu et al. 2015; Yavuz
2014; Hafer et al. 2013; Bus and Waaijman 2013; Waaijman and Bus 2012; Acharya
et al. 2012). Individuals with both type 1 and type 2 diabetes mellitus are at risk of
developing diabetic peripheral neuropathy (nerve damage due to uncontrolled or
long-standing chronic hyperglycemia) and peripheral arterial disease. Both compo-
nents predispose individuals to foot ulcers, but of particular relevance to foot
pressures is the development of diabetic peripheral neuropathy. Diabetic peripheral
neuropathy has sensory, motor, and autonomic components (Boulton et al. 2004).
The sensory components have received most of the attention in the literature, and a
“loss of protective sensation” has been implicated in the development of plantar foot
ulcer (wound) formation (Armstrong 2005; Wood et al. 2005). While the motor
component of neuropathy has received less attention in relation to foot ulcers, this
has also been implicated in the development of foot ulcers by its effect on foot and
lower limb biomechanics (Mueller et al. 1994a; Veves et al. 1992; Fernando et al.
2014, 2016b).
Individuals with diabetes mellitus acquire foot deformities (such as claw and
hammertoe deformities) and experience significant changes to their foot morphol-
ogy. These changes are accompanied by a gradual decline in peripheral propriocep-
tion, motor function, and protective sensation of the foot, after the onset of diabetic
peripheral neuropathy. The collective term for these changes is the “high-risk” foot,
and such changes are associated with elevated pressure beneath the foot (see Fig. 1).
Such changes contribute to higher magnitudes and durations of mechanical stress

within the foot during standing and walking. These higher foot pressures are thought
to predispose patients with diabetic peripheral neuropathy to the development of foot
ulcers (wounds) on the plantar aspect of the foot. Elevated foot pressures have also
been implicated in delayed wound healing, which in turn predisposes individuals
with foot ulcers to limb amputations due to wound infection and ischemia.
Therefore, foot pressure assessment in patients with diabetes mellitus has multi-
ple applications, and some of the main applications include the identification of sites
of potential ulceration, evaluation of the duration and magnitude of pressure
off-loading required for ulcer healing, and the assessment of the effectiveness of
off-loading techniques. Other applications include the assessment of gait
temporal–spatial parameters and balance in the presence of peripheral neuropathy.
Unfortunately, despite an increase in the clinical use of pedobarography, the results
obtained from the foot pressure assessments do not always reflect clinical expecta-
tions (i.e., sites of ulceration), leading to confusion and misdiagnosis (Choi et al.
2014). Hence the ability of pedobarography to assist in improving the identification
of sites of potential foot ulceration remains somewhat controversial (Choi et al.
2014). Therefore, clinicians and researchers need to be aware of both the strengths
and the shortcomings of foot pressure assessments in order to utilize it effectively.
Hence, the aims of this chapter are to provide the reader with an overview of foot
pressure assessment in relation to diabetes mellitus; to describe factors that influence
foot pressure assessments, including some of the key technical requirements; and to
provide a focused discussion of its importance in diabetes mellitus, utilizing the most
up-to-date literature on the topic.
State of the Art
The acquisition of foot pressure data in individuals with diabetes mellitus has
evolved over the last three decades from only barefoot assessments to both barefoot
and in-shoe foot pressure assessments (see Fig. 2) (Lord et al. 1986). This shift of
methodology was predicated on the assumption that in-shoe pressure measurements
are more representative of the foot pressures experienced by patients with diabetes
on a day-to-day basis. While this is true from an application point of view, barefoot
assessments and in-shoe assessments provide different types of biomechanical
information. For example, barefoot investigations allow the assessment of the
foot–ground interaction and the inherent foot pressures experienced due to the
presence of neuropathy or deformity, without the confounding influence of footwear.
In-shoe assessments, in contrast, allow for the identification of the influence of
footwear on foot pressures, to assess areas of high-pressure and the adequacy of
off-loading strategies. Therefore, both barefoot and in-shoe assessments continue to
have a role within the field.
Another key area of interest has been in assessing shear pressure, or the pressure
related to the horizontal components of the ground reaction force. The shear com-
ponents of foot pressure are thought to have an important role in the development of
Fig. 2 Barefoot and in-shoe foot pressure acquisition. The image to the left represents an example
of in-shoe foot pressure assessment. Here the pressure-detecting insoles lined with electronic
sensors, capable of detecting foot pressures, can be seen inside the footwear. The data obtained is
transmitted to a receiver (seen in the far left) (Image is courtesy of novel electronics incorporated,
obtained with permission.) The image to the right represents an example of how barefoot foot
pressure data is obtained using a pressure platform. The given example demonstrates a freestanding
platform, 2 m long, which is usually embedded within the floor. This platform contains electronic
sensors capable of detecting the foot pressure at various foot sites and estimating the vertical ground
reaction force utilizing the pressure value and contact areas at various foot sites, provided the
sensors are loaded uniformly
complications such as diabetic foot ulcers, in the presence of diabetic peripheral

neuropathy (Yavuz 2014; Yavuz et al. 2007a, 2008, 2009; Park and Kim 2007).
Although it was known over three decades ago that shear pressures had a profound
influence on tissue viability in individuals with diabetes mellitus (Pollard and Le
Quesne 1983), the quantification of the horizontal components of pressure has
proved to be technically challenging (Yavuz et al. 2007a). With the exception of a
few devices, most standard sensors used for pressure measurement do not measure
the fore–aft or medial–lateral shear forces that are obtained using force platforms
(Alexander et al. 1990; Orlin and McPoil 2000). Measurement of shear is technically
difficult compared to vertical pressure, as shear is also dependent on frictional
properties of the sensor surface (which may or may not be representative of the
surrounding footwear or the surface of contact). Consequently, some attempts have
been made to estimate shear forces by modeling the spatial change or gradient of
vertical peak pressures beneath the foot, termed the peak pressure gradient (Mueller
et al. 2005). Findings indicated that the peak pressure gradient was substantially
higher in the forefoot than in the rearfoot even when compared with the peak foot
pressure (Mueller et al. 2005). While insightful, the accuracy of these methods has
recently been questioned and has been suggested to be of limited clinical capacity
(Yavuz et al. 2007a).
Some technological advances have been able to provide the instrumentation
required to evaluate the role of shear pressure in leading to ulceration and hence
have assisted in advancing this area of knowledge (Yavuz 2014; Yavuz et al. 2009;
Hamatani et al. 2016; Lord and Hosein 2000; Davis et al. 1998). Davis and
colleagues (1998) were one of the first groups to develop a device to simultaneously
measure the vertical pressure and the anterior–posterior and medial–lateral distrib-
uted shearing forces under the plantar surface of the foot using strain gauge tech-
nology. However, this device had a low sampling frequency of 37 Hz and was
limited to a few plantar locations during simultaneous measurement (Davis et al.
1998). There is no commercially available measurement device for assessing shear
pressure at present. The inability to measure the shear pressure may have contributed
to the lack of clarity regarding pressure thresholds for foot ulceration and may also
provide a reason as to why ulcer sites do not always correspond with sites of peak
vertical foot pressure (Yavuz et al. 2007b).
Yavuz and colleagues (2007b) demonstrated that sites of peak pressure do not
always correlate with sites of peak shear pressure within the foot. In their study, peak
shear occurred at the same site as the peak pressure for 20 % of the cohort with
diabetes mellitus. For the other participants, the location of peak shear was more than
2.5 cm away from the site of peak pressure (Yavuz et al. 2007b). Hence, our current
understanding of shear and vertical pressure and how they are involved in foot
ulceration and ulcer healing continues to improve. These advancements may have a
large impact in terms of how foot ulcers are managed in the future.
Foot Pressure Sensors, Types, Spatial Resolutions, and Device

Specifications
Types of Sensors
There are a range of different foot pressure measurement devices. The foot pressure
systems used by clinicians and researchers vary in sensor configuration to meet
different application requirements. The types of pressure sensors that are utilized
include capacitive sensors, resistive sensors, piezoelectric sensors, and piezoresistive
sensors (Abdul Razak et al. 2012; Rosenbaum and Becker 1997; Urry 1999). Table 1
provides a summary of the various types of sensors. Capacitive sensors contain two
electrically charged plates separated by an elastic medium and measure pressure as a
voltage change proportional to the applied pressure. Resistive sensors work by
measuring the resistance of conductive medium sandwiched between two electrodes.
Application of pressure causes conductive particles to touch, increasing the current
through the sensors. Piezoelectric sensors produce an electric field (voltage) in direct
response to an applied force which can be measured. Piezoresistive sensors rely on
the piezoresistive effect, where the electrical resistance of a material changes in
Table 1 Types of foot pressure sensors and their characteristics.

Pressure detection
Type of sensor strategy Examples Advantage Disadvantage
Capacitive Measures voltage Emed# Uses a Sensor is larger
sensors change system calibration than most other
Pedar# curve sensor types
system developed for
each sensor
Resistive Measures resistance of Matscan# Smaller in size Can only
sensors conductive foam system Able to measure
between two electrodes Tekscan# measure both vertical force
system force and Conductive
Musgrave# pressure foam is prone
system to damage
Piezoelectric Produces an electric F-scan# Highly elastic, Highly
sensors voltage in response to show little temperature
pressure material sensitive
deformation
Piezoresistive Semiconductor ParoTec# May be able to High signal to
sensors material. Measures measure shear noise ratio due
electrical voltage forces to electrical
proportional to applied interference
force
The data in the table is adapted from the following review or original studies (Zhao et al. 2013; Orlin
and McPoil 2000; Abdul Razak et al. 2012).
response to an applied mechanical strain. Typically made of a semiconductor

material, the resistivity of the sensor is influenced by the force or pressure applied;
when the sensor is unloaded, resistivity is high, and when force is applied, resistance
decreases. When piezoresistors are placed in a Wheatstone bridge configuration and
attached to a pressure-sensitive diaphragm, a change in resistance is converted to a
voltage output, which is proportional to the applied pressure. The type of sensors
within the device are also a determinant of measurement accuracy and precision
(Giacomozzi 2010). For example, static and dynamic pressure tests have demon-
strated very high accuracy of capacitive, elastomer-based technologies, and a high
accuracy of resistive technology, despite a need for a complex ad hoc calibration
(Giacomozzi 2010).
Types of Devices
Pressure measurement devices are also broadly categorized into either pressure
distribution platforms or in-shoe systems. Key benefits of using in-shoe measure-
ment devices are the mobility, flexibility, and cost, as opposed to platform systems,
which are less portable and generally more expensive. Platform systems were
traditionally used for both static (standing) and dynamic (during walking) assess-
ments, although several new in-shoe systems are also cable of assessing these
parameters. Platform systems are typically constructed from a flat array of pressure
sensing elements and various types of sensors arranged in a matrix configuration and
embedded in the floor to allow the capture of foot pressure information during gait
(Abdul Razak et al. 2012). In-shoe sensors are flexible and embedded in the shoe
such that measurements reflect the interface between the foot and the shoe (Abdul
Razak et al. 2012), rather than the foot and the floor as in platform systems.
Other advantages of using platform devices compared to in-shoe devices include
a greater number of active sensors, the size of the active sensing area, a higher spatial
resolution (i.e., the number of sensors for a given absolute area), and the ideal
placement of the sensors parallel to the supporting surface to provide a “true” vertical
force measurement (Orlin and McPoil 2000). Disadvantages include the need to
target the pressure platform during gait, with smaller platforms which leads to altered
pressure measurements and the large number of required steps to obtain reliable and
reproducible data (Wearing et al. 1999; Abdul Razak et al. 2012; Orlin and McPoil
2000). A disadvantage of pressure measurement systems compared to force plates is
their relatively lower temporal resolution. Modern systems typically have sample
rates between 100 and 500 Hz (Abdul Razak et al. 2012). Although suitable for most
clinical gait applications, the sampling frequency of commercially available pressure
platforms may not be sufficient to accurately detect transient loads associated with
heel strike (Verdini et al. 2006; Gillespie and Dickey 2003).
A general advantage of in-shoe measurement systems is that multiple steps can be
easily averaged and analyzed with the same mask, and the influence of footwear and
orthotics can also be studied (Rosenbaum and Becker 1997). However, in-shoe
sensors are susceptible to damage and generally have lower spatial resolution, yet
they provide real-time information regarding foot pressures while wearing footwear.
In-shoe systems may also be limited by the fact that data using different insole sizes
may be incomparable given differences in spatial resolution between insoles of
different sizes. Consistent with platform systems, the ability to detect the transient
loads associated with heel strikes, which may be as high as 400 Hz in some footwear
conditions, is seemingly limited with in-shoe systems (Gillespie and Dickey 2003).
The decision of what type of device is based on the clinical or research requirement,
the loading characteristics, and the outcomes of interest.
Specifications for Devices

The required specifications for a pressure sensor in terms of sensor performance
include the linearity, hysteresis, temperature sensitivity, sensor size, type, and the
capture frequency (Abdul Razak et al. 2012; Rosenbaum and Becker 1997; Urry
1999). The linearity refers to whether sensor output is directly (linearly) related to
sensor input over a specified range over time, and hence a higher linearity is
preferred in a sensor. Hysteresis refers to a measure of the sensors performance on
loading and unloading. Hence, the ideal sensor should have a low hysteresis or loss
of energy. The temperature sensitivity refers to the sensitivity of the sensor to
ambient temperature and the sensor type and size as well as the type of the elastomer
used, and this determines the range of pressure able to be captured by the device. The
spatial resolution refers to the size rather than the number of sensors within the
capturing system (either a platform or an insole). The capture frequency is the rate of
capture or sampling frequency, which is dependent on a number of factors including
the natural frequency of the sensor. Hence, the sampling frequency is the number of
samples measured by each sensor per second recorded in cycles per second or Hertz.
Calibration refers to the priming of the sensors prior to capturing foot pressure data
(Abdul Razak et al. 2012; Orlin and McPoil 2000). Most importantly, the specifica-
tions that are crucial for obtaining reliable and valid foot pressure measurements are
the spatial resolution, temporal resolution, and accuracy of calibration (Orlin and
McPoil 2000; Rosenbaum and Becker 1997).
Types of Foot Pressure Outcomes
The two commonly used measures of foot pressure are peak pressure (mpp), by
definition the peak pressure measurement during the stance phase of the gait cycle,
and the pressure–time integral (pti), by definition the area under a peak pressure
versus time curve (see Fig. 3). The most common units for mean peak pressure
include newtons per centimeter squared (N/cm2) and kilopascals (kPa), and for the
pressure–time integral, Ns/cm2 and kPa/s. The SI unit for pressure is kPa and is
encouraged among the foot pressure measurement community. With regard to the
types of measurement outcomes used, there remains some ambiguity in the field as to
which is the better outcome to predict ulceration and ulcer healing in individuals
with diabetes and diabetic peripheral neuropathy (Melai et al. 2011; Bus and
Waaijman 2013). Therefore, there are differences between what measures are com-
monly used in studies (see Table 2 for a random selection of studies that reported
mpp and/or pti). From observing the outcomes used in these studies, it is evident that
Pressure-time integral (N/cm2 . s)

25
20
Pressure (N/cm2)
15
10
a
5
0
Stance phase of gait cycle (s)
Fig. 3 An example of a pressure versus time graph of the whole foot. The x-axis represents the
time or stance phase duration of the gait cycle in seconds (s), and the y-axis represents foot pressure
in N/cm2. The mean peak pressure (the peak of the pressure vs. time graph) is highlighted by the red
arrow and represents the magnitude of pressure. The pressure–time integral, the area under the
curve, is represented by the shaded area under the graph (a) and represents a measure of the
magnitude and duration of pressure. Hence, the two measurements provide different types of
information related to the duration of pressure and magnitude of pressure
Table 2 Examples of studies reporting peak pressure and/or pressure–time integral.

Study name Population studied Reported on mpp Reported on pti
(Melai et al. 2011) DPN Yes Yes
(Guldemond et al. 2008) DPN Yes No
(Bacarin et al. 2009) DPN Yes Yes
(Sacco et al. 2009) DPN Yes Yes
(Caselli et al. 2002) DPN Yes No
(Waaijman and Bus 2012) DPN Yes Yes
(Yuk San Tsung et al. 2004) DPN Yes Yes
(Owings et al. 2009) DMC Yes No
(Waldecker 2012) DMC Yes Yes
(Mueller et al. 2008) DPN Yes No
The table demonstrates a selection of studies in the area that were randomly selected to demonstrate
what outcomes were reported. mpp = peak plantar pressure, pti = pressure–time integral,
DPN = individuals with diabetic peripheral neuropathy, DM = individuals with diabetes mellitus
without neuropathy, DMC = individuals with a diabetes mellitus-related foot complication exclud-
ing neuropathy
the mpp has historically been the more commonly used outcome. As these outcomes
represent different types of information regarding the temporal and peak character-
istics of foot pressure, it may be appropriate to measure both outcomes. There is
evidence however that the level of reproducibility is different for the various foot
pressure outcome measures (Fernando et al. 2016a; Gurney et al. 2013; Price et al.
2016). In general, the mean peak pressures seem to be more reproducible compared
to the pressure–time integral in people with diabetes mellitus, although further
research is needed to support this finding (Fernando et al. 2016a).
Historically, an aggregated measure of foot pressure was reported for the entire
plantar surface of the foot as opposed to site-specific foot pressures. Aggregated
measures do not take into account the area of the foot that the pressure acts on, and,
especially in individuals with diabetes, the location of peak pressure gives important
information as to where the sites of ulceration may occur (Fernando et al. 2016a;
Ledoux et al. 2013). Hence reporting a single pressure value, which is not represen-
tative of the entire plantar surface of the foot, provides only limited information.
Rather site-specific measurements of pressure–time integral and mean peak pressure
provide more insight to the localization of foot pressure as opposed to aggregated
whole-foot pressure measurements. A benefit of foot pressure assessment is the
ability to measure the distribution of load under the foot, which is not possible, for
example, with force plate analyses.
The Procedures Used to Acquire Foot Pressure Data
Although there have been efforts to standardize the protocols used for foot pressure
measurements (Barnett 1998; Cavanagh et al. 2000; Cavanagh and Ulbrecht 1994;
Fernando et al. 2015), there is no agreed standardized protocol to date, and various
Fig. 4 A photo depicting the

mid-gait strategy using a
pressure platform (Image is
courtesy of novel electronics
incorporated, obtained with
permission)
methods have been used to obtain data. This has led to concerns with the compara-
bility of data obtained using different protocols. The traditional method for
collecting data using a platform system has been termed the “mid-gait” technique
(see Fig. 4) (Meyers-Rice et al. 1994). This method requires individuals to ambulate
along a walkway, and pressure data are collected from a single foot contact on the
sensor platform during steady-state walking. While this method of assessment is
suitable for healthy participants, it is no longer considered appropriate for use in
individuals with high-risk foot conditions (i.e., ulceration) and particularly in indi-
viduals with diabetic peripheral neuropathy due to increased repetitive stress and the
potential for skin ulceration (Orlin and McPoil 2000). This led to the development of
alternate methods of foot pressure assessment, including the one-step, two-step, and
three-step methods, relating to the number of steps a person must take before
stepping onto a platform to capture their foot pressure (McPoil et al. 1999; Bus
and de Lange 2005). While the one-step method is often used clinically, foot
pressures obtained using this technique are not representative of those collected
using the mid-gait method. The two-step method of capture, in contrast, produces
pressure values that are closer to those of the mid-gait technique and is widely
advocated for pressure assessment in the high-risk foot (Bus and de Lange 2005).
Irrespective of the gait capture protocol, at least three to five assessments of gait are
required to reliably assess pressures in individuals with and without diabetes mellitus
(McPoil et al. 1999; Bryant et al. 1999; Bus and de Lange 2005).
Whether foot pressures are acquired during the initiation or termination of gait
also influences foot pressure measurements using the two-step protocol; therefore
whether foot pressure data should be acquired in relation to gait initiation or
termination is largely dependent on the gait parameter of interest (Wearing et al.
1999). For example, assessing foot pressures during gait initiation has minimal effect
on peak pressures beneath the forefoot but markedly alters timing parameters of the
gait cycle compared to the mid-gait method (Wearing et al. 1999). Although the
assessment of foot pressure during the commencement of gait may be ideal for
assessing healthy individuals, this protocol may be less than ideal in individuals with
diabetes mellitus due to poor balance and proprioception and an unstable gait (Bus
and de Lange 2005). This is consistent with other work which indicates that a steady
gait is obtained between the second and third step after gait initiation (Miller and
Verstraete 1996). Generally, an approach that reduces the total stance time could
reduce the amount of mechanical loading. The three-step protocol may offer con-
sistent results and avoid unnecessary loading of the foot, especially in individuals
with diabetic peripheral neuropathy (Fernando et al. 2015, 2016a).
Currently, there are no standardized protocols for the assessment of in-shoe foot
pressures, and a range of different methods utilizing different numbers of steps have
been utilized to capture foot pressure (see Fig. 5) data (Arts and Bus 2011; Price et al.
2014, 2016; Woodburn and Helliwell 1996). This creates a challenge in comparing
results between studies that have used different methods of assessment. The other
limitation of in-shoe pressure measurements is the influence of footwear and orthot-
ics, which can have a confounding effect on the distribution and magnitude of foot
pressure. Hence pressure data from studies using different footwear types are not
directly comparable (Guldemond et al. 2007; Bus 2008; Bus et al. 2011; Fernando
et al. 2013).
Reliability and Validity of Foot Pressures
The validity of foot pressures refers to the ability of either a platform or an in-shoe
device to quantify the magnitude and duration of pressure accurately. The reliability
is the agreement between different pressure measurements of the same individual
(which is dependent on both natural variability of pressure in an individual as well as
measurement reproducibility of the system). The validity of foot pressure measure-
ments has received limited attention in the literature (Price et al. 2016; McPoil et al.
1995; Luo et al. 1998; Giacomozzi 2010). The number and size of sensors, sampling
frequency, and accuracy of calibration are likely to influence the validity of foot
pressure outcomes. On the contrary, the reproducibility of various foot pressure
platforms and in-shoe devices have been assessed in healthy populations (Hafer et al.
2013; Hughes et al. 1991; Gurney et al. 2008; Zammit et al. 2010) and in individuals
with diabetes mellitus (Fernando et al. 2016a). These reproducibility assessments are
Fig. 5 A photo depicting

in-shoe pressure measurement
(Image is courtesy of novel
electronics incorporated,
obtained with permission)
representative of the reliability of the pressure system and variability of the individ-
uals assessed.
Factors such as the protocol used to capture foot pressure data and the partici-
pant’s gait on the day (especially with respect to speed of walking) will also
influence foot pressure assessments, and therefore it is often indicated that gait
speed should be controlled (Hughes et al. 1991). These influences are discussed in
the next section. It is however reasonable to say that perfect reliability (i.e., 100 %
agreement between two measurements) cannot be expected in foot pressure assess-
ments, because of inherent differences in each gait assessment that cannot be
standardized, as it involves a dynamic system within an individual (Orlin and
McPoil 2000). However, the variability of foot pressure within individuals may
also be an important consideration in studying the risk of ulceration and ulcer healing
as it provides important information regarding cyclic loading and the range of
pressures experienced.
Hafer and co-workers (2013) calculated the intra-mat, intra-manufacturer, and
inter-manufacturer reliability of foot pressure parameters as well as the number of
foot pressure trials needed to reach a stable estimate for healthy participants using
intraclass correlation statistics. Their results indicated that the intra-platform reli-
ability correlations were greater than 0.70 and that the inter-platform correlations of
reliability were more than 0.70 for more than half of the foot pressure outcomes at
various sites, when using five gait assessments per participant (Hafer et al. 2013).
Intrasubject or platform comparisons of greater than 0.75 are considered to be of
good reliability, those between 0.5 and 0.75 moderate reliability, and those under 0.5
poor reliability (Ferrarin et al. 2011). In addition, all foot pressure parameters
improved in consistency and were within 90 % of the mean when five gait assess-
ments were used (Hafer et al. 2013). This is consistent with other work indicating
that the use of three to five gait assessments increases the reproducibility of foot
pressures (Gurney et al. 2008; Hughes et al. 1991; Wearing et al. 1999). However,
the comparisons by Hafer and co-workers (2013) were limited to a healthy popula-
tion and to two types of platform systems, and the intraclass correlation statistics in
general are known to be poor predictors of overall reliability (Lee et al. 2012).
With respect to measurements in patients with diabetic peripheral neuropathy
using platform systems, findings have indicated that the coefficient of variation and
the reproducibility are also site and outcome dependent as in healthy individuals
(Gurney et al. 2013). This is also the case in individuals with diabetic peripheral
neuropathy and active foot ulcers. Figure 6 demonstrates differences in the coeffi-
cient of variability of foot pressure measurements across different plantar sites and
when using two outcomes of foot pressures in individuals with diabetic foot ulcers,
in individuals with diabetes without ulcers, and in healthy controls (Fernando et al.
2016a). The overall reproducibility of foot pressures are, however, site dependent
(Gurney et al. 2008; Fernando et al. 2016a) and outcome dependent (Fernando et al.
2016a; Gurney et al. 2008) in individuals with and without diabetes mellitus.
Therefore, the site of measurement within the foot (i.e., plantar hallux), the size of
the mask or pressure capture location, and type of pressure outcome (pressure–time
integral or mean peak pressure) also determine the overall reproducibility of foot
pressure, as in healthy controls (Wearing et al. 1999). The mean peak pressure on
average has better reproducibility in diabetes participants compared to the
pressure–time integral (Fernando et al. 2016a).
When using in-shoe pressure assessments, Ahroni and colleagues (Ahroni et al.
1998) found that the peak foot pressures over the metatarsal heads had the best
reproducibility with low coefficients of variability and interclass correlation coeffi-
cients of greater than 0.7. However whether this constitutes an appropriate level of
reproducibility is questionable (Lee et al. 2012). The reproducibility of foot pressure
at the heel, the whole foot, and the hallux also showed fair to good reproducibility
(Ahroni et al. 1998). More recent findings have indicated that some in-shoe pressure
assessment systems may only be suitable for recording peak pressures in the range of
200 to 300 kPa (Price et al. 2016) and that these measurements often had poor
reproducibility. In this study, two in-shoe pressure measurement devices were most
accurate between 200 and 300 kPa, and the contact area was relatively repeatable for
all systems (Price et al. 2016). The largest error in peak pressures ranged from 50 kPa
for capacitive devices and up to 600 kPa for resistive devices (Price et al. 2016). The
authors speculated that the reduced accuracy and precision of the resistive systems
demonstrates higher variability between sensors and could be attributed to inherent
Fig. 6 The two feet demonstrate the coefficients of variation for foot pressure at ten plantar sites
assessed with a pressure platform in individuals with a plantar foot ulcer and neuropathy (n = 4)
(gray), in individuals with diabetes mellitus without neuropathy and foot ulcers (orange) (n = 5),
and in healthy controls (n = 5) (blue). The platform used was the Footscan ® pressure plate (RSScan
International, Olen, Belgium). This plate was 2 m in length and 0.4 m in width and contained 16,384
sensors, with individual sensor dimensions of 0.0076 m 0.0051 m. All pressure data were
captured at rate of 100 Hz. Foot pressure measurements were performed over 5 days and are the
average results from five assessments, averaged across the two feet. The two graphs represent two
outcomes of foot pressure. The top graph represents the coefficient of variation of mean peak
pressure; the bottom graph represents the coefficient of variation of pressure–time integral. The
x-axis represents the various sites of foot pressure measurement; the y-axis represents the coefficient
of variation as a percentage (%) (Adapted from Fernando et al. 2016a)
sensor noise (Price et al. 2016). Therefore, this questions the use of foot pressure
assessments in quantifying pressures above this level.
Arts and Bus (2011) carried out a study to assess the ideal number of steps needed
for reproducible and valid foot pressure data in individuals with peripheral diabetic
neuropathy when wearing custom-made footwear. Their results indicated that
12 steps per foot were required to obtain valid and reliable in-shoe foot pressure
data (Arts and Bus 2011) in this population. Between 7 and 17 footsteps per foot
were required to obtain reliable data, depending on the parameter of interest, but
12 steps per foot were required to obtain both valid and reliable in-shoe pressure data
in all parameters (Arts and Bus 2011). This provides some novel insight regarding
the type of protocol, which should be considered when measuring in-shoe foot
pressures in a high-risk population.
Factors that Influence Foot Pressure Measurements
Understanding the role of elevated foot pressure in the neuropathic foot has been
previously highlighted as an important consideration, if not the most important
consideration toward the prevention and treatment of foot ulcers in individuals
with diabetes mellitus and diabetic peripheral neuropathy (Brand 1979; Brand and
Ebner 1969; Cavanagh et al. 1993, 2000; Formosa et al. 2013; Boulton et al. 1983).
However, there still remains a lack of understanding regarding the determinants of
elevated foot pressure in individuals with diabetes mellitus with and without the
coexistence of peripheral neuropathy (Barn et al. 2015). These determinants can
broadly be divided into laboratory- and protocol-related influences and patient-
specific influences.
Laboratory Environment and Protocol-Related Influences

The effect of gait protocol, targeting, and laboratory environment on the foot
pressures of individuals with diabetes mellitus, in particular in individuals with
diabetic peripheral neuropathy, is largely unknown and needs to be investigated.
The gait speed is an important confounder affecting foot pressure measurements
(Burnfield et al. 2004; Rosenbaum et al. 1994). Hence, gait speed is often tightly
controlled under research conditions to obtain reproducible results and to avoid the
confounding effect of speed on foot pressures (Segal et al. 2004; Burnfield et al.
2004; Rosenbaum et al. 1994; Warren et al. 2004). However, the application of such
results to the everyday walking characteristics of individuals is unclear. Hence
studying the foot pressures of individuals during their natural speed of walking
may offer measurements that are more realistic (Fernando et al. 2015). A limitation
of using natural gait speed is that it is context specific and is dependent on the task or
activity that is being performed. Faster gait speeds (i.e., shorter stance phase
durations) have been shown to increase foot pressure at the heel, medial and central
forefoot, and lesser toes, while decreasing foot pressure beneath the midfoot and
lateral forefoot (Burnfield et al. 2004; Rosenbaum et al. 1994; Hughes et al. 1991).
This has been termed a medialization of the loading pattern (Rosenbaum et al. 1994).
As patients with neuropathy are likely to have a slower gait speed, especially after
the development of foot ulcers, controlling gait speed or adjusting for gait speed may
be an important consideration in this population (Fernando et al. 2016b).
Patient-Specific Influences
Restriction in range of motion of the foot (particularly at the first metatarso-
phalangeal joint and ankle joint) which is often associated with peripheral neurop-
athy has been associated with significantly elevated foot pressures and hence an
increased risk of ulceration in individuals with diabetic peripheral neuropathy
(Fernando et al. 1991; Payne et al. 2002; Armstrong and Lavery 1998; Goldsmith
et al. 2002; Hastings et al. 2000; Mueller et al. 1994b; Orendurff et al. 2006; Rao
et al. 2006). However, elevated foot pressures in the absence of neuropathy are not
attributed to ulceration, as the presence of neuropathy has been termed a requirement
for plantar ulceration (Masson et al. 1989; van Schie and Boulton 2002; van Schie
2005). Alterations in lower limb joint moments, which often results from diabetic
peripheral neuropathy and include an increased plantarflexon moment and longer
mid-stance phase, have been associated with elevated forefoot foot pressures in
individuals with diabetic peripheral neuropathy (Savelberg et al. 2009). It has also
been proposed that intrinsic foot muscle atrophy and extensor tendon substitution
coupled with flexor tendon stabilization (a compensatory biomechanical strategy)
can result in increased supinatory moments and increased pressure under the lateral
forefoot in individuals with foot ulcers (Stess et al. 1997).
The presence of foot deformity has repeatedly been observed to be an important
predictor for higher foot pressures in individuals with diabetes (Bus et al. 2005;
Guldemond et al. 2008; Mueller et al. 2003a). Foot deformity also often results in
plantar callus and thickening of the skin (Abouaesha et al. 2001). In a landmark
study by Cavanagh and colleagues (1997), foot structure as defined by 27 radiolog-
ical measures were compared to regional foot pressure distributions in healthy
participants. Using multiple regression analyses, their findings could only account
for up to 38 % of the variance in peak foot pressure at the heel and first metatarsal,
respectively (Cavanagh et al. 1997). They concluded that gait and individual vari-
ation in gait are more likely to exert the major influences on peak plantar foot
pressures during walking than foot structure. Since this study, other studies have
investigated the relationship between clinical and structural variables on in-shoe or
platform-based foot pressures in diverse populations with diabetes and obtained
variable results (Ahroni et al. 1999; Payne et al. 2002; Cavanagh et al. 1991; Mueller
et al. 2003a). Therefore, the influence of variables such as age, body weight, duration
of diabetes, plantar soft tissue thickness, and range of motion of foot joints is largely
unknown.
More recently, Barn and colleagues (2015) studied patients with diabetic periph-
eral neuropathy and a history of ulceration to determine the predictors of peak foot
pressure using multivariate linear regression analyses in this high-risk population.
They determined that different factors were likely to influence foot pressures beneath
foot sites. They were only able to predict between 6 % (heel) and 41 % (midfoot) of
the variation in peak foot pressure (Barn et al. 2015). The largest contributing factor
to peak foot pressure at the heel was glycosylated hemoglobin concentration (hbA1c)
(17.8 %), whereas in the midfoot it was the presence of Charcot neuroarthropathy
(a complication of severe neuropathy) (41.1 %) and the presence of prominent
metatarsal heads in the forefoot (37.3 %) (Barn et al. 2015). Presence of hammertoe
deformity (31.9 %) was the biggest contributing factor to elevated plantar pressures at
the lesser toes as opposed to a history of previous ulceration (32.9 %) which was the
biggest contributor for elevated plantar pressures at the hallux (Barn et al. 2015).
Overall, their results indicated that local influences such as the presence of foot
deformity were stronger predictors of foot pressure than global features such as the
body mass, age, gender, and diabetes duration (Barn et al. 2015). This is consistent
with findings in healthy populations indicating that the weight and age only account
for X % of the variability in foot pressures (Martínez-Nova et al. 2008).
Interestingly, other findings have highlighted the potential role of glycation of
collagen-rich tissues, particularly in the plantar fascia of the foot and the posterior
muscle compartments of the ankle, in the pathogenesis of diabetes-related foot
problems (D’Ambrogi et al. 2003; Giacomozzi et al. 2005). More recent investigations
have shown a relationship between potential glycation of collagen in the Achilles
tendon and higher forefoot pressures in individuals with poorly controlled diabetes
mellitus (Couppe et al. 2016). These findings are consistent with other work indicating
an association between skin autofluorescence (a surrogate measure of glycation of
collagen) and foot ulcers in patients with type 2 diabetes and vascular complications
(Liu et al. 2015). These findings indicate that the aetiology of the mechanical changes
that are observed as changes in foot pressure may be more complicated than previously
thought and could include more intricate structural changes within the foot. Novel
investigations such as the measurement of skin autofluorescence are providing new
and innovative data regarding the factors previously thought to influence foot pressure,
but which were previously unable to be assessed.
Foot Pressure in Individuals with Diabetes Mellitus
Identifying the foot pressure profiles of patients with diabetes mellitus has been a
topic of interest for a number of decades (Boulton et al. 1983, 1987; Veves et al.
1992; Armstrong et al. 1998; Lavery et al. 2003; Masson et al. 1989; Pham et al.
2000; Cavanagh et al. 1993, 2000). Following early work done by Brand and
colleagues in patients with leprosy (Brand 1979; Brand and Ebner 1969), it was
apparent that foot pressures could be used to potentially identify areas of the plantar
foot that were prone to ulceration. Hence, a mechanical aetiology to diabetes foot
ulcers was thought to exist (Stokes et al. 1975).
There are now at least four systematic reviews that have compiled results of
several studies which investigated the foot pressure features of individuals with
diabetes mellitus (Wrobel and Najafi 2010; Allet et al. 2008; Fernando et al. 2013,
2014). These reviews have suggested a progressive increase in the foot pressures of
people with diabetes neuropathy compared to controls without neuropathy and a
subsequent substantial increase in foot pressures following foot ulceration. However,
one of the key limitations has been that although many studies have proposed that
foot pressures can be used as a surrogate measure of trauma to the plantar foot,
current evaluation methods suffer from various shortcomings including the lack of a
threshold of peak foot pressure that predicts the development of foot ulcers (Wrobel
and Najafi 2010). Foot pressures are thought to be significantly higher in individuals
with diabetes mellitus, however, than in healthy controls (Cavanagh et al. 1993).
What is clinically relevant is that high foot pressures alone are unlikely to cause
ulceration in individuals with diabetes with adequate sensation in the absence of
diabetic peripheral neuropathy (Masson 1992; Masson et al. 1989; Fernando et al.
1991; Boulton 1998; Boulton et al. 1983, 1987, 2004). Therefore, assessments of
foot pressure are most important in individuals with diabetic peripheral neuropathy.
Foot Pressure in Individuals with High-Risk Feet
A large number of retrospective and prospective studies have shown that foot
pressures are elevated in patients with diabetic peripheral neuropathy at high risk
of foot ulceration, as well as in those with a past history of ulceration compared to
healthy controls (Veves et al. 1992; Wrobel and Najafi 2010; Frykberg et al. 1998;
Pham et al. 2000; Boulton et al. 1983; Fernando et al. 2013, 2014). The frequency,
magnitude, and duration of foot pressure can either individually or cumulatively
contribute to increased loading of the foot in patients with diabetic peripheral
neuropathy (van Schie 2005). This predisposes plantar tissues to trauma and to
skin breakdown and ulceration (van Schie 2005). In a meta-analysis of foot pressure
acquired using platform measurements, it was evident that individuals with diabetic
peripheral neuropathy experienced significantly higher foot pressures compared to
controls (Fernando et al. 2013). Specifically, the peak forefoot and rearfoot pressure
and pressure–time integrals were higher in individuals with neuropathy compared to
individuals with diabetes without neuropathy and to healthy controls (Fernando et al.
2013). A limitation of this finding was the fact that not all confounding factors were
controlled in individual studies.
A prior history of foot ulceration is thought to be associated with significantly
high foot pressure measurements in the forefoot in individuals with diabetic periph-
eral neuropathy, irrespective of whether the ulcer has healed (Fernando et al. 2014).
An important reason for the higher incidence of skin breakdown in the forefoot than
in the rearfoot is that the ratio of foot pressure in the forefoot compared to the
rearfoot is significantly higher during barefoot ambulation (Caselli et al. 2002).
Therefore, the risk of subsequent ulceration does not reduce in this group of
individuals and warrants appropriate off-loading interventions (Raspovic et al.
2000; Tsung et al. 2004; Bus et al. 2008a). A systematic discussion of off-loading
options in patients with diabetes mellitus is beyond the scope of this chapter.
However, when assessing the efficacy and appropriateness of off-loading interven-
tions, foot pressure assessments can be used and are recommended to obtain
quantitative evaluations of off-loading (Bus et al. 2013). Figure 7 demonstrates
site-specific differences in plantar foot pressure in a person with foot ulcers and
16
Mean peak pressure
Mean peak foot pressure (N/cm2)
14
12
10
8
Foot-ulcer + neuropathy
6
Diabetes without neuropathy
4
Healthy control
2
0
Location of foot pressure
Fig. 7 The graph demonstrates the mean peak foot pressure (N/cm2) at ten plantar sites assessed
with a pressure platform in a person with an active plantar foot ulcer and neuropathy (gray), a
person with diabetes mellitus without neuropathy and foot ulcers (orange), and in a healthy control
(blue) during self-selected gait speeds. The platform used was the Footscan ® pressure plate
(RSScan International, Olen, Belgium). This plate was 2 m in length and 0.4 m in width and
contained 16,384 sensors, with individual sensor dimensions of 0.0076 m 0.0051 m. All the
pressure data were captured at rate of 100 Hz. Gait speed was not adjusted in these analyses to
assess the natural pressure distribution, and this may have affected the pressure results. Foot
pressure measurements are averaged from five bilateral measurements, except in the person with
the foot ulcer, whose foot pressures are only representative of the ulcerated foot. The x-axis
represents the various sites of foot pressure measurement; the y-axis represents the mean peak
foot pressures in N/cm2 (Adapted from Fernando et al. (2016a))
diabetic peripheral neuropathy, compared to a person with diabetes mellitus without

peripheral neuropathy and a healthy control. A large difference in foot pressures in the
presence of neuropathy and ulceration can be appreciated. Foot pressures have also
been shown to increase in individuals with diabetes mellitus in the presence of digital
and partial foot amputations, thereby increasing the mechanical stress and trauma to
the tissue and increasing the risk of subsequent ulceration (Kanade et al. 2006a, b).
Therefore, foot pressures demonstrate a gradual increase in individuals with diabetes
mellitus, in the presence of high-risk feet, and subsequent foot complications.
Future Directions
Ultimately, Could a Threshold Be Found for Identifying Individuals at

Risk of Ulceration?
The ideal scenario is to be able to utilize foot pressures to predict sites of potential
ulceration, prior to ulcer occurrence, and to guide pressure off-loading of ulcerated
sites effectively. While progress has been made with such applications, proposed
thresholds have been inadequate due to a number of limitations including the poor
agreement between various methods and instrumentation used to measure foot
pressure, the lack of correlation between shod and barefoot foot pressure assess-
ments, and a large between-day variability in obtained measurements. This had led
to an uncertainty in what is clinically important as opposed to what is the expected
natural variation in measurements in patients. The extent to which areas of high or
elevated foot pressure corresponds with areas of ulceration is also controversial
(Cavanagh et al. 2000; van Schie 2005).
Several attempts have been made in the past to identify ulceration risk in diabetes
patients utilizing foot pressure. Lavery et.al investigated risk factors for foot ulcer-
ation in 225 patients with diabetes mellitus with and without ulceration and
established that a foot pressure >650 kPa contributed a high ulceration risk (Lavery
et al. 1998). In a case–control study of 219 participants with and without foot
ulceration, Armstrong et.al (Armstrong et al. 1998) determined that the optimal
cutoff point for ulceration was 700 kPa, which generated a sensitivity of 70 % and
a specificity of 65 %. As there was no optimal cutoff point for screening patients for
ulceration beyond this, it was concluded that the higher the peak pressure, the higher
the corresponding risk for ulceration.
In another longitudinal study comprising 1,666 consecutive patients with diabetes
mellitus, the optimum cutoff for ulceration was believed to be 875 kPa, with a
sensitivity of 64 % and a specificity of 46 % (Lavery et al. 2003). Owings and
colleagues (2009) studied 49 individuals with a history of diabetic foot ulcers and
identified that an in-shoe pressure cutoff of 207 kPa should be utilized as a threshold
target for footwear prescription for individuals with a history of foot ulcers (Owings
et al. 2009). However, barefoot peak pressure only predicted approximately 35 % of
the variance of in-shoe peak pressure, indicating that other factor such as individual
footwear prescriptions may have a bigger influence of in-shoe foot pressure mea-
surements (Owings et al. 2009). These findings have also suggested that foot
pressure assessment by itself was a poor predictive tool for ulceration. A systematic
review and meta-analysis, however, concluded that high foot pressures increased the
risk of foot ulceration (Crawford et al. 2007). Nonetheless, the results from these
studies have been inconclusive and the strength of evidence restricted in determining
a threshold pressure for ulceration. Hence, foot pressure measurements alone cannot
be utilized as a diagnostic tool for foot ulceration.
The threshold foot pressure at which human skin ulcerates is unknown, although
several attempts have been made to identify this silver bullet (Frykberg et al. 1998;
Armstrong et al. 1998; Veves et al. 1992). The reality may be that there is no such
threshold which is able to be applied to all individuals with diabetes and diabetic
peripheral neuropathy. Hence, the foot pressure value likely responsible for ulcer
development varies between individual to individual, based on body composition,
ethnicity, sex, the range of motion of lower limb joints, gait features, and on-foot
morphology and structure as earlier reported. Therefore a pressure threshold for
ulceration is influenced by other factors that alter the biological composition and
mechanical properties of plantar tissue and their viability (Bacarin et al. 2009).
Factors such as glycation of the plantar fascia and the Achilles tendon and fatty
infiltration on the intrinsic foot muscles due to diabetes mellitus have also been
suggested to alter both the stress within the plantar tissues and their response to stress
(Craig et al. 2008; Huijberts et al. 2008; Wrobel and Najafi 2010; D’Ambrogi et al.
2003; Giacomozzi et al. 2005). Until all the factors that influence foot pressures are
identified, it may be difficult to establish a threshold for ulceration. Perhaps this
means that thresholds are so individualized that it should be based on a formula or a
composite score that takes into account various measurements of which foot pres-
sures are one factor. A composite score should take into account variables such as the
range of movement and clinical and radiological measures together with foot
pressures to identify individuals at risk of ulceration and locations of plantar
ulceration. Such an approach is however not currently utilized.
What Research Is Required in order to Make Foot Pressures More

Effective and Beneficial?
Although important changes in foot pressures can be assessed throughout the

sequelae of foot complications of diabetes mellitus including the development of
diabetic peripheral neuropathy and foot ulceration, there is a lack of correlation
between sites of ulceration and sites of peak pressure. While this has resulted in some
questioning the utility of foot pressure assessments in diabetes mellitus, considerably
more research is required to improve our understanding of factors that influence foot
pressures and how these may be altered in diabetes. Irrespective of the absence of a
threshold for ulceration, international consensus should be made on the methods
used to collect both platform and in-shoe foot pressure data. Consensus and agree-
ment needs to take place in the protocols for foot pressure measurements, the
frequency of measurement, and the expected deviations due to measurement inac-
curacy and on how measurement error should be minimized within the field. This
may be challenging given the differences between laboratories, patients, and pro-
tocols used. As a bare minimum, reporting requirements should be adhered to in
studies allowing for open disclosure of protocols and methods used to obtain plantar
pressures. Such recommendations may allow for comparisons that are more appro-
priate and at a global standard between studies, patients, and institutions. Further-
more, consensus needs to be made on how and when elevated foot pressures should
be clinically managed in patients with diabetes mellitus at different stages of
diabetes-related foot complications. This has already started to take place with
international guidelines, and such guidelines need to be revised and advanced within
the field (Bus et al. 2015).
A major focus of future direction should be addressing the current paucity in the
reproducibility of pressure measurement between devices. As there is no interna-
tional recommendation on key requirements, there is no current requirement for
producers to adhere to when to manufacture devices. It is especially important to
consider the spatial factors and the temporal behavior of sensors that are used (Urry
1999). This is of particular importance with more recent focus on shear pressures
(Yavuz et al. 2008). The current dogma is that the choice of sensor is based on the
principle of matching the required performance profile to the particular measurement

task. Therefore, further research should focus on testing the special characteristic of
existing sensors and in developing newer sensors that are sensitive and have a range
of performance characteristics including the ability to detect shear pressures.
Foot pressure assessments are currently limited due to significant differences
between measurement protocols, measurement methods, types of sensors used,
and a lack of international consensus on key requirements for foot pressure assess-
ments in individuals with diabetes mellitus. Furthermore, there are currently no
commercially available devices to measure shear pressures, and hence currently
plantar pressures at best only provide partial observations regarding the mechanical
characteristics of individuals with diabetes-related foot complications. With
advancements in technology, it is anticipated that the foot pressure assessment
may be more readily employed in evaluating the mechanical characteristics of
individuals with diabetes-related foot complications.
Cross-References

▶ Assessing Pediatric Foot Deformities
Applications
▶ Plantar Pressure
▶ The Use of Low Resolution Pedoboragraphs
References
Abdul Razak AH, Zayegh A, Begg RK, Wahab Y (2012) Foot plantar pressure measurement
system: a review. Sensors 12:9884
Abouaesha F, van Schie CH, Griffths GD, Young RJ, Boulton AJ (2001) Plantar tissue thickness is
related to peak plantar pressure in the high-risk diabetic foot. Diabetes Care 24:1270–1274
Acharya UR, Ghista DN, Nergui M, Chattopadhyay S, Ng EYK, Sree SV, Tong JWK, Tan JH,
Meng LK, Suri JS (2012) Diabetes mellitus: enquiry into its medical aspects and bioengineering
of its monitoring and regulation. J Mech Med Biol 12
Ahroni JH, Boyko EJ, Forsberg R (1998) Reliability of F-scan in-shoe measurements of plantar
pressure. Foot Ankle Int 19:668–673
Ahroni JH, Boyko EJ, Forsberg RC (1999) Clinical correlates of plantar pressure among diabetic
veterans. Diabetes Care 22:965–972
Alexander IJ, Chao EY, Johnson KA (1990) The assessment of dynamic foot-to-ground contact
forces and plantar pressure distribution: a review of the evolution of current techniques and
clinical applications. Foot Ankle 11:152–167
Allet L, Armand S, Golay A, Monnin D, DE Bie R, de Bruin E (2008) Gait characteristics of
diabetic patients: a systematic review. Diabetes/Metabo Res Rev 24:173–191
Armstrong DG (2005) Detection of diabetic peripheral neuropathy: strategies for screening and
diagnosis. Adv Stud Med 5:S1033–S1037
Armstrong DG, Lavery LA (1998) Plantar pressures are higher in diabetic patients following partial
foot amputation. Ostomy Wound Manag 44:30–32, 34, 36 passim
Armstrong DG, Peters EJ, Athanasiou KA, Lavery LA (1998) Is there a critical level of plantar foot
pressure to identify patients at risk for neuropathic foot ulceration? J Foot Ankle Surg
37:303–307
Armstrong DG, Stacpoole-Shea S, Nguyen H, Harkless LB (1999) Lengthening of the Achilles
tendon in diabetic patients who are at high risk for ulceration of the foot. J Bone Joint Surg
Series A 81:535–538
Arts MLJ, Bus SA (2011) Twelve steps per foot are recommended for valid and reliable in-shoe
plantar pressure data in neuropathic diabetic patients wearing custom made footwear. Clin
Biomech 26:880–884
Bacarin TA, Sacco ICN, Hennig EM (2009) Plantar pressure distribution patterns during gait in
diabetic neuropathy patients with a history of foot ulcers. Clinics 64:113–120
Barn R, Waaijman R, Nollet F, Woodburn J, Bus SA (2015) Predictors of barefoot plantar pressure
during walking in patients with diabetes, peripheral neuropathy and a history of ulceration.
PLoS One 10:e0117443
Barnett S (1998) International protocol guidelines for plantar pressure measurement. Diabet Foot
1:137–140
Boulton AJ (1998) Lowering the risk of neuropathy, foot ulcers and amputations. Diabet Med 15
(Suppl 4):S57–S59
Boulton AJ, Hardisty CA, Betts RP, Franks CI, Worth RC, Ward JD, Duckworth T (1983) Dynamic
foot pressure and other studies as diagnostic and management aids in diabetic neuropathy.
Diabetes Care 6:26–33
Boulton AJ, Betts RP, Franks CI, Ward JD, Duckworth T (1987) The natural history of foot pressure
abnormalities in neuropathic diabetic subjects. Diabetes Res 5:73–77
Boulton AJ, Krisner RS, Vileikyte L (2004) Neuropathic diabetic foot ulcers. N Engl J Med
351:48–55
Brand PW (1979) Management of the insensitive limb. Phys Ther 59:8–12
Brand PW, Ebner JD (1969) Pressure sensitive devices for denervated hands and feet. A preliminary
communication. J Bone Joint Surg Am 51:109–116
Bryant A, Singer K, Tinley P (1999) Comparison of the reliability of plantar pressure measurements
using the two-step and midgait methods of data collection. Foot Ankle Int 20:246–250
Burnfield JM, Few CD, Mohamed OS, Perry J (2004) The influence of walking speed and footwear
on plantar pressures in older adults. Clin Biomech (Bristol, Avon) 19:78–84
Bus SA (2008) Foot structure and footwear prescription in diabetes mellitus. Diabetes Metab Res
Rev 24:S90–S95
Bus SA, de Lange A (2005) A comparison of the 1-step, 2-step, and 3-step protocols for obtaining
barefoot plantar pressure data in the diabetic neuropathic foot. Clin Biomech 20:892–899
Bus SA, Waaijman R (2013) The value of reporting pressure–time integral data in addition to peak
pressure data in studies on the diabetic foot: a systematic review. Clin Biomech (Bristol, Avon)
28:117–121
Bus SA, Ulbrecht JS, Cavanagh PR (2004) Pressure relief and load redistribution by custom-made
insoles in diabetic patients with neuropathy and foot deformity. Clin Biomech 19:629–638
Bus SA, Maas M, de Lange A, Michels RP, Levi M (2005) Elevated plantar pressures in neuro-
pathic diabetic patients with claw/hammer toe deformity. J Biomech 38:1918–1925
Bus S, van Deursen R, Kanade R, Wissink M, Manning E, van Baal J, Harding K (2008a) Plantar
pressure relief in the diabetic foot using forefoot offloading shoes. Gait Posture 29:618–620
Bus SA, Valk GD, van Deursen RW, Armstrong DG, Caravaggi C, Hlavacek P, Bakker K,
Cavanagh PR (2008b) The effectiveness of footwear and offloading interventions to prevent
and heal foot ulcers and reduce plantar pressure in diabetes: a systematic review. Diabetes Metab
Res Rev 24(Suppl 1):S162–S180
Bus SA, Haspels R, Busch-Westbroek TE (2011) Evaluation and optimization of therapeutic

footwear for neuropathic diabetic foot patients using in-shoe plantar pressure analysis. Diabetes
Care 34:1595–1600
Bus SA, Waaijman R, Arts M, DE Haart M, Busch-Westbroek T, van Baal J, Nollet F (2013) Effect
of custom-made footwear on foot ulcer recurrence in diabetes: a multicenter randomized
controlled trial. Diabetes Care 36:4109–4116
Bus SA, Armstrong DG, van Deursen R, Lewis J, Caravaggi C, Cavanagh PR, (IWGDF),
OBOTIWGOTDF (2015) IWGDF Guidance on footwear and offloading interventions to
prevent and heal foot ulcers in patients with diabetes [Online]. IWGDF. Available: http://
www.iwgdf.org/files/2015/website_footwearoffloading.pdf. Accessed 14th July 2015
Caselli A, Pham H, Giurini JM, Armstrong DG, Veves A (2002) The forefoot-to-rearfoot plantar
pressure ratio is increased in severe diabetic neuropathy and can predict foot ulceration.
Cavanagh PR, Ulbrecht JS (1994) Clinical plantar pressure measurement in diabetes: rationale and
methodology. Foot 4:123–135
Cavanagh PR, Sims DS Jr, Sanders LJ (1991) Body mass is a poor predictor of peak plantar pressure
in diabetic men. Diabetes Care 14:750–755
Cavanagh PR, Simoneau GG, Ulbrecht JS (1993) Ulceration, unsteadiness, and uncertainty: the
biomechanical consequences of diabetes mellitus. J Biomech 26(Suppl 1):23–40
Cavanagh PR, Morag E, Boulton AJ, Young MJ, Deffner KT, Pammer SE (1997) The relationship
of static foot structure to dynamic foot function. J Biomech 30:243–250
Cavanagh P, Ulbrecht J, Caputo G (2000) New developments in the biomechanics of the diabetic
foot. Diabetes Metab Res Rev 16(Suppl 1):S6–S10
Choi YR, Lee HS, Kim DE, Lee DH, Kim JM, Ahn JY (2014) The diagnostic value of
pedobarography. Orthopedics 37:e1063–e1067
Couppe C, Svensson RB, Kongsgaard M, Kovanen V, Grosset JF, Snorgaard O, Bencke J, Larsen
JO, Bandholm T, Christensen TM, Boesen A, Helmark IC, Aagaard P, Kjaer M, Magnusson SP
(2016) Human Achilles tendon glycation and function in diabetes. J Appl Physiol 120:130–137
Craig ME, Duffin AC, Gallego PH, Lam A, Cusumano J, Hing S, Donaghue KC (2008) Plantar
fascia thickness, a measure of tissue glycation, predicts the development of complications in
adolescents with type 1 diabetes. Diabetes Care 31:1201–1206
Crawford F, Inkster M, Kleijnen J, Fahey T (2007) Predicting foot ulcers in patients with diabetes: a
systematic review and meta-analysis. QJM 100:65–86
D’ambrogi E, Giurato L, D’agostino MA, Giacomozzi C, Macellari V, Caselli A, Uccioli L (2003)
Contribution of plantar fascia to the increased forefoot pressures in diabetic patients. Diabetes
Care 26:1525–1529
Davis B, Parry J, Neth DC, Waters KC (1998) A device for simultaneous measurement of pressure
and shear force distribution on the plantar surface of the foot. J Appl Biomech 14:93–107
Fernando DJ, Masson EA, Veves A, Boulton AJ (1991) Relationship of limited joint mobility to
abnormal foot pressures and diabetic foot ulceration. Diabetes Care 14:8–11
Fernando M, Crowther R, Lazzarini P, Sangla K, Cunningham M, Buttner P, Golledge J (2013)
Biomechanical characteristics of peripheral diabetic neuropathy: a systematic review and meta-
analysis of findings from the gait cycle, muscle activity and dynamic barefoot plantar pressure.
Clin Biomech (Bristol, Avon) 28(8):831–845
Fernando ME, Crowther RG, Pappas E, Lazzarini PA, Cunningham M, Sangla KS, Buttner P,
Golledge J (2014) Plantar pressure in diabetic peripheral neuropathy patients with active foot
ulceration, previous ulceration and no history of ulceration: a meta-analysis of observational
studies. PLoS One 9:e99050
Fernando ME, Crowther RG, Cunningham M, Lazzarini PA, Sangla KS, Golledge J (2015) Lower
limb biomechanical characteristics of patients with neuropathic diabetic foot ulcers: the diabetes
foot ulcer study protocol. BMC Endocr Disord 15:59
Fernando M, Crowther R, Cunningham M, Lazzarini P, Sangla K, Buttner P, Golledge J (2016a)
The reproducibility of acquiring three dimensional gait and plantar pressure data using
established protocols in participants with and without type 2 diabetes and foot ulcers. J Foot
Ankle Res 9:4
Fernando ME, Crowther RG, Lazzarini PA, Sangla KS, Buttner P, Golledge J (2016b) Gait
parameters of people with diabetes-related neuropathic plantar foot ulcers. Clin Biomech
37:98–107
Ferrarin M, Bovi G, Rabuffetti M, Mazzoleni P, Montesano A, Moroni I, Pagliano E, Marchi A,
Marchesi C, Beghi E, Pareyson D (2011) Reliability of instrumented movement analysis as
outcome measure in Charcot–Marie–Tooth disease: results from a multitask locomotor protocol.
Formosa C, Gatt A, Chockalingam N (2013) The importance of clinical biomechanical assessment
of foot deformity and joint mobility in people living with type-2 diabetes within a primary care
setting. Prim Care Diabetes 7:45–50
Frykberg RG, Lavery LA, Pham H, Harvey C, Harkless L, Veves A (1998) Role of neuropathy and
high foot pressures in diabetic foot ulceration. Diabetes Care 21:1714–1719
Giacomozzi C (2010) Appropriateness of plantar pressure measurement devices: a comparative
technical assessment. Gait Posture 32:141–144
Giacomozzi C, D’ambrogi E, Uccioli L, Macellari V (2005) Does the thickening of Achilles tendon
and plantar fascia contribute to the alteration of diabetic foot loading? Clin Biomech
20:532–539
Gillespie KA, Dickey JP (2003) Determination of the effectiveness of materials in attenuating high
frequency shock during gait using filterbank analysis. Clin Biomech (Bristol, Avon) 18:50–59
Goldsmith JR, Lidtke RH, Shott S (2002) The effects of range-of-motion therapy on the plantar
pressures of patients with diabetes mellitus. J Am Podiatr Med Assoc 92:483–490
Guldemond NA, Leffers P, Schaper NC, Sanders AP, Nieman F, Willems P, Walenkamp GHIM
(2007) The effects of insole configurations on forefoot plantar pressure and walking conve-
nience in diabetic patients with neuropathic feet. Clin Biomech 22:81–87
Guldemond NA, Leffers P, Walenkamp GHIM, Schaper NC, Sanders AP, Nieman FHM, van Rhijn
LW (2008) Prediction of peak pressure from clinical and radiological measurements in patients
with diabetes. BMC Endo Disord 8:16
Gurney JK, Kersting UG, Rosenbaum D (2008) Between-day reliability of repeated plantar pressure
distribution measurements in a normal population. Gait Posture 27:706–709
Gurney JK, Marshall PW, Rosenbaum D, Kersting UG (2013) Test-retest reliability of dynamic
plantar loading and foot geometry measures in diabetics with peripheral neuropathy. Gait
Posture 37:135–137
Hafer JF, Lenhoff MW, Song J, Jordan JM, Hannan MT, Hillstrom HJ (2013) Reliability of plantar
pressure platforms. Gait Posture 38:544–548
Hamatani M, Mori T, Oe M, Noguchi H, Takehara K, Amemiya A, Ohashi Y, Ueki K, Kadowaki T,
Sanada H (2016) Factors associated with callus in diabetic patients, focused on plantar shear
stress during gait. J Diabetes Sci Technol 10(6):1353–1359
Hastings M, Mueller M, Sinacore D, Salsich G, Engsberg J, Johnson J (2000) Effects of a tendo-
achilles lengthening procedure on muscle function and gait characteristics in a patient with
diabetes mellitus. J Orthop Sports Phys Ther 30:85–90
Hughes J (1993) The clinical use of pedobarography. Acta Orthop Belg 59:10–16
Hughes J, Pratt L, Linge K, Clark P, Klenerman l (1991) Reliability of pressure measurements: the
EM ED F system. Clin Biomech (Bristol, Avon) 6: 14–18.
Huijberts MS, Schaper NC, Schalkwijk CG (2008) Advanced glycation end products and diabetic
foot disease. Diabetes Metab Res Rev 24(Suppl 1):S19–S24
Kanade RV, Van Deursen RWM, Harding K, Price P (2006a) Walking performance in people with
diabetic neuropathy: benefits and threats. Diabetologia 49:1747–1754
Kanade RV, van Deursen RWM, Price P, Harding K (2006b) Risk of plantar ulceration in diabetic
patients with single-leg amputation. Clin Biomech 21:306–313
Kato H, Takada T, Kawamura T, Hotta N, Torii S (1996) The reduction and redistribution of plantar
pressures using foot orthoses in diabetic patients. Diabetes Res Clin Pract 31:115–118
Lavery LA, Armstrong DG, Vela SA, Quebedeaux TL, Fleischli JG (1998) Practical criteria for
screening patients at high risk for diabetic foot ulceration. Arch Intern Med 158:157–162
Lavery LA, Armstrong DG, Wunderlich RP, Tredwell J, Boulton AJM (2003) Predictive value of
foot pressure assessment as part of a population-based diabetes disease management program.
Ledoux WR, Shofer JB, Cowley MS, Ahroni JH, Cohen V, Boyko EJ (2013) Diabetic foot ulcer
incidence in relation to plantar pressure magnitude and measurement location. J Diabetes
Complications 27:621–626
Lee KM, Lee J, Chung CY, Ahn S, Sung KH, Kim TW, Lee HJ, Park MS (2012) Pitfalls and
important issues in testing reliability using intraclass correlation coefficients in orthopaedic
research. Clin Orthop Surg 4:149–155
Liu C, Xu L, Gao H, Ye J, Huang Y, Wu M, Xie T, Ni P, Yu X, Cao Y, Lu S (2015) The association
between skin autofluorescence and vascular complications in Chinese patients with diabetic foot
ulcer: an observational study done in Shanghai. Int J Low Extrem Wounds 14:28–36
Lord M, Hosein R (2000) A study of in-shoe plantar shear in patients with diabetic neuropathy. Clin
Biomechan 15:278–283
Lord M, Reynolds DP, Hughes JR (1986) Foot pressure measurement: a review of clinical findings.
J Biomed Eng 8:283–294
Luo ZP, Berglund LJ, An KN (1998) Validation of F-Scan pressure sensor system: a technical note.
J Rehabil Res Dev 35:186–191
Martínez-Nova A, Huerta JP, Sánchez-Rodríguez R (2008) Cadence, age, and weight as determi-
nants of forefoot plantar pressures using the biofoot in-shoe system. J Am Podiatr Med Assoc
98:302–310
Masson EA (1992) What causes high foot pressures in diabetes: how can they be relieved?.
Proceedings of the IDF satellite symposium on the diabetic foot, Washington 1991. Foot
2:212–217
Masson EA, Hay EM, Stockley I, Veves A, Betts RP, Boulton AJM (1989) Abnormal foot pressures
alone may not cause ulceration. Diabet Med 6:426–428
McPoil TG, Cornwall MW, Yamada W (1995) A comparison of two in-shoe plantar pressupe
measurement systems. Lower Extrem 2:95–103
McPoil TG, Cornwall MW, Dupuis L, Cornwell M (1999) Variability of plantar pressure data. A
comparison of the two-step and midgait methods. J Am Podiatr Med Assoc 89:495–501
Melai T, Ijzerman TH, Schaper NC, DE Lange TLH, Willems PJB, Meijer K, Lieverse AG,
Savelberg HHCM (2011) Calculation of plantar pressure time integral, an alternative approach.
Meyers-Rice B, Sugars L, McPoil T, Cornwall MW (1994) Comparison of three methods for
obtaining plantar pressures in nonpathologic subjects. J Am Podiatr Med Assoc 84:499–504
Miller CA, Verstraete MC (1996) Determination of the step duration of gait initiation using a
mechanical energy analysis. J Biomech 29:1195–1199
Mueller M, Minor S, Sahrmann S, Schaaf J, Strube M (1994a) Differences in the gait characteristics
of patients with diabetes and peripheral neuropathy compared with aged-matched controls. . .
including commentary by McPoil T and Cavanaugh PR with author response. Phys Ther
74:299–313
Mueller MJ, Sinacore DR, Hoogstrate S, Daly L (1994b) Hip and ankle walking strategies: effect on
peak plantar pressures and implications for neuropathic ulceration. Arch Phys Med Rehabil
75:1196–1200
Mueller MJ, Hastings M, Commean PK, Smith KE, Pilgram TK, Robertson D, Johnson J (2003a)
Forefoot structural predictors of plantar pressures during walking in people with diabetes and
peripheral neuropathy. J Biomech 36:1009–1017
Mueller MJ, Sinacore DR, Hastings MK, Strube MJ, Johnson JE (2003b) Effect of Achilles tendon
lengthening on neuropathic plantar ulcers: a randomized clinical trial. J Bone Joint Surg Series
A 85:1436–1445
Mueller MJ, Zou D, Lott DJ (2005) “Pressure gradient” as an indicator of plantar skin injury.
Mueller MJ, Zou D, Bohnert KL, Turtle LJ, Sinacore DR (2008) Plantar stresses on the neuropathic
foot during barefoot walking. Phys Ther 88:1375–1384
Orendurff MS, Rohr ES, Sangeorzan BJ, Weaver K, Czerniecki JM (2006) An equinus doformity of
the ankle accounts for only a small amount of the increased forefoot plantar pressure in patients
with diabetes. J Bone Joint Surg Series B 88:65–68
Orlin M, McPoil T (2000) Plantar pressure assessment. Phys Ther 80:389–409
Owings TM, Apelqvist J, Stenstrom A, Becker M, Bus SA, Kalpen A, Ulbrecht JS, Cavanagh PR
(2009) Plantar pressures in diabetic patients with foot ulcers which have remained healed.
Diabet Med 26:1141–1146
Park SW, Kim Y-H (2007) Local shear stress measurements in the normal and diabetic foot patients
during walking. In: Magjarevic R, Nagel JH (eds) World congress on medical physics and
biomedical engineering 2006: August 27 – September 1, 2006 COEX Seoul, Korea “imaging
the future medicine”. Springer, Berlin/Heidelberg
Payne C, Turner D, Miller K (2002) Determinants of plantar pressures in the diabetic foot. J
Diabetes Complications 16:277–283
Pham H, Armstrong DG, Harvey C, Harkless LB, Giurini JM, Veves A (2000) Screening techniques
to identify people at high risk for diabetic foot ulceration: a prospective multicenter trial.
Pollard JP, le Quesne LP (1983) Method of healing diabetic forefoot ulcers. Br Med J (Clin Res Ed)
286:436–437
Price C, Parker D, Nester CJ (2014) Validity and repeatability of three commercially available
in-shoe pressure measurement systems. J Foot Ankle Res 7:A67
Price C, Parker D, Nester C (2016) Validity and repeatability of three in-shoe pressure measurement
systems. Gait Posture 46:69–74
Qiu X, Tian DH, Han CL, Chen W, Wang ZJ, Mu ZY, Liu KZ (2015) Plantar pressure changes and
correlating risk factors in Chinese patients with type 2 diabetes: preliminary 2-year results of a
prospective study. Chin Med J (Engl) 128:3283–3291
Rao S, Saltzman C, Yack HJ (2006) Ankle ROM and stiffness measured at rest and during gait in
individuals with and without diabetic sensory neuropathy. Gait Posture 24:295–301
Raspovic A, Newcombe L, Lloyd J, Dalton E (2000) Effect of customized insoles on vertical plantar
pressures in sites of previous neuropathic ulceration in the diabetic foot. Foot 10:133–138
Rosenbaum D, Becker HP (1997) Plantar pressure distribution measurements. Technical back-
ground and clinical applications. Foot Ankle Surg 3:1–14
Rosenbaum D, Hautmann S, Gold M, Claes L (1994) Effects of walking speed on plantar pressure
patterns and hindfoot angular motion. Gait Posture 2:191–197
Sacco IC, Bacarin TA, Canettieri MG, Hennig EM (2009) Plantar pressures during shod gait in
diabetic neuropathic patients with and without a history of plantar ulceration. J Am Podiatr Med
Assoc 99:285–294
Savelberg HH, Schaper NC, Willems PJ, de Lange TL, Meijer K (2009) Redistribution of joint
moments is associated with changed plantar pressure in diabetic polyneuropathy. BMC
Musculoskelet Disord 10:16
Segal A, Rohr E, Orendurff M, Shofer J, O’Brien M, Sangeorzan B (2004) The effect of walking
speed on peak plantar pressure. Foot Ankle Int 25:926–933
Stess RM, Jensen SR, Mirmiran R (1997) The role of dynamic plantar pressures in diabetic foot
ulcers. Diabetes Care 20:855–858
Stokes IA, Faris IB, Hutton WC (1975) The neuropathic ulcer and loads on the foot in diabetic
patients. Acta Orthop Scand 46:839–847
Tsung BYS, Zhang M, Mak AFT, Wong MWN (2004) Effectiveness of insoles on plantar pressure
redistribution. J Rehabil Res Dev 41:767–774
Urry S (1999) Plantar pressure-measurement sensors. Meas Sci Technol 10
van Schie CH (2005) A review of the biomechanics of the diabetic foot. Int J Low Extrem Wounds
4:160–170
van Schie CM, Boulton AM (2002) Biomechanics of the diabetic foot: the road to foot ulceration.
In: Veves A, Giurini J, Logerfo F (eds) The diabetic foot. Humana Press, Totowa
Verdini F, Marcucci M, Benedetti MG, Leo T (2006) Identification and characterisation of heel
strike transient. Gait Posture 24:77–84
Veves A, Murray HJ, Young MJ, Boulton AJM (1992) The risk of foot ulceration in diabetic
patients with high foot pressure: a prospective study. Diabetologia 35:660–663
Waaijman R, Bus SA (2012) The interdependency of peak pressure and pressure–time integral in
pressure studies on diabetic footwear: no need to report both parameters. Gait Posture 35:1–5
Waldecker U (2012) Pedographic classification and ulcer detection in the diabetic foot. Foot Ankle
Surg 18:42–49
Warren GL, Maher RM, Higbie EJ (2004) Temporal patterns of plantar pressures and lower-leg
muscle activity during walking: effect of speed. Gait Posture 19:91–100
Wearing SC, Urry S, Smeathers JE, Battistutta D (1999) A comparison of gait initiation and
termination methods for obtaining plantar foot pressures. Gait Posture 10:255–263
Wood WA, Wood MA, Werter SA, Menn JJ, Hamilton SA, Jacoby R, Dellon AL (2005) Testing for
loss of protective sensation in patients with foot ulceration: a cross-sectional study. J Am Podiatr
Med Assoc 95:469–474
Woodburn J, Helliwell PS (1996) Observations on the F-Scan in-shoe pressure measuring system.
Clin Biomech (Bristol, Avon) 11:301–304
Wrobel JS, Najafi B (2010) Diabetic foot biomechanics and gait dysfunction. J Diabetes Sci
Technol 4:833–845
Yavuz M (2014) Plantar shear stress distributions in diabetic patients with and without neuropathy.
Clin Biomech (Bristol, Avon) 29:223–229
Yavuz M, Botek G, Davis BL (2007a) Plantar shear stress distributions: comparing actual and
predicted frictional forces at the foot-ground interface. J Biomech 40:3045–3049
Yavuz M, Erdemir A, Botek G, Hirschman GB, Bardsley L, Davis BL (2007b) Peak plantar
pressure and shear locations: relevance to diabetic patients. Diabetes Care 30:2643–2645
Yavuz M, Tajaddini A, Botek G, Davis BL (2008) Temporal characteristics of plantar shear
distribution: relevance to diabetic patients. J Biomech 41:556–559
Yavuz M, Ocak H, Hetherington VJ, Davis BL (2009) Prediction of plantar shear stress distribution
by artificial intelligence methods. J Biomech Eng 131:091007
Yuk San Tsung B, Zhang M, Fuk Tat Mak A, Wan Nar Wong M (2004) Effectiveness of insoles on
plantar pressure redistribution. J Rehabil Res Dev 41:4671–4674
Zammit GV, Menz HB, Munteanu SE (2010) Reliability of the TekScan MatScan(R) system for the
measurement of plantar forces and pressures during barefoot level walking in healthy adults. J
Foot Ankle Res 3:11
Zhao J, Guo Y, Wang L (2013) An insole plantar pressure measurement system based on 3D forces
piezoelectric sensor. Sens Transducers 160:49–54
Assessing the Impact of Aerobic Fitness
on Gait
Annet Dallmeijer, Astrid Balemans, and Eline Bolster
Abstract
Children with cerebral palsy often develop walking problems, like reduced
walking distance, or early fatigue during walking. These problems are primarily
caused by motor impairments that lead to gait deviations and increased energy
demands of walking which reduce activities in daily life. As a consequence,
physical inactivity and low fitness are frequent in this population. The combina-
tion of an increased energy demand of walking and reduced aerobic fitness brings
about high levels of physical strain of walking. Maintaining adequate (aerobic)
fitness levels by physical training is therefore important in children with high
energy demands of walking as this keeps up their metabolic reserve and reduces
fatigue-related walking problems and inactivity. Exercise testing is applied to
measure the energy demands of walking and aerobic fitness, guiding whether
treatment should focus on reducing energy cost, increasing fitness, or both.
Testing is especially indicated when walking problems and physical activity-
related fatigue are reported. To improve aerobic fitness in deconditioned children,
training of sufficient frequency, intensity and duration is required, preferably
combined with specific functional exercises.
A. Dallmeijer (*) • E. Bolster

Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, EMGO+ Institute
for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands
e-mail: a.dallmeijer@vumc.nl; e.bolster@vumc.nl
A. Balemans
Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, EMGO+ Institute
for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands
Brain Center Rudolf Magnus and Center of Excellence for Rehabilitation Medicine University
Medical Center, Utrecht, The Netherlands
De Hoogstraat Rehabilitation, Utrecht, The Netherlands
e-mail: a.balemans@vumc.nl

DOI 10.1007/978-3-319-30808-1_40-1
2 A. Dallmeijer et al.
Keywords
Aerobic fitness • Cerebral palsy • Energy consumption • Gait deviations • Phys-
ical strain • Training
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Walking Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Energy Demands of Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Energy Cost in Children with CP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Aerobic Fitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Aerobic Fitness Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Aerobic Fitness and Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Physical Strain and Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Physical Strain of Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Training Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Aerobic Training Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Practical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Specificity of Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Introduction
Children with cerebral palsy (CP) have motor impairments that often lead to gait
deviations and increased energy demands of walking. These children often develop
walking related problems, like reduced walking distance and fatigue during walking,
leading to inactivity and low aerobic fitness levels. The combination of an increased
energy demand of walking and reduced aerobic fitness evokes high levels of physical
strain of walking. This chapter describes how the physical strain and its underlying
components, the energy cost and aerobic fitness, can be measured and how this
information can be used to reduce walking problems in children with CP.
State of the Art
Cerebral Palsy
Cerebral palsy (CP) is described as a group of conditions with permanent disorders

of the development of movement and posture, causing activity limitation. The
condition is caused by a nonprogressive injury of the developing fetal or infant
brain and is the most common cause of physical disability in childhood with a
prevalence around 2–3 cases per 1000 live births. Motor disorders are the most
important clinical symptoms, including disturbances of motor control, abnormal
muscle tone, hyperreflexia, and muscle weakness. These lead to gait deviations
and limitations at the level of activity and participation as described in the
Assessing the Impact of Aerobic Fitness on Gait 3
International Classification of Function, Disability and Health (ICF). Despite the

nonprogressive nature of the primary brain injury, secondary impairments, like joint
contractures and bony deformities, may develop during growth, which can further
limit activities and societal participation across the lifespan. The condition is highly
variable in terms of severity and is often accompanied by sensory, behavioral, and
cognitive disturbances (Graham et al. 2016).
Walking Limitations
Limitations of walking ability and overall mobility are a predominant feature that is
highly variable among children with CP. To describe the impact of this heteroge-
neous condition on mobility and to facilitate communication among health pro-
fessionals, the Gross Motor Function Classification System (GMFCS) was
developed, a five-level functional classification system that classifies gross motor
ability in children with CP. Distinction between categories is based on self-initiated
movement, use of assistive walking devices, use of wheeled mobility, and to a lesser
extent, the quality of movement. Children in GMFCS level I can walk indoors and
outdoors and climb stairs without restriction but have limitations in more advanced
motor skills like running, GMFCS level II includes children who can walk indoors
and outdoors without assistive devices but experience limitations walking outdoors
on uneven surface and stairs, children in GMFCS level III walk indoors and outdoors
with an assistive mobility device, those in GMFCS level IV walk short distances
with a device but rely more on wheeled mobility in daily life, and children classified
as GMFCS level V have no means of independent mobility (Palisano et al. 1997).
Despite the nonprogressive nature of the primary brain injury, secondary impair-
ments, like joint contractures and bony deformities, may develop during growth,
which can further deteriorate gait deviations and limit walking ability. An important
aim of treatment during childhood is therefore to maintain walking ability and
optimize overall mobility. Walking problems are among the main reasons to seek
medical advice in ambulant children with CP. Problems that are frequently seen in
clinical practice are reduced walking distance, decreased walking speed, and the
experience of fatigue during walking, sports, or other daily life activities. It is
important to address these complaints in childhood, as it is known that walking
ability deteriorates in adults with CP starting in early adulthood (Opheim et al.
2009). Long-term aim of treatment is therefore to prevent early deterioration of
walking ability in adulthood.
Energy Demands of Walking
The walking limitations in children with CP often originate from a pathological gait
pattern. The motor disorders can lead to a variety of gait deviations. The most typical
gait deviations are crouch gait, where children walk with increased knee flexion
during stance phase and equines gait where children walk with increased plantar
flexion without heel contact, stiff knee gait, and excessive hip flexion with endo
rotation/adduction movement (Graham et al. 2016). Although the etiology is not

always well understood, these gait deviations generally lead to increased energy
demands of walking and thus a low walking economy, in comparison to typically
developing children (Bolster et al. 2017).
The energy demands of walking can be determined by measuring oxygen uptake
(VO2, ml/min), carbon dioxide output (VCO2, ml/min), and ventilation (l/min) with a
mobile metabolic system during walking at comfortable walking speed. As the oxida-
tive system needs time to start the oxygen delivery to the muscles and produce energy
(ATP), children have to walk for at least 3 min before the actual energy demands of
walking can be measured. After 2–3 min, there will be equilibrium between the oxygen
demands and oxygen supply, which is called “steady state.” The mean oxygen uptake
and carbon dioxide output over 2-min steady-state exercise can be used to calculate the
energy demands. The energy consumption, expressed in Joules, can be calculated by
converting the oxygen uptake (ml/min) to Joules (1 liter VO2 21 KJ, depending on
the respiratory exchange ratio: VCO2/VO2). The energy consumption of walking can
be expressed per meter, which is called the energy cost (J/kg/m).
Energy Cost in Children with CP
Energy cost of walking is increased in children with CP who have deviated gait
patterns. Values increase with higher GMFCS levels (i.e., more severe motor
involvement), with mean values of around 120% of age-predicted values for children
in GMFCS level I up to 220% of age-predicted values in children in GMFCS level
III. Values for individual children with GMFCS level III can be up to three times
higher than typically developing controls (Bolster et al. 2017) (Fig. 1). For the net
energy cost, which is calculated by subtracting the resting energy consumption from
the gross energy consumption, similar increases are found in children with CP,
ranging from 130% to 250% of age-predicted values (Bolster et al. 2017).
Gross energy cost during walking (J/kg/m)
Fig. 1 Energy cost of

walking at comfortable
walking speed for typically 20
developing children (TD,
n = 63) and children with CP
in GMFCS level I (n = 48), II 15
(n = 56), and III (n = 24).
Boxplots represent median
and interquartile ranges
10
0
TD I II III
GMFCS level
In typically developing children, it has been shown that gross energy cost declines
from around 5–6 J/kg/m at age 6 to 3–4 J/kg/m at age 18 (Bolster et al. 2017). This
decline is attributed to a (relative) decrease in resting energy consumption during
growth and has been shown to depend on differences in body height and size,
especially body surface area (BSA) (Kamp et al. 2014). This is important to note
as children with CP are generally smaller and show larger variability in body size.
The dependency of the gross energy cost on body size limits its use for monitoring
changes in walking economy with growth or for comparing values among children
of different ages and body sizes. For monitoring walking economy during growth,
the net energy cost is a more appropriate outcome. The net energy cost is less
dependent on age and height and is therefore assumed to give a better indication
of the walking economy during growth (Schwartz et al. 2006). Still, there is a small
dependency of net energy cost on age with a decline of around 0.073 J/kg/m per year,
which needs to be taken into account when comparing net energy cost values over
time (Bolster et al. 2017). The disadvantage of using the net energy cost is that the
measurement error is larger, because a second assessment (i.e., resting metabolic
rate) is required and thus reliability is lower (Brehm et al. 2007). Another normal-
ization procedure is described by Schwartz et al. (2006), which uses leg length to
adjust walking speed and energy consumption (J/kg/min) for growth, resulting in
nondimensional variables. The net nondimensional energy cost shows a similar
decline with age compared to the net energy cost and can therefore be used
interchangeably with the net energy cost (Bolster et al. 2017).
Common gait deviations such as a walking pattern with flexed knees (crouch gait)
and equines gait (plantar flexion ankles) are associated with increased energy cost of
walking and thus low walking economy in children with CP. Increased co-
contraction of agonist and antagonist muscles accounted for up to 50% of the
variability in oxygen uptake during walking in children with CP (Unnithan et al.
1996). Several treatments are aimed at improving the gait pattern, preventing further
deterioration of gait and the development of musculoskeletal deformities. These
treatments are also expected to improve walking economy by reducing the energy
cost. The most common treatments include orthopedic surgery, treatment to reduce
spasticity (e.g., botulinum toxin), or orthotic treatment with ankle foot orthoses
(Graham et al. 2016; Brehm et al. 2008; Scholtes et al. 2006). Evaluation of energy
cost before and after treatments that aim to improve gait and walking economy is
done to provide insight in the individual treatment effects.
Aerobic Fitness
Aerobic Fitness Assessment
A high energy consumption during walking results by definition also in a higher effort
or physical strain of walking. To determine the physical strain of walking, the maximal
(or peak) oxygen uptake (VO2peak) needs to be known as well. The VO2peak or
maximal aerobic capacity is considered as the gold standard for aerobic fitness (i.e.,
cardiorespiratory) and can be assessed with a cardiopulmonary exercise test (CPET).
During this laboratory-based test, VO2, VCO2, ventilation, and heart rate are measured
during standardized test conditions with an incremental exercise protocol.
Exercise testing in children with CP may be complicated by the prevalent motor
disorders such as spasticity, limited range of motion, impaired selective motor
control, and increased levels of coactivation. Nevertheless, adapted test protocols
have been developed for exercise testing in this population, including cycle ergom-
eter and treadmill tests (Verschuren and Balemans 2015).
In walking children with CP (GMFCS I–II), exercise tests are typically performed
on a bicycle ergometer or treadmill in laboratory conditions (Balemans et al. 2013a).
These tests enable a direct assessment of the VO2peak and determination of the
objective physiological criteria for the achievement of maximal exercise. These
criteria are (1) a heart rate above 180 beats per minute, (2) a respiratory exchange
ratio (RER) above 1.00, and (3) subjective signs of exhaustion in order to ensure a
cardiorespiratory maximum. The test is considered as a valid measurement of
VO2peak when at least two out of these three criteria are met. Protocols are
developed to reach maximal effort, which means that the child experiences exhaus-
tion within 8–12 min (Balemans et al. 2013b).
Treadmill testing is most specific for evaluating fitness when walking ability is of
interest. Most children who walk without a walking aid are able to perform a treadmill
test, but sufficient balance control is required, and optimal walking velocity needs to
be determined pretesting. For children with more extensive motor control problems,
i.e., those with GMFCS II or III, cycle ergometry is a more feasible alternative, despite
its lower specificity for walking. Both treadmill test protocols (GMFCS I–II,
Verschuren and Takken 2010) as well as cycle ergometry protocols (GMFCS I–III,
Brehm et al. 2014) showed good reliability in children with CP. However, although
most – but not all – children with GMFCS level III can be tested on a bicycle
ergometer, this does not apply to children with GMFCS IV or V. Due to their gross
motor limitations, these children are often not able to cycle against incremental
resistance. For this group, exercise tests need to be developed that are compliant
with their functional limitations. The recent development of training and exercise tests
on a racerunner (i.e., a tricycle without pedals that can be propelled by stepping
forward, Bolster et al. 2016) is promising for this group with more severe disability.
Another feasible alternative is measuring VO2peak during field-based tests, like
the shuttle run test that was adapted for children with CP (Verschuren et al. 2006). If
the above described criteria for maximal exercise can be measured to assure maximal
effort, these tests, though less standardized, can also be used to determine the
VO2peak (see, e.g., Dallmeijer and Brehm 2011). Children can be tested from age
6 or 7, and they need to be able to follow simple instructions. Exclusion criteria for
CPET in children with CP are similar to typically developing peers (i.e., no contra-
indications for maximal exercise).
Aerobic Fitness and Cerebral Palsy
Aerobic fitness levels in children with CP are lower than in typically developing
children (Balemans et al. 2013b, Verschuren and Takken 2010). When measured on
a bicycle ergometer, values were 15–30% lower in children with CP (GMFCS level
I–III) than in typically developing children. As expected from a reduced motor
control and concomitant lower mechanical efficiency, peak power output was even
more reduced (20–55%) and was much more dependent on GMFCS level than
VO2peak (Balemans et al. 2013b). Treadmill testing showed a decline of 17% for
VO2peak in children with GMFCS I and II compared to typically developing
children (Verschuren and Takken 2010).
Although values tend to be lower in children with higher GMFCS levels (i.e.,
more severe motor involvement), there were no significant differences between
GMFCS levels. The variability within groups was however high, indicating that
some children have adequate fitness levels, while others are highly deconditioned.
The proportion of children with CP with VO2peak levels below the tenth percentile
of age and gender-based predicted values for typically developing children was high,
ranging from around 45% in children with GMFCS level I to 65% in GMFCS II and
85% for those in GMFCS III or IV (unpublished observations). All these children
were seeking medical advice because of walking problems or fatigue complaints
during activities of daily living. This large proportion of children with low fitness
levels and the large variability within the GMFCS levels emphasize the need to
monitor aerobic fitness in this population, especially when fatigue-related walking
problems are present.
Inactivity most likely explains the low fitness levels of children with CP. It has
been shown that children with CP are less active in physical activities than their
typically developing peers and that activity levels decrease with larger motor
involvement (Van Wely et al. 2014). Walking children with CP spend only 4–10%
of their waking hours in moderate to vigorous activities, compared to 29% in
typically developing children. Also sedentary time, which is associated with an
increased risk for health problems in the general population, is as high as 73–82%
in children with CP, compared to 39% in typically developing children. Lower levels
of physical activity are related to low aerobic fitness levels (i.e., VO2peak) in
children with a bilateral CP (Maltais et al. 2005; Balemans et al. 2015a).
Physical Strain and Training
Physical Strain of Walking
The combination of an increased energy cost of walking and a decreased aerobic

fitness in children with CP leads to high levels of physical strain during walking and
other daily activities (Dallmeijer and Brehm 2011). The physical strain of walking
can be calculated by expressing the oxygen uptake during walking (in ml/kg/min) as
a percentage of the maximal oxygen uptake (in ml/kg/min). In walking children with
CP, the average physical strain for GMFCS levels I, II, and III has shown to be as
high as 55%, 62%, and 78% of VO2peak, respectively, compared to 40% in typically
developing children (Balemans et al. 2015b). Again, large variation within GMFCS
groups (Fig. 2) suggests that this problem differs among individual children and that
exercise testing is required to establish actual level of physical strain.
Fig. 2 Physical strain (in %
Physical strain during walking (%)

VO2peak) for typically 100
developing children (TD,
n = 20) and children with CP
in GMFCS levels I (n = 13), II 80
(n = 17), and III (n = 7).
Boxplots represent median
60
and interquartile ranges
40
20
0
TD I II III
GMFCS level
From the above, it is apparent that a high physical strain may arise from either an
increased energy cost of walking, a decreased aerobic fitness level, or both. When an
increased energy cost is the main factor causing high levels of physical strain,
treatment should focus on improving the walking economy. Most common walking
economy treatments are spasticity treatment (e.g., botulinum toxin treatment), treat-
ment with orthotics (e.g., ankle foot orthosis), or orthopedic interventions (Graham
et al. 2016). When however a reduced fitness level is causing the high physical
strain, an individualized physical training program is indicated. Distinguishing the
cause of an increased physical strain by exercise testing is therefore relevant for
appropriate treatment management. Recent data showed that less than 20% of the
children with CP who have walking problems had no deviations in either energy cost
or VO2peak, while 25% of the children had an increased energy cost in combination
with a normal VO2peak and also 25% had a decreased VO2peak in combination with
a normal energy cost. Around one third of the children with CP had both an
increased energy cost and decreased VO2peak (Balemans et al. 2015b). Walking
speed is often reduced, especially for those with increased energy cost. As a
consequence of the lower walking speed, energy consumption, and thus physical
strain, is lower in these children than in children with normal walking speed despite
similar deviations in energy cost and VO2peak.
Clearly, a high physical strain leaves a smaller metabolic reserve (i.e., the
difference between actual oxygen uptake during walking and VO2peak) that may
cause early fatigue during daily life activities, further limiting mobility and decreas-
ing activity (Dallmeijer and Brehm 2011). Although it seems counterintuitive, the
higher levels of physical strain do apparently not result in higher fitness levels but, in
contrast, may lead to physical inactivity and a concomitant deterioration in fitness
level. An individually targeted fitness training program may help to cut this vicious
cycle by increasing their fitness levels and thereby reducing the relative effort of their
daily activities.
Training Guidelines
As in the general population, exercise training can be undertaken safely in most

individuals with CP and is increasingly propagated in this population (Fowler et al.
2007, Maltais et al. 2014). Training guidelines for improving aerobic fitness are
suggested to be similar to the general population with a frequency of at least 2–4
times per week with a minimum duration of 20 min, at a moderate intensity of about
60–75% maximum heart rate, 40–80% of heart rate reserve, or 50–65% VO2peak
(Verschuren et al. 2016). For younger children, it is advised to incorporate an interval
training scheme rather than long bouts of continuous exercise, as to increase
compliance and to match their daily activity pattern.
Aerobic Training Effects
A number of aerobic training studies have been described for children and adoles-
cents with CP. In general, few studies evaluated training effects on VO2peak. Studies
that applied proper training guidelines regarding frequency, duration, and intensity
of exercise support the effectiveness of aerobic training interventions in children
with CP (see Verschuren et al. 2016). Reported increases in VO2peak varied from
23% for an 8-week intervention in young people with CP (10–16 years, Nsenga et al.
2013) to 18% for a 3-month intervention in adolescents with CP (14–18 years and
GMFCS II and III, Unnithan et al. 2007) and 9% for a 3-month intervention in
adolescents and young adults with CP (16–24 years, Slaman et al. 2015). These
results indicate that exercise training can effectively increase aerobic fitness in
children, adolescents, and young adults with CP, but evidence with regard to age
and disability subgroups needs to be extended.
Practical Implications
Maintaining adequate aerobic fitness levels (i.e., VO2peak) is especially important in

children who are exposed to high energy demands of walking. The increased physical
strain that results from the higher energy consumption of walking is even higher in
those with low VO2peak levels, which may lead to an exacerbation of the vicious cycle
of inactivity and decreased fitness. In children with normal energy cost of walking but
low VO2peak, physical strain will also be higher than normal, up to 40–50% of
VO2peak, but not as high as the 60–80% of VO2peak, as found in children with
both increased energy cost of walking and decreased VO2peak. The latter intensity
level equals vigorous exercise which is likely to affect daily activities. Figure 2 shows
that for the majority of children with GMFCS level II or higher, the effort of walking is
increased to such an extent that the physical strain is above 60%. For these children, it
is essential to lower the physical strain by either increasing the VO2peak or decreasing
the energy cost. The benefits of improving the VO2peak are therefore larger, and are
more relevant, in those with a high physical strain and highly increased energy cost of
walking. Generally, an increase of 20% of the VO2peak would result in a decrease of

the physical strain of around 10–15% (depending on the oxygen uptake of walking).
Obviously, next to improving the VO2peak, reducing the energy cost of walking is also
essential in order to reduce the physical strain of walking.
Specificity of Training
As the main goal of treatment is aimed at improving activities of daily living, it is

advised that aerobic training is incorporated in functional exercises that comply with
movement behavior in daily life. This especially applies to training in younger
children, in order to stimulate their motor development. Training with functional
exercises like walking, running, stepping, and stair climbing have been shown to
improve functional outcomes measured with the shuttle run test in children and
adolescents with GMFCS I–II by 41% (7–20 years, Verschuren et al. 2007).
Although it is not known to what extent these training effects can be attributed to
fitness improvements (i.e., increased VO2peak) or improvements in motor skills, it is
likely that both components contributed to a better performance on the shuttle run
test. These positive effects are with no doubt relevant for daily life activities.
Conclusion
Motor impairments of children with CP lead to gait deviations and increased energy
demands of walking. As a consequence, walking children with CP often develop
walking-related problems, causing inactivity and low aerobic fitness levels. The
combination of an increased energy cost of walking and reduced aerobic fitness
brings about high levels of physical strain during walking. Apart from reducing the
energy cost of walking, maintaining or improving adequate aerobic fitness levels is
of utmost importance in children who walk with increased energy cost levels as this
increases their metabolic reserve and may reduce fatigue-related walking problems
and inactivity. Exercise testing to identify children with high energy cost and/or low
maximal aerobic fitness is therefore indicated when walking problems or physical-
activity related fatigue is present. To improve aerobic fitness in deconditioned
children, training of sufficient frequency, intensity, and duration is required, prefer-
ably combined with functional exercises. Treatment of walking problems in children
with CP should not only focus on improving the walking pattern, but individualized
training interventions are indicated when aerobic fitness levels are low.
References
Balemans AC, Fragala-Pinkham MA, Lennon N, Thorpe D, Boyd RN, O'Neil ME, Bjornson K,
Becher JG, Dallmeijer AJ (2013a) Systematic review of the clinimetric properties of laboratory-
and field-based aerobic and anaerobic fitness measures in children with cerebral palsy. Arch
Phys Med Rehabil 94:287–301
Balemans AC, Van Wely L, de Heer SJ, Van den Brink J, de Koning JJ, Becher JG, Dallmeijer AJ
(2013b) Maximal aerobic and anaerobic exercise responses in children with cerebral palsy. Med
Sci Sports Exerc 45:561–568
Balemans AC, Van Wely L, Becher JG, Dallmeijer AJ (2015a) Longitudinal relationship among
physical fitness, walking-related physical activity, and fatigue in children with cerebral palsy.
Phys Ther 95:996–1005
Balemans AC, Bolster E, Bakels J, Blauw R, Becher JG, Dallmeijer AJ (2015b) Physical strain of
walking in cerebral palsy. Dev Med Child Neurol Suppl 57:65–66
Bolster EA, Balemans AC, Brehm MA, Buizer AI, Dallmeijer AJ (2017) Energy cost during walking
in association with age and body height in youth with cerebral palsy. Gait Posture 54:119–126
Bolster E, Dallmeijer AJ, de Wolf GS, Versteegt M, Schie PE (2016) Reliability and construct
validity of the 6-minute racerunner test in children and youth with cerebral palsy. GMFCS
Levels III and IV Phys Occup Ther Pediatr 17:1–12
Brehm MA, Balemans AC, Becher JG, Dallmeijer AJ (2014) Reliability of a progressive maximal
cycle ergometer test to assess peak oxygen uptake in children with mild to moderate cerebral
palsy. Phys Ther 94:121–128
Brehm MA, Becher J, Harlaar J (2007) Reproducibility evaluation of gross and net walking
efficiency in children with cerebral palsy. Dev Med Child Neurol 49:45–48
Brehm MA, Harlaar J, Schwartz M (2008) Effect of ankle-foot orthoses on walking efficiency and
gait in children with cerebral palsy. J Rehabil Med 40:529–534
Dallmeijer AJ, Brehm MA (2011) Physical strain of comfortable walking in children with mild
cerebral palsy. Disabil Rehabil 33:1351–1357
Fowler EG, Kolobe TH, Damiano DL, Thorpe DE, Morgan DW, Brunstrom JE, Coster WJ,
Henderson RC, Pitetti KH, Rimmer JH, Rose J, Stevenson RD (2007) Promotion of physical
fitness and prevention of secondary conditions for children with cerebral palsy: section on
pediatrics research summit proceedings. Phys Ther 87:1495–1510
Graham HK, Rosenbaum P, Paneth N, Dan B, Lin JP, Damiano DL, Becher JG, Gaebler-Spira D,
Colver A, Reddihough DS, Crompton KE, Lieber RL (2016) Cerebral palsy. Nat Rev Dis
Primers 2:15082
Kamp FA, Lennon N, Holmes L, Dallmeijer AJ, Henley J, Miller F (2014) Energy cost of walking
in children with spastic cerebral palsy: relationship with age, body composition and mobility
capacity. Gait Posture 40:209–214
Maltais DB, Pierrynowski MR, Galea VA, Bar-Or O (2005) Physical activity level is associated
with the O2 cost of walking in cerebral palsy. Med Sci Sports Exerc 37:347–353
Maltais DB, Wiart L, Fowler E, Verschuren O, Damiano DL (2014) Health-related physical fitness
for children with cerebral palsy. J Child Neurol 29:1091–1100
Nsenga AL, Shephard RJ, Ahmaidi S (2013) Aerobic training in children with cerebral palsy. Int J
Sports Med 34:533–537
Opheim A, Jahnsen R, Olsson E, Stanghelle JK (2009) Walking function, pain, and fatigue in adults
with cerebral palsy: a 7-year follow-up study. Dev Med Child Neurol 51:381–388
Palisano R, Rosenbaum P, Walter S, Russell D, Wood E, Galuppi B (1997) Development and
reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med
Child Neurol 39:214–223
Scholtes VA, Dallmeijer AJ, Knol DL, Speth LA, Maathuis CG, Jongerius PH, Becher JG (2006)
The combined effect of lower limb multilevel botilinun toxin type a and comprehensive
rehabilitation on mobility in children with cerebral palsy: a randomized clinical trial. Arch
Schwartz MH, Koop SE, Bourke JL, Baker R (2006) A nondimensional normalization scheme for
oxygen utilization data. Gait Posture 24:14–22
Slaman J, Roebroeck M, Dallmeijer A, Twisk J, Stam H, van den Berg-Emons R (2015) Can a
lifestyle intervention programme improve physical behaviour among adolescents and young
adults with spastic cerebral palsy? A randomized controlled trial. Dev Med Child Neurol
57:159–166
Unnithan VB, Dowling JJ, Frost G, Bar-Or O (1996) Role of cocontraction in the O2 cost of
walking in children with cerebral palsy. Med Sci Sports Exerc 28:1498–1504
Unnithan VB, Katsimanis G, Evangelinou C, Kosmas C, Kandrali I, Kellis E (2007) Effect of
strength and aerobic training in children with cerebral palsy. Med Sci Sports Exerc
39:1902–1909
Van Wely L, Dallmeijer AJ, Balemans AC, Zhou C, Becher JG, Bjornson KF (2014) Walking
activity of children with cerebral palsy and children developing typically: a comparison between
the Netherlands and the United States. Disabil Rehabil 36:2136–2142
Verschuren O, Takken T, Ketelaar M, Gorter JW, Helders PJ (2006) Reliability and validity of data
for 2 newly developed shuttle run tests in children with cerebral palsy. Phys Ther 86(8):
1107–1117
Verschuren O, Ketelaar M, Gorter JW, Helders PJ, Uiterwaal CS, Takken T (2007) Exercise training
program in children and adolescents with cerebral palsy: a randomized controlled trial. Arch
Pediatr Adolesc Med 161:1075–1081
Verschuren O, Takken T (2010) Aerobic capacity in children and adolescents with cerebral palsy.
Res Dev Disabil 31(6):1352–1357
Verschuren O, Balemans AC (2015) Update of the core set of exercise tests for children and
adolescents with cerebral palsy. Pediatr Phys Ther 27:187–189
Verschuren O, Peterson MD, Balemans ACJ, Hurvitz EA (2016) Exercise and physical activity
recommendations for people with cerebral palsy. Dev Med Child Neurol 58(8):798–808
Oxygen Consumption in Cerebral Palsy
Hank White, J. J. Wallace, and Sam Augsburger
Abstract
Oxygen consumption is a measure of aerobic fitness and the body’s ability to
deliver oxygen for energy generation during exercise. Oxygen consumption can
be measured directly or indirectly. To measure oxygen consumption, breath by
breath measurements of the oxygen inspired and the carbon dioxide exhaled are
measured by specialized equipment. For other indirect measures, heart rate or
distance walked/ran during exercise are used in regression equations to estimate
oxygen consumption. Examples of such measures of oxygen consumption and
energy expenditure are shuttle runs/rides, stair climbing tests, 6 min walk tests,
1 min walking tests, and mechanical energy estimation. Regardless of age
children, adolescents and adults diagnosed with cerebral palsy have decreased
physical activity and increased energy, oxygen cost and oxygen consumption
(measured with direct and indirect methods) when walking compared to able-
bodied persons. Decreases in physical activity may increase the risk of cardio-
vascular and cardiopulmonary compromise in children and adults diagnosed with
cerebral palsy and these impairments may contribute to further decreases in
physical activities. However, surgical interventions (single event multi-level
surgeries, and Rhizotomy) and therapy have been reported to increase walking
distances and decrease energy expended when walking for persons diagnosed
with cerebral palsy.
Keywords
Cerebral palsy • Children • Oxygen consumption • Oxygen cost • Energy
expenditure • Walking
H. White (*) • J.J. Wallace • S. Augsburger

Motion Analysis Center, Shriners Hospitals for Children Medical Center, Lexington, KY, USA
e-mail: hwhite@shrinenet.org; jwallace@shrinenet.org; saugsburger@shrinenet.org

DOI 10.1007/978-3-319-30808-1_41-1
2 H. White et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Exercise Testing and Children . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Exercise Testing and Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Equipment for Exercise Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Measures of Oxygen Consumption (VO2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Direct Measures of VO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Indirect Measures of VO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Relationship between Mechanical Energy and VO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Interventions That Affect VO2 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Therapy Effects on VO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Introduction
The international classification of functioning, disability and health (ICF) is a

classification system that provides a standard language and framework for the
description of health and health related states (Vargus-Adams and Majnemer
2014). Different domains describe changes in body function and structure, as well
as changes in activity and participation. The ICF differentiates what a person can do
in a standard environment (their level of capacity) versus what a person actually does
in their usual environment (their level of performance). Walking, running, and other
gross motor skills are considered measures of capacity while playing, working and
activities of daily living are considered levels of performance. The ICF can help
clinicians and researchers understand the relationship between a person’s capacity
and their level of performance. There is a cyclical nature between impairments in
walking abilities, mechanical energy and oxygen cost when walking and perfor-
mance that may affect participation in the home and community. For example, if a
person’s impairment worsens it may decrease his/her participation and if someone’s
participation decreases then a progression of impairments may occur. The goal of
this chapter to give clinicians and researchers the knowledge needed to understand
how walking efficiency is measured and to provide an overview of oxygen con-
sumption, oxygen cost and mechanical energy when walking for children with
cerebral palsy.
Cerebral Palsy
Cerebral Palsy (CP) is a clinical syndrome characterized by some type of insult to the
brain during development, birth, or in the first 2 years of life (Scholtes et al. 2006).
The incidence of Cerebral Palsy (CP) is approximately 2–3 children per 1,000 births
Oxygen Consumption in Cerebral Palsy 3
(Nene et al. 1993). The clinical presentation of CP includes a broad spectrum of

impairments of the neuromusculoskeletal system (Scholtes et al. 2006). Among
these impairments are decreases in motor control and increases in muscle spasticity
and muscle/joint contractures. These impairments have been proposed to cause an
increase in energy cost when walking (Augsburger and Tylkowski 2000; Norman
et al. 2004; Stallings et al. 1996; van den Hecke et al. 2007). This can result in
activity limitations such as a decreased ability to walk and perform transfers safely
(Kay et al. 2004). The end result can be participation restrictions in the home, school
and community.
CP is often classified based on topographical presentation. A child diagnosed
with hemiplegia presents as a person who has sustained a cerebral vascular accident
with involvement of the ipsilateral arm and leg. CP spastic diplegia presentation is
involvement of bilateral lower extremities with or without upper extremity involve-
ment. Hemiplegia and diplegia are the most common presentations of CP. CP spastic
quadriplegia presents with involvement of all four extremities, and trunk with or
without neck involvement (Morris 2007). Other less common presentations of CP
include hypotonia, ataxia, and mixed (spasticity with hypotonia, dystonia and/or
ataxia) (Morris 2007).
A better classification of CP is the Gross Motor Function Classification System
(GMFCS). The GMFCS is a valid and reliable five level classification system for
children diagnosed with CP based on self-initiated movements (sitting, standing,
walking) (Palisano et al. 2000; Wood and Rosenbaum 2000). The criteria for
classification are age based. In general for children 6 years or older GMFCS Level
I children demonstrate minimal impairments and mild gait and gross motor abnor-
malities. Level II will demonstrate more impairments of gait and gross motor
abilities, however, can walk without an assistive device on level and uneven
surfaces. Children classified as Level III require use of an assistive device (walker
or crutches) to ambulate. Children classified as level IV may be able to take steps or
perform standing transfers, however have decreased trunk control and cannot sit
without external supports. The highest level of impairments is present in children
classified as Level V who demonstrate few self-initiated movements.
Exercise Testing and Children
Cardiorespiratory health can be assessed by measuring the oxygen uptake during

either a maximal or submaximal exercise test (ACSM’s Guidelines for Exercise
Testing and Prescription/American College of Sports Medicine. Philadelphia:
Wolters Kluwer Health/ Lippincott Williams and Wilkins 2014). Measuring oxygen
consumption (VO2) during exercise is considered the gold standard for assessing
energy expenditure and exercise capacity (Rose et al. 1990). VO2 is also commonly
used to assess changes in walking efficiency after surgery, or rehabilitation in adults
and children with disabilities (Bar-Haim et al. 2004; Chan et al. 2008).
Oxygen consumption testing can be conducted on a treadmill, cycle ergometer or
on level ground in children as young as 3–4 years of age (ACSM’s Exercise
4 H. White et al.
Management for Persons with Chronic Diseases and Disabilities/American College

of Sports Medicine. Champaign and IL: Human 2016). Tests can be either maximal
or submaximal. Exercise testing is typically a graded test in which the work load is
incrementally increased until a plateau in oxygen uptake (L/min) is achieved regard-
less of the increasing load. Maximal exercise tests are used to elicit the highest level
of participation and can be quite demanding for subjects to complete. Alternatively,
submaximal tests are designed to be easier to complete and utilize regression
equations to estimate maximal exercise levels. The mode of exercise testing, run-
ning, walking and cycling, all produce different VO2 results. In able-bodied children
treadmill tests produce a 7%–10% higher VO2 than cycling tests, and running on a
treadmill will produce a 6%–10% higher VO2 maximum (VO2max) result compared
to walking on a treadmill (ACSM’s Exercise Management for Persons with Chronic
Diseases and Disabilities/American College of Sports Medicine. Champaign & IL:
Human 2016).
Exercise Testing and Cerebral Palsy
Assessment
The American College of Sports medicine (ACSM), has set specific guidelines for
VO2 assessment in children (ACSM’s Guidelines for Exercise Testing and Prescrip-
tion/American College of Sports Medicine. Philadelphia: Wolters Kluwer Health/
Lippincott Williams and Wilkins 2014). However, for children diagnosed with CP,
ACSM guidelines are not always appropriate (ACSM’s Exercise Management for
Persons with Chronic Diseases and Disabilities/American College of Sports Medi-
cine. Champaign & IL: Human 2016). Clinical exercise testing protocols like the
Bruce or Balke, require specialized equipment, are too intense and increase the work
load too steeply for children with CP (Nsenga Leunkeu et al. 2012). Cycle ergometer
tests are also not recommended for children with CP; due to the decreased coordi-
nation and spasticity that make reciprocal pedaling and maintaining pedal cadence
difficult (Nsenga Leunkeu et al. 2012). Testing on a treadmill can be performed in
less involved children with CP, with access to a harness system that can ensure
patient safety (Potter and Unnithan 2005). Therefore, for children with CP exercise
testing is typically performed on level ground (Potter and Unnithan 2005). Children
with GMFCS Level I may be able to follow the ACSM guidelines for testing;
however Levels II and III may need adaptive measures (ACSM’s Exercise Manage-
ment for Persons with Chronic Diseases and Disabilities/American College of Sports
Medicine. Champaign & IL: Human 2016). For children with CP Levels IV and V,
functional mobility testing may be more appropriate than exercise testing (ACSM’s
Exercise Management for Persons with Chronic Diseases and Disabilities/American
College of Sports Medicine. Champaign & IL: Human 2016). Previous literature
assessing VO2 reported that children with CP fatigue more quickly (Potter and
Unnithan 2005), do not tolerate exercise as well as peers (Rodda et al. 2006), are
less physically active (Gorter et al. 2009), have a lower VO2max during cycling and
treadmill exercise (Potter and Unnithan 2005; Verschuren et al. 2010), have a two to
three times increase in O2 cost when exercising (Bar-Haim et al. 2004; Potter and
Unnithan 2005; Bowen et al. 1998a; Thomas et al. 2004) and have a lower mechan-
ical efficiency at maximum load (Rose et al. 1990; Bar-Haim et al. 2004). These
differences could be due to co-contractions of muscles (Potter and Unnithan 2005;
Thomas et al. 2004), increased muscle tone, muscle spasticity, and bony abnormal-
ities (Thomas et al. 2004).
Equipment for Exercise Testing
Measuring energy expenditure during exercise can be accomplished through direct

or indirect calorimetry. Direct calorimetry is a direct measure of the heat production
while the subject is in an enclosed chamber called a calorimeter (Powers and Howley
2007). This method requires a large and extremely expensive device and is not
always feasible for clinical use (Battley 1995). To measure energy expenditure when
exercising the chamber must be large enough to have a treadmill or bike inside and
the heat produced by the device must be accounted for (Powers and Howley 2007).
Therefore, for clinical and research purposes indirect calorimetry, the measurement
of the oxygen consumed during rest or exercise, is more widely used (Powers and
Howley 2007). Indirect calorimeters are less expensive, require less space, and have
been proven to be just as reliable and valid as direct measures (Powers and Howley
2007; Battley 1995).
The most common type of indirect calorimetry used is open-circuit spirometry
(Powers and Howley 2007). Expired air is sent into a mixing chamber that uses
electronic gas analyzers to measure the amount of oxygen (O2) and carbon-dioxide
(CO2) and the results are processed in a computer. (Powers and Howley 2007).
Open-circuit spirometry can be either a dilution method or a breath-by-breath
analysis. With the dilution method the subject is connected to a hood or mask that
covers the entire face or head, the dilution pump samples air from the space, the data
is sent to a metabolic cart and data is averaged over 15 s intervals (Welch et al. 2015).
The breath by breath method analyzes the O2 and CO2 content of every breath,
which can be portable, is more applicable in real world scenarios outside the
laboratory, and is also more patient friendly and is less cumbersome in clinical
settings (Welch et al. 2015). Both methods have been proven to be reliable and valid,
however results between methods cannot be interchanged and data from different
methods should not be used for pre and post comparisons (Welch et al. 2015).
Another method that estimates energy expenditure is the use of the joint kine-
matics and kinetics to estimate the mechanical energy generated and absorbed by the
body when walking or running. The estimation of joint kinematics and kinetics
requires the use of force plates and a three-dimensional motion capture systems, both
of which are expensive and require large spaces. It is thought that the use of gait
mechanics to measure the energy cost of walking can give researchers and clinicians
a better understanding of factors that contribute to the elevated cost of walking seen
with pathology (Umberger et al. 2013). It is postulated that disruptions in the transfer
6 H. White et al.
of mechanical energy in body segments, as seen in pathologic gait, require compen-

sations by the body that can increase the metabolic cost (Umberger et al. 2013).
Therefore, utilizing mechanical energy estimations can enhance the interpretation of
data obtained from open-circuit spirometry.
Measures of Oxygen Consumption (VO2)
VO2 is a measure of aerobic fitness and the body’s ability to deliver oxygen for
energy generation during exercise (Verschuren et al. 2010). VO2 can be measured
directly or indirectly. To directly measure VO2, breath by breath measurements of
the oxygen inspired and the carbon dioxide exhaled is measured by specialized
equipment (Verschuren et al. 2010). For indirect measures, heart rate or distance
walked/ran during exercise are used in regression equations to estimate VO2.
Examples of indirect measures of oxygen consumption and energy expenditure are
shuttle runs/rides, stair climbing tests, 6 min walk tests (6-mwt), 1 min walking tests
(1-mwt), and mechanical energy estimation.
Direct Measures of VO2
Breath by Breath Oxygen Consumption (VO2) and CP

VO2 measurements when walking for children with CP (GMFCS levels I–III)
compared to that of a control group of typically developing children is presented
in Table 1. Walking speed was more variable for the control group (50–121.7 m/min)
compared to that of the children with CP (41.3–55.9 m/min) (Norman et al. 2004;
van den Hecke et al. 2007; Rose et al. 1989; Unnithan et al. 1996). The values for
VO2 in children with CP when walking ranged from 14.6 to 23.4 ml/kg/min (average
18.58 ml/kg/min), which is higher than that of the typically developing group
6.3–25.1 (average 13.65 ml/kg/min) (Norman et al. 2004; van den Hecke et al.
2007; Rose et al. 1989; Unnithan et al. 1996). When walking at the same speed
(3 km/h), children with CP use 54% of their VO2max compared to only 23% of
VO2max for typically developing children (Potter and Unnithan 2005). However,
children with CP can improve cardiorespiratory fitness through training. A 9%
increase in VO2max and a 7% increase in walking distance for children with CP
after a 9 week (2 a week) exercise program has been reported (Gorter et al. 2009).
Oxygen Cost and CP

Oxygen cost is the volume of oxygen in milliliters measured per kilogram of body
weight per meter walked and is reported as ml/kg/m (ACSM’s Guidelines for
Exercise Testing and Prescription/American College of Sports Medicine. Philadel-
phia: Wolters Kluwer Health/Lippincott Williams and Wilkins 2014). Oxygen cost is
a measure of the oxygen required to walk one meter and allows for comparison of
people walking at different speeds. Compared to adults, children can have a
40%–70% greater oxygen cost when walking at their maximum speed (DeJaeger
Table 1 Oxygen consumption in children with cerebral palsy and in normal controls
Walking VO2
Subjects # of Age speed (ml/kg/ Resting VO2
Study diagnosis Subjects (years) (m/min) min) (ml/kg/min)
Rose et al. CP 13 11.2 55.9(20) 23.4(7) 4.8(0.9)
(1989) (7–16)
Rose et al. Normal 18 12.5 121.7(14) 25.1(5) 4.9(1)
(1989) control (7–17)
Unnithan et al. CP 9 12.7 50 16.6
(1996) (2.8) (6.5)
Unnithan et al. Normal 8 13.6 50 10.2
(1996) control (2.1) (1.2)
Norman et al. CP 10 12.8 41.3(13.4) 19.7 5.1(3.2)
(2004) (2.9) (11.3)
Norman et al. Normal 15 11.7 66.7(9.6) 6.3(1.9) 2.9(1.1)
(2004) control (2.7)
Van Den Hecke CP 20 8.1 41.67(6.17) 14.6
et al. (2007) (1.6) (3.2)
Van Den Hecke Normal 6 9.9 75.5(4) 13(2)
et al. (2007) control (0.55)
et al. 2001). However, with increasing age oxygen cost has been shown to decrease
for typically developing populations (DeJaeger et al. 2001). In addition, a linear
relationship between the body surface area inverse and oxygen cost exists (Bowen
et al. 1998b). These findings demonstrate that as children grow their oxygen cost
decreases with both age and with increasing body size. When assessing oxygen cost
in children over time or pre/post intervention changes due to growth should be taken
into account.
The oxygen cost of children with CP during walking at different speeds and
across different GMFCS levels is presented in Table 2. In typically developing
children, only a small increase in oxygen cost is seen while walking at a self-
selected speed compared to resting oxygen cost. However, children with CP dem-
onstrate significant increases in oxygen cost when walking at a self-selected speeds
(Norman et al. 2004). Between day variability of VO2 measurements was found to
be similar between typically developing children and children with CP (variability of
13% for children with CP and 14% for typically developing children) (Bowen et al.
1998a).
For children diagnosed with CP, oxygen cost when walking has been reported to
be two to three times greater than able bodied children’s oxygen cost (Thomas et al.
2004). Children with diplegic CP have double the exercise oxygen cost of children
with hemiplegia; 0.56(0.18) ml/kg/min and 0.21(0.01)ml/kg/min respectively (Rose
et al. 1990). At self-selected walking speeds, children with CP have a net oxygen
cost of 0.18 ml/kg/m (Kerr et al. 2008). However, when stratified by GMFCS levels,
8
Table 2 Oxygen cost when walking for children with CP

# of Age Walking speed Exercising O2 cost Exercising NET O2 cost
Study Subjects diagnosis subjects (years) (m/min) (ml/kg/m) (ml/kg/m)
Rose et al. (1990) CP (3 hemiplegics, 13 11.2 46.0(11.4) 0.43(0.22)
10 diplegics) (7–16)
Rose et al. (1990) Hemiplegics 3 0.21(0.01)
Rose et al. (1990) Diplegics 10 0.56(0.18)
Keefer et al. CP 13 11.2 (3) 40.2 0.28(0.21–0.36) 0.15(0.07–0.22)
(2004)
Keefer et al. CP 13 11.2 (3) 53.4 0.25(0.19–0.31) 0.15(0.11–0.22)
(2004)
Keefer et al. CP 13 11.2 (3) 67.2 0.25(0.20–0.29) 0.16(0.12–0.23)
(2004)
Norman et al. CP 10 12.8(2.9) 41.3(13.4) 0.55(38)
(2004)
Kerr et al. (2008) Level I 57 12.06 Self-selected 0.18(0.1) 0.15(0.04)
(0.41)
Kerr et al. (2008) Level I 57 12.05 Self-selected 0.15(0.04)
(.41)
Kerr et al. (2008) Level II 91 9.93 Self-selected 0.17(0.07)
(0.37)
H. White et al.
Kerr et al. (2008) Level III 22 10.27 Self-selected 0.31(0.12)
(.89)
Kerr et al. (2008) Level III 14 11.75 Self-selected 0.36(0.27)
(0.98)
Kerr et al. (2008) CP ALL 184 10.76 Self-selected 0.18(0.1)
(3/61)
Kerr et al. (2008) Hemiplegics 94 10.62 Self-selected 0.14(0.04)
(3.49)
Kerr et al. (2008) Diplegics 84 10.89 Self-selected 0.24(0.12)
(3.70)
Oeffinger et al. Level I 179 10.75(4) SeIf-selected 0.37 (0.11–1.4)
(2004)
Oxygen Consumption in Cerebral Palsy
Oeffinger et al. Level II 134 11.17 Self-selected 0.47(0.06–1.1)

(2004) (4.25)
Oeffinger et al. Level III 106 10.58 Self-selected 0.78 (0.23–2.5)
(2004) (4.17)
9
10 H. White et al.
net oxygen cost rises as GMFCM levels rise; Level I 0.15(0.04)ml/kg/m, Level II
0.17(0.07)ml/kg/m, Level III 0.31(0.12)ml/kg/m and Level IV 0.36(0.27)ml/kg/m.
The minimum clinically important difference (MCID) is the magnitude of change of
a measure that is required for a patient to perceive as beneficial (Cook, 2008). The
MCID is based on the effect size of the measure. The MCID of large (0.8) effect size
for Oxygen cost (ml O2/kg/min) has been reported to be 0.06, 0.17 and 0.09 for
GMFCS levels I, II, and III receptively (Oeffinger et al. 2008).
GMFCS Levels, GMFM Relationships with Oxygen Cost and VO2

Energy cost, oxygen cost and oxygen consumption when walking increased with
increases in GMFCS level (Kamp et al. 2014). These differences are believed to be
due in part to the increased severity of impairments (muscle tone/spasticity,
co-contractures and increased gait abnormalities) (Johnston et al. 2004). Children
with CP (GMFCS level I) demonstrated similar mean oxygen cost to able-bodied
children (0.25 ml/kg/m) (Kamp et al. 2014). However, large inter-individual differ-
ences of energy cost were noted within GMFCS levels (Kamp et al. 2014).
The gross motor functional measure (GMFM) is a standardized objective measure
of gross mobility for children diagnosed with CP (Russell et al. 1993). A higher
score on the GMFM test was associated with lower energy cost when walking
(Kamp et al. 2014). The percent score of section D of GMFM had the strongest
inverse association with energy cost when walking. In 15 young adults (mean age
29 years) diagnosed with CP (GMFCS levels I and II), oxygen cost when walking
had a strong inverse association with GMFM scores (Sections D and E[r = -0.57
and -0.066]) (Maltais et al. 2012). This is because both GMFM and energy cost both
measure impairments of the musculoskeletal system.
One study assessed the relationship between oxygen cost when walking and
activity limitations and participation restrictions for 184 children diagnosed with
CP stratified across GMFCS levels (GMFCS level I = 57; II = 91, III = 22;
IV = 14) (Kerr et al. 2008). Net oxygen cost was measured with the Cosmed
K4b2 device (Cosmed, Rome, Italy) while subjects walked at their self-selected
walking speed. Net oxygen cost data was skewed, therefore the data had to be
transformed to log net oxygen cost. The PEDI questionnaire assesses child’s func-
tional abilities and is similar to PODCI. Moderate correlations were found between
log net oxygen cost and GMFM scores (r = -0.61), and with log net oxygen cost
and the mobility section of Pediatric Evaluation of Disability Inventory (PEDI)
(r = -0.28). The LAQ-CP questionnaire assesses the effect the child’s disability
has on child and family. No relationship was found between log net oxygen cost and
Lifestyle Assessment Questionnaire for Cerebral Palsy (LAQ-CP).
Indirect Measures of VO2
Mechanical Energy
Until the mid-1990s assessing movement and energy expenditure was limited to
cardiopulmonary indices, oxygen consumption, and gait kinematic analyses (Jones
and McLaughlin 1993). These techniques were limited by “their inability to control
work-load with enough precision to allow comparisons over time or to estimate the
relative inefficiency of a disabled individual compared with the able-bodied popu-
lation.” (Jones and McLaughlin 1993) With the onset of full body gait kinematic and
kinetic assessments, this list increased to include calculated mechanical energy
expenditure of the lower extremities during gait.
Mechanical energy expenditure is calculated by summing the integrals of the
rectified power curves of the lower extremities kinetic profiles (Augsburger and
Tylkowski 2000; Van de Walle et al. 2012a). This approximation does not include
the energy expenditure of the upper extremities and the trunk, or energy spent and
lost in the form of heat, but serves as a reasonable approximation of the mechanical
work performed during gait by the subject. It does not include the energy spent by
co-contractures or spastic muscles. It is simply a measure of the actual mechanical
energy output of the lower extremities. It can, however, be compared with the
oxygen consumption to give an estimation of the efficiency of the gait, and is an
indirect measure of co-contractures and/or spastic muscles that do not result in
mechanical output.
Relationship between Mechanical Energy and VO2
For children diagnosed with CP, the magnitude of impairments (spasticity, weakness,
co-contracture of muscles) can be approximated by comparing a patient’s volumetric
oxygen consumption (VO2) and gait mechanical energy (Eg) with able-bodied
subjects; graphically represented by VO2 versus velocity squared(V2) and by com-
paring VO2 with Eg (Augsburger and Tylkowski 2000). A linear relationship
(R2 0.938–0.999) between VO2 and velocity square has previously been reported
(Augsburger and Tylkowski 2000). The slope of this curve is an indication of an
individual’s differences in metabolic rate. In addition, Eg versus V2 also has a linear
relationship (R2 = 0.961) and these two curves can be combined to produce a
graphic representation of VO2 versus Eg.
For able-bodied persons who walk with a simulated crouched gait pattern and
patients diagnosed with CP who naturally walked with a crouched gait pattern (knee
flexion 25 –35 ) both demonstrate an increased energy cost when walking
(Augsburger and Tylkowski 2000). When plotting VO2 versus V2 both groups
demonstrate similar increases in VO2. However, when plotting VO2 versus Eg the
able-bodied subjects demonstrating simulated crouch to the subjects with CP who
naturally walked in a crouched gait pattern differences were noted. Specifically, able-
bodied subjects demonstrated increases in VO2 and Eg. However, subjects with CP
demonstrated increases in VO2 without increasing Eg. Because some subjects with
CP demonstrated increased VO2 and decreased Eg and others demonstrated
increased both VO2 and Eg it was proposed the these differences were not due to
just walking in a crouched posture but increased muscle tone/spasticity and
increased co-contractures of the lower extremity musculature (Augsburger and
Tylkowski 2000).
12 H. White et al.
This increase in energy cost when walking could be due to decreased efficiency of
the muscles or increased total work performed by the muscles. The increased muscle
work could be due to increased step frequency, muscle co-contraction, kinematic
abnormalities, or increased muscle tone/spasticity (Augsburger and Tylkowski 2000;
Norman et al. 2004; Stallings et al. 1996; van den Hecke et al. 2007). The mechan-
ical work and metabolic cost of walking was assessed in 20 children with CP, spastic
hemiplegia and 6 able-bodied children (van den Hecke et al. 2007). Total work was
calculated from external work needed to move the center of mass of the body
(calculated from ground reaction forces) and internal work forces needed to move
the body segments relative to the center of mass based on a summation of the
rotational and translational kinetic energy of each segment. Results of the study
indicated the increased cost was due to increased mechanical work performed by the
muscles and not decrease efficiency (van den Hecke et al. 2007). Therefore children
with CP use a larger proportion of their aerobic capacity resulting in decreased
endurance when walking.
The net joint moments are a summation of the mechanical energy absorbed,
generated and transferred between body segments when walking (Umberger et al.
2013). Disruption of this mechanical energy transfer is believed to be the cause of
increase oxygen cost when walking for children diagnosed with CP and other
disorders (Umberger et al. 2013). A seven segment dynamic model of the whole
body was developed by comparing the center of mass acceleration from force plate
data and joint moment and power data calculated from inverse dynamics. Results
indicated the ankle joint moments generated and transferred the largest magnitude of
energy during late single limb stance and weight release. While the hip joint
moments transfer similar magnitude of energy to the ipsilateral limb, contralateral
limb and the head, arms and trunk (HAT) from 30% to 70% of the gait cycle. The
authors concluded lower extremity extensor moments primarily cause energy flow
into the HAT and flexor moments cause energy flow out of the HAT (Umberger et al.
2013). It is the disruption of the transfer of mechanical energy between the ipsilateral
limb, contralateral limb and the (HAT) segment that results in less efficient gait. The
disruption of transfer of energy could result in increased metabolic cost in walking
because increased energy by the muscle would be required.
This inefficient transfer of energy is also supported by a prospective study of
children with CP, spastic diplegia (GMFCS level I and II) (n = 18) to able-bodied
children (n = 25) (Van de Walle et al. 2012b). Mechanical work calculated from
integration of joint powers for children with CP was found to be 1.5 times higher
than able-bodied children (Van de Walle et al. 2012b). The increase in work was
primarily due to increased movements of the trunk and head and not due to increased
movements of the arms. The integrated power approach first obtains a summation of
positive and negative net joint power for upper extremity (shoulder, elbow, wrist),
lower extremities (ankle, knee, hip) and for the neck. These net joint powers are
summed to provide a net joint work for the entire body (Van de Walle et al. 2012a).
The integrated power approach has been validated against oxygen consumption for
children with CP (Van de Walle et al. 2012a). The total net joint work and net oxygen
cost when walking for children with CP was significantly higher than able-bodied
children and able-bodied adults. The total net joint work, which included calcula-
tions of head trunk and arms and not just the lower extremity, was significantly
correlated with the oxygen cost of walking (Van de Walle et al. 2012a).
What about children that cannot walk the 3–10 min required for obtaining a
steady state of oxygen consumption? What about children who refuse to wear the
equipment required for breath by breath oxygen consumption testing? Additionally,
most clinicians do not have access to directly measure oxygen cost. Therefore, other
indirect measures of energy cost have been reported for children with CP (Keefer
et al. 2004; McDowell et al. 2005). These indirect measures of VO2 can be
performed utilizing variables like heart rate or distance measured during submaximal
exercise (Verschuren et al. 2010).
Shuttle Run/Ride
Heart rate during a shuttle run test can be used to indirectly measure aerobic capacity
in children with CP (GMFCS levels I–II) (Verschuren et al. 2010). This study
provided percentile curves for comparison across different ages and heights
(Verschuren et al. 2010). As age and height increased so did the number of shuttles
performed (Verschuren et al. 2010). Similarly, assessing VO2 indirectly using HR
during a shuttle ride test for 23 children with CP that use manual wheelchairs has
been reported (Verschuren et al. 2013). The shuttle ride test was compared to VO2
directly measured while using an arm ergometer (Verschuren et al. 2013). The shuttle
ride test was reliable (ICC = 0.99) for all VO2 variables. The shuttle ride test also
produced VO2max vales of 26(5) ml/kg/min, which is similar to that measured using
an arm ergometer 25.3(5.7) ml/kg/min.
Stair Climbing
Stair climbing can be used to measure mechanical efficiency in children with CP,
using body weight and step height (kg/m) to measure the work done when climbing
stairs. Typically developing children performed significantly more stair ascents
(34.8[2.7] v. 6.4[7.7]), had approximately 2–3 ml/kg/min more VO2 and had
increased gross mechanical energy (20% vs. 4%) when climbing stairs as com-
pared to children with CP (GMFCS II-IV) (Bar-Haim et al. 2004). Stair climbing
detected changes in post exercise training with children performing on average 1.3
more stair ascents and increasing gross mechanical efficiency by 2%(Bar-Haim
et al. 2004).
6-Minute Walk Test

For children with CP who cannot tolerate treadmill or cycle testing, waking on level
ground during the 6-min walk test (6-mwt) can be used. The 6-mwt is a submaximal,
indirect measure that uses distance walked as a predictor of aerobic capacity (Maher
et al. 2008; Thompson et al. 2008). This test may be preferable for children with CP
(GMFCS I–III) as it reflects activities of daily living and the ability to walk
community distances (Thomas et al. 2004). In addition, a walking test on level
ground is less expensive, simple and safer than traditional measures that require
testing equipment (Nsenga Leunkeu et al. 2012).
14 H. White et al.
The 6-mwt has a reported ICC value of 0.98 overall for children 4–18 years of age
with CP, GMFCS levels I-III (Maher et al. 2008; Thompson et al. 2008). Reported
distances walked stratified by GMFCS level were 486.6(84.4)m for Level I, 312.9
(77)m for Level II and 240.2(121.1)m for Level III. The validity of the 6-mwt, with
simultaneous direct collection of VO2, in children with CP, GMFCS levels I and II,
as compared to a direct measure during cycle ergometery has been reported (Nsenga
Leunkeu et al. 2012). For this study VO2 and peak HR results for the 6-mwt were
33.1(7.1)ml/kg/min 156.6(22.4)bpm and the cycle tests were 32.3(6.3)ml/kg/min)
and 148.4(25.1)bpm. The 6-mwt was found to be valid and reliable for children with
CP when measuring aerobic capacity during the 6-mwt (Nsenga Leunkeu et al.
2012).
One-Minute Walk Test

An alternative to the 6 min walk test is the 1-min walk test which measures the
distance walked in 1 min when a child walks at their maximum speed. The distance
walked in 1 min was found to be inversely related to GMFM scores (section D
r = 0.910, section E r = 0.872) and a measure of walking endurance (McDow-
ell et al. 2005). Additionally, the average distance walked decreased with an increase
in GMFCS levels (Level I = 100[12]m, Level II = 83[17]m, Level III 56[17]m,
Level IV 19[7]m). A large prospective multi-center study also found decrease in
walking distance with increase in GMFCS levels and reported minimum clinically
important differences to be 5.1 – 9.0(m) for GMFCS levels I and II and 3.86.3
(m) for GMFCS levels III (Hassani et al. 2014).
Additionally, the 600-Yard Walk-Run Test is a standardized physical fitness test
that is reported to be highly correlated (r = 0.80) with VO2 in children with
intellectual disabilities (Kerr et al. 2008). For children with CP, increasing
GMFCS levels reported an increase in the time to complete the 600-Yard test
(Mattsson and Andersson 1997).
Energy Expenditure Index (EEI)

Another indirect measure of VO2 uses changes in heart rate (HR) to measure the
energy expenditure index (EEI), which is expressed as EEI ¼ Exercise HRResting HR
walking Speed
(beats/m) (Keefer et al. 2004). This measure is also known as the Physiological Cost
Index (PCI) defined as PCI ¼ Finalwalking
HRResting HR
Speed (beats/m) (Raja et al. 2007). Similar
to oxygen cost, EEI and PCI are a measure of the economy of walking, with a high
values indicating poor walking economy (Rose et al. 1990; Provost et al. 2007). EEI
and PCI use the HR response during walking/exercising to assess the energy cost and
it is related to the change in velocity (Provost et al. 2007). EEI demonstrates a good
correlation (r = 0.61) with oxygen consumption for children with cerebral palsy
when walking at self-selected speeds (Norman et al. 2004). Furthermore, PCI was
calculated for specified distances (50, 100 and 150 m) at self-selected pace in
children with CP (Raja et al. 2007). The PCI when walking 50 m is reproducible
(ICC 0.80–0.88) for able-bodied children and children diagnosed with CP. The PCI
for able-bodied children (n = 100) increased from 0.10 beat/m to 0.11 beats/m when
walking on uneven ground and increased from 0.58 beats/m to 0.86 beats/m for
children with CP (n = 100).
Children with CP (spastic diplegia GMFCS level not reported) demonstrate a
threefold increase in EEI compared to peers, and EEI is less in hemiplegic CP versus
diplegic CP (Rose et al. 1990). This estimation of VO2 with HR can be achieved due
to the linear relationship between HR and VO2 during exercise (Rose et al. 1990).
Heart rate while walking at self-selected walking speed demonstrates good repeat-
ability (ICC = 0.81–0.96) with minimal detectable change 9.2–14.2 beats per
minute for able-bodied adults (Darter et al. 2013). However, variable results have
been reported when comparing EEI and VO2 for children with CP walking at
different speeds. One study reported a linear relationship (r = 0.84) between gross
VO2 and HR for children with CP (GMFCS not reported) walking at different speeds
(Rose et al. 1991). While another reported no relationship between gross EEI and
gross VO2 (Keefer et al. 2004), however they did report a low linear relationship
between net EEI and net VO2 (r = 0.50–0.64) across different walking speeds
(GMFCSI-II). Furthermore, previous research reports that only 38% of subjects
with CP (GMFCS I-II) demonstrate decreases in net VO2 and EEI with increases
in walking speed (Keefer et al. 2004). The reminder of subjects demonstrated
variable changes in net VO2 and EEI (Keefer et al. 2004). Because similar changes
in EEI and net VO2 were not demonstrated by all participants, the authors concluded
EEI is not a valid estimate of VO2 (Keefer et al. 2004). In another study, 10 children
diagnosed with CP (GMFCS I–III) and 15 able-bodied children were assessed using
a portable metabolic system when walking at their self-selected walking speed
(Norman et al. 2004). A stronger correlation was found between EEI and oxygen
consumption index for children with CP (GMFCS I-III) (r = 0.61) as compared to
able-bodied children (r = 0.40) (Norman et al. 2004). A possible reason for the
different results is that heart rate can be affected by numerous factors, including
stress, anxiety, anticipation of exercise, other impairments and medications used by
participants.
Interventions That Affect VO2 Measures
Surgery
Single-Event Multilevel Surgery (SEMLS) Effects on VO2

Single-event multilevel surgery (SEMLS) is often performed to improve gait for
children diagnosed with CP (Marconi et al. 2014). Instrumented gait analysis using
three-dimensional motion capture is recommended for developing appropriate treat-
ment plan for ambulatory children diagnosed with CP (Marconi et al. 2014;
Thomason et al. 2012). Instrumented gait analysis provides detailed information
regarding joint motions, moments and powers resulting in a summary of the biome-
chanics of walking. However, it does not provide a measure of the energy cost of
walking (Marconi et al. 2014).
16 H. White et al.
A systematic review of SEMLS literature for children with CP (GMFCS I-IV)

reported overall kinematics, kinetics and energy efficiency of walking improved
after SEMLS for children diagnosed with CP (McGinley et al. 2012). Additionally,
energy, oxygen cost and oxygen consumption when walking increased with
increases in GMFCS level (Kamp et al. 2014). In this study the children with CP
(GMFCS level I) demonstrated similar mean oxygen cost to able-bodied children
(0.25 ml/kg/m) (Kamp et al. 2014). However large inter-individual differences of
energy cost were noted within each GMFCS level. Energy cost when walking is
greater in children diagnosed with CP compared to typically developing children.
This is believed to be due to increased mechanical work and exaggerated displace-
ment of their center of mass when walking (Marconi et al. 2014).
In a small sample (n = 10) of children with CP (GMFCS I–III) that were assessed
before and after SEMLS, improvements in hip and knee motions, moments and
powers were seen (Marconi et al. 2014). However, the children’s mechanical work
(calculated from ground reaction forces, movement of center of mass and rotational
and translational energy calculated for each segment) was not significantly different
before and after surgery. Despite no change in mechanical work, oxygen cost when
walking did significantly decreased (Marconi et al. 2014). Therefore, the authors
proposed the decrease in energy cost was due to decrease in energy consumption
needed to maintain upright posture when walking (Marconi et al. 2014).
The oxygen cost of walking in patients who underwent femoral derotational
osteotomies as part of SEMLS was compared to subjects who underwent SEMLS
without femoral derotational osteotomies (McMulkin et al. 2015). The results of this
study show significant decreases in net oxygen cost (0.03 ml O2/kg/m) for both
groups for GMFCS levels I and II but a non-significant decrease (.05 ml O2/kg/m)
for GMFCS level III. However, sample size for level III participants was 10 com-
pared to 70 participants who were classified GMFCS levels I & II (McMulkin et al.
2015). In another study, the PCI in 35 children with CP (GMFCS levels not reported)
decreased from 1.13 beats/m to 0.83 beats/m after undergoing SEMLS surgery
( p < 0.01) (Raja et al. 2007).
Rhizotomy Effects on VO2

For children diagnosed with CP, muscle spasticity is proposed to be due to damage to
the motor cortex, which results in decreased cortical input to the corticospinal tract
(Valle et al. 2007). A decrease in descending input to the spinal interneuron pool can
result in increased activity of the gamma and alpha motor neurons (Valle et al. 2007;
Steinbok 2007; Verrotti et al. 2006). The end result of this increased activity is
muscle spasticity. Input to the spinal interneuron pool via the afferent nerves in the
dorsal roots have a net excitatory effect on the efferent nerves output via the alpha
motor neuron (Steinbok 2007). Therefore when a dorsal Rhizotomy surgery is
performed (cutting 50%–70% of dorsal sensory roots L1-S2) the result is a decrease
in the excitability of alpha motor neuron and decrease in muscle spasticity (Steinbok
2007). Results of a dorsal Rhizotomy are a permanent reduction in muscle spasticity
with improvements in gait and other gross motor skills (Carraro et al. 2014). Except
for excluding patients with dystonia, to date there is still no uniform criteria for
patient selection for dorsal Rhizotomy (Carraro et al. 2014). Two studies with small
sample size reported children diagnosed with CP (GMFCS I–III) demonstrated
significant decreases in stance phase knee flexion (crouch) and a trends of decreases
in heart rate and oxygen consumption when walking 1 year after undergoing a dorsal
Rhizotomy (Chan et al. 2008; Carraro et al. 2014).
Therapy Effects on VO2
Children classified as GMFCS levels III and IV must use an assistive device (walker
or crutches) to ambulate. The two most common types of walkers are forward
(walker in front of child) or reverse (walker behind child). Reverse walkers are
supposed to encourage more upright trunk posture and less crouched gait patterns for
children with CP (Park et al. 2001). One study assessed the differences between
anterior and posterior walkers for ten children diagnosed with CP (GMFCS Level
III) who used each type of walker (randomly assigned) for 1 month and then the used
other type. A decrease in average oxygen cost and oxygen consumption was found
when using a posterior walker compared to an anterior walker (Park et al. 2001).
Therefore, children demonstrated a less crouched gait pattern and a more energy
efficient gait when using a posterior walker compared to an anterior walker. Con-
versely, an older study assessed ten children with CP (GMFCS Level III) using both
type of walkers (randomly assigned) on the same day and reported no difference in
oxygen cost between the two walker types (Mattsson and Andersson 1997).
Regardless of age, physical activity of persons with CP is reportedly lower than
able-bodied person. A decrease in physical activity may increase the risk of cardio-
vascular and cardiopulmonary impairments. Therefore, it is important to understand
why persons with CP demonstrate decreases in physical activity compared to their
peers (Fowler et al. 2007). Previous research has demonstrated that children with CP
have decreased aerobic capacity and increased energy cost when walking compared
to typically developing children (Rose et al. 1990). A decrease in physical activity
can lead to decreases in aerobic capacity which can result in the cyclical problem of a
decreased aerobic capacity leading to a decrease in participation in physical
activities.
Short term benefits from cardiopulmonary and strength training has been
documented for persons with CP. A randomized controlled trial was performed to
assess the effects of cardiopulmonary and strength training for adolescents and
young children diagnosed with CP (GMFCS I–IV) (Slaman et al. 2014). The control
group continued regular care (approximately 2 h of traditional therapy), while the
treatment group received specialized cardiopulmonary and strength training. Body
composition, muscle strength and cardiopulmonary fitness were assessed at four
intervals: prior to randomization, 3, 6 and 12 months after initial evaluation. The
treatment group demonstrated 10%–30% increases in cardiopulmonary fitness at
6 months, while the control group was unchanged. Additionally, decreases were seen
in skin fold thickness, blood pressure and total cholesterol for the treatment group
after 1 year (Slaman et al. 2014).
18 H. White et al.
A 32% decrease in the EEI after body weight supported treadmill training has
been reported (Provost et al. 2007). A 9% decrease in EEI after a 9 week circuit
training and aerobic endurance training program (Gorter et al. 2009). Additionally,
another study found that the distance walked during the 6-mwt significantly
improved after an 8 week walking exercise program (Nsenga Leunkeu et al.
2012). Aerobic exercise in two additional studies reported improvements in VO2
for children with CP using stationary bikes or arm-ergometers (Fowler et al. 2007;
Slaman et al. 2013).
One study assessed adults (mean age 36 6 years) with CP (GMFCS levels
I–III) who participated in a prospective study assessing oxygen consumption while
walking at self-selected speed, peak oxygen consumption during progressive stress
test (cycle ergometer), and daily walking time (activity monitor wore for 48 h on
weekdays) (Slaman et al. 2013). There was not a relationship between daily walking
time and peak oxygen consumption or walking consumption when walking. How-
ever, there was a significant negative relationship between physical strain while
walking and total daily walking time. Walking speed for participants with CP was
32% lower and physical strain of participants with CP was twice that of able-bodied
subjects. Therefore adults with CP used a larger proportion of their metabolic reserve
for walking compared to able-bodied person resulting in them walking less with the
total daily walking time averaged 1 h and 24 min (Slaman et al. 2013).
Conclusions
Regardless of age, children, adolescents and adults diagnosed with CP have

decreased physical activity and increased energy, oxygen cost and oxygen consump-
tion (measured with direct and indirect methods) when walking, compared to able-
bodied persons. Decreases in physical activity may increase the risk of cardiovas-
cular and cardiopulmonary compromise in children and adults diagnosed with CP
and these impairments may contribute to decreases in physical activity. This can
cause capacity and participation to become circular problems, with capacity affect-
ing participation and vice versa. However, surgical interventions (SEMLS, and
Rhizotomy) and therapy have been reported to increase walking distances and
decrease energy expended when walking for persons diagnosed with CP.
References
Geoffrey E, Moore J, Larry Durstine, Patricia L, Painter American College of Sports Medicine
ACMS’s exercise management for persons with chronic diseases and disabilities/American
College of Sports Medicine. Champaign, IL: Human Kinetics; 2016
Walter R, Thompson, Neil F, Gordon Linda S, Pescatello, American College of Sports Medicine
ACSM’s guidelines for exercise testing and prescription/American College of Sports Medicine.
Philadelphia: Wolters Kluwer Health/Lippincott Williams and Wilkins; 2014. p 480
Augsburger S, Tylkowski C (2000) A comparison of volumetric oxygen consumption to gait
mechanical energy in normal and pathological gait. In: Harris GF, Smith PA (eds) A new
millennium in clinical care and motion analysis technology. Institute of Electical and Electronics
Engineers, Piscataway, pp 109–115
Bar-Haim S, Belokopytov M, Harries N, Frank A (2004) A stair-climbing test for ambulatory
assessment of children with cerebral palsy. Gait Posture 20:183–188
Battley E (1995) The advantages and disadvantages of direct and indirect calorimetry. Thermochem
Acta 250:337–352
Bowen TR, Lennon N, Castagno P et al (1998a) Variability of energy-consumption measures in
children with cerebral palsy. J Pediatr Orthop 18:738–742
Bowen TR, Cooley SR, Castagno PW et al (1998b) A method for normalization of oxygen cost and
consumption in normal children while walking. J Pediatr Orthop 18:589–593
Carraro E, Zeme S, Ticcinelli V et al (2014) Multidimensional outcome measure of selective dorsal
rhizotomy in spastic cerebral palsy. Eur J Paediatr Neurol 18:704–713
Chan SH, Yam KY, Yiu-Lau BP et al (2008) Selective dorsal rhizotomy in Hong Kong: multi-
dimensional outcome measures. Pediatr Neurol 39:22–32
Cook CE, (2008) Clinimetrics Coroner: The Minimal Clinically Important Change Score (MCID):
A Necessary Pretense. J Man Manip Ther 16(4): E82–E83
Darter BJ, Rodriguez KM, Wilken JM (2013) Test-retest reliability and minimum detectable change
using the K4b2: oxygen consumption, gait efficiency, and heart rate for healthy adults during
submaximal walking. Res Q Exerc Sport 84:223–231
DeJaeger D, Willems PA, Heglund NC (2001) The energy cost of walking in children. Pflugers
Arch 441:538–543
Fowler EG, Kolobe TH, Damiano DL et al (2007) Promotion of physical fitness and prevention of
secondary conditions for children with cerebral palsy: section on pediatrics research summit
proceedings. Phys Ther 87:1495–1510
Gorter H, Holty L, Rameckers EE et al (2009) Changes in endurance and walking ability through
functional physical training in children with cerebral palsy. Pediatr Phys Ther 21:31–37
Hassani S, Krzak JJ, Johnson B et al (2014) One-Minute Walk and modified Timed Up and Go tests
in children with cerebral palsy: performance and minimum clinically important differences. Dev
Johnston TE, Moore SE, Quinn LT, Smith BT (2004) Energy cost of walking in children with
cerebral palsy: relation to the Gross Motor Function Classification System. Dev Med Child
Neurol 46:34–38
Jones J, McLaughlin JF (1993) Mechanical efficiency of children with spastic cerebral palsy. Dev
Kamp FA, Lennon N, Holmes L et al (2014) Energy cost of walking in children with spastic cerebral
palsy: relationship with age, body composition and mobility capacity. Gait Posture 40:209–214
Kay RM, Rethlefsen SA, Kelly JP, Wren TA (2004) Predictive value of the Duncan-Ely test in distal
rectus femoris transfer. J Pediatr Orthop 24:59–62
Keefer DJ, Tseh W, Caputo JL et al (2004) Comparison of direct and indirect measures of walking
energy expenditure in children with hemiplegic cerebral palsy. Dev Med Child Neurol
46:320–324
Kerr C, Parkes J, Stevenson M et al (2008) Energy efficiency in gait, activity, participation, and
health status in children with cerebral palsy. Dev Med Child Neurol 50:204–210
Maher CA, Williams MT, Olds TS (2008) The six-minute walk test for children with cerebral palsy.
Int J Rehabil Res 31:185–188
Maltais DB, Robitaille NM, Dumas F et al (2012) Measuring steady-state oxygen uptake during the
6-min walk test in adults with cerebral palsy: feasibility and construct validity. Int J Rehabil Res
35:181–183
Marconi V, Hachez H, Renders A et al (2014) Mechanical work and energy consumption in children
with cerebral palsy after single-event multilevel surgery. Gait Posture 40:633–639
Mattsson E, Andersson C (1997) Oxygen cost, walking speed, and perceived exertion in children
with cerebral palsy when walking with anterior and posterior walkers. Dev Med Child Neurol
39:671–676
20 H. White et al.
McDowell BC, Kerr C, Parkes J, Cosgrove A (2005) Validity of a 1 minute walk test for children
with cerebral palsy. Dev Med Child Neurol 47:744–748
McGinley JL, Dobson F, Ganeshalingam R et al (2012) Single-event multilevel surgery for children
with cerebral palsy: a systematic review. Dev Med Child Neurol 54:117–128
McMulkin ML, Gordon AB, Caskey PM et al (2015) Outcomes of orthopaedic surgery with and
without an external femoral derotational osteotomy in children with cerebral palsy. J Pediatr
Orthop Gait and Posture 41:608–612
Morris C (2007) Definition and classification of cerebral palsy: a historical perspective. Dev Med
Child Neurol Suppl 109:3–7
Nene AV, Evans GA, Patrick JH (1993) Simultaneous multiple operations for spastic diplegia.
Outcome and functional assessment of walking in 18 patients. J Bone Joint Surg Br 75:488–494
Norman JF, Bossman S, Gardner P, Moen C (2004) Comparison of the energy expenditure index
and oxygen consumption index during self-paced walking in children with spastic diplegia
cerebral palsy and children without physical disabilities. Pediatr Phys Ther 16:206–211
Nsenga Leunkeu A, Shephard RJ, Ahmaidi S (2012) Six-minute walk test in children with cerebral
palsy gross motor function classification system levels I and II: reproducibility, validity, and
training effects. Arch Phys Med Rehabil 93:2333–2339
Oeffinger DJ, Tylkowski CM, Rayens MK, Davis RF, Gorton GE, D’Astous J, Nicholson DE,
Damiano DL, Abel MF, BAgley AM, Luan J (2004) Developmental Medicine and Child
Neurology 46:311–319
Oeffinger D, Bagley A, Rogers S, Gorton G, Kryscio R, Abel M, Damiano D, Barnes D, Tylkowski
C (2008) Outcome tools used for ambulatory children with cerebral palsy: responsiveness and
minimum clinically important differences. Dev Med Child Neurol 50(12): 918–925
Palisano RJ, Hanna SE, Rosenbaum PL et al (2000) Validation of a model of gross motor function
for children with cerebral palsy. Phys Ther 80:974–985
Park ES, Park CI, Kim JY (2001) Comparison of anterior and posterior walkers with respect to gait
parameters and energy expenditure of children with spastic diplegic cerebral palsy. Yonsei Med
J 42:180–184
Potter CR, Unnithan VB (2005) Interpretation and implementation of oxygen uptake kinetics
studies in children with spastic cerebral palsy. Dev Med Child Neurol 47:353–357
Powers S, Howley E (2007) Exercise physiology: theory and application to fitness and performance.
McGraw Hill, New York
Provost B, Dieruf K, Burtner PA et al (2007) Endurance and gait in children with cerebral palsy after
intensive body weight-supported treadmill training. Pediatr Phys Ther 19:2–10
Raja K, Joseph B, Benjamin S et al (2007) Physiological cost index in cerebral palsy: its role in
evaluating the efficiency of ambulation. J Pediatr Orthop 27:130–136
Rodda JM, Graham HK, Nattrass GR et al (2006) Correction of severe crouch gait in patients with
spastic diplegia with use of multilevel orthopaedic surgery. J Bone Joint Surg Am 88:2653–2664
Rose J, Gamble JG, Medeiros J et al (1989) Energy cost of walking in normal children and in those
with cerebral palsy: comparison of heart rate and oxygen uptake. J Pediatr Orthop 9:276–279
Rose J, Gamble JG, Burgos A et al (1990) Energy expenditure index of walking for normal children
and for children with cerebral palsy. Dev Med Child Neurol 32:333–340
Rose J, Gamble JG, Lee J et al (1991) The energy expenditure index: a method to quantitate and
compare walking energy expenditure for children and adolescents. J Pediatr Orthop 11:571–578
Russell D, Rosenbaum P, Gowland C et al (1993) Gross motor function measure manual. McMaster
University, Hamilton
Scholtes VA, Becher JG, Beelen A, Lankhorst GJ (2006) Clinical assessment of spasticity in
children with cerebral palsy: a critical review of available instruments. Dev Med Child Neurol
48:64–73
Slaman J, Bussmann J, van der Slot WM et al (2013) Physical strain of walking relates to activity
level in adults with cerebral palsy. Arch Phys Med Rehabil 94:896–901
Slaman J, Roebroeck M, van der Slot W et al (2014) Can a lifestyle intervention improve physical
fitness in adolescents and young adults with spastic cerebral palsy? A randomized controlled
trial. Arch Phys Med Rehabil 95:1646–1655
Stallings VA, Zemel BS, Davies JC et al (1996) Energy expenditure of children and adolescents
with severe disabilities: a cerebral palsy model. Am J Clin Nutr 64:627–634
Steinbok P (2007) Selective dorsal rhizotomy for spastic cerebral palsy: a review. Childs Nerv Syst
23:981–990
Thomas SS, Buckon CE, Piatt JH et al (2004) A 2-year follow-up of outcomes following orthopedic
surgery or selective dorsal rhizotomy in children with spastic diplegia. J Pediatr Orthop B
13:358–366
Thomason P, Rodda J, Sangeux M et al (2012) Management of children with ambulatory cerebral
palsy: an evidence-based review. Commentary by Hugh Williamson Gait Laboratory staff. J
Pediatr Orthop 32(Suppl 2):S182–S186
Thompson P, Beath T, Bell J et al (2008) Test-retest reliability of the 10-metre fast walk test and
6-minute walk test in ambulatory school-aged children with cerebral palsy. Dev Med Child
Neurol 50:370–376
Umberger BR, Augsburger S, Resig J et al (2013) Generation, absorption, and transfer of mechan-
ical energy during walking in children. Med Eng Phys 35:644–651
Unnithan VB, Dowling JJ, Frost G, Bar-Or O (1996) Role of cocontraction in the O2 cost of walking
in children with cerebral palsy. Med Sci Sports Exerc 28:1498–1504
Valle AC, Dionisio K, Pitskel NB et al (2007) Low and high frequency repetitive transcranial
magnetic stimulation for the treatment of spasticity. Dev Med Child Neurol 49:534–538
Van de Walle P, Hallemans A, Schwartz M et al (2012a) Mechanical energy estimation during
walking: validity and sensitivity in typical gait and in children with cerebral palsy. Gait Posture
35:231–237
Van de Walle P, Hallemans A, Truijen S et al (2012b) Increased mechanical cost of walking in
children with diplegia: the role of the passenger unit cannot be neglected. Res Dev Disabil
33:1996–2003
van den Hecke A, Malghem C, Renders A et al (2007) Mechanical work, energetic cost, and gait
efficiency in children with cerebral palsy. J Pediatr Orthop 27:643–647
Vargus-Adams JN, Majnemer A (2014) International Classification of Functioning, Disability and
Health (ICF) as a framework for change: revolutionizing rehabilitation. J Child Neurol
29:1030–1035
Verrotti A, Greco R, Spalice A et al (2006) Pharmacotherapy of spasticity in children with cerebral
palsy. Pediatr Neurol 34:1–6
Verschuren O, Bloemen M, Kruitwagen C, Takken T (2010) Reference values for aerobic fitness in
children, adolescents, and young adults who have cerebral palsy and are ambulatory. Phys Ther
90:1148–1156
Verschuren O, Zwinkels M, Ketelaar M et al (2013) Reproducibility and validity of the 10-meter
shuttle ride test in wheelchair-using children and adolescents with cerebral palsy. Phys Ther
93:967–974
Welch WA, Strath SJ, Swartz AM (2015) Congruent validity and reliability of two metabolic
systems to measure resting metabolic rate. Int J Sports Med 36:414–418
Wood E, Rosenbaum P (2000) The gross motor function classification system for cerebral palsy: a
study of reliability and stability over time. Dev Med Child Neurol 42:292–296
The Use of Kinematics for Pulmonary
Volume Assessment
Carlo Massaroni
Abstract
Since the movements of the lung are transmitted to the chest wall during
breathing, the interest in technologies for the noninvasive monitoring of the
kinematics of the thorax to study the breathing biomechanics is growing
up. The analysis of the kinematics of the chest wall can be used to: (i) measure
human chest wall movements, (ii) study the mechanics of breathing, and (iii)
evaluate respiratory volumes during breathing. Differently from the spirometer
and other equipment, tools based on the tracking of the chest wall deformities
allow patients to perform breathing exercises without any kind of constraints and
to investigate the unaltered breathing mechanics, as well as to separately study the
behavior of each compartment of the chest wall.
The most well-established technology to assess the breathing volume of the
chest wall and of its compartments (i.e., pulmonary and abdominal rib cage and
abdomen) is the optoelectronic plethysmography (OEP). OEP is a motion capture
system designed and validated to track a number of photo-reflective markers
placed on the human chest wall. By the reconstruction of the kinematics of the
chest, OEP algorithm allows the computation of the subject’s respiratory volumes
and other breathing-related features.
Since 1990 a lot of research group have been developed and tested non-
invasive optical technologies to assess respiratory pattern parameters, to measure
asynchronies inside chest wall, to investigate patient respiratory strategies and
volume moved, and to distinguish different respiratory disease. Specifically, the
increasing use of OEP in clinical evaluation context in the respiratory field is
widely proven by the growing number of published articles in the last few years.
C. Massaroni (*)
Unit of Measurements and Biomedical Instrumentation, Campus Bio-Medico di Roma University,
Rome, Italy
e-mail: c.massaroni@unicampus.it

DOI 10.1007/978-3-319-30808-1_42-1
2 C. Massaroni
Its noninvasiveness and the possibility of the use of other equipment (e.g.,
pressure sensors, electromyography, electrocardiogram) and devices (e.g., cycle
ergometer, treadmill) at the same time allow to study a wide range of patients with
different physiological and clinical condition (included the no-collaborative ones)
and provide new perspective on the evaluation of ventilatory parameters.
Keywords
Breathing • Respiratory volumes • Breathing mechanics • Chest wall kinematics •
Lung volumes
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Theory of OEP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Measurement Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Validity, Accuracy, Reliability, and applications of the Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 10
Clinical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Limitation and Drawbacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Conclusions/Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Cross-Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
In clinical practice and scientific research, the evaluation of the biomechanics of the
chest wall, and the analysis of the breathing volumes by the use of optical and
optoelectronic technologies is growing up.
The indirect measurement of the ventilation by the external measurement of chest
wall surface kinematics has some advantages compared to the classical tools used to
measure the pulmonary ventilation (i.e., spirometer or a pneumotachograph). First of
all, the optoelectronic technologies are less invasive than the flow measure that can
appear to be an easy procedure but that can be affected by several issues as the need
of the use of mouthpieces, face masks, and nose clips can thereby increase the tidal
volume calculation. Secondly, the gas composition, temperature, humidity, as well as
pressure generally influence the recording of lung volume collected with the
pneumotachographs. Thirdly, the uncooperation of the patient have to be considered:
technologies based on the flow integration during the time cannot be used or are
extremely difficult in children or uncooperative adults, as well as in some research
scenarios (e.g., intensive care unit).
Since 1960 scientific community starts to study respiratory system as a model and
founded its analysis on mechanical models of the different parts composing the
system (i.e., airways, lungs, respiratory muscles, thoraco-abdominal wall and every-
thing that surrounded the lungs). In the 1960s, Mead et al. published a series of
scientific papers in which the possibility of measuring lung volume variations by the
measure of thoracic and abdominal wall displacement were described (Mead 1963;
The Use of Kinematics for Pulmonary Volume Assessment 3
Mead et al. 1967; Konno and Mead 1967). He demonstrated a relationship between
chest wall volume variations and its kinematics (Barnas et al. 1987). These studies
gave the way to hundreds of studies on respiratory mechanics in the 1980s with
works of Mead, Macklem, Peslin, and others who contributed to understand the
respiratory mechanics, the mechanisms of pulmonary ventilation, and to know how
the ventilatory pump acts on respiratory structures (with their elastic and resistive
properties) to ventilate the lungs.
Until the 1990s, studies were performed by the use of resistive and variable
inductance displacement sensors or magnetometers stiffly apposed on the chest wall
surface. The real technological breakthrough occurred in 1990 with Pedotti, who
was the first to use a system based on motion analysis technology that provided the
theoretical possibility of measuring and monitoring the kinematics of a number of
points by photo-reflective markers. The thoraco-abdominal surface, the enclosed
volume, and volume changes during breathing were measured for the first time
(Ferrigno et al. 1994).
State of the Art
Theory of OEP
From a physiological point of view, the chest wall models is a three compartment
model, composed by the pulmonary rib cage (RCp), the abdominal rib cage (RCa),
and the abdomen (AB) as highlighted in Fig. 1. The three-compartment model
allows to considered (i) that RCp and RCa are exposed at different pressures during
the inspiration, (ii) that the diaphragm acts directly only on RCa, and that non-
diaphragmatic inspiratory muscles act largely on RCp and not on RCa. Regarding
the abdomen, AB volume change is defined as the volume swept by the abdominal
wall (Konno and Mead 1967), and it is the result of the conjunct action of the
diaphragm and expiratory abdominal muscles. While in healthy subjects the three
compartments generally move synchronously, in different diseases, different asyn-
chronous movement between the three chest wall compartments can be found. Total
chest wall volume or global volume (VCW) is generally considered like the sum of
VRCp, VRCa, and VAB.
RCp is defined as the area extending from the clavicles to a line around the thorax
at the level of the xifoid process (corresponding to the area of apposition of the
diaphragm to the rib cage at the expiratory volume in sitting posture, confirmed by
percussion); RCa as the area extending from this line to the costal margin anteriorly
down from the xifosternum and to the level of the lowest point of the lower costal
margin posteriorly; AB is defined as the area extending caudally from the lower rib
cage to the level of the anterior superior iliac crest.
4 C. Massaroni
Fig. 1 3D human chest wall obtained after the triangles mesh. Markers are showed as black points;
blue, green, and orange surfaces represent the pulmonary rib cage (RCp), the adbominal rib cage
(RCa), and the abdomen (AB), respectively
Measurement Principles
The OEP working principle is based upon the principle of 3D motion capture
system. Six to eight TV cameras are placed in a dedicated room in a circular
perimeter around the subject. Camera placement must ensure that a minimum of
two cameras capture the markers during movement.
IR-reflective passive markers are affixed to the skin of the subject with double-
sided tape. For chest wall capture, 6 mm and 9 mm markers are usually used. Marker
placement is a large source of potential error: investigators must have outstanding
palpation skills to orient markers to body anatomy. Optimally, the same examiner
sets up each subject to ensure consistent placement.
The definition of the marker set prior to data collection determines the number of
markers used. The full marker set, the most widely employed marker set in breathing
analysis, uses 89 markers to define the three compartments. However, other marker
sets can be used to compartmentalize the chest wall in newborns and to monitor the
volumes in different postures than the standing one, like the supine or lateral ones.
Before starting with the data collection, a calibration procedure is needed to
minimize systematic errors that may occur due to effects such as lens distortion,
lack of flatness of the imaging plane, warm-up effects, and unequal pixel spacing.
Motion capture systems customarily use manufacturer-provided calibration cubes
and/or calibration wands. Once the calibration procedures have been completed,
parameters are stored for transformation of image coordinates into 3D marker
coordinates. Therefore it is crucial that camera positions hold steady throughout
the data collection. With multiple cameras, after the calibration, OEP reconstruction
is the systematic integration of a flat image from each camera into a 3D coordinate
system. OEP uses the direct linear transformation (DLT) as reconstruction technique
(Chen et al. 1994).
Markers are illuminated by stroboscopic light by means of a group of IR emitting
diodes positioned circularly around each camera lens. The synchronization between
Workstation
IR-photoreflective 3D markers Geometrical

markers coordinates model
Computation
algorithm
Volume
calculation
Chest wall compartmental VRCp

volumes and volume VRCa
changes VAB
Cameras VRCp VRCa VAB Vcw
18
OEP Volume [L]
10
0 10 20 30 40 50 60
time [s]
Fig. 2 Data collection carried out by using the OEP. Working flow shows the steps to obtain the
chest wall compartmental volume and volume changes. Moreover, a typical breathing pattern
during quiet breathing is reported in the figure as well
cameras is possible by the use of phase locked loop, which operate at frequency
higher than 60 Hz (Fig. 2).
Markers are automatically recognized by means of the dedicated computer
algorithm by a pattern recognition technique for object identification in real time
and according to shape and size, not exclusively to light intensity: the marker
centroid is calculated with a resolution of about 1:65.536 of the field of view
(Ferrigno and Pedotti 1985). A dedicated motion analyzer synchronizes input and
output information to and from cameras, and then an ad-hoc designed software in the
motion analyzer computes the 3D trajectories of each marker.
To compute the chest volumes, geometrical models are developed by defining a
closed surface starting with connecting each triplet of markers to arrange a triangle.
From each closed surface, the area contained into this surface can be calculated.
Figure 2 shows the placement and the relative displacement of the markers using an
89 marker set along the X, Y, and Z axes.
Several geometrical models have been developed and validated to calculate the
volume in both standing and supine positions. Chest wall model based on
32 markers, 54 markers, 86 markers, and 89 markers have been developed and
validated in literature (Ferrigno et al. 1994; Aliverti et al. 2001; Cala et al. 1996;
Massaroni et al. 2017). The model choice depends on the numbers of the markers
6 C. Massaroni
Fig. 3 89 marker protocol setup in the configuration generally used in the seated and standing
assessment: (a) front view of the chest wall (42 markers), (b) and (c) Lateral views of the chest wall
(5 lateral markers for each side), (d) back view of the chest wall (37 markers)
placed on the skin as well as on the number of compartment that one would like to
monitor during the collection (e.g., only rib cage and abdomen, pulmonary and
abdominal rib cage, both lower and upper abdomen contribution). Higher the
number of marker on the skin, better the mesh of triangle allowing the compartmen-
talization of the chest.
In the 89 marker set up, markers are placed in seven horizontal rows between the
clavicles and the anterior superior iliac spine. Along the horizontal rows, markers are
arranged anteriorly and posteriorly in five vertical rows and there is an additional
bilateral row in the midaxillary line. The anatomical landmarks for the horizontal
rows are clavicular line, manubriosternal joint, nipples, xiphoid process, lower costal
margin, umbilicus, and anterior superior iliac spine. Landmarks for vertical rows are
midlines, anterior and posterior axillary lines, midpoint of the interval between
midline and the anterior and posterior axillary line, and midaxillary lines. An extra
marker is added bilaterally at the midpoint between the xiphoid and the most lateral
portion of the tenth rib to provide better detail of the costal margin; two markers are
added in the region of overlying the lung apposed rib cage and in the corresponding
posterior position (Fig. 3).
Sitting, supine, and prone evaluations are carried out without posterior and
anterior markers, respectively, with a 52 markers protocol.
In newborns, the marker protocol is slightly different because the space limitation
of the chest wall surface. 24 hemispheric reflective markers can be positioned on the
anterior thoracic-abdominal surface from the clavicles to the anterior superior iliac
spines by designing six horizontal lines and five vertical lines to define the rib cage
and abdomen compartment. 52 markers can be used on infants where seven circum-
ferential horizontal rows between the clavicles and the anterior superior iliac spine
are considered, RCp is described by four levels of markers, and RCa by two.
Volume calculation is based on the method to compute volumes using the Gauss
theorem. For each triangle identified by three markers, the surface (S) and the
!
direction of normal vector ( n ) are calculated and the volume contained in this
surface can be calculated using the Gauss theorem as in Eq. 1:
ð ð
! !
F n dS ¼ dV ¼ V (1)
S V
!
where F is an arbitrary vector, S is a closed surface, V is the volume closed by S, and
!
n is the normal unit vector on S. This procedure allows the computation of the
volume enclosed by the thoraco-abdominal surface.
OEP chest wall model is a three compartment model. That model allows to
consider (i) that RCp and RCa are exposed at different pressures during the inspi-
ration, (ii) that the diaphragm acts directly only on RCa, and that nondiaphragmatic
inspiratory muscles act largely on RCp and not on RCa. Regarding the abdomen, AB
volume change is defined as the volume moved by the abdominal wall (Romagnoli
et al. 2008; Konno and Mead 1967), and it is the result of the conjunct action of the
diaphragm and expiratory abdominal muscles. While in healthy subjects the three
compartments generally move synchronously, in different diseases, different asyn-
chronous movement between the three chest wall compartments can be found. Total
chest wall volume (VCW) is the sum of VRCp, VRCa, and VAB.
One of the parameters measured by OEP is chest wall kinematics. The compart-
mental analysis results are useful to assess the contribution of each single compartment
or of each hemithorax of the chest wall to the respiratory pattern. Results obtained from
the breath-by-breath analysis of compartmental volumes can be further processed to
assess if the thoracic-abdominal movement of the chest wall is synchronous (Fig. 4).
Asynchrony is defined as the difference in time of expansion or retraction between the
compartments. When this difference is so great that the movement among the com-
partments becomes opposite, paradoxical movement occurs.
Konno and Mead described one of the first models of movement of the chest wall,
assuming that during spirometry breathing the respiratory system is closed, and
hence, it a has a single degree of freedom. In this case, volume variations of the rib
cage must be equal and opposite to volume variations of the abdomen and the two
compartments must move in phase. Within a given lung volume, there is a single
linear relationship between the rib cage and the abdomen (isovolumic line). During
quiet breathing, the coordination of diaphragm, intercostal, and scalene muscles
allows the thoracic-abdominal system to move with one degree of freedom along its
“relaxation line,” which results from plotting abdominal versus rib cage movement
during lung passive inflation. Breathing along this line represents the most efficient
way to exploit the respiratory system (Konno and Mead 1967).
After the description of this model, several methods were designed in order to
describe the synchrony of the thoracic-abdominal movement. The following are the
most commonly used variables: phase angle and shift, inspiratory phase ratio
(PhRIB), total phase ratio (PhRTB), expiratory phase ratio (PhREB), and the
cross-correlation function (CCF).
8 C. Massaroni
Fig. 4 (a) Plot of rib cage (RC) versus abdominal (AB) excursion plot illustrating calculation of
phase angle ϕ. s = maximal AB excursion, m = horizontal width of RC–AB loop at halfway
between maximal and minimal RC excursion. (b) Time plot of AB and RC excursions showing
phase shift, (c) Konno-Mead loop for subjects presenting synchronous chest wall motion, asyn-
chronous motion (d), and paradoxical chest wall motion (e)
Another kinematic measure uses the phase angle analysis as a quantitative

evaluation of the phase shift between the rib cage (RC, as the sum of RCp and
RCa) and the AB and reflects the delay between excursions of the two compartments
of the CW. It is measured in degrees ( ), ranging from 0 to 180 , where 0 represents
perfect synchrony, and 180 represent paradoxical movement. The calculation of the
phase angle is performed through equations extracted from Konno-Mead loop
(Konno and Mead 1967), or Lissajous figure, in which movements of one compart-
ment during one respiratory cycle are plotted against the excursion of a second
compartment in an X-Y graph.
In the case of angles lower than 90 , Eq. 1 is used:
sin Φ ¼ m =s (1)
while when angles between 90 and 180 occur, Eq. 2 is valid:
Φ ¼ 180 μ (2)
where μ ¼ m =s .
The variable m is the width of the figure at the midpoint of the maximal excursion
of the compartment represented in the Y axis, and s, the maximal excursion of the
compartment shown in the X axis. The advantage of this kind of analysis is that data
gathered throughout the respiratory cycle are evaluated. Phase angle analysis can be
used when rib cage has an almost sinusoidal shape. Non-sinusoidal curves and/or
figure-eight curves should be removed from analysis as they may affect the results.
For curves with Ф greater than 20 , the direction of the curve can be identified. A
clockwise curve indicates that the rib cage precedes the abdomen, and a counter-
clockwise curve indicates that the abdomen precedes the rib cage. In healthy sub-
jects, the movement of thorax and abdomen during breathing is almost synchronous,
and therefore it results in a line which is similar to a straight line. As the movement
becomes asynchronous, the line becomes a loop which can range from an elliptical to
a circular shape as the asynchrony worsens. In physiological conditions in quiet
breathing, the abdominal compartment leads the rib cage.
Other kinematic measures include Inspiratory phase ratio (PhRIB), Expiratory
phase ratio (PhREB), and Total phase ratio (PhRTB). These values represent the
percentage of time of the respiratory cycle in which the compartments of the RC and
AB move asynchronously: 0% represents synchrony, whereas 100% indicates par-
adoxical movement. These variables quantify the asynchrony at each point of the
respiratory cycle. No sinusoidal curves or Konno-Mead loops are required to
perform this analysis (Aliverti et al. 2009). The cross-correlation function represents
the delay in seconds among the compartments. When the movement is synchronous
the delay is equal to 0 s. The higher is the cross-correlation function, the greater the
asynchrony between the compartments of the CW (Millard 1999). The paradoxical
inspiratory time is the parameter used to evaluate the asynchrony of the CW
compartments in patients with COPD during exercise on a cycle ergometer. This
variable is defined as the fraction of inspiratory time, in percentage, in which the
volume of RCa decreases (Parreira et al. 2012).
Breathing volumes, volume changes, and compartmental percentage contribu-
tions are further measure of chest wall function. Volume variations of the chest wall,
of its three compartments, and of each hemithorax are calculated as the difference
between the end-inspiratory volume and the end-expiratory volume of the same
compartment. Different volume variables, measured in liters, can be assessed
through OEP: tidal volume of CW and of its compartments, end-expiratory volume
of CW and of its compartments, end-inspiratory volume of CW and of its compart-
ments. Moreover, the volume changes are calculated as percentage (%) of contribu-
tion of each compartment to the chest wall tidal volume. If maximal inspirations are
performed repeatedly during the exercise, changes in chest volume can also be
calculated regarding the total lung capacity (TLC), and it is possible to assess
restriction in vital capacity when the end-inspiratory volume is close to TLC. By
the analysis of the volumes, it is also possible to indirectly estimate the following
lung volumes: Expiratory Reserve Volume (ERV), Inspiratory Reserve Volume
(IRV), Forced Expiratory Volume in the first second (FEV1).
Within the respiratory cycle, by the kinematical analysis it is possible to evaluate
the total time of the respiratory cycle, the inspiration and expiratory time, the ratio
10 C. Massaroni
between inspiratory time and total time of the cycle, respiratory rate, and minute
ventilation (product of respiratory rate and tidal volume). The time variables mea-
sured in seconds by means of optical technology are: inspiratory time (Ti), expiratory
time (Te), and total time of the respiratory cycle. Besides these, the following
variables can be calculated: inspiratory time in relation to the total time (Ti/Ttot),
RR in breathing incursions per minute, minute ventilation (VE) in liters per minute,
and mean inspiratory flow and mean expiratory flow in liters per second (Parreira
et al. 2012). Table 1 reports several parameters that can be estimated by the
kinematics of the chest wall and their clinical significance, as reported in Wilhelm
et al. (2003).
Validity, Accuracy, Reliability, and applications of the

Measurements
An important requirement for a breathing measurement system based on optoelec-

tronic technology is the noninvasiveness, i.e., a decreasing degree of interference in
the subject’s performance of a human natural movement. OEP performances have
been widely evaluated by using ad-hoc designed calibrator systems and by the
comparison with reference instruments (i.e., flowmeters and pneumotachometers).
OEP is able to detect linear marker’s displacement higher than 30 μm, correspondent
to a volume threshold around 8.92 mL. By increasing the numbers of the cameras
adopted, OEP accuracy grows when spherical markers are placed on the skin.
Moreover, the accuracy in the volume estimation appears to be not influenced by
thorax’s movement, magnitude, and breathing rate. OEP validity in measuring lung
volume changes has been evaluated in healthy seated and standing subjects in
different experimental settings (e.g., quiet breathing, incremental exercises): maxi-
mum difference between spirometer and OEP was always <4% (Aliverti et al.
2015). The VCW measurements between the two instruments were always highly
correlated with discrepancy lower than 5%, considering constrained postures like
supine and prone position in healthy subjects during quiet and deep breathing. The
volume accuracy is always better than 6.0% of measured volume (Massaroni et al.
2015a). The volume repeatability evaluated delivering repeated known volumes was
found to be 2.7 mL. Regarding the intra-rater and inter-rater reliability of OEP, an
intraclass correlation coefficient (ICC) higher than 0.75 and a coefficient of variation
of method error (CVME) always lower than 10% has been assessed at rest and
during submaximal exercise on a cycle ergometer (Vieira et al. 2013).
The clinical application of OEP has been extensively used to assess the chest wall
kinematics and the compartmental volume changes on healthy subjects (i.e., adults
and newborns) in relation to age, gender, posture, and different physiological
conditions. The detailed analysis of total and compartmental chest wall volumes
allows to better understand mechanisms underlying different functions provided by
the respiratory muscles.
In adult, OEP has been applied to study for the first time how the influence of
posture and age can affect breathing variables. On elderly population, the influence
Table 1 Respiratory parameters that can be estimated by the rib cage, abdominal, and chest wall
patterns and their clinical significance in the breathing assessment
Parameter Clinical significance
Breathing rate Highly sensitive but nonspecific sign of respiratory dysfunction
Tidal volume Variability helps differentiate between restrictive (less) and
obstructive pulmonary diseases (more) as well as acute anxiety
(more); elevated at baseline in panic disorder
Minute ventilation Equivalent to tidal volume times respiratory rate; reflects
metabolic activity (e.g., increased with exercise); increases
(hyperventilation) in acute anxiety, bronchospasm, acute
pulmonary edema, and cerebrovascular accidents; decreases
(hypoventilation) with obstructed breathing, narcotic, and
sedative overdoses
Peak inspiratory flow Reflects respiratory centre drive; normally rises with increased
metabolic activity; similar to mean inspiratory flow but less
dependent on accurate estimation of the onset and offset of
inspiration; elevates at rest in restrictive lung disease because of
increased neural afferent stimulation from intrapulmonary
receptors; elevates in obstructive pulmonary disease because of
increased activity of pulmonary stretch receptors in chest wall;
decreases with narcotic and sedative overdoses
Ventilation/peak inspiratory Measure of appropriateness of ventilation to respiratory drive;
flow indirect correlation with intensity of breathlessness (e.g., values
low during resistive loading and dynamic pulmonary
hyperventilation as in acute bronchospasm and in chronic
pulmonary emphysema); helps distinguish organic from
psychogenic breathlessness
Fractional inspiratory time “Duty cycle” (Ti/Ttot, ratio of inspiratory to total breath time);
low values may reflect severe airways obstruction because
expiratory time is prolonged, also occur in normal speech;
values increase during snoring
Ratio between breathing rate “Rapid shallow breathing index” reflects the work of breathing
and tidal volume
Peak/mean inspiratory and Reflects presence of upper airway flow limitation during
expiratory flow inspiration or expiration
%RC/CW Percentage contribution of the rib cage excursions to tidal
volume; higher in women than men and during acute
hyperventilation; increases variability with respiratory muscle
fatigue or dysfunction
Phase relation Measure of thoraco-abdominal coordination during breathing;
increases in severe airways obstruction and during respiratory
muscle fatigue or dysfunction
Forced expiratory volume in Requires a voluntary breathing manoeuvre; reflects severity of
1 s (FEV1) airway obstruction; similar to % vital capacity expired in 1 s,
which can also be measured
of age on ventilatory kinematics and on mechanisms adopted to overcome age-

related physiological changes has been explored (Muniz de Souza et al. 2016).
Moreover, gender differences have been investigated through the OEP (Romei
12 C. Massaroni
et al. 2010). In preterm and term newborns, OEP has been demonstrated as a good
tool to assess lung volume, compartmental changes, and relative compartmental
distribution, to assess disease severity and the response to pharmacological inter-
ventions in critically ill newborns as well as to guide mechanical respiratory support
(Dellaca et al. 2010).
Clinical Applications
OEP has been successfully used in several clinical scenarios to better understand and
investigate pathological conditions. For examples:
• In the pulmonary assessment in Chronic Obstructive Pulmonary Disease (COPD)

patient
• In the neuromuscular disease assessment
• For the patient’s monitoring in the intensive care unit
• As an alternative tool to assess the respiratory rehabilitation outcome
• To assess the chest and abdomen wall movements during exercise, in athletes
OEP has been extensively used to evaluate total and compartmental volume
variations during incremental exercise in COPD patients and neuromuscular dis-
eases (Georgiadou et al. 2007). Studies on clinically stable patients have been carried
out during incremental exercises to better understand the Pathogenesis of the
Dynamic Hyperinflation (Aliverti et al. 2004), to evaluate the effect of salbutamol
on dyspnoea and End-Expiratory Lung Volume (Aliverti et al. 2005). Other studies
aimed at understanding the effect of oxygen therapy on dyspnoea (Bruni et al. 2012),
the effect of exercise program on chest wall volume (Gagliardi et al. 2014), and the
effect of exercise-induced dyspnoea on severely obstructed patients (Aliverti et al.
2005).
Neuromuscular diseases are characterized by progressive loss of muscle strength,
resulting in cough ineffectiveness with its deleterious effects on the respiratory
system. OEP has been used to investigate pulmonary volumes, respiratory muscle
strength, peak cough flow, and chest wall kinematics in neuromuscular patients
affected by Duchenne Muscular Dystrophy (DMD) (Lo Mauro et al. 2010), noctur-
nal hypoxemia (NH) (D’Angelo et al. 2009), Becker’s muscular dystrophy
(D’Angelo et al. 2011), Facioscapulohumeral Dystrophy, Limb-girdle myotonic
dystrophy, and Spinal muscular atrophy (SMA) (Lissoni et al. 1996). Among several
studies, OEP has been used to investigate if abdominal contribution to global volume
should be considered an early indicator of respiratory impairment in neuromuscular
diseases. Moreover, OEP has been used (i) to distinguish between those patients who
are in the early stages of NH development and those who are not, (ii) to rule out the
role of pharmacological and surgical therapies in DMD, (iii) to test the hypothesis
that operating forces on the chest wall may impact on distribution of inspired gas
volume to RC compartments, resulting in RC distortion and decrease in cough
effectiveness, (iv) to compare the chest wall kinematics during spontaneous
breathing and while using mechanically assisted ventilation (SMA patients) (Parreira
et al. 2012). In the intensive care unit, the OEP system has been used to evaluate the
effect of invasive and noninvasive ventilatory strategies on tidal volume and
thoraco-abdominal synchrony and to monitor chest wall biomechanics and the
distribution of VCW changes in patients undergoing mechanical ventilation
(Chiumello et al. 2007)
Patients who underwent surgical procedures may benefit from pulmonary reha-
bilitation to restore correct function of respiratory muscles. OEP can be used to
measure the chest wall volumes before and after the lung transplantation which can
alter the kinematics (Bastianini et al. 2009). Moreover, it has been used to assess the
postoperative ventilatory asymmetries by the analysis of the chest wall dynamics
reflected by different changes in the volume of the two hemithoraxes. Efficacy of the
pulmonary rehabilitation has been demonstrated in COPD patient by assessing
patients receiving superior lobectomy immediately after surgery, as well as after
2-weeks and after 4-weeks rehabilitation. Indeed, the negative effect of chest wall
tumors on global mechanics during quiet breathing and exercise has been confirmed
(Georgiadou et al. 2007).
The function of the respiratory system during exercise is usually assessed by the
analysis of expired air to calculate breathing frequency, tidal volume, minute venti-
lation, oxygen consumption, and carbon dioxide production. Investigations into the
use of OEP to assess the movement of the chest and abdomen during exercise are
emerging (Levai et al. 2017; Massaroni et al. 2016a, b). The results show that the
end-inspiratory lung volume increase during exercise are mainly achieved by
increasing end-inspiratory volumes of the RCp and RCa reflecting the inspiratory
RC muscle contribution. Moreover, by the investigation of the response pattern of
the compartmental breathing volumes and the relative contribution of respiratory
muscle groups between men and women of similar fitness, significantly higher
operational lung volumes have been assessed in men than women during
symptom-limited exercise.
In the last years, OEP has been used as a reference instrument for chest wall
compartmental and global volume values for the design of wearable. Wearable t-shirt
for patient monitoring are gaining popularity because they are comfortable and easy
to wear; hence, they allow a continuous monitoring of respiratory parameters
without impairing patient’s mobility (Massaroni et al. 2015b).
Instead of using spirometry as reference instrument for lung volume as the
majority of the studies on wearable systems for respiratory monitoring reported,
OEP has been used in some works to collect the reference breathing compartmental
volumes (Massaroni et al. 2016c). This approach is more valid because both the OEP
and the smart textile measurements are based on chest wall compartments kinematics
(and strain) analysis. Results encourage researchers to use the OEP in further studies
on this topic.
14 C. Massaroni
Limitation and Drawbacks
Although the OEP is used in the clinical research, the use of huge number of markers
discourage the common use in the clinical practice as an alternative tool to assess the
breathing. The markers placement can be tedious and complicated especially in
subjects in whom the landmarks are difficult to identify (e.g., newborns and obese
subjects). Moreover, the right placement of the markers requires practice and is
different from men and women because of the different shapes of the chest wall. To
respect privacy of the subject and their comfort, the presence of a same-sex techni-
cian is generally a good practice suggestion, but it may result in additional staff
requirement. From a technical point of view, if a marker is blocked from the camera
field, the OEP will render the matrix indecipherable, and the values will have to be
manually computed or additional scripts are required.
A motion capture system exclusively dedicated to perform optoelectronic pleth-
ysmography is quite expensive. However, algorithms to compute breathing volumes
from markers have been recently described in literature, and they can be used with
high-resolution commercial motion capture systems.
Conclusions/Summary
OEP is gaining large acceptance in the medical field because of its great advantage of
measuring breathing patterns in any condition by the noninvasive visualization of
the chest wall and its compartments. OEP is highly accurate in the measurement of
both temporal (e.g., breathing rate and I:E ratio) and volumetric breathing features
(e.g., compartmental volume changes). It can be considered a good tool to better
understand the biomechanics of the breathing in several scenarios (e.g., breathing
during high-intensity exercise) allowing partitioning of the complex shape of the
chest wall into different compartments. In addition to being noninvasive, it requires
no connection to the patient and it has successfully used in ICU to monitor the
patient and lead the ventilatory strategy. Thanks to the possibility to distinctly study
the left and right side of the chest wall, OEP has been also used in the outcome
assessment of patients after lobectomy. Promising results have been also obtained in
using the OEP in the characterization of new tools like wearable smart textiles.
Cross-Reference

References
Aliverti A et al (2001) Compartmental analysis of breathing in the supine and prone positions by
optoelectronic plethysmography. Ann Biomed Eng 29(1):60–70. Available at: http://link.
springer.com/10.1114/1.1332084. Accessed 24 Sept 2014
Aliverti A, Stevenson N, Dellacà RL, et al (2004) Regional chest wall volumes during exercise in
chronic obstructive pulmonary disease. Thorax 59:210–216
Aliverti A et al (2005) Effect of salbutamol on lung function and chest wall volumes at rest and
during exercise in COPD. Thorax 60(11):916–924. Available at: http://www.pubmedcentral.nih.
gov/articlerender.fcgi?artid=1747234&tool=pmcentrez&rendertype=abstract. Accessed
24 Sept 2014
Aliverti A et al (2009) Paradoxical movement of the lower ribcage at rest and during exercise in
COPD patients. Eur Respir J 33(1):49–60. Available at: http://www.ncbi.nlm.nih.gov/pubmed/
18799505. Accessed 24 Sept 2014
Aliverti A et al (2015) Optoelectronic plethysmography. In: Springer-Verlag (ed) Mechanics of
breathing pathophysiology, diagnosis, and treatment. Springer-Verlag, Milano
Barnas GM et al (1987) Impedance and relative displacements of relaxed chest wall up to 4 Hz. J
Appl Physiol (Bethesda, Md.: 1985) 62:71–81
Bastianini F, Silvestri S, Magrone G, Gallotta E, Sterzi S (2009) A preliminary efficacy evaluation
performed by opto-electronic plethysmography of asymmetric respiratory rehabilitation. In
2009 Annual International Conference of the IEEE Engineering in Medicine and Biology
Society, pp. 849–852
Bruni GI et al (2012) Dyspnea, chest wall hyperinflation, and rib cage distortion in exercising
patients with chronic obstructive pulmonary disease. Med Sci Sports Exerc 44:1049–1056
Cala SJ et al (1996) Chest wall and lung volume estimation by optical reflectance motion analysis. J
Appl Physiol (Bethesda, Md.: 1985) 81:2680–2689
Chen L et al (1994) An investigation on the accuracy of 3-dimensional space reconstruction using
the direct linear transformation technique. J Biomech 27:493–500
Chiumello D et al (2007) Effects of volume shift on the pressure-volume curve of the respiratory
system in ALI/ARDS patients. Minerva Anestesiol 73(3):109–118. Available at: http://www.
scopus.com/inward/record.url?eid=2-s2.0-34248997789&partnerID=tZOtx3y1. Accessed
24 Sept 2014
D’Angelo MG et al (2009) M.P.2.05 Optoelectronic plethysmography in respiratory assessment in
Duchenne muscular dystrophy. Neuromuscul Disord 19(8–9):572–572. Available at: http://
www.sciencedirect.com/science/article/pii/S0960896609002508. Accessed 24 Sept 2014
D’Angelo MG et al (2011) Respiratory pattern in an adult population of dystrophic patients. J
Neurol Sci 306(1–2):54–61. Available at: http://www.sciencedirect.com/science/article/pii/
S0022510X1100178X. Accessed 24 Sept 2014
Dellaca RRL et al (2010) Measurement of total and compartmental lung volume changes in
newborns by optoelectronic plethysmography. Pediatr Res 67(1):11–16. Available at: http://
www.ncbi.nlm.nih.gov/pubmed/19755932. Accessed 24 Sept 2014
Ferrigno G, Pedotti A (1985) ELITE: a digital dedicated hardware system for movement analysis
via real-time TV signal processing. IEEE Trans Biomed Eng 32(11):943–950. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/3905583. Accessed 20 Feb 2015
Ferrigno G et al (1994) Three-dimensional optical analysis of chest wall motion. J Appl Physiol
77:1224–1231
Gagliardi E, Bruni G, Presi I (2014) Thoraco-abdominal motion/displacement does not affect
dyspnea following exercise training in COPD patients. Respir Physiol Neurobiol 190:124–130
Georgiadou O et al (2007) Effects of rehabilitation on chest wall volume regulation during exercise
in COPD patients. Eur Respir J 29(2):284–291. Available at: http://www.scopus.com/inward/
record.url?eid=2-s2.0-33846926545&partnerID=tZOtx3y1. Accessed 21 Sept 2014
Konno K, Mead J (1967) Measurement of the separate volume changes of rib cage and abdomen
during breathing. J Appl Physiol 22(3):407–422
16 C. Massaroni
Levai IK et al (2017) Optoelectronic plethysmography (OEP) in the assessment of dysfunctional

breathing (DB) in athletes. Eur Respir J 48 (suppl 60)
Lissoni A et al (1996) Spinal muscular atrophy: kinematic breathing analysis. Am J Phys Med
Rehabil 75(5):332–339. Available at: http://search.ebscohost.com/login.aspx?direct=true&db=
cin20&AN=1997019718&site=ehost-live
Lo Mauro A et al (2010) Abdominal volume contribution to tidal volume as an early indicator of
respiratory impairment in Duchenne muscular dystrophy. Eur Respir J 35(5):1118–1125. Avail-
able at: http://www.scopus.com/inward/record.url?eid=2-s2.0-77951970331&partnerID=
tZOtx3y1. Accessed 24 Sept 2014
Massaroni C et al (2015a) Evaluation of optoelectronic plethysmography accuracy and precision in
recording displacements during quiet breathing simulation. In 37th Annual International Con-
ference of the IEEE Engineering in Medicine and Biology Society, pp 1291–1294. Available at:
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7318604. Accessed 21 Jan 2016
Massaroni C, Saccomandi P, Schena E (2015b) Medical smart textiles based on fiber optic
technology: an overview. J Funct Biomater 6(2):204–221
Massaroni C et al (2016a) Breathing analysis during exercise: comparison between optoelectronic
system and breath-by breath analyzer. In European Society of Biomechanics (ESB). Available
at: https://scholar.google.it/citations?view_op=view_citation&hl=it&user=jEHMd9kAAAAJ&
sortby=pubdate&citation_for_view=jEHMd9kAAAAJ:L8Ckcad2t8MC
Massaroni C et al (2016b) Optical measurement of breathing: algorithm volume calibration and
preliminary validation on healthy trained subjects. In 38th annual international conference of the
IEEE engineering in medicine and biology society, pp 2153–2156
Massaroni C, Saccomandi P, Formica D, Lo Presti D, Caponero MA, Di Tomaso G, Giurazza F,
Muto M, Schena E (2016c) Design and feasibility assessment of a magnetic resonance-com-
patible smart textile based on fiber Bragg grating sensors for respiratory monitoring. IEEE Sens
J 16(22):8103–8110
Massaroni C, Cassetta E, Silvestri S (2017) A novel method to compute breathing volumes via
motion capture systems: design and experimental trials. J Appl Biomech (in press). http://www.
ncbi.nlm.nih.gov/pubmed/28338381
Mead J (1963) The control of respiratory frequency. Ann N Y Acad Sci 109:724–729
Mead J et al (1967) Pulmonary ventilation measured from body surface movements
Millard RK (1999) Inductive plethysmography components analysis and improved non-invasive
postoperative apnoea monitoring. Physiol Meas 20:175–186
Muniz de Souza H et al (2016) Acute effects of different inspiratory efforts on ventilatory pattern and
chest wall compartmental distribution in elderly women. Respir Physiol Neurobiol 227:27–33.
Available at: http://www.sciencedirect.com/science/article/pii/S1569904816300118. Accessed
20 Mar 2016
Parreira VF et al (2012) Optoelectronic plethysmography: a review of the literature. Rev Bras Fisioter
16(6):439–453. Available at: http://www.ncbi.nlm.nih.gov/pubmed/23184278. Accessed
24 Sept 2014
Romagnoli I et al (2008) Optoelectronic plethysmography has improved our knowledge of respi-
ratory physiology and pathophysiology. Sensors 8(12):7951–7972. Available at: http://www.
scopus.com/inward/record.url?eid=2-s2.0-58149203421&partnerID=tZOtx3y1. Accessed
24 Sept 2014
Romei M et al (2010) Effects of gender and posture on thoraco-abdominal kinematics during quiet
breathing in healthy adults. Respir Physiol Neurobiol 172(3):184–191. Available at: http://www.
ncbi.nlm.nih.gov/pubmed/20510388. Accessed 24 Sept 2014
Vieira DSR et al (2013) Optoelectronic plethysmography: intra-rater and inter-rater reliability in
healthy subjects. Respir Physiol Neurobiol 189(3):473–476. Available at: http://www.ncbi.nlm.
nih.gov/pubmed/24036178. Accessed 24 Sept 2014
Wilhelm FH, Roth WT, Sackner MA (2003) The Life Shirt. Behav Modif 27(5):671–691
Measures to Determine Dynamic Balance
Timothy A. Niiler
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Static Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Semi-Qualitative Functional Tests of Dynamic Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Quantitative Testing of Dynamic Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Mathematical Methods: Finding Patterns to Identify Balance Impairment . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Abstract
This chapter discusses the theory of dynamic balance as is currently understood
and the various methods of assessing it. In order to test dynamic balance, it must
first be defined, and this is where an understanding is still evolving. As such, a
number of balance assessment methods have evolved concurrently.
The chapter starts by introducing some of the concepts of static balance and
obvious signs of balance loss. Next, functional testing methods including the
TUGT, BBS, DGI, and TBGA are introduced. These are tests that are relatively
simple to administer and rely on a rater’s assessment of signs like postural
instability or sway in completing a task. From there, instrumentation typically
found in a movement analysis lab is described along with the assessments used to
quantify dynamic balance. Finally, more advanced mathematical methods of
teasing out balance impairments from data are described. These last analyses
include usage of Lyapunov exponents, autocorrelation, margin of support, and
deviation of the COM from the interfoot line.
T.A. Niiler (*)

Gait Laboratory, Nemours/AI duPont Hospital for Children, Wilmington, DE, USA
e-mail: tim.niiler@gmail.com

DOI 10.1007/978-3-319-30808-1_44-1
2 T.A. Niiler
Studies indicating the validity of all methods presented are also described in
greater or lesser detail as needed throughout the chapter. This serves to introduce
readers to the potential populations in which balance deficits exist, as well as the
relative validity and reliability of the tests. In some cases, tests are valid with one
population, but not with another. At this juncture, there are a myriad of balance
tests, and more are being developed as this is being written. Ultimately,
researchers and clinicians must choose the appropriate test based on who they
are testing, what specific symptoms they wish to identify or diagnose, and the
consideration that balance is multifactoral.
Keywords
Gait • Dynamic balance • Posture • Stability • Center of mass • Balance
assessment • Balance impairment
State of the Art
The state of the art in measuring dynamic balance depends largely on who is doing it
and for what purpose. Most tools attempt to measure variability in gait either
qualitatively or quantitatively as a proxy for balance. However, some physics-
based methods attempt to predict when a subject is out of balance. In physical
therapy clinics, it has become routine to use semi-qualitative functional assessments
such as the Berg Balance Scale (BBS) or the Timed Up and Go Test (TUGT) as a
way to estimate future risk of falling, and a number of different measures of this type
have emerged for different populations. For athletes, dual-task cognition/balance
testing has become the norm for concussion assessment since concussion is known
to affect both balance and cognitive functions. Dual-task testing has now moved to
other areas and is routinely used with the elderly or with other balance-impaired
populations. Other populations used in the validation of dynamic balance assessment
include individuals with cerebral palsy (CP), Parkinson’s disease, and multiple
sclerosis. In gait and movement analysis laboratories, more quantitative methods
are often used. While many of these methods analyze the motion of the body’s center
of mass, others are used to identify gait patterns and thereby surmise dynamic
balance. The most popular are the margin of support and the Lyapunov stability
indices. These quantitative methods are more sensitive to therapeutic interventions
than the semi-qualitative scales employed by many physical therapists but require
expensive and dedicated hardware as well as staff capable of analyzing and
interpreting the results. Other tools which have emerged for this usage are acceler-
ometers and pressure-sensitive mats whose data may be analyzed for variability. As
technology has improved, it has become possible to use accelerometers in cell
phones or tablets to assess balance. Likewise, the development of inexpensive
depth cameras like the Microsoft Kinect has the potential to revolutionize balance
assessment and training.
Measures to Determine Dynamic Balance 3
Introduction
Assessment of balance is clinically useful in determining impairment and monitoring

functional improvement after therapeutic interventions. Whereas the lay person may
envision the paradigm of good balance as being the gymnast who pirouettes on the
balance beam or who sticks a landing after a vault, in clinical practice, assessments
will be less spectacular and more mundane. Balance is not a hit or miss endeavor as
in gymnastics, but rather, it is a continuum, and as such, it must be quantified to be
useful.
Watching a child who attempts to walk a balance beam for the first time gives
insight into how we might conduct an assessment. Invariably, the child will sway
with arms out, and very likely she will miss her footing and step off and to the side,
and we will say that she has lost her balance. In some cases there will be sway or
vibration, but she will keep her footing. Here we might say that she has become
“unstable” although balance was maintained. In this later case, her body might
contort substantially in an effort to stay on the beam, while in the former case, her
posture might have been erect as she stepped off the beam. This brings us to a
conundrum: is dynamic balance about not falling, not stepping off a line, or
maintaining posture? In the first case, the determination of balance loss is clear,
but in the second, such a determination is much more nuanced.
Static Balance
In static balance assessment, both ideas are used. Physical therapists will routinely
assess patients using timed one-legged standing. A patient is asked to hold his
balance with one leg raised for as long as possible before setting the raised foot
down. A longer time is indicative of better static balance. In this case, it is clear that
the drill involves balance, strength, and endurance, so a better balanced person who
is weaker might not do as well as someone with poorer balance but who has great
endurance and strength.
In order to improve this test, we must have a better understanding of how humans
balance so as to isolate the factors that are controlling balance from aspects of
strength and endurance. The balance system is primarily influenced by three factors
(balance triad): the vestibular system, vision system, and somatosensory system. The
first component is fairly significant and depends on the fluid levels in the bony
labyrinth of the inner ear. These levels function as a type of gyroscope giving the
brain a reference point of the head’s orientation. Visual input to balance works
similarly – a view of the external environment allows humans to map their internal
sense of position to the exterior world. The final component involves the sensing of
ground reaction forces and body positioning and using these as a reference for
postural orientation. Of these, visual cues to balance are the easiest to train. When
subjects are trained to look at a fixed point while standing on one leg, their
one-legged stand time increases. To test other components of balance, one can
then have the subject close his eyes so that only the vestibular system and
4 T.A. Niiler
20
a Young Adult
b 20 Elderly
15 15
10 10
5 5
mm (AP)
0 0
–5 –5
–10 –10
–15 –15
–20 –20
–20 –15 –10 –5 0 5 10 15 20 –20 –15 –10 –5 0 5 10 15 20
mm (ML) mm (ML)
Fig. 1 Stabilogram trace collected over a 4 min quiet standing period from a young adult (left) and
elderly subject (right) (Adapted from Muir et al. 2013. Creative Commons Attribution License)
proprioception are in play. Invariably, this decreases the one-legged stand time, so
endurance and strength are less important (Springer et al. 2007).
There is still the issue of postural control. It is possible that one subject balances
longer than another in the above assessment and yet sways wildly, while the other
subject stands quietly. To truly quantify the amplitude of sway requires more than
just visual inspection. Labs and clinics often have force plates that can be used to
identify the position of a subject’s center of pressure (COP). Subjects with more
sway will have more variability in this position than those who stand more quietly.
Related to the center of pressure is the center of mass (COM) which is typically near
the naval anatomically. Ideally, when one is balanced and standing on 2 ft, the center
of mass is over the top of the center of pressure. In reality, the COM is not stationary,
and as it oscillates, the COP will also move in response. Balance loss occurs when
the COM moves outside the subject’s base of support as defined by a single foot in a
one-legged balance or the region between the outer edges of the feet during quiet
standing. When the COM deviates from this region and no corrective action is taken,
a gravitational torque will cause the subject to fall: that is to either take a step to
adjust the base of support (also moving the COP) or to quite literally topple over.
Since it is possible to estimate the COM from the COP using the gravity line
projection technique (Lafond et al. 2004), force plates can be used to estimate
COM movement.
The current thinking is that larger sway during quiet standing implies either
poorer proprioceptive or vestibular control of balance (Fig. 1).
If variability of COM defines static balance and the movement of the COM
outside the base of support implies balance loss, then we are left with a problem
when it comes to dynamic balance. Dynamic balance is balance on the fly: postural
and destination control when one is not quietly standing either on one or both legs.
During the swing phase of walking, only a single leg is planted, and the COM may
get quite far forward, relatively, from the planted foot. Although, from a static
viewpoint, the walker is unstable, so long as she is able to get her swing foot
down in time to “catch” herself, she will go on walking without incident. The
forward deviation of the COM in gait acts to help propel the walker rather than
upsetting her balance. Of course, due to the propulsive nature of its anterior motion,
some authors have expressed that lateral deviations of the COM are more meaning-
ful with regard to balance loss (Hsue et al. 2009a, b; Chou et al. 2003; Schrager et al.
2008).
This observed lateral deviation of the COM upon balance impairment is the basis
of the old standard walk and turn drunk driving test that was commonly used before
breathalyzers were commonplace. In addition to other attempts to quantify propri-
oceptive impairment (touching ones nose with closed eyes), police officers would
ask suspected drunk drivers to walk a straight line without deviating, taking nine
steps heel to toe, turning on one foot, and then returning. Suspects who showed
balance difficulties including use of arms for balance, an inability to stay on the line,
an inability to place each foot heel to toe, or a certain amount of sway were then
detained for further questioning (Stuster 2006). Dynamic balance testing has evolved
to involve some combination of functional testing as with the walk and turn drunk
driving test and real quantification of how the center of mass moves during gait.
Semi-Qualitative Functional Tests of Dynamic Balance
Single-Task Assessments
In daily life, balance impairment is a functional issue. Individuals with such impair-
ment either move more slowly to guard against balance loss or because they lack the
strength or motor control to move more quickly. Additionally, those with balance
deficits may simply be unable to complete certain tasks. With this in mind, a number
of functionally oriented balance tests have been designed to estimate dynamic
balance impairment. These include the Berg Balance Scale (BBS), the Tinetti
Balance and Gait Assessment (TBGA), the Dynamic Gait Index (DGI), and the
Timed Up and Go Test (TUGT).
The BBS is a quasi-dynamic balance assessment which includes, among other
things, ratings of sit to stand, stand to sit, transfer from bed to chair, reach forward,
pick up object from the floor, and look over shoulder and turn 360 (Berg et al.
1989). In all, the assessment is comprised of 14 tasks each of which may be scored
on a 5-point scale (from 0 to 4) with 56 being the maximum possible score. This test
is commonly used in the physical therapy world because it requires few tools other
than a checklist, a subject, and an assessor, and it has been validated a number of
times (Berg et al. 1992; Wrisley and Kumar 2010; Whitney et al. 2003). A 2006
survey indicated that among physical therapists in Canada, it was the most popular
balance test (Korner-Bitensky et al. 2006). The BBS has some mixed results in
identifying at risk for falling individuals. In a Brazilian study of 188 older adults, the
scale was used dichotomously to predict the score (49) at which fallers were
discriminated from non-fallers (Santos et al. 2011). A similar cutoff score of
6 T.A. Niiler
50 was found in a study of 44 community-dwelling adults which also found that the
BBS was better than other measures at predicting falls (Shumway-Cook et al. 1997).
However, in a study of 210 elderly adults, the BBS was capable of discriminating
fallers from non-fallers but only with more advanced analysis of the data using
receiver operating curves rather than a dichotomous cutoff score (Muir et al. 2008). It
should be noted that the BBS has different cutoffs for predicting fallers from
non-fallers dependent on the population being considered.
The TBGA, which is now sometimes known as the Performance Oriented
Mobility Assessment (POMA), was created to address the disconnect between
physical abilities of subjects and the diagnostic laboratory testing that was in place
in the 1980s (Tinetti 1986). Unlike some other tests, the TBGA differentiates balance
testing from gait testing. There are nine balance tasks, many of which are static.
Dynamic balance tasks include rising up, sitting down, and turning 360 . Gait testing
in the TBGA evaluates seven different characteristics of the subjects’ gait, focusing
mainly on symmetry, normalcy, path, and posture, while subjects walk for 25 ft
(Kegelmeyer et al. 2007). In many ways, the TBGA is a very similar test to the BBS,
but where questions overlap, there is often a key difference in the manner of
assessment. For example, both the TBGA test and BBS require subjects to turn
360 . While the BBS assesses performance on this item based mostly on timing
which is quantitative, the TBGA assesses performance based on a rater’s qualitative
judgment as to whether or not steps are continuous or discontinuous. Despite this,
the TBGA has a high reliability and predictive value in some populations. In a study
of 126 Parkinson’s patients, intra- and interrater ICCs ranged from 0.79 to 0.94, and
the positive fall predictive value of the test was 76% (Kegelmeyer et al. 2007). On
the other hand, in a test of 72 elderly adults, sensitivity and specificity of fall
prediction were 62.5% and 66.1%, respectively, indicating marginal predictive
value in this population (Faber et al. 2006).
The DGI is also similar to the BBS and has been validated against it with
moderate correlation in results (Whitney et al. 2003). The DGI is scored using
eight items with a maximum of three points each with higher scores indicating better
performance. Unlike the BBS, most of the items on the DGI are dual motor tasks
rather than cognitive/motor tasks as we will see later. For example, in the DGI,
subjects are asked to walk while turning their head right and then left when signaled
to do so (“gait with horizontal head turns”). In this case, the horizontal movement of
the head is designed to change the visual focus of the subject while maintaining the
vestibular sense and thereby keeps subjects from overreliance on visual cues as an
aid. Tasks such as this are sufficiently involved that even healthy subjects may not
score the full 24 points possible. As a result, it has been noted that the DGI identifies
balance deficit only when subjects score a 19/24 or below (Maranhão-Filho et al.
2011). The reliability of the DGI has been questioned as well. In a study of the DGI’s
reliability using 30 patients with vestibular disorders, Wrisley et al. (2003) found that
the interrater reliability per item on the assessment was poor to good and, as such,
recommended that the DGI be used with caution in vestibular patients. Two raters
were used. Another study of a modified DGI using seven raters and three older adults
with balance problems found that per item interrater reliability was again poor to
excellent (Krishnan et al. 2002). Research supporting the reliability of the DGI also
exists. In a study using 11 raters and 10 individuals with multiple sclerosis, the DGI
was found to have high reliability overall and also on the eight itemized scores
(McConvey and Bennett 2005). Likewise, when the DGI was compared to other
measures of dynamic balance such as the margin of support (discussed below), it was
found to have a moderate and significant correlation implying that despite its deficits,
in some populations, it may be a useful tool to assess dynamic balance (Vistamehr
et al. 2016).
The TUGT tests subjects’ ability to rise from a chair, walk to a line 3 m distant,
turn, return to the chair, and then sit (Podsiadlo and Richardson 1991). Generally,
adults who take longer to complete this test have more impaired balance as exhibited
by an increase in the number of historical falls. Slower gait is taken to be a guarded
behavior which is intended to preserve stability and give subjects a chance to
respond to a crisis of balance. Interpretation of the TUGT depends on the population:
studies will usually seek to identify a cutoff time that discriminates the more capable
subjects from the less capable ones. For example, in a study of 413 community-
dwelling and 78 institutionalized adults, 10% of institutionalized subjects performed
the test in less than 12.7 s, and 10% of community-dwelling subjects performed the
test in greater than 11.2 s (Bischoff et al. 2003). This suggested a rough cutoff of 12 s
to identify gait deficit. Dependent on the population being studied, the rate of falls
sharply increases somewhere between 12 s (Bischoff et al. 2003; Trueblood et al.
2001) and 15 s (Whitney et al. 2005). Of these commonly used methods of
assessments, at the current time, only the TUGT is recommended by the CDC for
dynamic balance assessment (CDC 2016).
Dual-Task Assessments of Balance

As functional tests become more complicated, those with balance impairments have
more difficulties. When elderly adults with balance impairments were surveyed as to
when they fell, it was found that carrying loads, hurrying, or other distracted
behavior led to an increase in the incidence of falling (Berg et al. 1997). A review
of studies involving balance and cognition have indicated that as cognitive function
declines, risk of falling increases (Amboni et al. 2013). When cognitively impaired
older adults were asked to walk and talk and the same time, those who were able to
maintain their gait during the activity were significantly less likely to fall over the
next 6 months than those who stopped for the conversation (Lundin-Olsson et al.
1997), and the test correctly predicted fallers 83% of the time. In application of the
TUGT under dual-task conditions compared to single-task conditions, Silsupadol
et al. (2006) found that all three subjects slowed in their gait taking longer to
complete the test. In larger studies, the dual- versus single-task increase in sensitivity
is maintained. A study of 38 patients with multiple sclerosis (MS) was conducted in
which one group received balance training, while the other group received balance
training while doing cognitive tasks (Monjezi et al. 2016). At the end of the training
period, no difference was seen in TUGT and 10 m timed walk test when performed
as a single task. However, when a cognitive component was added to the test
(counting by threes from a randomly selected number), the group with dual-task
8 T.A. Niiler
training did significantly better than the single-task group on both tests. These
findings have had some influence on the functional testing methodologies described
above, and a number of researchers modified the tests to include a concurrent
cognitive component. As discussed below, dual-task assessment is also now done
in many motion capture labs to better identify balance deficits.
Quantitative Testing of Dynamic Balance
The Need for More Refined Measures

The raw number of itemized balance tests has grown in recent years, partially due to
a sense that they are imperfect tools but also because they seem to capture different
aspects of balance. One test may be more appropriate with a given population than
another test given what is measured. Schlenstedt et al. (2016) noted that only certain
items in their chosen tests correlated with fall risk in their population of patients with
Parkinson’s disease (Schlenstedt et al. 2016). Others have noted floor or ceiling
effects with their tests (Herman et al. 2011; Vereeck et al. 2008; Boulgarides et al.
2003; Garland et al. 1997), lack of sensitivity of tests, or the inability to isolate the
specific part of the balance triad which is impaired. This has lent momentum to the
motion capture researchers who have attempted to characterize dynamic balance, in
particular, based on gait data. Correlations of temporal-spatial parameters such as
stance width, side-to-side symmetry, and timing of stepping with results of itemized
balance tests (Day et al. 2012) have indicated that motion capture may provide some
benefits in quantification of balance in cases where it can be applied. More specif-
ically, motion capture can enable a lab to more precisely assess balance impairment
or the effects of interventions to improve balance.
Pressure-Sensitive Mats
There are a number of commercially available pressure-sensitive mats which are
capable of tracking gait events such as heel strike and toe off, stance width, stride
length, velocity of progression, and timing of gait events. Considering that there is
often increased sway or decreased forward velocity in balance-impaired individuals,
these mats have been used to attempt to quantify various parameters that identify
balance deficits. A correlation between balance and temporal-spatial parameters of
gait was established by Lewek et al. (2014) who studied individuals with chronic
stroke by utilizing the BBS in conjunction with a pressure-sensitive walkway. Weak
to moderate negative correlations between the BBS score and asymmetries in step
length and swing time were found implying that greater asymmetries in gait lead to
decreased balance. Moderate positive correlations between the BBS score and
asymmetries in step width and swing time were also found. Another population
with known balance deficits due to weakness and decreased proprioception are
individuals with multiple sclerosis (MS). In a study of 88 patients with MS and
20 healthy controls, a pressure-sensitive mat was used to measure temporal-spatial
parameters of gait (Socie et al. 2013). When controlling for age disability level as
measured by the Expanded Disability Status Scale correlated positively with
variability in step time and step length, indicating that greater variability in gait was a
marker for disability.
Accelerometers
As technology has improved, triaxial accelerometers have become less expensive
and, therefore, more accessible. They are typically easy to put on, and unobstrusive
to wear, and, as such, do not interfere with the motion being studied. Accelerometers
are capable of detecting gait events including foot strikes and can be used to gauge
velocity and position. These data can be analyzed to detect variability and thereby
identify balance deficit.
A primary test of accelerometers in this role is to relate their data with commonly
used clinical balance tests. In one such study, 21 older patients were recruited to
investigate how the BBS and TUGT related to accelerometry data (O’Sullivan et al.
2009). These data were collected from an accelerometer placed on the torso while
subjects stood on firm ground or soft mats with eyes open or closed. A strong inverse
correlation (r = 0.829) between acceleration data from subjects standing on the
soft mat with eyes open and the BBS score was found. A weaker positive correlation
(r = 0.621) was found between these data and the TUGT time. These results suggest
that balance-impaired subjects had poorer postural control on soft surfaces leading to
higher accelerations. Increased variability of measured acceleration also distin-
guished the fallers from the non-fallers in this study.
Using the accelerometer to identify gait events and thereby quantify increased
variability as a proxy for balance impairment has also been successful. Lamoth et al.
(2011) studied 13 elderly patients with dementia along with 13 similarly aged
healthy controls using dual-task walking for 3 min. Compared with single-task
walking, all subjects in dual-task conditions walked more slowly and with less
variability. However, local stability exponents (see below) were higher indicating
lower stability under dual-task conditions. In intergroup comparisons, the cogni-
tively impaired subjects showed greater variability of gait timing than the healthy
controls.
With the decline in cost of accelerometers, they are now found almost univer-
sally in smart phones. Researchers have now started to make use of them to
measure temporal-spatial parameters in gait. A key issue that must be addressed
is the validity of these sensors compared to dedicated accelerometers. Cell phone-
based sensors may be inconsistent dependent on the manufacturer and model of
phone. In a study of static balance in 49 healthy individuals, an iPad2 was tested
against a NeuroCOM Smart Balance Master for validity (Alberts et al. 2015). The
NeuroCOM is a large machine that has force plates, railing, and a screen which
surrounds the subject. The iPad2 is a small handheld tablet computing device
which is very commonly owned. During testing, the iPad2 was attached to the
subjects at about the level of the COM, while the NeuroCOM administered the
Sensory Organization Test, a static balance test that involves all three parts of
the balance triad. The results indicated that the iPad2 measurements tracked the
NeuroCOM measurements rather closely and thus could be used to replace the
more expensive device for some balance testing. A second study, this one of
10 T.A. Niiler
dynamic balance during gait in Parkinson’s patients, was conducted using the
accelerometer on an iPhone in conjunction with a pressure-sensitive mat for
validation (Ellis et al. 2015). Twelve PD patients and 12 healthy age-matched
controls were enrolled. Regardless of device, results were similar to previous
studies of this population. For self-paced gait, PD patients showed increased
variability of gait parameters compared to the control group and decreased speed
and step length. Where devices showed significant differences in their measures,
the effect sizes were small and did not change the conclusions of the study.
Studies like these are only the beginning of what is likely to be a revolution in our
ability to quantify balance deficits among other things. A brief search of PubMed at
the time of this writing returned over 30 results pertinent to the field of biomechan-
ics. As smart phones are also minicomputers, they are programmable, networked,
and able to communicate via bluetooth with other wearable sensors such as
Samsung’s GearFit 2 or Apple’s Watch making it likely that consumer affordable
sensor arrays for biomechanical data collection cannot be far off.
Motion Capture and Tracking the Variability of the COM

Motion capture-based methods involve the tracking of the walker’s center of mass
using markers and high-speed video systems. It is beyond the scope of this chapter to
discuss such systems in detail, except to say that they are largely accurate to within a
millimeter (Richards 1999) and, as such, are able to relatively precisely estimate the
position of the COM in three dimensions. This means that the vertical excursion of
the center of mass may be monitored along with its horizontal excursion. Large
variations in the center of mass position not only indicate a larger amount of sway,
but also they require larger corrective forces if the walker is to continue on her way in
a consistent manner. Likewise, large variations may be indicative of muscle weak-
ness rather than problems in the balance triad. Weaker muscles take longer to correct
an action than do stronger ones. As such, this manner of assessment may not be able
to deconvolve the strength and endurance issues from those addressed by the balance
triad.
This increase in lateral sway is seen in the gait of balance-impaired populations.
Hsue et al. (2009a) studied dynamic balance in 32 children with cerebral palsy
(CP) as well as 10 typically developing (TD) children using motion analysis. Higher
peak mediolateral and lower anterior-posterior COM displacements were measured
in the CP group compared to the TD controls. A follow-up study with the same
group found that those children with hemiplegic CP had a larger mediolateral
component of COM velocity (Hsue et al. 2009b). A comparison of older adults
having balance deficits (both idiopathic and vestibular) with healthy age-matched
peers demonstrated that the motion of the mediolateral COM was higher in range and
in velocity for those with identified balance problems (Chou et al. 2003). Further
evidence that excess COM motion in gait is an indication of balance impairment
comes from a study of 34 older adults without a history of falling who were asked to
walk under normal and “narrow-base” conditions where the walkway was half the
width of each subject’s ASIS width (Schrager et al. 2008). Motion capture was used
to analyze COM motion. Under narrow-base conditions, mediolateral COM
Fig. 2 COM displacement in COM in AP, ML and vertical directions

gait for typically developing,
Displacement/leg length
TD
hemiplegic CP, and diplegic
1 Hemi
CP in anterior-posterior, Di
mediolateral, and vertical
direction (Adapted from Hsue
0.5 SD(TD)=0.07
et al. 2009b.)
SD(Hemi)=0.08
SD(Di)=0.09
0
0 20 40 60 80 100
SD(TD)=0.02
0.16
SD(Hemi)=0.04
0.14 SD(Di)=0.04
0.12
0.1
0.08
0.06
0 20 40 60 80 100
SD(TD)=0.02
0.98
SD(Hemi)=0.02
0.97 SD(Di)=0.03
0.96
0.95
0.94
0.93
0 20 40 60 80 100
% of modified gait cycle
displacement increased more rapidly with age than it did under normal conditions.
This suggested that increasing age was itself a risk for decreased balance. However,
the mediolateral COM velocity was higher under the narrow-base condition for most
ages, suggesting that this method of testing may have increased potential for
uncovering balance deficits in any population.
Although COM data is typically collected using expensive multi-camera motion
capture systems, recent advances in technology have made it possible to collect
such data on the cheap. The Microsoft Kinect, originally intended as a gaming
platform, has a depth camera as a sensor and software to detect body positioning. It
has already been adapted for use in static balance analysis and validated by a
clinical motion capture system (Yeung et al. 2014). Although errors in quantifica-
tion of sway (about 3 mm) were significantly larger than what is acceptable for
clinical use, the platform is evolving rapidly both in hardware and software.
Another unique use of the device is for providing real-time feedback for balance
training (Clark et al. 2013). While this is, in theory, possible with much more
expensive motion capture systems, the cost of professional systems makes their use
for such an application impractical. The advent and continued development of
12 T.A. Niiler
systems like the Kinect moves us into a new paradigm where balance measure-
ments and interventions are one in the same (Fig. 2).
Quantitative Dual-Task Assessment of Balance: Increased Sensitivity

of Balance Testing
When a cognitive task is added to walking, a number of clinical balance assessments
become more sensitive than their single-task counterparts (Muhaidat et al. 2014;
Silsupadol et al. 2006). It is therefore expected that COM-based motion capture
testing would similarly prove to be more sensitive in identifying balance deficits. In
doing such testing, it is desirable to test balance-impaired versus healthy subjects in
single and dual tasks. Difference scores between the single- and dual-task conditions
are taken for each group to establish the effect of cognitive distraction on the activity
at hand.
One group in whom this testing has become the norm in recent years are athletes
in contact sports who are at risk for concussion. A meta-analysis of ten studies of gait
in concussed athletes has shown that increased mediolateral sway in COM is seen in
concussed subjects both when compared to non-concussed controls and when
compared to pre-concussion gait (Lee et al. 2013). There is evidence this method-
ology can highlight other gait parameters are affected by concussion and thereby
better identify concussed athletes. In a study of 20 concussed adolescents and age-
and sport-matched healthy controls, motion capture was used to track the
mediolateral motion of the COM of subjects who walked while concurrently com-
pleting an auditory Stroop test (Howell et al. 2013). Under these conditions, the
mediolateral deviation of the COM was greater in the concussed group, as was the
peak mediolateral COM velocity. The anterior-posterior COM velocity was signif-
icantly lower in the concussed athletes suggesting a slowing of gait velocity as an
adaptation to balance impairment.
Dual-task assessment using motion capture or pressure mats has also proven to
be more sensitive to detecting balance impairment than single-task assessments.
Using three elderly subjects, Silsupadol et al. (2006) demonstrated some increased
variability in the mediolateral position of the COM during the narrow line and
obstacle crossing gait activities in dual-task versus single-task activities. Using
single- and dual-task assessments of gait, Azadian et al. (2016) studied how
cognitive and gait retraining can improve balance in older adults. Differences in
temporal-spatial parameters were found between single-task and dual-task condi-
tions for all groups. However, groups that received either cognitive dual-task
training or memory, processing speed, and inhibition training scored significantly
closer to the single-task condition than did the control group which received no
training. Dual-task complexity can also play a role in identifying balance impair-
ment. In a study involving 43 mildly cognitively impaired (MCI) subjects and
25 healthy controls, gait variability of stride time was measured using a pressure-
sensitive mat (Montero-Odasso et al. 2012). While gait variability was found to
increase with complexity of the task, it increased significantly faster among MCI
subjects than for controls (Fig. 3).
Fig. 3 Variability of gait variables as indicated by coefficient of variation for single- and dual-task
walking (Adapted from Montero-Odasso et al. 2009. Creative Commons 2.0 Attribution License)
Mathematical Methods: Finding Patterns to Identify Balance

Impairment
In addition to different testing protocols and tools, another way in which balance
deficits have been identified has been via mathematical analysis of data. Such
methods are sometimes based in the physics of balance as with the margin of
stability or normalized dynamic balance index, while others examine the structure
and timing of the data itself to look for clues that identify the differences between
balanced and unbalanced individuals.
Margin of Stability
For balance to be maintained, the projection of the COM (pCOM) must lie within the
base of support. The condition for balanced walking is unclear since the pCOM
exceeds the base of support during gait. Additionally, since the COM is moving, this
motion may lead to overbalancing even if at the moment, pCOM is within the base of
support.
To account for the COM motion, Hof et al. (2005) proposed that the pCOM
should be adjusted for velocity so that the velocity-adjusted pCOM always lies
within the base of support during dynamic balance. This physical behavior is
modeled as an inverted pendulum of length, l, where l is the length from the ankle
14 T.A. Niiler
to the COM. According to Hof, if the pCOM is designated by x and the COP by u,
the differential equation governing this behavior is given by
l €
x
u x ¼ x€ ¼ 2 : (1)
g ωo
where ωo2 = g/l is the angular frequency of pendular motion. Given the constraints
of balance, and assuming an initial COM position xo and initial forward COM
velocity vo, the solution to this equation is
vo
xo þ u (2)
ωo
where the quantity Xcom = xo + vo/ωo is the velocity-adjusted pCOM. This equation
can be interpreted to mean that the velocity-adjusted pCOM must stay behind the
COP. In the event that the motion is backward, the constraints and solution can be
adjusted accordingly. To quantify dynamic balance, one must calculate the differ-
ence between the outermost portion of the base of support (umax) and Xcom. This is
called the margin of stability, b, and is defined by:

vo
b ¼ umax xo þ (3)
ωo
A related measure called the time to contact is an estimation of the maximum time
available for a foot strike before falling ensues. This quantity is equal to the margin
of support divided by the velocity:
b
τ¼ (4)
vo
For maximum stability, both τ and b should be larger.
The margin of stability (MOS) has become a very popular metric of balance in
motion capture labs since it was introduced since it is relatively easy to understand
and apply to kinematic data (Hof et al. 2007, 2010; Rosenblatt and Grabiner 2010;
Carty et al. 2011; Peterson and Horak 2016; van Meulen et al. 2016). Validation
studies have used the MOS in balance-impaired and in control groups to assess the
differences, as well as tested MOS finds with respect to quasi-dynamic qualitative
assessments of balance.
In one validation study, 13 post-stroke patients were instrumented to measure
position, velocity, orientation, and ground reaction force of each foot, and from these
the body COM was determined (van Meulen et al. 2016). Subjects walked along a
10 m walkway twice while timed and also had the BBS assessment completed.
Results indicated that MOS was not correlated to BBS score. However, for subjects
scoring less than 45 on the BBS, Xcom was found to be outside the base of support.
While this is not a strong validation of the MOS as a balance metric, it does indicate
that past a dichotomous cutoff score, the MOS results and BBS results agree. This
may be more a function of the population or the limitations in the BBS than in the
theory behind the MOS.
Another earlier study by Hof et al. (2007) compared the MOS in six above-knee
amputees with six age-matched controls. COP during gait was collected using an
instrumented treadmill, and from this, the MOS was calculated. Gait of amputees
was much more asymmetrical, as could be expected, and while cadence increased in
both groups with speed, it increased less in the amputees. What could be surprising is
that the MOS was larger for the amputees than the controls. The theory of the MOS
leads us to expect that less balanced individuals will have a smaller MOS or an Xcom
which exceeds the base of support. However, in this case, the balance-impaired
individuals show the opposite. This may be guarding behavior in which an adapta-
tion to balance deficit is made to preserve stability. Such results highlight the need
for care in interpretation of results of physical measures such as the MOS (Fig. 4).
Moment of Inertia Ellipsoids and the Interfoot Line

Another interesting approach to increasing the sensitivity of COM-based balance
measures is to examine the variability of the COM in a body-centered versus a
lab-centered coordinate system.
Typically when measurements of mediolateral COM motion are given, they are
made in a laboratory frame: all deviations toward the left or right walls of the lab are
considered separately from deviations toward the front or back walls of the lab.
Despite a fair bit of evidence indicating that an analysis of such a mediolateral
component can discriminate fallers from non-fallers, this component as chosen may
be less sensitive than one which considers the exact body positioning at any given
time. Studies of robotic motion often consider the whole body moment of inertia in
the consideration of dynamic balance. According to the tennis racket theorem,
rotation of a rigid body is more stable about the first and third principle moments
of inertia than about other axes (Ashbaugh et al. 1991). As the full body moment of
inertia ellipsoid changes during gait, it stands to reason that the preferred axis of
rotation will also change accordingly rather than remain in the mediolateral direction
of the lab. Preliminary work in quantifying the body’s principle moments of inertia
has confirmed this and suggests that one principle axis tracks approximately with the
alignment of the feet (Dutta and Goswami 2010). Therefore, foot placement rather
than room alignment may serve as a better reference frame for measuring deviations
in the COM motion. It is also noteworthy that Dutta and Goswami found increased
deviations of the moment of inertia ellipsoid for a spinal cord injured subject
compared with a healthy control.
Further evidence that mediolateral deviations of the COM may not be the ideal
when considering dynamic balance comes from studies involving actual falling. As
noted by Berg et al. (1997), who surveyed 96 male and female subjects between the
ages of 60 and 68, the two primary ways in which falling occurred while walking
were tripping (34%) and slipping (25%). In theory, for tripping, the swing foot
catches and the initial fall is in the forward direction. Similarly, in slipping, it is
expected the forward foot slides out, and the fall will generally be toward the side or
back, dependent on the phase of the gait cycle when the slip occurs. In a study to test
16 T.A. Niiler
a b
CoM
VCoM VCoM
CoM XCoM XCoM
L L
g g
PCoM BSUmax PCoM

BSUmax
MS MS
(positive values) (negative values)
Fig. 4 Relationship of the parameters used in determining the margin of support. The figure on the
left is balanced, while the one on the right is unbalanced. In the unbalanced case, the Xcom exceeds
the base of support (Adapted from McCrum et al. 2014. Creative Commons Attribution License)
fall direction of different modalities of falling, Smeesters et al. (2001) largely

validated these expectations. In slow and normal gait, the majority of slips resulted
in backward or sideways movement 79% and 71% of the time, respectively. Trips
resulted in forward falls 93–100% of the time regardless of gait speed (Smeesters
2001). Other tested modalities of falling such as stepping down or fainting had a
majority of falls in the forward direction. These results imply that while the
mediolateral deviations may be valuable as an indicator of balance impairment,
examination of COM motion in other directions is warranted.
The balance index proposed by Niiler et al. (2007) attempts to account for these
motions by using the positioning of the feet as the basis for measurement. Due to the
relatively large mass of the lower body, the positioning of the feet disproportionately
controls the shape of the moment of inertia ellipsoid during gait. When taken in
COM inter-foot line

left foot
DIFL
pCOM
DIFL right foot
pCOM
Fig. 5 Geometry defining the interfoot line and DIFL. Note that in the current stance, DIFL projects
medially indicating that perturbations in the mediolateral direction would be most effective in
unbalancing the walker (Adapted from Niiler et al. 2017)
combination with the initial conditions of both slipping and tripping wherein both
feet are placed on the ground, the line joining the feet, from midfoot to midfoot,
becomes the relevant coordinate basis. The distance from the projection of the COM
on the ground to this axis (termed the interfoot line) is the effective instantaneous
moment arm to the gravitational torque exerted to effect the fall. If this moment arm
(termed the distance from the interfoot line, DIFL) is within the base of support
defined by the feet, the walker is balanced at that instant. However, when DIFL
exceeds the boundary of the base of support, balance is threatened (Fig. 5).
In typically developing balanced gait, the projection of the COM (pCOM) tracks
near to the interfoot line throughout the gait cycle, even during swing phase. At heel
strike, the pCOM races from the rear foot toward the forward foot within a few tenths
of seconds, and DIFL is relatively small. During swing, the pCOM moves forward
under the support foot until the contralateral foot approaches heel strike, and it
quickly crosses to the forward foot once more. The maximum forward deviation of
the pCOM occurs in mid-swing, and this gravitational moment serves to propel the
walker forward. The minimum deviation is during double support. It is in double
support that DIFL is pointing laterally from the interfoot line. The implications of this
for tripping and slipping are clear. In tripping DIFL points anteriorly in the lab frame,
and in slipping, DIFL points mediolaterally. As such, this change of reference frame
for analysis of COM motion holds clinical potential for analysis of balance within
18 T.A. Niiler
CP vs TD Comparison
1.00
CMS MS
CP
0.50 Slow
Normal
CHS
Fast
DN
0.00
–0.50
0.00 20.00 40.00 60.00 80.00 100.00
% Gait Cycle
Fig. 6 Plot of DN as a percentage of gait cycle for 14 CP subjects compared to 66 typically

developing (TD) subjects walking at self-selected slow, normal, and fast speeds. CMS is contralat-
eral mid-swing, CHS is contralateral heel strike, and MS is mid-swing. Note that DN is very much
smaller for TD subjects regardless of the speed (Adapted from Niiler et al. 2017)
real falling. DIFL is often normalized to half the foot length to ease inter-subject
comparisons. This new parameter is called DN (Fig. 6).
Application of this method to analyzing balance-impaired populations has shown
that it has an increased sensitivity over usual COM methods. In a study of the
dynamic balance of typically developing children compared to those with CP, an
average DN was computed for both groups. The average DN values for children with
CP were nearly nine times that of typically developing children. When compared to
data from a similar study by Hsue et al. (2008), wherein the ratio of CP to TD
mediolateral COM deviation was 1.7, it is clear that DN is more sensitive. When
children with CP have single-event multilevel surgery to correct gait, their DN values
across the gait cycle decrease, on average, by a factor of two bringing them closer to
the typically developing subjects (Niiler et al. 2009). As children develop, the
variability in their stride time decreases (Hausdorff et al. 1999), the coordination
of lower leg muscle control increases, and as a consequence, the variability in COM
motion decreases (Sundermier et al. 2001) and gait becomes more stable. A cross-
sectional study of developmental balance in 66 young children demonstrated that DN
decreased with increasing age, implying increasing stability and reinforcing previous
results (Niiler et al. 2010).
Autocorrelation and Balance

Another approach to the problem of dynamic balance is to define it in terms of the
mathematical stability of the walk. In this way of thinking, those who walk in a more
consistent manner are more stable and, therefore, more balanced. A number of
mathematical tools have been brought to bear to address the consistency of gait as a
proxy to balance. Of these, autocorrelation is, perhaps, the simplest.
In autocorrelation, multi-cyclic data (e.g., of the COM position) is compared with
itself at all time intervals within the data. If data at time point i is given by COM(i), at
some later time point, the data is given by COM(i + j), and the autocorrelated
function at point j is given by
1 X
Nj
ACFðjÞ ¼ COMðiÞCOMði þ jÞ: (5)
N j i¼1
Because this result could have a very large range in values, it is normalized by the
ACF at time zero:
1X N
ACFð0Þ ¼ COMðiÞ2 ; (6)
N i¼1
by dividing
ACFðjÞ
ACFnorm ðjÞ ¼ : (7)
ACFð0Þ
Periodic data as we would see in gait will have a periodic autocorrelation function
(ACF) which varies between one and negative one with the same period as the data.
As noise or perturbations are added to the data, the ACF is attenuated and may
change its periodicity. Individuals with impaired gait have been shown to have very
different ACFs compared with healthy subjects (Fig. 7).
Interestingly, ACFs are generally used with force plate data (Banks et al. 2004;
Stodolka et al. 2016) or accelerometer data (Matsushima et al. 2015) during quiet
standing. Under those circumstances, it has been found that PD patients who suffer
from impaired balance exhibit rapidly declining ACFs compared with healthy
controls, as well as a long rebound phase indicative of attempted recovery and
more variability in general (Banks et al. 2004) (other studies on static balance and
accelerometry?). However, it has also been noted that correlations exist between
features extracted from ACFs of trunk accelerations and the TUGT, implying that the
ACF is also a tool for identifying dynamic balance deficit (Yang et al. 2011).
Additionally, ACF analysis has been used to identify gait asymmetries in trans-
femoral amputees, a population with known balance deficits (Tura et al. 2010).
Lyapunov Analysis of Stability

Building on the method of ACF is that of Lyapunov analysis. In this type of analysis,
a signal such as the mediolateral component of the COM is time delayed and then
examined for divergence. If on average, a signal has a positive divergence, it means
that the movement being analyzed is increasingly different and, therefore, unstable.
The average rate of divergence across the signal is given by the Lyapunov
20 T.A. Niiler
Autocorrelation sequence-Healthy subject

1
–1
0 20 40 60 80 100 120 140 160 180 200
Samples
Autocorrelation sequence-PD subject
1
–1
0 20 40 60 80 100 120 140 160 180 200
Samples
Fig. 7 Difference in ACF between healthy subject and one with PD. Note the more rapid decrease
in ACF in the lower graph (Adapted from Yang et al. 2011. Creative Commons Attribution 3.0)
exponent, λ. Given the distance between the jth neighbors at timestep i, dj(i), a
timestep of Δt, and the initial separation in trajectories, do, the Lyapunov exponent is
defined by the equation (Rosenstein et al. 1993):

ln dj ðiÞ λðiΔtÞ þ lnðd o Þ (8)
A plot of ln(dj(i)) versus iΔt has a slope of λ. In real life data, λ is always positive, so
more mathematically stable walks have smaller exponents. The short-term slope of
this plot is also sometimes known as the local stability exponent (LSE). A number of
tool kits exist in languages such as R and MatLab, for example, to help with such
analysis for those who would embark on such an analysis (Fig. 8).
To validate the value of the LSE to predict falling, Toebes et al. (2012) recruited
134 elderly adults for treadmill walking. Subjects walked for 12–17 min with the
final 7 min recorded by an accelerometer strapped to their backs. Triaxial accelera-
tions and angular velocities of the trunk were recorded. Logistic regression was then
used to relate fall histories and treadmill data. Results indicated that fall history was
positively correlated with mediolateral variability of accelerometer data and LSE of
the combined angular velocity data. This suggested that for fallers, increased local
dynamic instability was at play. It should be noted that all subjects walked at the
same speed on the treadmill, and there was no attempt to adjust speed based on fall
history.
Interestingly, in this sort of analysis, the local stability exponent (LSE) for some
balance-impaired groups is often smaller than for typical controls. As diabetes
advances in severity, patients suffer from peripheral neuropathy leading to decreased
proprioception which, in turn, is thought to affect balance. A study of 14 diabetes
patients and 12 weight- and age-matched controls was conducted to quantify the
stability in the respective groups (Dingwell et al. 2000). Sagittal plane angular
Fig. 8 Origin of divergence exponent in Lyapunov analysis. λs is the local stability exponent (LSE)
(Adapted from Terrier and Dériaz 2011. Creative Commons 2.0 Attribution License)
kinematics of the hip, knee, and ankle and triaxial accelerometer data from the trunk
were analyzed using LSE. LSE values from the affected group were significantly
smaller than for the controls as was the walking speed. This leads the authors to
conclude that the decrease in walking speed served to compensate for the decrease in
balancing ability. While this result may seem contrary to the previous one, the key
difference is that in the later study, subjects walked overground at a self-selected
speed.
While LSE analysis has proven to be another valuable way of assessing balance,
it is not without criticisms. First, in theory, the Lyapunov exponent requires an
infinite number of cycles for true calculation (van Emmerik et al. 2016). In practice
one needs a large number of gait cycles to approximate the LSE. This means that for
certain balance-impaired populations who cannot walk for extended periods, the
method cannot be used. Second, the context of stability must be carefully defined. If
a population falls more, it is generally considered to be more unstable. Yet in the
22 T.A. Niiler
Dingwell et al. (2000) analysis of diabetic neuropathy, controls had higher instability
as indicated by the LSE. Therefore, the conditions of the assessment must be
specified in such a way as to properly evoke instability: for example, by controlling
gait speed as with Toebes et al. (2012). Finally, the occasional misstep or change in
movement might leave the COM over the base of support and, in fact, not result in
increased variability of the COM, yet, according to this analysis, it would increase
the LSE.
Conclusion/Summary
There are a vast array of methods for measuring dynamic balance, ranging from
semi-qualitative scales for functional assessments often used in the physical therapy
world to physical based methods favored by biomechanists and engineers. Most of
them have been validated either via testing against balance-impaired populations
versus healthy controls, testing against fall history, or by testing against each other.
Judging from the number of references that were encountered in literature searches,
the most popular functional assessments are the BBS and TUGT. Similarly, the MOS
and LSE analyses seem to be the numerical methods that are catching on where labs
have the resources to calculate them. The mediolateral variability of gait is a
universally recognized proxy for instability. Yet in some balance-impaired
populations, it is smaller than for healthy controls due to more guarded stepping
strategies. These results indicate that gait that is either too conservative or too
variable should signal to the observer that a balance deficit exists. Strategies to better
identify balance problems have emerged over the years from increasing the difficulty
of functional tasks to implementing dual-task protocols. Mathematical analysis of
gait data continues to become more sophisticated to better tease out clues that
identify balance impairments at earlier times. In the area functional assessments,
updates to the various assessments could be implemented to address some of their
limitations such as floor and ceiling effects. Finally, with the advent of wearable
sensors, it is likely that in the next decades techniques to identify balance deficits will
further mature and gain increasing accessibility in the community.
Cross-References

▶ Concussion Assessment During Gait
▶ Functional Capacity Evaluation Post Injury
▶ Functional Dystonias
▶ Gait and Multiple Sclerosis
▶ Gait Initiation, Turning, and Slopes
▶ Gait Retraining for Balance Improvement

Analysis
▶ Motor Patterns Recognition in Parkinson's Disease
▶ Pre-participation Physical Evaluation in Sport
▶ Slips and Falls
▶ Time Series Analysis in Biomechanics
References
Alberts JL, Hirsch JR, Koop MM, Schindler DD, Kana DE, Linder SM, . . . Thota AK (2015) Using
accelerometer and gyroscopic measures to quantify postural stability. J Athl Train 50
(6):578–588. doi:10.4085/1062-6050-50.2.01
Amboni M, Barone P, Hausdorff JM (2013) Cognitive contributions to gait and falls: evidence and
implications. Mov Disord 28(11):1520–1533. doi:10.1002/mds.25674
Ashbaugh MS, Chicone CC, Cushman RH (1991) The twisting tennis racket. J Dyn Diff Equat
3(1):67–85. doi:10.1007/BF01049489
Azadian E, Torbati HRT, Kakhki ARS, Farahpour N (2016) The effect of dual task and executive
training on pattern of gait in older adults with balance impairment: a randomized controlled trial.
Arch Gerontol Geriatr 62:83–89. doi:10.1016/j.archger.2015.10.001
Banks D, McMorris FR, Arabie P, Gaul W (eds) (2004) Classification, clustering, and data mining
applications. Springer, Berlin/Heidelberg. doi:10.1007/978-3-642-17103-1
Berg K, Wood-Dauphine S, Williams JI, Gayton D (1989) Measuring balance in the elderly:
preliminary development of an instrument. Physiother Can 41(6):304–311. doi:10.3138/
ptc.41.6.304
Berg KO, Wood-Dauphinee SL, Williams JI, Maki B (1992) Measuring balance in the elderly:
validation of an instrument. Can J Public Health 83(Suppl 2):S7–S11
Berg WP, Alessio HM, Mills EM, Tong C (1997) Circumstances and consequences of falls in
independent community-dwelling older adults. Age Ageing 26(4):261–268. doi:10.1093/age-
ing/26.4.261
Bischoff HA, Stähelin HB, Monsch AU, Iversen MD, Weyh A, von Dechend M, . . . Theiler R
(2003) Identifying a cut-off point for normal mobility: a comparison of the timed “up and go”
test in community-dwelling and institutionalised elderly women. Age Ageing 32(3):315–320.
doi:10.1093/ageing/32.3.315
Boulgarides LK, McGinty SM, Willett JA, Barnes CW (2003) Use of clinical and impairment-based
tests to predict falls by community-dwelling older adults. Phys Ther 83(4):328–339
Carty CP, Mills P, Barrett R (2011) Recovery from forward loss of balance in young and older adults
using the stepping strategy. Gait Posture 33(2):261–267. doi:10.1016/j.gaitpost.2010.11.017
Centers for Disease Control (CDC): STEADI Materials for Health Care Providers|STEADI – Older
Adult Fall Prevention|CDC Injury Center (n.d.) http://www.cdc.gov/steadi/materials.html.
Retrieved 10 Aug 2016
Chou L-S, Kaufman KR, Hahn ME, Brey RH (2003) Medio-lateral motion of the center of mass
during obstacle crossing distinguishes elderly individuals with imbalance. Gait Posture 18
(3):125–133
24 T.A. Niiler
Clark RA, Pua Y-H, Bryant AL, Hunt MA (2013) Validity of the Microsoft Kinect for providing
lateral trunk lean feedback during gait retraining. Gait Posture 38(4):1064–1066. doi:10.1016/j.
Day KV, Kautz SA, Wu SS, Suter SP, Behrman AL (2012) Foot placement variability as a walking
balance mechanism post-spinal cord injury. Clin Biomech (Bristol, Avon) 27(2):145–150.
Dingwell JB, Cusumano JP, Sternad D, Cavanagh PR (2000) Slower speeds in patients with diabetic
neuropathy lead to improved local dynamic stability of continuous overground walking. J
Biomech 33(10):1269–1277
Dutta A, Goswami A (2010) Human postural model that captures rotational inertia. Am Soc
Biomech. Retrieved from http://ambarish.com/paper/Dutta_Goswami_ASB2010.pdf
Ellis RJ, Ng YS, Zhu S, Tan DM, Anderson B, Schlaug G, Wang Y (2015) A validated smartphone-
based assessment of gait and gait variability in Parkinson’s disease. PLoS ONE 10(10).
doi:10.1371/journal.pone.0141694
Faber MJ, Bosscher RJ, van Wieringen PC (2006) Clinimetric properties of the performance-
oriented mobility assessment. Phys Ther 86(7):944–954
Garland SJ, Stevenson TJ, Ivanova T (1997) Postural responses to unilateral arm perturbation in
young, elderly, and hemiplegic subjects. Arch Phys Med Rehabil 78(10):1072–1077
Hausdorff JM, Zemany L, Peng C-K, Goldberger AL (1999) Maturation of gait dynamics: stride-to-
stride variability and its temporal organization in children. J Appl Physiol 86(3):1040–1047
Herman T, Giladi N, Hausdorff JM (2011) Properties of the “timed up and go” test: more than meets
the eye. Gerontology 57(3):203–210. doi:10.1159/000314963
Hof AL, van Bockel RM, Schoppen T, Postema K (2007) Control of lateral balance in walking:
experimental findings in normal subjects and above-knee amputees. Gait Posture 25
(2):250–258. doi:10.1016/j.gaitpost.2006.04.013
Hof AL, Vermerris SM, Gjaltema WA (2010) Balance responses to lateral perturbations in human
treadmill walking. J Exp Biol 213(15):2655–2664. doi:10.1242/jeb.042572
Howell DR, Osternig LR, Chou L-S (2013) Dual-task effect on gait balance control in adolescents
with concussion. Arch Phys Med Rehabil 94(8):1513–1520. doi:10.1016/j.apmr.2013.04.015
Hsue B-J, Miller F, Su F-C (2009a) The dynamic balance of the children with cerebral palsy and
typical developing during gait. Part I: spatial relationship between COM and COP trajectories.
Gait Posture 29(3):465–470. doi:10.1016/j.gaitpost.2008.11.007
Hsue B-J, Miller F, Su F-C (2009b) The dynamic balance of the children with cerebral palsy and
typical developing during gait. Part II: instantaneous velocity and acceleration of COM and
COP and their relationship. Gait Posture 29(3):471–476. doi:10.1016/j.gaitpost.2008.11.008
Kegelmeyer DA, Kloos AD, Thomas KM, Kostyk SK (2007) Reliability and validity of the Tinetti
Mobility Test for individuals with Parkinson disease. Phys Ther 87(10):1369–1378.
doi:10.2522/ptj.20070007
Korner-Bitensky N, Wood-Dauphinee S, Teasell R (2006) Best versus actual practices in stroke
rehabilitation: results of the Canadian National Survey. Stroke 37:631
Krishnan LVP, O’Kane KSP, Gill-Body KMP (2002) Reliability of a modified version of the
Dynamic Gait Index – a pilot study. Neurol Rep 26(1):8–14
Lafond D, Duarte M, Prince F (2004) Comparison of three methods to estimate the center of mass
during balance assessment. J Biomech 37(9):1421–1426. doi:10.1016/S0021-9290(03)00251-3
Lamoth CJ, van Deudekom FJ, van Campen JP, Appels BA, de Vries OJ, Pijnappels M (2011) Gait
stability and variability measures show effects of impaired cognition and dual tasking in frail
people. J Neuroeng Rehabil 8:2. doi:10.1186/1743-0003-8-2
Lee H, Sullivan SJ, Schneiders AG (2013) The use of the dual-task paradigm in detecting gait
performance deficits following a sports-related concussion: a systematic review and meta-
analysis. J Sci Med Sport 16(1):2–7. doi:10.1016/j.jsams.2012.03.013
Lewek MD, Bradley CE, Wutzke CJ, Zinder SM (2014) The relationship between spatiotemporal
gait asymmetry and balance in individuals with chronic stroke. J Appl Biomech 30(1):31–36.
doi:10.1123/jab.2012-0208
Lundin-Olsson L, Nyberg L, Gustafson Y (1997) “Stops walking when talking” as a predictor of

falls in elderly people. Lancet (London, England) 349(9052):617. doi:10.1016/S0140-6736(97)
24009-2
Maranhão-Filho PA, Maranhão ET, Lima MA, da Silva MM (2011) Rethinking the neurological
examination II: dynamic balance assessment. Arq Neuropsiquiatr 69(6):959–963. doi:10.1590/
S0004-282X2011000700022
Matsushima A, Kunihiro Y, Hirokazu G, Asuka M, Setsuko M, Katsuya N, Akinori N, Shu-ichi I
(2015) Clinical assessment of standing and gait in ataxic patients using a triaxial accelerometer.
Cerebellum Ataxias 2:9. doi:10.1186/s40673-015-0028-9
McConvey J, Bennett SE (2005) Reliability of the Dynamic Gait Index in individuals with multiple
sclerosis. Arch Phys Med Rehabil 86(1):130–133
McCrum C, Eysel-Gosepath K, Epro G, Meijer K, Savelberg HHCM, Brüggemann G-P,
Karamanidis K (2014) Deficient recovery response and adaptive feedback potential in dynamic
gait stability in unilateral peripheral vestibular disorder patients. Physiol Rep 2(12):e12222. doi:
10.14814/phy2.12222
Monjezi S, Negahban H, Tajali S, Yadollahpour N, Majdinasab N (2016) Effects of dual-task
balance training on postural performance in patients with multiple sclerosis: a double-blind,
randomized controlled pilot trial. Clin Rehabil. doi:10.1177/0269215516639735
Montero-Odasso M, Casas A, Hansen KT, Bilski P, Gutmanis I, Wells JL, Borrie MJ (2009)
Quantitative gait analysis under dual-task in older people with mild cognitive impairment: a
reliability study. J Neuroeng Rehabil 6:35. doi:10.1186/1743-0003-6-35
Montero-Odasso M, Muir SW, Speechley M (2012) Dual-task complexity affects gait in people
with mild cognitive impairment: the interplay between gait variability, dual tasking, and risk of
falls. Arch Phys Med Rehabil 93(2):293–299. doi:10.1016/j.apmr.2011.08.026
Muhaidat J, Kerr A, Evans JJ, Pilling M, Skelton DA (2014) Validity of simple gait-related dual-
task tests in predicting falls in community-dwelling older adults. Arch Phys Med Rehabil
95(1):58–64. doi:10.1016/j.apmr.2013.07.027
Muir SW, Berg K, Chesworth B, Speechley M (2008) Use of the Berg Balance Scale for predicting
multiple falls in community-dwelling elderly people: a prospective study. Phys Ther 88
(4):449–459. doi:10.2522/ptj.20070251
Muir JW, Kiel DP, Hannan M, Magaziner J, Rubin CT (2013) Dynamic parameters of balance
which correlate to elderly persons with a history of falls. PLoS One 8(8):e70566. doi:10.1371/
journal.pone.0070566
Niiler T, Miller F, Marchesi S, Henley J (2007) Comparison of CP and normal gait using a kinematic
balance index. In: Abstracts of the 16th annual meeting of ESMAC (European Society of
Movement Analysis for Adults and Children), Athens, 27–29 Sept 2007. Gait Posture 26
(Suppl 1):S29–S30.
Niiler T, Henley J, Marchesi S, Miller F (2009) Rectus transfer patients show improved dynamic
balance post-surgically. In Proceedings of the 14th annual meeting of the gait and clinical
movement analysis society (GCMAS), Denver, 10–13 Mar 2009
Niiler T, Henley J, Miller F (2010) Development of dynamic balance in adolescence. In: Pro-
ceedings of the Joint ESMAC and GCMAS (JEGM) conference, Miami, 12–15 May 2010
Niiler T, Henley J, Miller F (2017) Center of mass perturbation as a normalizable estimate of
dynamic balance in gait: an application comparing typically developing children with spastic
cerebral palsy. arXiv:1701.00507 [Physics]. Retrieved from http://arxiv.org/abs/1701.00507
O’Sullivan M, Blake C, Cunningham C, Boyle G, Finucane C (2009) Correlation of accelerometry
with clinical balance tests in older fallers and non-fallers. Age Ageing 38(3):308–313.
doi:10.1093/ageing/afp009
Peterson DS, Horak FB (2016) Effects of freezing of gait on postural motor learning in
people with Parkinson’s disease. Neuroscience 334:283–289. doi:10.1016/j.
neuroscience.2016.08.017
elderly persons. J Am Geriatr Soc 39(2):142–148. doi:10.1111/j.1532-5415.1991.tb01616.x
26 T.A. Niiler
Richards JG (1999) The measurement of human motion: a comparison of commercially available

systems. Hum Mov Sci 18(5):589–602. doi:10.1016/S0167-9457(99)00023-8
Rosenblatt NJ, Grabiner MD (2010) Measures of frontal plane stability during treadmill and
overground walking. Gait Posture 31(3):380–384. doi:10.1016/j.gaitpost.2010.01.002
Rosenstein MT, Collins JJ, DeLuca CJ (1993) A practical method for calculating largest Lyapunov
exponents from small data sets. Phys D 65:117–134
Santos GM, Souza ACS, Virtuoso JF, Tavares GMS, Mazo GZ (2011) Predictive values at risk of
falling in physically active and no active elderly with Berg Balance Scale. Brazilian J Phys Ther
15(2):95–101. doi:10.1590/S1413-35552011000200003
Schlenstedt C, Brombacher S, Hartwigsen G, Weisser B, Möller B, Deuschl G (2016) Comparison
of the Fullerton Advanced Balance Scale, Mini-BESTest, and Berg Balance Scale to predict falls
in Parkinson disease. Phys Ther 96(4):494–501. doi:10.2522/ptj.20150249
Schrager MA, Kelly VE, Price R, Ferrucci L, Shumway-Cook A (2008) The effects of age on
medio-lateral stability during normal and narrow base walking. Gait Posture 28(3):466–471.
doi:10.1016/j.gaitpost.2008.02.009
Shumway-Cook A, Baldwin M, Polissar NL, Gruber W (1997) Predicting the probability for falls in
community-dwelling older adults. Phys Ther 77(8):812–819
Silsupadol P, Siu K-C, Shumway-Cook A, Woollacott MH (2006) Training of balance under single-
and dual-task conditions in older adults with balance impairment. Phys Ther 86(2):269–281
Smeesters C, Hayes WC, McMahon TA (2001) Disturbance type and gait speed affect fall direction
and impact location. J Biomech 34(3):309–317. https://doi.org/10.1016/S0021-9290(00)00200-1
Socie MJ, Motl RW, Pula JH, Sandroff BM, Sosnoff JJ (2013) Gait variability and disability in
multiple sclerosis. Gait Posture 38(1):51–55. doi:10.1016/j.gaitpost.2012.10.012
Springer BA, Marin R, Cyhan T, Roberts H, Gill NW (2007) Normative values for the unipedal
stance test with eyes open and closed. J Geriatr Phys Ther 30(1):8–15
Stodolka J, Golema M, Migasiewicz J (2016) Balance maintenance in the upright body position:
analysis of autocorrelation. J Hum Kinet 50(1). doi:10.1515/hukin-2015-0140
Stuster J (2006) Validation of the standardized field sobriety test battery at 0.08% blood alcohol
concentration. Hum Factors 48(3):608–614
Sundermier L, Woollacott M, Roncesvalles N, Jensen J (2001) The development of balance control
in children: comparisons of EMG and kinetic variables and chronological and developmental
groupings. Exp Brain Res 136(3):340–350
Terrier P, Dériaz O (2011) Kinematic variability, fractal dynamics and local dynamic stability of
treadmill walking. J Neuroeng Rehabil 8:12. doi:10.1186/1743-0003-8-12
Tinetti ME (1986) Performance-oriented assessment of mobility problems in elderly patients. J Am
Geriatr Soc 34(2):119–126. doi:10.1111/j.1532-5415.1986.tb05480.x
Toebes MJP, Hoozemans MJM, Furrer R, Dekker J, van Dieën JH (2012) Local dynamic stability
and variability of gait are associated with fall history in elderly subjects. Gait Posture 36
(3):527–531. doi:10.1016/j.gaitpost.2012.05.016
Trueblood PR, Hodson-Chennault N, McCubbin A, Youngclarke D (2001) Performance and
impairment-based assessments among community-dwelling elderly: sensitivity and specificity.
Issues Aging 24:2–6
Tura A, Raggi M, Rocchi L, Cutti AG, Chiari L (2010) Gait symmetry and regularity in trans-
femoral amputees assessed by trunk accelerations. J Neuroeng Rehabil 7:4. doi:10.1186/1743-
0003-7-4
van Emmerik REA, Ducharme SW, Amado AC, Hamill J (2016) Comparing dynamical systems
concepts and techniques for biomechanical analysis. J Sport Health Sci 5(1):3–13. doi:10.1016/
j.jshs.2016.01.013
van Meulen FB, Weenk D, Buurke JH, van Beijnum B-JF, Veltink PH (2016) Ambulatory
assessment of walking balance after stroke using instrumented shoes. J Neuroeng Rehabil 13.
doi:10.1186/s12984-016-0146-5
Vereeck L, Wuyts F, Truijen S, de Heyning PV (2008) Clinical assessment of balance: normative
data, and gender and age effects. Int J Audiol 47(2):67–75. doi:10.1080/14992020701689688
Vistamehr A, Kautz SA, Bowden MG, Neptune RR (2016) Correlations between measures of
dynamic balance in individuals with post-stroke hemiparesis. J Biomech 49(3):396–400.
doi:10.1016/j.jbiomech.2015.12.047
Whitney S, Wrisley D, Furman J (2003) Concurrent validity of the Berg Balance Scale and the
Dynamic Gait Index in people with vestibular dysfunction. Physiother Res Int 8(4):178–186
Whitney JC, Lord SR, Close JCT (2005) Streamlining assessment and intervention in a falls clinic
using the timed up and go test and physiological profile assessments. Age Ageing 34
(6):567–571. doi:10.1093/ageing/afi178
Wrisley DM, Kumar NA (2010) Functional gait assessment: concurrent, discriminative, and
predictive validity in community-dwelling older adults. Phys Ther 90(5):761–773.
doi:10.2522/ptj.20090069
Wrisley DM, Walker ML, Echternach JL, Strasnick B (2003) Reliability of the Dynamic Gait Index
in people with vestibular disorders. Arch Phys Med Rehabil 84(10):1528–1533. doi:10.1016/
S0003-9993(03)00274-0
Yang C-C, Hsu Y-L, Shih K-S, Lu J-M (2011) Real-time gait cycle parameter recognition using a
wearable accelerometry system. Sensors (Basel, Switzerland) 11(8):7314–7326. doi:10.3390/
s110807314
Yeung LF, Cheng KC, Fong CH, Lee WCC, Tong K-Y (2014) Evaluation of the Microsoft Kinect as
a clinical assessment tool of body sway. Gait Posture 40(4):532–538. doi:10.1016/j.
Slip and Fall Risk Assessment
Feng Yang
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Slip and Fall Risk Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Analytical Aspect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Empirical Aspect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Summary and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Abstract
The cumulative effect of falls on older adults and on the healthcare system is
immense; the results are debilitating injuries, loss of independence, and transfer to
a healthcare institution, or even death. Slip-related falls contribute up to 40 % of
all outdoor falls in seniors, which often cause serious injuries such as hip fractures
and high economic burden. Therefore, it is pressing to develop fall prevention
interventions to prevent falls from happening. To identify individuals with high
risk of falls could be equivalently important to the development of fall prevention
paradigms because it allows the limited resources assigned to fall prevention to be
directed to those who truly need the interventions. It is thus essential to establish
accurate and effective fall risk assessment tools to identify those with elevated
risk of falls. This chapter intends to review the approaches which can be applied
in research or clinics to evaluate one’s risk of falls after a slip. Dynamic gait
stability and lower limb support have been identified as two key factors leading to
slip-initiated falls. In addition, muscle weakness was associated with high risk of
slips and falls. Other measures including dynamic stability based on linear and
F. Yang (*)
Department of Kinesiology, The University of Texas at El Paso, El Paso, TX, USA
e-mail: fyang@utep.edu

DOI 10.1007/978-3-319-30808-1_45-1
2 F. Yang
nonlinear dynamics and gait parameters could also play a certain role in
predicting a fall initiated by a slip. These measurements provide some useful
and objective tools for the early identification of individuals with a higher risk of
falling, which is important for effective deployment of fall prevention
interventions.
Keywords
Fall prevention • Muscle strength • Muscle power • Dynamic stability • Limb
support • Variability • Balance loss
Introduction
Falls among the growing population of elderly present a significant medical and
socioeconomic challenge (Tinetti 2003; Heinrich et al. 2010), affecting not only the
frail or impaired but the active and vigorous (Rubenstein et al. 1994; Morley 2002;
Rubenstein and Josephson 2006). Approximately one third of persons aged 65 or
older report one or more falls each year (Tinetti 2003). Falls are the leading cause of
injury-related hospitalization and death in this population (Zijlstra et al. 2007;
Heinrich et al. 2010; Kannus et al. 1999). Even if physical injury is avoided, the
incident of falls can often induce a reduction in activity arising from a heightened
fear of falling that can, in turn, lead to nursing home admittance, loss of indepen-
dence, and greatly diminished life quality (Tinetti 1986; Hausdorff et al. 1997;
Zijlstra et al. 2009; Rubenstein and Josephson 2006). Fall-related injuries are also
pricey (Heinrich et al. 2010); the cost of fall-related injuries is expected to reach
$43.8 billion by 2020 (Stevens et al. 2006).
Slips cause approximately 30–40 % of outdoor falls among community-living
adults (Luukinen et al. 2000). Backward falls initiated by forward slips are partic-
ularly dangerous because they frequently cause hip fractures (Smeesters et al. 2001;
Parkkari et al. 1999). In this connection, an estimated 81–98 % of hip fractures are
the result of falls per year (Parkkari et al. 1999; Hedlund and Lindgren 1987). These
fractures require surgical intervention and extensive postsurgical management,
which comes at great cost to patients, families, and the healthcare system (Hayes
et al. 1996; Grisso et al. 1991; Englander et al. 1996; Heinrich et al. 2010).
Because of the many costly social and economic consequences of falls, it is
imperative to prevent or reduce falls among older adults (El-Khoury et al. 2013).
One perspective to approach this goal is to develop effective and efficient fall
prevention protocols; the other would be to identify appropriate populations to
whom the interventions and resources should target with priority. Such a screening
process permits the identification of high-risk populations based on known risk
factors. Therefore, accurate, objective, and effective tools to assess individuals’
risk of falls are highly desired. To develop efficient fall risk assessment instruments
and scales, a critical step is to identify the risk factors of falls and to understand how
and to what degree each factor contributes to falls among older adults.
Slip and Fall Risk Assessment 3
Falls are a complex, multifactorial phenomenon resulting from interactions

among over 400 known intrinsic and extrinsic risk factors (Heinrich et al. 2010).
The intrinsic factors consist of some unalterable factors such as age and gender
(Tinetti 1986), ethnicity (Nevitt et al. 1989), and previous falls (Nevitt et al. 1989) as
well as factors that are somewhat modifiable with targeted interventions like muscle
strength (Moreland et al. 2004; Horlings et al. 2008), gait and balance problems
(Muir et al. 2010; Berg et al. 1997), reaction time (Lajoie and Gallagher 2004),
vision (Heinrich et al. 2010), cognitive function (Muir et al. 2012), fear of falling
(Berg et al. 1997; Koepsell et al. 2004), etc. The extrinsic factors mainly include the
potential hazards from the environment or surroundings, such as slippery or uneven
surfaces, poor stair design, improper use of assistive device, etc. (Rubenstein 2006).
The majority of real-life outdoor falls happen during gait (Berg et al. 1997), and
walking is one of the most fundamental dynamic tasks in daily living. Gait instability
(quantified by either dynamic stability or variability) and inadequate lower limb
support have been identified as major risk factors of falls, particularly among
geriatric populations (Lockhart and Liu 2008; Yang and Pai 2014; Hamacher
et al. 2011; Toebes et al. 2012; Yang et al. 2009). Additionally, muscle weakness
was associated with risk of falls in older adults (Moreland et al. 2004; Horlings
et al. 2008; Han and Yang 2015; Ding and Yang 2016). The muscle weakness could
be related to the insufficient lower limb support during the recovery from a slip (Pai
et al. 2006; Ding and Yang 2016). Since gait instability and muscle weakness could
be modifiable by designing specific training strategies, this chapter primarily covers
them although considerable number of risk factors can possibly contribute to falls.
Given the severe consequence and prevalence of slip-precipitated falls, the fall
incidences initiated by a slip is the concentration of this chapter. Depending on
when a slip occurs during a gait cycle, a slip perturbation could be either forward or
backward. A backward slip typically results in a forward fall, similar to a tip
perturbation in gait. This chapter is mainly devoted to forward slips. The principal
purpose of this chapter is to review the currently available slip-related fall risk
evaluation approaches in terms of gait instability and muscle weakness.
State of the Art
Effective interactions between the neuromuscular-skeletal system and the environ-

ment enable an individual to generate necessary ground reaction force (GRF) for
maintaining balance during movements. Slips usually occur when the friction force
between the footwear and the walking surface is insufficient to prevent the relative
motion between them (Redfern et al. 2001). After a slip is initiated, one’s leading or
slipping foot accelerates forward and results in a significant forward displacement
with respect to the trunk. Under the action of gravity, the body experiences a
backward rotating angular moment around the leading ankle joint. To counteract
or neutralize such an angular momentum, one must react to the perturbation in a
timely manner and regulate the angular impulse to avoid a materialized fall. The
4 F. Yang
reaction strategies refer to the body’s inherent corrective responses, mediated by the
central nervous system, attempting to attenuate the backward rotating angular
movement, maintain the upright posture, and thus resume the forward progression.
The reactive responses include the effort on the slipping limb to bring the body’s
center of mass (COM) close to the base of support (BOS) to regain the balance in the
horizontal direction and the effort on the trailing limb to stepping quickly to
reestablish the BOS, stop the descent of the body, and reverse the falling in the
vertical direction. As a result, the generated GRF can propel and redirect the COM
into the desired direction of progress. Therefore, the ability to reliably and safely
regain stability and retard the body descent is critical to prevent a fall. The quanti-
fication of stability is an important component to investigate how stability can be
restored after a slip perturbation. Also, sufficient muscle strength and the capability
of producing muscle strength quickly (i.e., the muscle power), which are essential to
provide sufficient limb support, are both crucial to a successful recovery stepping.
Extensive theoretical and empirical efforts have been dedicated to the examina-
tion of how dynamic stability, limb support, and muscle weakness lead to falls after a
slip in gait. Advanced understandings have been obtained regarding the effects of
those risk factors of falls on increasing the likelihood of slip-related falls. A theory
upon which body balance and stability can be quantified under dynamic situation has
been established using computer simulation assisted by dynamic optimization. By
incorporating the limb support, this theory has been further advanced to quantify the
resistance against a backward fall commonly induced by a slip. On the other hand,
numerous empirical experiments were undertaken to inspect the roles of muscle
weakness in slip-related falls.
Slip and Fall Risk Assessment
As stated above, the evaluation of risk of falls in response to a slip during gait has
been approached from two perspectives: analytical and empirical. These two aspects
work in a complementary way. Specifically, the former provides theoretical founda-
tions for the design and analysis of the experimental work, while the latter offers
solid verifications to the former one.
Analytical Aspect
The Feasible Stability Region Theory

Stability during human locomotion is defined as the ability to restore or maintain the
upright posture without altering the existing BOS when encountering an external or
internal disturbance to the movement (Hamacher et al. 2011). Bipedal human motion
is inherently unstable due to the human body’s multi-link inverted pendulum
structure with a small BOS but a high COM (Maki and McIlroy 1997). Stability
limits are conventionally defined solely by the position of one’s COM with respect to
the BOS (Borelli 1680; O’Sullivan 2007), meaning that one has to confine the
projection of the COM within the BOS to maintain a stable upright posture while
standing still (Fig. 1a, b, point A in Fig. 1b) (Borelli 1680). However, this conven-
tional theory might not be applicable to dynamic situations, like walking in which
individuals volitionally displace their projection of COM outside the feasible stabil-
ity region to achieve mobility (Fig. 1c). For instance, the COM is behind the BOS at
liftoff during gait as marked by point B (in Fig. 1b) that is posterior to the rear edge
of the BOS. Based on the classical stability theory, the person must lose balance
since the COM is outside the BOS. No one loses balance backward during regular
walking which seems violating the prediction from the classic stability theory.
Significant efforts have been spent to study the dynamic balance condition during
human gait. Only recently has this dominant concept been expanded to dynamic
conditions (like walking) by demonstrating that one’s control of COM velocity is as
essential for stability as the control of one’s COM position relative to the BOS (Pai
and Patton 1997; Mille et al. 2003; Hof et al. 2005). To quantify dynamic stability,
the feasible stability regions (FSR) aiming at human regular gait (Pai and Patton
1997; Yang et al. 2007; Hof et al. 2005) and gait-slip (Pai and Patton 1997; Yang
et al. 2008a) (Fig. 1d) have been established by using computer simulation and
modeling technique based on the constraints imposed by human’s anatomical and
physiological limitations as well as on the interactions between human body and
environment. Dynamic gait stability can be characterized by the relationship
between the body’s COM motion state (i.e., the combination of COM position and
velocity related to the BOS) and the analytically derived dynamic stability FSR
(Fig. 1d).
The dynamic FSR is comprised of two limits: the limit against backward balance
loss (the thick solid line in Fig. 1d) and the one against forward balance loss (the
thick dashed line). These two stability limits encompass all possible motion states of
the COM relative to the BOS which ensure a person to keep the balanced upright
body posture during gait. If the COM motion state is within the FSR region enclosed
by the two limits (e.g., point B in Fig. 1d), dynamic stability, calculated as the
shortest distance from the COM motion state to the limit against backward balance
loss (s in Fig. 1d), takes a positive value meaning the person is stable. This person is
not obligated to alter the existing BOS to maintain balance (Hof et al. 2005).
Based on this generalized framework, when the COM motion state is below the
FSR (or the limit against backward balance loss, e.g., point C in Fig. 1d), the COM
has no sufficient forward momentum to carry it over the BOS when its velocity
diminishes. The person is instable against backward balance loss since he/she must
take a backward compensatory step to keep the body from falling backward. Thus,
the person encounters a backward balance loss. In this case, the person takes a
negative dynamic stability value, representing instability against backward loss of
balance or falling. This is what typically occurs after a forward slip perturbation in
gait. Conversely, a person whose COM motion state is above the FSR (or the limit
against forward balance loss), the COM possesses excessive forward momentum
that would move the COM anteriorly beyond the BOS when its movement termi-
nates; thus, a forward step becomes inevitable resulting in a forward balance loss. In
this scenario, the person is instable against forward balance loss. A forward recovery
6 F. Yang
a Standing c Walking
: Center of mass (COM)

BOS: Base of support
STD: Slipping foot touchdown
RLO: Recovery foot liftoff
RTD: Recovery foot touchdown
b Feasible stability region - static d Feasible stability region - dynamic

0.6
A B
COM velocity
B
Anterior
0.3
0 C
STD
RLO
-0.3 RTD
-2 -1 0 1
BOS BOS
COM position COM position
Fig. 1 Feasible stability region (FSR) under static and dynamic conditions. (a) When a person
stands still, only if his or her projection of the center of mass (COM) is within the base of support
(BOS) which consists of the outline area of each foot in contact with the ground and the area between
the feet, he or she can maintain a balanced or stable state. (b) The shaded section indicates the
feasible stability region (FSR) for the static condition. Whenever the body COM is located within
this region (such as point A), the person is assuming a stable posture. (c) At liftoff during human gait,
no backward balance loss occurs although the COM is behind the BOS (point B in subfigure b). This
disobeys the static feasible stability region theory, implying that the classical static theory is
inapplicable to dynamic conditions. The FSR theory has been recently advanced to dynamic
situation, particularly walking. (d) Schematic illustration of the dynamic FSR, which is surrounded
by two boundaries: the limit against backward balance loss (the lower boundary) and the one against
forward balance loss (the upper boundary) (Yang and Pai 2014). The stability measurement (s, the
length of the thin solid lines) indicates the magnitude of the instantaneous stability of the COM
against backward balance loss and is defined as the shortest distance from the COM motion state (i.e.,
the x-coordinates represent the COM anteroposterior position, and the positive y-coordinates indicate
its forward velocity) to the limit against backward balance loss. When the instantaneous COM
motion state is below/above this limit, the stability value is negative/positive, respectively, denoting
unstable/stable state against backward balance. For instance, COM motion state B is stable, while C
is unstable. Also shown is a representative trajectory (the thin dashed line) depicting the COM
motion state from slipping (or leading) foot touchdown (STD), through recovery (or trailing) foot
liftoff (RLO), to the instant of its touchdown (RTD) in a falling responding to a sudden and
unrehearsed overground slip. Position and velocity of the COM relative to the BOS are dimension-
less as a fraction of the foot length and the square root of the product of the gravitational acceleration
and body height
step becomes necessary to keep the balance, that is what happens when stepping
forward during normal human gait. Although this dynamic FSR was initially derived
by a 2-link model (Pai and Patton 1997; Hof et al. 2005) and further confirmed and
extended by a complicated 7-link gait model (Yang et al. 2007, 2008a), it has been
extensively verified with experimental data in a range of task conditions (Patton
et al. 2000; Wang et al. 2011, 2012; Yang et al. 2008a). Modeling, experimental, and
observational studies support that the threshold against backward balance could
reach a high sensitivity of predicting backward balance loss trials (87.5 %) and a
perfect specificity of predicting non-loss of balance trials (100 %) among 927 trials
collected from 99 young subjects who were exposed to a novel and unannounced
slip produced by a movable platform during overground walking (Yang et al. 2008a).
Instability, representing an inability to regulate or terminate the horizontal motion
of the body COM over the BOS, will lead to a fall unless stability is regained through
an adjustment in BOS, typically by a protective stepping. Dynamic gait stability,
measured based on this theoretical framework, has been identified as a major factor
associated with slip-related falls (Yang et al. 2009). Quantifications of dynamic gait
stability during regular walking have also been shown to be able to predict an
impending slip-related fall better than some motor performance-based clinical eval-
uations of the fall risk, such as Berg Balance Scale test, Timed-Up-and-Go test, and
postural sway (Bhatt et al. 2011). These motor performance tests are generally not
capable of providing a quantitative and sensitive prediction owing to their ceiling
effect or insensitivity (Haines et al. 2008). Therefore, dynamic gait stability based on
the dynamic FSR theory could be a promising assessment means for the risk of slip-
related falls. The detailed calculation method of dynamic gait stability can be found
elsewhere (Yang et al. 2008b).
The Limit of Recovery

The dynamic gait stability quantifies the COM-BOS relationship in the horizontal
direction. However, falls mainly occur in the vertical direction resulting from vertical
descent of the body although a loss of balance due to instability initiates a fall (World
Health Organization 2007). Such a descent could be attributed to two reasons: the
instability and insufficient lower limb support (Fig. 2a). The limb support (Ship) here
is characterized by the quotient of hip vertical velocity to hip height (Fig. 2c) (Yang
et al. 2011). The instability entails that the COM is behind the BOS. Thus, the
backward-tilted body around the ankle joint results in the descent of the COM. Limb
collapse due to insufficient strength or inability of producing strength quickly at
lower limbs would also lead to the descent of the body. The limb collapse mainly
takes place in the fashion of knee buckling (Pavol and Pai 2007). The resulting
descent from limb collapse would be more significant than the one from the
backward leaning body (Yang et al. 2011, 2012).
To elucidate the biomechanical causations of falls from both the dynamic gait
stability and limb support, well-controlled slip perturbations have been created in a
protective laboratory environment to successfully and safely reproduce inadvertent
falls during gait (Yang et al. 2009; Wang et al. 2011). After a slip, both dynamic gait
stability and limb support deteriorate severely (Fig. 2b, c). The inability to provide
8 F. Yang
a Stick figure - fall Slipping side d Limit of recovery

Recovery side
Limb support (Ship)

0
STD RLO RTD Fall -0.1

Limit
b Stability - horizontal Fall (n=11)
-0.2 Recovery (n=58)
-0.1
Stability (s)
-1.6 -1.2 -0.8 -0.4 0

-0.3 STD Stability
RLO
RTD
e Probability of fall
-0.5
Probability of fall (%)

100
c Limb support - vertical 80
Limb support (Ship)
0.06 60
5
- 0.2
40 .22s
= -0
0 S hip
20
0
-0.06 Hip vertical velocity -0.3
Type I error
S = -0.2
Hip height
-0.1
-0.12 0
-0.8 -0.4 0
0.1 -1.6 -1.2
0 50 100
(STD) Time (%) (Fall)
Fig. 2 (a) Stick figures showing the sequence of body configuration at four characteristic instants
of time during falling after an unannounced slip induced during gait over ground. The four instants
include the slipping foot touchdown (STD), the recovery foot liftoff (RLO), its touchdown (RTD),
and fall. The close and open triangles mark the location of the movable platform at STD for the
slipping and recovery side, respectively. The actual slip occurs approximately 30 ms later than STD.
Time history of (b) dynamic gait stability and (c) limb support of the trial shown in (a). The stability
(s) is calculated as the shortest distance from the center of mass (COM) motion state to the limit
against backward balance loss (Fig. 1d). Limb support (Ship) is quantified as the quotient of hip
vertical velocity to hip height.(d) Relationship between COM stability (x-coordinate), the
corresponding limb support (y-coordinate) at the instant of RTD, and the gait-slip outcome.
Individuals within the shaded area would be classified as likely to fall by the logistic regression
1
model: pðfallÞ ¼ 1þe22:80þ20:01sþ90:35Ship
( p < 0.05 for all model coefficients). The limit of recovery (the
thick solid line) is derived based on above logistic regression equation by assigning p(fall) = 0.5. It
is Ship = 0.22s – 0.25. The boundary correctly classified 100 % of fall incidences. The thin line
illustrates an example of the shortest distance from the instantaneous s-Ship state to the limit against
backward falls. (e) Surface plot of the logistic regression that can best predict the likelihood of falls
following a sudden, unannounced slip during gait as function of its two best determinants: s and Ship
at the instant of RTD. Also shown is the boundary (the outline of the limit of recovery) formed at the
middle (50 %) bisecting the surface. The projection of the boundary on the bottom plane (i.e., the s-
Ship domain) is also demonstrated, which is the boundary demonstrated in (d). One recovery trial
(the open diamond) was falsely classified as a fall by the limit being a type I or α error (Yang
et al. 2011)
timely and sufficient limb support due to deficient amount of upward impulse
generated from GRF (Yang et al. 2009) can cause limb collapse. Examination of
the interdependence and the tradeoffs between stability and limb support has been
the focus of several studies (Yang et al. 2009, 2011; Pavol and Pai 2007). Together,
instability of the body COM and inadequate limb support at recovery step touch-
down predict the subsequent slip outcome (recovery or fall occurring ~500 ms later)
of gait-slip with high sensitivity (100 %) and specificity (98.3 %) among 69 young
adults (age: 25.8 4.5 years) who were exposed to an unannounced overground
slip (Fig. 1d, e) (Yang et al. 2011).
On the stability-limb support phase space (Fig. 2d), when the combination of
COM stability and limb support at recovery foot touchdown is below the limit of
recovery, the slip-related fall resistance index (r in Fig. 2d) becomes negative; the
person would have fallen inadvertently onto the ground without the protection of the
harness. By contrast, the fall resistance would be positive, meaning this individual
could recover from a slip, when the combination of COM stability and limb support
is above the limit of recovery. The limit of recovery derived in the s-Ship domain
explains the reason of slip-initiated falls from biomechanical standpoint (i.e., due to
the excessive instability that coupled with inadequate limb support quotient). The
limit of recovery quantifies the stability that must be provided in conjunction with an
adequate amount of limb support to prevent a slip-related fall from being material-
ized. A decrease in either the stability or limb support leads to the increase of the
probability of falls post slip (Fig. 2e). This boundary also quantifies the trade-off that
may take place between the control of stability and limb support where a greater
stability can compensate for an insufficiency in limb support (Fig. 2d, e). This theory
of falls provides us a novel way to predict an actual slip-related fall in the controlled
laboratory settings. The slip-related fall resistance index r offers us a gold standard to
characterize the risk of a slip-related fall induced in gait.
Empirical Aspect
Muscle Weakness Is Associated with Slip-Related Falls

Though it is well accepted that aging is associated with a progressive decline in
overall mechanical muscle function (Goodpaster et al. 2006; Skelton et al. 2002) and
muscle weakness has been implicated as a risk factor for falls, sound evidence
regarding the extent to which muscle weakness directly and independently relates
to falls, particularly slip-related falls in older adults, is largely lacking. A recent
review article suggested that the evidence for the cause and effect link between
muscle function and balance performance in older population is still weak (Orr
2010).
Several possible reasons could be responsible for this. First, the traditional self-
report method collecting real-life falls utilized in previous studies is subjected to
10 F. Yang
imprecision, bias, and omission resulting from deteriorated memory or cognitive

dysfunction in seniors (Moreland et al. 2004; Jenkins et al. 2002). Second, the
physical activity level and the exposure to possible fall hazards are factors affecting
the likelihood of falls in older adults. There could be a trade-off between the
exposure to fall hazards and the risk of falls among older adults (Horlings
et al. 2008). The self-report method does not account for these factors, possibly
leading to underestimation of actual fall counts (Graafmanc et al. 2003; Wijlhuizen
et al. 2010). Third, the self-reported data often lack information on the specific
details (like types and circumstances of falls) of the actual falls (Feldman and
Robinovitch 2006), which could vary considerably from person to person. Without
considering or controlling for the circumstances of falls and level of the exposure to
fall hazards, it is very difficult, if not impossible, to precisely investigate the causal
relationship between muscle weakness and falls. Last, the (retrospective or prospec-
tive) fall collection using self-report methods is usually a significant period (like
months) away from the evaluation of muscle strength (Moreland et al. 2004;
Horlings et al. 2008). As muscle strength decline is a serious problem among
older adults (Goodpaster et al. 2006; Skelton et al. 2002), the muscle strength
measured could be substantially different from the one at the instant when the fall
occurs. This mismatch raises another question as to how accurately the findings
derived from the information collected at various time instants reflect the real causal-
effect linkage between fall incidences and the muscle weakness. The only way to
truly quantify the relationship between muscle weakness and falls is to evaluate how
all subjects respond to the same gait perturbation administered in a controlled
laboratory condition and to evaluate the muscle strength performance at the same
time as the laboratory-induced falls.
A recent study has been conducted to pinpoint the role of muscle weakness in
slip-related falls among older adults upon an ActiveStep treadmill (Simbex, Leba-
non, NH, Fig. 3a) (Ding and Yang 2016). After being measured for their right knee
joint isometric strength capacity, 36 healthy older adults (71.3 4.7 years) were
subject to an identical unexpected slip perturbation while walking on the treadmill
under the protection of a safety harness. The perturbation level of the slip was 8 m/s2
acceleration, 16 cm slip distance, and 1.6 m/s peak slip velocity (Fig. 3b).
Responding to this slip, 17 participants (47.2 %) fell and 19 (52.8 %) recovered
successfully (Fig. 3c). Results indicated that the maximum knee strength in both
directions were significantly lower among fallers than among those who did not fall
with high effect size ( 0.67) (Fig. 3d). Logistic regression models using the
strength measurements as predictors reached good classification as evidenced by
the large area under the receiver operating characteristic curve (Fig. 4a–d). The
optimal cutoff values to classify fallers and those who recovered were 1.05 and
1.10 Nm/kg for knee extensor and flexor, respectively (Fig. 4e, f). One with the knee
joint strength less than 1.05 Nm/kg at extensor or 1.10 Nm/kg at flexor would have a
higher likelihood of falls than those who possess knee strength greater than the cutoff
values. Results further indicated that a decrease of 1 standard deviation
(SD) (= 0.36 Nm/kg across all subjects) in knee extensor strength capacity increased
a Experimental setup c Slip outcome
Harness
Load cell
d Strength comparison
1.8
Fall (n = 17)
Belt speed ** Recovery (n = 19)
Strength (Nm/kg)
b Perturbation profile (8 m/s 2) *
1.2
1
Slip perturbation D 0.8 m/s
Belt speed (m/s)
(8 m/s2)
0.6
0
0 A
7 14 21 **: p < 0.01
G
*: p < 0.05
0.2 s 0
B C E F
-1 -0.8 m/s -0.8 m/s Extensor Flexor
Time (s) Muscle group
Fig. 3 Schematics of (a) the treadmill used to produce slip-like perturbation and (b) a represen-
tative profile of the treadmill perturbation for a slip trial with the self-selected walking speed of
0.8 m/s and an acceleration of 8 m/s2. The slip trial began with a 2-s ramp up (point A to B),
followed by a steady state with a backward-moving belt speed of 0.8 m/s (B to C). After 10–12
regular steps in the slip trial, approximately 80–120 ms following the touchdown of the leading foot,
the top belt was suddenly accelerated forward within 0.2 s (point C to D) without the subjects’
knowledge. Following the slip perturbation, the top belt speed slowly returned to backward
direction at 0.8 m/s (point D to E). Subjects were protected by a full-body safety harness during
all trials on the treadmill. (c) Images showing two possible slip outcome: fall and recovery, which
was determined by the force exerted by the participant on the ropes connecting the harness to the
overhead arch. Such a force was measured by a load cell. A trial was classified as a fall if the peak
load cell force exceeds 30 % body weight (Yang and Pai 2011). (d) Comparisons of the maximum
voluntary strength (or torque) of the knee extensor and flexor under isometric contraction condition
between falls and recoveries. Both muscle measurements were normalized to body mass (Ding and
Yang 2016)
the probability of fall 3.23 times. A reduction of 1 SD (=0.29 Nm/kg) of knee flexor
strength capacity raised the odds of falling by a factor of 2.30. The probability of a
fall induced by a slip could be characterized by the following two equations based on
the knee joint strength:
1
pðfallÞ ¼ (1)
1 þ e4:05þ3:28EXT
1
pðfallÞ ¼ (2)
1 þ e2:84þ2:90FLE
12 F. Yang
Extensor Flexor
Probability of fall
a b
1 1
1 1
Probability (%)
p ( fall ) = p ( fall ) =
0.75 1 + e −4.05+3.28EXT 0.75 1 + e −2.84+ 2.90FLE
0.5 0.5
0.25 0.25
0 0
0 1 2 3 0 1 2 3
Strength (Nm/kg) Strength (Nm/kg)
Receiver operating characteristic curve
c 1 d 1
0.75
Sensitivity
0.75
0.5 0.5
0.25 0.25
AUC = 0.780 AUC = 0.693
0 0
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
1-Specificity 1-Specificity
Youden index
e 0.6 f 0.4
1.1 Nm/kg
Youden index
0.4
0.2
0.2
1.05 Nm/kg
0 0
0.5 1 1.5 2 0.5 1 1.5 2
Strength (Nm/kg) Strength (Nm/kg)
Fig. 4 Logistic regression in which the probability of fall from a slip induced on the ActiveStep
treadmill is predicted using the strength capacity of (a) knee extensor and (b) knee flexor. The
receiver operating characteristic curve illustrating sensitivity and specificity of knee strength
capacity of (c) extensor and (d) flexor to predict the slip outcome. The area under curve (AUC) is
0.78 and 0.69 for extensor and flexor, respectively. Also shown are the changes in the Youden Index
values with (e) knee extensor and (f) knee flexor strength capacity ranging from 0.5 to 2.0 Nm/kg.
The Youden Index was calculated as the difference between the sensitivity and 1 – specificity. The
best cutoff point for the knee strength capacities to predict slip-initiated falls was selected as the
value which maximized the Youden Index. The best cutoff value for knee extensor was 1.05 Nm/kg
and 1.10 Nm/kg for flexor (Ding and Yang 2016)
where EXT and FLE stand for the maximum isometric torque produced by the knee
extensor and flexor, respectively.
The bond between muscle weakness and slip-related fall incidences could be
explained by the necessary body reactions after a postural perturbation during gait.
Following a slip, to regain body balance and prevent an actual fall, one must generate
quick and sufficient corrective reactions during the recovery stepping (Cham and
Redfern 2001; Yang et al. 2009). Two major reactions to a slip perturbation have
been identified empirically (Cham and Redfern 2001; Moyer et al. 2009) and
analytically (Yang and Pai 2010a, b). One reaction is a knee flexor moment which
has the potential to drag the BOS toward the body’s COM. Such reaction improves
the dynamic stability and increases the chance of successfully recovering from the
unexpected perturbation (Cham and Redfern 2001; Yang and Pai 2010b). Another
reactive response is the extensor moment from the recovery leg providing sufficient
antigravity support to prevent a limb collapse (Cham and Redfern 2001; Pai
et al. 2006). Adequate lower extremity muscle strength is demanded to execute
both reaction strategies. If one can produce great muscle strength, the chance to
retard and even reverse the falling after the slip would increase, reducing the
probability of fall.
Based on the predictive models, one whose knee muscle strength is above the
thresholds would have small susceptibility of slip-related falls. These cutoff values
could become a target for any muscle-strengthening programs in addition to its
potential use of distinguishing those with high risk of falls. This slip-related fall risk
could be simply used in clinic settings by using a handheld device to measure the
knee joint muscle strength. By substituting the strength measurements to the two
equations listed above, the probability of falling following a slip in gait could be
estimated. Therefore, those with elevated risk of falls can be identified quickly.
Muscle Power Is More Important than Muscle Strength in Resisting Falls

After a Slip
The importance of lower extremity strength in recovering from a slip has been
demonstrated as those with low strength fall due to the inability to support the
body during the reactive stepping (Cham and Redfern 2001; Ding and Yang 2016).
Decline with aging in muscle power (the product of muscle strength and contraction
velocity) is more visible than in strength (3–4 % vs. 1–2 % per year) (Skelton
et al. 1994), and the loss of muscle power due to normal aging has greater functional
impact on mobility than the loss of strength among older adults (Foldvari et al. 2000;
Bean et al. 2003). Emerging evidence suggests that muscle power may also play a
vital role in preventing falls (Evans 2000). It has been reported that senior fallers
demonstrated less power in the lower limbs than their non-faller counterpart (Skelton
et al. 2002; Orr et al. 2006), suggesting that lower muscle power may be an early
indicator of the risk of falls (Orr et al. 2006). Muscle power requires the integration
of muscle strength with the ability to develop contraction velocity (Kraemer and
Newton 2000). Since falls usually occur in a very short duration (approximately
200–500 ms) after a perturbation (Pijnappels et al. 2005; Yang et al. 2009), the
reactions to the perturbation must be quick enough. Post slip, the slipping limb needs
to generate flexor moment to pull the BOS backward close to the body’s COM to
restore stability (Yang and Pai 2010b; Cham and Redfern 2001), while the recovery
limb must provide sufficient limb support rapidly to impede the falling (Pai
et al. 2006; Yang et al. 2009). From this perspective, muscle power could be more
14 F. Yang
Fig. 5 (a) The profile of the a Perturbation profile (24 cm and 12 m/s 2)
treadmill belt speed for a slip
trial with an acceleration level 1.2 m/s
Slip perturbation
Belt speed (m/s)

of 12 m/s2 and peak slip 1
(12 m/s2)
velocity of 2.4 m/s.
Comparisons of the maximum 0 7 14 21
0
strength (or torque) and power
Time (s)
of the knee joint associated
with maximum voluntary -1 0.2 s
isometric and isokinetic
-1.2 m/s -1.2 m/s
contractions (MVC) for
(b) extensor and (c) flexor b Strength Power
between falls and recoveries. Extensor
8
Strength (Nm/kg)
The muscular performance 4
Power (W/kg)
variables of interest include *: p < 0.05
isometric joint torque capacity
*
during MVC in both extension 2 4
and flexion directions and the
joint power capacity during
concentric isokinetic MVC in 0 0
both directions at three
different speeds: 60 /s, 120 / c Recovery (n = 14)
Flexor
s, and 180 /s. Both joint
Strength (Nm/kg)
2 Fall (n = 12) 4
Power (W/kg)
torque and power were
normalized to body mass (Han *
and Yang 2015)
1 * 2
0 0
0 deg/s 60 deg/s 120 deg/s 180 deg/s
Isometric Isokinetic
Contraction condition
essential than muscle strength to recover balance during an impending fall initiated
by a slip.
A study using the ActiveStep treadmill focused on the relative importance of
muscle power and strength in resisting a slip-related fall (Han and Yang 2015).
Twenty-six healthy young adults (23.8 4.3 years) participated in this study. Their
muscle strength (torque) and power capacities were assessed at the right knee under
maximum voluntary isometric (extension and flexion) and isokinetic (concentric
extension and flexion at three different contraction speeds: 60 /s, 120 /s, and 180 /s)
contractions, respectively. They were then subject to a standardized and unannounced
slip while walking on the ActiveStep treadmill under the protection of a safety harness.
The acceleration level of the slip was 12 m/s2 with the slip distance of 24 cm and the
peak slip velocity of 2.4 m/s (Fig. 5a). Upon the unannounced slip, 12 participants
(46.2 %) fell and 14 recovered successfully. Accuracy of predicting slip outcome (fall
vs. recovery) was examined for each muscle performance measure using logistic
regression. Results showed that overall the joint power capacity measures predicted
the slip outcome with higher accuracy than did the joint torque capacity measurements.
Specifically, the predictive capability of knee power under the conditions of 60 /s
flexion and 120 /s flexion reached the greatest sensitivity (66.7 %) and overall predic-
tion accuracy (73.1 %) among all predictors (Fig. 5b, c). Knee strength capability in
both directions demonstrated the lowest overall prediction accuracy among all muscular
performance measurements (53.8 % for extensor and 61.5 % for flexor). A SD decrease
in all knee power capability measurements increased the probability of falling by higher
factors than did the strength measurements (1.77–2.28 vs. 1.25–1.37).
Such results illustrated that muscle power could be more closely related to a fall
initiated by a slip during gait. The findings could provide another potential tool to
identify individuals at increased risk of falls using the joint power capacity mea-
surement and to design effective fall prevention training paradigms aiming at
maximizing muscle power among older adults and others with physical disabilities.
This assessment tool could also be clinically applied. The chair-rise test has been
clinically used to easily quantify the lower limb power (Hardy et al. 2010). In this
connection, the evaluation of the risk of slip-related falls could be deployed to
clinical settings.
Other Tools to Assess Risk of Slip-Related Falls

Alternatively, gait variability has been applied to quantify its stability. Based on the
nonlinear dynamic theory for cyclical movement, variability in kinematics is indic-
ative of stability (Dingwell et al. 2001; England and Granata 2007; Hausdorff
et al. 2001). Indices, such as the maximum Floquet multiplier (Dingwell
et al. 2007) and Lyapunov exponents (Dingwell and Cusumano 2000), have been
employed to continuous joint or trunk kinematics (Bruijn et al. 2010; Dingwell and
Kang 2007; Toebes et al. 2012) to, respectively, evaluate body orbital and local
stability. Some studies have proposed that the orbital stability (Hamacher et al. 2011)
and the local stability (Lockhart and Liu 2008; Toebes et al. 2012) may be able to
differentiate fall-prone individuals from their health counterpart. Further, simpler
yet, the linear variability measures (i.e., the standard deviation) of spatiotemporal
gait parameters such as the step length, step width, or step/stride/swing time can also
yield useful information reflecting a person’s control of gait stability (Hausdorff
et al. 2001; Owings and Grabiner 2004; Woledge et al. 2005; Callisaya et al. 2011).
However, how these variability indices relate to slip-related falls are not well
understood. A study has been conducted to inspect the relative predictive power of
slip-related falls in gait by using these variability/stability measures among
community-dwelling older adults (Yang and Pai 2014).
Orbital Dynamic Stability

Orbital stability is defined using the maximum Floquet multiplier that quantifies how
a purely periodic system responds to transient perturbations discretely from one
cycle to the next (Dingwell and Kang 2007). The maximum Floquet multiplier can
be estimated based on the well-developed methods (Donelan et al. 2004; Hurmuzlu
and Basdogan 1994; Dingwell and Kang 2007). Some studies have shown that the
maximum Floquet multiplier might be associated with high risk of falls (Granata and
16 F. Yang
Lockhart 2008), while others reported weak linkages (Riva et al. 2013). Therefore,
the connection between orbital dynamic stability with general falls is still unclear
(Bruijn et al. 2010). To our knowledge, only one study has investigated the predic-
tion capacity of Floquet multiplier in identifying slip-related falls (Yang and Pai
2014). A group of 187 community-dwelling adults (age: 71.9 5.1 years) were
included in this study (Yang and Pai 2014). After approximately 10 normal walking
trials, all participants experienced a novel and unannounced overground slip created
by a movable platform. Among them, 98 fell and 89 recovered following the slip.
Their orbital dynamic stability was quantified by the maximum Floquet multiplier in
terms of the trunk’s movement in six dimensions (three translations and three
rotations) during the normal walking trials. No significant difference in the maxi-
mum Floquet multiplier was detected between fallers and those who did not fall. The
maximum Floquet multiplier predicted the impending slip-related falls with moder-
ate and nonsignificant sensitivity (71.4 %) and specificity (43.8 %). Although this
study was based on a large sample size (n = 187), it suffered from the technical
difficulty which refrained from collecting sufficient number of steps continuously
(Yang and Pai 2014). This finding implies that orbital stability might not be used as a
major risk factor to predict the probability of slip-related falls in order to identify
those with heightened risk of slip-related falls. Nevertheless, more studies with
comprehensive design along this research line are demanded.
Local Dynamic Stability

Local stability, defined by local divergence exponents, measures how sensitive a
system responds to very small perturbations continuously in real time (Dingwell and
Kang 2007). In the context of human gait, local dynamic stability measures the
ability of the human neuromuscular control system to attenuate those disturbances
manifested from either internal or external perturbations (Hamacher et al. 2011). A
study based on 13 adults (five healthy young, four healthy older, and four fall-prone
older adults) examined how local dynamic stability (the maximum Lyapunov expo-
nent) could be employed to distinguish those with higher risk of slip-related falls
from those with lower risk (Lockhart and Liu 2008). The maximum Lyapunov
exponent was calculated from the hip joint anteroposterior acceleration data to
quantify the local dynamic stability. The results suggested that the local dynamic
stability was significantly lower among the fall-prone older adults in comparison
with their healthy counterpart.
The same study conducted by Yang and Pai (2014) also attempted to inspect the
capability of local dynamic stability in predicting an imminent slip-related fall
among the group of 187 older adults. Their local dynamic stability was quantified
by the maximum Lyapunov exponent in terms of the trunk’s movement in six
dimensions. Results indicated that the Lyapunov exponent could predict correctly
up to 57.2 % slip outcomes. Findings from these two studies suggest that local
dynamic stability may be a valid predictor to differentiate adults with high risk of
experiencing a slip-related fall. However, the exact connection between local
dynamic stability and slip-related falls in modeling, experimental, and observational
studies is not yet established (Hamacher et al. 2011; Bruijn et al. 2011). It requires
further in-depth investigations.
Gait Parameters Variability

Gait variability is considered a linear and indirect assessment of stability measure
(Hamacher et al. 2011). The variability of these gait parameters, such as step length,
step width, and step time, is calculated as SD of these parameters over a sufficient
number of step cycles. From a large sample of older adults (n = 597), the variability
of the swing time and stride length has been associated with the risk of falls
(Verghese et al. 2009). A later study based on 52 community-dwelling adults also
examined the relationship between gait variability and falls (Hausdorff et al. 2001).
It was discovered that the stride and swing time variability differs significantly
between those who fell within 1-year period and those who did not report any falls
within the same period (Hausdorff et al. 2001). Logistic regression further exhibited
that stride time variability is able to predict future falls, with one SD increase in the
stride time variability increasing the odds of fall 5.3 folds. As for slip-initiated falls, a
study indicated that the step width variability during normal walking could signif-
icantly predict an impending slip-related fall (Yang and Pai 2014). The variability of
the step width may be the most practical and clinically friendly measurement to
quantify the risk of slip-related falls due to its simplicity to apply in clinics.
Gait Parameters
Some gait parameters collected during normal walking were also related to slip-related
falls. In a study, 11 older and 16 young individuals were exposed to an unexpected slip
when walking over a slippery surface over ground (Moyer et al. 2006). Gait parameters
collected at the heel strike prior to the slip onset were compared between those who
experienced high slip severity and those who did not. It was indicated that greater
normalized step length to leg length, larger and more rapidly varying foot landing angle
at heel strike, and decreased step frequency could be tied with the elevated probability
of a fall after a slip in gait. Therefore, these gait parameters collected during unperturbed
walking could play a role in the severity of a slip induced in gait and could be used as a
simple tool to assess the risk of slip-initiated falls.
Summary and Future Directions
Falls and fall-related injuries are expected to be a major healthcare problem facing
older people for the anticipatable future (Sleet et al. 2008). Consequently, fall
prevention will remain a critical public health priority. The development and imple-
mentation of accurate and effective fall risk assessment tools is one of the most
important components for fall prevention. The early identification and screening of
individuals subjected to high risk of falls can maximize the use of limited resources
allocated to fall prevention. Substantial work has been carried out to identify, testify,
and evaluate the approaches for assessing the risk of falls, particularly falls initiated
by slips among older adults. The relationships between slip-related falls and
18 F. Yang
dynamic gait stability, limb support, muscle weakness, and gait variability have been
quantitatively examined, and the risk assessment approaches based on these factors
have been established.
Although fall prevention is receiving increasing clinical and economical atten-
tion, the assessment tool designed to identify fall-prone individuals remains in its
early stage. Further work needs to be conducted to understand the role of certain
other risk factors, such as sensation loss, vision impairment, cognition deficit,
reaction time deterioration, fear of falling, etc. To rank the relative efficacy of
these factors leading to falls could be another research direction, which would
provide useful guidance to prioritize and develop objective, accurate, cost-effective,
and clinically-applicable fall risk assessment systems on an individual basis. Some
neurological disorders (such as multiple sclerosis, stroke, Parkinson’s disease, etc.)
and clinical conditions (like obesity, diabetes, musculoskeletal pain, etc.) increase
the risk of falls. It is, therefore, worthy to explore how these conditions interplay
with slips and falls in order to design population-specific fall reduction protocols.
Another possible future direction could be the examination of how the risk assess-
ment can be integrated with practical interventions for fall reduction. For example,
although studies reported that muscle weakness is a limiting factor for balance
recovery after a slip perturbation among older adults, the question remains whether
slip-related falls can be reduced or prevented by targeting this factor through
exercise-based muscle-strengthening interventions. To date, outcomes from strength
training programs for individuals with high risk of falls are still inconsistent or even
controversial (Melzer et al. 2004; Orr 2010). Lastly, barriers still exist when
implementing these risk assessment tools to clinical settings due to its demand of
special equipment and technologies. However, with the rapid development of smart,
easy-to-use, and cost-effective measurement techniques, such as the wearable health
system, inertial measurement unit, accelerometers, and video recording (Kavanagh
and Menz 2008; Labrador and Delahoz 2014), the fall risk assessment tools
reviewed in this chapter could be increasingly applied in clinics pertaining to fall
prevention.
References
Bean JF, Leveille SG, Kiely DK, Bandinelli S, Guralnik JM, Ferrucci L (2003) A comparison of leg
power and leg strength within the InCHIANTI study: which influences mobility more?
J Gerontol Ser A: Biol Sci Med Sci 58:M728–M733
Berg WP, Alessio HM, Mills EM, Tong C (1997) Circumstances and consequences of falls in
independent community-dwelling older adults. Age Ageing 26:261–268
Bhatt T, Espy D, Yang F, Pai Y-C (2011) Dynamic stability, clinical correlates, and prognosis of falls
among community-dwelling older adults. Arch Phys Med Rehabil 92:799–805
Borelli GA (1680) De Motu Animalium (On the movement of animals). Springer, Berlin
Bruijn SM, Kate WR, Faber GS, Meijer OG, Beek PJ, van Dieen JH (2010) Estimating dynamic
gait stability using data from non-aligned inertial sensors. Ann Biomed Eng 38:2588–2593
Bruijn SM, Bregman DJJ, Meijer OG, Beek PJ, van Dieen JH (2011) The validity of stability
measures: A modelling approach. J Biomech 44:2401–2408
Callisaya ML, Blizzard L, Schimidt MD, Martin KL, McGinley JL, Sanders LM, Srikanth VK
(2011) Gait, gait variability and the risk of multiple incident falls in older people: a population-
based study. Age Ageing 40:481–487
Cham R, Redfern MS (2001) Lower extremity corrective reactions to slip events. J Biomech
34:1439–1445
Ding L, Yang F (2016) Muscle weakness is related to slip-initiated falls among community-
dwelling older adults. J Biomech 49:238–243
Dingwell JB, Cusumano JP (2000) Nonlinear time series analysis of normal and pathological
human walking. Chaos 10:848–863
Dingwell JB, Kang HG (2007) Differences between local and orbital dynamic stability during
human walking. J Biomech Eng 129:586–593
Dingwell JB, Cusumano J, Cavanagh P, Sternad D (2001) Local dynamic stability versus kinematic
variability of continuous overground and treadmill walking. J Biomech Eng 123:27–32
Dingwell JB, Gu KH, Marin LC (2007) The effects of sensory loss and walking speed on the orbital
dynamic stability of human walking. J Biomech 40:1723–1730
Donelan JM, Shipman DW, Kram R, Kuo AD (2004) Mechanical and metabolic requirements for
active lateral stabilization in human walking. J Biomech. 37:827–835
El-Khoury F, Cassou B, Charles M-A, Dargent-Molina P (2013) The effect of fall prevention
exercise programmes on fall induced injuries in community dwelling older adults: systematic
review and meta-analysis of randomised controlled trials. BMJ 2013:1–13
England SA, Granata KP (2007) The influence of gait speed on local dynamic stability of walking.
Englander F, Hodson TJ, Terregrossa RA (1996) Economic dimensions of slip and fall injuries. J
Forensic Sci 41:733–746
Evans WJ (2000) Exercise strategies should be designed to increase muscle power. J Gerontol Ser
A: Biol Sci Med Sci 55:M309–M310
Feldman F, Robinovitch SN (2006) Recalling the mechanics of falls. Young adults cannot accu-
rately describe the sites of impact immediately after a fall occurs. Osteoporos Int 17:955
Foldvari M, Clark M, Laviolette LC, Bernstein MA, Kaliton D, Castaneda C, Pu CT, Hausdorff JM,
Fielding RA, Singh MA (2000) Association of muscle power with functional status in commu-
nity-dwelling elderly women. J Gerontol Ser A: Biol Sci Med Sci 55:M192–M199
Goodpaster BH, Park SW, Harris TB, Kritchevsky SB, Nevitt M, Schwartz AV, Simonsick EM,
Tylavsky FA, Visser M, Newman AB (2006) The loss of skeletal muscle strength, mass, and
quality in older adults: the health, aging and body composition study. J Gerontol Ser A: Biol Sci
Med Sci 61:1059–1064
Graafmanc WC, Lips P, Wijlhuizen GJ, Pluijm SM, Bouter LM (2003) Daily physical activity and
the use of a walking aid in relation to falls in elderly people in a residential care setting. Z
Gerontol Geriatr 36:23–28
Granata KP, Lockhart TE (2008) Dynamic stability differences in fall-prone and healthy adults. J
Grisso JA, Kelsey JL, Strom BL, Chiu GY, Maislin G, O’Brien LA, Hoffman S, Kaplan F (1991)
Risk factors for falls as a cause of hip fracture in women. N Engl J Med 324:1326–1331
Haines T, Kuys SS, Morrison G, Clarke J, Bew P (2008) Balance impairment not predictive of falls
in geriatric rehabilitation wards. J Gerontol: Med Sci 63A:523–528
Hamacher D, Singh NB, Van Dieen JH, Heller MO, Taylor WR (2011) Kinematic measures for
assessing gait stability in elderly individuals: a systematic review. J R Soc Interface 8:1682–1698
Han L-Z, Yang F (2015) Strength or power: which is more important to prevent slip-related falls?
Hum Mov Sci 44:192–200
Hardy R, Cooper R, Shah I, Harridge S, Guralnik J, Kuh D (2010) Is chair rise performance a useful
measure of leg power? Aging Clin Exp Res 22:412–418
Hausdorff JM, Edelberg HK, Mitchell SL, Goldberger AL, Wei J-Y (1997) Increased gait unstead-
iness in community-dwelling elderly fallers. Arch Phys Med Rehabil 78:278–283
20 F. Yang
Hausdorff JM, Rios DA, Edelberg HK (2001) Gait variability and fall risk in community-living
older adults: a 1-year prospective study. Arch Phys Med Rehabil 82:1050–1056
Hayes WC, Myers ER, Robinovitch SN, Van Den Kroonenberg A, Courtney AC, McMahon TA
(1996) Etiology and prevention of age-related hip fractures. Bone 18:77s–86s
Hedlund R, Lindgren U (1987) Trauma type, age, and gender as determinants of hip fracture. J
Orthop Res 5:242–246
Heinrich S, Rapp K, Rissmann U, Becker C, Kunig H-H (2010) Cost of falls in old age: a systematic
review. Osteoporos Int 21:891–902
Hof AL, Gazendam MG, Sinke WE (2005) The condition for dynamic stability. J Biomech 38:1–8
Horlings CGC, van Engelen BGM, Allum JHH, Bloem BR (2008) A weak balance: the contribution
of muscle weakness to postural instability and falls. Nat Clin Pract Neurol 4:504–515
Hurmuzlu Y, Basdogan C (1994) On the measurement of dynamic stability of human locomotion. J
Biomech Eng 116:30–36
Jenkins P, Earle-Richardson G, Slingerland DT, May J (2002) Time dependent memory decay. Am J
Ind Med 41:98–101
Kannus P, Parkkari J, Koskinen S, Niemi S, Palvanen M, Jarvinen M, Vuori I (1999) Fall-induced
injuries and deaths among older adults. J Am Med Assoc 281:1895–1899
Kavanagh JJ, Menz HB (2008) Accelerometry: a technique for quantifying movement patterns
during walking. Gait Posture 28:1–5
Koepsell TD, Wolf ME, Buchner DM, Kukull WA, LaCroix AZ, Tencer AF, Frankenfeld CL,
Taulvydas M, Larson EB (2004) Footwear style and risk of falls in older adults. J Am Geriatr
Soc 52:1495–1501
Kraemer WJ, Newton RU (2000) Training for muscular power. Phys Med Rehabil Clin N Am
11:341–368
Labrador MA, Delahoz YS (2014) Survey on fall detection and fall prevention using wearable and
external sensors. Sensors 14:19806–19842
Lajoie Y, Gallagher SP (2004) Predicting falls within the elderly community: comparison of
postural sway, reaction time, the Berg balance scale and the Activities-specific Balance Confi-
dence (ABC) scale for comparing fallers and non-fallers. Arch Gerontol Geriatr 38:11–26
Lockhart TE, Liu J (2008) Differentiating fall-prone and healthy adults using local dynamic
stability. Ergonomics 51:1860–1872
Luukinen H, Herala M, Koski K, Honkanen R, Laippala P, Kivela SL (2000) Fracture risk
associated with a fall according to type of fall among the elderly. Osteoporos Int 11:631–634
Maki BE, McIlroy WE (1997) The role of limb movements in maintaining upright stance: the
“change-in-support” strategy. Phys Ther 77:488–507
Melzer I, Benjuya N, Kaplanski J (2004) Postural stability in the elderly: a comparison between
fallers and non-fallers. Age Ageing 33:602–607
Mille ML, Rogers MW, Martinez K, Hedman LD, Johnson ME, Lord SR, Fitzpatrick RC (2003)
Thresholds for inducing protective stepping responses to external perturbations of human
standing. J Neurophysiol 90:666–674
Moreland JD, Richardson JA, Goldsmith CH, Clase CM (2004) Muscle weakness and falls in older
adults: a systematic review and meta-analysis. J Am Geriatr Soc 52:1121–1129
Morley JE (2002) A fall is a major event in the life of an older person. J Gerontol Ser A Biol Sci
Med Sci 57:M492–M495
Moyer BE, Chambers AJ, Redfern MS, Cham R (2006) Gait parameters as predictors of slip
severity in younger and older adults. Ergonomics 49:329–343
Moyer BE, Redfern MS, Cham R (2009) Biomechanics of trailing leg response to slipping -
Evidence of interlimb and intralimb coordination. Gait Posture 29:565–570
Muir SW, Berg KO, Chesworth B, Klar N, Speechley M (2010) Balance impairment as a risk factor
for falls in community-dwelling older adults who are high functioning: a prospective study. Phys
Ther 90:338–347
Muir SW, Gopaul K, Montero Odasso MM (2012) The role of cognitive impairment in fall risk
among older adults: a systematic review and meta-analysis. Age Ageing 41:299–308
Nevitt MC, Cummings SR, Kidd S, Black D (1989) Risk factors for recurrent nonsyncopal falls. A
prospective study. J Am Med Assoc 261:2663–2668
O’Sullivan SB (2007) Examination of motor function. In: O’Sullivan SB, Schmitz TJ (eds) Physical
rehabilitation. F.A. Davis Company, Philadelphia, pp 249–253
Orr R (2010) Contribution of muscle weakness to postural instability in the elderly. Eur J Phys
Rehabil Med 46:183–220
Orr R, de Vos NJ, Singh NA, Ross DA, Stavrinos TM, Fiatarone-Singh MA (2006) Power training
improves balance in healthy older adults. J Gerontol Ser A: Biol Sci Med Sci 61A:78–85
Owings TM, Grabiner MD (2004) Step width variability, but not step length variability or step time
variability, discriminates gait of healthy young and older adults during treadmill locomotion. J
Biomech 37:935–938
Pai Y-C, Patton JL (1997) Center of mass velocity-position predictions for balance control.
J Biomech 30:347–354
Pai Y-C, Yang F, Wening JD, Pavol MJ (2006) Mechanisms of limb collapse following a slip among
young and older adults. J Biomech 39:2194–2204
Parkkari J, Kannus P, Palvanen M, Natri A, Vainio J, Aho H, Vuori I, Jarvinen M (1999) Majority of
hip fractures occur as a result of a fall and impact on the greater trochanter of the femur: a
prospective controlled hip fracture study with 206 consecutive patients. Calcif Tissue Int
65:183–187
Patton JL, Lee WA, Pai Y-C (2000) Relative stability improves with experience in a dynamic
standing task. Exp Brain Res 135:117–126
Pavol MJ, Pai Y-C (2007) Deficient limb support is a major contributor to age differences in falling.
J Biomech 40:1318–1325
Pijnappels M, Bobbert MF, van Dieen JH (2005) Control of support limb muscles in recovery after
tripping in young and older subjects. Exp Brain Res 160:326–333
Redfern MS, Cham R, Gielo-Perczak K, Gronqvist R, Hirvonen M, Lanshammar H, Marpet M, Pai
Y-C, Powers C (2001) Biomechanics of slips. Ergonomics 44:1138–1166
Riva F, Bisi MC, Stagni R (2013) Orbital stability analysis in biomechanics: a systematic review of
a nonlinear technique to detect instability of motor tasks. Gait Posture 37:1–11
Rubenstein LZ (2006) Falls in older people: epidemiology, risk factors and strategies for prevention.
Age Ageing 35(Suppl 2):ii37–ii41
Rubenstein LZ, Josephson KR (2006) Falls and their prevention in elderly people: what does the
evidence show? Med Clin N Am 90:807–824
Rubenstein LZ, Josephson KR, Robbins AS (1994) Falls in the nursing home. Ann Intern Med
121:442–451
Skelton DA, Greig CA, Davies JM, Young A (1994) Strength, power and related functional ability
of healthy people aged 65-89 years. Age Ageing 23:371–377
Skelton DA, Kennedy J, Rutherford OM (2002) Explosive power and asymmetry in leg muscle
function in frequent fallers and non-fallers aged over 65. Age Ageing 31:119–125
Sleet DA, Moffett DB, Stevens J (2008) CDC’s research portfolio in older adults fall prevention: A
review of progress, 1985 - 2005, and future research directions. J Safety Res 39:259–267
Smeesters C, Hayes WC, McMahon TA (2001) Disturbance type and gait speed affect fall direction
and impact location. J Biomech 34:309–317
Stevens JA, Corso PS, Finkelstein EA, Miller TR (2006) The costs of fatal and non-fatal falls
among older adults. Inj Prev 12:290–295
Geriatr Soc 34:119–126
Tinetti ME (2003) Preventing falls in elderly persons. N Engl J Med 388:42–49
Toebes MJ, Hoozemans MJ, Furrer R, Dekker J, van Dieen JH (2012) Local dynamic stability and
variability of gait are associated with fall history in elderly subjects. Gait Posture 36:527–531
Verghese J, Holtzer R, Lipton RB, Wang C (2009) Quantitative gait markers and incident fall risk in
older adults. J Gerontol Med Sci 64:896–901
22 F. Yang
Wang T-Y, Bhatt T, Yang F, Pai Y-C (2011) Generalization of motor adaptation to repeated-slip
perturbation across tasks. Neuroscience 180:85–95
Wang T-Y, Bhatt T, Yang F, Pai Y-C (2012) Adaptive control reduces trip-induced forward gait
instability among young adults. J Biomech 45:1169–1175
Wijlhuizen GJ, Chorus AMJ, Hopman-Rock M (2010) The FARE: a new way to express FALLs
Risk among older persons including physical activity as a measure of exposure. Prev Med
50:143–147
Woledge RC, Birtles DB, Newham DJ (2005) The variable component of lateral body sway during
walking in young and older humans. J Gerontol Ser A: Biol Sci Med Sci 60:1463–1468
World Health Organization (2007) Global age-friendly cities: A guide. Geneva, Switzerland, pp 76
Yang F, Pai Y-C (2010a) Reactive control and its operation limits in responding to a novel slip in
gait. Ann Biomed Eng 38:3246–3256
Yang F, Pai Y-C (2010b) Role of individual lower limb joints in reactive stability control following a
novel slip in gait. J Biomech 43:397–404
Yang F, Pai Y-C (2011) Automatic recognition of falls in gait-slip training: Harness load cell based
criteria. J Biomech 44:2243–2249
Yang F, Pai Y-C (2014) Can stability really predict an impending slip-related fall among older
adults? J Biomech 47:3876–3881
Yang F, Anderson FC, Pai Y-C (2007) Predicted threshold against backward balance loss in gait. J
Biomech 40:804–811
Yang F, Anderson FC, Pai Y-C (2008a) Predicted threshold against backward balance loss follow-
ing a slip in gait. J Biomech 41:1823–1831
Yang F, Passariello F, Pai Y-C (2008b) Determination of instantaneous stability against backward
balance loss: Two computational approaches. J Biomech 41:1818–1822
Yang F, Bhatt T, Pai Y-C (2009) Role of stability and limb support in recovery against a fall
following a novel slip induced in different daily activities. J Biomech 42:1903–1908
Yang F, Bhatt T, Pai Y-C (2011) Limits of recovery against slip-induced falls while walking.
J Biomech 44:2607–2613
Yang F, Espy D, Bhatt T, Pai Y-C (2012) Two types of slip-induced falls among community
dwelling older adults. J Biomech 45:1259–1264
Zijlstra GA, van Haastregt JC, van Rossum E, van Eijk JT, Yardley L, Kempen GI (2007)
Interventions to reduce fear of falling in community-living older people: a systematic review.
J Am Geriatr Soc 55:603–615
Zijlstra GA, van Haastregt JCM, Ambergen T, van Rossum E, van Eijk JTM, Tennstedt SL,
Kempen GIJM (2009) Effects of a multicomponent cognitive behavioral group intervention
on fear of falling and activity avoidance in community-dwelling older adults: Results of a
randomized controlled trial. J Am Geriatr Soc 57:2020–2028
Detecting and Measuring Ataxia in Gait
Mariano Serrao and Carmela Conte
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Stationary Locomotion: Steady-State Overground Linear Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Time–Distance Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Lower Limb Joint Kinematics, Kinetics, and Coordination Patterns . . . . . . . . . . . . . . . . . . . . . . . . 6
Upper Body Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Muscle Activation Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Nonstationary Locomotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Gait Initiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Gait Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Abstract
Gait ataxia is traditionally described as clumsy, staggering movements with a
wide-based gait which resembles the gait of drunken people. Recent modern
motion analysis systems have been used to quantitatively characterize the nature
and degree of walking dysfunction. These findings have revealed that the whole
range of locomotor activities is impaired, including linear steady-state gait,
M. Serrao (*)
Department of Medical and Surgical Sciences and Biotechnologies, Sapienza University of Rome,
Latina, Italy
Movement Analysis LAB, Rehabilitation Centre Policlinico Italia, Rome, Italy
e-mail: mariano.serrao@uniroma1.it
C. Conte
Movement Analysis LAB, Rehabilitation Centre Policlinico Italia, Rome, Italy
e-mail: conte_carmela@libero.it

DOI 10.1007/978-3-319-30808-1_46-1
2 M. Serrao and C. Conte
turning, gait initiation, and gait termination. All these locomotor abnormalities
reflect poor limb coordination and impaired balance, which greatly restrict
patients in their daily life activities and predispose them to falls (van de
Warrenburg et al. 2005).
Detecting and measuring gait in patients with ataxia gives further insights on
the motor deficit and may allow to discern the complex relationship between the
primary deficits and the compensatory mechanisms, to recognize specific abnor-
malities and their impact on clinical decision-making, and to individualize reha-
bilitative treatment and better evaluating its effects over the time.
Keywords
Gait ataxia • Cerebellum • Gait analysis • Muscle coordination • Gait variability
Introduction
The cerebellum is a complex brain organ involved in the coordination of precise

movements. It functions as a comparator working in concert with the cerebral cortex
and basal ganglia (Kozio et al. 2014). In humans, 60–80 % of brain neurons are
found within the cerebellum (Herculano-Houzel 2010; Grimaldi and Manto 2012),
highlighting the importance of this brain organ in the phylogenetic evolution of
motor control.
A lesion or degeneration of the cerebellum or its afferents does not cause a loss of
movement but gives rise to several signs and features involving motor control. The
term ataxia (“lack of order” in Greek) was originally referred to a lack of motor
coordination, although a growing body of evidence indicates that the cerebellum
contributes to the regulation of other non-motor features including linguistic, cogni-
tive, and affective functions. The classic cerebellar motor syndrome includes a broad
array of signs, with the most commonly noted being dysmetria, asynergia or
dyssynergia, a- or dysdiadochokinesia, overshoot/impairment of the check reflex,
tremor, oculomotor abnormalities, speech disturbances, abnormalities of posture and
gait, and hypotonia (Bodranghien et al. 2016). Gait ataxia is a key feature of cerebellar
disorders and greatly impacts on patients’ independence, daily life activities, quality
of life, and risk of falls (Fonteyn et al. 2010; van de Warrenburg et al. 2005).
Gait ataxia has long been acknowledged and described as “clumsy, staggering
movements with a wide-based gait” which resembles the gait of drunken people.
While such traditional description still remains useful for clinicians to identify
cerebellar disease, modern motion analysis systems have been used to quantitatively
characterize the nature and degree of walking dysfunction. These findings have
revealed that the whole range of locomotor activities is impaired, including linear
steady-state gait, turning, gait initiation, and gait termination. All these locomotor
abnormalities reflect poor limb coordination and impaired balance, which greatly
restrict patients in their daily life activities and predispose them to falls (van de
Warrenburg et al. 2005).
Detecting and Measuring Ataxia in Gait 3
State of the Art
In the last two decades, the technology advancement provided different devices for
measuring and quantifying three-dimensional gait in ataxic patients. The most used
are optical motion analysis systems with active or passive markers (Palliyath et al.
1998; Earhart and Bastian 2000; Ilg et al. 2008; Serrao et al. 2012b; Conte et al.
2014; Martino et al. 2014) and, more recently, inertial sensors such as accelerometers
and gyroscopes (Chini et al. 2016).
The optoelectronic motion analysis system represents the laboratory gold stan-
dard and has been widely used in detecting and measuring gait ataxia (Bodranghien
et al. 2016). The system consists of infrared cameras that detect the coordinates of a
series of markers fixed on the body landmarks. The system can be equipped with
force plates for the acquisition of the ground reaction forces and with surface
electromyography devices for the acquisition of muscle activation. Using optoelec-
tronic systems, it is possible to obtain a full motion description in terms of kinematic,
kinetic, and electromyographic measures of human gait (Serrao et al. 2012a). Inertial
sensors are a possible alternative method to the laboratory gold standard and have
been recently used to detect ataxia (Chini et al. 2016). The inertial sensors measures
physical quantities such as acceleration and angular velocity, therefore converting
these data into kinematic data required mathematical process such as integration
(Tong and Granat 1999). However, the integration process introduces, in the calcu-
lated values, a shift from the true values causing data imprecisions. In conclusion,
inertial sensor has small size, portability properties, and limited costs; therefore these
sensors could be an attractive option for measuring human movement; however,
their use is still site and task specific (Cuesta-Vargas et al. 2010).
When analyzing and detecting gait ataxia, several biomechanical abnormalities
have been revealed in terms of spatiotemporal parameters, joint ROMs, joint torques,
joint coordination, upper body kinematics, and muscle activation patterns.
If the gait speed is controlled as influencing factor, the main differences, between
patients with ataxia and healthy controls, in spatiotemporal, joint kinematics, and
kinetics, regard the step width, step length, and ankle joint range of motion (Martino
et al. 2014; Chini et al. 2016; Mari et al. 2014). Particularly, patients with ataxia
show an increased step width, a reduced step length, and a reduced ankle joint
excursion. All of these abnormalities seem to represent compensatory mechanisms
aimed at reducing dynamic unbalance by either increasing the margin of stability on
the frontal plane, reducing the time in the less stable configuration (single support),
and increasing the ankle joint strategy to control the center of mass (CoM) displace-
ments, respectively (Bodranghien et al. 2016).
One of the most characterizing feature of gait ataxia is a marked variability of all
global and segmental kinematic and kinetic gait parameter values which has been
reported in almost all studies (for review see Bodranghien et al. 2016). It reflects the
inability of ataxic patients to maintain the dynamic balance through a regular
walking pattern leading to an increased risk of falls (Schniepp et al. 2014, 2016).
As regards the joint coordination deficit, abnormal intra-limb joint coupling
during walking, in terms of both joint movements and interaction torques, has
been reported by several studies (for review see Bodranghien et al. 2016). Particu-
larly, increased temporal variability of intra-limb coordination has been found
to be related with dynamic balance and irregular foot trajectories in ataxic gait
(Ilg et al. 2007). The impaired inter-joint coordination leads to an extremely irregular
pattern of alternating proximal/distal joint as compared to the well-shaped pattern of
healthy controls (Serrao et al. 2012b). Recently, an abnormal control of upper body
segment leading to an increase of movements and a lack of local stability of the trunk
has been observed (Conte et al. 2014; Chini et al. 2016), suggesting that upper and
lower body coordination may play a pivotal role in explaining several aspects of gait
ataxia.
Very recently, a detailed analysis of muscle activity patterns during locomotion in
cerebellar ataxia has been performed (Mari et al. 2014; Martino et al. 2015), showing
both impaired spatial and temporal profiles of EMG activity in individual muscles
and an impaired antagonist muscle activation. In essence, ataxic patients seem to
activate both individual and paired antagonist muscles more intensely and for a
longer time, possibly in an attempt to stiffen the limb and to compensate for the
instability due to poor muscle coordination.
Stationary Locomotion: Steady-State Overground Linear Walking
Time–Distance Parameters
Several time–distance parameters have been reported to be abnormal by some

authors, but within normal range by others (Palliyath et al. 1998; Ebersbach et al.
1999; Mitoma et al. 2000; Earhart and Bastian 2000; Stolze et al. 2002; Morton and
Bastian 2003; Ilg et al. 2008; Serrao et al. 2012b; Martino et al. 2014), while some
authors have described significant changes only for severe ataxias (Mitoma et al.
2000). Such abnormalities consisted in reduced step length, gait speed and cadence,
and increased step width, double support, stride, and stance durations.
However, it should be considered that when comparing the gait parameters
between patients and controls, the gait speed may be a confounding factor. Indeed,
considering that patients with ataxia walk slower than healthy controls and that
many global and segmental gait parameters are speed dependent, the final results
may be misleading. When matching the gait speed, essentially only an increased step
width and a reduced step length are revealed (Martino et al. 2014; Mari et al. 2014;
Fig. 1a).
The increased step width is the main compensatory mechanism applied on the
frontal plane and is aimed at increasing the safety margin between the center of mass
from the edge of base of support. The reduced step length may reflect, at least in part,
a mechanism aimed at reducing the time spent in the most unstable configuration
(single support). However, it has been reported that increasing the base of support
determines per se a reduction of the step length (McAndrew Young and Dingwell
2012).
Fig. 1 Kinematic gait parameters. (a) Comparison of general gait parameters for healthy subjects
(HS) and cerebellar ataxic patients (CA). (b) Ensemble-averaged (means SD) hip, knee. and
ankle joint angles and trunk roll and pitch orientation angles in 19 ataxic patients (right) and
20 age-matched healthy subjects (left). Data were normalized to the cycle duration and represented
in percentage of gait cycle. *Significant group differences (P 0.05, unpaired t-tests) (By Martino
et al. 2014, with authors’ permission)
Lower Limb Joint Kinematics, Kinetics, and Coordination Patterns
Reduced motion and torques of hip, knee, and ankle joints have been reported by
several authors, even though with contrasting results (Palliyath et al. 1998; Mitoma
et al. 2000; Stolze et al. 2002; Morton and Bastian 2003; Serrao et al. 2012b). It
should be noted that both the lack of gait speed matching procedure and the presence
of a high temporal variability of the joint kinematics and kinetics, across the gait
cycle, may represent a confounding factors in detecting biomechanical differences
between healthy controls and ataxic patients. When matching the gait speed and/or
examining joint kinematics and joint torques as continuous and not as discrete
variables (Serrao et al. 2012b; Martino et al. 2014), the main emerging differences
involve the ankle joint (reduced joint excursion) (Serrao et al. 2012b; Mari et al.
2014; Martino et al. 2014; Fig. 1b). Reduced RoM in the ankle joint could be related
to shorter steps and impaired intersegmental coordination and suggests that patients
with ataxia try to stiffen the ankle joint in order to increase gait stability by
maintaining the CoM oscillations within the base of support, even at the expense
of an efficient ankle movement.
Abnormal intra-limb joint coordination during walking, in terms of both joint
movements and interaction torques, has been reported in several studies with
different methods including decomposition index using joint angle–angle plots
(Earhart and Bastian 2000, 2001), variability of joint coordination patterns
(Ilg et al. 2007), continuous relative phase (Serrao et al. 2012b), and planar law
of intersegmental coordination (Martino et al. 2014). From these studies, it
clearly emerges that damage to/degeneration of the cerebellum results in
an impaired spatiotemporal multi-joint coordination, while single-joint movements
may be relatively preserved. This lower limb inter-joint incoordination is
expressed by an extremely irregular pattern of alternating proximal/distal joints
(Serrao et al. 2012b) and leads to the lack of flexibility of the kinematic pattern and
consequently to the inability to adapt to different dynamic conditions (Martino
et al. 2014).
As regards the force production and interaction torques between joints, in patients
with ataxia, the vertical ground reaction force demonstrated a prominent transient
following the heel strike during the loading response phase (0–10 % of the gait
cycle) (Martino et al. 2014) (Fig. 2). In a normal condition, the intensity of this
impact is adjusted by shock-absorbing reactions at the ankle, knee, and hip joints
(Perry and Burnfield 1992), and it is small or absent at low and normal walking
speeds (Fig. 2). The cause of this impaired loading may originate from the unbal-
anced control and preparation to the foot touchdown. As seen for the upper limb
during grasping tasks, cerebellar patients cannot generate the appropriate muscle
torques to control the mechanical consequences of dynamic interaction forces that
occur when a movement is executed (Topka et al. 1998; Nowak et al. 2007). An
increase in the impact transient could also be a consequence of patients’ leg
stiffening or reduced push-off of the contralateral limb in late stance. Anyway, the
prominent transient of the GRF may be a further source of perturbation in the ataxic
gait worsening gait instability and variability.
Fig. 2 Vertical ground reaction forces (GRFs) during walking. (a) Vertical GRF in one represen-
tative HS (top left) and one ataxic patient (top right). Every trace refers to a single cycle. (b) Bottom
ensemble-averaged (means SD) vertical GRF in HS (n = 20) and ataxic patients (n = 19). The
patterns are normalized to body weight and plotted versus normalized stance. Note the prominent
transient (Δ1) during the weight-acceptance period (marked by a shaded area) in ataxic patients
(By Martino et al. 2014, with authors’ permission)
Upper Body Kinematics
Three-dimensional kinematic analysis of the upper body motions in patients with

gait ataxia has shown significantly larger head and trunk RoMs in all three spatial
planes (sagittal, frontal, and yaw plane) (Conte et al. 2014; Fig. 3a). Furthermore, a
recent study on local stability of the trunk, as evaluated by Lyapunov exponent
(Chini et al. 2016), displayed marked trunk instability in all three spatial directions
indicating that the locomotor control system in patients with cerebellar ataxia does
not recover from small perturbations and hence becomes unstable (Fig. 3b).
It is known that upper body control is essential in ensuring clear vision, facilitat-
ing the integration of vestibular information, and contributing to maintenance of
balance. Thus, both the increased movements and the reduced stability of the trunk
suggest that impaired intersegmental coordination (lower–upper body segments)
may play a central role in determining ataxic gait. Patients with ataxia seem to be
subjected to small perturbations from which are unable to recover and which trigger
Fig. 3 (a) Three-dimensional representation of head and trunk displacements (mean values) in a
representative ataxic patient (grey lines) and healthy control (black lines) (Conte et al. 2014). (b) 3D
reconstructed state space of the acceleration and its time-delayed copies (time delay of ten data
samples) in the vertical (VT), latero–lateral (LL), and anterior–posterior (AP) directions for a
representative ataxic patient (black line) and healthy subject (gray line) (By Chini et al. 2016,
with authors’ permission)
a vicious circle that transforms the upper body itself into a generator of perturbations
(Chini et al. 2016).
Muscle Activation Patterns
The analysis of muscle activation has shown distinctive features in patients with gait
ataxia compared to controls (Mari et al. 2014; Martino et al. 2014). In particular, an
a 120 120
100 100
80 80
VL [%]
TA [%]
60 60
40 40
20 20
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
% cycle % cycle
120 120
100 100
80 80
GM [%]
BF [%]
60 60
40 40
20 20
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
% cycle % cycle
b Ataxic patients Controls

25% 25%
VL 50% 0% 50% 0%
75% 75%
25% 25%
BF 50% 0% 50% 0%
75% 75%
25% 25%
TA 50% 0% 50% 0%
75% 75%
25% 25%
GM 50% 0% 50% 0%
75% 75%
Fig. 4 (a) Mean muscle activity traces of a representative ataxic patient and a control subject. The
grey curve refers to the control subject and the black one to the ataxic patient (Mari et al. 2014). (b)
EMG peak event in ataxic patients and controls. Each dot on the circle represents the subject’s peak
event expressed as a percentage of the normalized gait cycle. The vector is the mean resultant
vector; it indicates the mean direction of the dots, and its length is an indicator of the spread of the
dots. *Between-groups significant differences (By Mari et al. 2014, with authors’ permission)
increased co-activation index of antagonist muscles and an increased amplitude and

duration of EMG bursts of all individual muscles have been revealed (Fig. 4a).
As regards the increased co-activation, analyzing muscles of the knee joint
emerged that biceps center of activity of the biceps femoris muscles moved from
90–100 % (late swing) to 0–10 % (loading response phase) overlapping the vastus
lateralis activity (Fig. 4b). This finding suggests that patients stiffen the knee joint
during the loading response phase, in order to compensate for their instability created
by the load transfer from one limb to another. Similarly, an increase for the tibialis
anterior muscle activity, which overlapped the activity of the gastrocnemius
medialis, was observed during the single-support phase indicating that ataxic
patients need to increase the ankle joint stiffness in order to control the forward
progression (Mari et al. 2014).
In essence, ataxic patients need to activate antagonist muscle more and for longer,
possibly in attempt to stiffen the limb and to compensate for the instability due to
poor muscle coordination.
As regards the single muscle activation, on average the amplitude of muscle
activity in ataxic patients was found to be about twice than in healthy subjects
(Martino et al. 2014). The widening of EMG bursts was found also to be asymmetric
since the center of the EMG activity was shifted to later phases of the gait cycle in
proximal muscles and to earlier phases in distal muscles (Martino et al. 2014).
Although the increased EMG widening as well as the antagonist muscle
co-activation may constitute an attempt to provide mechanical stability, by stiffening
the limb, an alternative explanation could be that the broader activity bursts is a
result of primary deficit linked to the cerebellar degeneration. The observed widen-
ing of EMG bursts can possibly be compared to that observed in infants (Dominici
et al. 2011; Ivanenko et al. 2013), which may be determined by the developmental
state of the cerebellum. Maturation of gait is accompanied by a more selective and
flexible control of muscles, with shorter activations and an evident separation of the
distinct bursts (Dominici et al. 2011; Ivanenko et al. 2013). Thus in cerebellar ataxia
patients, widening of muscle activation patterns and shifts in the center of activity
may be a consequence of improper motor planning (feed-forward control) and
processing of proprioceptive information (Bastian 2011) leading to inaccurate
movements. Broader activation patterns likely imply higher metabolic cost and
may also limit adaptation to different walking conditions and coordination with
voluntary movements that require appropriate activation timings/duration (Ivanenko
et al. 2005).
Nonstationary Locomotion
Nonstationary locomotion, including gait initiation, turning while walking, and

stopping, are all motor tasks frequently performed in the course of everyday activ-
ities. Epidemiological studies have revealed a high rate of falls during nonstationary
locomotion both in the elderly and in several neurological diseases (Gryfe et al.
1977; Masud and Morris 2001; Stolze et al. 2004).
Since falls are very common in patients with cerebellar ataxias, with 93 %
reporting one or more falls in 12 months (van de Warrenburg et al. 2005), a full
characterization of these nonstationary locomotor activities could be useful in order
to develop a task-specific rehabilitation approach aimed at improving the mainte-
nance of balance and preventing falls.
Gait Initiation
Gait initiation is a sequence of voluntary commands and coordinated anticipatory

postural adjustments (APAs) with the two lower extremities playing two embedded
functional roles. One role is to control the postural perturbation at the beginning
of the gait initiation process, i.e., to shift bodyweight prior to foot clearance
(Brenière et al. 1981, 1987) and also to counteract the postural perturbation due
to heel off (Dietrich et al. 1994). The other role is to contribute to generating the
propulsive forces (Brenière and Do 1986; Brenière et al. 1987; Michel and Do
2002) aimed at transitioning the body into steady-state walking. These two func-
tional roles are integrated and sequenced into one motor program (Crenna and
Frigo 1991).
Only few abnormalities of anticipatory postural adjustments have been found in
cerebellar subjects (Timmann and Horak 2001), in spite of the hypothesized role of
the cerebellum in initiating movements (Bodranghien et al. 2016). Both in the
unperturbed and perturbed (backward translation) step conditions, force produc-
tion and step length and step velocity were reduced in cerebellar subjects compared
to controls, suggesting compensatory slowing. Cerebellar subjects also showed an
inability to use predictive information of perturbation amplitude to scale anticipa-
tory postural adjustments compared to control subjects. Nevertheless, in
unperturbed steps, temporal parameters of anticipatory postural adjustments were
preserved in cerebellar subjects. When subjects voluntarily initiated a step in
response to the surface translation, both controls and ataxic patients adapted
by executing the anticipatory postural adjustments for step more rapidly. Further-
more, both controls and patients were able to use online information regarding
perturbation amplitude to scale parameters of step initiation in perturbed stepping.
Overall, these findings suggest that the cerebellum is neither critical for the basic
motor program underlying unperturbed step initiation nor for many adaptive
changes occurring during perturbed step initiation (Timmann and Horak 2001).
Like its role in predictive scaling of automatic postural responses to external
perturbations, the cerebellum appears to be important for predictive adaptation of
anticipatory postural adjustments during step initiation (Timmann and Horak
2001).
Gait Termination
Planned Gait Termination

Gait termination is the transformation from steady-state walking to quite stance
(Hase and Stein 1998). Planned gait termination is achieved through a series of
postural adjustments that serve to control the forward progression of the body and
dissipate all kinetic energy generated by the motion of walking, in order to achieve a
stable, upright posture (Hase and Stein 1998).
In planned gait termination, the central nervous system attributes a role to each
limb. The leading limb produces a braking ground reaction force that reduces the
forward velocity of the CoM (Jian et al. 1993), while the trailing limb supports the
bodyweight at the beginning of the gait termination process, i.e., during the leading
foot heel strike, and then terminates the planned gait termination with a second step,
bringing the feet parallel and shifting the body down and backward.
In patients with cerebellar ataxia, the pattern of gait termination is characterized
by a significant reduction of the whole body CoM speed and step length, as well as a
significant increase of the step width and number of steps required to complete the
task (Conte et al. 2012; Serrao et al. 2013a).
Ataxic patients use two important mechanisms to improve their stability during
gait termination: the slowly gait and the increased step width. In fact, if they cannot
reduce their gait speed, increase the step width; conversely, if they cannot increase
their step width, reduce the gait speed. Similarly, in order to walk and stop safely,
patients reduce step length and increase the number of steps.
The analysis of the stability index, which is a measure of balance, indicates that
patients need to maintain an adequate “safety margin” by increasing the stability
index in both sagittal and frontal planes (Fig. 5).
Finally, the timing of the gait termination phases appears to be conserved in
cerebellar patients, suggesting that their abnormal pattern arises from compensatory
mechanisms instead of motor planning dysfunctions (Conte et al. 2012). These
evidences strongly indicate that cerebellum has an essential role in motor control
during the execution of planned gait termination rather than in the planning of gait
termination as observed in the gait initiation (Conte et al. 2012).
Unplanned Gait Termination (Sudden Stopping)

Sudden stopping occurs in everyday life, for instance, when we stop abruptly in
response to unexpected events (Hase and Stein 1998). Sudden stopping is accom-
plished in a single step, always with one foot forward and the other back (Tirosh and
Sparrow 2003). It has been found that ataxic patients have a real difficulty in sudden
stopping (Serrao et al. 2013a). Patients take longer to stop and with a great number of
steps and are unable to perform a well-defined flexion–extension pattern of the hip
and knee joints. This inability to stop in a coordinated harmonic way induces them to
adopt alternative stopping strategies. Patients occasionally stop with their feet
parallel and the head, trunk, and pelvis were more displayed toward the opposite
a b
0.4 0.4
Lateral position (m)

0.3 0.3
0.2 0.2
XcoM XcoM
CoM CoM
0.1 CoP
0.1 CoP
McoPt McoPt
McoPl McoPl
0 0
3.4 3.6 3.8 4 4.2 4.4 4.6 4.8 5 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
time (s) time (s)
c d
0.55 0.55

0.5 0.5
0.45 0.45
0.4 0.4
0.35 0.35
XcoM XcoM
0.3 CoM 0.3 CoM
CoP CoP
0.25 McoPt 0.25 McoPt
McoPl McoPl
0.2 0.2
3.5 4 4.5 5 5.5 7 7.5 8 8.5 9
time (s) time (s)
Fig. 5 Lateral position of center of pressure (CoP, light grey line), center of mass (CoM, black
line), extrapolated center of mass (XCoM, dashed line), and mean center of pressure (Mcop) during
the trailing and leading foot contact as a function of time. The vertical dashed line represents the
trailing foot contact, and the dotted one represents the contralateral toe off and heel strike during the
trailing and leading foot contact, respectively. (a, b) Ataxic patient and control traces unmatched for
speed. (c, d) Ataxic patient and control traces matched for speed (By Conte et al. 2012, with
authors’ permission)
side of the leading foot in the mediolateral direction (Fig. 6; Serrao et al. 2013a). The
observed increased distance between the contralateral arm and leading foot may help
ataxic patients to stabilize themselves on the frontal plane and further support the
notion that they transfer a forward momentum of the body from the sagittal to frontal
plane (Serrao et al. 2013a). This constitutes an attempt to transform a sudden
stopping and unplanned stopping task into a slow, planned, and safer one.
Turning
Turning while walking is an everyday motor task (Glaister et al. 2007), but it is also a
demanding one, requiring the central nervous system to make anticipatory postural
adjustments (Xu et al. 2004), decelerate the forward motion, and coordinate axial
body-segment reorientation to the new direction of travel and maintaining dynamic
body balance while continuing the ongoing step cycle (Patla et al. 1999). Analysis of
Anterior-posterior displacement Medio-lateral displacement

Head Trunk Pelvis Head Trunk Pelvis
–30 Anterior 25
* Medial * * *
–20 20
15
–10
10
(cm)
(cm)
0 5
0
10
-5
20 Posterior -10 Lateral
a c
30 -15
ARM ipsi ARM contra 80 ARM ipsi ARM contra
–20
Anterior 60 Medial
–15
–10 40
(cm)
(cm)
–5 20
0 0
5 –20
10 –40
Lateral
15 Posterior b –60 d
* p<0.05 Ataxic patients Controls
Fig. 6 (a–c) Mean values (SD) of the minimum head, trunk, and pelvis displacement, with
respect to the leading foot during the last step, in the anterior–posterior and mediolateral directions.
Negative value occurs when the segment exceeds the leading foot (By Serrao et al. 2013a, with
authors’ permission). (b–d) Mean values (SD) of the maximum arm displacement, with respect to
the leading foot during the last step, in the anterior–posterior and mediolateral directions. Negative
value occurs when the segment exceeds the leading foot (By Serrao et al. 2013a, with authors’
permission)
turning has revealed several biomechanical abnormalities in patients with cerebellar

ataxia involving several degrees of turning (Mari et al. 2012; Serrao et al. 2013b). In
particular, it was shown that: (i) ataxic patients are less able than controls to complete
turning tasks; (ii) patients approach turns with a wider stride width and a shorter step
length; (iii) ataxic patients are unable to flexibly adjust their stride width according to
the turning task; (iv) ataxic patients need an increased number of steps to complete
turns. This multistep turning strategy may reflect a sacrificing of turning efficiency
for the sake of stability (Stack et al. 2006). Moreover, the ataxic patients avoid the
spin–turn strategy in order to accomplish the task more safely (Fig. 7). Although
patients took more steps than the controls to turn, the total duration of the turning
task was not significantly different between the two groups. This finding, as well as
the substantially comparable duration of segmental reorientation, suggests that
cerebellar impairments have a very little influence on the time program of turning
as seen for both the gait initiation and gait termination tasks. From this perspective,
the role of cerebellum in turning may be mainly to produce coordinated and smooth
movements by real-time processing (Salman 2002) rather than to regulate the timing
of the necessary motor commands.
b
–1
a
Linear walking 0
cross-over
stride width (spin turn)
1
0
3
1 2
30° ipsil. turn (i):
30° contral. turn (c): increased stride width
reduced stride width
2i -1
2c
0
3c 3i multi-step
1
4i
4c
2
3
4
c Ataxic patient d Control subject

Anterior-posterior direction [m]
Anterior-posterior direction [m]
1 1
0.5 0.5
cue
cue
0 0
right step
right step before the cue
before the cue
–0.5 –0.5
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Medio-lateral direction [m] Medio-lateral direction [m]
Fig. 7 Schematic representation of the stride width modulation according to the turning side (a)
and of the different turning strategies adopted in 90 -contralateral turns (b). Ipsil, ipsilateral.
Contral, contralateral. Footsteps of a representative ataxic patient (five trials) (c) and control subject
(d) during 90 contralateral turns. The white and black triangles indicate the right and left footsteps,
respectively. The consecutive footsteps of the same limb are connected by a gray line. Gray bars
indicate the ideal trajectory. It can be noticed that the control subject has a very regular footsteps
pattern among trials, whereas the ataxic patient shows a great variability (By Mari et al. 2012, with
authors’ permission)
Conclusions
The recent use of modern 3D motion analysis systems to quantify gait in patients
with ataxia gives further insights on the motor deficit and may allow (i) to discern the
complex relationship between the primary deficits and the compensatory mecha-
nisms, (ii) to recognize specific abnormalities and their impact on clinical decision-
making, and (iii) to individualize rehabilitative treatment and better evaluate its
effects over the time.
Cross-References

▶ Surface EMG
References
Bastian AJ (2011) Moving, sensing and learning with cerebellar damage. Curr Opin Neurobiol
21:596–601
Bodranghien F, Bastian A, Casali C, Hallett M, Louis ED, Manto M, Mariën P, Nowak DA,
Schmahmann JD, Serrao M, Steiner KM, Strupp M, Tilikete C, Timmann D, van Dun K
(2016) Consensus paper: revisiting the symptoms and signs of cerebellar syndrome. Cerebellum
15:369–391
Brenière Y, Do MC (1986) When and how does steady state gait movement induced from upright
posture begin? J Biomech 19:1035–1040
Breniere Y, Do MC, Sanchez J (1981) A biomechanical study of gait initiation process. J Biophys
Nucl Med 5:197–205
Brenière Y, Cuong Do M, Bouisset S (1987) Are dynamic phenomena prior to stepping essential to
walking? J Mot Behav 19:62–76. doi:10.1080/00222895.1987.10735400
Chini G, Ranavolo A, Draicchio F, Casali C, Conte C, Martino G, Leonardi L, Padua L, Coppola C,
Pierelli F, Serrao M (2016) Local stability of the trunk in patients with degenerative cerebellar
ataxia during walking. Cerebellum. doi:10.1007/s12311-016-0760-6
Conte C, Serrao M, Casali C, Ranavolo A, Mari S, Draicchio F, Di Fabio R, Monami S, Padua L,
Iavicoli S, Sandrini G, Pierelli F (2012) Planned gait termination in cerebellar ataxias. Cerebel-
lum 11:896–904
Conte C, Pierelli F, Casali C, Ranavolo A, Draicchio F, Martino G, Harfoush M, Padua L,
Coppola G, Sandrini G, Serrao M (2014) Upper body kinematics in patients with cerebellar
ataxia. Cerebellum 13:689–697. doi:10.1007/s12311-014-0586-z
Crenna P, Frigo C (1991) A motor programme for the initiation of forward-oriented movements in
humans. J Physiol 437:635–653
Cuesta-Vargas AI, Galán-Mercant A, Williams JM (2010) The use of inertial sensors system for
human motion analysis. Phys Ther Rev 15:462–473
Dietrich G, Brenière Y, Do MC (1994) Organization of local anticipatory movements in single step
initiation. Hum Mov Sci 13:195–210
Dominici N, Ivanenko YP, Cappellini G, d’Avella A, Mondì V, Cicchese M, Fabiano A, Silei T, Di
Paolo A, Giannini C, Poppele RE, Lacquaniti F (2011) Locomotor primitives in newborn babies
and their development. Science 334:997–999
Earhart GM, Bastian AJ (2000) Form switching during human locomotion: traversing wedges in a
single step. J Neurophysiol 84:605–615
Earhart GM, Bastian AJ (2001) Selection and coordination of human locomotor forms following
cerebellar damage. J Neurophysiol 85:759–769
Ebersbach G, Sojer M, Valldeoriola F, Wissel J, Müller J, Tolosa E, Poewe W (1999) Comparative

analysis of gait in Parkinson’s disease, cerebellar ataxia and subcortical arteriosclerotic enceph-
alopathy. Brain 122:1349–1355
Fonteyn EM, Schmitz-Hubsch T, Verstappen CC, Baliko L, Bloem BR, Boesch S et al (2010) Falls
in spinocerebellar ataxias: results of the EuroSCA Fall Study. Cerebellum 9(2):232
Glaister BC, Bernatz GC, Klute GK, Orendurff MS (2007) Video task analysis of turning during
activities of daily living. Gait Posture 25:289–294
Grimaldi G, Manto M (2012) Topography of cerebellar deficits in humans. Cerebellum 11:336–351
Gryfe CI, Amies A, Ashley MJ (1977) A longitudinal study of falls in an elderly population:
I. Incidence and comorbidity. Age Ageing 6:201–210
Hase K, Stein RB (1998) Analysis of rapid stopping during human walking. J Neurophysiol
80:255–261
Herculano-Houzel S (2010) Coordinated scaling of cortical and cerebellar numbers of neurons.
Front Neuroanat 4:12. doi:10.3389/fnana.2010.00012
Ilg W, Golla H, Thier P, Giese MA (2007) Specific influences of cerebellar dysfunctions on gait.
Brain 130:786–788
Ilg W, Giese MA, Gizewski ER, Schoch B, Timmann D (2008) The influence of focal cerebellar
lesions on the control and adaptation of gait. Brain 131:2913–2927
Ivanenko YP, Cappellini G, Dominici N, Poppele RE, Lacquaniti F (2005) Coordination of
locomotion with voluntary movements in humans. J Neurosci 25:7238–7253
Ivanenko YP, Dominici N, Cappellini G, Paolo AD, Giannini C, Poppele RE, Lacquaniti F (2013)
Changes in the spinal segmental motor output for stepping during development from infant to
adult. J Neurosci 33:3025–3036
Jian Y, Winter DA, Ishac MG, Glichrist L (1993) Trajectory of the body COG and COP during
initiation and termination of gait. Gait Posture 1:9–22
Kozio LF, Budding D, Andreasen N et al (2014) Consensus paper: the cerebellum’s role in
movement and cognition. Cerebellum 13:151–177. doi:10.1007/s12311-013-0511-xo
Mari S, Serrao M, Casali C, Conte C, Ranavolo A, Padua L, Draicchio F, Iavicoli S, Monamì S,
Sandrini G, Pierelli F (2012) Turning strategies in patients with cerebellar ataxia. Exp Brain Res
222:65–75. doi:10.1007/s00221-012-3197-2
Mari S, Serrao M, Casali C, Conte C, Martino G, Ranavolo A, Coppola G, Draicchio F, Padua L,
Sandrini G, Pierelli F (2014) Lower limb antagonist muscle co-activation and its relationship
with gait parameters in cerebellar ataxia. Cerebellum 13:226–236. doi:10.1007/s12311-013-
0533-4
Martino G, Ivanenko YP, Serrao M, Ranavolo A, d’Avella A, Draicchio F, Conte C, Casali C,
Lacquaniti F (2014) Locomotor patterns in cerebellar ataxia. J Neurophysiol 112:2810–2821
Martino G, Ivanenko YP, d’Avella A, Serrao M, Ranavolo A, Draicchio F, Cappellini G, Casali C,
Lacquaniti F (2015) Neuromuscular adjustments of gait associated with unstable conditions. J
Neurophysiol 114:2867–2882. doi:10.1152/jn.00029.2015
Masud T, Morris RO (2001) Epidemiology of falls. Age Ageing 30:3–7
McAndrew Young PM, Dingwell JB (2012) Voluntary changes in step width and step length during
human walking affect dynamic margins of stability. Gait Posture 36:219–224. doi:10.1016/j.
Michel V, Do MC (2002) Are stance ankle plantar flexor muscles necessary to generate propulsive
force during human gait initiation? Neurosci Lett 325:139–143
Mitoma H, Hayashi R, Yanagisawa N, Tsukagoshi H (2000) Characteristics of parkinsonian and
ataxic gaits: a study using surface electromyograms, angular displacements and floor reaction
forces. J Neurol Sci 174:22–39
Morton SM, Bastian AJ (2003) Relative contributions of balance and voluntary leg-coordination
deficits to cerebellar gait ataxia. J Neurophysiol 89:1844–1856
Nowak DA, Timmann D, Hermsdörfer J (2007) Dexterity in cerebellar agenesis. Neuropsychologia
45:696–703
Palliyath S, Hallett M, Thomas SL, Lebiedowska MK (1998) Gait in patients with cerebellar ataxia.
Mov Disord 13:958–964
Patla AE, Adkin A, Ballard T (1999) Online steering: coordination and control of body center of
mass, head and body reorientation. Exp Brain Res 129:629–634
Perry J, Burnfield JM (1992) Gait analysis: normal and pathological function. SLACK Incorpo-
rated, Thorofare
Salman ME (2002) The cerebellum: it’s about time! But timing is not everything – new insights into
the role of the cerebellum in timing motor and cognitive tasks. J Child Neurol 17:1–9
Schniepp R, Wuehr M, Schlick C, Huth S, Pradhan C, Dieterich M, Brandt T, Jahn K (2014)
Increased gait variability is associated with the history of falls in patients with cerebellar ataxia. J
Neurol 261:213–223
Schniepp R, Schlick C, Pradhan C, Dieterich M, Brandt T, Jahn K, Wuehr M (2016) The
interrelationship between disease severity, dynamic stability, and falls in cerebellar ataxia. J
Neurol 263:1409–1417. doi:10.1007/s00415-016-8142-z
Serrao M, Bartolo M, Ranavolo A, Pierelli F (2012a) Analisi quantitativa del cammino e nuove
tecnologie in riabilitazione neurologica. In: Sandrini G, Dattola R (eds) compendio neuroriabi-
litazione. Verduci Editore, Rome, Italy
Serrao M, Pierelli F, Ranavolo A, Draicchio F, Conte C, Don R, Di Fabio R, LeRose M, Padua L,
Sandrini G, Casali C (2012b) Gait pattern in inherited cerebellar ataxias. Cerebellum
11:194–211
Serrao M, Conte C, Casali C, Ranavolo A, Mari S, Di Fabio R, Perrotta A, Coppola G, Padua L,
Monamì S, Sandrini G, Pierelli F (2013a) Sudden stopping in patients with cerebellar ataxia.
Cerebellum 12:607–616
Serrao M, Mari S, Conte C, Ranavolo A, Casali C, Draicchio F, Di Fabio R, Bartolo M, Monamì S,
Padua L, Pierelli F (2013b) Strategies adopted by cerebellar ataxia patients to perform U-turns.
Cerebellum 12:460–468
Stack EL, Ashburn AM, Jupp KE (2006) Strategies used by people with Parkinson’s disease who
report difficulty turning. Parkinsonism Relat Disord 12:87–92
Stolze H, Klebe S, Petersen G, Raethjen J, Wenzelburger R, Witt K, Deuschl G (2002) Typical
features of cerebellar ataxic gait. J Neurol Neurosurg Psychiatry 73:310–312
Stolze H, Klebe S, Zechlin C, Baecker C, Friege L, Deuschl G (2004) Falls in frequent neurological
diseases – prevalence, risk factors and aetiology. J Neurol 251:79–84
Timmann D, Horak FB (2001) Perturbed step initiation in cerebellar subjects: 2. Modification of
anticipatory postural adjustments. Exp Brain Res 141:110–120
Tirosh O, Sparrow WA (2003) Gait termination in young and older adults: effects of stopping
stimulus probability and stimulus delay. Gait Posture 19:243–251
Tong K, Granat MH (1999) A practical gait analysis system using gyroscopes. Med Eng Phys
21:87–94
Topka H, Konczak J, Schneider K, Boose A, Dichgans J (1998) Multijoint arm movements in
cerebellar ataxia: abnormal control of movement dynamics. Exp Brain Res 119:493–503
van de Warrenburg BP, Steijns JA, Munneke M, Kremer BP, Bloem BR (2005) Falls in degenerative
cerebellar ataxias. Mov Disord 20:497–500
Xu D, Carlton LG, Rosengren KS (2004) Anticipatory postural adjustments for altering direction
during walking. J Mot Behav 36:316–326
Gait During Real-World Challenges: Gait
Initiation, Gait Termination, Acceleration,
Deceleration, Turning, Slopes, and Stairs
Michael Orendurff
Abstract
The gait of daily living involves more than walking straight ahead at constant
speed. To be functional in the real world, individuals must competently maneuver
over slopes and stairs, and around corners to reach their desired destinations.
These forms of walking are more challenging than walking straight ahead, and
often gait pathologies are more apparent. Specific strategies are often used to walk
on slopes, stairs, and around corners that involves changes in speed, foot place-
ment, joint kinetics, and stride length – all of which challenge stability and
economy to a much greater degree than walking straight ahead at constant speed.
Keywords
Turning • Corners • Stairs • Slope • Walking • Acceleration • Deceleration
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Gait Initiation and Gait Speed Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Gait Initiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Gait Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Gait Deceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Slopes, Stairs, and Turning Maneuvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Slopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Stairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
M. Orendurff (*)
Motion & Sports Performance Laboratory, Lucile Packard Children’s Hospital Stanford, Sunnyvale,
CA, USA
e-mail: morendurff@stanfordchildrens.org

DOI 10.1007/978-3-319-30808-1_47-1
2 M. Orendurff
Introduction
Human gait has been the focus of a substantial number of investigations. Many have
focused on joint motions (Hughes and Jacobs 1979; Murray et al. 1964, 1966, 1970,
1984; Saunders et al. 1953), moments, and powers using three-dimensional inverse
dynamics approaches (Davis et al. 1991; Kadaba et al. 1989, 1990; Ramakrishnan
and Kadaba 1991). Straight ahead, steady-state walking across a range of speeds
comprises the vast majority of nondisabled human gait studies, although work has
expanded into turning (Courtine and Schieppati 2003a, b; Glaister et al. 2007a, b, c, ,
2008, 2009; Huxham et al. 2006, 2008; Orendurff et al. 2006a, b, c; Strike and
Taylor 2009; Taylor et al. 2005, 2007; Thigpen et al. 2000), obstacle avoidance
(Cumming and Klineberg 1994; Xu 1994; Xu et al. 2004, 2006), walk-to-run
transitions (Hreljac 1993, 1995, 2005; Hreljac et al. 2007; Neptune and Sasaki
2005; Prilutsky and Gregor 2001; Sasaki and Neptune 2006a, b), and even backward
walking (Carpenter et al. 1998; Lacquaniti et al. 2012; Yang et al. 2005). Human gait has
been quite well described and is reasonably well understood, but there has been very
limited research into how humans link steps together for functional community mobility
in real-world settings.
In typical metropolitan settings where individuals work, recreate, shop, and
commute, a particular level of mobility is necessary to meet the travel demands of
daily activities within a specific community. These travel demands have a substantial
environmental component and may be dependent upon terrain, climate, season,
weather, architecture and building practices, transportation options, surface condi-
tions, ambient lighting, or other factors. For individuals with gait disabilities to
become fully integrated in the community in which they live, they must meet or
exceed the functional capacity required to be mobile within their specific environ-
ment. This mobility demand has been assessed by several authors using question-
naires or surveys asking individuals to state what they would like to accomplish in
the community, which activities they perform, and which activities they avoid
(Legro et al. 1999; Shumway-Cook et al. 2005). The effects of an intervention
should be evaluated while individuals perform the type of gait they need to achieve
functional mobility, the type of gait they find especially challenging, or the type of
gait that is perceived to be the most important by the individual. Most current gait
outcome studies evaluate a type of gait that is simply convenient to study. For
example, 92.5% of individuals with lower-limb amputations who use prosthetic
limbs for mobility identified walking on sidewalks and streets as important to
them (Legro et al. 1999). Unfortunately, rehabilitation therapies are not always
focused upon functional community ambulation activities (Latham et al. 2005) but
instead focus on the types of walking possible within the therapeutic setting (hospital
or clinic corridors). The long, straight hallways in many institutional and clinical
settings bias gait observation and gait rehabilitation to long-duration straight walking
and may not reflect the specific challenges typically encountered in the gait of daily
living. Gait velocity has been chosen by several authors as an indicator of overall
functional ability (De Quervain et al. 1996; Shumway-Cook et al. 2007), success as a
Gait During Real-World Challenges: Gait Initiation, Gait Termination. . . 3
community ambulator (Lord et al. 2004), or other morbidity or mortality metrics.

Some researchers have focused on endurance, suggesting that individuals who are
unable to walk 400 m at once have a greater level of disability than those individuals
who can walk for this extended distance (Gill et al. 2006). Others have used both,
suggesting that speed on a 4 m walk test can predict success on a 400 m walk test for
older adults (Stepien et al. 2007). Using gait velocity or gait endurance as an
outcome measure in judging the efficacy of an intervention may be less useful
than focusing on improving performance on more typical walking behaviors needed
for successful and competent community ambulation. It is hypothesized that func-
tional walking endurance might be best described as low-intensity interval training,
with many short-duration walking bouts strung together with short rest periods. All
of the tests developed by researchers and clinicians alike are based on logical
assumptions of the types of walking necessary for community ambulation and
have obvious face validity but lack any strong evidence to support the assumptions.
Few authors have evaluated the relationship between self-reported problems with
community ambulation and direct observational studies of selected activities that
challenge walking ability. Shumway-Cook et al.(Shumway-Cook et al. 2005) com-
pared self-reported community mobility to direct observation of ambulation ability
during frequently occurring trips into the community in elderly subjects with
(n = 17) and without (n = 34) walking difficulty. The results revealed statistically
significant correlations between the self-reported mobility and observed mobility,
although r values of the eight dimensions did not account for a large portion of the
variance in either group (range of r = 0.02–0.48 represents 0.04–23% of the
variance in common). Some questions remain as to the level of speed, endurance,
maneuverability, efficiency, stability, safety, etc., of human gait that are necessary for
optimal functional community ambulation.
Initiation, termination, and maneuvering aspects of gait are often neglected
because they generally occur outside the capture volumes of the typical gait analysis
laboratory. These movements are much too complex to be easily described in
repeating cyclic curves because each step serves a unique purpose. However, nearly
all human gait is likely of very short durations (Glaister et al. 2007a; Orendurff et al.
2008b; Roos et al. 2012), and therefore to meet the functional performance require-
ments of moving about in the real world, gait initiation, acceleration, maneuvering
and turning, deceleration, and termination are required.
State of the Art
Data collected on humans walking in ecologically valid environments is extremely

sparse (Glaister et al. 2007a, b, c; Shumway-Cook et al. 2007). The real world – both
the built and the natural environment – is extremely varied and requires gait
initiation, acceleration, turning and maneuvering, deceleration, and termination to
negotiate the encountered challenges of everyday life. Humans must also climb and
descend stairs, ramps and side slopes, as well as uneven surfaces that challenge
4 M. Orendurff
balance and control. Greater movement capability means greater exposure to chal-
lenging environments and therefore more complex environments to negotiate.
These environmental barriers – speed changes, slopes, and corners are more
challenging than walking straight ahead at constant speed – more challenging from
a mechanical, metabolic, and control standpoint. Individuals with gait disabilities
have greater difficulties with these complex multidirectional acceleration, turning,
and deceleration challenges and often exhibit larger deviations during these events
than during walking at constant speed straight ahead (Gill et al. 2006; Hase and Stein
1999). For example, individuals with gait disability are slowest during the sit to
stand and turning portion of a get-up-and-go test when compared to individuals
without gait pathology (Deathe and Miller 2005; Podsiadlo and Richardson 1991;
Wall et al. 2000).
How humans modulate their joint kinematics and kinetics to accomplish these
complex gait tasks has received some attention in the literature, at least enough for a
better understanding of the strategies and adaptations with some gait pathologies.
But the complexities associated with quantifying a dynamic and progressive move-
ment pattern has made it more difficult for scientists to study. The lack of a cyclic
context for many of these complex movements makes it difficult to graph and
understand in terms of the gait cycle. However for many individuals turning and
purposefully altering their gait speed remain everyday challenges that limit their
participation. The extent that these additional movement types are utilized during
activities of daily living is poorly understood at present.
Evidence is extremely sparse on how humans use walking for mobility in the real
world. Technology to record human gait in real-world environments is in its infancy
and is poorly developed with incomplete kinematic data and short-lived power
profiles that reduce the time span that real-world human gait can be accurately
recorded. What evidence that is available is limited to simple multiaxial sensors if
the duration is over many days, weeks, or months. The evidence that is published
suggests that for all environments studied most human walking ends after extremely
short distances (Klute et al. 2006; Orendurff et al. 2008a, b, 2010; Roos et al. 2012).
In a small study of healthy employed adults in a western metropolitan setting, the
number of steps taken before stopping was predominately short: the most common
walking “bout” (the number of steps in a row) was 4 1 steps in a row, accounting
for about 15% of all walking bouts. Forty percent of all walking bouts were 12 1
steps in a row or less, and 75% of all walking bouts were 40 1 steps in a row or less
(Orendurff et al. 2008a, b). Even individuals in the cohort who walked for health for
an hour straight each day or who were participating in organized sports had a similar
distribution of short walking bouts (Orendurff et al. 2008a, b). The least active
individual (<4000 steps a day) and the most active individual (>10,000 steps a day)
had very similar bout distributions dominated by very short durations (4 1 steps in
a row) that accounted for about 15% of total daily walking bouts (Orendurff et al.
2008a, b). Although this sample is quite small, the range represented in activity level
is quite broad. These data suggest that perhaps a third of all steps humans take in the
real world are not straight ahead at constant speed on level ground. Walking indoors
often involves turning and maneuvering around obstacles like chairs, sofas, tables,
hallways, doorways, pets, children, toys, and other architectural barriers.
In a video task analysis of typical real-world walking challenges (walking from
one office to another, walking through a cafeteria, shopping in a convenience store,
or walking from an office out to a parking lot), turning was frequent, and between
15% and 50% of all steps were taken to complete these representative real-world
tasks (Glaister et al. 2007a). All of these typical walking tasks are likely to involved
gait initiation, acceleration, turning, deceleration, and gait termination and began or
ended in standing or seated postures.
Gait Initiation and Gait Speed Changes
Gait Initiation
Even the relatively simple task of initiating gait is a fairly complex challenge. From a
standing posture, with the weight borne equally on both feet, gait is initiated first by
applying greater vertical and posterolateral force onto the limb that will become the
swing limb (Breniere and Do 1986, 1991; Brunt et al. 1991; Elble et al. 1994; Mann
et al. 1979; Miller and Verstraete 1999; Nissan and Whittle 1990). These paradoxical
mechanisms suggest that the individual must enter single limb stance before the
contralateral swing limb advances forward. This confounded the first investigators of
gait initiation when they examined the force platform center of pressure (CoP) data
of the first studies of gait initiation. Counter intuitively, the CoP first moves toward
the swing limb as greater force is applied to accelerate the center of mass toward the
(opposite) stance limb. The CoP also moves posteriorly, and acceleration due to
gravity and the decrease in plantarflexion activity allows the body to fall forward
(Breniere and Do 1991; Elble et al. 1994; Mann et al. 1979; Nissan and Whittle
1990). Once the CoP is posterior and the forward fall has been initiated, the swing
limb is advanced forward, and single limb stance begins. The first step is generally
quite short, commensurate with the slow but increasing CoM forward speed
(Breniere and Do 1991; Elble et al. 1994; Mann et al. 1979; Nissan and Whittle
1990). After the initial swing limb contacts the ground, weight is shifted to the lead
limb, the training limb is advanced into swing, and a near typical walking pattern has
begun, with subtle differences related to gait acceleration. These first few steps are at
a slower speed than the preferred walking, but within a few steps the CoM forward
velocity has been accelerated to the desired walking speed. Gait initiation may be the
only portion of human walking that is “controlled falling,” as manipulating the
relationship between the center of mass and the center of pressure into a non-
equilibrium orientation allows gravity to accelerate the body downward and there-
fore forward. After these initial gait initiation events, acceleration due to gravity does
not appear to be utilized to increase or decrease walking speed, but the exchange of
potential energy (position related, mass times acceleration due to gravity) and kinetic
energy (velocity related, one half mass times velocity squared) is strategically
modulated using tactical joint motions and joint forces.
6 M. Orendurff
Gait Acceleration
Very little work has also been published on how humans modulate their walking
speed, but it is clear from any observation of human walking that changes of speed
are frequent and may cover a large range. The top walking speed for most individuals
is about 2 m/s (dependent upon stature and leg length), and generally after this speed,
a running gait is adopted (Hreljac 1993). There is a plethora of research on the walk-
to-run transition (Hreljac 1993, 1995; Hreljac et al. 2007; Lai et al. 2015; Neptune
and Sasaki 2005; Nilsson and Thorstensson 1989; Prilutsky and Gregor 2001; Ruina
et al. 2005; Sasaki and Neptune 2006a, b), but precious little on the 0.2–2 m/s
acceleration that must take place in functional human walking to navigate the real
world. Changing speed while walking obviously occurs – walking faster in a
crosswalk to avoid oncoming traffic, increasing walking speed to achieve an ener-
getically optimal velocity, or hurrying to an eagerly awaited location. There is
limited published data on what mechanisms humans might use to increase or
decrease their walking speed and only a single proposed hypothesis: the modulation
of the braking shear forces during early stance by modulating the sagittal ankle
moment with the triceps surae (Orendurff et al. 2008a, b). This results in a reduced
slope in the sagittal ankle moment, reducing the braking force and preventing the
slowing of the CoM during early single limb stance.
At constant walking speed, the CoM forward velocity is oscillating by about
12 % of the average walking speed twice in each gait cycle: the fastest forward
velocity occurs a little after the lowest vertical position in double support, and the
slowest forward velocity occurs slightly after the highest position of the CoM in
single limb stance (Orendurff et al. 2008a, b). It is this deceleration during the
elevation of the CoM that is hypothetically reduced to accelerate the body and
increase walking speed. Despite this the CoM still appears to reach the same vertical
maximum during acceleration, providing some evidence against “controlled falling”
as a mechanism of human walking (Orendurff et al. 2004, 2008a, b). If anything, it
might be more correctly called “controlled rising,” because this is what appears to be
altered during acceleration.
It is very attractive to consider the amount of ankle plantarflexor power as a
possible strategic mechanism of increasing walking speed, but what little research
there has been into methods of increasing walking speed has suggested that this
greater power peak at the ankle occurs only after the now faster walking speed has
been achieved (Orendurff et al. 2008a, b). Therefore, it appears that increased power
generation at the ankle is likely a response to walking fast rather than a mechanism
of achieving a faster walking speed. Increased trunk lean and enhanced downward
and therefore forward acceleration due to gravity are also an attractive mechanism,
but this also occurs well after the new faster walking speed has been achieved. In any
case, since this potential to kinetic energy change is cyclic and should achieve a
general equilibrium across gait cycles, it is difficult to imagine using acceleration due
to gravity in a way that would not require some additional joint kinetic mechanism to
return energy to the system and elevate the body to a position when gravity might be
utilized again. Steady-state walking speeds appear to integrate the acceleration due
to gravity into the strategy for achieving very low energy costs, but altering this
complex interchange relationship to increase walking speed for a step or two is
perhaps too complex to make it an attractive mechanism.
Many gait kinematic and kinetic characteristics scale with walking speed: peak
sagittal ankle push-off power, peak sagittal knee flexion in early stance, sagittal hip
range of motion, stride length, minimum CoM position during double support, and
many others (Andriacchi et al. 1977; Kirtley et al. 1985; Lelas et al. 2003; Murray
et al. 1966, 1984, 1985; Peterson et al. 2011; Riley et al. 2001; Segal et al. 2004;
Tesio et al. 1998; Winter 1983), but these all appear to be an adaptation to walking
faster, rather than mechanisms to alter walking speed.
Although the published data here is extremely sparse, separating the mechanism
to alter gait speed from the mechanisms to adapt to a given gait speed might have
substantial and robust control advantages (Orendurff et al. 2008a, b; Peterson et al.
2011). By shifting the responsibility for increasing walking speed to the reduction of
a braking force, acceleration might be achieved by a more energetically favorable
system – essentially saving kinetic energy to accelerate. The “dynamic walking”
hypothesis might be utilized to model this mechanism (Anderson and Pandy 2001;
Buczek et al. 2006; Collins et al. 2005; Full and Koditschek 1999; Garcia et al. 1998;
Kuo 2002, 2007; Morgenroth et al. 2011; Neptune et al. 2004a, b; Weyand et al.
2010; Zelik et al. 2015), but to date no modeler has published data to support or
refute this hypothesis of acceleration in human walking. In conclusion, one hypoth-
esis has been proposed, but little published evidence in support of this hypothesis is
available at present.
Gait Deceleration
The ability to decrease walking speed when desired is also a key skill for functional
community ambulation, but little is definitively known about the joint kinematic and
kinetic strategies used to slow walking velocity. This skill is often used when an
object suddenly appears ahead, for example, when turning a corner and encountering
another person in your path. The strategy here is to attempt to plantarflex rapidly,
increasing the relative forward position of the center of pressure to the forefoot, and
applying greater braking shear to the body to slow rapidly, and then sidestep the
obstacle. There is limited published evidence of a mechanism to decelerate walking
speed, but it appears that deceleration may be achieved by increasing the braking
shear in early single limb stance by creating a slightly increased sagittal plantarflexor
moment (Orendurff et al. 2008a, b), giving a slightly convex appearance to the
usually linear sagittal ankle moment during single limb stance. This subtle kinetic
change is produced without any observable increase in plantarflexion, shifting the
position of the CoP forward relative to the CoM. This results in an increase in the
braking shear impulse in early single limb stance, slowing the body to achieve the
desired decrease in forward velocity.
8 M. Orendurff
Slopes, Stairs, and Turning Maneuvers
Slopes
The basic descriptive data for walking on slopes has received some research
attention (Franz and Kram 2013a, b, 2014; Franz et al. 2012) and has been included
in several studies of rehabilitation devices designed to improve walking ability
(Burnfield et al. 2012; Darter and Wilken 2014; Ehara et al. 1993; Eshraghi et al.
2015; Fisher and Lord 1986; Highsmith et al. 2013; LaPre and Sup 2011; Lura et al.
2014; Stansfield and Nicol 2002; Sup et al. 2010; Tulchin et al. 2010; Vrieling et al.
2008). Walking up, down, or across slopes has been recognized by some as an
important functional task, one that might be hindered by a device designed only for
level walking. For example, rigid (solid) ankle foot orthoses may improve level
walking ability, but might make walking down a ramp more difficult since increased
ankle plantarflexion is needed to achieve foot-flat early in stance phase. With a rigid
AFO the only way to achieve foot-flat is to flex the knee, increasing the effort at the
quadriceps to maintain this position.
The basic mechanism of walking up a slope is somewhat similar to walking at a
faster relative speed. Elevating the body with each step requires greater ankle push-off
power with the trailing limb and greater knee and hip extensor moments with the
lead limb. Step and stride lengths are shorter, and the metabolic work rate is
obviously greater. The sagittal coupling between the hip, knee, and ankle is shifted
temporally in the kinematic and kinetic patterns consistent with a greater need for
vertical displacement of the CoM and in general a slower forward walking speed that
becomes more pronounced as the slope increases. There is progressively greater
ankle dorsiflexion in the first half of the gait cycle (Tulchin et al. 2010) as the slope
of the hill increases.
Uphill walking can be taxing to the cardiopulmonary system and walking up even
small inclines can be noticeable. The joint biomechanics may be a surmountable
challenge for some with musculoskeletal pathology, but the cardiopulmonary load
can limit the duration of uphill walking that individuals with gait pathology can
perform due to their detrained cardiopulmonary state.
Downhill walking is a different challenge that is primarily focused on preventing
the body from continuing to accelerate with each step. In some ways, this resembles
decelerating gait in the lower extremity kinematic and kinetic patterns, but it
becomes markedly different as the slope increases. Increased plantarflexion after
initial contact results in greater eccentric force on the tibialis anterior and also
increases eccentric loads on the quadriceps to prevent the knee from flexing too
rapidly or collapsing completely. This eccentric loading with controlled knee flexion
to lower the body is especially challenging to individuals with limb loss who use a
transfemoral prosthesis. The controlled flexing of the prosthetic knee joint down
slopes is a primary improvement seen in some advanced prosthetic knee joints
(primarily microprocessor controlled prosthetic knees). To aid the quadriceps in
eccentrically slowing knee flexion, the ankle plantarflexors are also more active,
shifting the CoP more anteriorly to provide an extensor moment to the knee. This
plantarflexor-knee extensor couple is an essential force sharing strategy for walking

down slopes (or stairs) that appears to smooth the velocity of the CoM and make the
task less fatiguing overall. Walking down slope is nevertheless a more energetically
demanding task than level walking, due primarily to the increase eccentric loads on
the hip, knee, and ankle.
Side slopes present a different challenge, as the ankle must invert or evert to place
the foot flat on the side slope surface (Damavandi et al. 2010, 2012; Dixon and
Pearsall 2010). Many prosthetic feet and orthotics designed to make level walking
easier can have very negative influences on safe side slope walking since they do not
move easily in the coronal plane. This is rarely considered for many of the prescrip-
tion decisions, but generally those who need orthotics or prosthetics feel that the
tradeoff of gaining level walking stability and economy is worth the reduction in
stability for side slope walking and simply avoid this walking challenge whenever
possible.
Stairs
Stair ascent and descent have been the focus of a large number of investigations in
both normal (Andriacchi et al. 1980; Yu et al. 1997a, b) and pathological populations
(Alimusaj et al. 2009; Bellmann et al. 2012; Hajizadeh et al. 2016; Hall et al. 2016;
Lee et al. 2007; Lessi et al. 2012; Salsich et al. 2001; Spanjaard et al. 2008;
Strutzenberger et al. 2011; Wu et al. 2005). Stairs are somewhat similar to sloped
walking in that the center of mass must be elevated or lowered in a controlled
fashion, and this alone requires increased muscular effort and cardiovascular work
compared to level walking. However, the level surface for the foot contact platform
makes ankle kinematics and kinetics quite different when comparing slopes and
stairs (Andriacchi et al. 1980; Bosse et al. 2012; McFadyen and Winter 1988; Mital
et al. 1987; Moseley et al. 2008; Reeves et al. 2008; Sinitski et al. 2012; Tulchin et al.
2010; Yu et al. 1997a, b). Also, stairs are generally of much greater pitch than slopes
in the built environment, and stairs are safer than ramps for similar slopes (Sheehan
and Gottschall 2012). Stairs are generally more tightly governed by building codes
for the tread and riser dimensions (Nemire et al. 2016), making stair ascent and
descent generally a more patterned motor control task due to generally predictable
riser and tread dimensions; slopes are found in many different pitches and often with
natural surfaces that can be more unpredictable. Nearly all stairs in the built
environment have handrails as stairs are recognized as a major injury risk during
falls (Jacobs 2016; Johnson 2012; McCrory et al. 2014; Pennock et al. 2014; Roys
2001; Startzell et al. 2000).
The increased joint moments required to elevate the center of mass during stair
ascent have specific challenges for those with osteoarthritis (Fok et al. 2013; Guo
et al. 2007; Hajizadeh et al. 2016; Hall et al. 2016; Igawa and Katsuhira 2014;
Koyama et al. 2015; Lessi et al. 2012; McAlindon et al. 1999; Peat et al. 2012;
Varnell et al. 2011; Whitchelo et al. 2014; Yu et al. 1997a, b; Zeni and Snyder-
Mackler 2010), limb loss (Aldridge Whitehead et al. 2014; Andrysek et al. 2005; Au
10 M. Orendurff
et al. 2008; Bellmann et al. 2012; Dou et al. 2006; Eshraghi et al. 2015; Hobara et al.
2011; Jones et al. 2006; Lee et al. 2007; Sinitski et al. 2012; Vanicek et al. 2014;
Wolf et al. 2009), old age (Bosse et al. 2012; Hamel and Cavanagh 2004; Hamel
et al. 2005; Hockings et al. 2013; Hsue and Su 2014; Mayagoitia et al. 2017; Shin
and Yoo 2016; Simoneau et al. 1991; Singhal et al. 2015; Startzell et al. 2000), or
other neuromuscular pathology (Bar-Haim et al. 2004, 2008, 2009; Bohannon and
Walsh 1991; Chen et al. 2014; Harries et al. 2015; Lee and Seo 2014; Novak and
Brouwer 2012, 2013; Prasomsri et al. 2014). Stairs require an increased effort to
raise the body with each stride, and this leads to higher compressive forces on joints,
usually at angles of increased flexion for the hip and knee (Andriacchi et al. 1980;
Bjerke et al. 2014; Dimitriou et al. 2015; Fu et al. 2014; Lamontagne et al. 2011;
Mundermann et al. 2008; Rasnick et al. 2016; Salsich et al. 2001; Shelley et al. 1996;
Stansfield and Nicol 2002; Taylor et al. 2004). Therefore, stair ascent is a major
mobility challenge for a wide range of pathologies and a necessary functional task
for the gait of daily living.
Generally walking speed is much slower when ascending stairs and the develop-
ment of peak moments occur at a similarly methodical pace, but considering that the
peak moments for the hip, knee, and ankle are generally higher than walking, this
may be an acceptable strategy overall. During stair ascent, the hip is flexed to ~60
and the knee is flexed to ~50 at initial contact, and both extend ~50 during stance
phase, requiring large extensor moments produced by the hip and knee extensors of
the lead limb, primarily the gluteus maximus and the quadriceps. At the same time
the ankle plantarflexors of the training limb must produce a large moment to
plantarflex the ankle from ~10 dorsiflexion to ~25 of plantarflexion as stance
phase progresses. Therefore, to elevate the body to the next stair, the hip and knee
extensors of the lead limb and the training limb ankle plantarflexors must produce
muscular efforts that are much greater than level walking. These biomechanical
factors also have cardiovascular consequences, taxing both systems substantially
during stair ascent. This can have an important training effect and improve overall
health by initiating adaptive increases in muscular strength and cardiovascular
endurance (Jennings et al. 2016).
Stair descent is a slightly more complex motor task, but with lower peak moment
demands that are centered on slowing the acceleration due to gravity rather than
elevating the body against gravity as in stair ascent. Therefore, the risks are
descending too quickly (falling), and the joint kinematic and kinetic strategy appears
to be focused on a controlled slowing of the movement of the center of mass. For the
lead limb, the hip is flexed to ~20 and the knee near full extension at initial contact,
but the ankle in ~25 of plantarflexion. The ankle plantarflexors produce a substan-
tial plantarflexor moment during early stance phase, contributing to knee extension
through a plantarflexor-knee extension couple. This tends to slow the progress of the
center of mass by placing the center of pressure well forward at initial contact. The
hip extensors contribute primarily to stabilize the hip position maintaining ~20 of
hip flexion throughout early stance and finally flexing to ~35 in late stance. The
knee flexes at a very near constant rate as stance progresses during stair descent.
Combining the hip, knee, and ankle produces substantial extensor moments in late
stance to lower the body onto the lead limb on the next stair tread, and the eccentric
nature of these contractions leads to substantial power absorption, most notably at
the knee extensors.
Turning
Walking straight ahead is a generally rare event that occurs mainly in larger
institutional buildings or outdoors. In one’s home a turn usually occurs every five
steps (Sedgeman et al. 1994) and may be 8–50% of all steps in typical real-world
tasks (Glaister et al. 2007a, b, c). Maneuvering around a corner is a more challenging
task than walking straight ahead (Thigpen et al. 2000) and involves deceleration,
initiating the turn, continuing the turn, terminating the turn, and then accelerating
back to the preferred walking speed. Each step in the process involves accelerating
the center of mass in a different direction, and maintaining stability throughout can
be a challenge for those with gait pathologies. Turning while walking has a higher
risk for falling, and if a fall occurs, the individual is eight times more likely to suffer a
hip fracture. (Cumming and Klineberg 1994).
Although turning is likely an essential skill for safe household and community
mobility, the gait mechanisms to accomplish a turn are little studied and not fully
understood. Adding to the complexity of understanding safe and effective turning
gait mechanisms and strategies, turning involves deceleration prior to the turn and
acceleration after the turn, masking many of the joint kinematic and kinetic changes
and blurring these two distinct tasks that are used in turning gait. The ability to alter
walking speed is also not well understood, and only a single paper has offered a
hypothesis as to a method that might be used to accomplish deceleration and
acceleration during walking (Orendurff et al. 2008a, b), even though changing
walking speed likely occurs often in the gait of daily living.
The hypothesis put forth for acceleration during walking is a subtle reduction in
the ankle plantarflexor moment slope during single limb stance; increased ankle
push-off power appears a mechanism for maintaining walking speed (Orendurff
et al. 2008a). Deceleration is proposed to be accomplished by a subtle increase in the
ankle plantarflexor moment slope during single limb stance (Orendurff et al. 2008a).
Center of mass position and acceleration data do support the hypothesis of “con-
trolled falling” to increase or decrease walking speed. The ankle moment strategies
appear to attenuate (acceleration) or amplify (deceleration) the aft shear impulse
during single limb stance, and this has little effect on the height of the center of mass.
Altering the slope of the ankle plantarflexor moment results in a reduction of the
slowing of the center of mass in single limb stance to initiate acceleration by
reducing the braking impulse, or an increase in the braking impulse in single limb
stance to slow the center of mass. Essentially, reducing the braking force accelerates
the center of mass, and increasing the braking force does the opposite. Once the
desired speed has been achieved, the sagittal ankle power profile returns to the level
needed to maintain the desired speed.
12 M. Orendurff
Turning is initiated by slightly different methods depending on the rate of the

desired turn and the amount of time for preparation. First the body is decelerated by a
subtle increase in the slope of the sagittal ankle plantarflexor moment. Then the
placement of the foot in a more medial position on the side of the desired turn: the
right foot may be planted more medially to initiate a turn to the right (Hase and Stein
1999). Several authors have differentiated turning by “step” and “spin” turns
(Courtine and Schieppati 2003a, b; Hase and Stein 1999; Taylor et al. 2005; Xu
et al. 2006). A “spin” turn is defined by the rotation of a planted foot in the direction
of the desired turn, while a “step” turn is accomplished by multiple foot contacts in
the direction of the turn. Video task analysis of human gait in real-world settings has
reported never observing the “spin” turn, instead suggesting that this maneuver
occurs only in the laboratory of those investigating turning, or as the author dryly
quipped during “ceremonial military marching and do not occur during activities of
daily living.” (Glaister et al. 2007a)
Turning is initiated by foot placement that accelerates the center of mass laterally
in the direction of the desired turn. Lateral foot placement accelerates the center of
mass in the contralateral direction, and medial foot placement accelerates the center
of mass in the ipsilateral direction. This also tends to rotate the trunk in the opposite
direction of the turn, and activation of the hip internal (inside limb) and external
(outside limb) rotators is needed to counteract this tendency and turn the trunk
toward the direction of progression (Glaister et al. 2007b, c, 2008; Orendurff et al.
2006a, b); this appears to occur during single limb stance (Orendurff et al. 2006a, b,
c). The body leans in the direction of the turn, more so at faster turning speeds, and
this requires a relative lengthening of the outside limb and a relative shortening of the
inside limb. This is accomplished by a combination of subtle ankle and knee
kinematics and a shortening of stride length for the inside limb and a maintenance
of stride length (compared to straight walking) of the outside limb. There are several
interacting joint kinematic and kinetic features, and determining which are involved
in initiating the turn and which are compensatory strategies to maintain balance
while leaning, key among these are ankle rotation moments about the transverse
plane. However the dependent variable with the largest effect size – more than ten
times the values of other kinematic or kinetic dependent variables – appears to be the
mediolateral shear forces that are associated with foot placement location during the
initiation, maintenance, and termination of a turn (Orendurff et al. 2006a, b, c). This
suggests that altering shear forces by precisely controlling foot placement locations
relative to the path of the center of mass is the effecting strategy for turning. Small
errors in foot placement locations can cause rapid instability and can initiate falls, so
it is little wonder that those with gait pathologies choose to slow more during turns
than healthy individuals (Wall et al. 2000).
Conclusion
In conclusion, functional mobility in the real world requires more than simple level
walking, straight ahead at constant speed. Gait initiation, termination, speed
changes, turning, and walking up and down slopes and stairs are all required for
safe and effective mobility in both indoor and outdoor environments. The challenges
require modulated control strategies and increased joint kinetics for many of these
tasks. These may tax the musculoskeletal and motor control system to a greater
extent than simple level walking and should be considered when evaluating the
functional mobility of individuals with gait pathology.
Cross-References

▶ Assessing the Impact of Cardiovascular Conditioning on Gait
▶ Clinical Gait Assessment by Video Observation and 2D–Techniques
▶ Crowd Formation Generation and Control
▶ Effects of Knee Replacement on Gait
▶ Effects of Total Hip Arthroplasty on Gait
▶ Functional Capacity Evaluation in Rehabilitation
▶ Functional Crowds
▶ Gait Parameters Estimated Using Magneto and Inertial Measurement Units
▶ Influence of Prosthetic Socket Design and Fitting on Gait
▶ Pedestrians in Traffic Accidents
▶ Pre-participation Physical Evaluation in Sport
▶ Prosthetic Foot Principles and their Influence on Gait
▶ Slips and Falls
▶ Three-dimensional Human Kinematics Estimation Using Magneto and Inertial
Measurement Units
14 M. Orendurff
References
Aldridge Whitehead JM, Wolf EJ, Scoville CR, Wilken JM (2014) Does a microprocessor-
controlled prosthetic knee affect stair ascent strategies in persons with transfemoral amputation?
Clin Orthop Relat Res 472(10):3093–3101
Alimusaj M, Fradet L, Braatz F, Gerner HJ, Wolf SI (2009) Kinematics and kinetics with an
adaptive ankle foot system during stair ambulation of transtibial amputees. Gait Posture 30
(3):356–363
Anderson FC, Pandy MG (2001) Dynamic optimization of human walking. J Biomech Eng 123
(5):381–390
Andriacchi TP, Ogle JA, Galante JO (1977) Walking speed as a basis for normal and abnormal gait
measurements. J Biomech 10(4):261–268
mechanics during stair-climbing. J Bone Joint Surg Am 62(5):749–757
Andrysek J, Naumann S, Cleghorn WL (2005) Design and quantitative evaluation of a stance-phase
controlled prosthetic knee joint for children. IEEE Trans Neural Syst Rehabil Eng 13
(4):437–443
Au S, Berniker M, Herr H (2008) Powered ankle-foot prosthesis to assist level-ground and stair-
descent gaits. Neural Netw 21(4):654–666
Bar-Haim S, Belokopytov M, Harries N, Frank A (2004) A stair-climbing test for ambulatory
assessment of children with cerebral palsy. Gait Posture 20(2):183–188
Bar-Haim S, Belokopytov M, Harries N, Loeppky JA, Kaplanski J (2008) Prediction of mechanical
efficiency from heart rate during stair-climbing in children with cerebral palsy. Gait Posture 27
(3):512–517
Bar-Haim S, Harries N, Al-Oraibi S, Lahat E, Waddah M, Loeppky JA, Belokopytov M (2009)
Repeatability of net mechanical efficiency during stair climbing in children with cerebral palsy.
Pediatr Phys Ther 21(4):320–324
Bellmann M, Schmalz T, Ludwigs E, Blumentritt S (2012) Stair ascent with an innovative
microprocessor-controlled exoprosthetic knee joint. Biomed Tech (Berl) 57(6):435–444
Bjerke J, Ohberg F, Nilsson KG, Foss OA, Stensdotter AK (2014) Peak knee flexion angles during
stair descent in TKA patients. J Arthroplast 29(4):707–711
Bohannon RW, Walsh S (1991) Association of paretic lower extremity muscle strength and standing
balance with stair-climbing ability in patients with stroke. J Stroke Cerebrovasc Dis 1
(3):129–133
Bosse I, Oberlander KD, Savelberg HH, Meijer K, Bruggemann GP, Karamanidis K (2012)
Dynamic stability control in younger and older adults during stair descent. Hum Mov Sci 31
(6):1560–1570
Breniere Y, Do MC (1986) When and how does steady state gait movement induced from upright
posture begin? J Biomech 19(12):1035–1040
Breniere Y, Do MC (1991) Control of gait initiation. J Mot Behav 23(4):235–240
Brunt D, Lafferty MJ, McKeon A, Goode B, Mulhausen C, Polk P (1991) Invariant characteristics
of gait initiation. Am J Phys Med Rehabil 70(4):206–212
Buczek FL, Cooney KM, Walker MR, Rainbow MJ, Concha MC, Sanders JO (2006) Performance
of an inverted pendulum model directly applied to normal human gait. Clin Biomech (Bristol,
Avon) 21(3):288–296
Burnfield JM, Eberly VJ, Gronely JK, Perry J, Yule WJ, Mulroy SJ (2012) Impact of stance phase
microprocessor-controlled knee prosthesis on ramp negotiation and community walking func-
tion in K2 level transfemoral amputees. Prosthetics Orthot Int 36(1):95–104
Carpenter MG, Bellos A, Patla AE (1998) Is backward stepping over obstacles achieved through a
simple temporal reversal of forward stepping? Int J Neurosci 93(3–4):189–196
Chen CL, Teng YL, Lou SZ, Chang HY, Chen FF, Yeung KT (2014) Effects of an anterior ankle-
foot orthosis on walking mobility in stroke patients: get up and go and stair walking. Arch Phys
Med Rehabil 95(11):2167–2171
Collins S, Ruina A, Tedrake R, Wisse M (2005) Efficient bipedal robots based on passive-dynamic
walkers. Science 307(5712):1082–1085
Courtine G, Schieppati M (2003a) Human walking along a curved path. I. Body trajectory, segment
orientation and the effect of vision. Eur J Neurosci 18(1):177–190
Courtine G, Schieppati M (2003b) Human walking along a curved path. II. Gait features and EMG
patterns. Eur J Neurosci 18(1):191–205
Cumming RG, Klineberg RJ (1994) Fall frequency and characteristics and the risk of hip fractures.
J Am Geriatr Soc 42(7):774–778
Damavandi M, Dixon PC, Pearsall DJ (2010) Kinematic adaptations of the hindfoot, forefoot, and
hallux during cross-slope walking. Gait Posture 32(3):411–415
Damavandi M, Dixon PC, Pearsall DJ (2012) Ground reaction force adaptations during cross-slope
walking and running. Hum Mov Sci 31(1):182–189
Darter BJ, Wilken JM (2014) Energetic consequences of using a prosthesis with adaptive ankle
motion during slope walking in persons with a transtibial amputation. Prosthetics Orthot Int 38
(1):5–11
Davis RI, Ounpuu S, Tyburski D, Gage J (1991) A gait data collection and reduction technique.
Hum Mov Sci 10(5):575–587
De Quervain IA, Simon SR, Leurgans S, Pease WS, McAllister D (1996) Gait pattern in the early
recovery period after stroke. J Bone Joint Surg Am 78(10):1506–1514
Deathe AB, Miller WC (2005) The L test of functional mobility: measurement properties of a
modified version of the timed “up & go” test designed for people with lower-limb amputations.
Phys Ther 85(7):626–635
Dimitriou D, Tsai TY, Li JS, Nam KW, Park KK, Kwon YM (2015) In vivo kinematic evaluation of
total hip arthroplasty during stair climbing. J Orthop Res 33(7):1087–1093
Dixon PC, Pearsall DJ (2010) Gait dynamics on a cross-slope walking surface. J Appl Biomech 26
(1):17–25
Dou P, Jia X, Suo S, Wang R, Zhang M (2006) Pressure distribution at the stump/socket interface in
transtibial amputees during walking on stairs, slope and non-flat road. Clin Biomech (Bristol,
Avon) 21(10):1067–1073
Ehara Y, Beppu M, Nomura S, Kunimi Y, Takahashi S (1993) Energy storing property of so-called
energy-storing prosthetic feet. Arch Phys Med Rehabil 74(1):68–72
Elble RJ, Moody C, Leffler K, Sinha R (1994) The initiation of normal walking. Mov Disord 9
(2):139–146
Eshraghi A, Abu Osman NA, Gholizadeh H, Ali S, Abas WA (2015) Interface stress in socket/
residual limb with transtibial prosthetic suspension systems during locomotion on slopes and
stairs. Am J Phys Med Rehabil 94(1):1–10
Fisher LD, Lord M (1986) Bouncy knee in a semi-automatic knee lock prosthesis. Prosthetics
Orthot Int 10(1):35–39
Fok LA, Schache AG, Crossley KM, Lin YC, Pandy MG (2013) Patellofemoral joint loading
during stair ambulation in people with patellofemoral osteoarthritis. Arthritis Rheum 65
(8):2059–2069
Franz JR, Kram R (2013a) Advanced age affects the individual leg mechanics of level, uphill, and
downhill walking. J Biomech 46(3):535–540
Franz JR, Kram R (2013b) How does age affect leg muscle activity/coactivity during uphill and
downhill walking? Gait Posture 37(3):378–384
Franz JR, Kram R (2014) Advanced age and the mechanics of uphill walking: a joint-level, inverse
dynamic analysis. Gait Posture 39(1):135–140
Franz JR, Lyddon NE, Kram R (2012) Mechanical work performed by the individual legs during
uphill and downhill walking. J Biomech 45(2):257–262
Fu YC, Simpson KJ, Brown C, Kinsey TL, Mahoney OM (2014) Knee moments after
unicompartmental knee arthroplasty during stair ascent. Clin Orthop Relat Res 472(1):78–85
Full RJ, Koditschek DE (1999) Templates and anchors: neuromechanical hypotheses of legged
locomotion on land. J Exp Biol 202(Pt 23):3325–3332
16 M. Orendurff
Garcia M, Chatterjee A, Ruina A, Coleman M (1998) The simplest walking model: stability,
complexity, and scaling. J Biomech Eng 120(2):281–288
Gill TM, Allore HG, Hardy SE, Guo Z (2006) The dynamic nature of mobility disability in older
persons. J Am Geriatr Soc 54(2):248–254
Glaister BC, Bernatz GC, Klute GK, Orendurff MS (2007a) Video task analysis of turning during
activities of daily living. Gait Posture 25(2):289–294
Glaister BC, Orendurff MS, Schoen JA, Klute GK (2007b) Rotating horizontal ground reaction
forces to the body path of progression. J Biomech 40(15):3527–3532
Glaister BC, Schoen JA, Orendurff MS, Klute GK (2007c) Mechanical behavior of the human ankle
in the transverse plane while turning. IEEE Trans Neural Syst Rehabil Eng 15(4):552–559
Glaister BC, Orendurff MS, Schoen JA, Bernatz GC, Klute GK (2008) Ground reaction forces and
impulses during a transient turning maneuver. J Biomech 41(14):3090–3093
Glaister BC, Schoen JA, Orendurff MS, Klute GK (2009) A mechanical model of the human ankle
in the transverse plane during straight walking: implications for prosthetic design. J Biomech
Eng 131(3):034501
Guo M, Axe MJ, Manal K (2007) The influence of foot progression angle on the knee adduction
moment during walking and stair climbing in pain free individuals with knee osteoarthritis. Gait
Posture 26(3):436–441
Hajizadeh M, Hashemi Oskouei A, Ghalichi F, Sole G (2016) Knee kinematics and joint moments
during stair negotiation in participants with anterior cruciate ligament deficiency and recon-
struction: a systematic review and meta-analysis. PM R 8(6):563–579. e561
Hall M, Wrigley TV, Kean CO, Metcalf BR, Bennell KL (2016) Hip biomechanics during stair
ascent and descent in people with and without hip osteoarthritis. J Orthop Res. Aug 30. doi:
10.1002/jor.23407. [Epub ahead of print]
Hamel KA, Cavanagh PR (2004) Stair performance in people aged 75 and older. J Am Geriatr Soc
52(4):563–567
Hamel KA, Okita N, Higginson JS, Cavanagh PR (2005) Foot clearance during stair descent: effects
of age and illumination. Gait Posture 21(2):135–140
Harries N, Loeppky JA, Shaheen S, Al-Jarrah M, Molteni F, Hutzler Y, Bar-Haim S, Project M
(2015) A stair-climbing test for measuring mechanical efficiency of ambulation in adults with
chronic stroke. Disabil Rehabil 37(11):1004–1008
Hase K, Stein RB (1999) Turning strategies during human walking. J Neurophysiol 81
(6):2914–2922
Highsmith MJ, Kahle JT, Miro RM, Mengelkoch LJ (2013) Ramp descent performance with the
C-Leg and interrater reliability of the Hill Assessment Index. Prosthetics Orthot Int 37
(5):362–368
Hobara H, Kobayashi Y, Nakamura T, Yamasaki N, Nakazawa K, Akai M, Ogata T (2011) Lower
extremity joint kinematics of stair ascent in transfemoral amputees. Prosthetics Orthot Int 35
(4):467–472
Hockings RL, Schmidt DD, Cheung CW (2013) Single-leg squats identify independent stair
negotiation ability in older adults referred for a physiotherapy mobility assessment at a rural
hospital. J Am Geriatr Soc 61(7):1146–1151
Hreljac A (1993) Preferred and energetically optimal gait transition speeds in human locomotion.
Med Sci Sports Exerc 25(10):1158–1162
Hreljac A (1995) Determinants of the gait transition speed during human locomotion: kinematic
factors. J Biomech 28(6):669–677
Hreljac A (2005) Etiology, prevention, and early intervention of overuse injuries in runners: a
biomechanical perspective. Phys Med Rehabil Clin N Am 16(3):651–667. vi
Hreljac A, Imamura R, Escamilla RF, Edwards WB (2007) Effects of changing protocol, grade, and
direction on the preferred gait transition speed during human locomotion. Gait Posture 25
(3):419–424
Hsue BJ, Su FC (2014) Effects of age and gender on dynamic stability during stair descent. Arch
Hughes J, Jacobs N (1979) Normal human locomotion. Prosthetics Orthot Int 3(1):4–12
Huxham F, Gong J, Baker R, Morris M, Iansek R (2006) Defining spatial parameters for non-linear
walking. Gait Posture 23(2):159–163
Huxham F, Baker R, Morris ME, Iansek R (2008) Footstep adjustments used to turn during walking
in Parkinson’s disease. Mov Disord 23(6):817–823
Igawa T, Katsuhira J (2014) Biomechanical analysis of stair descent in patients with knee osteoar-
thritis. J Phys Ther Sci 26(5):629–631
Jacobs JV (2016) A review of stairway falls and stair negotiation: lessons learned and future needs
to reduce injury. Gait Posture 49:159–167
Jennings CA, Yun L, Loitz CC, Lee EY, Mummery WK et al (2016) Am J Prev Med 52(1):106–114
Johnson D (2012) Stair safety: bottom of flight illusion. Work 41(Suppl 1):3358–3362
Jones SF, Twigg PC, Scally AJ, Buckley JG (2006) The mechanics of landing when stepping down
in unilateral lower-limb amputees. Clin Biomech (Bristol, Avon) 21(2):184–193
(6):849–860
during level walking. J Orthop Res 8(3):383–392
Kirtley C, Whittle MW, Jefferson RJ (1985) Influence of walking speed on gait parameters.
J Biomed Eng 7(4):282–288
Klute GK, Berge JS, Orendurff MS, Williams RM, Czerniecki JM (2006) Prosthetic intervention
effects on activity of lower-extremity amputees. Arch Phys Med Rehabil 87(5):717–722
Koyama Y, Tateuchi H, Nishimura R, Ji X, Umegaki H, Kobayashi M, Ichihashi N (2015)
Relationships between performance and kinematic/kinetic variables of stair descent in patients
with medial knee osteoarthritis: an evaluation of dynamic stability using an extrapolated center
of mass. Clin Biomech (Bristol, Avon) 30(10):1066–1070
Kuo AD (2002) Energetics of actively powered locomotion using the simplest walking model.
J Biomech Eng 124(1):113–120
Kuo AD (2007) The six determinants of gait and the inverted pendulum analogy: a dynamic
walking perspective. Hum Mov Sci 26(4):617–656
Lacquaniti F, Ivanenko YP, Zago M (2012) Patterned control of human locomotion. J Physiol 590
(10):2189–2199
Lai A, Lichtwark GA, Schache AG, Lin YC, Brown NA, Pandy MG (2015) In vivo behavior of the
human soleus muscle with increasing walking and running speeds. J Appl Physiol (1985) 118
(10):1266–1275
Lamontagne M, Varin D, Beaule PE (2011) Does the anterior approach for total hip arthroplasty
better restore stair climbing gait mechanics? J Orthop Res 29(9):1412–1417
LaPre AK, Sup F (2011) Simulation of a slope adapting ankle prosthesis provided by semi-active
damping. Conf Proc IEEE Eng Med Biol Soc 2011:587–590
Latham NK, Jette DU, Slavin M, Richards LG, Procino A, Smout RJ, Horn SD (2005) Physical
therapy during stroke rehabilitation for people with different walking abilities. Arch Phys Med
Rehabil 86(12 Suppl 2):S41–S50
Lee J, Seo K (2014) The effects of stair walking training on the balance ability of chronic stroke
patients. J Phys Ther Sci 26(4):517–520
Lee WC, Frossard LA, Hagberg K, Haggstrom E, Branemark R, Evans JH, Pearcy MJ et al (2007)
Clin Biomech (Bristol, Avon) 22(6):665–673
Legro MW, Reiber G, del Aguila M, Ajax MJ, Boone DA, Larsen JA, Smith DG, Sangeorzan B
(1999) Issues of importance reported by persons with lower limb amputations and prostheses.
J Rehabil Res Dev 36(3):155–163
Lelas JL, Merriman GJ, Riley PO, Kerrigan DC (2003) Predicting peak kinematic and kinetic
parameters from gait speed. Gait Posture 17(2):106–112
18 M. Orendurff
Lessi GC, da Silva Serrao PR, Gimenez AC, Gramani-Say K, Oliveira AB, Mattiello SM (2012)
Male subjects with early-stage knee osteoarthritis do not present biomechanical alterations in the
sagittal plane during stair descent. Knee 19(4):387–391
Lord SE, McPherson K, McNaughton HK, Rochester L, Weatherall M (2004) Community ambu-
lation after stroke: how important and obtainable is it and what measures appear predictive?
Arch Phys Med Rehabil 85(2):234–239
Lura DJ, Wernke MM, Carey SL, Kahle JT, Miro RM, Highsmith MJ (2014) Differences in knee
flexion between the Genium and C-Leg microprocessor knees while walking on level ground
and ramps. Clin Biomech (Bristol, Avon) 30(2):175–181
Mann RA, Hagy JL, White V, Liddell D (1979) The initiation of gait. J Bone Joint Surg Am 61
(2):232–239
Mayagoitia RE, Harding J, Kitchen S (2017) Identification of stair climbing ability levels in
community-dwelling older adults based on the geometric mean of stair ascent and descent
speed: the GeMSS classifier. Appl Ergon 58:81–88
McAlindon TE, Wilson PW, Aliabadi P, Weissman B, Felson DT (1999) Level of physical activity
and the risk of radiographic and symptomatic knee osteoarthritis in the elderly: the Framingham
study. Am J Med 106(2):151–157
McCrory JL, Chambers AJ, Daftary A, Redfern MS (2014) Ground reaction forces during stair
locomotion in pregnant fallers and non-fallers. Clin Biomech (Bristol, Avon) 29(2):143–148
McFadyen BJ, Winter DA (1988) An integrated biomechanical analysis of normal stair ascent and
descent. J Biomech 21(9):733–744
Miller CA, Verstraete MC (1999) A mechanical energy analysis of gait initiation. Gait Posture 9
(3):158–166
Mital A, Fard HF, Khaledi H (1987) A biomechanical evaluation of staircase riser heights and tread
depths during stair-climbing. Clin Biomech (Bristol, Avon) 2(3):162–164
Morgenroth DC, Segal AD, Zelik KE, Czerniecki JM, Klute GK, Adamczyk PG, Orendurff MS,
Hahn ME, Collins SH, Kuo AD (2011) The effect of prosthetic foot push-off on mechanical
loading associated with knee osteoarthritis in lower extremity amputees. Gait Posture 34
(4):502–507
Moseley AM, Descatoire A, Adams RD (2008) Observation of high and low passive ankle
flexibility in stair descent. Percept Mot Skills 106(1):328–340
Mundermann A, Dyrby CO, D'Lima DD, Colwell CW Jr, Andriacchi TP (2008) In vivo knee
loading characteristics during activities of daily living as measured by an instrumented total
knee replacement. J Orthop Res 26(9):1167–1172
Murray MP, Drought AB, Kory RC (1964) Walking patterns of normal men. J Bone Joint Surg Am
46:335–360
Murray MP, Kory RC, Clarkson BH, Sepic SB (1966) Comparison of free and fast speed walking
patterns of normal men. Am J Phys Med 45(1):8–23
Murray MP, Kory RC, Sepic SB (1970) Walking patterns of normal women. Arch Phys Med
Rehabil 51(11):637–650
Murray MP, Mollinger LA, Gardner GM, Sepic SB (1984) Kinematic and EMG patterns during
slow, free, and fast walking. J Orthop Res 2(3):272–280
Murray MP, Spurr GB, Sepic SB, Gardner GM, Mollinger LA (1985) Treadmill vs. floor walking:
kinematics, electromyogram, and heart rate. J Appl Physiol 59(1):87–91
Nemire K, Johnson DA, Vidal K (2016) The science behind codes and standards for safe walkways:
changes in level, stairways, stair handrails and slip resistance. Appl Ergon 52:309–316
Neptune RR, Sasaki K (2005) Ankle plantar flexor force production is an important determinant of
the preferred walk-to-run transition speed. J Exp Biol 208(Pt 5):799–808
Neptune RR, Zajac FE, Kautz SA (2004a) Muscle force redistributes segmental power for body
progression during walking. Gait Posture 19(2):194–205
Neptune RR, Zajac FE, Kautz SA (2004b) Muscle mechanical work requirements during normal
walking: the energetic cost of raising the body’s center-of-mass is significant. J Biomech 37
(6):817–825
Nilsson J, Thorstensson A (1989) Ground reaction forces at different speeds of human walking and
running. Acta Physiol Scand 136(2):217–227
Nissan M, Whittle MW (1990) Initiation of gait in normal subjects: a preliminary study. J Biomed
Eng 12(2):165–171
Novak AC, Brouwer B (2012) Strength and aerobic requirements during stair ambulation in persons
with chronic stroke and healthy adults. Arch Phys Med Rehabil 93(4):683–689
Novak AC, Brouwer B (2013) Kinematic and kinetic evaluation of the stance phase of stair
ambulation in persons with stroke and healthy adults: a pilot study. J Appl Biomech 29
(4):443–452
Orendurff MS, Segal AD, Klute GK, Berge JS, Rohr ES, Kadel NJ (2004) The effect of walking
speed on center of mass displacement. J Rehabil Res Dev 41(6A):829–834
Orendurff MS, Schoen J, Glaister BC, Bernatz GC, Huff EA, Segal A, Klute GK (2006a) Joint
rotation torques during a common turning task. Gait Posture 24S:201–203
Orendurff MS, Schoen JA, Glaister BC, Bernatz GC, Huff EA, Segal A, Klute GK (2006b). Joint
rotation torques during a common turning task. Joint ESMAC-GCMAS Meeting, Amsterdam,
Gait Posture.
Orendurff MS, Segal AD, Berge JS, Flick KC, Spanier D, Klute GK (2006c) The kinematics and
kinetics of turning: limb asymmetries associated with walking a circular path. Gait Posture 23
(1):106–111
Orendurff MS, Bernatz GC, Schoen JA, Klute GK (2008a) Kinetic mechanisms to alter walking
speed. Gait Posture 27(4):603–610
Orendurff MS, Schoen JA, Bernatz GC, Segal AD, Klute GK (2008b) How humans walk: bout
duration, steps per bout, and rest duration. J Rehabil Res Dev 45(7):1077–1089
Orendurff MS, Do VK, Newman C, Williams A (2010) How children walk: bout length during real-
world locomotor behavior. Int J Exerc Sci 4(1):1–2
Peat G, Duncan RC, Wood LR, Thomas E, Muller S (2012) Clinical features of symptomatic
patellofemoral joint osteoarthritis. Arthritis Res Ther 14(2):R63
Pennock AT, Gantsoudes GD, Forbes JL, Asaro AM, Mubarak SJ (2014) Stair falls: caregiver’s
“missed step” as a source of childhood fractures. J Child Orthop 8(1):77–81
Peterson CL, Kautz SA, Neptune RR (2011) Braking and propulsive impulses increase with speed
during accelerated and decelerated walking. Gait Posture 33(4):562–567
elderly persons. J Am Geriatr Soc 39(2):142–148
Prasomsri J, Jalayondeja C, Bovonsunthonchai S, Khemthong S (2014) Walking and stair climbing
abilities in individuals after chronic stroke with and without mental health problem. J Med
Assoc Thail 97(Suppl 7):S10–S15
Prilutsky BI, Gregor RJ (2001) Swing- and support-related muscle actions differentially trigger
human walk-run and run-walk transitions. J Exp Biol 204(Pt 13):2277–2287
Ramakrishnan HK, Kadaba MP (1991) On the estimation of joint kinematics during gait. J Biomech
24(10):969–977
Rasnick R, Standifird T, Reinbolt JA, Cates HE, Zhang S (2016) Knee joint loads and surrounding
muscle forces during stair ascent in patients with total knee replacement. PLoS One 11(6):
e0156282
Reeves ND, Spanjaard M, Mohagheghi AA, Baltzopoulos V, Maganaris CN (2008) The demands of
stair descent relative to maximum capacities in elderly and young adults. J Electromyogr
Kinesiol 18(2):218–227
Riley PO, DellaCroce U, Kerrigan DC (2001) Effect of age on lower extremity joint moment
contributions to gait speed. Gait Posture 14(3):264–270
Roos MA, Rudolph KS, Reisman DS (2012) The structure of walking activity in people after stroke
compared with older adults without disability: a cross-sectional study. Phys Ther 92
(9):1141–1147
Roys MS (2001) Serious stair injuries can be prevented by improved stair design. Appl Ergon 32
(2):135–139
20 M. Orendurff
Ruina A, Bertram JE, Srinivasan M (2005) A collisional model of the energetic cost of support work
qualitatively explains leg sequencing in walking and galloping, pseudo-elastic leg behavior in
running and the walk-to-run transition. J Theor Biol 237(2):170–192
Salsich GB, Brechter JH, Powers CM (2001) Lower extremity kinetics during stair ambulation in
patients with and without patellofemoral pain. Clin Biomech (Bristol, Avon) 16(10):906–912
Sasaki K, Neptune RR (2006a) Differences in muscle function during walking and running at the
same speed. J Biomech 39(11):2005–2013
Sasaki K, Neptune RR (2006b) Muscle mechanical work and elastic energy utilization during
walking and running near the preferred gait transition speed. Gait Posture 23(3):383–390
Saunders JB, Inman VT, Eberhart HD (1953) The major determinants in normal and pathological
gait. J Bone Joint Surg Am 35-A(3):543–558
Sedgeman R, Goldie P, Iansek R 1994. Development of a measure of turning during walking.
Proceedings of the Inaugural Conference of the Faculty of Health Sciences, La Trobe University,
Melbourne, Australia, pp 26–31
Segal A, Rohr E, Orendurff M, Shofer J, O’Brien M, Sangeorzan B (2004) The effect of walking
speed on peak plantar pressure. Foot Ankle Int 25(12):926–933
Sheehan RC, Gottschall JS (2012) At similar angles, slope walking has a greater fall risk than stair
walking. Appl Ergon 43(3):473–478
Shelley FJ, Anderson DD, Kolar MJ, Miller MC, Rubash HE (1996) Physical modelling of hip joint
forces in stair climbing. Proc Inst Mech Eng H 210(1):65–68
Shin SS, Yoo WG (2016) The effects of gait time and trunk acceleration ratio during stair climbing
in old-old adult females. J Phys Ther Sci 28(7):2025–2026
Shumway-Cook A, Patla A, Stewart AL, Ferrucci L, Ciol MA, Guralnik JM (2005) Assessing
environmentally determined mobility disability: self-report versus observed community mobil-
ity. J Am Geriatr Soc 53(4):700–704
Shumway-Cook A, Guralnik JM, Phillips CL, Coppin AK, Ciol MA, Bandinelli S, Ferrucci L
(2007) Age-associated declines in complex walking task performance: the walking InCHIANTI
toolkit. J Am Geriatr Soc 55(1):58–65
Simoneau GG, Cavanagh PR, Ulbrecht JS, Leibowitz HW, Tyrrell RA (1991) The influence of
visual factors on fall-related kinematic variables during stair descent by older women.
J Gerontol 46(6):M188–M195
Singhal K, Kim J, Casebolt J, Lee S, Han KH, Kwon YH (2015) Gender difference in older adult's
utilization of gravitational and ground reaction force in regulation of angular momentum during
stair descent. Hum Mov Sci 41:230–239
Sinitski EH, Hansen AH, Wilken JM (2012) Biomechanics of the ankle-foot system during stair
ambulation: implications for design of advanced ankle-foot prostheses. J Biomech 45
(3):588–594
Spanjaard M, Reeves ND, van Dieen JH, Baltzopoulos V, Maganaris CN (2008) Influence of step-
height and body mass on gastrocnemius muscle fascicle behavior during stair ascent. J Biomech
41(5):937–944
Stansfield BW, Nicol AC (2002) Hip joint contact forces in normal subjects and subjects with total
hip prostheses: walking and stair and ramp negotiation. Clin Biomech (Bristol, Avon) 17
(2):130–139
Startzell JK, Owens DA, Mulfinger LM, Cavanagh PR (2000) Stair negotiation in older people: a
review. J Am Geriatr Soc 48(5):567–580
Stepien JM, Cavenett S, Taylor L, Crotty M (2007) Activity levels among lower-limb amputees:
self-report versus step activity monitor. Arch Phys Med Rehabil 88(7):896–900
Strike SC, Taylor MJ (2009) The temporal-spatial and ground reaction impulses of turning gait: is
turning symmetrical? Gait Posture 29(4):597–602
Strutzenberger G, Richter A, Schneider M, Mundermann A, Schwameder H (2011) Effects of
obesity on the biomechanics of stair-walking in children. Gait Posture 34(1):119–125
Sup F, Varol HA, Goldfarb M (2010) Upslope walking with a powered knee and ankle prosthesis:
initial results with an amputee subject. IEEE Trans Neural Syst Rehabil Eng 19(1):71–78
Taylor WR, Heller MO, Bergmann G, Duda GN (2004) Tibio-femoral loading during human gait
and stair climbing. J Orthop Res 22(3):625–632
Taylor MJ, Dabnichki P, Strike SC (2005) A three-dimensional biomechanical comparison between
turning strategies during the stance phase of walking. Hum Mov Sci 24(4):558–573
Taylor MJ, Strike SC, Dabnichki P (2007) Turning bias and lateral dominance in a sample of able-
bodied and amputee participants. Laterality 12(1):50–63
Tesio L, Lanzi D, Detrembleur C (1998) The 3-D motion of the centre of gravity of the human body
during level walking. I. Normal subjects at low and intermediate walking speeds. Clin Biomech
(Bristol, Avon) 13(2):77–82
Thigpen MT, Light KE, Creel GL, Flynn SM (2000) Turning difficulty characteristics of adults aged
65 years or older. Phys Ther 80(12):1174–1187
Tulchin K, Orendurff M, Karol L (2010) The effects of surface slope on multi-segment foot
kinematics in healthy adults. Gait Posture 32(4):446–450
Vanicek N, Strike SC, Polman R (2014) Kinematic differences exist between transtibial amputee
fallers and non-fallers during downwards step transitioning. Prosthetics Orthot Int 39
(4):322–332
Varnell MS, Bhowmik-Stoker M, McCamley J, Jacofsky MC, Campbell M, Jacofsky D (2011)
Difference in stair negotiation ability based on TKA surgical approach. J Knee Surg 24
(2):117–123
Vrieling AH, van Keeken HG, Schoppen T, Otten E, Halbertsma JP, Hof AL, Postema K (2008)
Uphill and downhill walking in unilateral lower limb amputees. Gait Posture 28(2):235–242
Wall JC, Bell C, Campbell S, Davis J (2000) The Timed Get-up-and-Go test revisited: measurement
of the component tasks. J Rehabil Res Dev 37(1):109–113
Weyand PG, Smith BR, Puyau MR, Butte NF (2010) The mass-specific energy cost of human
walking is set by stature. J Exp Biol 213(Pt 23):3972–3979
Whitchelo T, McClelland JA, Webster KE (2014) Factors associated with stair climbing ability in
patients with knee osteoarthritis and knee arthroplasty: a systematic review. Disabil Rehabil 36
(13):1051–1060
Winter DA (1983) Biomechanical motor patterns in normal walking. J Mot Behav 15(4):302–330
Wolf SI, Alimusaj M, Fradet L, Siegel J, Braatz F (2009) Pressure characteristics at the stump/
socket interface in transtibial amputees using an adaptive prosthetic foot. Clin Biomech (Bristol,
Avon) 24(10):860–865
Wu WL, Huang PJ, Lin CJ, Chen WY, Huang KF, Cheng YM (2005) Lower extremity kinematics
and kinetics during level walking and stair climbing in subjects with triple arthrodesis or subtalar
fusion. Gait Posture 21(3):263–270
Xu D (1994). Ground reaction force analysis of changing direction during walking. Amercian
Society of Biomechanics Annual Conference, Columbus, OH
Xu D, Carlton LG, Rosengren KS (2004) Anticipatory postural adjustments for altering direction
during walking. J Mot Behav 36(3):316–326
Xu D, Chow JW, Wang YT (2006) Effects of turn angle and pivot foot on lower extremity kinetics
during walk and turn actions. J Appl Biomech 22(1):74–79
Yang YR, Yen JG, Wang RY, Yen LL, Lieu FK (2005) Gait outcomes after additional backward
walking training in patients with stroke: a randomized controlled trial. Clin Rehabil 19
(3):264–273
Yu B, Kienbacher T, Growney ES, Johnson ME, An KN (1997a) Reproducibility of the kinematics
and kinetics of the lower extremity during normal stair-climbing. J Orthop Res 15(3):348–352
Yu B, Stuart MJ, Kienbacher T, Growney ES, An KN (1997b) Valgus-varus motion of the knee in
normal level walking and stair climbing. Clin Biomech (Bristol, Avon) 12(5):286–293
Zelik KE, Takahashi KZ, Sawicki GS (2015) Six degree-of-freedom analysis of hip, knee, ankle and
foot provides updated understanding of biomechanical work during human walking. J Exp Biol
218(Pt 6):876–886
Zeni JA Jr, Snyder-Mackler L (2010) Preoperative predictors of persistent impairments during stair
ascent and descent after total knee arthroplasty. J Bone Joint Surg Am 92(5):1130–1136
Natural History of Cerebral Palsy
and Outcome Assessment
Erich Rutz and Pam Thomason
Abstract
Cerebral palsy (CP) is defined as “a group of permanent disorders of the devel-
opment of movement and posture, causing activity limitation, that are attributed
to nonprogressive disturbances that occurred in the developing fetal or infant
brain. The motor disorders of CP are often accompanied by disturbances of
sensation, perception, cognition, communication, and behavior, by epilepsy;
and by secondary musculoskeletal problems.” Though the neurological distur-
bance is nonprogressive, the natural history is for deterioration in gait and motor
function due to musculoskeletal pathology.
The natural history and outcomes of musculoskeletal problems are discussed.
The measurement tools and comprehensive assessment used to evaluate gross
motor function and gait impairments in children with CP are presented.
Keywords
Cerebral palsy • Musculoskeletal deformity • Assessment • Gait • Hip displace-
ment • Outcome
E. Rutz (*)
Pediatric Orthopaedic Department, University Children’s Hospital Basel, Basel, Switzerland
e-mail: erich_rutz@hotmail.com
P. Thomason
Hugh Williamson Gait Analysis Laboratory, Royal Children’s Hospital, Melbourne, VIC, Australia
e-mail: Pam.thomason@rch.org.au

DOI 10.1007/978-3-319-30808-1_49-1
2 E. Rutz and P. Thomason
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Natural History, Spasticity, Muscle Force, and Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
The Hip and Spine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Orthopedic Surgery to Improve Gait and Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Outcome Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Body Structure and Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Functional Mobility Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Functional Assessment Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
State of the Art
Natural History, Spasticity, Muscle Force, and Growth
Natural history of gait in children with cerebral palsy (CP) is one of deterioration.
Several studies report on these findings (Norlin and Odenrick 1986; Johnson et al.
1997; Bell et al. 2002; Rose et al. 2010; Morgan and McGinley 2014). The most
important factors that affect the natural history of deterioration, in our opinion, are
the following: increased muscle tone during growth, the development of deformities,
and a decrease of muscular force.
Primary features in children with CP are spasticity, muscle weakness, and lack of
selective motor control. All these factors are in conflict with growth in children with
CP. Lower limbs deformities are worsened by the increase of body weight (Davids
et al. 2015).
A longitudinal study (Johnson et al. 1997) over a mean of 32 months on 18 sub-
jects with spastic diplegia, ranging in age from 4 to 14 years was conducted.
Instrumented gait analyses (IGA) were performed to compare the temporal and
kinematic data across the two time intervals. The comparison revealed a deteriora-
tion of gait stability, evidenced by increases in double support and decreases in
single support time with time and growth. In this study, kinematic analysis revealed a
loss of excursion about the knee, ankle, and pelvis. Additionally, passive range-of-
motion analysis revealed a decrease in the popliteal angle over time. They concluded
that, in contrast to the gait of children with intact motor function, ambulatory ability
tends to worsen over time in spastic CP (Johnson et al. 1997).
Bell et al. (2002) showed a deterioration of gait temporal and stride parameters
and kinematics over time in 28 children with CP. There was no surgical intervention
in this cohort during the study period of 4.4 years. Of the 28, 19 children were
classified with the diagnosis of diplegia, seven with hemiplegia, and two with
quadriplegia. In general, CP is considered a static deformity. But in this study
(Bell et al. 2002), changes in function commonly result with growth and age. At
Natural History of Cerebral Palsy and Outcome Assessment 3
the time of the first 3DGA, recommendations for 14 were conservative (e.g., bracing,
botulinum toxin A). After 4.4 years, at the end of the study, surgery was
recommended in all but two children. Hagglund and Wagner, (2008) reported the
development of spasticity with age in children with CP. In 1994, a register and a
health-care program for children with CP in southern Sweden were initiated. In the
program, the child’s muscle tone according to the Modified Ashworth Scale is
measured twice a year until 6 years of age and then once a year. They used this
data to analyze the development of spasticity with age in a total population of
children with CP. In their study, all measurements of muscle tone in the
gastrocnemius-soleus muscle in all children with CP from 0 to 15 years during the
period 1995–2006 were analyzed. They found that in the total sample, the degree of
muscle tone increased up to 4 years of age. After 4 years of age, the muscle tone
decreased each year up to 12 years of age. A similar development was seen when
excluding the children operated with selective dorsal rhizotomy, intrathecal baclofen
pump, or tendo achilles lengthening. At 4 years of age, 47% of the children had
spasticity in their gastro-soleus muscle graded as Ashworth 2–4. After 12 years of
age, 23% of the children had that level of spasticity.
Hagglund and Wagner, (2008) concluded that, in children with CP, the muscle
tone as measured with the Modified Ashworth Scale increases up to 4 years of age
and then decreases up to 12 years of age.
Morgan and Mc Ginley conducted a systematic review on gait function and
decline in adults with CP. Sixteen studies met the inclusion criteria, and the reported
mean ages of the study groups varied from 22 to 42.6 years of age. Results suggest
that mobility decline occurs in 25% or more of adults with CP. Those at higher risk of
gait decline are those with worse initial gait ability, bilateral rather than unilateral
motor impairment, older age, and higher levels of fatigue and pain. They conclude
that many ambulant adults with CP experience mobility decline earlier than their
nondisabled peers (Morgan and McGinley 2014).
The relationship of strength, weight, age, and function in ambulatory children
with CP was investigated by Davids et al. (2015). The study design was prospective,
case series of 255 subjects, aged 8–19 years, with diplegic type of CP. Linear
regression was used to predict the rate of change in lower extremity muscle strength,
body weight, and strength normalized to weight (STR-N) with age. The cohort was
analyzed as a whole and in groups based on functional impairment as reflected by
Gross Motor Function Classification System (GMFCS) level. In their results, they
found that strength increased significantly over time for the entire cohort at a rate of
20.83 N/y (P = 0.01). Weight increased significantly over time for the entire cohort
at a rate of 3.5 kg/y (P < 0.0001). Lower extremity STR-N decreased significantly
over time for the entire cohort at a rate of 0.84 N/kg/y (P < 0.0001). The rate of
decline in STR-N (N/kg/y) was comparable among age groups of the children in the
study group. Interestingly they stated that there were no significant differences in the
rate of decline of STR-N (N/kg/y) among GMFCS levels. There was a 90% chance
of independent ambulation (GMFCS levels I and II) when STR-N was 21 N/kg (49%
predicted relative to typically developing children). The results of this study (Davids
et al. 2015) support the long-standing clinically based observation that STR-N
Fig. 1 Indicates the typical posture of a child with spastic diplegia
decreases with age in children with CP. This decrease occurs throughout the growing
years and across GMFCS levels I to III. Independent ambulation becomes less likely
as STR-N decreases.
The natural history of gross motor development in children with CP aged
1–15 years was reported by Beckung et al. (2007). These curves may be useful for
monitoring and predicting motor development, for planning treatment, and for
evaluating outcome after interventions.
In all of these studies, the natural history of ambulation in children with CP
consists of deterioration over time. (Norlin and Odenrick 1986; Bell et al. 2002;
Rose et al. 2010; Rutz et al. 2012; Morgan and McGinley 2014; Davids et al. 2015;
Graham et al. 2016).
Figures 1 and 2 present illustrative postures of the lower extermities (1 in bilateral
spastic CP, 2 in unilateral spastic CP).
The Hip and Spine
Soo et al. (2006) reported that hip displacement is common in children with CP, with an
overall incidence of 35% found in their study. The risk of hip displacement is directly
related to gross motor function as graded with the GMFCS (Palisano et al. 1997). There
are four studies from hip surveillance programs, and all report very similar results (Soo
et al. 2006; Connelly et al. 2009; Terjesen 2012; Hagglund et al. 2014). Hips in
ambulatory children with CP are usually stable or easily made so, but hips in children
with GMFCS levels IV and V have unstable hips (Rutz 2012; Rutz et al. 2015).
The untreated hip dislocation is not an easy problem. Pritchett (1983) reported the
incidence of pain and other complications associated with the untreated spastic
dislocated or subluxed hip in 80 institutionalized patients. The average age was
19 years, and the average length of follow-up study was 12 years. Eighty-five
percent of the patients had scoliosis, and 56% had significant pelvic obliquity. A
dislocated hip predisposed to lower extremity fractures but did not present problems
of pain, decubitus ulcers, or difficulties with perineal hygiene. The loss of sitting
balance, scoliosis, and pelvic obliquity are correlated with the severity of neurologic
Fig. 2 Presents a child with spastic hemiplegia on the right side
involvement rather than with the mechanics of a dislocated hip. Several studies
reported pain and other problems as a significant finding (Samilson et al. 1972;
Moreau et al. 1979). Knapp et al. (Knapp and Cortes 2002) reported seven dislocated
hips (18%, out of 29 hips) were definitely painful and four hips (11%) produced only
mild or intermittent pain. Twenty-seven hips (71%) were not painful.
Pruszczynski and Miller (2015) performed a review and reported that when hip
displacement in children with CP is identified early, treatment is more successful.
The review included ten studies with sample sizes greater than 20 children. They
found that the GMFCS level has a strong impact on subluxation risk and that the risk
continues to the end of growth.
Wawrzuta et al. 2016 studied “hip health” in a population-based cohort of
adolescents and young adults with CP to investigate associations between hip
morphology, pain, and gross motor function. Ninety-eight young adults were
reviewed at a mean age of 18 years, 10 months. Hip morphology was classified
using the Melbourne Cerebral Palsy Hip Classification Scale (MCPHCS)
(Murnaghan et al. 2010). Hip pain was reported in 72% of participants. Poor hip
morphology at skeletal maturity was associated with high levels of pain. Limited hip
surveillance and access to surgery, rather than GMFCS, were associated with poor
hip morphology. The majority of young adults with access to hip surveillance and
preventive and reconstructive surgery had satisfactory hip morphology at skeletal
maturity and less pain.
The role of pelvic obliquity is not entirely clear, but its management deserves
closer scrutiny in children and adolescents with CP (Heidt et al. 2015).
Natural history of scoliosis in spastic CP was described very well in the classical
paper of Saito et al. (1998). They concluded that the risk factors for progression of
scoliosis in spastic CP are having a spinal curve of 40 before age 15 years, having
total body involvement, being bedridden, and having a thoracolumbar curve.
Patients with these risk factors might benefit from early surgical intervention to
prevent progression to severe scoliosis (Saito et al. 1998). Gu et al. (2011) performed
a retrospective study to analyze the development and progression of scoliosis in
children and adolescents with nonambulatory spastic quadriplegic CP. A total of
110 children and adolescents <18 years of age with scoliosis were included. Cobb
angle, age, weight, height, history of hip dislocation, and tracheostomy were inves-
tigated as potential risk factors for the development of scoliosis. In conclusion, age
was found to be the most significant predictor of Cobb angle, and the effect of age
was greater in the tracheostomy group than in the nontracheostomy group. After
adjustment for age, the weight and height were not significant predictors of Cobb
angle. Cobb angles of >40 by the age of 12 years were associated with greater
increases in Cobb angle with age (Gu et al. 2011).
Orthopedic Surgery to Improve Gait and Function
The current concept for the management of musculoskeletal deformities in children

with CP is single-event multilevel surgery (SEMLS).
In a systematic review of SEMLS evidence was found for large improvements in
gait dysfunction, moderate improvements in health-related quality of life, and only
small changes in gross motor function (McGinley et al. 2012). In the first random-
ized clinical trial of SEMLS, a 50% improvement in gait function and a 4.9%
improvement in gross motor function (GMFM-66) were reported (Thomason et al.
2011). The 5-year results of this clinical trial show that these improvements were
largely maintained at 5 years after SEMLS (Thomason et al. 2013).
In the SEMLS approach, the gait pattern is identified and evaluated by IGA as
part of the diagnostic matrix (Davids et al. 2003). SEMLS in bilateral spastic CP can
be considered an exercise in correcting anatomical deformities based on the clinical
findings. It is necessary to consider all components of the matrix so that surgical
planning is optimized for the individual child.
Careful preoperative discussions about setting realistic goals help to ensure that
the patient’s, parents’, and surgeon’s goals are consistent and achievable (Thomason
and Graham 2013). A comprehensive plan is then developed for the correction of all
muscle tendon contractures, torsional malalignments, and joint instabilities in one
operative session. Rehabilitation requires at least 1 year and improvements continue
into the second year, postoperatively. A carefully tailored and carefully monitored
rehabilitation program can ensure that the child will reach a higher level of function
(Thomason et al. 2013; Thomason and Graham 2013). Follow-up until skeletal
maturity is recommended, to detect for new or recurrent deformities.
The orthotic prescription must be carefully monitored throughout the first year
after surgery. Functional recovery and orthotic prescription can be monitored by
a gait laboratory visit every 3 months for the first year after surgery and yearly
thereafter. Our approach to SEMLS rehabilitation has been described in more detail
elsewhere (Thomason and Graham 2013).
Multilevel orthopedic surgery for older children and adolescents with severe
crouch gait is effective for relieving stress on the knee extensor mechanism, reducing
knee pain, and improving function and independence (Rodda et al. 2006).
Surgical treatment for equinus gait in children with spastic diplegia was success-
ful, at a mean of 7 years, in the majority of cases when combined with multilevel
surgery, orthoses, and rehabilitation. No patient developed crouch gait, and the rate
of revision surgery for recurrent equinus was 12.5% (Firth et al. 2013).
Classification
Classification of functional levels in children with CP permits description of the level

of impairment, encourages clear communication, and is important to establish gross
motor prognosis, monitoring progress and in choice of appropriate interventions
(Rosenbaum et al. 2002, 2007a). Development of the Gross Motor Functional
Classification System (GMFCS) has given clinicians and parents a common lan-
guage to communicate about CP (Palisano et al. 1997, 2008). The GMFCS is a
classification system, and its use is essential when discussing gross motor function in
children with CP. It should be used alongside the classification of upper limb
function (Manual Ability Classification System) and communication abilities (Com-
munication Function Classification System) to provide the essential context for
considering the individual child’s prognosis, goal setting, management, and outcome
evaluation (Eliasson et al. 2006; Hidecker et al. 2011).
The descriptors and accompanying illustrations are found in Fig. 3a, b. Although
GMFCS levels are considered to be stable over time, it is important to note that the
descriptors for youth aged 13–18 years allow for deterioration in gross motor
function, a decrease in independent walking abilities, and an increase in the need
for both assistive devices and wheeled mobility. This is particularly true at GMFCS
levels II, III, and IV (McCormick et al. 2007).
Certain musculoskeletal features and deformities are closely related to GMFCS
level. The shape of the proximal femur shows a strong correlation with GMFCS
level. (Robin et al. 2008). The incidence and severity of hip displacement are directly
predicted by GMFCS level (Soo et al. 2006). The relationship between GMFCS and
hip displacement has implications for screening and management protocols.
Changes in GMFCS levels should be carefully documented. The GMFCS may not
be stable in the very young child. However, the most common reason for a change in
GMFCS level is an error in interpretation in the previous or current examination.
8
Fig. 3 (a) GMFCS levels descriptors for ages 6–12 years (a) and 13–18 years (b)
E. Rutz and P. Thomason
Given that the GMFCS is a categorical grading system, true changes in GMFCS
level sometimes occur which may be in both directions, namely, improvement or
deterioration. After major intervention such as selective dorsal rhizotomy (SDR) or
single-event multilevel surgery (SEMLS), a few children move up a level. This is
uncommon and should not be expected in more than 5–10% of children (Rutz et al.
2012). Deterioration in GMFCS level is more common. For example, lengthening of
the Achilles’ tendons in children in GMFCS level II can result in progressive crouch
gait and the need for assistive devices. For these children, their gait and function
deteriorate, and GMFCS level changes from II to III (Rodda et al. 2006). The relative
stability of the GMFCS means that this is not an outcome measure and was never
meant to be one. The GMFCS provides an excellent guide to long-term prognosis
and influences our decision making and management plan.
The development of gross motor function in children with CP can be described by
a series of curves (Rosenbaum et al. 2002). Understanding the position of a child’s
development in relation to their gross motor curve provides a rational basis for
understanding management strategies, goal setting, and long-term gross motor
function (Hanna et al. 2008). The popularity of many forms of intervention in
early childhood in children with CP is the mistaken attribution of improvements in
gross motor function to the intervention, when natural history has such a large effect.
Association is not causation. In most children, gross motor function reaches a
plateau between 3 and 6 years, with some regression in later childhood. One of the
causes for this regression in gross motor function is progressive musculoskeletal
pathology (Graham 2004). After age 6 years, gait parameters deteriorate as contrac-
tures and bony deformities increase. It should be noted that the gross motor curves
(Rosenbaum et al. 2002) include GMFM data up to the age of 15 years. There is
much less information regarding changes in gross motor function in the 15–20 age
group and beyond. Changes in gross motor function and gait during the plateau/early
decline phase can be more realistically attributed to intervention.
Outcome Assessment
Previously we have discussed the natural history of the musculoskeletal pathology in

CP. Musculoskeletal impairments affect many aspects of the child’s physical func-
tioning, limiting their levels of physical activity and participation (Graham et al. 2016;
Rosenbaum et al. 2007b). Children with CP often undergo interventions designed to
modify the natural history of musculoskeletal pathologies and improve their gross
motor and gait function, including botulinum toxin A injections, physiotherapy, and
orthopedic surgery (Graham et al. 2016). It is extremely important to be able to
accurately assess the outcomes of these interventions. Outcome measures must con-
sider a child’s level of function across multiple domains (Oeffinger et al. 2009). The
World Health Organization’s International Classification of Functioning, Disability
and Health (ICF) provides a useful framework for the measurement of outcomes
following interventions in children with CP. The ICF considers health conditions
with regard to three domains: body structure and function and activities and
participation. These domains are influenced by environmental and personal factors

(WHO 2001). Various tools exist to measure outcomes relevant to children with CP in
the ICF domains, and new measurement tools are being developed. These are shown
in Fig. 4.
When considering measurement of outcomes in children with CP, it is important
to consider all of the components of the ICF. It is also important to choose the correct
measurement tool to use at any given time or after an intervention. Choice of
measurement tool should be based on the psychometric properties of the tool, the
aspect of the ICF being measured, as well as the age and GMFCS level of the child.
Fig. 4 Measurement tools used for gross motor function in children with CP per ICF domains
A detailed discussion of measurement tools is beyond the scope of this chapter.

Some of the tools that we use commonly are discussed below.
The Diagnostic Matrix
Assessment tools can be considered in the context of a diagnostic matrix (Davids
et al. 2003). Davids and colleagues described a diagnostic matrix consisting of
clinical history, physical examination, diagnostic imaging, instrumented gait analy-
sis (IGA), and examination under anesthesia as five components of a diagnostic
matrix useful in clinical decision making, in relation to gait correction surgery in
children with CP. The diagnostic matrix consists of measurements across multiple
domains of the ICF and includes many aspects of gross motor function. Incorporat-
ing these measurements builds up a comprehensive picture of the individuals’
function which assists with diagnosis and identification of impairments and assist
treatment planning especially in relation to major interventions such as gait improve-
ment surgery. Major interventions, such as selective dorsal rhizotomy and multilevel
orthopedic surgery, are designed to improve gait and functioning in children with
CP. Ideally these major interventions are most appropriately conducted following the
most comprehensive and objective assessment possible. Over time we have added
additional measurement tools to the diagnostic matrix used in the gait laboratory.
Figure. 5 shows the diagnostic matrix and assessment tools included in the matrix.
Fig. 5 Assessment Matrix used for assessment of motor function in CP

Body Structure and Function
Instrumented Gait Analysis

The role of instrumented gait analysis (IGA) is crucial to evaluating gait dysfunction
in children with CP, especially in relation to planning and assessing the outcome of
major interventions such as selective dorsal rhizotomy (SDR) and single-event
multilevel surgery (SEMLS), and is central to the diagnostic matrix.
Three-dimensional kinematics, kinetics, and dynamic electromyography provides
a comprehensive description of joint movements, moments, and powers and muscle
timing which is essential to the management of gait disorders in children with CP
(Baker 2013).
The objectivity and relative freedom from bias of IGA are factors of major
importance in establishing objective outcomes. The use of IGA and a composite
measure of gait such as the Gillette Gait Index (Schutte et al. 2000), the Gait
Deviation Index (McMulkin and MacWilliams 2015), or the gait profile score
(GPS) (Baker et al. 2009) are useful to describe outcomes, of prospective cohort
studies as well as randomized trials of multilevel surgery in children with CP. The
movement analysis profile (MAP) and overall gait profile score (GPS) have been
developed to summarize kinematic data. The root mean square (RMS) difference
between nine clinically relevant kinematic variables for a particular child and the
average values of that variable from typically developing children are calculated.
This represents a clinical meaningful measure of gait variables as it measures
difference in degrees and is of value both clinically and in the research setting to
evaluate change following surgery (Baker et al. 2009). These measures have revo-
lutionized gait outcome assessment. We are now able to document change in gait,
and we can compare kinematic data collected from different centers, allowing the
possibility of large cohort collaborative studies.
Video Gait Analysis

Video gait analysis (VGA) is a central part of the diagnostic matrix. A visual record
of a child’s gait and functioning on digital video can be of much greater value than
observational gait analysis and a written report. Digital video is objective and can be
shared by multiple observers over a long period. It can be replayed in slow motion
and can be reviewed repeatedly. VGA can be used when IGA is either not appro-
priate or not available. We found VGA particularly useful for the objective docu-
mentation of younger children commencing BoNT-A therapy. At this stage, they are
too small, too young, and uncooperative for IGA. It is also useful for the selection
and monitoring of the use of ankle-foot orthoses (AFOs) and for the monitoring of
children after major intervention such as selective dorsal rhizotomy (SDR) or
multilevel surgery. In an effort to quantify and objectify the outcome of observa-
tional gait analysis, several gait scores have been developed. These include the
Physician Rating Scale, the Observational Gait Scale, and the Edinburgh Visual
Gait Score (Koman et al. 1993; Mackey et al. 2003; Read et al. 2003; Wren et al.
2005).
We do not consider VGA an adequate substitute for IGA when decisions regard-
ing major intervention such as selective dorsal rhizotomy or multilevel surgery have
to be made. Nor is VGA adequate for outcome measurement in clinical trials of gait
correction surgery.
Clinical History and Standardized Physical Examination

The clinical history is obtained by a careful review of all current and previous
medical records complemented by an up to date interview of the child in the context
of his family or care providers. Associated medical comorbidities and the response to
previous interventions are also crucial to the planning of interventions. Goals and
expectations can be established using specific tools such as goal attainment scales,
Canadian Occupational Performance Measure and the newly developed Gait Out-
comes Assessment List.
A standardized routine physical examination is an essential part of the diagnostic
matrix. The clinical protocol used in The Hugh Williamson Gait Analysis Labora-
tory has been published elsewhere (Thomason et al. 2013) and will not be described
in detail here.
Sagittal Gait Patterns

We believe that classification of sagittal gait patterns initially from VGA and then
from the information from IGA (sagittal gait kinematics) to be very important. The
classification of sagittal gait patterns in spastic hemiplegia (Winters et al. 1987) is a
valid and reliable tool which helps in framing logical management strategies. The
sagittal gait classification described by Rodda and Graham can be useful in planning
intervention in spastic diplegia (Rodda et al. 2004). Both gait classifications suggest
common patterns of musculoskeletal deformity and may assist in identifying defor-
mities that require intervention.
Standardized Radiology
Medical imaging is important in the diagnostic matrix as high proportions of
children with CP have skeletal deformities including torsional deformities of
long bones and instability of the hip and foot specifically the subtalar and midtarsal
joints. Radiology of the hips, including plain radiographs, supplemented by CT
measurements of femoral torsion and tibial torsion and more recently, biplanar
radiography using EOS (Escott et al. 2013), can be very useful as additional
information in the planning of multilevel surgery. Weight-bearing radiographs
can be analyzed by the measurement of a series of key radiological indices
which can help identify segmental malalignments in the hindfoot, midfoot, and
forefoot in a systematic manner (Davids et al. 2005). This contributes greatly to the
analysis of segmental foot deformity and the planning of intervention.
Instrumented gait analysis most typically interprets the foot as a rigid segment
and does not provide detailed information on segmental malalignments within the
foot. Until better foot models are in routine use, standardized weight-bearing
radiographs remain the cornerstone of analysis of deformities within the foot.
This information can be augmented by dynamic pedobarography which is in use in

a number of gait laboratories.
Longitudinal Assessments with Radiology: Hip Surveillance

Centers in Europe and Australia have developed formal “hip surveillance programs”
(Dobson et al. 2002; Hagglund et al. 2005; Kentish et al. 2011). Children with a
confirmed diagnosis of CP are offered regular clinical and radiographic examination
of their hips and access to both preventive and reconstructive surgery. In both
Victoria, Australia, and Southern Sweden, the prevalence of late dislocation has
decreased, and the need for salvage surgery has been reduced (Dobson et al. 2002;
Hagglund et al. 2005).
The most useful radiographic index for measuring hip displacement in children
with CP is the migration percentage of Reimers 1980. This measures the percent-
age of the femoral head that lies outside the acetabulum. Migration percentage can
be reliably measured from anterior-posterior hip radiographs, taken in supine with
good positioning and a standardized technique (Parrott et al. 2002). It is the key
index for making decisions about surgical management and to monitor hip
displacement both before and after operative intervention. The incidence of hip
displacement (migration percentage >30%) is 35% of children with CP in
population-based studies and is directly related to the child’s GMFCS level
(Soo et al. 2006; Hagglund et al. 2007; Connelly et al. 2009). Early hip displace-
ment is silent, and formal screening by radiographs of the hips is advised. The
frequency of such radiographs should be directly related to the risk of hip
displacement, which is in turn related to the child’s GMFCS level. Evidence-
based recommendations on hip surveillance in children with CP have been
published (Wynter et al. 2011).
Activity
Functional Mobility Scale
The FMS is a six level ordinal scale that rates the mobility of children with CP over
three distances according to their need for assistive devices (Graham et al. 2004).
The three distances of 5, 50, and 500 Meters represent mobility in the home, school,
and wider community settings, respectively. The scale is clinician administered
through parent or child report and should reflect performance rather than capability,
i.e., what the child actually does do rather than what they can do. For each of the
three distances, a rating of 1–6 is assigned. The FMS was designed as an outcome
measure and is sensitive to change (Thomason et al. 2011). Following optimum
biomechanical realignment and correction of spastic contractures, children can often
progress to lesser levels of support than used preoperatively (Thomason et al. 2011).
These important changes can be monitored and reported using the FMS (Graham
et al. 2004; Harvey et al. 2007).
Functional Assessment Questionnaire
The Functional Assessment Questionnaire (FAQ) is a 10-level, parent-reported

walking scale, which describes a range of walking abilities across the entire spec-
trum of CP, from nonambulatory to independent ambulation at a high level
(Novacheck et al. 2000). In addition to the 10-level walking scale, there is an
additional list of 22 items describing a variety of higher-level functional activities
requiring varying degrees of walking ability, balance, strength, and coordination.
The FAQ is a valid and reliable scale and has been shown to be sensitive to change
(Gorton et al. 2011). It is a simple scale, which can be quickly completed by parents
or caregivers and provides an excellent longitudinal view of the child’s gross motor
and walking abilities. It is a good measure of parental perspective however and
covers a wide variety of activities of daily living.
The FMS and FAQ are complementary scales and are used to assess outcomes in
children with CP after intervention.
The Gross Motor Function Measure

The Gross Motor Function Measure (GMFM) (Russell and Rosenbaum 1989) is the
gold standard for the measurement of gross motor function in children with CP and
has been shown to be valid, reliable, and responsive to change (Russell and
Rosenbaum 1989, Nordmark et al. 1997, 2000; Bjornson et al. 1998a, 1998b,
Russell et al. 2000, Russell and Leung 2003, Russell and Gorter 2005).
There are two versions of the GMFM. The original version consisted of 88 items,
which were grouped into five dimensions of gross motor function: lying and rolling,
sitting, crawling and kneeling, standing and walking, and running and jumping
(Russell and Rosenbaum 1989). Following Rasch analysis, a revised version,
which consists of 66 items, was developed, is interval scaled, and features item
maps. It is quicker to administer. However, it is limited for children who are very
young or severely involved (Russell and Leung 2003). The use of GMFM requires
an experienced physiotherapist and will take between 45 and 60 min to administer. It
is an essential tool in clinical outcome studies to assess change in gross motor
function. It is also a useful clinical measure to assess function in CP and guide the
management and treatment planning.
The Pediatric Evaluation of Disability Inventory (PEDI) (Haley 1997) may be
used to assess motor function. The PEDI may be a more appropriate tool for use with
children in GMFCS levels IV and V.
A detailed examination of the child’s level of activity and participation, using
measures such as the Canadian Occupational Performance Measure, Children’s
Assessment of Participation and Enjoyment, or the Activity Scale for Kids (Law
et al. 1990; Young et al. 2000; King et al. 2004), may be useful.
Self-Reported Questionnaires and Health-Related Quality of Life

In recent years, the assessment of quality of life has become a major goal in health
management including children with CP. A number of generic and specific
instruments have been developed which address aspects of health, functioning, and
quality of life. The Child Health Questionnaire (CHQ) (Waters et al. 2000) is a
widely used tool and is not disease specific. It has the advantage that the scores of
children with CP can be compared to children with typical health or with other
disease conditions. The Pediatric Orthopaedic Data Collection Instrument (PODCI)
(Daltroy et al. 1998) has a more musculoskeletal focus and contains several domains
directly relevant to children with CP and gait problems. Although some information
exists on the use of both of these instruments in children with CP, the responsiveness
to change and the value of using these as outcome instruments are not yet fully
established. It is also the case that neither can be considered to be a true quality of life
measure.
Other questionnaires include the PedsQL and the CPQoL-Child. The CPQoL-
Child is a specific quality of life measure developed for children with CP; however,
its responsiveness to physical interventions such as gait improvement surgery is not
yet known (Waters et al. 2006).
In order to judge the effectiveness of any intervention in children with CP, it is
important to understand the priorities and expectations of the child and parent
(Oeffinger et al. 2009; Novak et al. 2012). The Gait Outcomes Assessment List
(GOAL) is a new outcome measure to evaluate gait priorities and functional mobility
for ambulant children with CP (Narayanan et al. 2015). There are two versions of the
questionnaire, parent and child. There are 48 items in both versions grouped into
seven domains. Recent research has establishes the validity of the GOAL in mea-
suring the gross motor and gait function of ambulant children with CP. Evidence was
found for the discriminative validity of the GOAL, and correlations were demon-
strated with standard measures of gross motor function and gait. The GOAL pro-
vides meaningful information about a child’s function across multiple dimensions,
accounts for the environmental and personal factors that may contribute to function,
and measures the priorities and expectations of children and their parents. The
GOAL will allow clinicians to better understand the motor abilities, priorities, and
expectations of ambulant children with CP and enable better decision making about
appropriate interventions. The GOAL will be an invaluable addition to the measure-
ment tools available for gross motor function in CP.
Summary
In conclusion, natural history of gait in children with CP is one of deterioration.

There is level II evidence that single-event multilevel surgery (SEMLS)
improves the gait of children with spastic diplegic CP 12 months after surgery
(Thomason et al. 2011). SEMLS (Rutz et al. 2013) results in clinically and
statistically significant improvements in gait and function, in children with bilat-
eral spastic CP, which were maintained at 5 years after surgery (Rutz et al. 2013;
Thomason et al. 2013), and GMFCS stability was confirmed in the majority of
children (Rutz et al. 2012).
Cross-References

▶ Assessing Clubfoot and Cerebral Palsy by Pedobarography
▶ Gait Scores: Interpretations and Limitations
▶ Muscle Structure in Cerebral Palsy
▶ Surface Electromyography to Study Muscle Coordination
▶ Torsional Deformities on Cerebral Palsy
▶ Trunk and Spine Models for Instrumented Gait Analysis
References
Baker R (2013) Measuring walking: a handbook of gait analysis. Mac Keith Press, London
Baker R, McGinley JL, Schwartz MH et al (2009) The gait profile score and movement analysis
profile. Gait Posture 30:265–269
Beckung E, Carlsson G, Carlsdotter S, Uvebrant P (2007) The natural history of gross motor
development in children with cerebral palsy aged 1 to 15 years. Dev Med Child Neurol 49
(10):751–756
Bell KJ, Ounpuu S, DeLuca PA, Romness MJ (2002) Natural progression of gait in children with
cerebral palsy. J Pediatr Orthop 22(5):677–682
Bjornson KF, Graubert CS, Buford VL, McLaughlin J (1998a) Validity of the gross motor function
measure. Pediatr Phys Ther 10:43–47
Bjornson KF, Graubert CS, McLaughlin J, Kerfeld CI, Clark EM (1998b) Test retest reliability of
the gross motor function measure in children with cerebral palsy. Phys Occup Ther Pediatr
18:51–61
Connelly A, Flett P, Graham HK, Oates J (2009) Hip surveillance in Tasmanian children with
cerebral palsy. J Paediatr Child Health 45(7–8):437–443
Daltroy LH, Liang MH, Fossel AH, Goldberg M (1998) The POSNA pediatric musculoskeletal
functional health questionnaire: report on reliability, validity and sensitivity to change. The
pediatric outcomes instrument development group. J Pediatr Orthop 18:561–571
Davids JR, Ounpuu S, DeLuca PA, Davis RB (2003) Optimization of walking ability of children
with cerebral palsy. J Bone Joint Surg Am 85:2224–2234
Davids JR, Gilson T, Pugh LI (2005) Quantitative segmental analysis of weight-bearing radiographs
of the foot an ankle for children. J Pediatr Orthop 25(6):769–776
Davids JR, Oeffinger DJ, Bagley AM, Sison-Williamson M, Gorton G (2015) Relationship of
strength, weight, age, and function in ambulatory children with cerebral palsy. J Pediatr Orthop
35(5):523–529
Dobson F, Boyd RN, Parrott J, Nattrass GR, Graham HK (2002) Hip surveillance in children with
cerebral palsy: impact on the surgical management of spastic hip disease. J Bone Joint Surg Br
84:720–726
Eliasson A-C, Krumlinde-Sundholm L, Rosblad B et al (2006) The manual ability classification
system (MACS) for children with cerebral palsy: scale development and evidence of validity
and reliability. Dev Med Child Neurol 48:549–554
Escott BG, Ravi B, Weathermon AC, Acharva J, Gordon CL, Babyn PS, Kelley SP, Narayanan UG
(2013) EOS low-dose radiography: a reliable and accurate upright assessment of lower-limb
lengths. J Bone Joint Surg Am 95(23):e1831–e1837
Firth GB, Passmore E, Sangeux M, Thomason P, Rodda J, Donath S, Selber P, Graham HK (2013)
Multilevel surgery for equinus gait in children with spastic diplegic cerebral palsy: medium-term
follow-up with gait analysis. J Bone Joint Surg Am 95(10):931–938
Gorton GE 3rd, Stout JL, Bagley AM, Bevans K, Novacheck TF, Tucker CA (2011) Gillette
functional assessment questionnaire 22.-item skill set: factor and rasch analyses. Dev Med
Graham HK (2004) Mechanisms of deformity. In: Scrutton D, Damiano D, Mayston M (eds)
Management of the motor disorders of children with cerebral palsy, Clinics in developmental
medicine, vol 161, 2nd edn. Mac Keith Press, London, pp 105–129
Graham HK, Harvey A, Rodda J, Nattrass GR, Pirpiris M (2004) The functional mobility scale
(FMS). J Pediatr Orthop 24(5):514–520
Graham HK, Rosenbaum P, Paneth N, Dan B, Lin JP, Damiano DL, Becher JG, Gaebler-Spira D,
Colver A, Reddihough DS, Crompton KE, Lieber RL (2016) Cerebral palsy. Nat Rev Dis
Primers 2:15082
Gu Y, Shelton JE, Ketchum JM, Cifu DX, Palmer D, Sparkman A, Jermer-Gu MK, Mendigorin M
(2011) Natural history of scoliosis in nonambulatory spastic tetraplegic cerebral palsy. PM R 3
(1):27–32
Hagglund G, Wagner P (2008) Development of spasticity with age in a total population of children
with cerebral palsy. BMC Musculoskelet Disord 9:150
Hagglund G, Andersson S, Duppe H, Lauge-Pedersen H, Nordmark E, Westbom L (2005)
Prevention of dislocation of the hip in children with cerebral palsy. The first ten years of a
population-based prevention programme. J Bone Joint Surg Br 87:95–101
Hagglund G, Lauge-Pedersen H, Wagner P (2007) Characteristics of children with hip displacement
in cerebral palsy. BMC Musculoskelet Disord 8:101–107
Hagglund G, Alriksson-Schmidt A, Lauge-Pedersen H, Rodby-Bousquet E, Wagner P, Westbom L
(2014) Prevention of dislocation of the hip in children with cerebral palsy: 20-year results of a
population-based prevention programme. Bone Joint J 96-B(11):1546–1552
Haley SM (1997) The pediatric evaluation of disability inventory. J Rehab Outcome Meas 1:61–69
Hanna SE, Bartlett DJ, Rivard LM, Russell DJ (2008) Reference curves for the gross motor function
measure: percentiles for clinical description and tracking over time among children with cerebral
palsy. Phys Ther 88:596–607
Harvey A, Graham HK, Morris ME, Baker RJ, Wolfe R (2007) The functional mobility scale:
ability to detect changed following single event multilevel surgery. Dev Med Child Neurol 49
(8):603–607
Heidt C, Hollander K, Wawrzuta J, Molesworth C, Willoughby K, Thomason P, Khot A, Graham
HK (2015) The radiological assessment of pelvic obliquity in cerebral palsy and the impact on
hip development. Bone Joint J 97-B(10):1435–1440
Hidecker MJ, Paneth N, Rosenbaum PL, Kent RD, Little J, Eulenberg JB, Chester K Jr, Johnson B,
Michalsen L, Evatt M, Taylor K (2011) Developing and validating the communication function
classification system (CFCS) for individuals with cerebral palsy. Dev Med Child Neurol
53:704–710
Johnson DC, Damiano DL, Abel MF (1997) The evolution of gait in childhood and adolescent
Kentish M, Wynter M, Snape N, Boyd R (2011) Five-year outcome of state-wide hip surveillance of
children and adolescents with cerebral palsy. J Pediatr Rehabil Med 4:205–217
King G, Law M, King S et al (2004) Children’s assessment of participation and enjoyment (CAPE)
and preferences for activities of children (PAC). Harcourt Assessment, San Antonio
Knapp DR Jr, Cortes H (2002) Untreated hip dislocation in cerebral palsy. J Pediatr Orthop 22
(5):668–671
Koman LA, Mooney JF, Smith B (1993) Management of cerebral palsy with botulinum-A toxin:
preliminary investigation. J Pediatr Orthop 4:489–495
Law M, Baptiste S, McColl M, Opzoomer A, Polatajko H, Pollock N (1990) The Canadian
occupational performance measure: an outcome measure for occupational therapy. Can J
Occup Ther 57:82–87
Mackey AH, Lobb GL, Walt SE, Stott NS (2003) Reliability and validity of the observational gait
scale in children with spastic diplegia. Dev Med Child Neurol 45(1):4–11
McCormick A, Brien M, Plourde J, Wood E, Rosenbaum P, McLean J (2007) Stability of the gross
motor function classification system in adults with cerebral palsy. Dev Med Child Neurol
49:265–269
McGinley J, Dobson F, Ganeshalingam R, Shore BJ, Rutz E, Graham HK (2012) Single-event
multilevel surgery for children with cerebral palsy: a systematic review. Dev Med Child Neurol
54:117–128
McMulkin ML, MacWilliams BA (2015) Application of the Gillette gait index, gait deviation index
and gait profile score to multiple clinical pediatric populations. Gait Posture 41:608–612
Moreau M, Drummond DS, Rogala E, Ashworth A, Porter T (1979) Natural history of the
dislocated hip in spastic cerebral palsy. Dev Med Child Neurol 21(6):749–753
Morgan P, McGinley J (2014) Gait function and decline in adults with cerebral palsy: a systematic
review. Disabil Rehabil 36(1):1–9
Murnaghan ML, Simpson P, Robin JG, Shore BJ, Selber P, Graham HK (2010) The cerebral palsy
hip classification is reliable: an inter- and intra-observer reliability study. J Bone Joint Surg Am
92:436–441
Narayanan U, Moline R, Encisa C, Yeung C, Weir S (2015) Validation of the GOAL questionnaire:
an outcome measure for ambulatory children with cerebral palsy. Dev Med Child Neurol. 57(29)
doi:1111/dmcn.45_12887
Nordmark E, Hagglund G, Jarnlo GB (1997) Reliability of the gross motor function measure in
cerebral palsy. Scand J Rehabil Med 29:25–28
Nordmark E, Jarnlo GB, Hagglund G (2000) Comparison for the gross motor function measure and
paediatric evaluation of disability inventory in assessing motor function in children undergoing
selective dorsal rhizotomy. Dev Med Child Neurol 42:245–252
Norlin R, Odenrick P (1986) Development of gait in spastic children with cerebral palsy. J Pediatr
Orthop 6(6):674–680
Novacheck TF, Stout JL, Tervo R (2000) Reliability and validity of the Gillette functional
assessment questionnaire as an outcome measure in children with walking disabilities. J Pediatr
Orthop 20:75–81
Novak I, Hines M, Goldsmith S, Barclay R (2012) Clinical prognostic messages from a systematic
review on cerebral palsy. Pediatrics 130(5):e1285–312.
Oeffinger DJ, Rogers SP, Bagley A, Gorton G, Tylkowski CM (2009) Clinical applications of
outcome tools in ambulatory children with cerebral palsy. Phys Med Rehabil Clin N Am
20:549–565
Palisano R, Rosenbaum P, Walter S, Russell D, Wood E, Galuppi B (1997) Development and
reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med
Child Neurol 39(4):214–223
Palisano RJ, Rosenbaum P, Bartlett D, Livingston MH (2008) Content validity of the expanded and
revised gross motor function classification system. Dev Med Child Neurol 50:744–750
Parrott J, Boyd RN, Dobson F et al (2002) Hip displacement in spastic cerebral palsy: repeatability
of radiologic measurement. J Pediatr Orthop 22:660–667
Pritchett JW (1983) The untreated unstable hip in severe cerebral palsy. Clin Orthop Relat Res
173:169–172
Pruszczynski B, Sees J, Miller F (2015) Risk factors for hip displacement in children with cerebral
palsy: systematic review. J Pediatr Orthop 36(8):829–833
Read HS, Hazlewood ME, Hillman SJ, Prescott RJ, Robb JE (2003) Edinburgh visual gait score for
use in cerebral palsy. J Pediatr Orthop 23:296–301
Reimers J (1980) The stability of the hip in children: a radiological study of the results of muscle
surgery in cerebral palsy. Acta Orthop Scand Suppl 184:1–100
Robin J, Graham HK, Selber P, Dobson F, Smith K, Baker R (2008) Proximal femoral geometry in
cerebral palsy. A population-based cross-sectional study. J Bone Joint Surg Br 90:1372–1379
Rodda JM, Graham HK et al (2004) Sagittal gait patterns in spastic diplegia. J Bone Joint Surg Br
86(2):251–258
Rodda JM, Graham HK, Nattrass GR, Galea MP, Baker R, Wolfe R (2006) Correction of severe
crouch gait in patients with spastic diplegia with use of multilevel orthopaedic surgery. J Bone
Joint Surg Am 88(12):2653–2664
Rose GE, Lightbody KA, Ferguson RG, Walsh JC, Robb JE (2010) Natural history of flexed knee
gait in diplegic cerebral palsy evaluated by gait analysis in children who have not had surgery.
Rosenbaum PL, Walter SD, Hanna SE, Palisano RJ, Russell DJ, Raina P, Wood E, Bartlett DJ,
Galuppi BE (2002) Prognosis for gross motor function in cerebral palsy: creation of motor
development curves. JAMA 288:1357–1363
Rosenbaum P, Paneth N, Leviton A et al (2007a) A report: the definition and classification of
cerebral palsy April 2006. Dev Med Child Neurol Suppl 109:8–14
Rosenbaum PL, Livingston MH, Palisano RJ, Galuppi BE, Russell DJ (2007b) Quality of life and
health-related quality of life of adolescents with cerebral palsy. Dev Med Child Neurol
49:516–521
Russell DJ, Gorter JW (2005) Assessing functional differences in gross motor skills in children with
cerebral palsy who use an ambulatory aid or orthoses: can the GMFM-88 help? Dev Med Child
Neurol 47:462–467
Russell DJ, Leung KM (2003) Accessibility and perceived clinical utility of the GMFM-66:
evaluating therapists’ judgements of a computer-based scoring program. Phys Occup Ther
Pediatr 23(2):45–58
Russell DJ, Rosenbaum PL (1989) The gross motor function measure: a means to evaluate the
effects of physical therapy. Dev Med Child Neurol 31(3):341–352
Russell DJ, Avery LM, Rosenbaum PL, Raina PS, Walter SD, Palisano RJ (2000) Improved scaling
of the gross motor function measure for children with cerebral palsy: evidence of reliability and
validity. Phys Ther 80:873–885
Rutz E (2012) Are hips stable in children with cerebral palsy? Dev Med Child Neurol 54(10):878
Rutz E, Tirosh O, Thomason P, Barg A, Graham HK (2012) Stability of the gross motor function
classification system after single-event multilevel surgery in children with cerebral palsy. Dev
Med Child Neurol 54(12):1109–1113
Rutz E, Donath S, Tirosh O, Graham HK, Baker R (2013) Explaining the variability improvements
in gait quality as a result of single event multi-level surgery in cerebral palsy. Gait Posture 38
(3):455–460
Rutz E, Vavken P, Camathias C, Haase C, Juenemann S, Brunner R (2015) Long-term results and
outcome predictors in one-stage hip reconstruction in children with cerebral palsy. J Bone Joint
Surg Am 97(6):500–506
Saito N, Ebara S, Ohotsuka K, Kumeta H, Takaoka K (1998) Natural history of scoliosis in spastic
cerebral palsy. Lancet 351(9117):1687–1692
Samilson RL, Tsou P, Aamoth G, Green WM (1972) Dislocation and subluxation of the hip in
cerebral palsy. Pathogenesis, natural history and management. J Bone Joint Surg Am 54
(4):863–873
Schutte LM, Narayanan U, Stout JL, Selber P, Gage JR, Schwartz MH (2000) An index for
quantifying deviations from normal gait. Gait Posture 11:25–31
Soo B, Howard JJ, Boyd RN, Reid SM, Lanigan A, Wolfe R, Reddihough D, Graham HK (2006)
Hip displacement in cerebral palsy. J Bone Joint Surg Am 88(1):121–129
Terjesen T (2012) The natural history of hip development in cerebral palsy. Dev Med Child Neurol
54(10):951–957
Thomason P, Graham HK (2013) Rehabilitation of cerebral palsy. In: Iansek R, Morris M (eds)
Rehabilitation in movement disorders. Cambridge University Press, Cambridge, UK
Thomason P, Baker R, Dodd K, Taylor N, Selber P, Wolfe R, Graham HK (2011) Single-event
multilevel surgery in children with spastic diplegia: a pilot randomized controlled trial. J Bone
Joint Surg Am 93(5):451–460
Thomason P, Selber P, Graham HK (2013) Single event multilevel surgery in children with bilateral
spastic cerebral palsy: a 5 year prospective cohort study. Gait Posture 37(1):23–28
Waters E, Salmon L, Wake M, Hesketh K, Wright M (2000) The child health questionnaire in
Australia: reliability, validity and population means. Aust N Z J Public Health 24:207–210
Waters E, Davis E, Boyd R, Reddihough D, Mackinnon A, Graham HK, Lo SK, Wolfe R,
Stevenson R, Bjornson K, Blair E, Ravens-Sieberer U (2006) Cerebral palsy quality of life
questionnaire for children (CP QOL-child) manual. Deakin University, Melbourne
Wawrzuta J, Willoughby KL, Molesworth C, Ang SG, Shore BJ, Thomason P, Graham HK (2016)
Hip health at skeletal maturity: a population-based study of young adults with cerebral palsy.
Dev Med Child Neurol 58(12):1273–1280
Winters TF, Gage JR, Hicks R (1987) Gait patterns in spastic hemiplegia in children and young
adults. J Bone Joint Surg Am 69:437–441
World Health Organization (2001) International classification of functioning, disability and health.
World Health Organization, Geneva, pp 121–160; cited 2016. http://www.who.int/classifica
tions/icf/en/
Wren TAL, Rethlefsen SA, Healy BS, Do KP, Dennis SW, Kay RM (2005) Reliability and validity
of visual assessments of gait using a modified physician rating scale for crouch and foot contact.
J Pediatr Orthop 25:646–650
Wynter M, Gibson N, Kentish M, Love S, Thomason P, Graham HK (2011) The consensus
statement on hip surveillance for children with cerebral palsy; Australian standards of care.
J Pediatr Rehab Med 4:183–195
Young N, Williams JI, Yoshida KK, Wright JG (2000) Measurement properties of the activities
scale for kids. J Clin Epidemiol 53:125–137
Swing Phase Problems in Cerebral Palsy
Ana Presedo
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Swing Phase Abnormal Kinematic Patterns and Causes of Pathological Function . . . . . . . . . . . . . 4
Hip and Pelvis Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Knee Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Ankle and Foot Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Management of Swing Phase Problems in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Excessive Hip Adduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Stiff-Knee Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Insufficient Ankle Dorsiflexion and Foot Varus Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Abstract
The normal human gait cycle is commonly divided in two phases: stance and
swing. The objective of stance phase is to provide support, stability, and propul-
sion and contribute to the advancement of the limb in swing. Ground clearance
and appropriate pre-positioning of the foot are prerequisites of normal gait during
swing. Normal characteristics of stance and swing phases ensure adequate step
length and energy conservation during walk. Swing phase problems in cerebral
palsy tend to be related to spasticity, abnormal muscular control, poor opposite
limb balance, and lack of power. From a clinical point of view, knee stiffness and
ankle insufficient dorsiflexion constitute the two major abnormalities in the
sagittal plane, whereas in the frontal and transverse planes, excessive hip adduc-
tion and foot deviations can cause problems with limb advancement. Rectus
A. Presedo (*)
Pediatric Orthopaedics Department, Robert Debré University Hospital, Paris, France
e-mail: ana.presedo@aphp.fr; a.presedo@gmail.com

DOI 10.1007/978-3-319-30808-1_53-1
2 A. Presedo
femoris (RF) abnormal activity has been proposed as the main cause of knee
stiffness in patients with cerebral palsy. Spasticity and/or contractures of the hip
adductors and tibialis posterior muscles can cause abnormal hip and ankle
kinematics and can lead to foot deformities if left untreated. Indications for
treatment should be based on patient’s age and functional status. Different
therapeutic approaches are described and commented in this chapter. A review
of the literature based on the major articles in this field is provided.
Keywords
Swing phase • Stiff-knee gait • Rectus femoris release • Split anterior tibial tendon
transfer
Introduction
Swing phase represents about 40 % of the gait cycle. The purposes of swing are to
advance the limb, to provide foot clearance, to allow variation in cadence, and to
participate to the energy conservation mechanisms. The swing phase is commonly
subdivided in three periods: initial swing, mid-swing, and terminal swing. The
objective of the first period is to accelerate the limb, and this function is dependent
on muscular work occurring during terminal stance and pre-swing, the second period
represents a transition period, and the third period’s functional goal is to decelerate
the limb to prepare it for initial contact. At normal self-selected walking speed, the
limb swings with little muscle intervention; only the tibialis anterior is active
throughout swing (Piazza and Delp 1996). In terminal swing, hamstring contraction
generates simultaneous hip extension and knee flexion. The action of hip extension
contributes to level the pelvis and to maintain trunk alignment, whereas the knee
flexion moment decelerates the shank and controls knee extension at initial contact.
The main role of the rectus femoris (RF) muscle in normal gait is to act as a hip flexor
to propel the limb forward into swing (Perry 1992). Using fire-wire electrodes, two
bursts of activity have been described (Annaswamy et al. 1999). The first burst
occurs during the loading response phase of gait where the RF acts along with the
vastii, and the second burst occurs during pre- and initial swing. During fast walking,
the RF contraction helps to accelerate the shank in initial swing, and the hamstrings
act to decelerate it in terminal swing (Nene et al. 1999).
Swing phase important kinematic events occur mainly in the sagittal plane.
During initial swing, the peak of knee flexion (PKF), which accounts for the
maximum degree of knee flexion during the gait cycle, compensates for ankle plantar
flexion and helps with foot clearance (Anderson et al. 2004). Since the lower limb
acts as a compound pendulum, hip and knee flexion occur with very little muscular
intervention during normal swing phase (Piazza and Delp 1996). In mid-swing, this
mechanism generates knee extension, whereas the hip keeps going into flexion.
During terminal swing, hip flexion reaches its maximum and the knee is almost
extended before initial contact. The only active muscle throughout swing phase is
Swing Phase Problems in Cerebral Palsy 3
the tibialis anterior that brings the ankle into neutral position acting against gravity.
The actions of hamstrings and tibialis anterior stabilize the knee and ankle positions
for landing. In the transverse and coronal planes, pelvic internal rotation and hip
adduction in terminal swing provide adequate step length and proper foot orientation
at heel contact. Hip abduction is observed during initial swing and helps with limb
advancement. Abnormal kinematics during swing are mainly represented by insuf-
ficient or delayed peak of knee flexion (stiff-knee gait) and excessive ankle plantar
flexion. Transverse plane abnormal kinematics are frequently seen in patients with
hemiplegia and tend to be related to excessive pelvic rotation and foot deviations.
State of the Art
Swing phase problems in cerebral palsy (CP) are mainly represented by limited knee
flexion (stiff knee) and insufficient ankle dorsiflexion. An insufficient or delayed
PKF and similarly excessive ankle plantar flexion during swing compromise foot
clearance and adequate foot position at initial contact. Although terminal stance
kinematics and kinetics influence swing phase events, spasticity of the RF and poor
distal motor control constitute the main causes of these gait deviations.
Stiff-knee management is addressed to diminish RF spasticity or to treat RF
contractures. In children with CP abnormal RF activity is almost always present.
Rectus femoris distal tendon transfer to knee flexor muscles has been advocated to
avoid the knee extensor moment generated by RF abnormal activity during swing
and to create a knee flexion moment. Studies based on cine phase-contrast magnetic
resonance imaging (MRI) showed that the RF is not converted to a knee flexor after
its distal tendon is transferred to the posterior side of the knee, but its capacity for
knee extension is diminished by the surgery. Also, the three-dimensional reconstruc-
tion (MRI) of RF muscle geometry after the transfer demonstrated that the muscle
didn’t follow a straight course from its origin to its new insertion, but a sharp
deviation was frequently seen (Asakawa et al. 2004). In addition, scar tissue between
RF and the underlying muscles was visible and could be the cause of relapses after
RF transfer. Long-term follow-up studies showed variable results regarding
improvement of the magnitude of PKF, but improvements in timing of knee flexion
and overall knee range of motion remained stable over time (Dreher et al. 2012;
Thawrani et al. 2012). Distal RF tendon resection has been proposed as an alterna-
tive to the transfer. A short-term follow-up study showed that RF distal tendon
resection was associated with improved knee range of motion and timing of PKF in
swing, and the absolute values of PKF became normal for those patients who
showed abnormal preoperative values (Presedo et al. 2012).
Ankle excessive plantar flexion and dynamic varus deviations are commonly
observed in patients with spastic hemiplegia (Gage 1991). Botulinum toxin and
orthoses are currently used to diminish spastic plantar flexion and varus deviations
and to improve foot clearance in patients with ankle dorsiflexor weakness. Surgical
muscular-tendon lengthenings may be indicated to treat fixed contractures. Davids
4 A. Presedo
et al. (2011) demonstrated an improvement in ankle dorsiflexion following ankle

plantar flexor surgery in selected children with CP. They explained this by the fact
that correction of a fixed constraining equinus contracture of the ankle plantar flexors
may unmask preexisting ankle dorsiflexion function. Split anterior tibial tendon
transfer, combined with gastrocnemius release and/or tibialis posterior lengthening,
has been advocated to treat spastic equinovarus deformity. Hoffer et al. (1985)
demonstrated that this procedure was reliable and effective to correct dynamic
varus deviations in patients with CP, after a follow-up of 10 years. Other soft-
tissue procedures like tibialis posterior transfer to the forefoot tend to be abandoned
in patient with spasticity because of danger of overcorrection and resultant
calcaneus-valgus deformity. Bony procedures (calcaneal osteotomies and triple
arthrodesis) may be indicated to correct fixed deformities.
Swing Phase Abnormal Kinematic Patterns and Causes

of Pathological Function
Computerized gait analysis has become the gold standard for the evaluation of
complex gait problems in patients with cerebral palsy (Narayanan 2007; Gough
and Shortland 2008). Based on the kinematic study of lower limb motion, common
abnormal patters for the individual joints have been described. The recognition of
these patterns has conducted to the elaboration of gait classifications (Rodda and
Graham 2001). Without any intervention, these patterns of gait are extremely
predictable over time. Although therapy decisions are particular to each patient,
the knowledge of these abnormal patterns should allow for a better comprehension
of pathology and conduct to a more appropriate treatment.
Hip and Pelvis Pathology
The hip and pelvis play a major role during gait, ensuring power generation and
balance of the upper body segment. Hip extensor muscles provide 50 % of power
during normal walking (the other 50 % coming from ankle plantar flexors), and in
patients with CP, since motor control is somehow preserved around the hip, com-
pensatory mechanisms occur mainly at this level (Winter 1991; Õunpuu 1994).
During swing phase, pelvic internal rotation and hip adduction in terminal swing
provide adequate step length and proper foot orientation at heel contact. Hip
abduction is observed during initial swing and helps with limb advancement.
Excessive hip flexion throughout the cycle is commonly seen in patients with
severe involvement (GMFCS III–IV). The typical kinematic pattern called “double-
bump pelvic pattern” (Fig. 1) is characterized by an increased anterior pelvic tilt in
stance phase and a second one in swing phase (Gage 1991). This abnormal pattern is
related to a poor dissociation between pelvic and hip motion. Anterior pelvic tilt
peak occurs always at single-limb stance under the influence of spastic hip flexors
and/or weak hip extensors. “Single-bump pelvic pattern” is seen in patients with
30.0 Pelvic Tilt
25.0
Ant
20.0
15.0
Angle (degrees)
10.0
5.0
0.0
Post
−5.0
0 10 20 30 40 50 60 70 80 90 100
Angle (degrees)
Fig. 1 The green line represents normal pelvis kinematics in the sagittal plane. The blue and red
lines represent abnormal “double-bump pattern” kinematics for the right and left lower limb,
respectively
hemiplegia (Fig. 2). Anterior pelvic tilt peak occurs in terminal stance, only on the
hemiplegic side, and pelvis returns to normal alignment in swing phase as the stretch
of the hip flexors ends when the hip moves into flexion. Anterior pelvic tilt
throughout the gait cycle is commonly seen in young children with spastic diplegia
who walk with increased cadence to compensate for the lack of stability related to
spasticity of ankle plantar flexors. In this case, the shank does not advance normally
over the foot, and so the trunk and pelvis bend forward in order to move the center of
mass along the line of progression.
Hip and pelvic coronal and transverse plan abnormal patterns present often
combined. In diplegic and hemiplegic patients with distal involvement, these pat-
terns may represent coping responses. In hemiplegic patients, abnormal rotation of
pelvis and hip are very common and may be related to asymmetric neurological
involvement or to torsional bony deviations. Internal hip rotation on the hemiplegic
side is compensated by external rotation of the ipsilateral hemipelvis (Fig. 3).
Increased pelvic rotation can be a compensation for reduced sagittal plane motion,
in order to increase step length. During swing, hip abduction or circumduction
represents a typical compensation for reduced sagittal plane motion. As a conse-
quence of increased pelvic rotation, coronal hip motion will also increase. Increased
6 A. Presedo
Pelvic Tilt
30.0
Ant
20.0
Angle (degrees)
10.0
0.0
Post
0 10 20 30 40 50 60 70 80 90 100
Normalised (percent)
Fig. 2 In a patient with right hemiplegia, a “single-bump” pattern of pelvis kinematics is typically
observed
coronal pelvic motion may be seen to compensate for a reduced hip abductor
moment in stance (hip abductor insufficiency). In this case, the patient will lean
the trunk laterally to shift the body center of mass and bring the GRF vector close to
the hip center of rotation. Kinematic curves will show ipsilateral pelvic depression in
stance and elevation in swing with decreased hip adduction in mid-stance and
increased hip adduction in terminal stance and initial swing. In quadriplegic and
diplegic patients with more severe involvement, hip dysfunction is commonly
related to excessive femoral anteversion and adductor muscle contractures.
Knee Pathology
The peak of knee flexion is generated by the action of ankle plantar flexors in
terminal stance and by the hip flexors in pre-swing and initial swing, so the lower
limb acts as a compound pendulum (Anderson et al. 2004). During initial swing,
additional knee flexion is performed by the short head of biceps femoris to facilitate
foot clearance (Perry 1992). The magnitude of the motion and the timing are critical
to achieve the necessary knee flexion to clear the foot. The isometric contractions of
RF and hamstrings during swing phase act to regulate knee flexion/extension in
order to adjust to changes in walking speed (Nene et al. 1999).
Excessive knee flexion pattern is characterized by greater than normal knee
flexion throughout the stance phase. Increased knee flexion is usually maintained
a 30 Ab/Adduction Hanche
Add 20
10
Angle (degrees)
–10
–20
Abd
–30
0 10 20 30 40 50 60 70 80 90 100
Normalized (percent)
b 20.0 Obliquité bassin
15.0
up
10.0
Angle (degrees)
5.0
0.0
–5.0
–10.0
Down
–15.0
–20.0
0 10 20 30 40 50 60 70 80 90 100
c 30 Rotation bassin
20
Int
10
Angle (degrees)
–10
–20
Ext
–30
0 10 20 30 40 50 60 70 80 90 100
Fig. 3 (a) Patient with right hemiplegia. Hip frontal plane kinematics show an asymmetric pattern,
with the right side being abducted throughout the cycle. This pattern of hip motion is related to
asymmetric pelvic frontal (b) and transverse (c) alignment
8 A. Presedo
Fig. 4 (a) This patient shows knee stiffness and difficulty to clear the foot during swing phase.
(b) Kinematics curve shows insufficient peak knee flexion during initial swing with decreased knee
range of motion. (c) There is a premature and prolonged activity of the rectus femoris muscle during
terminal stance and swing phases
during swing phase, and global knee range of motion tends to be very limited
(“crouch gait”). In cerebral palsy, this pattern is usually associated to hamstring
spasticity or contracture but can be also secondary to plantar flexor or hip extensor
weakness. Insufficient knee flexion in swing phase has been defined as diminished
and/or delayed peak of knee flexion and is referred to as “stiff-knee gait” (Sutherland
and Davids 1993). Stiff-knee gait was first described as one of the gait classification
patterns; however, this pattern of gait is not considered as a separate entity by Rodda
and Graham classification (Rodda and Graham 2001) since knee stiffness can be part
of different types of gait. Stiff-knee gait is frequent in patients with jump and crouch
gait patterns and also in hemiplegic patients with knee involvement. This kinematic
pattern has been initially described as a deficit of knee flexion during swing (Fig. 4).
However, since patients who walk with a permanent knee flexion throughout the
cycle may have normal amount and timing of PKF but very limited knee range of
motion that may be difficult to clear the foot, decreased knee range of motion is also
considered one of the criteria to define stiff-knee gait (Moreau and Tinsley 2005). In
children with CP, stiff-knee gait is almost always associated to abnormal activity of
the rectus femoris during swing.
Abnormal knee function is almost always present in children with spastic diple-
gia. In patients with spasticity, selective distal motor control is impaired, and so they
are stance stability and propulsion. Young patients with mild involvement can
compensate for this with an increase of cadence, so they preserve speed and swing
knee flexion. However, propulsion tends to deteriorate over time and so it does
swing knee function. Because of the lack of distal motor control, biarticular muscles
(rectus femoris and hamstrings) tend to favor proximal function and act as hip
flexors/extensors rather than control knee motion. It is common in diplegic and
hemiplegic patients to show prolonged activity of RF and hamstring co-contraction
during swing phase. Thus, both muscles are active through mid-swing when they
should be silent. The RF acts as a primary hip flexor and secondarily extends the
knee, reducing the peak knee flexion, and the hamstrings act primarily as hip
extensors and not being able to counterbalance the extension caused by the rectus
(Gage et al. 1987; Perry 1987; Sutherland et al. 1990). Given the pathology of these
muscles, the ability to allow variation of cadence can be largely compromised in
patients with CP. In addition to that, poor propulsion and deficit of passive hip
flexion that results are frequent in patients with slow walks and also contribute to
knee stiffness.
Ankle and Foot Pathology
In patients with CP, there is almost always a dominance of the triceps over the ankle
dorsiflexors (Gage 1991). During swing phase, the shape of the ankle kinematic
curve can be normal or it can lack dorsiflexion (Fig. 5a). In patients with spastic
diplegia, foot position relative to the shank is usually normal, and forefoot initial
contact is commonly related to abnormal knee flexion rather to ankle plantar flexion.
The gastrocnemius is usually spastic and can develop some degree of contraction,
but the soleus tends to maintain normal length, so fixed plantar flexion due to triceps
contracture is relatively uncommon in these children. During swing phase, although
the dorsiflexors may still be overwhelmed by the plantar flexors, some degree of
dorsiflexion is in general possible, and thus, ankle kinematics tends to occur within
normal limits. In children with hemiplegia, contracture is common in both the soleus
and gastrocnemius. As a result of the overactivity of the triceps and the dominance of
the tibialis anterior and posterior over the peroneals and toe extensors, the foot is
typically postured in equinovarus. This deformity tends to become rigid over time.
Thus, patients with hemiplegia may show a permanent plantar flexion kinematic
pattern (Fig. 5b). During swing phase, problems with foot clearance and inadequate
foot position in terminal swing can result from abnormal ankle kinematics.
10 A. Presedo
Fig. 5 (a) Ankle kinematics in a spastic diplegic patient is characterized by plantar flexion at initial
contact, followed by limited dorsiflexion in stance and insufficient dorsiflexion in swing phase.
(b) Ankle kinematics in a hemiplegic patient with gastrocsoleus contracture shows permanent
plantar flexion
Management of Swing Phase Problems in Cerebral Palsy
Treatment strategies should be based on the age of the patient and the severity of
neurologic involvement. Primary problems (spasticity) should receive specific treat-
ment with the aim of prevent secondary problems (muscular contractures and bone
deformities). If orthopedic surgery procedures are needed, these should be planned
within a multidisciplinary team, taking into consideration the whole child and not
just his motor-skeletal parts.
Excessive Hip Adduction
In diplegic and hemiplegic patients with distal involvement, abnormal hip and pelvis
kinematic patterns may represent coping responses. In hemiplegic patients, abnor-
mal rotation of pelvis and hip are very common and may be related to asymmetric
neurological involvement and/or to torsional bony deviations. Increased hip adduc-
tion in terminal stance and initial swing is commonly seen in diplegic patients with
more severe involvement and tends to be related to excessive femoral anteversion
and adductor muscle contractures. Excessive hip adduction in swing can compro-
mise foot clearance and cause tripping and falling. Therapeutic goals are oriented to
treat adductor muscle contractures and bony torsions.
Stiff-Knee Gait
Stiff-knee gait management in children with cerebral palsy will focus on treatment of
RF spasticity and/or contractures. Rectus femoris surgical procedures are often
performed as part of multilevel surgery; therefore, improvement on stance phase
parameters can largely contribute to a better knee flexion during swing. Adequate
knee flexion velocity at toe-off has been proved to be the most important kinematic
factor to achieve normal swing phase knee flexion (Goldberg et al. 2004). Hip
flexion moment, generated by the iliopsoas muscle, and ankle plantar flexion,
generated by the gastrocnemius, were identified as the parameters that contribute
most to increasing knee velocity during double support.
Since prolongation of phasic, late stance, RF activity into swing phase constitutes
an electromyographic (EMG) pattern commonly associated with stiff-knee gait, RF
proximal or distal release procedures have been proposed to improve knee flexion in
patients with abnormal RF activity in swing phase (Sutherland et al. 1990).
According to these authors, RF proximal release did not influence pelvic alignment
or hip range of motion. They concluded that RF release would reduce the extensor
properties of the muscle and thus facilitate passive knee flexion in swing. Since
patients with cerebral palsy tend to walk slower than normal, hip flexion moment in
pre-swing is often diminished. For this reason, Perry suggested to abandon proximal
RF release and transfer the RF posterior to the axis of knee to enhance active knee
flexion (Perry 1987). Indications for transfer included (a) diminished range of knee
12 A. Presedo
flexion during swing phase, (b) excessive RF EMG activity during swing, and (c) a
positive Duncan-Ely test. Õunpuu et al. (1993a) compared the outcomes of RF
release versus transfer. Although the differences were not significant in terms of
amount of variation in PKF and time to PKF in swing, this study showed that there
was a tendency toward more normal values after the transfer. This tendency could be
explained by the notion that the RF, transferred posterior to the axis of knee flexion,
would generate a knee flexion moment in swing. The choice of the transfer site did
not influence PKF, knee ROM, or transverse plan kinematics values (Õunpuu
et al. 1993b). Hemo et al. (2007) compared the outcomes of two different techniques:
RF distal release and RF transposition to the iliotibial band. They found similar
improvement in knee ROM, PKF, and time to PKF after 1-year follow-up. Factors
related to good results after RF transfer have been indicated: (a) preoperative RF
prolonged activity limited to swing phase (Miller et al. 1997), (b) preoperative
positive Duncan-Ely test (Kay et al. 2004), and (c) postoperative increase in knee
flexion velocity at toe-off (Goldberg et al. 2006).
More recent studies attempted to explain the action of the RF muscle following
distal tendon transfer. Riewald and Delp (1997) investigated whether the RF con-
verts to a knee flexor after being transferred to the semitendinosus muscle or to the
iliotibial band. Rectus femoris EMG activity showed that the muscle generated an
extensor moment in all of their subjects. Based on cine phase-contrast MRI,
Asakawa et al. (2002, 2004) examined RF motion in vivo and muscle geometry
after tendon transfer surgery. In the tendon transfer group, the RF moved in the
direction of the knee extensors, and fiber excursions were reduced compared to
vastus intermedius. These authors concluded that the RF was not converted to a knee
flexor after its distal tendon was transferred to the posterior side of the knee, but its
capacity for knee extension was diminished by the surgery. They suggested that scar
tissue could form after RF transfer, making the RF adhere to the underlying muscles.
They also examined three-dimensional MRI of patients who had a RF transfer and
observed abnormal, low-signal intensity images that could represent scar tissue
between the transferred muscle and the underlying vastii in each of the patients.
Tridimensional models showed that the transferred muscles followed angular, devi-
ated paths to their new insertions, suggesting that RF tendons were probably
constrained by adhesions to the underlying muscles. Following these results, we
believed there was insufficient evidence supporting transferring the RF, rather than
performing a distal release. We also thought a complete tendon resection would
reduce adherences and prevent relapses (Fig. 6). We therefore determined to assess
the outcome of children with spastic diplegia following RF distal tendon resection as
a part of multilevel surgery, in order to (a) evaluate the improvement of knee ROM
during gait, (b) measure changes in maximum knee flexion and time to PKF during
swing phase, and (c) compare our results with those reported in the literature after RF
distal release or transfer, since we are not aware of any previous reports on RF distal
tendon resection. Our results, after 2-year follow-up, showed that RF distal tendon
resection was associated with improved knee ROM and timing of peak knee flexion
in swing, and the absolute values of peak knee flexion became normal for those
patients who showed abnormal preoperative values (Presedo et al. 2012).
Fig. 6 Rectus femoris distal

tendon resection. The tendon
is transected at the muscular
junction and removed
completely in order to avoid
subsequent adherences to the
underlying muscles
Long-term follow-up studies showed variable results in terms of PKF magnitude

and timing but agreement on knee range of motion improvement that was maintained
at final follow-up. Moreau and Tinsley (2005) compared the outcomes of a group of
patients who had RF transfer as part of multilevel surgical procedures to another
group of patients who did not undergo any RF procedure. They found the RF transfer
helped to maintain knee ROM and PKF values over time, after a minimum follow-up
of 3 years. Dreher et al. (2012) found a significant improvement in timing of PKF,
knee range of motion, and knee flexion velocity after 9 years follow-up. Thawrani
et al. (2012) reported an improvement in PKF magnitude and timing after 7-year
follow-up.
Insufficient Ankle Dorsiflexion and Foot Varus Deviation
Sagittal and coronal plan deviations of the ankle and foot are common in hemiplegic
patients. These patients may show a permanent plantar flexion pattern, combined with
hind foot varus deviation. A gait classification based on kinematic abnormal patterns in
the sagittal plan is commonly used for hemiplegic patients (Winters et al. 1987).
Type 1 hemiplegia is characterized by a “drop foot” during swing phase. This is
due to a lack of control of ankle dorsiflexors. There is not plantar flexor contracture
although they tend to be spastic. Because of the absence of muscle contracture, ankle
dorsiflexion is normal in mid-stance, but the ankle is in plantar flexion at the initial
contact. The treatment of this type of pattern is a leaf spring or an articulated ankle
foot orthosis (AFO). Spasticity treatment can be indicated in order to improve ankle
dorsiflexion during swing in patients with relative good motor control and active
dorsiflexion in physical exam.
Type 2 hemiplegia is characterized by a permanent ankle plantar flexion, during
stance and swing phases. There is spasticity and/or contracture of plantar flexors.
The management of this type of gait is based on spasticity treatment by botulinum
14 A. Presedo
toxin injections and/or muscular lengthenings, if necessary. If there is a mild

contracture, spasticity management combined with walking casts can be very effec-
tive to improve ankle range of motion. We tend to use knee immobilizers during
nighttime in order to maintain the gastrocnemius in a stretched position. In case of
more severe contracture, gastrocnemius lengthening or tendo-Achilles lengthening
will be indicated. Most of the children will require the use of orthoses. Leaf spring
orthoses will be adequate in absence of knee recurvatum; however, if the knee is
fully extended or in recurvatum, articulated AFO with limitation of plantar flexion
may be more convenient.
Most of the diplegic and some hemiplegic patients with proximal involvement
(type 4 hemiplegia) show combined coronal and transverse plane problems. In the
coronal plane, excessive hip adduction secondary to adductor spasticity or contrac-
ture and hip subluxation represent the main issues. In the transverse plane, pelvic and
hip abnormal rotations, femoral anteversion, tibial torsion, and foot deviations are
commonly observed. Whereas management of sagittal plane problems is mainly
address to spasticity and muscle contracture, bony surgery is often required to treat
coronal and transverse plane abnormalities. Patients with type 4 hemiplegia present
proximal involvement and a pattern of sagittal kinematics that can be similar to that
seen on diplegic patients. As a result of the triceps overactivity and the dominance of
the tibialis anterior and posterior over the peroneals, the foot typically in equinovarus
tends to become rigid over time. During swing phase, due to spasticity or contracture
of tibialis posterior and the predominance of tibialis anterior and extensor hallucis
longus, the foot is deviated in varus and supination. In patients with severe spasticity
and/or contracture of hip flexors and adductors, hip subluxation is not uncommon.
Management of this type of hemiplegia may include spasticity treatment, but more
often, muscular lengthening is required. Split anterior tibial tendon transfer, com-
bined with gastrocsoleus and tibialis posterior lengthening, is recommended to
correct dynamic equinovarus foot deviation (Hoffer et al. 1985; Davids
et al. 2011). Bony surgery to address foot deformities and femoral torsion can be
also part of the surgical program. After surgery, the most common type of orthoses
used is a solid AFO that will help with knee extension and will also avoid ankle
plantar flexion during swing. With time, most of the patients tend to keep a leaf
spring AFO to help with clearance.
Summary
Swing phase abnormal kinematics can compromise foot clearance and adequate limb
pre-positioning for initial contact. In children with cerebral palsy, these problems are
related to spasticity, muscular weakness, and poor motor control. Stiff-knee gait
represents the most common swing phase kinematic problem in these patients.
Although knee flexion during swing is related to terminal stance and pre-swing
kinematics, rectus femoris prolonged activity during swing is almost always present,
and stiff-knee treatment is oriented to diminish the extensor effect of this muscle.
Long-term studies have demonstrated that after rectus femoris surgery (distal tendon
resection or transfer), swing phase knee kinematics and knee range of motion tend to
improve, and this improvement lasts over time. Difficulties with foot clearance
caused by ankle dorsiflexor weakness, tendo-Achilles contracture, or hindfoot
varus deviation are frequent in hemiplegic patients. Treatment focuses on spasticity
management, musculotendinous releases, tibialis anterior transfer, or bony proce-
dures to correct fixed deformities.
References
Anderson FC, Goldberg SR, Pandy MG, Delp SL (2004) Contributions of muscle forces and toe-off
kinematics to peak knee flexion during the swing phase of normal gait: an induced position
analysis. J Biomech 37:731–737
Annaswamy TM, Giddings CJ, Della Croce U, Kerrigan DC (1999) Rectus femoris: its role in
normal gait. Arch Phys Med Rehabil 80:930–934
Asakawa DS, Blemker SS, Gold GE, Delp SL (2002) In vivo motion of the rectus femoris muscle
after tendon transfer surgery. J Biomech 35:1029–1037
Asakawa DS, Blemker SS, Rab GT, Bagley A, Delp SL (2004) Three dimensional muscle-tendon
geometry after rectus femoris tendon transfer. J Bone Joint Surg 86:348–354
Davids JR, Rogozinski BM, Hardin JW, Davis RB (2011) Ankle dorsiflexion function after plantar
flexor surgery in children with cerebral palsy. J Bone Joint Surg Am 93:e 138, 1–7
Dreher T, Wolf SI, Maier M, Hagmann S, Vegvari D, Gantz S, Heitzmann D, Wenz W, Braatz F
(2012) Long-term results after distal rectus femoris transfer as a part of multilevel surgery for the
correction of stiff-knee gait in spastic diplegic cerebral palsy. J Bone Joint Surg Am 94:e
142, 1–10
Gage JR (1991) Gait analysis in cerebral palsy. Mac Keith Press, London, pp 61–95
Gage JR, Perry J, Hicks RR, Koop S, Werntz JR (1987) Rectus femoris transfer to improve knee
function of children with cerebral palsy. Dev Med Child Neurol 29:159–166
Goldberg SR, Anderson FC, Pandy MG, Delp SL (2004) Muscles that influence knee flexion
velocity in double support: implications for stiff-knee gait. J Biomech 37:1189–1196
Goldberg SR, Õunpuu S, Arnold AS, Gage JR, Delp SL (2006) Kinematic and kinetic factors that
correlate with improved knee flexion following treatment for stiff-knee gait. J Biomech
39:689–98.
Gough M, Shortland AP (2008) Can clinical gait analysis guide the management of ambulant
children with bilateral spastic cerebral palsy? J Pediatr Orthop 28:879–883
Hemo Y, Aiona MD, Pierce RA, Dorociack R, Sussman M (2007) Comparison of rectus femoris
transposition with traditional transfer for the treatment of stiff knee gait in patients with cerebral
palsy. J Child Orthop 1:37–41
Hoffer MM, Barakat G, Koffman M (1985) 10-year follow-up of split tibial tendon transfer in
cerebral palsied patients with spastic equinus deformity. J Pediatr Orthop 5:432–434
Kay RM, Rethlefsen SA, Kelly JP, Wren TAL (2004) Predictive value of the Duncan-Ely test in
distal rectus femoris transfer. J Pediatr Orthop 24:59–62
Miller F, Dias R, Lipton GE, Albarracin JP, Dabney KW, Castagno P (1997) The effect of rectus
EMG patterns on the outcome of rectus femoris transfers. J Pediatr Orthop 17(5):603–607
Moreau N, Tinsley S (2005) Progression of knee joint kinematics in children with cerebral palsy
with and without rectus femoris transfers: a long-term follow up. Gait Posture 22:132–137
Narayanan UG (2007) The role of gait analysis on the orthopaedic management of ambulatory
cerebral palsy. Curr Opin Pediatr 19:38–43
Nene A, Mayagoitia R, Veltink P (1999) Assessment of rectus femoris function during initial swing
phase. Gait Posture 9:1–9
Õunpuu S (1994) The biomechanics of walking and running. Clin Sports Med 13:843–863
16 A. Presedo
Õunpuu S, Muik E, Davis RB 3rd, Gage JR, DeLuca PA (1993a) Rectus femoris surgery in children
with cerebral palsy. Part I. The effect of rectus femoris transfer location on knee motion. J
Pediatr Orthop 13:325–330
Õunpuu S, Muik E, Davis RB 3rd, Gage JR, DeLuca PA (1993b) Rectus femoris surgery in children
with cerebral palsy. Part II. A comparison between the effect of transfer and release of the distal
rectus femoris on knee motion. J Pediatr Orthop 13:331–335
Perry J (1987) Distal rectus femoris transfer. Dev Med Child Neurol 29:53–58
Perry J (1992) Gait analysis: normal and pathological function. Slack, Thorofare, pp 1–19
Piazza SJ, Delp SL (1996) The influence of muscles on knee flexion during the swing phase of gait.
Presedo A, Megrot F, Ilharreborde B, Mazda K, Penneçot GF (2012) Rectus femoris distal tendon
resection improves knee motion in patients with spastic diplegia. Clin Orthop Relat Res 470
(5):1312–1319
Riewald SA, Delp SL (1997) The action of the rectus femoris muscle following distal tendon
transfer: does it generate knee flexion moment? Dev Med Child Neurol 39:99–105
Rodda J, Graham HK (2001) Classification of gait patterns in spastic hemiplegia and spastic
diplegia: a basis for a management algorithm. Eur J Neurol 8(Suppl 5):98–108
Sutherland DH, Davids JR (1993) Common gait abnormalities of the knee in cerebral palsy. Clin
Orthop Relat Res 288:139–147
Sutherland DH, Santi M, Abel MF (1990) Treatment of stiff-knee gait in cerebral palsy: a
comparison by gait analysis of distal rectus femoris transfer versus proximal rectus release. J
Thawrani D, Haumont T, Church C, Holmes L, Dabney KW, Miller F (2012) Rectus femoris
transfer improves stiff-knee gait in children with spastic cerebral palsy. Clin Orthop Relat Res
470(5):1303
Winter DA (1991) The biomechanics and motor control of human gait: normal, elderly and
pathological, 2nd edn. University of Waterloo Press, Waterloo, pp 35–52, 75–85
Winters TF, Gage JR, Hicks R (1987) Gait patterns in spastic hemiplegia in children and young
adults. J Bone Joint Surg Am 69:437–441
Stance Phase Problems in Cerebral Palsy
(Strength)
Justin Connor and Mutlu Cobanoglu
Abstract
Normal human gait results from a combination of several complex coordinated
activities. The variable loss of control that is associated with cerebral palsy (CP)
lesions can cause multiple interruptions in the gait cycle. To understand and
properly evaluate pathologic gait patterns and to provide proper medical remedies
for gait impairments due to CP, the clinician must understand the normal gait
cycle. Instrumented gait analysis is a tool that provides detailed information and
quantitative measurements throughout the gait cycle to evaluate individual gait
patterns that help surgeons plan appropriate interventions. There are several
prevailing abnormal gait patterns associated with spastic CP. These patterns are
categorized into those affecting stance vs swing phases of gait. The patterns at the
knee most often responsible for impeding the gait cycle by causing stance phase
instability are crouch gait and back-kneeing in sagittal plane kinematic. Patterns
such as back-kneeing cause knee hyperextension which is especially bad for
delaying push-off and propelling forward; crouch knee gait causes dropping
down into hip and knee flexion. These two types of gait patterns require different
types of intervention due to differing muscle responses. Identification of the
correct gait pattern in cerebral palsy is required to create the correct management
algorithm.
J. Connor
duPont Hospital for Children, Wilmington, DE, USA
e-mail: jconnor97@aol.com
M. Cobanoglu (*)
Department of Orthopedics and Traumatology, Adnan Menderes University Faculty of Medicine,
Aydın, Turkey
e-mail: drmutlu79@hotmail.com

DOI 10.1007/978-3-319-30808-1_54-1
2 J. Connor and M. Cobanoglu
Keywords
Cerebral palsy • Stance phase • Crouch gait • Back-knee • Ankle-foot orthosis •
Knee-ankle-foot orthosis • Ground reaction force
Abbreviations
AFO Ankle-foot orthosis
CP Cerebral palsy
GRAFO Ground reaction ankle-foot orthosis
HAT Head, arm, trunk
KAFO Knee-ankle-foot orthosis
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Back-Kneeing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Crouch Knee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Introduction
Normal human gait is composed of a complex combination of coordinated motor

activities which require precise control from the central nervous system. These motions
control balance, motor activity, and cognitive decision-making. The gait cycle is
divided into stance and swing phases. For gait to be productive and effective, there
are five prerequisites as described by Perry: stability in stance, swing phase clear-
ance, foot preposition in terminal swing, adequate step length, and energy conser-
vation (Gage 1993). The role of stance phase (comprises 60% of the normal gait
cycle) is to provide support for weight bearing and stability to provide power for
push-off. Swing phase (comprises 40% of the normal gait cycle) advances the limb
forward, thus allowing forward motion. These phases are supported by forces which
are created at the ankle, knee, and hip joints by internal moments produced
by attached muscles. The measurement of these forces at each joint through instrumented
gait analysis allows for evaluation of kinetic moments. These kinetic measures create
resultant vector forces by muscles on the bones of the body. Directly measuring
muscle force is clinically impractical; however, net joint forces may be indirectly
measured as the opposing ground reaction force. The residual of the ground reaction
force at each joint has a direction and distance from the defined center of the joint
that impacts gait. Concentric contraction of hip extensors, knee extensors, and ankle
plantar flexors in stance phase maintain a ground reaction force in front of the knee
joint during the stance phase, leading the erect posture during gait (Fig. 1) (Kadhim
and Miller 2014).
Stance Phase Problems in Cerebral Palsy (Strength) 3
Fig. 1 As shown in the

figure, concentric contraction
of hip extensors, knee
extensors, and ankle plantar
flexors in stance phase
maintain a ground reaction
force (red arrows) in front of
the knee joint during the
stance phase, leading the erect
posture during gait
What is happening during stance phase? On the first contact to the ground, the
knee is slightly flexed (approximately 5 ) to absorb the shock during initial heel
strike. At this moment the quadriceps and hamstrings contract isometrically to
stabilize the knee joint. Then the knee further flexes to 20 , shifting the center of
gravity forward to allow advancement. After the short period of flexion, second half
of stance starts. In this phase, the knee starts to extend with the activation of plantar
flexion–knee extension couple: the soleus contracts and slows the progression of the
tibia. This extension moment without activation of quadriceps muscle keeps the
ground reaction force anterior to the knee. The hamstrings function in a closed chain
to act as hip extensors. In the terminal phase of stance, limb acceleration occurs
with the concentric contraction of the plantar flexors, providing power. This complex
coordination of muscular activities is often difficult to maintain due to the neurologic
insult associated with CP. Orthopedists must understand normal and abnormal gait
cycle patterns and measurement techniques to be able to accurately evaluate patho-
logic gait and to improve function.
State of the Art
Muscular spasticity and contractures leading to loss of functional abilities can be

observed in ambulatory children with cerebral palsy resulting in pathologic gait
patterns (Bell et al. 2002; Johnson et al. 1997). These primary motor deficits can lead
to secondary biomechanical and growth problems such as muscle tightness, joint
contractures, and secondary bony malformations. Rodda et al. described a classifi-
cation based on the pelvis, hip, knee, and ankle position during stance to assist in
understanding these abnormal gait patterns. According to this classification, there
is a subset of patients in which the ankle is in equinus and knee and hip extend fully.
Spasticity of the hamstrings and hip flexors increases gradually and leads to flexion
of the knee and hip. This has been described as a jump gait. The classic crouch gait is
manifested by ankle dorsiflexion with hip and knee flexion throughout the stance
phase (Rodda et al. 2004). Increased body mass associated with the pubertal growth
spurt may advance a child from a jump gait pattern to a crouch gait pattern (Rodda
et al. 2006). The gait pattern also becomes asymmetric (Rodda et al. 2004).
Crouch gait and back-kneeing (genu recurvatum) are the common types of gait
patterns in ambulatory children with CP (Wren et al. 2005; Klotz et al. 2013). Both
patterns affect stance phase stability by affecting pre-postponing the limb for proper
push-off. The aim of this chapter is to investigate the properties and treatment results
of these gait patterns.
Back-Kneeing
The back-kneeing position is defined as hyperextension of the knee during the

midstance phase and is frequently found in children with bilateral spastic cerebral
palsy (Fig. 2). Eventually it may cause instability of the knee as the result of
progressive stretching of the posterior knee capsule and ligamentous structures
and over time cause secondary malformations of the tibial condyles as growth
proceeds (Simon et al. 1978; Klotz et al. 2014). This condition can primarily occur
as a result of three patterns. The first pattern is characterized by overactivity of the
Fig. 2 (a) The back-kneeing position is defined as hyperextension of the knee during the midstance
phase of stance. (b) Ground reaction force in stance phase of the left lower limb is shown as red
arrow which goes anterior to the knee axis. Average moments of lower limb joints in sagittal plan
are shown. (c) Ten to forty percent of the gait cycle increased the internal hip extension moment
seen because during the stance phase, the hip extensors try to compensate excessive hip flexion.
This creates an increased internal extension moment. (d) Due to the severe back-kneeing during the
stance phase, the knee flexor and capsule activate excessively to balance the high internal flexor
moment caused by the hyperextension. (e) Because of persistent plantar flexion, the length of
plantar flexors is shortened which limits the maximum internal plantar flexor moment needed after
midstance (red line demonstrates right leg; blue line demonstrates left leg in c, d, and e)
gastrocsoleus muscles. The overactive gastrocsoleus causes the hyperextended knee

through the coupling of ankle plantar flexion and knee extension. The second pattern
is movement of the HAT (head, arm, trunk) center of gravity anterior to the knee in
the presence of a weak gastrocnemius. The third pattern is movement of the HAT
center of gravity posterior of the hip joint but anterior of the knee joint (Simon
et al. 1978). Klotz et al. reported that equinus was a major underlying factor in
primary back-knee gait (Klotz et al. 2014). Furthermore, this problem may occur
secondarily after hamstring lengthening for the correction of flexed knee gait (Dreher
et al. 2012).
Treatment
The first consideration in the treatment of back-kneeing is to assess whether the

gastrocnemius has adequate length to allow for dorsiflexion past neutral with knee
extension. A hinged ankle-foot orthosis (AFO) with a 90 degree plantar flexion stop
is the preferred orthosis for children whose back-kneeing results from overactivity
of gastrocnemius spasticity. If dorsiflexion of the ankle with knee extension is
possible, articulated AFO that allows 3–5 of dorsiflexion while limiting plantar
flexion to minus 5 should be prescribed (Miller 2005). By setting the plantar flexion
stop at 5 of dorsiflexion, these children will be forced into knee flexion in stance if
they are independent ambulators. If the ground reaction force is moving either
significantly in front or behind the knee in the presence of a plantar flexor weakness,
a non-articulated solid ankle AFO should be used to control the ankle.
If individuals have any shortness of the gastrocsoleus, the knee will hyperextend
and go into back-kneeing. On the other hand, if individuals have plantar flexor
weakness, back-kneeing can also occur due to increased ankle dorsiflexion and a
HAT segment anterior to the knee. Instrumented gait analysis has shown an increased
trunk lean in patients walking with assistive devices. In spite of appropriate orthotics,
these assistive devices lead to an increased anterior trunk lean which may cause
progressive back-kneeing and the development of pain (Simon et al. 1978; Klotz
et al. 2014; Miller 2005). In the presence of this kind of progressive back-kneeing,
the treatment method with AFO must be changed with knee-ankle-foot-orthosis
(KAFO) with extension blocking hinges. But it is important to make sure that
there is no contracture of the gastrosoleus. Ankle dorsiflexion has to be 5–10 in
knee extension or the gastrocnemius should be lengthened (Dreher et al. 2012;
Miller 2005).
Crouch Knee
Crouch gait is common in ambulatory children with cerebral palsy (CP). The
definition of a crouch gait is increased knee flexion in midstance with increased
ankle dorsiflexion and usually increased hip flexion (Fig. 3). The crouch pattern may
be seen in all levels of severity; however, it is primarily encountered in moderate and
Fig. 3 (a) The definition of a crouch gait is increased knee flexion in midstance with increased
ankle dorsiflexion and usually increased hip flexion. (b) Ground reaction force in stance phase of
the lower limbs is shown as red arrow which goes posterior to the knee axis. Average moments of
lower limb joints in sagittal plan are shown. (c) In this the gait cycle increased internal hip flexion
moment seen at the end of stance phase in order to prepare the clearance during swing phase. (d)
Due to the persistent knee flexion, knee extensors activate throughout the stance phase to balance
the high knee internal extension moment caused by the knee flexion to prevent the possible collapse.
(e) Because of persistent dorsiflexion, the plantar flexors activate to balance high ankle internal
dorsiflexion moment; however, due to weakness of plantar flexors, the balance has to be maintained
with strong quadriceps contraction (red line demonstrates right leg; blue line demonstrates left leg in
c, d, and e)
severe diplegia. The ground reaction force is maintained close to the centers of the
hip, knee, and ankle joints, reducing the demands on the antigravity support muscles.
Failure of the total body extensor moment as a result of diminished ability of the hip,
knee, or ankle plantar flexor moments may result in collapse of the extension posture
into a flexion posture, described as crouch gait (Rodda et al. 2006). The rapid onset
of weight and height growth may cause knee flexion contractures, hamstring con-
tractures, deficient foot moment arm, and gastrocsoleus weakness which can con-
tribute to the classic crouch gait. Knee flexion in stance phase increases and the foot
starts to dorsiflex, and severe planovalgus foot deformities, characterized by heel
equinus, talonavicular joint dislocation, and midfoot break during stance, develop
and reduce lever arm (Kadhim and Miller 2014). The toe walking with knee flexion
pattern (jump gait) is typically not seen in full adolescence or nearly adult-sized
individuals. The muscles and joints are not strong enough to support the body weight
for chronic ambulation with the typical early childhood toe-walking pattern.
The other lever arm dysfunctions affecting stance stability are rotational devia-
tions at the femurs, tibias, and feet. External tibial torsion reduces the extension
capacity of the soleus and hip extensors, as it misdirects the forward propulsion
vector, thus decreasing power, especially if the torsion is greater than 30 of normal
(Hicks et al. 2007). Furthermore, the planovalgus foot is usually associated with
external tibial torsion. In the presence of gastrosoleus weakness in crouch gait,

the ground reaction force creates a flexion moment on the knee and hip by passing
through the knee posteriorly and the hip anteriorly. To prevent the knee from
collapsing into flexion, quadriceps activity must increase during stance. The rectus
femoris however not only acts as a hip extensor of the quadriceps but also contrib-
utes to hip flexion in a closed chain. Hip extensors are mechanically disadvan-
taged and the hamstrings cause increasing knee flexion. Over time, fixed hamstrings
and knee flexion contracture, which are responsible for further exacerbation of
crouch, may develop. The patellar tendon elongates gradually and extensor mech-
anism weakens.
A secondary etiology for crouch may be a significant hip flexion contracture,
which can limit knee extension in midstance. Although hamstring contracture is
considered a primary cause for crouch, many patients have normal length hamstrings
(Arnold et al. 2006; Delp et al. 1996). And also it has also been observed that
excessive knee flexion typically accompanies excessive hip flexion due to shortened
hip flexors throughout the gait cycle (Delp et al. 1996). The shortened hip flexors
lead to anterior pelvic tilt facilitating hamstring tightness and causing knee flexion
in stance phase (Kedem and Scher 2016).
As a result of increasing crouch, the stress on the knee extensor mechanisms to
support weight bearing may lead to the complaint of knee pain. Standing with >30
of knee flexion increases the forces acting on the quadriceps, patella, and proximal
part of the tibia and requires the quadriceps muscle to work at >50 % of its
maximum moment-generating capacity in order to stabilize the knee joint (Perry
et al. 1975). Tibial tubercle apophysitis or even chronic fractures through the distal
pole of the patella may occur, especially during rapid growth. Progressive failure of
the knee extensor mechanism is associated with gait deterioration, increased depen-
dence on walkers or crutches, and the need for wheelchair use in the community
(Rodda et al. 2006). As gait deteriorates the position of the feet may progress into
increased. Orthotics lose their ability to support the collapsing feet, and ultimately
the child may lose the ability and motivation to walk (Miller 2005).
A child’s weight plays an important role in the evolution of gait patterns through
maturity. It should be monitored on every clinic visit particularly throughout the
pubertal growth spurt. During this period the child may exhibit new complaints of
pain due to increasing stress on the knees or feet. Also, the physical examination
should be monitored, being mindful of passive knee extension and popliteal angle
measurements, to monitor progressive hamstring contractures or fixed knee flexion
contractures that may impede gait.
Objective measures to evaluate crouch gait through instrumented gait analysis
include kinematic evaluation of the magnitude of the knee flexion in midstance,
excessive dorsiflexion of the ankle, and the knee excessive internal knee extension
moments in midstance. If the range of motion (ROM) of ankle is normal or below
normal, and ROM does not increase with the knee flexion, the ankle weakness and
foot moment arm are the most likely reasons of the crouch. If the knee extends to
the limits of the fixed knee flexion contracture measured on physical examination,
the knee joint contracture is most likely a cause. If the ankle has a high plantar
flexion moment with a high knee flexion moment, a combination of contracture of

the gastrocnemius and the hamstrings is most likely a cause. If the hip extension
peak occurs early, and a significant hip flexion contracture is positive in the physical
examination, hip flexion contracture may also contribute to knee flexion deformity in
the midstance phase (Miller 2005).
Treatment
Appropriate treatment for crouch gait should focus on early detection, by monitoring
examinations every 6 months during middle childhood and intervention before the
problem becomes severe. Mild crouch can be initially controlled conservatively,
with spasticity management, physical therapy, and foot orthoses. But the role of
physical therapy and strengthening exercises in the treatment of crouch
is controversial. A meta-analysis of randomized trials concluded that strength-
ening interventions had no effect on strength, walking speed, or activity level in
children with CP (Scianni et al. 2009), though there are reports that find that
strengthening exercises may lead to some functional improvement in patients with
crouch (Damiano et al. 1995, 2009).
The aim of the interventions in these children is to reduce knee flexion and prevent
gait deterioration ultimately maintaining gait efficiency and walking activity in daily
life. Initially a solid AFO provides stability to the ankle and foot and provides a
stable base of support for children to stand. These can usually be used at the
preambulatory stage between the ages of 18 and 24 months. It is easy to done and
works well for child less than 30 kg. As children gain better stability and start to walk
using a walker, usually between the ages of 3 and 4 years, the ankle hinge can be
added to allow dorsiflexion but limit plantar flexion. Most children who have good
walking ability with diplegic and hemiplegic pattern involvement benefit from the
transition to a hinged AFO at approximately 3 years of age. But hinged AFO is
contraindicated if children have severe planovalgus or varus foot deformity and
increased knee flexion in stance (Miller 2005). A rigid AFO can compensate for
weakness of ankle plantar flexors and may normalize knee kinematics and kinetics
effectively, but it has the disadvantage of inhibiting push-off power (Kerkum et al.
2015). Although AFO may affect ankle joint dorsiflexion during stance, ankle power
and ankle plantar flexion moment, it does not affect proximal joint movement during
gait (Rethlefsen et al. 1999). A ground reaction ankle-foot orthosis (GRAFO) can be
used to enhance push-off power and is a commonly applied intervention in children
with CP walking with crouch. A GRAFO features a solid pretibial shell to more
effectively redirect the GRF vector and slow tibial progression as the center of
pressure moves distally under the foot during stance (Kane et al. 2010). Extremities
with knee and hip flexion contractures of <10 demonstrate significantly better peak
knee extension in midstance with the GRAFO than the extremities with knee and hip
flexion contractures of >15 (Rogozinski et al. 2009). The GRAFO limits ankle
dorsiflexion and reduces knee flexion in stance. This improved ankle and knee
position reduces the knee extensor moment in stance. The ankle can be brought to
neutral dorsiflexion with the knee in full extension. If this is not possible, the orthosis
cannot work and these children will require gastrocnemius and hamstring lengthen-
ing to accommodate the orthosis. The other point is that the foot-to-knee axis has
to be in a relatively normal alignment which defines less than 20 of internal or
external tibial torsion (Miller 2005). Transverse plane skeletal malalignments and
foot segmental malalignments (planovalgus deformities) may lead to an increased
external foot progression angle which moves the line of action of the ground reaction
force lateral to the knee joint center. Recent studies utilizing dynamic computer
models of the musculoskeletal system have demonstrated that increased external
tibial torsion impedes the capacity of the soleus to extend the knee during stance,
which supports the hypothesis of lever arm dysfunction as an important contributor
to crouch gait (Hicks et al. 2007). As these children get heavier, this orthosis
becomes more effective; however, it also has to be made stronger. As children
approach 50–70 kg, the orthosis has to be constructed with a composite of carbon
fiber or laminated copolymer to withstand the applied forces (Miller 2005).
The surgery for crouch gait often includes many procedures at different joints.
Single-event multilevel surgery is widely performed in cerebral palsy and focuses
on improving gait function and pattern (Rodda et al. 2006). The preferred order is
to start from proximal to distal. First the hip rotation should be corrected, with
iliopsoas lengthening if needed. Distal hamstring lengthening if indicated should
also be performed. Repeat hamstring lengthening may prevent or delay progres-
sive crouch in patients with CP but does not result in long-term improvement in
crouch gait (Rethlefsen et al. 2013). The hamstring lengthening followed by knee
capsulotomy or femoral extension osteotomy (Novacheck et al. 2009) is
performed if indicated. The surgery for foot deformity is done next; then an
intraoperative assessment of the torsional alignment is used to make the final
determination of the need for a tibial osteotomy. After the tibial osteotomy, the hip
fully extends and the knee can be fully extended and lies in approximately 10 of
external rotation. The foot-to-thigh alignment should be 20 external to neutral
with neutral dorsiflexion (Kadhim and Miller 2014; Miller 2005). If the foot has a
significant planovalgus or a midfoot break, it must be corrected. A stable and
correctly aligned foot is mandatory in the correction of crouch because the ground
reaction force has to be controlled through the foot as a functional moment arm
(Kadhim and Miller 2014). Postoperative rehabilitation should start in the hospi-
tal with the goal of having children at least stand before discharge and plan for
immediate home rehabilitation. Parents need to expect that the acute rehabilita-
tion will take 3 months until these individuals are close to their preoperative
function, and then it will take at least 1 year of rehabilitation to reach maximum
function. If there is weakness or a tendency for the gastrocsoleus not to have good
strength, a GRAFO has to be used postoperatively. This is the ideal time to use the
articulated GRAFO, which will allow the gastrocsoleus to gain strength, and over
1–2 years, the orthotic can be weaned away and the correction will be maintained
(Miller 2005).
Conclusion
There are a variety of factors that affect stance phase stability of the lower limb
segment in gait. These factors may lead to progressive deterioration of gait and
function and may ultimately lead to pain and disability. Through the use of routine
monitoring of pubertal maturation, physical examination, and analysis of kinematics
and moment kinetics through instrumented gait analysis, appropriate treatment plans
can be acted upon to prevent progression of deformity and loss of function. Appro-
priate bracing adjuvants should also be utilized in addition to surgical interventions.
Cross-References
▶ Diagnostic Gait Analysis

▶ Natural History of CP and Outcome Assessment
▶ Spasticity Effect in CP Gait
References
Arnold AS, Liu MQ, Schwartz MH, Ounpuu S, Delp SL (2006) The role of estimating muscle-
tendon lengths and velocities of the hamstrings in the evaluation and treatment of crouch gait.
Gait Posture 23(3):273–81. Epub 2005 Jun 17
Bell KJ et al (2002) Natural progression of gait in children with cerebral palsy. J Pediatr Orthop 22
(5):677–682
Damiano DL, Kelly LE, Vaughn CL (1995) Effects of quadriceps femoris muscle strengthening on
crouch gait in children with spastic diplegia. Phys Ther 75(8):658–667. discussion 668–71
Damiano DL et al (2009) Can strength training predictably improve gait kinematics? A pilot study
on the effects of hip and knee extensor strengthening on lower-extremity alignment in cerebral
palsy. Phys Ther 90(2):269–279
Delp SL et al (1996) Hamstrings and psoas lengths during normal and crouch gait: implications for
muscle-tendon surgery. J Orthop Res 14(1):144–151
Dreher T et al (2012) Development of knee function after hamstring lengthening as a part of
multilevel surgery in children with spastic diplegia. J Bone Joint Surg 94(2). doi:10.2106/jbjs.
j.00890
Gage JR (1993) Gait analysis. An essential tool in the treatment of cerebral palsy. Clin Orthop Relat
Res 288:126–134
Hicks J et al (2007) The effect of excessive tibial torsion on the capacity of muscles to extend the hip
and knee during single-limb stance. Gait Posture 26(4):546–552
Johnson DC, Damiano DL, Abel MF (1997) The evolution of gait in childhood and adolescent
Kadhim M, Miller F (2014) Crouch gait changes after planovalgus foot deformity correction in
ambulatory children with cerebral palsy. Gait Posture 39(2):793–798
Kane K, Kyra K, John B (2010) Comparison of ground reaction and articulated ankle-foot orthoses
in a child with lumbosacral myelomeningocele and tibial torsion. J Prosthetics Orthot 22
(4):222–229
Kedem P, Scher DM (2016) Evaluation and management of crouch gait. Curr Opin Pediatr 28
(1):55–59
Kerkum YL et al (2015) The effects of varying ankle foot orthosis stiffness on gait in children with
spastic cerebral palsy who walk with excessive knee flexion. PLoS One 10(11):e0142878
Klotz MCM et al (2013) Reduction in primary genu recurvatum gait after aponeurotic calf muscle
lengthening during multilevel surgery. Res Dev Disabil 34(11):3773–3780
Klotz MCM et al (2014) The association of equinus and primary genu recurvatum gait in cerebral
palsy. Res Dev Disabil 35(6):1357–1363
Miller F (2005) Cerebral palsy. Springer Science & Business Media, New York
Novacheck TF et al (2009) Distal femoral extension osteotomy and patellar tendon advancement to
treat persistent crouch gait in cerebral palsy. Surgical technique. J Bone Joint Surg 91(Suppl
2):271–286
Perry J, Antonelli D, Ford W (1975) Analysis of knee-joint forces during flexed-knee stance. J Bone
Joint Surg 57(7):961–967
Rethlefsen S et al (1999) The effects of fixed and articulated ankle-foot orthoses on gait patterns in
subjects with cerebral palsy. J Pediatr Orthop 19(4):470–474
Rethlefsen SA et al (2013) Repeat hamstring lengthening for crouch gait in children with cerebral
palsy. J Pediatr Orthop 33(5):501–504
Rodda JM et al (2004) Sagittal gait patterns in spastic diplegia. J Bone Joint Surg 86(2):251–258
Rodda JM et al (2006) Correction of severe crouch gait in patients with spastic diplegia with use of
multilevel orthopaedic surgery. J Bone Joint Surg 88(12):2653–2664
Rogozinski BM, Davids JR, Davis RB 3rd, Jameson GG, Blackhurst DW (2009) The efficacy of the
floor-reaction ankle-foot orthosis in children with cerebral palsy. J Bone Joint Surg Am 91
(10):2440–2447
Scianni A et al (2009) Muscle strengthening is not effective in children and adolescents with
cerebral palsy: a systematic review. Aust J Physiother 55(2):81–87
Simon SR et al (1978) Genu recurvatum in spastic cerebral palsy. Report on findings by gait
analysis. J Bone Joint Surg Am 60(7):882–894
Wren TAL, Rethlefsen S, Kay RM (2005) Prevalence of specific gait abnormalities in children with
cerebral palsy: influence of cerebral palsy subtype, age, and previous surgery. J Pediatr Orthop
25(1):79–83
Spasticity Effect in Cerebral Palsy Gait
Marlene Cristina Neves Rosa and André Gonçalo Gomes Roque
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Spastic Gait Disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Methodologies to Assess Spasticity During Gait in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Spasticity Treatments with Impact in CP Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Abstract
A high number of children with cerebral palsy (CP) have spastic gait and
consequently abnormalities in joint patterns. Several factors have been contrib-
uting to the lack of consensus on the spasticity effect in cerebral palsy gait and
would be summarized and discussed in this chapter, e.g., spastic gait patterns are
in constant evolution during the process of growth; there are still considerable
limitations in the methodologies used to assess spasticity during gait; a wide
range of rehabilitation strategies have been explored to control spasticity during
gait. Spastic gait patterns are divided in hemiplegic (5 types) and diplegic
(4 types) with the most prevalent joint abnormalities described in the sagittal
plane. Ashworth, Tardieu, and DAROM scales and Pendulum Tests are widely
used to assess spasticity but do not reliably explain the spasticity effects during
gait. Orthotics, adequate exercise or handling techniques, botulinum injections, or
surgical procedures have been used to manage spasticity effects.
M.C.N. Rosa (*)

Piaget Institute, Viseu, Portugal
e-mail: planyergos@planycorpo.pt
A.G.G. Roque (*)
Physiotherapy, University of Averio, Aveiro, Portugal
e-mail: andreroque87@hotmail.com

DOI 10.1007/978-3-319-30808-1_55-1
2 M.C.N. Rosa and A.G.G. Roque
Keywords
Cerebral palsy • Gait • Spasticity
Introduction
Spasticity is one of the sequelae of neuromuscular disabilities, affecting 260–340 per

100,000 children with cerebral palsy (Dana and Cub 2013). Spasticity is the result of
pyramidal tract lesions, but there are several theories to explain the neurophysiologic
mechanisms related with this phenomenon (Dana and Cub 2013). The most com-
monly used definition of spasticity is that of Lance (1980) that explains it as
“a velocity-dependent increase in tonic stretch reflexes (muscle tone) with exagger-
ated tendon jerks. . .” (Hobart et al. 1994). It depends on overexcitability of spinal
alpha motor neurons, as a consequence of the interruption of descending modulatory
carried by the corticospinal, vestibulospinal, and reticulospinal tracts (Filloux 1996).
In specific, spasticity can be explained by the following pathophysiologic mecha-
nisms: reduced reciprocal inhibition of antagonist motor neuron pools by Ia affer-
ents, decreased presynaptic inhibition of Ia afferents, and decreased nonreciprocal
inhibition by Ib terminals (Hobart et al. 1994).
Despite this lack of consensus explaining the neurophysiologic mechanisms
underlying spasticity, it is one of the most common disabling consequences in
cerebral palsy, causing contractures, joint subluxations, and, indirectly, fatigue,
loss of dexterity and coordination, and balance disorders (Dana and Cub 2013). As
cerebral palsy (CP) is a life-lasting condition, spasticity also affects the growth of the
individual, hampering both muscle and skeletal development and resulting in sym-
metric or asymmetric biomechanical deficiencies depending on the PC type (Bar-On
et al. 2015). Also, one has to consider that CP originates in the infancy or childhood,
and the consequences of spasticity have to be separated from that resulting from
adult lesions. In the second case, motor control has been normally developed, as
opposed by the first case in which spasticity-related gait alterations will direct the
motor learning lifelong. These consequences have serious impact in patients´ func-
tionality, specifically in gait. Gait disorders due to spasticity can assume different
patterns, depending on the muscles affected, which may cause different effects in
joint kinematics (Miller 2004). The first section (“Spastic Gait Disorders”) of this
chapter will summarize the types of spastic gait disorders and its characteristics,
according to the muscles affected.
In addition, efficient methodologies to assess spasticity are crucial to understand
its effect in gait pattern (Scholtes et al. 2006). Most of the instruments described in
the literature are not able to assess spasticity under rhythmic motor tasks such as gait
(Scholtes et al. 2006). Alternative methodologies (e.g., electromyography) have
critical aspects that would be discussed in this chapter as they are not aligned with
the original definition of spasticity. Section two (“Methodologies to Assess Spas-
ticity During Gait in Cerebral Palsy”) will summarize weaknesses and strengths of
different scales and instruments that have been used to assess spasticity during gait
or the related phenomenon.
Spasticity Effect in Cerebral Palsy Gait 3
Section three (“Spasticity Treatments with Impact in CP Gait”) of this chapter

will give a simple perspective of the widely accepted approaches to manage spas-
ticity and its effects on CP gait, not focusing on specific results or merits of each.
State of the Art
The knowledge of CP spastic gait patterns is well established and despite some children
may present with mixed patterns, often they walk in a typical bilateral equinus. Also
spasticity contribution to the different gait patterns is defined in the literature; however,
the specific association between spasticity and different components and phases of gait
is only recently subjected of throughout investigation and will not therefore be
exploited in this chapter. This fact is related to the newly available research methods
that allows to study spasticity during gait but also to the capacity to establish associ-
ations between clinical and research measures of CP gait and spasticity.
The methods and therapeutics that address spasticity and CP gait are numerous;
some of them are well established for a long time with different degrees of success
(orthopedic surgery, stretching, botulinum toxin A, orthotics), while some are more
recent yet promising (selective dorsal rhizotomy, intrathecal baclofen, virtual reality,
and transcranial magnetic stimulation). The choice between different approaches is a
truly clinical decision, which should be based on CP gait analysis, the available
evidence, and the individual presented to evaluation. The only conclusion that our
readers should obtain toward the treatment theme is that a throughout evaluation of
the characteristics of CP gait is fundamental to the success of each of the selected
interventions. As such, it is not our objective in this chapter to demonstrate the
results of the different approaches but to indicate the existing managing possibilities
and, when they were studied, the expected gait outcomes of those interventions and
possible implications for the future of the adolescent and adult with CP.
Spastic Gait Disorders
Cerebral palsy (CP) is traditionally classified by the clinical type (motor dysfunc-
tional) and topography (anatomical region of the lesion). The motor dysfunctional
patterns are classified as (i) spastic, which is the most common pattern, and
(ii) ataxic, hypotonic, dyskinetic, and mixed (Morais Filho et al. 2014). Approxi-
mately two thirds of all patients with CP suffer from spasticity (Awaad and Rizk
2012). The most affected muscles are the antigravity muscles. Commonly, and as a
result of this spasticity, legs are in extended and adducted position and arms are in
flexed, internal rotation and pronated position (Awaad and Rizk 2012).
In gait, several deviations have been reported as a consequence of spasticity in
hemiplegic and diplegic patients, most of them based on joint deviations in the
sagittal plane (O’Byrne et al. 1998). Ultimately, these deviations reduce walking
speed and increase energy expenditure, which cause functional inability (Piccinini
et al. 2007; Novacheck et al. 2000). Spastic motor deviations are consistent from
stride to stride and day to day (Rodda and Graham 2001). Therefore, detectable
changes are generally seen as a result of an intervention or as a change related with
the aging process and therefore should be monitored during the rehabilitation
process (Rodda and Graham 2001).
There are at least four different gait patterns in hemiplegic patients with CP all
characterized by more distal involvement, which means that true equinus is the basis of
the most common gait hemiplegic patterns. In hemiplegia type 1, the most prominent
characteristic is drop foot during swing phase caused by inability to control
dorsiflexors. Two subtypes characterize hemiplegia type 2. The subtype 2.A includes
equinus, neutral knee, and extended hip, while subtype 2.B is characterized by equinus,
recurvatum knee, and extended hip. This is the most common type found in clinical
practice and it is observed in the stance phase (spasticity in gastrocnemius). Drop foot is
also noted during the swing phase because tibialis anterior is impaired. Associated with
the predominance of ankle plantarflexion, the knee may adopt the position of
recurvatum or extension. The main characteristics of the hemiplegic gait type 3 is the
impaired ankle flexion in swing due to gastrocnemius spasticity or contracture and the
“stiff knee gait” because of co-contraction between hamstrings and quadriceps. Finally,
the hemiplegia type 4 is similar to the pattern of spastic diplegia (but unilateral,
therefore asymmetric) with increasing proximal involvement. The characteristics in
the sagittal plane are flexed knee, flexed hip, and anterior pelvic tilt. Two other planes
present abnormalities in lower limb posture. In the coronal plane, there is hip adduction
and in the transverse plane, there is internal rotation. Table 1 summarizes gait hemi-
plegic patterns in people with cerebral palsy (Rodda and Graham 2001).
In contrast with hemiplegic gait, diplegic gait is characterized by more proximal
involvement, which means that apparent equinus and crouch gait are very common
(Rodda et al. 2004). The five different diplegic gait patterns are schematically
described in Table 2.
In the up-down direction across the table, there is a decreasing equinus, an
increasing proximal involvement, and a change in the direction of the ground
reaction force, i.e., from the front of the knee to behind it. The key muscles in
these patterns are hip and knee flexors and ankle plantarflexors. These deformities in
sagittal gait plane are probably responsible for others found in transverse plane such
as torsional deformities of the femur and tibia. The true equinus hemiplegic gait
pattern (type 1) is one of the most incident patterns and it is characterized by an
equinus and a fully extended knee in late stance. About the jump gait (type II), it is
characterized by hip and knee flexion during all stance phase, and then, in late
stance, the main characteristic is the equinus. These deformities give the appearance
that the individual is jumping up and down, which explains the identification of this
gait pattern. In apparent equinus (type III), there are knee and hip flexion contrac-
tures that explain why the heel is not in contact with the ground, despite it is not a
true equinus foot. In the crouch gait (type IV), three simultaneous conditions are
confirmed: hip and knee flexed and ankle kinematics in calcaneus range. Finally,
each lower limb presents a different kinematics pattern in the asymmetrical gait
(type V), i.e., one lower limb is usually classified as belonging to group III and the
other is considered as belonging to group II (Rodda et al. 2004).
Table 1 Characteristics of gait patterns in spastic cerebral palsy hemiplegia (Rodda and Graham
2001)
Characteristics
Spastic Joint Plane of Problematic Origin of the
hemiplegia deviation movement gait phase problem Management
Type I Drop foot Sagittal SW Loss of Ankle foot
Drop foot selective orthosis
control in
dorsiflexors
No calf
contractures
Type II Equinus + Sagittal ST Spasticity/ Botulinum toxin
2.a True neutral knee contracture of type A (spasticity)
equinus + extended gastrocnemius/ Orthotic support
2.b. True hip soleus muscles (mild contracture)
equinus/ Equinus + Tendo Achilles +
recurvatum recurvatum calf lengthening
knee knee + (fixed
extended hip contractures)
Type III Ankle Sagittal SW Spasticity/ Botulinum toxin
True dorsiflexion contracture of type A (spasticity)
equines/ impaired gastrocnemius/ Tendo Achilles +
jump knee Flexed stiff soleus muscles hamstring
knee Hamstring/ lengthening
quadricep (contractures)
co-contraction Orthotic support
(plantar flexion +
knee extension
couple)
Type IV Equinus + Sagittal ST Spasticity/ Botulinum toxin
Equinus/ flexed stiff Coronal SW contracture of type A (spasticity)
jump knee knee + Transverse gastrocnemius/ Tendo Achilles +
Pelvis flexed hip + soleus muscles hamstrings + hip
rotation/hip anterior Hamstring/ adductors +
flexion/ pelvic tilt quadricep iliopsoas
adducted/ Hip co-contraction lengthening
internal adduction Spasticity/ (contractures)
rotation Hip internal contracture of Orthotic support
rotation hip adductors (plantar flexion +
and hip flexors knee extension
couple)
Joint problems may be primary or secondary, with the related problem in the joint
or elsewhere. Secondary joint adjustments can be spontaneously reversed with the
solving of the primary affection. However, if left untreated, it can become a primary
problem that progresses into adulthood. Usually asymmetrical gait has one lower
limb toe walking due to spasticity; however, if the child is strong enough, the
tendency is for gait symmetry. Thus, the contralateral lower limb will also toe
walk and develop a fixed contracture that will need surgical addressing, becoming
itself a primary joint problem (Miller 2004).
Table 2 Characteristics of gait patterns in spastic cerebral palsy diplegia (Rodda et al. 2004)
Characteristics
Spastic Plane of Problematic Origin of the
diplegia Joint deviation movement gait phase problem
Type I Anterior/normal pelvic tilt Sagittal Stance phase Contracture of the
True + recurvatum/normal knee hamstrings
equinus position + drop foot Contracture of calf
muscles
Type II Anterior/normal pelvic tilt Sagittal Stance phase Flexion
Jump gait + knee and hip from contractures of knee
excessively flexed to and hip
extended
Type III Anterior/normal pelvic tilt Sagittal Stance phase Hamstrings and
Apparent + knee and hip flexed psoas contraction
equinus
Type IV Posterior/normal pelvic Sagittal Stance phase Hamstrings and
Crouch gait tilt + knee and hip flexed + psoas contraction
foot in excessive
dorsiflexion
Type V One lower limb – pattern Sagittal Stance phase Lower limb (pattern
Asymmetric III II) – Flexion
gait The other – pattern II contractures of knee
and hip
Lower limb (pattern
III) – Hamstrings
and psoas
contraction
During adolescence there is a rapid decrease in the strength ratio (Miller 2004).
The association of this factor with a spasticity growth impairs normal muscle
development, leaving the young adults even weaker.
In adults, muscle contractures due to spasticity develop much more rapidly than
in childhood and results in severe amplitude limitations. The following fixed con-
tractures are not associated with joint structural alterations, but they reduce the
variability of movements available to motor control, promoting its decrease to a
more gross pattern (Miller 2004).
Nevertheless, joint deformities occur in adults with CP independently of the
ambulatory status, diagnosis, or severity of involvement (Kembhavi et al. 2011).
The combined impairments contribute to a sense of frustration in the adolescents,
who lose the motivation to walk (Miller 2004).
Adolescents with CP are usually heavier, and as such they are more prone to fall
and be hurt during the adolescent clumsy stage (a normal feature of gait develop-
ment). These falls may cognitively repress the gait behavior in this population,
worsening the functional prognosis. However, it is difficult to assess if deterioration
of mobility is due to the pathophysiology associated with CP, or the physical effects
associated with aging, or an interaction of both (Kembhavi et al. 2011).
Height and weight growth contributes to the development of the crouch gait. The
muscles and joints become unable to support the body in the child typical toe-walking
pattern, and knee flexion, hip flexion, and dorsiflexion increases, which can cause a
midfoot and hindfoot collapse and a consequential severe planovalgus (Miller 2004).
Young adults may also develop a back-kneeing pattern, particularly those using
walking aids, with severe weakness of the gastrocnemius and often those who had
tendon Achilles transections (Miller 2004).
Methodologies to Assess Spasticity During Gait in Cerebral Palsy
Clinical evaluation of spasticity and its contribution to gait alterations in PC was

traditionally a subjective analysis. Manual testing of spasticity uses non standardized
velocities, which are frequently under the threshold for eliciting the stretch reflex,
and is dependent on a relaxed state of the individual. As such it cannot differentiate
between the contributions of neural and nonneural factors to spasticity. This is of
relevance in the selection of treatment options, with medication used primarily in the
first situation and casts or orthotics in the second. Also selective dorsal rhizotomy
produces results only in a neural predominant spasticity (Bar-On et al. 2015).
Different subjective scales have been used for clinical tone assessment, such as
(i) the Ashworth Scale (AS)/Modified Ashworth scales (MAS) (Table X), (ii) the
Tardieu Scale (TS) and the Modified Tardieu Scale (MTS) (Table Y), (iii) the
Dynamic Evaluation Range of Motion (DAROM), and (iv) the Pendulum Test.
However, as stated before, the reliability of these instruments have been criticized
because the results obtained are dependent on some differences across inter-raters or
day-testing conditions, e.g., the range of motion (ROM), the velocity of movement,
and the position of the tested muscle (Boyd and Graham 1999). In specific, the
Tardieu Scale and the Dynamic Evaluation of Range of Movement (DAROM) have
reported more specifications that improved reliability in assessing spasticity, as, for
example, the ROM is defined as slow or fast passive stretching. Moreover, the
DAROM identifies a “range of motion deficit,” which means a value from the
minimal muscle stretch position (Pandyan et al. 1999).
When using the Ashworth Scales, an assessor tests the resistance to passive
movement about a joint. Then, the resistance perceived while moving a joint through
its full range of movement – except in grade “4” (Table 3) – is graded in a 5-Likert
scale. The Ashworth Scale and the Modified Ashworth Scale can be used as
measures of resistance to passive movement, but not as an ordinal level measure
of spasticity (Pandyan et al. 1999; Damiano et al. 2002).
The Tardieu Scale (Table 4) rates the spasticity as the difference between the
reactions to stretch at two extreme velocities (Gracies et al. 2005):
– The slowest: below the threshold of any significant stretch reflex (output =
passive range of motion).
– The fastest: maximizes the involvement of the stretch reflex, and if any spasticity
is present, then the rater feels the sensations of catch and release/clonus/fatigable.
Table 3 The Ashworth and modified Ashworth scales

Score Ashworth scale Modified Ashworth scale
0 No increase in tone No increase in muscle tone
1 Slight increase in tone giving a catch Slight increase in muscle tone, manifested
when the limb was moved in flexion or by a catch and release or by minimal
extension resistance at the end of the range of motion
when the affected part(s) is moved in flexion
or extension
1+ Slight increase in tone giving a catch Slight increase in muscle tone, manifested
when the limb was moved in flexion or by a catch, followed by minimal resistance
extension throughout the remainder (less than half) of
the range of movement (ROM)
2 More marked increase in tone but limb More marked increase in muscle tone
easily flexed through most of the ROM, but affected
part(s) easily moved
3 Considerable increase in tone, passive Considerable increase in muscle tone,
movement difficult passive movement difficult
4 Limb rigid in flexion or extension Affected part(s) rigid in flexion or extension
Table 4 The Tardieu Scale principles and grading system

Principles Grading
1. Grading always performed: X = spasticity angle (threshold)
Muscle at rest before the stretch maneuver (Angle of arrest at slow speed (V1) – angle of
Reproducible velocity of stretch: once the fast catch at fast speed (V3))
velocity is selected for a muscle, it remains Y = spasticity grade
always the same 0 = no resistance throughout passive
At the same time of the day movement
2. Velocity of stretch
1 = slight resistance throughout passive
Slow: V1, as slow as possible/slower than the
movement
rate of natural drop of the limb segment under
gravity 2 = clear catch at precise angle, interrupting
Fast: either V2 and V3; V2 is equivalent to the passive movement, followed by release
speed of the limb segment falling under 3 = fatigable clonus (<10s when maintaining
gravity; V3 is equivalent to faster than the rate pressure) occurring at a precise angle, followed
of natural drop of the limb segment under by release
gravity 4 = unfatigable clonus (>10s when
maintaining pressure occurring at a precise
angle)
Catch without release
Catch with minimal release
Angle 0
For grades 0 and 1, spasticity angle X = 0 by
definition
Comparing the characteristics of Ashworth and Tardieu Scales, it can be con-

cluded that spasticity measured by the AS is confounded by muscle contracture,
while the TS is able to differentiate spasticity from contracture and therefore is a
more valid tool (Emily Patrick and Ada 2006).
Table 5 Dynamic Evaluation of Range of Motion (DAROM) and pendulum tests outcomes
Measurement Parameter Description
Range of motion deficit DROM I Range of motion deficit following a slow
(DROM) for DAROM tests: T1, velocity stretch (V1) – expressed in degrees
T2, T3, T4 DROM II Range of motion deficit after a fast velocity
stretch (V3) – expressed in degrees
ASO Value calculated as the difference between the
DROM II and DROM I – expressed in
degrees
Pendulum Test Ex. First swing excursion – difference between
the starting angle and the first angle of reversal
of the swinging limb; expressed in degrees
t Duration of the pendulum swings (sec.)
n Counting the maxima of the sinusoidal waves
produced by the swinging limb after the heel
was released
The Dynamic Evaluation of Range of Motion (DAROM) and the Pendulum Test
complete the most widely used instruments for assessing spasticity in children
with CP. The DAROM relies on the same velocity principle of TS, as it considers
at least two different velocities of passive muscle stretching. The DAROM identifies
a “range of motion deficit” (DROM), defined as a value from the minimal muscle
stretch position. Using this instrument, two joint angles are measured: DROM I,
defined as the range of motion deficit following a slow velocity stretch, and
DROM II, defined as the angle of catch after a fast velocity stretch. The difference
between DROM II and DROM I indicates the examined muscle group’s level of
contracture and is called the angle of spasticity (ASO) (Bax et al. 2005). Otherwise,
the Pendulum Test was first described by Wartenberg and it consists of a biome-
chanical method that evaluates muscle tone (discriminating between various degrees
of spasticity) using gravity to induce the muscle stretch reflex during passive
swinging of the lower leg. The following table (Table 5) shows a summary of the
outcome measures that can be collected using the DAROM and Pendulum Tests
(Domagalska et al. 2013).
When considering how to treat spasticity, in addition to documenting changes in
the resistance to passive movement (Pandyan et al. 1999), it may also be relevant to
document changes in function. To further understand the role of spasticity in
functional status of patients with CP, the literature has been focused in exploring
possible relationships between spasticity and other gait parameters. Considering this
problematic, there are still relevant controversies that may affect the quality of
spasticity management in gait. For example, while Damiano and Abel (Abel
et al. 2003) reported a significant correlation between the knee extensor Ashworth
score and gait impairments, Gage (2004) and Boyd and Graham (1999) confirmed
that rectus femoris spasticity is a possible cause of stiff knee gait; however, Wren
et al. (2007) questioned the correlations between gait deviations and knee flexors
spasticity and contractures. Moreover, spasticity in the knee (both in quadriceps and
hamstrings) and hip muscles (adductors) (Ross and Engsberg 2007) has also been
related with gait velocity and stride length decreasing (Damiano et al. 2006). This
finding corroborates the clinical importance of spasticity in general functioning, as
gait velocity is a major determinant of patients’ integration in community, i.e.,
someone that walks faster can easily and safety cross a street and therefore would
not avoid community environments (Middleton et al. 2015).
Spasticity and co-contraction have been critically assessed as being the same
phenomenon (Diane et al. 2000). Despite the co-contraction is most of the times
related with spasticity and therefore simultaneously present in CNS disorders, these
are not exactly the same phenomenon. In fact, co-contraction becomes excessive, and
the agonist force decreases because spasticity transforms, at long term, the peripheral
muscle components (configuration and properties of muscle fibers) and, conse-
quently, generates abnormalities in dynamic muscle activation patterns. The applica-
tion of EMG has been tested as a novel approach to quantitative study of muscle
activity during dynamic motor activities. For example, during gait, the CP children
stretching of calf muscles (agonist) in the post-contact stride period is known to be
frequently accompanied by synchronous bursts in triceps surae (antagonist), gener-
ating abnormal muscle co-contraction patterns. This abnormal muscle patterns tend to
be congruent with mechanical changes in ankle joint (e.g., absence of dorsiflexion).
Even though EMG is able to detect these abnormal muscle patterns, it cannot reliably
separate two possible causes for this phenomenon: the augmented myotatic reflex
effect or the peripheral component contributions (muscle rigidity) (Crenna 1998).
Instrumented gait analysis (IGA) has the unique ability to measure and provide
information critical to the distinction between complex gait patterns. By assessing
the tridimensional position of the different segments, including pelvis, at any
moment of the gait cycle, it facilitates pattern recognition and problem addressing
(Miller 2004). The integration of video analysis with dynamic electromyography
(D-EMG) allows the identification of the contribution of muscles in each phase of
the gait cycle, enhancing understanding of the pathological gait patterns (Miller
2004). Also, it makes it possible to assess the tridimensional gait results of the
different proposed interventions (Miller 2004). By also incorporating clinical mea-
sures, IGA is the gold standard in gait evaluation in the CP population, with none
other video analysis protocol reaching its consistency (Rathinam et al. 2014). There-
fore, the simultaneous analysis of IGA and D-EMG during gait does not provide a
direct information on the spasticity effect in locomotion patterns in children with
cerebral palsy but can reliably assess consequences of spasticity in segment postures
and spatiotemporal parameters.
As a conclusion, the original definition of spasticity (Lance and McLeod 1981)
“does not include impaired movement and an abnormal posture” and only consid-
ered observations under static conditions (muscles relaxed; muscles tonically acti-
vated), which explains the poor report of correlations between the pathological
responsiveness to stretch measured at rest and motor deficits assessed during natural
actions, such as gait (Crenna 1998). Therefore, the lack of adequate instruments to
assess spasticity during gait has been limiting the consensus about the effect of
spasticity in gait pattern.
Spasticity Treatments with Impact in CP Gait
For some years now, a spinal cord mechanism that controls the excitability of the
stretch reflex has been described, which does not depend on supraspinal inhibition.
The so-called postactivation depression is an intrinsic property of the Ia afferent
fibers that is associated with a decreased release of neurotransmitters and has been
shown to be reduced with limb immobilization in a shortened position. As such,
muscle atrophy and reduced activity seen in CP also results in spasticity. More
importantly this mechanism seems to be affected by active movement, passive
mobilization, and prolonged muscle stretching, all of which can prevent muscle
hypertonia (Trompetto et al. 2014). This could be one mechanism subjacent to
improvements seen with both splits/orthotics and physiotherapy movement
modalities.
Bracing and orthotics are classic treatment options for lower limb muscle and
tendon lengthening in CP. Orthotics applied in the ankle, holding the foot in a
plantigrade position and correcting planovalgus foot deformities, may provide the
stable fundamental base that allow CP children to develop balance and train task-
specific activities in stance but also give a movement focus in the knee and hip. This
favors a more symmetric and “normal” gait pattern. However, by restraining active
plantar flexion, ankle-foot orthotics (AFO) reduce the lift-off moment of gait,
producing secondary adaptations, such as the hip extensors being the main respon-
sible for the push-off phase and forward movement, which is frequently combined
with pelvis rotation. A flexible leaf-spring AFO may allow some degree of plantar
flexion but only maintains the foot in dorsiflexion if there is absent spasticity of the
gastrocnemius and soleus.
Baclofen is a gamma-aminobutyric acid (GABA) agonist and thus mimics an
inhibitory neurotransmitter, reducing spasticity. It has been used both orally and,
more recently, intrathecally. The latter reduces some side effects such as excessive
sedation and also increases the bioavailability of the drug in its action sites: the
cerebrospinal fluid and brain. Its evidence is insufficient in regard to decreasing
spasticity, improving gross motor actions, and transferring achievements to activities
and participation (Martin et al. 2010). Despite the lack of evidence, baclofen is a
treatment option for generalized spasticity, such as the one presented in
tetraparetic CP.
Other alternatives to the management of generalized spasticity are pharmacolog-
ical, including diazepam and tizanidine with moderate evidence and dantrolene with
insufficient evidence. However, the pharmacological options have not demonstrated
changes in gross motor function or quality of life (Martin et al. 2010).
Botulinum toxin A (BTA) is a neurotoxin that inhibits acetylcholine release into
the neuromuscular junction, reducing the muscle response to efferent neural stimuli.
As such, the tonic and phasic stretch reflexes and consequently spasticity are
diminished in the selected injected muscles or muscular groups. However, the
duration of its effect is estimated in 3–6 months and the action potential of the
muscle and active movements are reduced, which means that BTA injections by
themselves are not a treatment for spasticity-derived gait alterations but can provide
the means to facilitate a more biomechanical correct movement and posture during
gait (Klemens Fheodoroff et al. 2016). This facilitation is achieved by providing the
correct afferent exteroceptive and proprioceptive stimulus in an intensive physio-
therapy program or the use of orthotics. Its direct impact is therefore in the structural
and functional dimensions of the International Classification of Functioning, Dis-
ability and Health (ICF), with only a latter translation to activities and participation,
due to the time required to motor relearning of the treated limb to occur (Love
et al. 2010). Precocious use of BTA in young children with CP may reduce the
evolution to a crouch gait and the need for surgical corrective procedures but also
increase the probability of the child to achieve the motor milestones crucial to a
normal neurodevelopment (Klemens Fheodoroff et al. 2016). Furthermore it is
considered a safe procedure in CP children with localized spasticity (Delgado
et al. 2010).
An international consensus has stated in 2010 that BTA is effective in the
management of spastic equines to improve gait (level A), however, only probably
effective to improve goal attainment and function in the management of spastic
equinus (level B). Also, adductor injections may help attain some goals (level B) and
may delay hip displacement (level A) but does not improve gross motor function and
does not affect long-term outcome (level A). Multiple lower limb injections have
conflicting information as respect to gait, goal attainment, and function (level U)
(Love et al. 2010).
Another treatment option for spasticity-related gait disorders in PC is selective
dorsal rhizotomy (SDR). By selectively impairing afferent stimuli to the medullae,
this procedure inhibits the efferent response of the stretch reflex, resulting in
decreased tone and therefore reducing spasticity. Since the clinical effects are
somewhat similar to BTA or baclofen, it is expected to act merely in the structural
and functional dimensions of ICF, with the considerations made to BTA regarding
transfer to activities and participation, being also valid. However, the benefits
observed in the sagittal plane at all lower limb joints with SDR are associated with
an augmented misalignment in the pelvis. These postural alterations include
crouching and horizontalization of the sacrum (Roberts et al. 2015), which will
progress to the adult age due to the irreversibility of the technique. Furthermore, the
procedure may reveal underlying muscle weakness that worsens the performance of
previous ambulatory children, especially if the antigravity postural capacity is
affected. Thus, a multidimensional evaluation and biomechanical analysis of gait
should be performed to guide patients’ election to this procedure (Grunt et al. 2014).
Indeed, spasticity of the extensor apparatus of the lower limb may be a useful
characteristic of CP gait, as it maintains the possibility of placing the foot (even if in
equinus) on the ground and thus relieve the body mass loading in the contralateral
limb. This is a fundamental characteristic of a normal gait. When spasticity is
removed, for instance, by the abovementioned dorsal rhizotomy, this extension
maintenance is impaired, resulting in incapacity of body mass transfer to the affected
limb. The correct consideration of the contribution of spasticity to CP gait is crucial
to the success or failure of a spasticity management procedure, when it relates to
activity, participation, and transfer to the child environment (Grunt et al. 2014).
Despite the absent role in the management of tone abnormalities of physiother-

apy, it is a fundamental adjunctive therapy to increase muscle strength and aerobic
conditioning but also in improving lower extremity coordination and speed (Martin
et al. 2010). Excessive exercise and strength training was said to aggravate spasticity
in CP and therefore a cause to gait deterioration; however, studies have challenged
this traditional belief (DL 2006). As such intensive task-oriented practice is nowa-
days a central practice in CP rehabilitation, which in the case of gait precludes
intensive treadmill training with or without assistive weight supports, despite a lack
of demonstrated evidence (Martin et al. 2010).
Other physiotherapy modalities that address spasticity and gait have long been
used in clinical practice. Neurodevelopmental therapy (NDT) and functional training
(FT) evaluate the individual as a whole, directing its practice according to the
identified incapacities, limitations, or handicaps (Martin et al. 2010). By focusing
in the particular characteristics of the different presentations of gait in CP instead of
intervening in a cluster of typical patterns, these modalities have struggled to produce
conclusive evidence in investigational studies designed as randomized control trials
(RCT) but have nevertheless a role in neurorehabilitation of the consequences of
spasticity in CP gait, either individually or as multimodal approaches.
Virtual reality (VR) is an emerging computer-assisted multisensorial feedback
approach to neurorehabilitation that promotes neuroplasticity induced by repetition
in an enjoyable environment but depends on the capacity of the individual to
integrate the “scene.” By conferring the possibility of manipulation of the environ-
ment, VR facilitates motor learning which in the case of the present chapter is
directed to improvements in the spatiotemporal features of CP gait (E. Monge
Pereiraa et al. 2014). Transcranial magnetic stimulation (TMS) has also recently
been therapeutically used, and it is characterized by noninvasive stimulation of the
motor cortex, promoting supraspinal inhibition of spasticity and therefore reducing
CP spasticity-induced gait changes (Gunduz et al. 2014).
When not efficiently managed, spasticity can progress to the complications
previously mentioned, and orthopedic surgery has a pivotal role in addressing
those consequences. As deformity evolves from dynamic to fixed, the focal tone
management modalities already referred become contraindicated, as bone deformity
or soft tissue contractures are not tone dependent and therefore susceptible.
A wide range of orthopedic interventions are available in CP management, some
of which address soft tissues such as tendon lengthening and ligament or capsular
releases, while others intervene in the skeletal system with osteotomies and arthrod-
eses as the mainstay of the possibilities. After a biomechanical evaluation, there is
also the possibility of combining surgical procedures to different anatomical loca-
tions, in order to optimize gait, being referred to as single-event multilevel surgery
(SEMS) (Thomason et al. 2013).
We previously described the pattern of flexed knee gait, characterized by the
incapacity to perform full hip and knee extension in late stance. The surgical
lengthening of the biarticular gastrocnemius, hamstrings, and iliopsoas is one com-
ponent of surgical procedures targeting improved stance and gait in spastic diplegia.
Indeed, a surgical procedure that lengthens the gastrocnemius may overcome the hip
secondary compensation abovementioned, as it allows some active plantar flexion

and also reduces the contracture in midstance, resulting in global gait improvements.
Also hamstring lengthening is effective in correcting knee extension; however, since
the hamstrings contribute to hip extension in stance, the procedure affects to some
degree pelvic control. Other surgical options have been developed, such as the
supracondylar extension osteotomy to correct the flexed knee deformity combined
with patellar tendon advancement, which is described to improve knee extensor lag,
knee extension in gait, and overall gait. Due to increased stability in the knee, these
procedures are also associated with reduced spasticity in the knee extensors (Sossai
et al. 2010). Surgical selection should thus be decided based in gait analysis of the
CP individual potentiating the after gains.
From the treatments referred above, orthopedic surgery (including tendon trans-
fer, muscle lengthening, or others), orthotics, and spasticity control (including
rhizotomy, BTA, phenol blocks, or others) have succeeded in improving gait veloc-
ity, decreasing cadence, and increasing stride length (Paul et al. 2007). Other
treatments such as serial casting, muscle strengthening exercises, and biofeedback
have not accumulated the necessary evidence to affirm their effects in improving CP
spatiotemporal components of gait. A multifactorial approach to gait alterations
should thus be considered in detrimental of more isolated approaches.
Future Directions
There is a scarce of consolidated information on the evolution of the gait pattern

throughout adolescent and adulthood, specially from CP individuals who are subjected
to orthopedic and invasive tone management procedures (Wilson et al. 2014). This is an
issue that must be address in future studies, as the best medical, developmental, and
rehabilitative science for optimizing mobility, motor functioning, and fitness for adults
with CP are unknown (Frisch and Msall 2013). Also, it is not clear for now what the
activity limitations that impair adults with CP from participating in their community are
and also which environmental factors and in what measure they affect CP adults’
handicaps, despite some works investigating this theme (Frisch and Msall 2013).
Given the difficulty to distinguish between spasticity and intrinsic hypertonia
with the use of current evaluation measures, question remains if the actual treatments
for lower limb hypertonia really address spasticity or are actually just influencing the
biomechanical factors responsible for the intrinsic hypertonia as a result of immo-
bilization. This is an important issue when deciding for intervention use, orthopedic
or tone management. In other words where and when are the limits to tone manage-
ment potential? When does rigidity become a primary issue, contraindicating spas-
ticity control? The use of biomechanical measures with EMG recordings and
perhaps accelerometers should be included in future clinical trials of CP gait
evaluation and spasticity management, in order to differentiate between the focus
of the treatment and provide a more comprehensive knowledge of the individual
alterations (Chang et al. 2010), and on where in the gait cycle the velocity threshold
to elicit the tonic or phasic stretch reflex occurs.
References
Abel MF, Damiano DL, Blanco JS, Conaway M, Miller F, Dabney K, Sutherland D, Chambers H,
Dias L, Sarwark J, Killian J, Doyle S, Root L, LaPlaza J, Widmann R, Snyder B (2003)
Relationships among musculoskeletal impairments and functional health status in ambulatory
Awaad Y, Rizk T (2012) Spasticity in children. J Taibah Univ Med Sci 7(2):53–60
Bar-On L, Molenaers G, Aertbelien E, Van Campenhout A, Feys H, Nuttin B, Desloovere K (2015)
Spasticity and its contribution to hypertonia in cerebral palsy. BioMed Res Int 2015:1–10
Bax M, Goldstein M, Rosenbaum P, Leviton A, Paneth N, Dan B, Jacobsson B, Damiano D (2005)
Proposed definition and classification of cerebral palsy. Dev Med Child Neurol 47:571–576
Boyd R, Graham HK (1999) Objective measurement of clinical findings in the use of botulinum
toxin type A for the management of children with CP. Eur J Neurol 6:23–35
Chang FM, Rhodes JT, Flynn KM, Carollo JJ (2010) The role of gait analysis in treating gait
abnormalities in cerebral palsy. Orthop Clin N Am 41(4):489–506
Crenna P (1998) Spasticity and ‘Spastic’ gait in children with cerebral palsy. Neurosci Biobehav
Rev 22(4):571–578
Damiano DL (2006) Activity, activity, activity: rethinking our physical therapy approach to cerebral
palsy. Phys Ther 86(11):1534–1540
Damiano DL, Quinlivan JM, Owen BF, Payne P, Nelson KC, Abel MF (2002) What does the
Ashworth scale really measure and are instrumented measures more valid and precise? Dev Med
Child Neurol 44(2):112–118
Damiano DL, Laws E, Carmines DV, Abel MF (2006) Relationship of spasticity to knee angular
velocity and motion during gait in cerebral palsy. Gait Posture 23(1):1–8
Dana R, Cub DA (2013) Cerebral palsy gait, clinical importance. MAEDICA – J Clin Med 8
(4):388–393
Delgado MR, Hirtz D, Aisen M, Ashwal S, Fehlings DL, McLaughlin J, Morrison LA, Shrader
MW, Tilton A, Vargus-Adams J (2010) Practice parameter: pharmacologic treatment of spas-
ticity in children and adolescents with cerebral palsy (an evidence-based review): report of the
quality standards subcommittee of the American academy of neurology and the practice
committee of the child neurology society. Neurology 74(4):336–343
Diane L, Martellotta L, Sullivan J, Granata P, et al. (2000) Muscle force production and functional
performance in spastic cerebral palsy: Relationship of cocontraction. Archives of Physical
Medicine and Rehabilitation 81(7):895–900
Domagalska M, Szopa A, Syczewska M, Pietraszek S, Kidoń Z, Onik G (2013) The relationship
between clinical measurements and gait analysis data in children with cerebral palsy. Gait
Posture 38(4):1038–1043
Fheodoroff K, Jacinto J, Geurts A, Molteni F, Franco JH, Santiago T, Rosale R, Gracies J-M (2016)
How can we improve current practice in spastic paresis? Eur Neurol Rev 11(2) (Epub ahead of
print)
Filloux FM (1996) Neuropathophysiology of movement disorders in cerebral palsy. J Child Neurol
11(Suppl 1):S5–S12
Frisch D, Msall ME (2013) Health, functioning, and participation of adolescents and adults with
cerebral palsy: a review of outcomes research. Dev Disabil Res Rev 18(1):84–94
Gage JR (2004) The treatment of gait problems in cerebral palsy. Mac Keith, London
Gracies J-M, Burke K, Clegg NJ, Browne R, Rushing C, Fehlings D, Matthews D, Tilton A,
Delgado MR (2005) Reliability of the Tardieu Scale for assessing spasticity in children with
cerebral palsy. Arch Phys Med Rehabil 91(3):421–428
Grunt S, Fieggen AG, Vermeulen RJ, Becher JG, Langerak NG (2014) Selection criteria for
selective dorsal rhizotomy in children with spastic cerebral palsy: a systematic review of the
literature. Dev Med Child Neurol 56(4):302–312
Gunduz A, Kumru H, Pascual-Leone A (2014) Outcomes in spasticity after repetitive transcranial
magnetic and transcranial direct current stimulations. Neural Regen Res 9(7):712–718
Hobart JC, Riazi A, Thompson AJ (1994) Spasticity: a review. Neurology 44(11):S12–S20

Kembhavi G, Johanna D, Kelsey P, Devon P (2011) Adults with a diagnosis of cerebral palsy: a
mapping review of long-term - Review Development medicine and child neurology 53
(7):610–614
Lance JW (1980) Symposium synopsis. In: Feldman RG, Young RR, Koella WP (eds) Spasticity:
disordered motor control. Chicago, Yearbook Medical Publishers
Lance JW, McLeod JG (1981) A physiological approach to clinical neurology. Butterworths,
London
Love SC, Novak I, Kentish M, Desloovere K, Heinen F, Molenaers G, O’Flaherty S, Graham HK
(2010) Botulinum toxin assessment, intervention and after-care for lower limb spasticity in
children with cerebral palsy: international consensus statement. Eur J Neurol 17:9–37
Martin L, Baker R, Harvey A (2010) A systematic review of common physiotherapy interventions
in school-aged children with cerebral palsy. Phys Occup Ther Pediatr 30(4):294–312
Middleton A, Fritz SL, Lusardi M (2015) Walking speed: the functional vital sign. J Aging Phys Act
23(2):314–322
Miller F (2004) Cerebral palsy. Springer, New York
Monge Pereiraa E, Rueda FM, Diego IMA, de la Cuerda RC, de Mauroc A, Miangolarra JC (2014)
Empleo de sistemas de realidad virtual como método de propiocepción en parálisis cerebral:
guía de práctica clínica. Neurologia 29(9):550–559
de Morais Filho MC, Kawamura CM, Lopes JAF, Neves DL, Cardoso M d O, Caiafa JB (2014)
Most frequent gait patterns in diplegic spastic cerebral palsy. Acta Ortop Bras 22:197–201
Novacheck TF, Stout JL, Tervo R (2000) Reliability and validity of the Gillette Functional
Assessment Questionnaire as an outcome measure in children with walking disabilities. J
Pediatr Orthop 20(1):75–81
O’Byrne JM, Jenkinson A, O’Brien TM (1998) Quantitative analysis and classification of gait
patterns in cerebral palsy using a three-dimensional motion analyzer. J Child Neurol 13
(3):101–108
Pandyan AD, Johnson GR, Price CIM, Curless RH, Barnes MP, Rodgers H (1999) A review of the
properties and limitations of the Ashworth and modified Ashworth Scales as measures of
spasticity. Clin Rehabil 13(5):373–383
Patrick E, Ada L (2006) The Tardieu Scale differentiates contracture from spasticity whereas the
Ashworth Scale is confounded by it. Clin Rehabil 20(2):173–182
Paul SM, Siegel KL, Malley J, Jaeger RJ (2007) Evaluating interventions to improve gait in cerebral
palsy: a meta-analysis of spatiotemporal measures. Dev Med Child Neurol 49(7):542–549
Piccinini L, Cimolin V, Galli M, Berti M, Crivellini M, Turconi AC (2007) Quantification of energy
expenditure during gait in children affected by cerebral palsy. Eura Medicophys 43(1):7–12
Rathinam C, Bateman A, Peirson J, Skinner J (2014) Observational gait assessment tools in
paediatrics – a systematic review. Gait Posture 40(2):279–285
Roberts A, Stewart C, Freeman R (2015) Gait analysis to guide a selective dorsal rhizotomy
program. Gait Posture 42(1):16–22
Rodda J, Graham HK (2001) Classification of gait patterns in spastic hemiplegia and spastic
diplegia: a basis for a management algorithm. Eur J Neurol 8:98–108
Rodda JM, Graham HK, Carson L, Galea MP, Wolfe R (2004) Sagittal gait patterns in spastic
diplegia. J Bone Joint Surg Br 86(2):251–258
Ross SA, Engsberg JR (2007) Relationships between spasticity, strength, gait, and the GMFM-66 in
persons with spastic diplegia cerebral palsy. Arch Phys Med Rehab 88(9): 1114–1120
Scholtes VAB, Becher JG, Beelen A, Lankhorst GJ (2006) Clinical assessment of spasticity in
children with cerebral palsy: a critical review of available instruments. Dev Med Child Neurol
48(1):64–73
Sossai R, Vavken P, Brunner R, Camathias C, Graham HK, Rutz E (2010) Patellar tendon
shortening for flexed knee gait in spastic diplegia. Gait Posture 41(2):658–665
Thomason PJ, Selber P, Graham HK (2013) Single event multilevel surgery in children with spastic
cerebral palsy: 5 year prospective cohort study. Gait and Posture 37(1):23–28
Trompetto C, Marinelli L, Mori L, Pelosin E, Currà A, Molfetta L, Abbruzzese G (2014) Patho-

physiology of spasticity: implications for neurorehabilitation. BioMed Res Int 2014:1–8
Wilson NC, Chong J, Mackey AH, Stott NS (2014) Reported outcomes of lower limb orthopaedic
surgery in children and adolescents with cerebral palsy: a mapping review. Dev Med Child
Neurol 56(9):808–814
Wren TA, Do KP, Hara R, Dorey FJ, Kay RM, Otsuka NY (2007) Gillette gait Index as a gait
analysis summary measure: comparison with qualitative visual assessments of overall gait. J
The Arm Pendulum in Gait
Jaques Riad
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Arm Pendulum and Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Mechanical Effects of Arm Movement on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
How Information on Arm Pendulum Can Help in Medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Psychological and Social Aspects of Deviation in Arm Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
Because of the physical length of the arms and the relatively large range of motion
in the shoulder and elbow in gait, any deviation from normal is detected imme-
diately and attracts attention. The deviation can consist of increased flexion in the
elbow, decreased range of motion, movement out of phase with the lower
extremities, and asymmetry between the movement of right and left arms, either
in isolation or in many more or less noticeable combinations. Although human
evolution means that we no longer walk on our arms, arm movement has impact
on our stability, balance, and appearance while walking. In addition, we can carry
things, make gestures, or do other things with the arms and hands while walking.
Despite an evolution toward corticospinal control of arm and hand movements,
quadrupedal limb coordination persists during locomotion. We do not think about
how we coordinate our arms and legs when walking. It just happens.
Individuals with deformity, limited range of motion, or movement disorders
affecting the arm show a disturbance of the normal arm pendulum in gait. It can
J. Riad (*)
Skaraborg Hospital Skövde, Skövde, Sweden
e-mail: jacques.riad@vgregion.se

DOI 10.1007/978-3-319-30808-1_56-1
2 J. Riad
be difficult to understand the consequences of the primary pathology and the

influence on the movement pattern as well as the possible development of
compensation mechanisms. Studying the arm pendulum is important for diagno-
sis and treatment and to follow progression over time. In addition, our sensitivity
to deviations from normal highlights the importance of arm movement for
communication and appearance.
Keywords
Arm movement • Gait • Motion analysis • Upper extremity • Cerebral palsy •
Symmetry
Introduction
Smooth and symmetrical arm movements in gait are expected and part of the general
movement pattern in human locomotion. The movement in gait sends signals about
the walker’s personality, health, and functional status. Symmetry of movement in
particular is important for appearance and well-being. Deviation from normal draws
attention to abnormalities and implies poor and inadequate function. Asymmetry of
movements in gait is immediately apparent to the beholder and can be or become a
physiological and psychological problem for the walker.
Arm movements in gait, as well as movements of the lower extremity, may be
difficult to define and understand. Is the movement pattern primarily an expression
of functioning at the lowest possible energy cost? Or are there secondary changes as
a consequence of impaired control, injury, deformity, or any other cause? Even
tertiary changes might develop owing to previous treatment or progressive deformity
with continued growth in children. Obviously, it is essential to separate and identify
these differences to make it possible to plan follow-up and treatment. It goes without
saying that it is not ideal to treat the secondary or compensatory changes if the
primary cause itself can be addressed. This is especially true when the treatment is
irreversible, such as surgical interventions.
Arm and hand function are essential for independence, something that is in focus
in the modern industrial society and crucial for survival of the family in the
developing world. In addition, upper extremity movement and function reflect
personality and – along with facial expression – contribute strongly in human
communication and appearance (Chouchourelou et al. 2006; Ikeda and Watanabe
2009; Meeren et al. 2005; Montemare 1987; Schneider et al. 2014). However, arm
movements during gait also play a role in locomotion, contributing to stability of the
trunk and keeping energy costs low. Therefore, arm pendulum in gait is of
importance.
Braune and Fischer included arm movement in their description “The gait of
humans” from 1895, but not much attention has been directed specifically toward
arm pendulum (Braune 1895). The purpose and mechanism of arm pendulum are
The Arm Pendulum in Gait 3
still under discussion, although modern techniques have given us a better under-
standing and to some extent allowed us to appreciate their medical value.
State of the Art
Different models for analysis of the three-dimensional (3D) motion of the upper
extremity have been developed, although overall the system is largely consistent
with lower extremity gait analysis (Kadaba et al. 1990). A 3D analysis provides more
detailed information than a two-dimensional assessment (video). Nevertheless,
video capturing provides a lot of information on the dynamics and appearance of
the movement of a moving subject. Video can be especially useful to provide
information about transitions such as gait initiation and accelerating to walking or
running at a higher speed. Arm posturing with increased elbow flexion and
decreased range of motion in the sagittal plane (flexion/extension) is a typical
observation in video captures when a child with hemiplegic cerebral palsy
(CP) walks or starts running (Fig. 1). Hemiplegic CP is defined as increased muscle
tone (spasticity) in the arm and the leg on either the right or the left side and is caused
by a brain injury before the age of 2 years. However, the data from a video are
difficult to quantify, and the exact degree of flexion and the range of motion can only
be measured with a 3D analysis. In addition, the video assessment naturally does not
include the rotational profile (i.e., the third dimension).
Previously, only the lower extremity – what Perry called the locomotor system –
was segmented for calculations (two feet, two shanks, two thighs, and the pelvis)
(Perry 1992). The head, arms, and trunk were treated as a single unit, the passenger
segment. With higher computer capacity, most human motion analysis laboratories
now segment also this passenger segment into the head, the trunk, two upper arms,
two forearms, and two hands. For recording purposes, the segments are defined by
several retroflective markers, which are attached to the skin at specific anatomical
landmarks. Several cameras identify the markers when the person walks in the
laboratory, and movements relative to other segments can be calculated in three
dimensions. If the subject walks on the force plates situated in the floor, information
on the forces acting on the body (ground reaction forces) can be obtained. From the
movement and the force data, the moments over separate joints can be calculated.
Hence, movements (kinematics) and forces (kinetics) are collected, and the
information is presented in graphs with the gait cycles for the right and the left
side and usually the mean. The gait cycle consists of stance phase (60% of the gait
cycle) and swing phase (40%) (Fig. 2). From these data we can obtain exact degrees
of movement at any point during the gait cycle. In addition, temporal-spatial data,
information on gait speed and step length, stride length, support time, step width,
etc., is captured.
To obtain a measure of degree of deviation from normal in the upper extremity in
gait, the Arm Posture Score (Riad et al. 2011) and the Arm Profile Score (Jaspers
et al. 2011) were developed, analogous to the Gait Profile Score (Baker et al. 2009).
4 J. Riad
Fig. 1. Arm posture. In

hemiplegic cerebral palsy, the
typical one-sided spastic
position is with elbow flexion,
forearm pronation, and palmar
flexion and with little range of
motion when walking
Fig. 2. Elbow motion (y-axis) during the gait cycle (x-axis) with the first 60% being the stance
phase and the last 40% the swing phase. The gray band in the background is the normal/reference
group here; the left (red) and right (blue) sides are plotted with all the gait trials obtained from each
side. To the right the means from both sides are illustrated. It is clear that the left side (red) has
increased flexion and decreased range of motion
The deviation is calculated as the root mean square during the entire gait cycle
compared to the laboratory’s age- and gender-controlled references (Fig. 3). This
score can be useful as a comprehensive assessment of the degree of deviation and to
calculate symmetry between both upper and lower extremities. However, the degree
of deviation does not tell us what sorts of changes are occurring, merely the degree of
deviation. For example, it does not reveal whether the movement is increased or
decreased or if the range of motion is increased or decreased relative to normal.
Control
Patient Non-involved side
RMSD Involved side
Flexion 60.0 60.0

40.0 40.0
GPS Knee
20.0 20.0
0.0 0.0
Extension
60.0 60.0
Flexion
40.0 40.0
APS Elbow
20.0 20.0
Extension 0.0 0.0
Fig. 3. The mean of the sagittal movements of knee and elbow on the involved (hemiplegic) side
and the noninvolved side in a child with hemiplegic cerebral palsy. The area between the mean
represents the root mean square and is the measurement of deviation
It is especially in children with cerebral palsy that modern gait analysis has been
evolved as a medical assessment tool. Bonnefoy-Mazure and co-workers, in addi-
tion, reported comprehensively on upper limb patterns during gait in individuals
with CP (Bonnefoy-Mazure et al. 2014).
In this chapter we discuss only free arm movements unless stated otherwise. That
is, we assume an independent gait without assistive devices other than possible
bracing and orthosis of the foot and ankle.
Arm Pendulum and Gait
In this chapter, arm pendulum in gait, we will discuss the possible purpose of arm
pendulum and the mechanical effects on gait. We will discuss how arm pendulum
can be useful in medicine to make diagnoses, how to follow development over time,
and how arm pendulum assessment can help in treatment. The medical part will be
focused on individuals with neurological impairments, with special emphasis on
cerebral palsy, past stroke, and Parkinson’s disease. We will also discuss psycholog-
ical and social aspects of deviations from normal arm movement in gait.
Mechanical Effects of Arm Movement on Gait
Arm pendulum has been reported to serve the purposes of decreasing energy
consumption, increasing gait stability, and improving balance (Kuhtz-Buschbeck
and Jing 2012; Meyns et al. 2013; Ortega et al. 2008). Some state that arm pendulum
6 J. Riad
is “a relic from quadrupedal walking” without purpose (Jackson et al. 1983; Murray
et al. 1967). There is also a debate about the mechanism of arm pendulum.
Muscle activity of the upper limb in gait has been assessed by electromyography
(EMG) (Ballesteros et al. 1965; Kuhtz-Buschbeck and Jing 2012) under different
conditions: free arms, arms held by the side, bound to the body, and in gait at
different speeds. The purpose was to determine whether arm pendulum is active or
passive. Muscle activity was noted in normal gait but also in some of the other
conditions, and Kuhtz-Buschbeck and Jing concluded that normal arm swing most
likely has both active and passive components. In a simulation study with the aim to
quantify the components, a significant decrease of arm swing amplitude was found
when there was no EMG activity (Goudriaan et al. 2014). The same authors could
also show alteration in the pattern (interlimb coordination) when muscle activity was
absent.
It is not fully clear, but it appears that arm pendulum contributes toward lowering
energy consumption, even though there is an energy cost for the swinging (Collins
et al. 2009; Elftman 1939). The mechanism is the decrease of vertical ground
reaction moment, and the net effect is a reduction of energy consumption by 8%
compared to walking with arms constrained (Umberger 2008).
Although there are some contradictory results regarding arm swing and lateral-
medial stability (Collins et al. 2009), it seems that arm swing contributes to stability
(Ortega and Farley 2015; Ortega et al. 2008). Obviously, the small child, the toddler,
has the high guard position for stability and balance, and the elderly individual
achieves increased trunk stability through increased arm swing (Nakakubo et al.
2014; Kubo and Ulrich 2006). Elderly also had longer recovery time after perturba-
tion to return to steady state gait, with normalized arm swing (Nakakubo 2014). In
running, step width is narrower than in walking, and arms swing to minimize energy
cost and improve lateral-medial balance (Arellano and Kram 2011).
In the neural control of arm pendulum in gait, it appears that a remnant from our
quadrupedal days is still in charge of the coordination between the forelimbs and
hind limbs, which is mediated by long neurons from the cervical and lumbar spine
(Dietz 2011). An uncoupling occurs when the human wants to perform a voluntary
arm and hand task, at which point direct control from the cortex takes over. Dietz
points out that in patients with Parkinson’s disease, the so-called quadrupedal limb
coordination is intact; therefore, verbal, visual, and other stimuli of arm swing can be
beneficial in the treatment. In patients who have had a stroke, however, the afferent
nerve signal does not lead to an appropriate response (Dietz 2011).
How Information on Arm Pendulum Can Help in Medicine
Diagnosis
Posturing with the arm – the “high guard” position – is common in toddlers for
stabilization (Kubo and Ulrich 2006) and may persist in children with cerebral palsy
(CP) (Meyns et al. 2012a). Typically in diplegic CP gait, there is high variability in
arm movement and increased arm abduction, as a sign of balance disturbance owing
to spasticity and decreased motor control in the lower extremity (Meyns et al. 2012a;
Romkes et al. 2007).
When attempting to differentiate between hemi- and diplegic cerebral palsy in a
young child, the arm posturing often reveals the hemiplegic involvement even in
mild cases of CP. In hemiplegic CP, the one-sided spastic position with elbow
flexion, forearm pronation, and palmar flexion and with little range of motion
(Riad et al. 2007; Romkes et al. 2007) is different from the bilateral elbow flexion
pattern in diplegic CP (Fig. 1).
It has been found that in children born at term with hemiplegic CP that the arm is
more involved than the leg, whereas in children born prematurely, the arm and
leg are more equally involved (Uvebrant 2000). Thus, the patient’s history of
birth and the clinical sign of arm posturing when walking (and especially when
running, which enhances the arm posturing) can contribute toward making a
diagnosis.
Spastic paraparesis and diplegic CP may be difficult to differentiate. According
to Bonnefoy-Mazure and co-workers (Bonnefoy-Mazure et al. 2013), 3D analysis of
the upper extremity can help in the diagnoses; they found differences by studying the
trunk and arm movement patterns that compensate for gait deviation in the lower
extremity. Apart from the fact that patients with spastic paraparesis were older (mean
16.7 years) than the diplegic CP group (mean 12.3 years) and walked slower (mean
1.13 m/s vs. mean 1.42 m/s in the diplegic CP group), no differences between the
groups were found: their lower extremity movement in the sagittal plane during gait
was indistinguishable in 3D analysis. However, the diplegic group showed increased
flexion and more variable arm movements in the arms, similar to the high guard
position described in toddlers. In those with spastic paraparesis, compensation for
the lower extremity deviation occurred in the trunk and pelvis, which was interpreted
as an attempt to stabilize the head. Hence, upper extremity 3D analysis can be
helpful in this clinical situation.
The gait of patients with Parkinson’s disease is characterized by low velocity,
short, shuffling steps and the highly typical reduced arm swing, beside the freezing
and difficulties initiating gait. Early diagnosis can be tricky, but is important, since
early treatment may have a positive impact on the progression of the disease (Fahn
et al. 2004). Reduced arm swing is an early sign of Parkinson’s disease (Lewek et al.
2010), and a study by Mirelman and co-workers reported that use of body-affixed
sensors to examine arm swing might provide a new prodromal marker for the disease
(Mirelman et al. 2016).
How One Limb Affects the Other Limbs

Interlimb coordination is studied to determine if the arms and legs are moving in
phase, i.e., if they are normally synchronized, in gait. Meyns and co-workers
reported that several children with hemiplegic CP had a 2:1 arm-to-leg ratio on the
involved side and a 1:1 ratio on the noninvolved side (1:1 ratio being normal). In
children with diplegic CP, also with altered interlimb coordination, increased walk-
ing speed improved the synchronization, and the authors suggest that exercise
increasing walking speed could be beneficial for the general gait pattern (Meyns
8 J. Riad
et al. 2012b). However, this has neither been tested nor validated, and of course the
question arises, what influences what?
In early Parkinson’s disease, newly diagnosed patients show changes in interlimb
coordination, and thus coordination analyses help in the diagnosis (Winogrodzka
et al. 2005). In addition, it has been suggested that improving interlimb coordination
could improve gait in Parkinson’s disease (Dietz 2011).
Another interesting study, investigating the effect of elbow contracture on gait
revealed significant decreases in gait speed with three different degrees of elbow
flexion contractures simulated with a brace in healthy participants (Trehan et al.
2015). In other words, the arm pendulum affects gait also in subjects with no
neurological impairment.
The arm posture in hemiplegic CP may have an impact on gait and vice versa, and
studying the possible influence of different limbs can be quite complex (Lundh et al.
2014). How does the gait respond to the increased elbow flexion and the decreased
range of elbow motion? And how does the noninvolved lower extremity/leg respond
or rather compensate for the hemiplegic leg? What does this do to the noninvolved
arm? The interactions and deviations in relation to each of the two upper and two
lower extremities in gait in hemiplegic CP were studied by Lundh (Lundh et al.
2014). By calculating the degree of deviation and symmetry between involved and
noninvolved side in 47 adolescents and young adults with hemiplegic CP, four
distinct different groups were identified: close to normal, deviations mainly in the
leg, deviations mainly in the arm, and deviation in both the leg and the arm. This
information can help in diagnoses, as previously described, but can also be useful in
long-term follow-up. It is important to identify and understand changes of movement
pattern caused by the development of deformity and/or new compensation mecha-
nisms, so as to differentiate between what to address and attempt to treat and what to
leave alone.
Natural Development and Treatment of Arm Posture in Children

and Young Adults with Cerebral Palsy
In children with hemiplegic CP, the arm posture of the involved side is mainly
dependent on spasticity and decreased motor control. Children with diplegic CP,
however, use their arms more for balance to compensate for lower limb spasticity
and impaired motor control. The natural development of arm posturing, elbow
flexion in gait, in patients with hemiplegic CP, is that it spontaneously resolves
with time (Riad et al. 2007). In a study with 175 individuals (mean age 9.2 years,
range 4 to 21 years), there was a significant decrease of elbow flexion with increased
age in gait on the hemiplegic side, but not on the noninvolved side. Elbow extension
normalized on both sides with increased age. The range of motion increased
significantly on the noninvolved side with age, but not on the hemiplegic side
(Fig. 4). It was also noted that the variability of elbow flexion in gait decreased
with increased age.
In the same study, the outcome of elbow flexor lengthening on the natural history
of elbow flexion and motion during gait was investigated. Among the 175 children
who had severe elbow flexion, some had surgery with elbow flexor lengthening and
Fig. 4. Elbow flexion in gait in 175 individuals with hemiplegic cerebral palsy, with no surgical
treatment. The dark area represents the hemiplegic side and reveals decreasing elbow flexion with
increasing age, which is the natural development
some did not. The outcome for the two groups was examined by comparison of two
gait analyses of the same patient over time. At a follow-up time of 3.3 years, the
outcome in the surgical and nonsurgical groups was similar, suggesting that elbow
flexor surgery had no effect on elbow motion during gait. The explanation for this
might be the indication and in particular the individuals and their treatment goals. In
addition, children with hemiplegic CP born at term has a higher degree of brain
abnormalities, which can have implications when considering surgical correction of
arm posture and hand and arm function, since these children would not be expected
to cooperate in the rehabilitation program as compliantly as those with normal
mental development.
Taken together, these observations highlight how important it is to inform the
child and concerned parents about the natural history of elbow flexion, and that one
should be careful with early surgical intervention. The indications for and optimal
timing of surgical intervention are not clear. Attention should be directed toward the
child’s possible concerns, and treatment should be offered if the spontaneous
correction is not satisfactory (Riad et al. 2007).
Treatment with botulinum toxin has been reported to benefit young, mainly high-
functioning adolescents with hemiplegic CP; the patients experienced a very positive
cosmetic effect, even though parents and caregivers could not detect any change in
movement pattern (Corry et al. 1997). The explanation could be decreased tone,
leading to a perception of better control. It is possible that having better control –
especially emotional in situations or in transitions of movement (gait initiation,
running, etc.) – was beneficial for these patients.
10 J. Riad
In Treatment of Parkinson’s Disease

Verbal instructions including “walking while deliberately swinging the arms” were
found to normalize gait pattern regarding gait speed and step length, in addition to
arm swing (Behrman et al. 1998). The authors stated that this cognitive strategy
could be superior to use of visual cues such as lines drawn in the floor and also
compared to hearing cues that are not accessible in daily life. Of course numerous
studies have evaluated medical/drug treatment for normalization of arm and gait
pattern, and many included symmetry assessments. Here, the 3D analysis can prove
to be a useful tool for early diagnosis, evaluation of treatment, and following the
progress of the disease.
In Treatment of Adults Poststroke

In rehabilitation after stroke, handheld training devices for arm function have been
reported to have good effect. Virtual reality games have increased patients’ motiva-
tion and noticeably facilitated training. After stroke, the arm movement in gait is
often abnormal/heavily affected, and this most likely affects exercise and training to
a great extent. A study using movable handrails in a treadmill setup demonstrated
changes in muscle activation in the lower limb, most likely due to changes in
postural stability that occurred when the patient performed arm movements (Ste-
phenson et al. 2010). Interestingly, during swing phase, muscle activity was signif-
icantly increased in the tibialis anterior muscle when walking with free/more normal
arm movements. This increased muscle activity can help clear the foot in swing
phase, which is otherwise a common problem after stroke (Stephenson et al. 2010).
Botulinum toxin in elbow flexors resulted in increased gait speed (Esquenazi et al.
2008), and botulinum toxin in the hand, the forearm, and the elbow flexor
brachioradialis improved the range of motion in the knee and ankle of “slow
striding” patients (Hirsch et al. 2005).
Psychological and Social Aspects of Deviation in Arm Pendulum
Our appearance – how we look and move – is of importance and contributes to our
behavior, to how we are perceived, and, consequently, to how we feel.
One method to study how humans perceive other humans’ movements is to use
point-light biological motion sequences. This involves attaching lights to major
joints and filming a body in motion. Although the resulting visual sequence shows
only the movement patterns of the joints, without showing limbs or body, viewers
are readily able to recognize the point lights as representing a human walking. It has
been found that the visual detection of motion occurs in the superior temporal sulcus,
an area of the brain that is closely connected with the emotional processing area, the
amygdala (Johansson 1973). There is evidence that gait characteristics, including
arm swing, differentiate emotions (Montemare 1987). There is a clear visual sensi-
tivity to “angry walkers” (Chouchourelou et al. 2006; Ikeda and Watanabe 2009). In
addition, arm movement in gait is important for the recognition of individuals, and
“exaggerated” movements of the arms facilitated the task of recognition (Hill and
Pollick 2000). In their paper “Show me how you walk and I tell you how you
feel. . .,” Schneider and co-workers suggested that full understanding of emotion
perception requires the investigation of dynamic representation and means of
expression other than the face (Schneider et al. 2014).
It has been reported that self-esteem and self-concept can be affected by physical
impairment even in highly functioning patients with unilateral cerebral palsy (Russo
et al. 2008). Arm posturing and gait deviations may be perceived as cosmetic and
social impediments when the individual enters adolescence and becomes more self-
conscious (Decety and Grezes 2006; Wake et al. 2003).
In a group of high-functioning adolescents and young adults with hemiplegic CP,
self-esteem was significantly lower than in a control group (Riad et al. 2013). In
addition, movement deviations in the upper extremity in gait, mainly elbow flexion,
correlate with both self-esteem and sense of coherence. The higher the deviation, the
lower the self-esteem and sense of coherence (Fig. 5). Interestingly, no such corre-
lations were seen with deviations in the lower extremity. The conclusion was that
movement pattern in the upper extremity should not be assessed solely from a
functional perspective. Possible concerns about appearance and influence on self-
esteem should also be considered.
Fig. 5. Self-esteem assessed by questionnaire “I think I am” plotted with the Gait Profile Score
(GPS) and the Arm Posture Score (APS) expressing degree of deviation from normal. Correlation
coefficient ( 0.397) with APS. The higher the deviation, the lower the self-esteem
12 J. Riad
Symmetry
In ancient times, both the Egyptians and the Greeks noted that facial symmetry was
important in human judgment of beauty; this has also proven to be true for body
symmetry (Brown et al. 2008) (Figs. 6 and 7). In one study, dances performed by
symmetrical men, as opposed to asymmetrical men, were rated as more attractive,
suggesting that dynamic movement can signal underlying quality independently of
static appearance (Brown et al. 2005). It is not well studied or clear what possible
impact asymmetrical gait pattern, including arm swing, could have on the individ-
ual’s own perceptions and those of other people.
Fig. 6. Facial symmetry
Fig. 7. Body symmetry

Future Directions
In the future, gait assessments should also include the upper extremity, head, and
trunk, since this contributes to a more comprehensive understanding of human
movement patterns in locomotion. Without doubt, inclusion of total body movement
is important for a better understanding, not only for medical purposes but also for a
wider understanding of human communication and behavior.
Cross-References

▶ Gait scores: Interpretations and Limitations
Analysis
▶ Motor Patterns Recognition in Parkinson’s Disease
▶ Surface EMG
▶ Upper Extremity Models for Clinical Movement Analysis
References
Arellano CJ, Kram R (2011) The effects of step width and arm swing on energetic cost and lateral
balance during running. J Biomech 44:1291–1295
Baker R, Mcginley JL, Schwartz MH, Beynon S, Rozumalski A, Graham HK, Tirosh O (2009) The
gait profile score and movement analysis profile. Gait Posture 30:265–269
Ballesteros ML, Buchthal F, Rosenfalck P (1965) The pattern of muscular activity during the arm
swing of natural walking. Acta Physiol Scand 63:296–310
Behrman AL, Teitelbaum P, Cauraugh JH (1998) Verbal instructional sets to normalise the temporal
and spatial gait variables in Parkinson’s disease. J Neurol Neurosurg Psychiatry 65:580–582
Bonnefoy-Mazure A, Turcot K, Kaelin A, De Coulon G, Armand S (2013) Full body gait analysis
may improve diagnostic discrimination between hereditary spastic paraplegia and spastic
diplegia: a preliminary study. Res Dev Disabil 34:495–504
Bonnefoy-Mazure A, Sagawa Y Jr, Lascombes P, De Coulon G, Armand S (2014) A descriptive
analysis of the upper limb patterns during gait in individuals with cerebral palsy. Res Dev
Disabil 35:2756–2765
14 J. Riad
Braune W (1895) Der gang des Mencschen I. Abh K Sachs Ges Wiss Math-Phys 21:153
Brown WM, Cronk L, Grochow K, Jacobson A, Liu CK, Popovic Z, Trivers R (2005) Dance
reveals symmetry especially in young men. Nature 438:1148–1150
Brown WM, Price ME, Kang J, Pound N, Zhao Y, Yu H (2008) Fluctuating asymmetry and
preferences for sex-typical bodily characteristics. Proc Natl Acad Sci U S A 105:12938–12943
Chouchourelou A, Matsuka T, Harber K, Shiffrar M (2006) The visual analysis of emotional
actions. Soc Neurosci 1:63–74
Collins SH, Adamczyk PG, Kuo AD (2009) Dynamic arm swinging in human walking. Proc Biol
Sci 276:3679–3688
Corry IS, Cosgrove AP, Walsh EG, Mcclean D, Graham HK (1997) Botulinum toxin A in the
hemiplegic upper limb: a double-blind trial. Dev Med Child Neurol 39:185–193
Decety J, Grezes J (2006) The power of simulation: imagining one’s own and other’s behavior.
Brain Res 1079:4–14
Dietz V (2011) Quadrupedal coordination of bipedal gait: implications for movement disorders. J
Neurol 258:1406–1412
Elftman H (1939) The function of arms in walking. Hum Biol 11:529–535
Esquenazi A, Mayer N, Garreta R (2008) Influence of botulinum toxin type A treatment of elbow
flexor spasticity on hemiparetic gait. Am J Phys Med Rehabil 87:305–310 quiz 311, 329
Fahn S, Oakes D, Shoulson I, Kieburtz K, Rudolph A, Lang A, Olanow CW, Tanner C, Marek K,
Parkinson Study Group (2004) Levodopa and the progression of Parkinson’s disease. N Engl J
Med 351:2498–2508
Goudriaan M, Jonkers I, Van Dieen JH, Bruijn SM (2014) Arm swing in human walking: what is
their drive? Gait Posture 40:321–326
Hill H, Pollick FE (2000) Exaggerating temporal differences enhances recognition of individuals
from point light displays. Psychol Sci 11:223–228
Hirsch MA, Westhoff B, Toole T, Haupenthal S, Krauspe R, Hefter H (2005) Association between
botulinum toxin injection into the arm and changes in gait in adults after stroke. Mov Disord
20:1014–1020
Ikeda H, Watanabe K (2009) Anger and happiness are linked differently to the explicit detection of
biological motion. Perception 38:1002–1011
Jackson KM, Joseph J, Wyard SJ (1983) The upper limbs during human walking. Part 2. Function.
Electromyogr Clin Neurophysiol 23:435–446
Jaspers E, Feys H, Bruyninckx H, Klingels K, Molenaers G, Desloovere K (2011) The Arm Profile
Score: a new summary index to assess upper limb movement pathology. Gait Posture
34:227–233
Johansson G (1973) Visual perception of biological motion and a model for its analysis. Percept
Psychophys 14:201–211
during level walking. J Orthop Res 8:383–392
Kubo M, Ulrich B (2006) A biomechanical analysis of the ‘high guard’ position of arms during
walking in toddlers. Infant Behav Dev 29:509–517
Kuhtz-Buschbeck JP, Jing B (2012) Activity of upper limb muscles during human walking. J
Lewek MD, Poole R, Johnson J, Halawa O, Huang X (2010) Arm swing magnitude and asymmetry
during gait in the early stages of Parkinson’s disease. Gait Posture 31:256–260
Lundh D, Coleman S, Riad J (2014) Movement deviation and asymmetry assessment with three
dimensional gait analysis of both upper- and lower extremity results in four different clinical
relevant subgroups in unilateral cerebral palsy. Clin Biomech (Bristol, Avon) 29:381–386
Meeren HK, Van Heijnsbergen CC, De Gelder B (2005) Rapid perceptual integration of facial
expression and emotional body language. Proc Natl Acad Sci U S A 102:16518–16523
Meyns P, Desloovere K, Van Gestel L, Massaad F, Smits-Engelsman B, Duysens J (2012a) Altered
arm posture in children with cerebral palsy is related to instability during walking. Eur J Paediatr
Neurol 16:528–535
Meyns P, Van Gestel L, Bruijn SM, Desloovere K, Swinnen SP, Duysens J (2012b) Is interlimb
coordination during walking preserved in children with cerebral palsy? Res Dev Disabil
33:1418–1428
Meyns P, Bruijn SM, Duysens J (2013) The how and why of arm swing during human walking. Gait
Posture 38:555–562
Mirelman A, Bernad-Elazari H, Thaler A, Giladi-Yacobi E, Gurevich T, Gana-Weisz M, Saunders-
Pullman R, Raymond D, Doan N, Bressman SB, Marder KS, Alcalay RN, Rao AK, Berg D,
Brockmann K, Aasly J, Waro BJ, Tolosa E, Vilas D, Pont-Sunyer C, Orr-Urtreger A, Hausdorff
JM, Giladi N (2016) Arm swing as a potential new prodromal marker of Parkinson’s disease.
Mov Disord 31(10):1527–1534
Montemare JM (1987) The identification of emotions from gait information. J Nonverbal Behavior
11:33–42
Murray MP, Sepic SB, Barnard EJ (1967) Patterns of sagittal rotation of the upper limbs in walking.
Phys Ther 47:272–284
Nakakubo S, Doi T, Sawa R, Misu D, Tsutsumimoto K, Ono R (2014) Does arm swing emphasized
deliberately increase the trunk stability during walking in the elderly adults? Gait Posture
40:516–520
Ortega JD, Farley CT (2015) Effects of aging on mechanical efficiency and muscle activation
during level and uphill walking. J Electromyogr Kinesiol 25:193–198
Ortega JD, Fehlman LA, Farley CT (2008) Effects of aging and arm swing on the metabolic cost of
stability in human walking. J Biomech 41:3303–3308
Perry J (1992) Gait analysis, normal and pathological function. Slack Incorporated, Thorofare
Riad J, Coleman S, Miller F (2007) Arm posturing during walking in children with spastic
hemiplegic cerebral palsy. J Pediatr Orthop 27:137–141
Riad J, Coleman S, Lundh D, Brostrom E (2011) Arm posture score and arm movement during
walking: a comprehensive assessment in spastic hemiplegic cerebral palsy. Gait Posture
33:48–53
Riad J, Brostrom E, Langius-Eklof A (2013) Do movement deviations influence self-esteem and
sense of coherence in mild unilateral cerebral palsy? J Pediatr Orthop 33:298–302
Romkes J, Peeters W, Oosterom AM, Molenaar S, Bakels I, Brunner R (2007) Evaluating upper
body movements during gait in healthy children and children with diplegic cerebral palsy. J
Pediatr Orthop B 16:175–180
Russo RN, Goodwin EJ, Miller MD, Haan EA, Connell TM, Crotty M (2008) Self-esteem, self-
concept, and quality of life in children with hemiplegic cerebral palsy. J Pediatr 153:473–477
Schneider S, Christensen A, Haussinger FB, Fallgatter AJ, Giese MA, Ehlis AC (2014) Show me
how you walk and I tell you how you feel – a functional near-infrared spectroscopy study on
emotion perception based on human gait. NeuroImage 85(Pt 1):380–390
Stephenson JL, De Serres SJ, Lamontagne A (2010) The effect of arm movements on the lower limb
during gait after a stroke. Gait Posture 31:109–115
Trehan SK, Wolff AL, Gibbons M, Hillstrom HJ, Daluiski A (2015) The effect of simulated elbow
contracture on temporal and distance gait parameters. Gait Posture 41:791–794
Umberger BR (2008) Effects of suppressing arm swing on kinematics, kinetics, and energetics of
human walking. J Biomech 41:2575–2580
Uvebrant G (2000) Congenital hemiplegia. Mac Keith Press, London
Wake M, Salmon L, Reddihough D (2003) Health status of Australian children with mild to severe
cerebral palsy: cross-sectional survey using the Child Health Questionnaire. Dev Med Child
Neurol 45:194–199
Winogrodzka A, Wagenaar RC, Booij J, Wolters EC (2005) Rigidity and bradykinesia reduce
interlimb coordination in Parkinsonian gait. Arch Phys Med Rehabil 86:183–189
Foot and Ankle Motion in Cerebral Palsy
Jon R. Davids and Sean A. Tabaie
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Normal Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Disruption of Foot, Ankle Function, and Its Effect on Gait in Cerebral Palsy . . . . . . . . . . . . . . . . . 9
Clinical Decision-Making: Management of Foot and Ankle Problems in CP . . . . . . . . . . . . . . . . . . 13
Clinical History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Physical Examination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Radiographic Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Quantitative Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Levels of Deformity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Treatment Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Surgical Treatment Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Surgical Treatment Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Summary/Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Abstract
There are three common ankle/foot segmental malalignment patterns seen in
children with cerebral palsy (CP): equinus, equinoplanovalgus, and
equinocavovarus. Each type of ankle/foot malalignment can be classified into
three levels based upon the presence of dynamic muscle imbalance, fixed or
myostatic deformity of the muscle-tendon unit, and fixed or skeletal deformities.
Management of foot and ankle deformities in children with CP can consist of both
nonsurgical (orthotics and pharmacotherapy) and surgical (muscle-tendon unit
lengthening or transfer, skeletal osteotomies or arthrodeses) modalities. The goals
of these surgeries are to rebalance muscle activity, restore range of motion, and
realign the skeletal segments of the foot. Clinical decision-making for the
J.R. Davids (*) • S.A. Tabaie

Northern California Shriner’s Hospital for Children, Sacramento, Sacramento, CA, USA
e-mail: jdavids@shrinenet.org; tabaiesa@gmail.com

DOI 10.1007/978-3-319-30808-1_58-1
2 J.R. Davids and S.A. Tabaie
management of the ankle/foot in children with CP involves a diagnostic matrix

utilizing data from the clinical history, physical examination, radiographic imag-
ing, and quantitative gait analysis.
Keywords
Cerebral palsy • Foot • Ankle • Management
Introduction
Foot and ankle problems are common in children with cerebral palsy (CP). They
affect both the ambulatory and nonambulatory children either by compromising the
efficiency of gait or leading to issues with orthotic and shoe wear (Davids 2010). To
approach the management of foot and ankle problems in a child with CP, it is
important to understand normal gait and how it is disrupted in children with
CP. This chapter will discuss normal gait at the foot and ankle and the disruption
of normal gait in CP and then present the current paradigm for clinical decision-
making for surgical management about the foot and ankle in children with CP.
Normal Gait
The interaction between the ankle, foot, and the floor is the basis of normal gait. The
function of the foot and ankle is based on a complex interplay between anatomy,
physiology, and physics. Proper leads to efficient function of the knee and hip during
gait (Davids et al. 2007, Inman 1969; Inman et al. 1981; Perry 1992). Standardized,
consistent terminology should be used to describe the alignment of the separate
segments of the ankle and foot (Ponseti et al. 1981). The understanding of foot and
ankle function during normal gait is facilitated by considering the lower leg as
consisting of four segments and the foot as three segments (Inman 1966; Inman
et al. 1981; Perry 1992; Davids 2009a, 2010, 2016; Fig. 1):
• Tibial or shank
• Hindfoot (talus and calcaneus)
• Midfoot (navicular, cuneiforms, and cuboid)
• Forefoot (metatarsals and phalanges)
The foot should also be considered as consisting of two columns (Fig. 2):
• Medial column (talus, navicular, cuneiforms, great toe metatarsal, and phalanges)
• Lateral column (calcaneus, cuboid, lesser toe metatarsals, and phalanges)
Movement of the segments should be described as (Fig. 3):

Foot and Ankle Motion in Cerebral Palsy 3
Fig. 1 Three segments of the foot. (a) Diagram of the AP view of the foot. The hindfoot segment is
red, the midfoot segment is yellow, and the forefoot segment is white. (b) Diagram of the lateral
view of the foot. The hindfoot segment is red, the midfoot segment is yellow, and the forefoot
segment is white
Fig. 2 Two columns of the foot. (a) Diagram of the AP view of the foot. The medial column is red,
and the lateral column is yellow. (b) Diagram of the lateral view (medial side) of the foot. The
medial column is red. (c) Diagram of the lateral view (lateral side) of the foot. The lateral column is
yellow
Fig. 3 (a) Terminology for describing foot segmental alignment, utilizing the hindfoot as an
example. When the plantar aspect of the hindfoot is deviated toward the body’s midline (red
arrow), the segment is described as being inverted. When the plantar aspect is deviated away
from the midline (yellow arrow), the segment is described as being everted. (b) When the distal
aspect of the hindfoot is deviated toward the midline (red arrow), the segment is described as being
adducted. When the distal aspect is deviated away from the midline (yellow arrow), the segment is
described as being abducted
• Plantar aspect of segment

– Inversion (toward the midline)
– Eversion (away from the midline)
• Distal portion of segment
– Adduction (toward the midline)
– Abduction (away from the midline)
• Multi-planar motion
– Supination (inversion and adduction)
– Pronation (eversion and abduction)
• Rotation of the segment about its longitudinal axis
– External rotation (away from midline)
– External rotation (away from midline)
Function of the foot and ankle during the gait cycle is best appreciated by
consideration of kinematics, kinetics, EMG (electromyography), and skeletal align-
ment (Fig. 4).
Fig. 4 Illustration of kinematics, moment, and EMG throughout the gait cycle (IC initial contact,
MSt midstance, TSt terminal stance, PSw preswing, ISw initial swing, MSw midswing, TSw terminal
swing)
The interaction of the foot and ankle with the floor during the stance phase is best
understood utilizing the concept of three rockers (Inman et al. 1981; Perry 1992;
Davids 2009a, 2010).
• First or heel rocker (Fig. 5)

– Occurs from heel strike to foot flat during the loading response subphase of
stance.
– As the body progresses forward over the foot, the ground reaction force moves
progressively distally through the foot, creating an external dorsiflexion
moment.
– The tibia advances forward over the foot, which is achieved by ankle
dorsiflexion.
Fig. 5 The first or heel

rocker. The clinical
photograph shows a heel
strike at initial contact. The
normal ankle kinematics,
kinetics, and dynamic EMG
during the gait cycle are
plotted below the clinical
photograph
35.0
Dorsiflexion
Ankle Angle (degrees) 7.5
Plantar Flexion
–20.0
2.0
Plantar Flexor
Moment (Nm/kg) 0.8
Dorsiflexor
–0.5
Gastrocnemius
Tibialis Anterior
0 25 50 75 100
% Gait Cycle
– This motion is controlled by eccentric activity of the ankle plantar flexor

muscle group, which generates an internal plantar flexion moment.
– This provides shock absorption during loading response.
• Second or ankle rocker (Fig. 6)
– Occurs as the tibia advances over the foot during the midstance subphase of
stance.
Fig. 6 The second or ankle

photograph shows the tibia
advancing over the foot in
midstance. The normal ankle
kinematics, kinetics, and
dynamic EMG during the gait
cycle are plotted below the
clinical photograph
35.0
Dorsiflexion
Ankle Angle (degress) 7.5
Plantar Flexion
–20.0
2.0
Plantar Flexor
Moment (Nm/kg) 0.8
Dorsiflexor
–0.5
Gastrocnemius
Tibialis Anterior
0 25 50 75 100
% Gait Cycle
– With forward progression, the ground reaction force remains distal to the ankle
joint, creating an increasing external dorsiflexion moment.
– This motion is controlled by the concentric activity of the ankle plantar flexor
muscle group, which continues to generate an internal plantar flexion moment.
– This provides stability during midstance.
• Third or forefoot rocker (Fig. 7)
Fig. 7 The third or forefoot

photograph shows the heel
rising and toes dorsiflexing in
terminal stance. The normal
ankle kinematics, kinetics,
and dynamic EMG during the
gait cycle are plotted below
the clinical photograph
35.0
Dorsiflexion
Ankle Angle (degress) 7.5
Plantar Flexion
–20.0
2.0
Plantar Flexor
Moment (Nm/kg) 0.8
Dorsiflexor
–0.5
Gastrocnemius
Tibialis Anterior
0 25 50 75 100
% Gait Cycle
– Prior to the initial contact of the opposite foot, the heel of the reference foot
rises off the ground, and dorsiflexion occurs through the metatarsophalangeal
joints of the forefoot.
– The ankle starts to plantar flex as the reference limb is unloaded during
preswing subphase of stance.
– Largest moment generated by any single muscle group during the gait cycle is
the internal plantar flexion moment generated by the plantar flexor group
during third rocker in terminal stance.
– This is essential for normal gait as it provides a rigid lever (i.e., stability)
during terminal stance.
Normal gait is the consequence of couple movements between the segments of the
foot described earlier in this chapter. The alignments of the ankle and foot, and the
activity of the ankle plantar flexor muscles, determine the location of the ground reaction
force about the knee in mid and terminal stance which influences knee alignment. In
general, ankle plantar flexion is associated with knee extension, and ankle dorsiflexion is
associated with knee flexion. This biomechanical relationship is known as the plantar
flexion/knee extension couple, and it is through this mechanism that control of the foot
and ankle alignment can contribute to the control of knee alignment in stance phase
(Inman et al. 1981; Inman 1966; Perry 1992; Davids 2009a; Fig. 8).
In the swing phase of normal gait, the foot and ankle contribute to clearance and
pre-positioning for the subsequent stance phase.
• Initial swing
– The combination of hip flexion (active), knee flexion (passive), and ankle
dorsiflexion (active) serves to functionally shorten the limb to promote
clearance.
– The tibia or shank segment is externally rotating, and the ankle is
plantarflexing.
• Midswing
– The tibia or shank segment is rotating internally and the ankle is dorsiflexing.
– Maximum shortening is achieved and clearance within 1 cm of the ground
(proprioception) occurs.
• Terminal swing
– The coupled movements described in midswing continue, and the foot is
maintained in a plantigrade position, perpendicular to the anatomical axis of
the tibia or shank segment. This pre-positioning of the foot during terminal
swing will result in a heel strike at the initial contact.
Disruption of Foot, Ankle Function, and Its Effect on Gait

in Cerebral Palsy
Foot and ankle deformities in children with CP are the consequence of a dynamic
imbalance between the extrinsic muscles of the lower leg that control segmental foot
and ankle alignment (Davids 2009b, 2010). The causes of imbalance are linked to:
• Spasticity
• Disrupted motor control
• Impaired balance function
Fig. 8 The plantar flexion/

knee extension couple. In the
midstance of subphase of the
gait cycle, increasing ankle
plantar flexion (red arrow)
shifts the ground reaction
force (red arrow) anterior to
the knee joint center, which
causes the knee to extend.
Increasing ankle dorsiflexion
(black arrow) shifts the
ground reaction force (black)
behind the knee joint center,
which causes the knee to flex
Most often, the ankle plantar flexor muscles are overactive, and the ankle
dorsiflexor muscles are ineffective. Additional imbalance patterns can be evident
between the muscle groups that control foot and ankle supination and pronation.
Ultimately, these muscle imbalances can lead to three common coupled foot and
ankle segmental malalignment patterns in children with spastic-type CP. Other, more
complex, or uncoupled segmental malalignment may occur but are much less
common (Davids 2009b, 2010; Stevens 1988; Mosca 1995; Etnyre et al. 1993;
Tylkowski et al. 2009).
• Equinus deformity (Fig. 9)

– Characterized by excessive plantar flexion of the hindfoot relative to the ankle.
– In a purely equinus deformity, there is normal midfoot and forefoot alignment.
• Equinoplanovalgus deformity (Fig. 10)
– Characterized by equinus deformity of the hindfoot, along with pronation
deformities of the midfoot and forefoot.
– In this deformity pattern, the lateral column of the foot is shorter (anatomically
and functionally) than the medial column.
– Associated deformities of ankle and/or hallux valgus are commonly seen with
equinoplanovalgus foot segmental malalignment.
Fig. 9 Plain radiographs of the foot in a child with equinus deformity. (a) AP view shows normal
segmental alignment. (b) Lateral view shows hindfoot plantar flexion (diminished calcaneal pitch,
normal is 25–30 ), with otherwise normal segmental alignment
• Equinocavovarus deformity (Fig. 11)

– Characterized by equinus deformity of the hindfoot, along with supination
deformity of the midfoot and variable malalignment of the forefoot.
– In this deformity pattern, the lateral column is functionally and/or anatomically
longer than the medial column.
– Associated deformities of the ankle may be seen with equinocavovarus seg-
mental foot malalignment.
The three common malalignment patterns described above are often correctable
on manipulation in younger children with milder forms of CP. However, the foot and
ankle deformities can become rigid and uncorrectable due to the following factors:
• Increasing age and growth

• Fixed shortening or myostatic deformities of the muscles
• Development of permanent skeletal deformities in response to the malalignment
pattern
Foot and ankle segmental malalignment in children with CP can alter function
during both the stance and swing phases of the gait cycle. The three segmental
Fig. 10 Plain radiographs of the foot in a child with equinoplanovalgus deformity. (a) AP view
shows hindfoot pronation, talonavicular uncoverage, and forefoot abduction. (b) Lateral view
shows hindfoot plantar flexion, midfoot pronation, and forefoot pronation
Fig. 11 Plain radiographs of the foot in a child with equinovarus deformity. (a) AP view shows
hindfoot supination and forefoot adduction. (b) Lateral view shows hindfoot varus, midfoot
supination, and forefoot supination
malalignment patterns are associated with a variety of gait disruptions and deviations
(Davids 2010).
• Common alterations to the normal gait cycle seen with all three segmental
malalignment patterns:
– Lack of heel strike at initial contact, disrupting the first rocker and shock
absorption function in the loading response of the gait cycle
– Compromised ankle plantar flexor muscles ability to generate an adequate
internal plantar flexor moment during the second and third rockers
– Inhibition of ankle dorsiflexion in swing phase, compromising clearance in
midswing and proper positioning of the foot and ankle in terminal swing
• Common alterations to the normal gait cycle seen with equinus and
equinocavovarus malalignment patterns:
– Disruption of the second rocker by blocking ankle dorsiflexion, thereby
compromising stability in midstance.
– Due to the hindfoot malalignment with these two deformity patterns, the
lengths of the plantar flexor muscles are shortened, and their ability to generate
tension is adversely affected.
• Common alterations of the normal gait cycle with equinoplanovalgus segmental
malalignment:
– In this deformity pattern, the midfoot and forefoot segments remain in an
unlocked alignment leading to excessive loading of the plantar, medial portion
of the midfoot.
– The malalignment of the midfoot and forefoot further compromises the
moment-generating capacity of the ankle plantar flexor muscles and effec-
tively shortens the lever arm available to this muscle group during the third
rocker.
– External tibial torsion seen with equinoplanovalgus malalignment may con-
tribute to an external foot progression angle, further affecting the lever arm
available to the ankle plantar flexor muscles in terminal stance.
Clinical Decision-Making: Management of Foot and Ankle

Problems in CP
The management of foot and ankle deformities in CP can be very challenging. In

order to provide the best care for this patient population, it is best to utilize a
diagnostic matrix that incorporates the following (Davids 2009b, 2010; Davids
et al. 2005, 2007; Inman et al. 1981; Perry 1992):
• Clinical history
• Physical examination including observational gait analysis
• Plain radiographs
• Quantitative gait analysis
Clinical History
Often patients present with specific complaints of gait abnormalities, such as pain
with ambulation, tripping when ambulating, and/or issues with shoe wear. It is
imperative to inquire and document the specifics of the symptoms, inciting and
alleviating factors, duration of symptoms and their progression over time, and finally
the treatment history to date.
Physical Examination
The sequence of the physical exam in children with CP who have foot and ankle
deformities should begin with a thorough inspection and progress to a hands-on
evaluation of the foot segmental alignment in both weight-bearing and non-weight-
bearing conditions. Inspection of the plantar aspect of the feet should be done to
evaluate for the presence of excessive or inadequate skin callous formation that
could be due to abnormal loading patterns and/or problems with shoe/orthotic wear.
Manual examination is performed to determine and document:
• Static standing alignment – assessed from the front, back, and both sides of the
patient
• Intra- and intersegmental flexibility
• Rotational profile from the hips to the foot
• Range of motion – passive and active
• Muscle strength and selective motor control
The final key portion of the physical exam is the observational gait analysis.
Observing ambulation needs to be done from multiple viewpoints in both the coronal
and sagittal planes. The patient should be barefoot and wearing tight cycling style
shorts in order for the examiner to adequately visualize the entire lower extremity
including the thigh, knee, lower leg, ankle and foot. When observing a patient’s gait,
the following events of the gait cycle related to dynamic foot function should be
documented:
• Foot position at initial contact – heel strike, flatfoot, or toe strike

• Foot alignment in midstance – varus/valgus in the coronal plane and internal/
external in the transverse plane (foot progression angle)
• Foot alignment at toe-off – varus/valgus in the coronal plane and dorsi-/plantar-
flexed in the sagittal plane
• Foot clearance in swing phase
Radiographic Evaluation
Radiographic images for evaluation of foot and ankle deformities must be routinely
performed with the child fully weight bearing. Non-weight-bearing images are not of
value in assessing foot segmental alignment and should not be utilized. Three
standardized views should always be obtained:
• AP foot
• Lateral foot
• AP ankle
When analyzing radiographs and designating deformities, it is best to divide the

foot into three segments and two columns as was described earlier in this chapter,
then determining the relative alignment of each segment and the relative length of
each column. A comprehensive technique of quantitative segmental analysis of the
foot and ankle has been developed by Davids et al. (2005) and was derived from the
foot model originally developed by Inman and colleagues (Inman et al. 1981).
The technique of quantitative segmental analysis was developed from normative
values of a cohort of 60 normal feet in children between ages of 5 and 17 years old,
utilizing ten radiographic measurements to determine the alignment of the segments
and the lengths of the columns of the foot and ankle (Davids et al. 2005; Table 1).
Quantitative Gait Analysis
The calculation of foot and ankle kinematics and kinetics involves modeling
assumptions and approximations concerning the relationship between the skin
markers and the underlying skeletal anatomy. The standard ankle and foot model
most commonly used in clinical gait analysis was developed in the 1980s, uses
markers at the malleoli and forefoot, and considers the foot as a single segment
(Perry 1992). It is assumed that the foot segment is rigid from the hindfoot to the
forefoot. Ankle motion in the sagittal plane is calculated from the location of the foot
axis relative to the tibial axis. Any movement between the three segments of the foot
(i.e., hindfoot to midfoot, midfoot to forefoot) that occurs between the malleolar and
forefoot markers is captured by this simple foot model and described as ankle
motion. Significant measurement artifact occurs when the normal foot segmental
alignment is disrupted (i.e., equinoplanovalgus foot malalignment in children with
CP). This artifact creates apparent discrepancies within the diagnostic matrix
between the data derived from the physical examination, observational gait analysis,
and quantitative gait analysis. Failure to appreciate the causes for these apparent
Table 1 Radiographic measurements of normative values for feet in children

Radiographic measurements: quantitative and categorical definitions
Normal
(mean Abnormal high value Abnormal low value
3.75) (> mean +1 SD) (> mean +1 SD)
Hindfoot
Tibiotalar angle (degrees) 1.1 3.75 Eversion Inversion
Calcaneal pitch (degrees) 17 6.0 Calcaneus Equinus
Tibio-calcaneal angle 69 8.4 Equinus Calcaneus
(degrees)
Talocalcaneal angle (degrees) 49 6.9 Eversion Inversion
Midfoot
Naviculo-cuboid overlap (%) 47 13.8 Pronation Supination
Talonavicular coverage angle 20 9.8 Abduction Adduction
(degrees)
Lateral talo-first metatarsal 13 7.5 Pronation Supination
angle (degrees)
Forefoot
Anteroposterior talo-first 10 7.9 Abduction Adduction
metatarsal angle (degrees)
Metatarsal stacking angle 82 9 Supination Pronation
(degrees)
Columns
Medial-lateral column ratio 0.9 0.3 Abduction Adduction
discrepancies may result in confusion for clinicians and compromise clinical deci-
sion-making.
Technological improvements have allowed for the development of more sophis-
ticated, multisegment foot models that more accurately approximate the complex
anatomy and biomechanics of the foot (MacWilliams et al. 2003). However, these
models are difficult to apply to children with CP, because of small foot size and
deformity.
Additionally, the dynamic EMG, which is part of the quantitative gait analysis, is
very relevant in the evaluation foot deformities in children with CP (Sutherland
1993). The incorporation and use of surface/fine-wire EMG provides information on
the timing of muscle activity during the gait cycle (Hoffer et al. 1985; Scott and
Scarborough 2006). This data is imperative in determining:
• The relative activity of tibialis anterior and posterior muscles in both stance and
swing phases
• Selection of a particular muscle-tendon unit for lengthening or transfer
The final component of quantitative gait analysis is dynamic pedobarography,

which measures the spatial and temporal distribution of force over the plantar aspect
of the foot during the stance phase of the gait cycle (Jameson et al. 2008; Davids
Fig. 12 Examples of abnormal foot loading patterns (each figure is of a left, as if looking down
from above, toes at the top, lateral to the left, medial to the right): (a) displacement of the center of
pressure progression (COPP – solid red line) medially, describing a valgus loading pattern.
(b) Displacement of the COPP laterally, resulting in a varus loading pattern. (c) Displacement of
COPP distally, describing an equinus loading pattern). (d) Displacement of the COPP proximally
(i.e., excessive duration of the COPP in the hindfoot segment, reflected by increased thickness of the
red line) resulting in a calcaneus loading pattern
2009a). Pedobarography provides the following information regarding the patient’s

dynamic foot function:
• Foot contact patterns

• Foot pressure distribution and magnitude
• Progression of the center of pressure through the foot
As described above, foot function during gait in children with CP is disrupted by

several patterns of skeletal segmental malalignment that can ultimately affect all
three rockers in stance phase. This biomechanical disruption has been termed lever
arm deficiency and is best characterized by the center of pressure progression
(COPP) relative to the foot. Deviations in the location and duration of the COPP
relative to the segments of the foot when compared to normative values of COPP in
children can be used to describe common abnormal loading patterns (Davids 2009a;
Fig. 12).
• Valgus loading pattern

– Displacement of the COPP medially due to an everted, abducted, or pronated
segmental malalignment of the foot segment
• Varus loading pattern

– Displacement of the COPP laterally, which is usually due to an inverted,
adducted, or supinated segmental malalignment of the foot segment
• Equinus loading pattern
– Prolonged duration of the COPP in the forefoot segment
• Calcaneus loading pattern
– Prolonged duration of the COPP in the hindfoot segment
Levels of Deformity
Foot and ankle deformities in children with CP are sequential and progressive with
growth and development. The deformities can be classified into three levels (Davids
2010):
• Level I
– Dynamic soft tissue imbalance with no skeletal pathoanatomy
• Level II
– Fixed or myostatic soft tissue imbalance with flexible and correctable skeletal
segmental malalignment
• Level III
– Fixed myostatic soft tissue imbalance with associated structural skeletal
deformities
Understanding the levels of deformity in children with CP can guide in both the
nonoperative and surgical interventions for these patients.
Treatment Principles
Management of foot and ankle deformities in children with CP can consist of both
nonsurgical and surgical modalities. Treatment options include (Davids et al. 2004,
2007; Gage 1994, 1995, 2004):
• Orthotics
– May be used to protect the outcome of a surgical procedure during the healing
and rehabilitation phases, to prevent the development or worsening of muscu-
loskeletal deformities with growth, and to improve gait.
– When the goal is to improve gait, the physician should clearly identify the gait
deviations and functional deficits to be addressed by the orthosis.
– Clinicians must be familiar with the biomechanical characteristics and clinical
indicators of six common orthotic designs (Table 2):
1. Foot orthosis
2. Supramalleolar orthosis
3. Posterior leaf spring orthosis
4. Articulating ankle-foot orthosis
5. Floor-reaction ankle-foot orthosis
Table 2 Indications for the use of common orthoses to improve gait in children with CP
Indications for use of orthoses to improve gait in children with cerebral palsy
Physical exam Gait deviation
Initial Terminal Swing
Orthosis Hip Knee Ankle Foot contact Midstance stance phase
UCBL Normal Normal Normal Mild, Normal Normal Normal Normal
correctible
Supramalleolar Normal Normal Normal Mild, Normal Normal Normal Normal
orthosis correctible
Foot and Ankle Motion in Cerebral Palsy
PLSO Normal Normal Dorsiflexion to 5 Mild, Absent Normal Normal Increased

correctible heel plantar
strike flexion
Articulating Normal Normal Dorsiflexion to 5 Mild, Absent Increased plantar flexion, Normal Increased
AFO correctible heel increased knee extension plantar
strike (mild) flexion
Solid AFO Normal Normal Dorsiflexion to Moderate, Absent Increased plantar flexion, Increased Increased
neutral partially heel increased knee extension, or plantar plantar
correctible strike increased knee flexion flexion flexion
(mild)
FRAFO Extension Extension Dorsiflexion to Moderate, Absent Increased dorsiflexion, Increased Increased
30 20 neutral thigh-foot partially heel increased knee flexion, dorsiflexion dorsiflexion
angle 15 correctible strike increased hip flexion
extension
UCBL University of California at Berkeley Laboratory, PLSO posterior leaf spring orthotic, AFO ankle-foot orthotic, FRAFO floor reaction ankle-foot orthotic
19
6. Solid ankle-foot orthosis

• Pharmacologic/Neurosurgical
– Level I deformities resulting from dynamic soft tissue imbalance with no
skeletal deformities can be treated with interventions to manage muscle tone
and spasticity (Preiss et al. 2003; Boyd et al. 2000; Bjornson et al. 2007).
1. Pharmacologic – botulinum toxin (or phenol) injection
2. Neurosurgical – selective dorsal rhizotomy or intrathecal baclofen
– Early management with the abovementioned options is favored to avoid
development of fixed deformities of the muscle-tendon unit.
Surgical Treatment Goals
Surgical management to correct foot and ankle deformities in children with CP may
be selected to improve gait or function, shoe wear, and cosmesis. These goals may be
achieved by surgical procedures that are designed to improve foot shape (Davids
2010; Sutherland 1993; Mosca 1998).
• Improved foot shape following soft tissue and skeletal surgery can restore the
stability function of the foot during:
– The second rocker in midstance and the skeletal lever arm function of the foot
during the third rocker in terminal stance
• Cosmetic improvements following foot surgery are related to:
– Improved visual assessment of static standing foot alignment – restoration of
the medial longitudinal arch and toe alignment
– Improved foot progression angle during stance phase
Additionally, it is presumed that surgically improved foot shape can correct pain
by improving foot loading and stability in stance phase. In regard to pain, it is
important to make the distinction between the younger children with CP who may
tolerate mild or moderate foot deformities versus the older teenage/adult population.
Deformities that may be tolerated in a younger CP population are often poorly
tolerated in teenage and adult life as the body weight increases, leading to a greater
magnitude of the abnormal loading. The cumulative effect of the magnitude of
abnormal loading results in premature degenerative changes of the joints of the
foot and ankle. Presumptively, surgery to improve foot shape in childhood will
improve the loading of the foot and decrease the possibility of early degenerative
arthritis in adulthood.
Surgical Treatment Techniques
As noted earlier in this chapter, there are three common coupled foot and ankle
segmental malalignment patterns in children with spastic-type CP: equinus,
equinoplanovalgus, and equinocavovarus. Additionally, hallux valgus is a common
secondary malalignment that may be associated with any of the three principle
malalignment patterns. The surgical options for these coupled segmental deformities
are based on the level of deformity and will be discussed below.
• Equinus – Pure plantar flexor malalignment is usually the consequence of

overactivity (level I) or tightness (level II) of the ankle plantar flexor muscle
group. Assessment of foot segmental alignment with plain radiographs is essen-
tial to establishing that there are no associated deformities at the level of the mid-
or forefoot (Davids 2010).
– Surgical management starts with level II deformities.
– Careful assessment is required to determine the relative contributions of the
gastrocnemius and soleus muscles to the fixed shortening of the plantar flexor
group.
1. This is best achieved by assessing ankle dorsiflexion range of motion with
the knee both flexed and extended.
2. Isolated limitation with the knee flexed suggests involvement of the soleus
muscle, limitation with the knee extended suggests involvement of the
gastrocnemius muscle, and limitation regardless of knee position represents
involvement of both muscles.
– The goal of surgical lengthening of the plantar flexor group is to achieve five
degrees of dorsiflexion when the knee is extended.
1. When only the gastrocnemius is involved, a selective fractional lengthening
is best performed proximally at the level of the muscle belly.
2. When both muscles are involved and 15 or less of correction is required,
selective fractional lengthening mid-calf at the level of the myotendinous
junction is preferred.
3. When both muscles are involved and greater than 15 of correction is
required, nonselective lengthening distally at the Achilles tendon level is
necessary.
– In most cases, a fractional lengthening at the level of the muscle belly or
myotendinous junction is sufficient. However, with proper patient selection
and careful surgical technique, all three techniques described above may be
effective and excessive lengthening and weakness of the ankle plantar flexor
muscle groups can be avoided (Etnyre et al. 1993; Tylkowski et al. 2009).
• Equinoplanovalgus – This malalignment pattern, which is often referred to as
flatfoot, is the consequence of overactivity (level I) or tightness (level II) of the
ankle plantar flexor and evertor muscle groups. Physical examination is needed to
determine whether the muscle-tendon unit deformities are dynamic or mysostatic.
Fixed skeletal segmental malalignment (level III) is assessed with plain radio-
graphs which are essential for preoperative planning (Davids 2010).
– As with equinus malalignment, in this pattern orthopedic surgical intervention
starts with level II deformity and is usually seen in children between 4 and
7 years of age.
1. The treatment choice is lengthening of the ankle plantar flexor muscle
group, as described earlier in the equinus section.
Fig. 13 The technique for the intraoperative stress assessment of foot segmental alignment. The
assistant stabilizes the knee while the surgeon loads the foot and ankle with the foot pusher. A
fluoroscopic image is taken in the loaded position and is used to assess the segmental alignment
2. Additionally, transfer of the peroneus brevis muscle to the peroneus longus

muscle should be performed when fixed deformity of the former is present.
3. At the time of surgery, it is essential to determine that normal foot segmen-
tal alignment has been restored following soft tissue surgeries (Fig. 13).
– Failure to restore normal skeletal alignment with soft tissue surgery alone
reclassifies the deformity to level III (Davids 2009b; Mosca 1996, 1998; Yoo
et al. 2005; Danko et al. 2004).
1. The primary procedure used to correct level III equinoplanovalgus
malalignment is lateral column lengthening, which can be performed at
the following locations in the foot:
(a) Neck of the calcaneus
(b) Calcaneocuboid (CC) joint
(c) Body of the cuboid
2. Lateral column lengthening usually allows for correction of all three seg-
ments of the foot, most likely secondary to ligamentotaxis (Fig. 14).
(a) The lengthening is approximately 1–2.5 cm.
(b) Interposition grafting with tricortical iliac crest or allograft.
(c) Lengthening through the CC joint (arthrodesis) is often done in
children over 12 years of age. Internal fixation should be used when
the graft size is greater than 1.5 cm. in order to promote early mobi-
lization and minimize late graft collapse during phases of healing.
3. Following lateral column lengthening for a level III equinoplanovalgus foot
deformity, assessment of the medial column is performed.
(a) If there is residual forefoot varus deformity, or if the medial column is
hypermobile in the sagittal plane (i.e., unmasked deformities), then a
Fig. 14 Correction of moderate level III equinoplanovalgus deformity. (a, b) Lateral and AP
radiographs in a child with CP showing equinoplanovalgus segmental malalignment. (c, d) Lateral
and AP radiographs of the foot after gastrocsoleus fractional lengthening and lateral column
lengthening through the neck of the calcaneus
plantar flexion osteotomy (or a combination of both) is made in the

medial cuneiform or base of the great toe metatarsal (if the child is
skeletally mature.
(b) If there is residual forefoot abduction deformity (i.e., incomplete
correction), then a talonavicular arthrodesis needs to be performed.
(c) If lengthening of the lateral column fails to correct the hindfoot
deformity, then arthrodesis of the subtalar, talonavicular, and
calcaneocuboid joints are required to achieve optimal alignment of
the foot. Fortunately, this is rarely required, even for feet with signif-
icant malalignments.
• Equinocavovarus – This malalignment pattern that is often referred to as varus

foot is usually the consequence of overactivity (level I) or tightness (level II) of
the ankle plantar flexor and invertor muscle groups. As with the other
malalignment patterns, physical examination is needed to determine whether
the muscle-tendon unit deformities are dynamic or mysostatic. Fixed skeletal
segmental malalignment (level III) is assessed with plain radiographs and is
essential for preoperative planning (Davids 2010).
– In children over the 6 years of age with equinocavovarus foot malalignment,
surgical management with muscle-tendon transfer can be an option.
1. Information from the physical exam, kinematics, kinetics, dynamic EMG,
and pedobarography are used to determine the relative contributions of the
tibialis anterior and tibialis posterior muscles to the dynamic varus defor-
mity that occurs during stance and swing phases (Sutherland 1993; Scott
and Scarborough 2006; Hoffer et al. 1985).
(a) Split transfer of the tibialis anterior muscle should only be performed
when there is ankle dorsiflexion appreciated in midstance.
(b) Split transfer of the tibialis posterior should only occur when the
dynamic EMG shows the timing of the activation of this muscle
corresponds with the presence of varus malalignment during specific
subphases of the gait cycle.
(c) When kinematics and dynamic EMG assessments are not available,
dynamic varus deformity is best treated by concomitant split transfer
of the tibialis anterior muscle and fractional lengthening of the tibialis
posterior muscle (Barnes and Herring 1991).
– Level II deformity involves sequential correction of the hindfoot and midfoot
soft tissue contractures (Fig. 15).
1. Correction of the deformity is performed by fractional lengthening of the
following muscle groups:
(a) Ankle plantar flexor muscle group
(b) Tibialis posterior muscle
(c) Flexor hallucis and digitorum longus muscles – rarely necessary
(d) Abductor hallucis
(e) Plantar fascia and short intrinsic muscle of the foot
2. At the time of surgery, it is essential to determine the normal foot segmental
alignment has been restored following the soft tissue surgeries.
(a) Confirmation of normal alignment is best achieved with intraoperative
stress radiographs of the foot as was described earlier in the chapter.
(b) Failure to restore normal skeletal alignment should result in reclassi-
fication of the deformity to level III.
– Correction of level III equinocavovarus segmental malalignment differs from
correction of equinoplanovalgus malalignment, because there is not one single
skeletal procedure that can achieve adequate correction of all three segments of
the foot (Fig. 16).
1. Gross alignment and dynamic loading of the foot may be greatly improved
by performing sequential osteotomies that create deformities to compensate
for segmental malalignments.
Fig. 15 Technique for the correction of level II equinocavovarus deformity. (a) Three incisions are
used to lengthen the appropriate soft tissue structures. (b) The plantar fascia and intrinsic muscles of
the foot are released through the plantar incision. (c) Fractional lengthening of the abductor hallucis
muscle is performed through the distal medial incision. (d) Fractional lengthening of the
gastrocsoleus muscle group is performed through the medial calf incision. (e) Fractional lengthen-
ing of the tibialis posterior muscle is performed through the same medial calf incision. The flexor
digitorum and hallucis longus muscles can also be lengthened, when necessary, though this incision
(a) Residual hindfoot varus malalignment may be corrected by calcaneal

slide or laterally based closing wedge osteotomies (Koman et al.
1993).
(b) Residual midfoot supination deformity may be corrected by lateral
column shortening through the cuboid.
(c) Residual forefoot pronation may be corrected by dorsiflexion
osteotomy of the medial column.
(d) Residual forefoot supination deformity may be corrected by plantar
flexion osteotomy of the medial column.
2. Arthrodesis strategy is reserved for the most severe cases of level III
equinocavovarus segmental malalignment.
(a) Double arthrodesis of the calcaneocuboid and talonavicular joints may
be required to correct extreme midfoot cavus deformity.
(b) Triple arthrodesis of the subtalar, calcaneocuboid, and talonavicular
joints may be required to achieve optimal foot alignment.
Fig. 16 Correction of level III equinocavovarus deformity. (a, b) Lateral and AP radiographs of the
foot in a child with CP showing equinocavovarus segmental malalignment. (c, d) Lateral and AP
radiographs of the foot after gastrocsoleus fractional lengthening, posterior tibialis fractional
lengthening, radical plantar fascia release, calcaneal slide osteotomy, and medial column osteotomy
through the cuneiform
• Hallux Valgus – This deformity in children with CP may be the consequence of

intrinsic and/or extrinsic factors. The gait deviations secondary to these factors
tend to medialize the forces across the great to metatarsophalangeal (MTP) joint,
resulting in an external abduction or valgus moment (Davids 2010).
– Radiographic assessment is essential for determining the elements and magnitude

of deformity at the great to MTP joint and the presence of associated foot
segmental malalignments.
– Correction of the deformity is indicated to address the following issues:
1. Pain at the great toe MTP joint
2. Treat hygiene problems related to toe positioning
3. Facilitate shoe and orthotic wear
– For hallux valgus deformities in this patient population, there is little evidence to
support lengthening or release to correct or improve foot intrinsic muscle tight-
ness (level II deformity).
Fig. 17 Correction of hallux valgus deformity (in association with comprehensive foot segmental
malalignment correction with lateral column lengthening through the neck of the calcaneus).
(a) Standing AP radiograph of the foot in a child with CP. The great toe deformity consists of
metatarsus varus and MTP joint valgus. (b) Intraoperative fluoroscopy AP showing great toe MTP
arthrodesis. (c) Standing AP radiograph of the foot 1 year after great toe arthrodesis (note the
indirect correction of the increased inter-metatarsal angle)
1. Poor results for soft tissue balancing procedures are most likely a consequence
of the failure to address significant extrinsic causes of hallux valgus deformity
in children with CP.
– The preferred treatment of level III hallux valgus deformity in children with CP is
great toe MTP arthrodesis (Davids et al. 2001; Bishay et al. 2009; Fig. 17).
1. Due to the proximal location of the physis of the proximal phalanx, this
procedure should not be performed in children who have more than 2 years
of growth remaining.
2. The shape of the foot and the child’s gait pattern determines optimal alignment
of the MTP arthrodesis.
(a) Correction of the coronal plane should align the phalanges of the great toe
with the lesser toes. Sagittal plane alignment of the arthrodesis should
include between 15 and 20 of dorsiflexion (relative to the floor when the
foot is weight bearing) to facilitate the forefoot rocker.
Summary/Conclusion
• The normal interaction of the ankle/foot with the floor is best described by the
three rockers (heel, ankle, and forefoot).
• There are three common ankle/foot segmental malalignment patterns seen in

children with CP: equinus, equinoplanovalgus, and equinocavovarus.
• Each type of ankle/foot malalignment can be classified into three levels based
upon the presence of dynamic muscle imbalance, fixed or myostatic deformity of
the muscle-tendon unit, and fixed skeletal deformities.
• Clinical decision-making for the management of the ankle/foot in children with
CP involves a diagnostic matrix utilizing data from the clinical history, physical
examination, radiographic imaging, and quantitative gait analysis.
• Common orthopedic surgical interventions used to address ankle/foot problems in
children with CP include muscle-tendon unit lengthening or transfer and skeletal
osteotomies or arthrodeses. The goals of these surgeries are to rebalance muscle
activity, restore range of motion, and realign the skeletal segments of the foot.
Acknowledgments Sean Brown, M.A. for assistance in preparation of figure.
References
Barnes MJ, Herring JA (1991) Combined split anterior tibial-tendon transfer and intramuscular
lengthening of the posterior tibial tendon. Results in patients who have a varus deformity of the
foot due to spastic cerebral palsy. J Bone Joint Surg Am 73(5):734–738
Bishay SN, El-Sherbini MH, Lotfy AA, Abdel-Rahman HM, Iskandar HN, El-Sayed MM (2009)
Great toe metatarsophalangeal arthrodesis for hallux valgus deformity in ambulatory adoles-
cents with spastic cerebral palsy. J Child Orthop 3(1):47–52
Bjornson K, Hays R, Graubert C, Price R, Won F, McLaughlin JF, Cohen M (2007) Botulinum
toxin for spasticity in children with cerebral palsy: a comprehensive evaluation. Pediatrics 120
(1):49–58
Boyd RN, Pliatsios V, Starr R, Wolfe R, Graham HK (2000) Biomechanical transformation of the
gastroc-soleus muscle with botulinum toxin A in children with cerebral palsy. Dev Med Child
Neurol 42(1):32–41
Danko AM, Allen B Jr, Pugh L, Stasikelis P (2004) Early graft failure in lateral column lengthening.
Davids JR (2009a) Normal function of the ankle and foot: biomechanics and quantitative analysis.
In: McCarthy JJ, Drennan JC (eds) Drennan’s the child foot and ankle. Lippincott Williams and
Wilkins, Philadelphia, pp 54–63
Davids JR (2009b) Orthopaedic treatment of foot deformities. In: Schwartz M, Gage J, Koop S,
Novacheck T (eds) The identification and treatment of gait problems in cerebral palsy. MacKeith
Press, London, pp 514–533
Davids JR (2010) The foot and ankle in cerebral palsy. Orthop Clin North Am 41(4):579–593
Davids JR (2016) Biomechanically based clinical decision making in pediatric foot and ankle
surgery. In: Sabharwal S (ed) Pediatric lower limb deformities. Springer, Cham, pp 153–162
Davids JR, Mason TA, Danko A, Banks D, Blackhurst D (2001) Surgical management of hallux
valgus deformity in children with cerebral palsy. J Pediatr Orthop 21(1):89–94
Davids JR, Ounpuu S, DeLuca PA, Davis RB 3rd (2004) Optimization of walking ability of children
with cerebral palsy. Instr Course Lect 53:511–522
Davids JR, Gibson TW, Pugh LI (2005) Quantitative segmental analysis of weight-bearing radio-
graphs of the foot and ankle for children: normal alignment. J Pediatr Orthop 25(6):769–776
Davids JR, Rowan F, Davis RB (2007) Indications for orthoses to improve gait in children with
cerebral palsy. J Am Acad Orthop Surg 15(3):178–188
Davis RB, Jameson E, Davids JR, Christopher LM, Rogozinski BM, Anderson JP (2007) The
design, development, and initial evaluation of a multisegment foot model for routine clinical gait
analysis. In: Smith P, Harris GF, Marks R (eds) Foot and ankle motion analysis: clinical
treatment and technology. CRC Press, Boca Raton, pp 425–444
Etnyre B, Chambers CS, Scarborough NH, Cain TE (1993) Preoperative and postoperative assess-
ment of surgical intervention for equinus gait in children with cerebral palsy. J Pediatr Orthop 13
(1):24–31
Gage JR (1994) The role of gait analysis in the treatment of cerebral palsy. J Pediatr Orthop 14
(6):701–702
Gage JR (1995) The clinical use of kinetics for evaluation of pathologic gait in cerebral palsy. Instr
Course Lect 44:507–515
Gage JR (2004) Orthotics: a comprehensive clinical approach. The treatment of gait problems in
cerebral palsy. MacKeith Press, London, pp 273–285
Hoffer MM, Barakat G, Koffman M (1985) 10-year follow-up of split anterior tibial tendon transfer
in cerebral palsied patients with spastic equinovarus deformity. J Pediatr Orthop 5(4):432–434
Inman VT (1966) The human foot. Manit Med Rev 46(8):513–515
Inman VT (1969) The influence of the foot-ankle complex on the proximal skeletal structures. Artif
Limbs 13(1):59–65
Inman VT, Ralston J, Todd F (1981) Human walking. Williams & Wilkins, Baltimore
Jameson EG, Davids JR, Anderson JP, Davis RB 3rd, Blackhurst DW, Christopher LM (2008)
Dynamic pedobarography for children: use of the center of pressure progression. J Pediatr
Orthop 28(2):254–258
Koman LA, Mooney JF 3rd, Goodman A (1993) Management of valgus hindfoot deformity in
pediatric cerebral palsy patients by medial displacement osteotomy. J Pediatr Orthop 13
(2):180–183
MacWilliams BA, Cowley M, Nicholson DE (2003) Foot kinematics and kinetics during adolescent
Mosca VS (1995) Calcaneal lengthening for valgus deformity of the hindfoot. Results in children
who had severe, symptomatic flatfoot and skewfoot. J Bone Joint Surg Am 77(4):500–512
Mosca VS (1996) Flexible flatfoot and skewfoot. Instr Course Lect 45:347–354
Mosca VS (1998) The child’s foot: principles of management. J Pediatr Orthop 18(3):281–282
Perry J (1992) Gait analysis: normal and pathologic function. Slack, Thorofare
Ponseti IV, El-Khoury GY, Ippolito E, Weinstein SL (1981) A radiographic study of skeletal
deformities in treated clubfeet. Clin Orthop Relat Res (160):30–42
Preiss RA, Condie DN, Rowley DI, Graham HK (2003) The effects of botulinum toxin (BTX-A) on
spasticity of the lower limb and on gait in cerebral palsy. J Bone Joint Surg Br 85(7):943–948
Scott AC, Scarborough N (2006) The use of dynamic EMG in predicting the outcome of split
posterior tibial tendon transfers in spastic hemiplegia. J Pediatr Orthop 26(6):777–780
Stevens PM (1988) Effect of ankle valgus on radiographic appearance of the hindfoot. J Pediatr
Orthop 8(2):184–186
Sutherland DH (1993) Varus foot in cerebral palsy: an overview. Instr Course Lect 42:539–543
Tylkowski CM, Horan M, Oeffinger DJ (2009) Outcomes of gastrocnemius-soleus complex
lengthening for isolated equinus contracture in children with cerebral palsy. J Pediatr Orthop
29(7):771–778
Yoo WJ, Chung CY, Choi IH, Cho TJ, Kim DH (2005) Calcaneal lengthening for the planovalgus
foot deformity in children with cerebral palsy. J Pediatr Orthop 25(6):781–785
Walking and Physical Activity Monitoring
in Children with Cerebral Palsy
Kristie F. Bjornson and Nancy Lennon
Abstract
Interventions, surgery, and rehabilitation strategies for children with CP are often
focused on enhancing walking and physical activity. Until recently, outcomes to
capture the effectiveness of these interventions have been primarily measures of
mobility that were collected within laboratory and clinical settings. Monitoring of
walking and physical activity via accelerometry allows ecologically based mea-
surement of treatment outcome. Advances in accelerometry technology, transfer
of data, and processing software now allow the clinician and researcher the ability
to quantify walking and physical activity within the context of daily life. This
chapter (1) provides an overview of walking (WA) and physical activity
(PA) measurement within the context of The International Classification of
Functioning and Disability Framework, (2) describes current methodology for
monitoring walking and physical activity in context of daily life, (3) summarizes
walking activity (WA) and physical activity (PA) via activity monitors in cerebral
palsy (CP), and (4) reviews proposed implications for clinical community mon-
itoring of WA and PA in CP.
Keywords
Physical activity (PA) • Walking activity (WA) • Light physical activity (LPA) •
Moderate physical activity (MPA) • Moderate vigorous physical activity
(MVPA) • Cerebral palsy (CP)
K.F. Bjornson (*)

Seattle Children’s Research Institute, University of Washington, Seattle, WA, USA
e-mail: Kristie.bjornson@seattlechildrens.org
N. Lennon
Nemours/Alfred I. duPont Hospital for Children, Wilmington, DE, USA
e-mail: Nancy.lennon@nemours.org

DOI 10.1007/978-3-319-30808-1_59-1
2 K.F. Bjornson and N. Lennon
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Conceptual Framework for Monitoring WA and PA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Physical Versus Walking Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Methods to Document Walking and Physical Activity: ICF Perspective . . . . . . . . . . . . . . . . . . . . 5
Monitoring Technology to Capture Walking and Physical Activity in CP . . . . . . . . . . . . . . . . . . 8
State of the Science: Walking and Physical Activity in Cerebral Palsy (CP) . . . . . . . . . . . . . . . 16
Monitoring of Walking and Physical Activity: Clinical Implications for Children CP . . . . . 21
Considerations and Limitations of Walking and Physical Activity Monitoring . . . . . . . . . . . . . 25
Conclusions/Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Introduction
Children with cerebral palsy (CP) experience walking and physical activity limita-
tions which negatively influence their ability to physically participate in day-to-day
life. Overall walking and physical activity for youth with CP is significantly less than
youth without disability (Bjornson et al. 2014b). School-aged children with CP are
less physically active than typically developing children, such that children with CP
would need to exercise 2.5 h/day to reach the daily physical activity levels of peers
(van den Berg-Emons et al. 1995). The importance of physical and walking activity
is revealed in findings that motor function has also been shown to be predictive of
restrictive participation in mobility, education, and social relations for children with
CP (Beckung and Hagberg 2002). Functional level as defined by the Gross Motor
Function Classification System (GMFCS) (Palisano et al. 2007) is strongly associ-
ated with self-care, activities of daily living, and participation (Voorman et al. 2006).
Children with the greatest motor limitations (quadriplegia, GMFCS level IV/V)
participated in fewer activities with less intensity than those with unilateral or
bilateral impairment (Imms et al. 2008).
Interventions (orthopedic/neurosurgery, injection therapy, oral medications) and
rehabilitation strategies (orthotics, gait or strength training, constraint-induced ther-
apy) for children with CP focus on enhancing walking and physical activity. Until
recently, outcomes to capture the effectiveness of these interventions have been
primarily measures of mobility that were collected within laboratory and clinical
settings. Advances in accelerometry technology, transfer of data, and processing
software now allow the clinician and researcher the ability to quantify walking and
physical activity within the context of daily life. Monitoring of walking and physical
activity via accelerometry allows ecologically based measurement of treatment
outcome. Thus the specific day-to-day functional movement or activity that is
hopefully enhanced by our treatments is captured in real time as well as the
environment in which the child needs to perform the activity. To support the
application of walking and physical activity monitoring in children with CP, this
chapter will provide the following:
Walking and Physical Activity Monitoring in Children with Cerebral Palsy 3
1. An understanding of the measurement of walking activity (WA) and physical

activity (PA) in context of the International Classification of Functioning and
Disability (WHO-ICF Framework) (World Health Organization 2002)
2. An exploration of current methods of monitoring walking and physical activity in
context of daily life
3. A description of current practices of collecting WA and PA data via activity
monitors in cerebral palsy (CP)
4. A review of current and future implications for clinical monitoring of WA and PA
in CP
State of the Art
Conceptual Framework for Monitoring WA and PA
The International Classification of Functioning and Disability Framework (ICF) was

developed to provide a unified and standard language for the description and testing
of influences on health and health-related states (see Fig. 1; World Health Organi-
zation 2002). Within the ICF framework, a child with spastic diplegia cerebral palsy
could be described as having limitations across some and/or all of the components.
For example, impairments such as decreased strength, limited range of motion, and
spasticity would be limitations in “body function and structure.” Difficulty in
walking or climbing stairs would be considered a limitation in “activity,” while
difficulty in walking at school or the mall would be an example of a limitation in life
The International Classification of Functioning, Disability and Health

(ICF): Interaction of Components*
Cerebral Palsy-Spastic Diplegia
Body Functions & Structures Activities Participation

Decreased strength Difficulty Walking Difficulty Walking at
spasticity & ROM climbing strairs school/mall
Environmental Factors Personal Factors

Hilly community Motivation to be active
Interanational Classification of Functioning, Disability & Health. 2002; WHO
Fig. 1 International classification of functioning, disability and health (ICF): interaction of com-
ponents for a child with spastic diplegia cerebral palsy (World Health Organization 2002)
“participation.” If walking were impacted by living in a hilly community, this would

present as an “environmental” factor. Having low motivation to walk or be physi-
cally active would be a contributing “personal” factor to overall health.
“Activity” is defined within the ICF as the execution of a task or action by an
individual, while participation is considered involvement in a life situation. The
component of “activity” can be measured through the constructs of capacity and
performance. “Capacity” is the execution of a task in a controlled environment, such
as the standardized gross motor function tests administered by clinicians and ther-
apists. Capacity-based measures tell us what a child is “capable” of doing when
asked to perform in a structured clinical situation and represent most clinical out-
comes of interventions today. “Performance” is the execution of a task within day-to-
day life as a metric of what the child “really does” when out on the playground at
their school, at home, or in the community.
Physical Versus Walking Activity
“Physical Activity” (PA) is defined as any body movement that expends energy
(Caspersen et al. 1985) and is quantified by total body movement “counts” via single
to multiple axis accelerometers that are worn at the waist/wrist. Accelerometry has
been employed to document optimal levels of physical activity (total body move-
ment) to support optimal health outcomes. Derived from large population-based
monitoring with accelerometers validated to measure PA, the CDC recommends
60 min/day of moderate-to-vigorous physical activity for children to achieve and
maintain optimal weight and health (HHS 2008).
In contrast, “walking” activity (WA) is defined as when a stride or a step is taken
and is a category or subset of PA. A stride is the phase of gait between two
successive points of foot contact of the same foot. A step is the phase of gait made
up by two successive points of contact of opposite feet. A stride is made up of two
steps, a right and left. Gait cycle strides or steps can be quantified by accelerometry
per unit time (i.e., day, min) or by intensity (rate per unit time) by specifically
designed and validated accelerometers (pedometers) that capture each step or stride
taken (Tudor-Locke et al. 2002). As a subset of overall physical activity and
employing the metric of “walking” via pedometers, there are population-based
daily walking recommendations for elementary-school-aged children. A total of
13,000–15,000 strides/day (26,000–30,000 steps) for boys and 11,000–12,000
strides/day for girls (22,000–24,000 steps) has been proposed to approximate the
60 min/day of moderate-to-vigorous physical activity for optimal growth and health
(Tudor-Locke et al. 2011a).
Thus, depending on the purpose of a particular intervention and rehabilitation
strategy, the goal may be to increase walking activity and/or physical activity of
children with CP. The quantification of “real-world” walking and physical activity of
our patient pre-/post-intervention longitudinally provides essential information to
promote best practices in the clinical management of children with CP. This knowl-
edge includes ecological relevant data about the relative effectiveness of
interventions; it allows comparison to public health standards and it has the potential
to document the progression or decline in walking and physical activity across the
life span. Additionally, such information will allow optimal timing of interventions
and comparison between interventions and clarify the unique factors of each child
with CP that lead to positive functional outcomes.
Methods to Document Walking and Physical Activity: ICF Perspective
Body Structure-Function Level

As previous chapters have described, measurement tools available for the study of
walking or gait cover a broad range of its ICF components. Many laboratory-based
technologies and clinic-based tests quantify altered gait conditions at the body
function and activity capacity levels, providing information about what the body is
doing in a controlled environment. Motion capture systems analyze segmental
kinematic and kinetic patterns of the body and measure the spatial and temporal
characteristics of strides and steps such as length, width, and duration. Dynamic
electromyography (EMG) recordings, which are obtained via skin-contact surface
electrodes or indwelling needle electrodes, highlight potential mechanisms control-
ling body movements and forces. Recordings of dynamic foot pressure are collected
by walking over a mat containing a matrix of multiple sensors that measure the force
acting on each sensor while the foot is in contact with the supporting surface. Such
measurements provide valuable information about the capacity of the foot to accept
and transfer body weight during gait. The metabolic analysis of oxygen uptake and
carbon dioxide production during walking provides an estimate of the energy cost of
walking. Use of these technologies in specialty medicine has led to a sophisticated
understanding of gait pathologies in cerebral palsy, which, in turn, has advanced a
variety of interventions to address altered mechanical alignment and manage muscle
spasticity. Motion analysis data collected pre- and post-intervention provides objec-
tive measurement of changes in walking at the body function and activity capacity
levels, often toward normalization of alignment, propulsion, and efficiency
(Gannotti et al. 2007; Thomas et al. 2004).
For clinicians without access to advanced, laboratory-based kinematic collection
tools, observational rating scales are a low-tech option. Observational gait scales
also provide information at the body function and activity capacity levels and are
intended to provide a quantitative description of gait patterns. The Visual Gait
Assessment Scale, The Edinburgh Visual Gait Scale, and The Observational Gait
Scale have been used in clinical practice to quantify gait deviations in children with
cerebral palsy. The Gait Deviation Index (GDI) is being employed to summarize
overall gait deviations and is derived from the three-dimensional gait analysis
(3DGA). They all demonstrate moderate intra- and inter-rater reliability for charac-
terizing general gait events, such as foot strike or swing phase and for classifying
temporal-spatial parameters such as speed and symmetry (Gor-García-Fogeda et al.
2016). The literature comparing visual rating scales with instrumented motion
capture systems though shows poor validity for evaluating multi-segment three-
dimensional motions (del Pilar Duque Orozco et al. 2016). Although observational
scales can provide a very general snapshot of body function during gait, they are
inadequate for the measurement of detailed kinematic gait deviations.
Activity Capacity Level

Walking speed is the most commonly reported outcome measure of surgical and
rehabilitation interventions intended to improve gait and is often cited as evidence of
improved function. Indeed, the relevance of speed as a powerful indicator of
mobility is widely acknowledged. Walking speed is a simple measurement that
provides fundamental information about both global function and underlying pathol-
ogy. It is valid, reliable, and sensitive, sometimes referred to as the “sixth vital sign”
(Cesari et al. 2005; Fritz and Lusardi 2009; Bohannon 1997). The ability to walk at a
functional speed predicts long-term mobility status in individuals with CP and other
neuro-muscular conditions and is associated with positive health outcomes in adult-
hood (Haak et al. 2009).
Walking speed and temporal-spatial gait parameters of step lengths and support
times can be measured with highly advanced laboratory-based motion capture
systems or with portable lower-cost alternative technologies such as the Mobility
Lab by APDM Inc., Portland, Ore, or the GAITRite from CIR Systems, Sparta,
NJ. These portable systems are used in PT clinics as alternatives to fully
instrumented motion capture laboratories. The Mobility Lab uses sensors attached
to the individual, and the GAITRite is a portable walkway system. Gait speed can
also be measured with very simple clinical methods based on precise distance and
time. The 10-Meter Walk Test (10MWT) is a widely used assessment of gait speed
with established validity and reliability in many adult neurologic conditions, and it is
gaining acceptance as a mobility capacity measure in pediatric settings (Thompson
et al. 2008; Chrysagis et al. 2014). The individual is instructed to walk at a natural
pace over a 10-m distance, and the middle 6 m is timed. The average of three trials is
reported. The 6-Minute Walk Test (6MWT) is an endurance capacity measure in
which the individual is asked to walk at a fast pace to cover as much distance as
possible over 6 min. Originally developed for cardio-pulmonary patients as a field
test of submaximal aerobic capacity, it has not served that purpose for youth with CP
but has been reported as a mobility capacity measure (Slaman et al. 2013; Thompson
et al. 2008).
Walking speed is closely related to global gross motor ability in youth with
cerebral palsy, and the examination of gross motor skills beyond walking provides
more detail about mobility capacity. The Gross Motor Function Measure (GMFM) is
the gold standard observational assessment for reporting gross motor capacity in
children and youth with CP. The GMFM is widely used as a measure of mobility
capacity to monitor a child’s motor development, to evaluate the outcome of
interventions, and to predict future mobility status. It is intended to describe the
status of and evaluate changes in gross motor skills over time. There are 88-item and
66-item versions, each with five domains, including domains of standing and
walking/running/jumping. Multiple authors have found it to be reliable and respon-
sive with good content and construct validity (Ko and Kim 2013; Alotaibi et al.
2014; Beckung et al. 2007; Sullivan et al. 2007; Oeffinger et al. 2008). Its relation-
ship to walking capacity has been demonstrated in studies comparing GMFM scores
to the oxygen cost and energy cost of gait in youth with CP. Kerr found moderate
correlations (r = 0.39, p < 0.001) between O2 cost and the GMFM-66 total score
and strong correlations between O2 cost and dimensions D and E scores of the
GMFM (r = 0.61, p < 0.001) in a group of 184 youth with CP. In a separate review
of 276 youth with CP, GMFM was a powerful correlate of energy cost; for every 1%
increase in GMFM score, there was a 0.12 J/kg/m decrease in energy cost with 42%
of the differences in energy cost explained by differences in GMFM (P < 0.0001)
(Kerr et al. 2008; Kamp et al. 2014).
Another tool for evaluating mobility capacity, the Functional Independence
Measure (FIM), is commonly used in a variety of neuro-motor conditions, including
cerebral palsy (Granger and Hamilton 1993). It is administered through observation,
interview, and/or self-report. The FIM and the WeeFIM, a child version for 1–7 year
olds, consist of a minimal data set of 18 items that measure ability in self-care,
mobility, and cognition. The WeeFIM is a valid measure for tracking disability in
preschool and school-age children with CP. It captures the impact of the develop-
mental condition on independence at home, in school, and in the community which
allows care planning for interventions to enhance outcomes and support families.
Activity Performance Level

At the walking/physical activity performance level, there are some instruments
commonly used for individuals with CP to describe typical performance of gait-
related or mobility-related tasks in day-to-day life. The gold standard classification
scale of gross motor function for individuals with cerebral palsy is the Gross Motor
Functional Classification Scale (GMFCS). It is based on well-informed patient or
proxy report, which categorizes children from age 2 to 18 years into one of five
levels of gross motor function. Classifying a child on the GMFCS is intended to
guide healthcare providers in recommending treatment and equipment and provide
an understanding of how a child’s movement abilities could evolve over time,
including prediction of future walking in a young child. The GMFCS has excellent
inter-rater reliability (0.93) and is stable over time, with an overall test–retest
reliability of 0.79 (Wood and Rosenbaum 2000).
The Functional Mobility Scale (FMS) is another tool used to classify walking
performance. The FMS scores functional mobility on a six-point ordinal scale of
performance over three different distances representing mobility in the home, at
school, and in the wider community. The developers have shown it to be valid and
reliable in a sample of 310 children with cerebral palsy (Graham et al. 2004).
The Gillette Functional Assessment Questionnaire (FAQ) is a self- or proxy-
report measure that includes a 10-level classification of ambulatory function and
22 other locomotor skills rated on a five-point ordinal scale. The FAQ examines
function with the support of walking devices and/or orthoses. It is a valid tool to
evaluate current status as well as change over time or after intervention (Gorton et al.
2011).
The Activities Scale for Kids (ASK) is a questionnaire intended to assess a wide
spectrum of physical abilities in children ages 5–15 years through their self-report. It
has both a capability version (ASK-c) and a performance version (ASK-p). It is a
practical and inexpensive option to measures constructs of motor ability important to
families. ASK items and overall scores have a significant association with related
measures of mobility capacity and performance. As an evaluation tool, it is intended
to help clinicians better understand limitations in mobility and their impact on
participation function (Bagley et al. 2011).
Personal and Environmental Factors

The Lifestyle Assessment Questionnaire (LAQ-CP) is a survey measure that cap-
tures the impact of disability on the lives of children with cerebral palsy and their
families. It examines health-related quality of life and family burden through dimen-
sions of physical independence, mobility, and social integration. The LAQ-CP
physical independence and mobility scores are shown to be related to movement,
manual, and intellectual disability and to be influenced by the presence of
comorbidities. LAQ-CP scores for economic/clinical burden and social integration
though are not closely related to mobility capacity. These dimensions of the LAQ-CP
provide family assessment representing societal/community domains in ICF frame-
work and provide additional explanation of participation habits (Morris et al. 2006).
Monitoring Technology to Capture Walking and Physical Activity

in CP
Devices to capture walking and physical activity were developed in response to the
logistic challenge and intrusiveness of direct observations, the poor reliability of
self-report surveys, and the complexity of heart rate monitoring over the last decades
(Puyau et al. 2002). These devices are basically accelerometers, which are designed
to measure accelerations produced as a body segment or limb part moves in space.
By definition, acceleration is defined as the change in velocity over time of the body
or body part as it moves (i.e., foot during walking). Electric transducers and
microprocessors convert recorded accelerations into digital signals, which are the
“counts” or steps taken. The gold standard for total body movement or physical
activity is indirect calorimetry of total body energy expenditure. Thus
accelerometry-based devices of physical activity have been validated by indirect
calorimetry or calibrated in the metric of resting metabolic equivalents (MET).
Today we have established total body energy cost or MET values for typically
developing children (Puyau et al. 2004; Trost 2007; Trost et al. 2006). In contrast,
accelerometers to capture steps or strides taken are validated against actual observed/
counted strides taken.
Specific to walking as a part of total body physical activity, step, or stride
monitoring is defined as the direct measurement of the amount of walking and/or
steps taken over time (Tudor-Locke and Myers 2001). Within a scoping discussion
of activity measurement, Tyron in 1991 noted that “the step is the preferred unit of
measure, since it is the natural unit of ambulation (Tyron 1991).” Consistent with
Tyron’s observations, today there is a focus on walking activity because of its
ecological clinical relevance and because it can be easily captured within a person’s
day to day life longitudinally. Such information has become the basis for public
health recommendations for optimal daily walking levels for adults and children for
optimal health outcomes (Tudor-Locke et al. 2011a, b).
In contrast to capturing total body physical activity, unique devices are designed
to estimate mileage walked, and/or steps taken over a period of time are often
referred to as “pedometers.” The technology to capture steps taken by pedometers
can be either mechanical spring-loaded mechanical arm (i.e., Yamax SW-200,
Yamax-Digiwalker, 1044 Pulinski Road, Warminster, PA 18974) (Kozey et al.
2010; Yamax-Digiwalker.com 2016) or piezoelectric using inertial sensors/acceler-
ometers in 1, 2, or 3 axis detection (i.e., Omron HF-100, FitBit, StepWatch)
(Healthcare 2016; FitBit 2016; Health 2015). In the mechanical devices, a spring-
suspended horizontal level arm moves up and down as the body part (typically the
hip) moves (i.e., Yamax SW-2000). Piezoelectric devices (accelerometers) have
become quite popular commercially and are worn on various body parts (i.e., FitBit)
and in recent years are embedded in mobile phones (i.e., Virtual Walk Treadmill or
GPS) (Play 2016). Costs for these devices varies ($20–$500) as does their accuracy
and precision to count strides taken depending on the location of attachment to the
body, speed of walking, and/or walking patterns (Bjornson et al. 2011; Foster et al.
2005; Melanson et al. 2004).
As technology has advanced, there are multiple consumer-based devices on the
market that propose to measure walking and/or physical activity (i.e., FitBit Zip)
(FitBit 2016). What is not always clearly apparent is the algorithms employed to
determine the output of these devices (physical activity total body movement
“counts” and/or “strides or steps” taken) as many devices purport to measure both.
A comparison of eight consumer-based devices to measure fitness (PA) in 2014
documented that only one device had consistent measurement to the gold standard of
energy consumption (Lee et al. 2014). When selecting a device to monitor walking
and/or physical activity in children with cerebral palsy, it is essential that the device
has been tested for the key measurement attributes of validity and reliability specif-
ically in children with CP.
Validity addresses whether the device measures what it proposes to measure (i.e.,
total body movement to gold standard of MET or steps counted to actual steps
taken). Reliability speaks to the overall consistency of a device to measure what it is
designed to measure or produce similar outcomes under consistent conditions
(Portney and Watkins 2009). Reliable monitor output data are accurate, consistent
from one session to another when change is not expected, and reproducible. Accu-
racy of a device to monitor steps can be compared to actual steps taken (observed/
videotaped). Consistency and reproducibility would be examined by average num-
ber of steps or strides taken at two time points where the same number of steps or
strides have been taken (i.e., treadmill walking or consistent distance/speed walked).
Devices can be donned or worn at various sites on the body, attached by belts and
carried in pockets and/or special cuffs. The reliability and validity of a device/
monitor should be established for relative age range (children versus adults), popu-
lation (typically developing children versus CP), and body site worn (i.e., wrist,
waist) (Bjornson et al. 2007, 2014; Capio et al. 2012).
Raw data from the accelerometry devices can be summarized by time period of
monitoring from milliseconds up to minutes/hours and across days and weeks
depending on the specific clinical or research question and/or analysis software.
Devices to capture walking and physical activity will provide amount and frequency
of sedentary time behaviors (i.e., walking or not walking or moving or not moving)
which has been extensively examined relative to multiple public health outcomes
(Services and U. S. D. O. H. A. H. 2012). Intensity of walking and/or PA (i.e.,
minutes in moderate-to-vigorous PA, walking a high step or stride rate) can be
described with the outputs from these devices and has similar implications for
overall health and fitness. Cut points for defining intensity levels of physical and
walking activity have been established through monitoring of population-based
samples and/or large clinical/population cohorts (Bjornson et al. 2014; Trost et al.
2005). Patterns of walking and/or PA versus sedentary activity can also be described
with the output from raw step or movement counts. Population-based patterns of
physical activity, sedentary behaviors, and diet have been documented in US ado-
lescents related to obesity levels (Iannotti et el. 2012). Recent work has documented
functional classification accuracy across all five Gross Motor Function Classification
System levels in young children with CP (Keawutan et al. 2014, 2016). In ambula-
tory children with CP, strides spent in moderate-to-high stride rates have been
documented to be associated with higher levels of mobility-based participation in
daily life (Bjornson et al. 2014). Recently, walking activity and intensity was
employed to examine the influence of orthotics on daily walking activity. Bjornson
and colleagues reported the level and intensity of day-to-day walking was not
consistently improved by the current orthotic prescription in a clinical series
(n = 11) (Bjornson et al. 2016), suggesting sensitivity to intervention and change
in walking over time. Thus, clinicians and researchers have multiple approaches to
monitor output analysis to address their clinical or research questions.
StepWatch
The StepWatch (SW) device is a two-dimensional individually calibrated acceler-
ometer that is worn on the outside of the ankle and quantifies when the heel of that
leg leaves the ground or one stride of a single leg during walking. Thus, to compare
the SW data to data from other devices designed to count steps take for each leg, the
average SW counts should be doubled, or alternatively, the “step” count from
another device should be halved. Walking activity of children utilizing the SW
was first reported by McDonald and colleagues in 2005 (McDonald et al. 2005).
Authors reported the StepWatch (SW) monitor activity in 21 boys and 21 girls, ages
7–21 years with average total strides per day at 5,428 strides (10,856 steps)
(McDonald et al. 2005). Employing a two-week sampling period, Song et al. in
2006 reported a total of 6 weeks of daily SW data on 20 youth in the age groups 5–7
and 9–11 years with total strides per day ranging from 4,864 to 14,591 (Song et al.
2006). Bjornson et al. in 2010 documented the walking activity of typically
developing cohort of children ages 2–15 years examining the influence of leg length
on walking activity with the SW (Bjornson et al. 2010). Donning of the SW for this
work was by a knit cuff worn always on the left ankle versus a strap in previous work
by Song and McDonald (McDonald et al. 2005; Song et al. 2006). Decline in stride
frequency and intensity throughout childhood increased with adjustment for leg
length as captured by the SW in this cohort. The SW documented higher stride
counts than published waist mounted pedometer-based pediatric data with the
greatest discrepancy in 4–5 year olds. This discrepancy is proposed to be due to
function of device type, attachment site, and the relative potential undercounting of
steps or strides by waist-mounted devices.
A common device question is how many days of wearing a device is needed in
order to capture a representative sample of their habitual walking activity. Kang and
colleagues, in 2014, examined SW data from a large gender-balanced cohort of
typically developing children (2–15 years) (Kang et al. 2014). Employing a gener-
alizability theory analysis, results suggest that 7 days of data collection was suffi-
cient to achieve the desired reliability level for all but one group; the 10 to 11-year-
old girls (i.e., need 12 days of data collection). Stable estimates of stride or step
counts were achieved with less than 5 days of measurement for the majority of
groups. Interpretation of the data collected by the SW has been examined for “idle”
versus productive walking to decrease any extraneous counts and improve the
accuracy of the output (Wininger and Bjornson 2016). A data-driven analysis of a
large cohort of typically developing children with the SW suggests that a filter or
threshold stride count of ten strides per minute captured 90% of samples in a
weighted average among isolated non-zero stride-count samples offset by inactivity.
This filter for productive versus nonproductive stride data with the SW decreased
strides/day by 8–10% by age group and thus has potential implications for the SW
data pre-/post-interventions clinically and within research protocols.
Excellent accuracy and precision of the SW compared to manual counts of
walking across speeds of treadmill walking and free walking was reported by Foster
et al. as compared to two common belt-worn pedometers for adults (Foster et al.
2005). The StepWatch counted 99.7 0.67% (mean SEM) of the manual counts,
while the Omron pedometer counted 61 3.3%, and the Accusplit pedometer
counted 26 2.8% of the manual counts at 1 mph. At 3 mph, all three monitors
were accurate (>98% of manual counts). The SW monitor was found to have
negligible variance (sd 0.36) over all speeds, whereas the pedometers showed a
large amount of variance at all speeds (sd 4–13). SW accuracy (manual counts of
visual observation of strides to SW count) documented during participant monitor
calibration of free walking and running in typically developing (TD) children
6–14 years of age averaged 97% (sd 3) (Song et al. 2006) and ranged from 99.6 to
100.1 9 (sd 3.7–4.5) for children ages 2–13 years (Bjornson et al. 2014).
The SW has been documented to have superior accuracy for stride counts as
compared with waist-mounted pedometers during treadmill walking in lean and
obese youth ages 10–12 years (Mitre et al. 2009). Excellent accuracy and precision
of the SW was documented for treadmill walking speeds up to 4 mph (ICC = 0.995)
(Bjornson et al. 2010). In a review of pedometers and accelerometers in 2010, David
Bassett and John Dinesh reported that the SW is the most accurate pedometer ever
designed for walking and is capable of capturing actual strides taken to within 3%
for speeds from 1 to 5 mph (Bassett and Dinesh 2010; Karabulut et al. 2005). Thus,
the StepWatch device has one of the highest levels of accuracy for detecting the
movement of stepping across walking speeds of the commercially available monitors
today. This high level of accuracy and ability to be individually calibrated to slow
and/or unusual walking patterns makes the StepWatch ideally suited for populations,
such as cerebral palsy, with evolving pathological gait patterns and/or numerous
interventions focused on walking (i.e., bracing, botulinum toxin, orthopedic/neuro-
logical surgeries). The individualized calibration also makes the StepWatch suitable
for pre-/post-testing of interventions expected to impact walking activity levels and
intensity.
Actigraph
Unlike the Step Watch, the ActiGraph is a portable monitoring device designed to
capture and report total body movement and by extension total body energy expen-
diture. ActiGraph physical activity monitors (ActiGraph, Pensacola, FL, USA) have
been extensively used to measure physical activity in pediatric health research over
the past decade; a review in 2013 identified 273 studies using Actigraph between
2005 and 2010 (Cain et al. 2013). The ActiGraph is a small lightweight sensor most
commonly worn at the waist that provides raw data output in the form of axis activity
counts and vector magnitudes. The first device, the AM7164, developed in the
1980s, is made up of a piezoelectric sensor and an electronic circuit board. It detects
acceleration, converts it to a proportional electric charge, rectifies it to an absolute
acceleration value, and converts signals to activity counts. The GT1M (from the
mid-2000s) replaced the internal sensor with a dual-axis capacitive accelerometer
capable of detecting both static (force of gravity) and dynamic accelerations. The
GTX3 (in 2009) then employed a triaxial capacitive sensor that uses vector magni-
tude data from all three axes to provide inclinometer output (standing, lying, or
sitting). Sampling rates range from 10 to 100 Hz with epoch lengths from 1 to 60 s.
Some studies have shown comparability between models in moderate-to-vigorous
physical activity (Corder et al. 2007; Kozey et al. 2010), while others have shown
differences in overall counts and in sedentary and light activity. Though there is
growing evidence of differences in sensitivity (Rothney et al. 2008; Kozey et al.
2010), it is unclear how model differences affect interpretation. Output from the
newer ActiGraph models includes raw accelerations, vector magnitudes, and incli-
nation (i.e., standing, sitting, etc.). ActiGraph software translates raw outputs into
units of activity counts, steps, or metabolic equivalents (METs) using regression-
based models. Methodological protocols define cut points to quantify PA into
sedentary, light, moderate, and vigorous, although opinions and practices differ
regarding the definition of intensity cut points for various phases of childhood.
Testing in childhood populations demonstrates consistently high instrument
reliability (ICCs = 0.83–0.98), good validity compared with indirect calorimetry
(r = 0.82–0.89), and good agreement with step counts (r = 0.77–0.82) in a variety of
studies, though differences exist in defining count cut points for MET levels (Puyau
et al. 2004, Sirard et al. 2005; Evenson et al. 2008; Wen et al. 2010). In typically
developing children, one study found that wrist-worn placement of the ActiGraph
had good inter-device reliability (r = 0.72) and validity against indirect calorimetry
(r = 0.8, p < 0.01).
Clanchy, in 2011, examined Actigraph validity in 30 youth (age 12.6 2 years)
with CP who walked on a treadmill at a comfortable, brisk, and fast pace after seated
rest. The Actigraph AM7164 had high classification accuracy for sedentary (92%)
and MVPA (91%) and fair classification accuracy for light intensity physical activity
(67%) (Clanchy et al. 2011). These findings established cut points for PA intensity in
children and adolescents with CP. Similarly, Actigraph cut points to interpret PA
intensity have also been validated in toddlers with CP and adults with CP (Claridge
et al. 2015; Oftedal et al. 2014).
In a small study (n = 8) of the reliability and validity of the ActiGraph GT3X in
youth with CP (11.9 3 years), excellent inter-instrument reliability (ICC - .96-.99,
CI [95] 0.81-0.99) was established for the GT3X between left and right waist/hip
placement as well as fair/good (rho = 0.67, p < 0.001) concurrent validity between
count data and VO2. In this small study though, the correlation between step count
and VO2 was not significant (rho = 0.29, p = 0.2) (O’Neil et al. 2014). This study
included only participants classified at GMFCS levels I and II who walked at three
self-selected speeds during a 6-Minute Walk Test (6MWT). Subsequently, in 2016,
O’Neil examined the validity of ActiGraph to distinguish a broader range of
sedentary to light to moderate to vigorous activities in 54 children and youth with
CP, GMFCS levels I, II, and III. The 2016 study found good concurrent validity of
the Actigraph with VO2 (rho = 0.82–0.85, p < 0.001). The ActiGraph exhibited
excellent intra-device reliability for right/left placement. Comparing the measure-
ment of steps/strides taken between three devices, the Actigraph GT3X, the Step
Watch, and the Body Media SenseWear arm band, there was excellent inter-
instrument reliability (ICC_.93–.99). These findings suggest that the Actigraph is
an appropriate choice for the measurement of “real-world” PA or WA and estimation
of overall energy expenditure in youth with CP.
Body Media SenseWear

The SenseWear armband monitor (BodyMedia Inc., Pittsburgh, Pennsylvania) is a
multi-sensor device, worn on the upper arm that combines accelerometry data with
information from several heat-related channels including skin temperature, skin
galvanic response, and heat flux to estimate EE using proprietary algorithms. Studies
to evaluate the validity of SenseWear algorithms versus indirect calorimetry for
estimating energy expenditure (EE) in children have looked at treadmill walking and
a variety of real-world activities. A 2007 study by Andreacci in 33 children
(7–10 years) during treadmill walking examined the accuracy of child-specific
exercise algorithm for the assessment of energy expenditure (EE) using the
SenseWear™ Pro2 armband. Estimation of EE by the SWA was not significantly
different from indirect calorimetry (IC) (Andreacci et al. 2007). A 2016 study by Lee
(Lee et al. 2016) examined 45 TD children (7–13 years) during a subset of 24 various
5-min physical activities. Errors were significantly smaller using a newer versus
older BMSW algorithm (0.25 0.09 kcal min 1 vs. 1.04 0.09 kcal min 1). The
mean absolute percent errors were 17.0 12.1% and 31.4 11.1%. The newer
algorithms yielded nonsignificant ( p > 0.5) differences in EE estimates for walking
and stationary cycling activities (14.5%) (Lee et al. 2016).
Two recent studies examined the validity and reliability of the SenseWear monitor
in youth with CP. They reported fair to good validity between the SenseWear
armband monitor and indirect calorimetry and good inter-instrument reliability
between SenseWear armband monitor left and right arm placement in youth with
CP (Koehler et al. 2015; Ryan et al. 2014).
VitaMove
The VitaMove is a commercially available wireless system (2 M Engineering,
Veldhoven, The Netherlands) with body-fixed, triaxial accelerometers (Freescale
MMA7260Q, Denver, USA) designed specifically to monitor free-living physical
activity in individuals with physical disabilities and specifically those who use
wheeled mobility. The VitaMove activity monitor is the successor of the Vitaport/
Rotterdam Activity Monitor, validated in wheelchair users (Postma et al. 2005). The
VitaMove’s accelerometers are worn at the trunk, wrists, and legs in order to capture
movements in both sitting and standing positions. The VitaMove detects physical
activity intensity, movement patterns, and sedentary time and has demonstrated good
validity for quantifying physical behavior in ambulatory and wheelchair-dependent
adults with daily activity counts correlated with manual scores during periods of
activity (r = 0.506, p = 0.000) (Warms et al. 2008; Bussmann et al. 2009). A study
in 2015 of children with spina bifida (SB) and CP, whose mobility was both
ambulatory and wheeled, revealed good validity of the VitaMove to quantify
physical activity in a non-laboratory setting. For children who are wheelchair-
dependent, the VitaMove consists of one recorder on the sternum and one on each
wrist. In children who are partly ambulatory, additional recorders are attached to
each thigh. In this small sample (n = 12), the absolute percentage error ranged from
25% to +20% when compared to video recordings of actual physical activity
behavior. For wheelchair activities, the mean agreement was 84% (sensitivity = 80%
and specificity = 85%) and for ambulatory activities, the mean agreement was 83%
(sensitivity = 78% and specificity = 90%) (Nooijen et al. 2015). The researchers
suggest that the VitaMove is a reasonable choice to measure total duration of active
time and sedentary time in children with mobility limitations such as CP although
they caution its use in children with severe motor impairment.
Uptimer
The Uptimer (National Aging Research Institute of Melbourne, Melbourne,
Australia) is a remote activity monitor utilizing an inclinometer that has been
validated in adults to measure time spent in the upright position (Diggory et al.
1994). Its simplicity allowed a group of clinical researchers to document duration of
upright activity for children and youth with CP cared for at their institution (Pirpiris
and Graham 2004).
Global Positioning System (GPS)

Over the past 20 years, there has been growing interest in measuring walking and
physical activity relative to the environment as a key factor in the overall health of
populations (McCrorie et al. 2014). The advent of small wearable global positioning
systems (GPS) devices has allowed this work to capture the ecologically based
environmental constraints for constructs such as walking and physical activity, travel
to school, and/or the influence of the build environment on these behaviors. GPS is
able to capture geographical location information whenever a device is within
unobstructed view of four or more GPS satellites above the earth. Wieters et al. in
2012 documented that the GPS combined with geographic location was accurate for
static as well as dynamic movement in the environment (Wieters et al. 2012). Oliver
et al. in 2010 described the combining of GPS, Geographical Information System
(GIS), and accelerometry to describe the location and intensity of travel behaviors
(Oliver et al. 2010). GPS and accelerometry data are each synched by the continuous
time stamp of the accelerometer. The body movement behavior (walking or physical
activity) captured by the accelerometry is then matched to the location (GPS) and
interpreted with geographic location (GIS, i.e., Google Maps) and information (i.e.,
near home, school, sports/recreational facilities, sidewalks, bike paths, etc.).
This combined technology has potential to capture and address multiple metrics
of movement within the natural environment of daily life. For example (Chillón et al.
2015), documented that children walked further to school as they age and described
the distance that best defines active walking to school from non-walking to school as
1,421 m in 10 year olds, 1,627 m in 11 year olds, and 3,046 m in 14 year olds (Chilló
et al. 2015). Recently, triaxial accelerometry and GPS supported the documentation
that primary school-age children achieve the majority of moderate to vigorous
physical activity (MVPA) on streets, playgrounds near schools, or at recreational
facilities (Bürgi et al. 2015). Similarly, Carlson and colleagues documented that
adolescents demonstrate a greater proportion of MVPA at schools with the next
largest site “near home” (Carlson et al. 2016). The efficiency of GPS combined with
accelerometry also supports the consideration of environmental factors in
interpreting changes in activity behaviors with interventions (Demant Klinker
et al. 2015). Thus, GPS combined with accelerometry has potential to accurately
provide environmental, context-specific information about the relative location of
walking or physical activity and offers information about potential environmental
facilitators and/or barriers to activity.
Summary
Over the past decade, a handful of monitoring devices (SW, AG, BMSW, VM, UT,
GPS) have been examined for their validity and reliability to objectively measure
WA and PA in children and youth with CP. Evidence is building for good accuracy
and stability of this small set of WA and PA data acquisition devices. Their ability to
provide accurate valid and reliable data for youth with CP in real-world settings
allows for the identification of population, age, motor classification, regional, and
demographic-specific physical activity and walking activity habits of children and
youth with CP. It allows researchers and clinicians to evaluate changes in PA and WA
with maturation or with interventions for children with CP. Such information can
inform interventions intended to promote functional mobility (walking), PA, and
daily participation in youth with CP. The current body of evidence supports imple-
mentation of these devices with standard protocols for measurement in children and
youth with CP who have functional community ambulation, those classified at
GMFCS levels I and II. There is emerging work demonstrating the utility of the
SW, AG, and VM for children and youth classified at GMFCS level III, and there is a
more urgent need to examine reliability and validity of PA and WA devices for
children and youth with restricted mobility, those classified at GMFCS levels III
and IV.
Beyond the validly of these monitors to measure PA and WA, clinicians and
researchers are interested in using this technology to estimate energy expenditure
and classify activity intensity. Historically, in commercial and research-grade
accelerometry devices, the equations to estimate energy expenditure have been
developed from the general adult population. The output from these equations
cannot be expected to account for either the increased energy cost of locomotion
among people with CP or the variations in metabolic cost among individuals with
CP. These variations are the likely sources of inaccuracies of reported METS from
some devices. Calibration of CP-specific algorithms should improve the ability of
these devices to estimate EE in this population. Recent progress has been made by
Trost et al. who report improved accuracy, by as much as 40%, of Actigraph,
GMFCS I, II, and III-specific cut points to differentiate sedentary activity, LPA,
and MVPA over previously reported CP specific cut points (Trost et al. 2016).
Further work is needed to examine estimation equations and cut points in other
devices and for children and youth classified at GMFCS level IV.
State of the Science: Walking and Physical Activity in Cerebral Palsy

(CP)
Walking Activity with StepWatch

In 2007, Bjornson and colleagues documented the first published community-based
walking activity in 81 adolescents with CP between 10 and 14 years of age
employing the SW (Bjornson et al. 2007). Employing the SW, this cohort was
expanded to a cross-sectional cohort of 209 children with CP ages 2–13 years and
compared to 368 typically developing children (TD) balanced by gender across age
bands (Bjornson et al. 2014). Intense walking in adult US population samples has
been documented at a peak stride/min rate of 101 (Tudor-Locke et al. 2012). In
contrast, intense walking bursts for children (<18 years of age) are documented at
108–146 peak stride/min rates (Barreira et al. 2012; Bjornson et al. 2010). Thus, the
children with CP in Bjornson 2014 cohort were compared for level and intensity of
walking to the pediatric cut points/definitions for low, medium, and high walking
activity based on the SW normative cohort (Bjornson et al. 2010, 2014). As
expected, the children with CP walked significantly less each day and differ by
Gross Motor Function Classification System (GMFCS) levels (I, II, and III). Chil-
dren who are capable of climbing stairs without upper extremity assist (GMFCS
level I) walked at the highest daily levels and intensity while those using hand-held
assistive devices for mobility (GMFCS level III) walked significantly less each day
and rarely reached high stride rates. Relative to walking patterns with the SW, TD
children walk a similar number of strides in low and moderate stride rates each day
while youth with CP do not. The TD children attained high stride rates on average
(460 strides/min) for 8 min/day, while the children with CP at levels I to III spent 4.0,
3.2, and 0.53 min/day respectively.
Daily walking patterns of children with CP differ than typically developing
(TD) children (Bjornson et al. 2011). Employing the daily walking data from
5 days of SW data, 81 children with CP ages 10–13 years were compared to a
cohort of age- and gender-matched TD children. The daily walking trajectories
across increasing stride rates of the children with CP were documented to vary by
GMFCS levels and as expected spent significantly less time daily not walking. The
highest functioning children with CP (GMFCS level I) were not significantly
different in their mean group trajectory or peak stride rates of walking than the TD
children. This is clinically consistent with observations that those youth generally
can keep up with their TD peers. In contrast, youth at GMFCS levels II and III
demonstrated significantly lower trajectories of walking variability and peak stride
rates, with youth using hand-held mobility devices the most limited as compared to
TD children. In 2012, Van Weyl and colleagues examined the influence of day of the
week (school days) on the walking activity of 61 Dutch children with spastic CP
(van Wely et al. 2012). Data suggests that the children walked more during the
weekday versus the weekends. Greater mobility limitations (higher GMFCS), bilat-
eral CP, and age were related to lower walking levels on school days. On the
weekends, bilateral CP, age, and no sport club participation appeared associated
with lower walking activity. Nicholson reported in 2016 that a clinical series of
preoperative ambulatory children with CP averaged fewer steps/day than published
levels (Nicholson et al. 2017).
In 2013, Ishikawa et al. examined the minimum number of days required to
obtain a stable measure of daily walking activity in children with CP (Ishikawa et al.
2013) with the SW. Examining a cohort of 209 children across GMFCS levels I–III,
they report that variance in step counts due to participants walking activity ranged
from 33.6% to 65.4%. Thus, they suggest a minimum of 8, 6, and 2 days were
required to reach acceptable reliability for GMFCS levels I, II, and III, respectively,
for children ages 2–5 years. In contrast, for youth 6–14 years of age, a minimum of
6, 5, and 4 days were required to reach stable estimates of walking activity for
GMFCS levels I, II, and III, respectively (Ishikawa et al. 2013). This work suggests
that the duration of monitoring with the SW should be informed by age and
functional ability (GMFCS) in young children and youth with CP.
The post-monitoring processing of SW data in children with CP can be influenced
by the definition of a “day.” Wilson and colleagues examined two definitions of a
“day” for the inclusion and/or exclusion of walking data with the SW in a clinical
cohort of 61 youth with CP with 2 days of SW data (Wilson et al. 2016). They
compared the application of “day” defined as either number of steps in 24-h period
(>100 steps) or number of hours walking (>10 h) recorded in a 24-h period. Daily
SW counts were lower across 2 days of monitoring when a valid day was defined as
10 h recorded activity per 24-h period and higher when the definition >100 steps
recorded in a 24-h period. They also noted that only 75% of their sample was able to
capture 5 days of monitoring. These results may be influenced by the attachment of
the SW with a strap versus a knit cuff, as has been employed in previous work
(Bjornson et al. 2007, 2014).The definition of the metric of “day” should be clearly
defined and consistent in the processing of raw SW data with the proprietary
software, to allow accurate interpretation pre-/post-intervention and across studies.
The construct validity of SW watch monitoring in day-to-day life was examined
by comparison to gait-lab-based walking speed, a summary score of gait deviation
(Gait Deviation Index, GDI) and distance walked in 6 min (Wilson et al. 2015).
Examining a cohort of 55 youth with CP, primarily at GMFCS level I, the authors
documented a moderate relationship between GDI and SW average strides/day
(r = 0.58) and that as the strides/day increase so does GDI. SW stride activity was
also significantly correlated to lab-based walking speed and walking distance. This
work suggests that interventions that improve gait kinematics (orthopedic surgery,
orthotics) or decrease overall gait deviations (increase in GDI score toward normal)
may enhance community walking activity as well. Replication of this work across
lower functioning ambulatory children and pre-/post-intervention is needed to fur-
ther understand this potential relationship.
StepWatch capture of habitual walking activity has been employed as an outcome
within a randomized intervention testing the effect of a 6-month physical activity
stimulation program in the Netherlands (van Wely et al. 2014). No significant
difference was documented in walking level or intensity between intervention
groups at 6 and 12 months post. Habitual walking activity with and without current
AFO prescription was examined with daily walking activity and intensity with the
SW in 2016 (Bjornson et al. 2016). This work documented no group level condition
effect (AFO ON vs. OFF) for community walking levels for clinically prescribed
AFOs in 11 children with diplegia, across GMFCS levels I–III and various barefoot
walking patterns (equinus, jump gait, and crouch gait). Two participants who did
exhibit improved walking activity levels were prescribed similar AFO prescriptions.
These studies suggest that monitoring with the SW is feasible and has potential to
provide “real-world” information to inform clinical care and research outcomes.
Physical Activity in Children with CP (Actigraph)

A 2012 study utilizing the ActiGraph GT1M examined 23 ambulatory and
non-ambulatory adolescents with CP (13.5, SD 2.6 year) at GMFCS levels I–IV
over a 7-day period. Participants wore the device from 540.5 to 859.2 min per day
and engaged in 89.5 47.1 min of LPA, 17.8 16.9 min of MPA,
12.0 14.4 min of VPA, and 30.7 30.3 min of MVPA. Youth classified at
GMFCS level IV presented with lower levels of LPA, MPA, and MVPA compared
with level I (P < 0.05). Similarly, youth at level III demonstrated lower levels of
MPA and MVPA compared with level I. No differences were seen between levels I
and II for any intensity. MVPA and GMFCS levels were negatively correlated
(minutes/day: τ = 0.65, P < 0.001) (Gorter et al. 2012).
A 2015 study of 102 children (11 2 years) with spastic hemiplegia classified at
GMFCS levels I and II found that only 25% of participated in 60 min of MVPA on at
least one of 4 days. The mean totals for the group were 438 counts/min and 7,541
steps per day and spent 8:36 h (72% of recorded time) in sedentary time, 2:38 h in
light activities (22% of recorded time), and 0:44 h (6% of recorded time) in MVPA.
There were no significant differences in physical activity between children classified
at GMFCS levels I and II. Children were significantly more physically active than
adolescents. Boys were significantly more physically active than girls. More steps
were recorded on weekdays than on weekend days (Mitchell et al. 2015).
Physical Activity in Children with CP Uptimer

The Uptimer is a device validated to measure time spent upright and was used in a
study of 300 children and youth with CP. The authors found that compared to 5.6 h
per day of “uptime” in able-bodied peers, youth with CP described to have hemi-
plegia (n = 115) spent 5.1 h/day in upright time, youth with diplegia (n = 113) spent
2.5 h/day in upright time, and youth with quadriplegia (n = 72) spent 0.5 h/day in
upright time. Analysis of variance revealed significant differences between all
groups (P < 0.001), including the nondisabled comparison group (Pirpiris and
Graham 2004).
Walking Accelerometry Combined with Global Positioning System (GPS)

in Children with CP
A synchronization of SW walking activity and Global Positioning System (GPS)
data has been piloted in 12 ambulatory children with CP (R21 HD 077186) who
underwent 20 sessions of short burst interval treadmill training (SBLTT) (Bjornson,
Moreau, Hurvitz, Kerfeld 2016 unpublished data). SW accelerometry walking activ-
ity data were time matched with GPS records to document walking in the home and
community over a 7-day sample through distance walked, average strides per day,
and percent of time each day spent in low (1–30 strides/min), moderate (31–60
stride/min), and high (>60 stride/min) stride rates. Combined StepWatch/GPS pilot
data demonstrated an increase in the percentage of overall strides per day ambulated
in the community setting from 44.2% to 49.8%, suggesting an increase in commu-
nity walking participation at 6 weeks post-SBLTT training (Fig. 2).
Figure 3 illustrates a StepWatch/GPS a synchronization map for a child with CP
at GMFCS level I over one measurement day of walking activity in the Seattle,
Washington, area. The StepWatch data were classified into intensity levels based on
stride rate. Each GPS location was matched with the walking intensity level at the
same time point, allowing measurement of where walking across these levels
occurred in a spatial context. For this child, the moderate stride activity occurred
at school and a soccer field. This preliminary data provides feasibility of this novel
combined SW/GPS methodology and potential sensitivity for the amount and
location of community strides and intensity levels.
Fig. 2 Percentage of total walking (avg. strides/day) in the community (blue) vs in the home (red)
after 20 sessions of interval treadmill training (n = 12)
Fig. 3 Synchronized
StepWatch/GPS map in
Seattle for a child with CP
(GMFCS I). X = no walking
(nw); Ο = low stride rates;
☐ = moderate stride rates;
Δ = home
Summary
Our knowledge of the real-world physical activity and walking activity habits of
children and youth with CP has expanded over the past decade. The StepWatch, the
Actigraph, and Uptimer have provided data to show that ambulatory youth with CP,
those at GMFCS levels I, II, and III, are more sedentary, spend less time upright,
spend less time being physically active, and take fewer steps per day than TDY.
Studies using these devices have shown that time upright, time active, and amount of
walking are strongly associated with motor ability, as reported by GMFCS classifi-
cation levels. Walking and activity habits have similarities to TDY in that school-age
children with CP take more steps and have higher PA levels than adolescents with
CP. The PA and walking activity habits of school-age children have also revealed
that youth with CP tend to take more steps on school days compared to non-school
days. This recent addition of knowledge about the activity habits of children and
youth with CP can be utilized on a broad scale to tailor PA opportunities and
programs to promote greater activity and participation and can be used on an
individual level in physical therapy and home programming.
We did not identify for this review studies utilizing the BMSW or the VitaMove
monitors to report on PA and WA for children and youth with CP, although we
acknowledge that this is not an exhaustive review.
Monitoring of Walking and Physical Activity: Clinical Implications

for Children CP
Interventions for Walking Activity

The literature suggests that the use of walking and physical activity monitoring in
daily life is clinically feasible and has significant potential to be employed as a
clinical outcome to inform the care and management of children with cerebral palsy.
Community-based monitoring with devices reviewed in this chapter can be
employed within numerous rehabilitation strategies in this population including
but not limited to medications for movement disorders, injection therapy to lower
and upper extremity, upper extremity constraint-induced therapy, orthotic manage-
ment, use of assistive mobility devices, (walker versus crutches) gait training (i.e.,
overground, treadmill), orthopedic surgery, and neurosurgical interventions (intra-
thecal baclofen, selective dorsal rhizotomy) as well as lifestyle intervention to
enhance physical activity. We will review preliminary and published data describing
the application of monitoring devices to describe community-based walking and
physical activity in children with CP.
All children with cerebral palsy (CP) exhibit some type of movement disorder
(i.e., spasticity, dystonia etc.). Often oral medications are employed to decrease the
influence of these movement disorders on activities of daily life such as sitting,
walking, and upper extremity tasks of daily life. A monitoring device (Actigraph on
the wrist) can be employed to document the influence of oral medications on overall
excessive movement in a child with upper extremity dystonia during feeding, for
example. In this clinical scenario, we would hope to see less extraneous limb
movement if the medication is having the desired effect. Similarly, pre-/post-upper
extremity injections to optimize upper extremity use for feeding could employ the
Actigraph during feeding of one arm. Devices (i.e., three-dimensional accelerome-
ters) on both wrists could be used to document the relative change in bilateral upper
extremity use post-constraint-induced therapy treatment.
Walking activity in daily life captured by wearable devices can offer valuable
information to guide interventions which are employed to optimize walking within
the context of the child’s environment. A 2016 clinical pilot study of 11 children with
CP and bilateral impairment employed 2 weeks monitoring with the StepWatch to
examine the influence of current orthotic prescriptions on community walking

(Bjornson et al. 2016). StepWatch data captured strides/day, time walking, as well
as walking intensity. Depending on the walking (SW) outcome examined, only two
to four of the 11 participants demonstrated improvements with their current orthotic
prescription as compared to walking without their orthotics.
Recent work has applied StepWatch monitoring pre- and post-short-burst interval
locomotor treadmill training (SBLTT) in ambulatory school-aged children with CP
(unpublished Bjornson et al. 2016 R21 results). R21 pilot data of pre-/post-SBLTT
(n = 12, Figs. 4 and 5) documented enhanced community walking levels (average
strides/day +948 strides/day, p < 0.001) at 6 weeks post-SBLTT training. Percent
time walking at higher stride/min intensities of medium/high stride (>39 strides/
min) rates increased (+3.8%, p = 0.04) including absolute number of strides/day at
medium/high stride rates (+627, p < 0.001).
Studies of orthopedic surgery outcomes in lab settings for youth with cerebral
palsy (CP) who undergo surgical correction to improve gait reveal positive changes
in body structure/function measures such as kinematic and kinetic patterns (Wren
et al. 2013) and improvement in activity capacity measures such as gait speed
(Gannotti et al. 2007) after surgery. The validation of physical activity
(PA) monitors for youth with CP now allows for investigation of performance
level outcomes such as change in habitual level of walking activity post-surgery.
Lennon and colleagues used the StepWatch to examine the recovery of walking
activity in youth who underwent orthopedic surgery to correct gait (Lennon et al.
2015). Preoperative clinical gait analysis at the Nemours duPont Hospital for
Children includes collection of habitual walking activity pre- and postoperatively.
The surgeons, physiatrists, and physical therapists at the hospital evaluate walking
activity data as part of rehabilitation planning during the child’s recovery from
Fig. 4 Average strides/day at baseline, post-SBLTT, and 6 weeks post-SBLTT (n = 12)

Fig. 5 Percent of total strides/day in low, moderate and high stride rates for 368 typically
developing youth (TDY), (Bjornson et al. 2014) 209 youth with CP and 12 youth with CP, Pre/post
short-burst LTT. Green- high= >60 stride/min, Red-moderate=30-60 stride/min and Blue- low
<30 stride/min
surgery. Preliminary data from 49 youth with CP at GMFCS levels I (8), II (30), and
III (11), with mean age of 11 years, who followed this clinical care pathway, reveals
differences between 30 youth who underwent a low versus 19 youth who underwent
a high burden of surgery. Low burden was defined as soft tissue surgery with or
without a single unilateral osteotomy and high burden as bilateral or multiple
unilateral osteotomies. GMFCS levels were similar between groups. The percent
change from baseline was significantly different at 12 months post-op between the
low (+31%) and high ( 21%) burden groups ( p < 0.05). Mean daily strides were
significantly higher than baseline at 24-m follow-up for the low burden group (3,014
vs. 4,692; p < 0.05) and no different for the high burden group (2,866 vs. 3506;
p > 0.05) (Lennon et al. 2015). Figure 6 shows recovery trends of both groups
relative to expected strides for GMFCS level (Bjornson et al. 2007).
Interventions for Physical Activity

The Carol and Paul Hatfield Cerebral Palsy Sports and Rehabilitation Center at The
St. Louis Children’s Hospital provide children with cerebral palsy and related
childhood disabilities the opportunity to experience sport activities in fun, social
environments. Known as “Camp Independence,” it is a summer program of intensive
physical activities in a sports camp format. Participants are assisted by physical
therapists, nurses, aides, and volunteers to play many different sporting activities.
Campers spend 7 h a day for up to 4 weeks participating in sport activities such as
tennis, swimming, yoga, martial arts, basketball, soccer, cycling, baseball, and
dance. Each sport is adapted as necessary to meet the needs of the child. During
the summer of 2014, Miros and colleagues recruited 34 campers to wear a FitBit Flex
activity tracking device for 2 weeks before camp, while they were attending camp,
and a minimum of 2 weeks after camp (St. Louis Childrens Hospital 2015). All
activity data were collected using the FitBit app and the fitbit.com website (FitBit
Fig. 6 Percentage of the % Strides Relative to Expected for GMFCS Level

average strides/day relative to
100%
expected strides per day by
GMFCS level and surgical
burden. Expected levels based 80%
on published work (Bjornson
et al. 2007)
60%
40%
Surgical Burden (Low)

20%
Surgical Burden (High)
0%
Baseline 6 Month 12 Month 24 Month
2016). Campers also completed a number of standardized physical outcome mea-

sures at the beginning and end of camp to evaluate the functional benefits of camp.
Participants were between the ages of 7 and 18 and ranged in physical ability from
independently being able to run to requiring a power wheelchair for community
mobility. Although there were no significant changes in standardized physical
outcome measures, participants were more active on camp days versus non-camp
days. The participants averaged 8,693 steps on days they attended camp and 5,730
steps on days they did not attend camp. Activity levels decreased after camp, which
families attribute to the lack of community physical activity opportunities accom-
modating children with CP (St. Louis Childrens Hospital 2015).
Active video games (AVGs) are another promising strategy to provide access to
fun, active recreation for youth with CP. O’Neil and colleagues examined PA
intensity levels in 57 youth with CP (mean = 12 years) at GMFCS levels I (28), II
(16), and III (13) during 5 min of play on the X-Box 360 Kinect™ Adventures:
(River Rush and Space Pops) games (O’Neil et al. 2916). Data was collected
utilizing Polar heart rate monitors, the Cosmed K4b2 indirect calorimeter device,
the OMNI rate of perceived exertion (RPE) scale and the Actigraph SW, and BMSW
activity monitors. They found that youth played AVGs at light-to-moderate PA
intensity. Median MET values were River Rush = 3.0, IQR = 2.5–3.8, and Space
Pops = 3.3, IQR = 2.6–3.8. Median OMNI RPE values were River Rush = 3.8,
IQR = 1–4, and Space Pops = 4.0 (IQR = 2–6). Findings suggest that commercial
AVGs may provide opportunities to promote PA in ambulatory youth with CP but
are limited in their ability to promote higher levels of PA intensity and longer bouts
of PA (O’Neil et al. 2916, 2016).
Researchers and entrepreneurs are now working together to develop custom
gaming platforms to meet the needs of youth with physical disabilities. A game
that allows physical therapists to change the speed and intensity of the game through
a back-end portal allows the therapist to structure game conditions to reach the heart
rate and level of physical activity intensity individuals most benefit from. At the
same time, the game system can collect data on a player’s progress over a session,
several sessions, or several weeks. The flexibility of gaming parameters allows the
therapist to adjust the game to address therapeutic goals. Having the players wear
heart rate monitors and activity monitors during game play allows the therapist to see
if the child is reaching the aerobic and activity levels that are health promoting. To
date, O’Neil’s pilot study has enrolled 12 children and youth to play three newly
developed 20-min games (O’Neil et al. 2016, unpublished work). Youth have
enjoyed playing, which is an important factor. Having a fun way to keep kids active
has the potential to mitigate.
In 2014, van Wely and colleagues conducted a randomized trial examining the
effect of 6-month physical activity stimulation program (van Wely et al. 2010, 2014).
The investigators employed the StepWatch to capture habitual walking activity. No
significant difference was documented in walking level or intensity between inter-
vention groups at 6 and 12 months post. A randomized control trial of an internet-
based physical activity intervention was completed by Maher and colleagues in 2010
(Maher et al. 2010). Employing a waist-mounted pedometer (NL1000), no signifi-
cant differences between intervention and comparison group were documented for
number of steps taken per week or self-reported physical activity at 10 and
20 weeks post.
Considerations and Limitations of Walking and Physical Activity

Monitoring
When selecting a monitor or device to capture WA or PA, there are several factors to
consider. First, it is important to select a device that is validated and most directly
related to the behavior you are aiming to capture and/or influence with intervention.
For example, for walking activity post-lower extremity surgical intervention, you
may choose the StepWatch, while physical activity after a home-based AVG pro-
gram maybe most appropriately measured with the Actigraph. Second, the costs of
these devices vary from $30 to $500 per device, and this may or may not include the
processing software to access meaningful information to interpret. Third, the ease of
processing and getting to outputs that can be readily interpreted should be considered
as well. The addition of bluetooth versions (wireless/cloud-based) of devices will
now make downloading relatively easier (i.e., StepWatch and Actigraph). Lastly, in a
clinical setting, billing for the assessment of walking and/or physical activity can be
included within a clinical gait or mobility evaluation and/or a post-intervention
outcome assessment physical therapy charge by a physical therapist.
The emerging data on WA and PA in children and youth with CP reveals high
variability among many of the study samples. Thus, it is important that this vari-
ability is taken into account in developing study designs/sample sizes to insure that
work is adequately powered to capture clinically meaningful as well as statistical
significance. For WA activity in children with CP, a potential “minimal clinical
important difference” (MCID) benchmark could be the mean difference between
walking activity between GMFCS levels. For example, enroll a sample size adequate
to capture the mean difference between WA of children with CP at GMFCS level I

versus II (based on current published WA levels) (Bjornson et al. 2014).
PA monitors have potential weaknesses that should be considered with their
implementation as well. The accuracy and precision of instruments worn on the
waist or upper arm can be limited for certain types of upright behaviors that have a
low ambulatory component and that may involve upper body work. Similarly WA
devices worn on the wrist may not adequately capture the abnormal walking patterns
of children with CP and/or after interventions (i.e., orthotics, orthopedic surgery).
Recent efforts to enhance the ability of physical activity monitors to capture these
behaviors include more densely sampled data and more sophisticated prediction
equations. Information about the location or purpose of individual activities is
limited, unless information from other sources is integrated with information from
the monitor. Technological solutions synchronizing GPS with accelerometers previ-
ously employed in populations without physical disability are being examined in
children with CP (Bjornson et al. 2016, unpublished data).
Conclusions/Summary
In summary, accelerometers have been validated and reliably employed to document

walking and physical activity of children with cerebral palsy across functional levels.
Relative to walking activity, published information to date offers reference informa-
tion for walking activity levels, walking patterns, and intensity of walking by
GMFCS levels as compared to children without motor limitations with the
StepWatch. Physical activity has been documented for ambulatory and
non-ambulatory children with CP with several devices (i.e., Actigraph, BodyMedia,
etc.) and validated to energy cost in ambulatory children. Cut points for interpreta-
tion of PA intensity with the Actigraph in children with CP of varying ages are
emerging in the literature. Based on the body of knowledge reviewed in this chapter,
accelerometry can be effectively employed to capture walking and physical activity
in children and youth with CP. This knowledge should inform clinical and/or
research questions relative to performance by functional levels, documenting natural
history and directing rehabilitation strategies as ecologically based outcomes of WA
and PA within the context of daily life. Future development of WA and PA moni-
toring in children with CP should focus on the capturing upper extremity functional
activities, expand the use of bluetooth technology and/or cloud-based capture of data
and synchronization accelerometry with GPS and patient/parent reported qualitative
outcomes.
Cross-References

▶ Assessing Clubfoot and Cerebral Palsy by Pedobarography
▶ Clinical Gait Assessment by Video Observation and 2D Techniques

Analysis
▶ Natural History of Cerebral Palsy and Outcome Assessment
▶ Oxygen Consumption in Cerebral Palsy
References
Alotaibi M, Long T, Kennedy E, Bavishi S (2014) The efficacy of GMFM-88 and GMFM-66 to
detect changes in gross motor function in children with cerebral palsy (CP): a literature review.
Disabil Rehabil 36:617–627
Andreacci JL, Dixon C, Dube JJ, McConnell TR (2007) Validation of SenseWear Pro2 armband to
assess energy expenditure during treadmill exercise in children 7–10 years of age. J Exerc
Physiol Online 10:35–42
Bagley AM, Gorton GE, Bjornson K, Bevans K, Stout JL, Narayanan U, Tucker CA (2011) Factor-
and item-level analyses of the 38-item activities scale for kids–performance. Dev Med Child
Neurol 53:161–166
Barreira TV, Katzmarzyk PT, Johnson WD, Tudor-Locke C (2012) Cadence patterns and peak
cadence in US children and adolescents: NHANES, 2005–2006. Med Sci Sports Exerc
44:1721–1727
Bassett DR, Dinesh J (2010) Use of pedometers and accelerometers in clinical populations: validity
and reliability issues. Phys Ther Rev 15(3):135–142
Beckung E, Hagberg G (2002) Neuroimpairments, activity limitations, and participation restrictions
in children with cerebral palsy. Dev Med Child Neurol 44:309–316
Beckung E, Carlsson G, Carlsdotter S, Uvebrant P (2007) The natural history of gross motor
development in children with cerebral palsy aged 1 to 15 years. Dev Med Child Neurol
49:751–756
van den Berg-Emons HJ, Saris WH, de Barbanson DC, Westerterp KR, Huson A, van Baak MA
(1995) Daily physical activity of school children with spastic diplegia and of healthy control
subjects. J Pediatr 127:578–584
Bjornson KF, Belza B, Kartin D, Logsdon R, McLaughlin JF (2007) Ambulatory physical activity
performance in youth with cerebral palsy and youth who are developing typically. Phys Ther
87:248–257
Bjornson K, Song K, Lisle J, Robinson S, Killien E, Barrett T, Zhou C (2010) Measurement of
walking activity throughout childhood: influence of leg length. Pediatr Exerc Sci 22:581–595
Bjornson KF, Song K, Zhou C, Coleman K, Myaing M, Robinson SL (2011) Walking stride rate
patterns in children and youth. Pediatr Phys Ther 23:354–363
Bjornson KF, Zhou C, Stevenson RD, Christakis D (2014a) Relation of stride activity and
participation in mobility-based life habits among children with cerebral palsy. Arch Phys Med
Rehabil 95:360–368
Bjornson K, Zhou C, Stevenson RD, Christakis D, Song K (2014b) Walking activity patterns in
youth with cerebral palsy and youth developing typically. Disabil Rehabil 36:1279–1284
Bjornson K, Zhou C, Fatone S, Orendurff M, Stevenson R, Rashid S (2016) The effect of ankle-foot
orthoses on community-based walking in cerebral palsy: a clinical pilot study. Pediatr Phys Ther
28:179–186
Bohannon R (1997) Comfortable and maximum walking speed of adults aged 20–79 years:
reference values and determinants. Age Ageing 26:15–19
Bürgi R, Tomatis L, Murer K, de Bruin ED (2015) Localization of physical activity in primary
school children using accelerometry and global positioning system. PLoS One 10:e0142223
Bussmann JBJ, Ebner-Priemer U, Fahrenberg J (2009) Ambulatory activity monitoring: progress in
measurement of activity, posture, and specific motion patterns in daily life. Eur Psychol
14:142–152
Cain K, Salis JF, Conway TL, van Dyck D, Calhoon L (2013) Using accelerometers in youth
physical activity studies: a review of methods. J Phys Act Health 10:437–450
Capio CM, Sit CHP, Abernethy B, Masters RSW (2012) Fundamental movement skills and
physical activity among children with and without cerebral palsy. Res Dev Disabil
33:1235–1241
Carlson JA, Schipperijn J, Kerr J, Saelens BE, Natarajan L, Frank LD, Glanz K, Conway TL,
Chapman JE, Cain KL, Sallis JF (2016) Locations of physical activity as assessed by GPS in
young adolescents. Pediatrics 137:e20152430
Caspersen CJ, Powell KE, Christenson GM (1985) Physical activity, exercise, and physical fitness:
definitions and distinctions for health-related research. Public Health Report 100:126–131
Cesari M, Kritchevsky SB, Penninx BWHJ, Nicklas BJ, Simonsick EM, Newman AB, Tylavsky
FA, Brach JS, Satterfield S, Bauer DC, Visser M, Rubin SM, Harris TB, Pahor M (2005)
Prognostic value of usual gait speed in well-functioning older people – results from the health,
aging and body composition study. J Am Geriatr Soc 53:1675–1680
Chillón P, Panter J, Corder K, Jones AP, van Sluijs EMF (2015) A longitudinal study of the distance
that young people walk to school. Health Place 31:133–137
Chrysagis N, Skordilis EK, Koutsouki D (2014) Validity and clinical utility of functional assess-
ments in children with cerebral palsy. Arch Phys Med Rehabil 95:369–374
Clanchy K, Tweedy S, Boyd R, Trost S (2011) Validity of accelerometry in ambulatory children and
adolescents with cerebral palsy. European J Appl Physiol 111(12):2951–9. doi: 10.1007/
s00421-011-1915-2.
Claridge EA, McPhee PG, Timmons BW, Martin Ginis KA, Macdonald MJ, Gorter JW (2015)
Quantification of physical activity and sedentary time in adults with cerebral palsy. Med Sci
Sports Exerc 47:1719–1726
Corder K, Brage S, Ramachandran A, Snehalatha C, Wareham N, Ekelund U (2007) Comparison of
two Actigraph models for assessing free-living physical activity in Indian adolescents. J Sports
Sci 25:1607–1611
Del Pilar Duque Orozco M, Abousamra O, Church C, Lennon N, Henley J, Rogers KJ, Sees JP,
Connor J, Miller F (2016) Reliability and validity of Edinburgh visual gait score as an evaluation
tool for children with cerebral palsy. Gait Posture 49:14–18
Demant Klinker C, Schipperijn J, Toftager M, Kerr J, Troelsen J (2015) When cities move children:
development of a new methodology to assess context-specific physical activity behaviour
among children and adolescents using accelerometers and GPS. Health Place 31:90–99
Diggory P, Gorman M, Schwarz J, Helme R (1994) An automatic device to measure time spent
upright. Clin Rehabil 8:353–357
Evenson KR, Catellier DJ, Gill K, Ondrak KS, McMurray RG (2008) Calibration of two objective
measures of physical activity for children. J Sports Sci 26:1557–1565
FITBIT (2016) Fitbit Products [Online]. Available: https://www.fitbit.com/
Foster RC, Lanningham-Foster LM, Manohar C, McCrady SK, Nysse LJ, Kaufman KR, Padgett
DJ, Levine JA (2005) Precision and accuracy of ankle-worn accelerometer-based pedometer in
step counting and energy expenditure. Prev Med 41:778–783
32:46–49
Gannotti ME, Gorton GEI, Nahorniak MT, Masso PD, Landry B, Lyman J, Sawicki R, Hagedorn K,
Ross E, Warner J (2007) Postoperative gait velocity and mean knee flexion in stance of
ambulatory children with spastic diplegia four years or more after multilevel surgery. J Pediatr
Orthop 27:451–456
Gor-García-Fogeda MD, Cano de la Cuerda R, Carratalá Tejada M, Alguacil-Diego IM, Molina-
Rueda F (2016) Observational gait assessments in people with neurological disorders: a
systematic review. Arch Phys Med Rehabil 97:131–140
Gorter JW, Noorduyn SG, Obeid J, Timmons BW (2012) Accelerometry: a feasible method to
quantify physical activity in ambulatory and nonambulatory adolescents with cerebral palsy. Int
J Pediatr 2012:6
Gorton GE III, Stout JL, Bagley AM, Bevans K, Novacheck TF, Tucker CA (2011) Gillette
functional assessment questionnaire 22-item skill set: factor and Rasch analyses. Dev Med
Graham HK, Harvey A, Rodda J, Nattrass GR, Pirpiris M (2004) The functional mobility scale
(FMS). J Pediatr Orthop 24:514–520
Granger CV, Hamilton BB (1993) The uniform data system for medical rehabilitation report of first
admissions for 1991. Am J Phys Med Rehabil 72:33–38
Haak P, Lenski M, Hidecker MJC, Li M, Paneth N (2009) Cerebral palsy and aging. Dev Med Child
Neurol 51:16–23
Health M (2015) StepWatch [Online]. Available: https://modushealth.com/. Accessed 17 Nov 2015
Healthcare O (2016) Omron Pedometers [Online]. Available: https://omronhealthcare.com/fitness/
activity-trackers-and-pedometers/
Imms C, Reilly S, Carlin J, Dodd K (2008) Diversity of participation in children with cerebral palsy.
Dev Med Child Neurol 50:363–369
Iannotti RJ, Wang J (2013) Patterns of physical activity, sedentary behavior, and diet in U.S.
adolescents. J Adolescent Health 53(2):280–6. doi: http://dx.doi.org/10.1016/j.
jadohealth.2013.03.007.
Ishikawa S, Kang M, Bjornson KF, Song K (2013) Reliably measuring ambulatory activity levels of
children and adolescents with cerebral palsy. Arch Phys Med Rehabil 94:132–137
Kamp FA, Lennon N, Holmes L, Dallmeijer AJ, Henley J, Miller F (2014) Energy cost of walking
in children with spastic cerebral palsy: relationship with age, body composition and mobility
capacity. Gait Posture 40:209–214
Kang M, Bjornson KF, Barreira TV, Ragan BG, Song K (2014) The minimum number of days
required to establish reliable physical activity estimates in children aged 2–15 years. Physiol
Meas 35:2229
Karabulut M, Crouter S, Bassett DR (2005) Comparison of two waist-mounted and two ankle-
mounted electronic pedometers. Eur J Appl Physiol 96:334–335
Keawutan P, Bell K, Davies PSW, Boyd RN (2014) Systematic review of the relationship between
habitual physical activity and motor capacity in children with cerebral palsy. Res Dev Disabil
35:1301–1309
Keawutan P, Bell KL, Oftedal S, Davies PSW, Boyd RN (2016) Validation of accelerometer
cut-points in children with cerebral palsy aged 4 to 5 years. Pediatr Phys Ther 28:427–434
Kerr C, Parkes J, Stevenson M, Cosgrove AP, McDowell BC (2008) Energy efficiency in gait,
activity, participation and health status in children wtih cerebral palsy. Dev Med Child Neurol
50:204–210
Ko J, Kim M (2013) Reliability and responsiveness of the gross motor function measure-88 in
children with cerebral palsy. Phys Ther 93:393–400
Koehler K, Abel T, Wallmann-Sperlich B, Dreuscher A, Anneken V (2015) Energy expenditure in
adolescents with cerebral palsy: comparison of the SenseWear armband and indirect calorimetry.
J Phys Act Health 12:540–545
Kozey SL, Staudenmayer JW, Troiano RP, Freedson PS (2010) Comparison of the ActiGraph 7164
and the ActiGraph GT1M during self-paced locomotion. Med Sci Sports Exerc 42:971–976
Lee J-M, Kim Y, Welk GJ (2014) Validity of consumer-based physical activity monitors. Med Sci
Sports Exerc 46:1840–1848
Lee J-M, Kim Y, Bai Y, Gaesser GA, Welk GJ (2016) Validation of the SenseWear mini armband in
children during semi-structure activity settings. J Sci Med Sport 19:41–45
Lennon N, Hulbert R, Church C, Miller F (2015) Surgical burden and recovery of walking
performance in youth with cerebral palsy. Dev Med Child Neurol 57:96–97
Maher CA, Williams MT, Olds TIM, Lane AE (2010) An internet-based physical activity interven-
tion for adolescents with cerebral palsy: a randomized controlled trial. Dev Med Child Neurol
52:448–455
McCrorie PRW, Fenton C, Ellaway A (2014) Combining GPS, GIS, and accelerometry to explore
the physical activity and environment relationship in children and young people – a review. Int J
Behav Nutr Phys Act 11:93
McDonald CM, Widman L, Abresch T, Walsh S, Walsh D (2005) Utility of a step activity monitor
for the measurement of daily ambulatory activity in children. Arch Phys Med Rehabil
86:793–801
Melanson EL, Knoll JR, Bell ML, Donahoo WT, Hill JO, Nysse LJ, Lanningham-Foster L, Peters
JC, Levine JA (2004) Commercially available pedometers: considerations for accurate step
counting. Prev Med 39:361–368
Mitchell LE, Ziviani J, Boyd RN (2015) Habitual physical activity of independently ambulant
children and adolescents with cerebral palsy: are they doing enough? Phys Ther 95:202
Mitre N, Lanningham-Foster L, Foster R, Levine JA (2009) Pedometer accuracy in children: can we
recommend them for our obese population? Pediatrics 123:e127–e131
Morris C, Kurinczuk JJ, Fitzpatrick R, Rosenbaum PL (2006) Do the abilities of children with
cerebral palsy explain their activities and participation? Dev Med Child Neurol 48:954–961.
Dev Med Child Neurol. 2007 Feb;49(2):122
Nicholson K, Lennon N, Hulbert R, Church C, Miller F (2017) Pre-operative walking activity in
youth with cerebral palsy. Res Dev Disabil 60:77–82
Nooijen CF, de Groot JF, Stam HJ, van den Berg-Emons RJ, Bussmann HB (2015) Validation of an
activity monitor for children who are partly or completely wheelchair-dependent. J Neuroeng
Rehabil 12:11
O’Neil ME, Fragala-Pinkham MA, Forman JL, Trost SG (2014) Measuring reliability and validity
of the ActiGraph GT3X accelerometer for children with cerebral palsy: a feasibility study. J
Pediatr Rehabil Med 7:233–240
O’Neil ME, Fragala-Pinkham M, Lennon N, George A, Forman J, Trost SG (2016) Reliability and
validity of objective measures of physical activity in youth with cerebral palsy who are
ambulatory. Phys Ther 96:37–45
O’Neil ME, Fragala-Pinkham M, Lennon N, Trost SG (2916) Active video games to promote
physical activity in youth with cerebral palsy. In: IV STEP 2016. Columbus: Pediatric and
Neurology Sections, American Physical Therapy Association
Oeffinger D, Bagley A, Rogers S, Gorton G, Kryscio R, Abel M, Damiano D, Barnes D, Tylkowski
C (2008) Outcome tools used for ambulatory children with cerebral palsy: responsiveness and
minimum clinically important differences. Dev Med Child Neurol 50:918–925
Oftedal S, Bell KL, Davies PSW, Ware RS, Boyd RN (2014) Validation of accelerometer cut points
in toddlers with and without cerebral palsy. Med Sci Sports Exerc 46:1808–1815
Oliver M, Badland H, Mavoa S, Duncan MJ, Duncan S (2010) Combining GPS, GIS, and
accelerometry: methodological issues in the assessment of location and intensity of travel
behaviors. J Phys Act Health 7:102–108
Palisano R, Rosenbaum P, Bartlett D, Livingston MH (2007) GMFCS- R & E gross motor function
classification system expanded and revised, 2007. CanChild Centre for Childhood Disability,
McMasters University, Hamilton
Pirpiris M, Graham HK (2004) Uptime in children with cerebral palsy. J Pediatr Orthop 24:521–528
Play G (2016) Virtual walk treadmill or GPS [Online]. Available: https://play.google.com/store/
apps/details?id=com.virtualwalk.android&hl=en
Portney LW, Watkins MP (2009) Foundations of clinical research: applications to practice. Pearson/
Prentice Hall, Upper Saddle River
Postma K, van den Berg-Emons HJG, Bussmann JBJ, Sluis TAR, Bergen MP, Stam HJ (2005)
Validity of the detection of wheelchair propulsion as measured with an activity monitor in
patients with spinal cord injury. Spinal Cord 43:550–557
Puyau MR, Adolph AL, Vohra FA, Butte NF (2002) Validation and calibration of physical activity
monitors in children. Obes Res 10:150–157
Puyau MR, Adolph AL, Vohra FA, Zakeri I, Butte NF (2004) Prediction of activity energy
expenditure using accelerometers in children. Med Sci Sports Exerc 36:1625–1631
Rothney MP, Apker GA, Song Y, Chen KY (2008) Comparing the performance of three generations
of ActiGraph accelerometers. J Appl Physiol 105:1091–1097
Ryan JM, Walsh M, Gormley J (2014) A comparison of three accelerometry-based devices for
estimating energy expenditure in adults and children with cerebral palsy. J Neuroeng Rehabil
11:116
Services, U. S. D. O. H. A. H. (2012) Physical activity guidelines for Americans: fact sheet for
professionals [Online]. Available: http://www.health.gov/paguidelines/factsheetprof.aspx.
Accessed 06 Feb 2013
Sirard JR, Trost SG, Pfeiffer KA, Dowda M, Pate RR (2005) Calibration and evaluation of an
objective measure of physical activity in preschool children. J Phys Act Health 3:345–357
Slaman J, Dallmeijer A, Stam H, Russchen H, Roebroeck M, van den Berg-Emons R (2013) The
six-minute walk test cannot predict peak cardiopulmonary fitness in ambulatory adolescents and
young adults with cerebral palsy. Arch Phys Med Rehabil 94:2227–2233
Song KM, Bjornson KF, Capello T, Coleman K (2006) Use of the StepWatch activity monitor for
characterization of normal activity levels in children. J Pediatr Orthop 26:245–249
St Louis Childrens Hospital (2015) Fitbit fun: does Camp Independence increase activity of
children with cerebral palsy? Pediatr Perspect [Online], Fall Available: http://www.
stlouischildrens.org/sites/default/files/health_professionals/images/PedPerspective_
Fall2015web.pdf. Accessed 11 Dec 2016
Sullivan E, Barnes D, Linton JL, Calmes J, Damiano DL, Oeffinger D, Abel M, Bagley A,
Gorton G, Nicholson D, Rogers S, Tylkowski C (2007) Relationship among functional outcome
measures used for assessing children with ambulatory CP. Dev Med Child Neurol 49:338–344
Thomas SS, Buckon CE, Piatt JH, Aiona MD, Sussman MD (2004) A 2-year follow-up of
outcomes following orthopedic surgery or selective dorsal rhizotomy in children with spastic
diplegia. J Pediatr Orthop B 13:358–366
Thompson P, Beath T, Bell J, Jacobson G, Phair RSNM, Wright FV (2008) Test-retest reliability of
the 10-metre fast walk test and 6 minute walk test in ambulatory school-aged children with
cerebral palsy. Dev Med Child Neurol 50:370–376
Trost SG (2007) State of the art reviews: measurement of physical activity in children and
adolescents. Am J Lifestyle Med 1:299–314
Trost SG, McIver KL, Pate RR (2005) Conducting accelerometer-based activity assessments in
field-based research. Med Sci Sports Exerc 37:S531–S543
Trost SG, Way R, Okely AD (2006) Predictive validity of three Actigraph energy expenditure
equations for children. Med Sci Sports Exerc 38:380–387
Trost SG, Fragala-Pinkham M, Lennon N, O’Neil ME (2016) Decision trees for detection of activity
intensity in youth with cerebral palsy. Med Sci Sports Exerc 48:958–966
Tudor-Locke CE, Myers AM (2001) Methodological considerations for researchers and practi-
tioners using pedometers to measure physical (ambulatory) activity. Res Q Exerc Sports
72:1–12
Tudor-Locke C, Williams JE, Reis JP, Pluto D (2002) Utility of pedometers for assessing physical
activity: convergent validity. [Review] [49 refs] Sports Med 32:795–808
Tudor-Locke C, Craig C, Beets M, Belton S, Cardon G, Duncan S, Hatano Y, Lubans D, Olds T,
Raustorp A, Rowe D, Spence J, Tanaka S, Blair S (2011a) How many steps/day are enough? For
children and adolescents. Int J Behav Nutr Phys Act 8:78
Tudor-Locke C, Craig CL, Brown WJ, Clemes SA, de Cocker K, Giles-Corti B, Hatano Y, Inoue S,
Matsudo SM, Mutrie N, Oppert J-M, Rowe DA, Schmidt MD, Schofield GM, Spence JC,
Teixeira PJ, Tully MA, Blair SN (2011b) How many steps/day are enough? For adults. Int J
Behav Nutr Phys Act 8:79–79
Tudor-Locke C, Brashear M, Katzmarzyk PT, Johnson WD (2012) Peak stepping cadence in free-
living adults: 2005–2006 NHANES. J Phys Act Health 9:1125–1129
Tyron WW (1991) Activity measurement in psychology and medicine. Plenum Press, New York
U.S. Dept of Health and Human Services (2008) Physical activity guidelines for Americans
summary [Online]. Available: http://www.health.gov/PAGuidelines/. Accessed 30 Nov 2016
Voorman JM, Dallmeijer AJ, Schuengel C, Knol DL, Lankhorst G, Becher JG (2006) Activities and
participation of 9- to 13-year-old children with cerebral palsy. Clin Rehabil 20:937–948
Warms CA, Whitney JD, Belza B (2008) Measurement and description of physical activity in adult
manual wheelchair users†. Disabil Health J 1:236–244
van Wely L, Becher J, Reinders-Messelink H, Lindeman E, Verschuren O, Verheijden J, Dallmeijer
A (2010) LEARN 2 MOVE 7-12 years: a randomized controlled trial on the effects of a physical
activity stimulation program in children with cerebral palsy. BMC Pediatr 10:77
van Wely L, Becher JG, Balemans ACJ, Dallmeijer AJ (2012) Ambulatory activity of children with
cerebral palsy: which characteristics are important? Dev Med Child Neurol 54:436–442
van Wely L, Balemans ACJ, Becher JG, Dallmeijer AJ (2014) Physical activity stimulation program
for children with cerebral palsy did not improve physical activity: a randomised trial. J
Physiother 60:40–49
Wen LM, van der Ploeg HP, Kite J, Cashmore A, Rissel C (2010) A validation study of assessing
physical activity and sedentary behavior in children aged 3 to 5 years. Pediatr Exerc Sci
2:408–420
Wieters KM, Kim J, Lee C (2012) Assessment of wearable global positioning system units for
physical activity research. J Phys Act Health 9:913–923
Wilson NC, Signal N, Naude Y, Taylor D, Stott NS (2015) Gait deviation index correlates with daily
step activity in children with cerebral palsy. Arch Phys Med Rehabil 96:1924–1927
Wilson NC, Mudge S, Stott NS (2016) Variability of total step activity in children with cerebral
palsy: influence of definition of a day on participant retention within the study. BMC Res Notes
9:411
Wininger M, Bjornson K (2016) Filtering for productive activity changes outcomes in step-based
monitoring among children. Physiol Meas 37(12):2231
Wood E, Rosenbaum P (2000) The gross motor function classification system for cerebral palsy: a
study of reliability and stability over time. Dev Med Child Neurol 42:292–296
World Health Organization (2002) International classification of functioning, disability and health
(ICF). World Health Organization, Geneva
Wren TAL, Lening C, Rethlefsen SA, Kay RM (2013) Impact of gait analysis on correction of
excessive hip internal rotation in ambulatory children with cerebral palsy: a randomized
controlled trial. Dev Med Child Neurol 55:919–925
YAMAX-DIGIWALKER.COM (2016) Yamax-Digiwalker [Online]. Warminster. Available:
https://www.yamax-digiwalker.com/
Idiopathic Toe Walking
Karen Davies, Lise Leveille, and Christine Alvarez
Abstract
Idiopathic toe walking in children is a diagnosis of exclusion. Toe walkers require
investigation with a thorough clinical examination, observational gait assessment,
and potential additional review to exclude an underlying diagnosis. Three-
dimensional instrumented gait analysis is helpful to characterize the observed
gait pattern and identify children that may benefit from intervention. It is most
useful in the child with reduced range of motion at the ankle who has a toe-toe
gait pattern but appears to have a heel strike on observational gait analysis. These
children often compensate with either hyperextension of the knee in mid-stance,
external foot progression, and/or an early heel rise, frequently missed with
observational gait analysis. In this clinical scenario, instrumented three-
dimensional gait analysis enables characterization of the toe walking pattern
and identification of objective intervention outcomes.
Keywords
Idiopathic toe walking • Gait analysis • Ankle kinematics • Ankle kinetics •
Electromyography • Pedobarography
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Instrumented Gait Analysis in Idiopathic Toe Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
K. Davies (*)
Shriners Gait Lab, Sunny Hill Health Centre for Children, Vancouver, BC, Canada
e-mail: krdavies@cw.bc.ca
L. Leveille • C. Alvarez
Shriners Gait Lab, Sunny Hill Health Centre for Children, Vancouver, BC, Canada
British Columbia Children’s Hospital, Vancouver, BC, Canada
e-mail: lleveille@cw.bc.ca; calvarez@cw.bc.ca

DOI 10.1007/978-3-319-30808-1_60-1
2 K. Davies et al.
Typical Function of the Foot and Ankle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Function of the Foot and Ankle in Idiopathic Toe Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gait Accommodations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Pedobarography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Electromyography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Temporal Spatial Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Severity Classification of Idiopathic Toe Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Clinical Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Case Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Case Study No. 1: Mild (Type 1) Idiopathic Toe Walking (Fig. 4) . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Case Study No. 2: Moderate (Type 2) Idiopathic Toe Walking (Fig. 5) . . . . . . . . . . . . . . . . . . . . . 11
Case Study No. 3: Severe (Type 3) Idiopathic Toe Walking (Fig. 6) . . . . . . . . . . . . . . . . . . . . . . . . 14
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Introduction
Idiopathic toe walking (ITW) is a condition of childhood characterized by a bilateral

toe-toe gait pattern of unknown cause. In typically developing gait, heel strike at
initial contact is consistently present between 3 and 50 weeks following the emer-
gence of independent ambulation (Sutherland et al. 1980; Burnett and Johnson
1971). Persistent toe walking is considered abnormal in typically developing chil-
dren after 2–3 years of age (Engström and Tedroff 2012; Sobel et al. 1997). ITW is a
diagnosis of exclusion when other musculoskeletal and neurological pathology have
been ruled out. Potential differential diagnoses include cerebral palsy, autistic
spectrum disorders, muscular dystrophy, peripheral neuropathy, neuromuscular dis-
orders, spinal cord abnormalities, talipes equinovarus, dystonia, ankylosing spon-
dylitis, leg length discrepancy, venous malformation or tumor in the gastrocnemius
muscle, schizophrenia, and trauma (Hicks et al. 1988; Shulman et al. 1997; Le Cras
et al. 2011; Engelbert et al. 2011).
The reported prevalence of ITW in all children ranges between 4.9% and 24%
with an estimate of 2.1% by the age of 5.5 years (Engström and Tedroff 2012; Sobel
et al. 1997; Accardo et al. 1992). ITW is reported more frequently in males at an
occurrence rate of 55–68% (Engström and Tedroff 2012; Fox et al. 2006; Eastwood
et al. 2000; Stricker and Angulo 1998). Furthermore, a positive family history is
reported in 10–40% of both males and females (Engström and Tedroff 2012; Sobel
et al. 1997; Fox et al. 2006; Stricker and Angulo 1998; Pomarino et al. 2016). Family
history studies suggest a genetic component may play a role in the etiology of toe
walking. A relationship between ITW and language delays, learning disabilities,
prematurity, persistent elements of the tonic labyrinthine reflex, and immature motor
control has also been described (Accardo et al. 1992; Stricker and Angulo 1998; Fox
et al. 2006; Accardo and Barrow 2015; Clark et al. 2010). Further, challenges with
motor skills and sensory processing as well as diminished thresholds for vibration
perception have been linked with ITW (Williams et al. 2012, 2013), potentially
indicating the presence of subtle neurological changes in children with ITW. Toe
Idiopathic Toe Walking 3
walking has been reported in autism and pervasive developmental disorders with an
incidence as high as 19–20% (Ming et al. 2007; Barrow et al. 2011). While these
more recent studies suggest subtle neurological abnormalities are associated with
ITW, the original study describing ITW proposed that it was due to a congenital short
tendo calcaneus (Hall et al. 1967). However, older and exclusive toe walkers show
more restricted ankle dorsiflexion than younger or intermittent toe walkers, alterna-
tively suggesting that prolonged toe walking leads to diminished range of motion
(ROM) as opposed to limited ROM leading to ITW (Furrer and Deonna 1982; Sobel
et al. 1997; Stricker and Angulo 1998). Increased time spent toe walking may also
explain the high prevalence of Type 1 muscle fibers found in the gastrocnemius of
children with ITW histologically (Eastwood et al. 1997). Despite these established
associations with ITW, a distinct etiology remains unclear.
Children with ITW may experience pain in their legs or feet, diminished balance,
recurrent tripping or falling, difficulty with footwear, and reduced passive ankle
dorsiflexion (DF) ROM (Sobel et al. 1997; Engelbert et al. 2011; Fox et al. 2006;
Clark et al. 2010; Hirsch and Wagner 2004). Diminished ankle DF has been linked to
injuries or pathology of the ankle, forefoot, midfoot, and/or hindfoot (Tabrizi et al.
2000; DiGiovanni et al. 2002; Hill 1995). The reported pathology and clinical
manifestations associated with ITW are considered impairments in body structures
and body functions from the International Classification of Functioning, Disability
and Health for Children and Youth (ICF-CY) frame of reference (World Health
Organization 2007), requiring appropriate assessment and recommendations for
management. Parental and peer pressure to walk with a typical heel-toe gait pattern
may lead to accommodations in gait that are imperceptible to observational gait or
video analysis alone, thereby indicating the usefulness of an instrumented three-
dimensional (3D) gait analysis.
State of the Art
The role of clinical gait analysis in children with idiopathic toe walking is to identify
the condition and describe the necessary accommodations and impairments affecting
their gait. It is based on objective instrumented measurement of individual joints or
segments and the interpretation of the biomechanics and pathology of these mea-
surements (Baker 2013; Pierz and Õunpuu 2013). Prior to clinical gait analysis,
published evidence-based care guidelines are recommended for obtaining history;
assessment of pain, speech and language development, and sensory processing
function; physical examination; gross motor skill assessment; and observational
gait analysis (Le Cras et al. 2011). Physical examination determines active and
passive ROM, muscle strength, muscle tone, rotational alignment, and selective
motor control. Imaging results are also used to rule out suspected underlying
neurological and orthopedic causes of toe walking. An understanding of typical
gait is required to identify and interpret the atypical gait patterns observed in
idiopathic toe walking.
4 K. Davies et al.
Instrumented Gait Analysis in Idiopathic Toe Walking
Typical Function of the Foot and Ankle
In typical gait, weight bearing takes place on a stable foot, muscles absorb and
generate power adequately, the knee and ankle joints progress in the appropriate
plane, and there is sufficient ROM at the ankle. Ankle ROM approximates an arc
from neutral to 10 of dorsiflexion (DF) and neutral to 20 of plantar flexion (PF) in
mature heel-toe gait (Perry 1992). Typical passive ankle DF is 54 at birth and
decreases to approximately 41 passive ankle DF by the time children are learning to
walk and to 8–25 by young adulthood (Walker 1991; Grimston et al. 1993; Bovens
et al. 1990). Ankle position contributes to stability in stance, foot clearance,
pre-positioning the foot in swing, appropriate step length, and energy conservation –
all considered determinants of normal gait (Perry 1985; Gage 2004). Normal gait is
further divided into periods of initial contact, loading response, mid-stance, terminal
stance, pre-swing, mid-swing, and terminal swing (Perry 1992). Perry (1992) also
describes ankle and foot movement during stance in terms of three ankle rockers: the
first (heel) rocker occurs from initial contact to flat foot weight bearing with controlled
PF, the second (ankle) rocker takes place when the tibia moves forward over the foot in
stance, and the third (forefoot) rocker happens when the foot advances into PF for
push-off. The first rocker starts with initial contact, continuing to about 8% of the gait
cycle. It overlaps somewhat with the second rocker, which occurs from approximately
5–45% of the gait cycle. The third rocker starts with heel lift at approximately 30% of
the gait cycle, overlapping with the second rocker and continuing until the end of
stance (Perry 1992). Typical ankle kinetics demonstrate an early ankle DF moment at
loading response with a transition to a late PF moment corresponding with the third
rocker and toe off.
Function of the Foot and Ankle in Idiopathic Toe Walking
Clinically, absence of normal ankle motion during gait presents as forefoot or flat
foot weight acceptance at initial contact with an early heel rise prior to mid-stance.
Children with ITW demonstrate an absent first rocker and/or shortened loading
response (Crenna et al. 2005). Following initial contact, there is progressive ankle
DF instead of the controlled PF typically observed (Hicks et al. 1988). Diminished
ankle DF ROM limits the forward progression of the tibia over the foot resulting in a
reversal of the second rocker and early transition from progressive DF into progres-
sive PF in mid-stance (Hicks et al. 1988). This can also be described as an early heel
rise or an early third rocker, which is a premature progression into PF at or before
30% of the gait cycle (Alvarez et al. 2007). It is the atypical timing of peak ankle DF
during stance that differentiates ITW from other gait pathology.
Swing phase ankle kinematics are also helpful in distinguishing ITW from other
foot pathology such as a contralateral drop foot. Idiopathic toe walkers typically
present with an early excursion toward ankle DF from early to mid-swing followed
Ankle Moment
AM1
1.6 AM2
Flex-Ext
0.5
–0.5
0.0 50.0 100.0
% Gait Cycle
Fig. 1 Representative sagittal plane internal ankle moment. Graphical illustration demonstrating a
predominant internal ankle plantar flexion moment in early stance (AM1) and smaller peak plantar
flexion moment in late stance (AM2). The gray band represents sagittal ankle kinetics based on
normative data from 34 healthy children aged 5 to 18 years within 1 SD of the mean
by PF or diminished peak DF in mid- to late swing, leading to equinus

pre-positioning by initial contact (Kelly et al. 1997; Westberry et al. 2008;
McMulkin et al. 2016). In contrast, children with mild spastic diplegia typically
demonstrate progressive excursion toward DF throughout swing phase (Kelly et al.
1997).
Ankle kinetics in ITW demonstrate an early internal ankle moment resulting from
a premature heel rise or early third rocker. Alvarez et al. (2007) label the early
predominant internal ankle moment seen in ITW as the first ankle moment (AM1), to
delineate the peak PF moment in early stance from the typical peak PF moment seen
in late stance (AM2) (Fig. 1). The early first internal ankle moment (AM1) may be
larger or smaller than the typical late peak ankle PF moment (AM2). This is
alternatively described as a “double bump” ankle PF moment pattern (Stott et al.
2004; Hemo et al. 2006), indicating an early increased internal PF moment.
It is important to note that children with ITW often have significant variability in
their gait patterns, spontaneously alternating between bilateral toe-toe and heel-toe
gait patterns with variable ankle motion (Hicks et al. 1988; Crenna et al. 2005;
Westberry et al. 2008). In contrast, children with spastic diplegia tend to have more
repeatable gait patterns with less deviation in ankle motion (Hicks et al. 1988). This
subtle finding may be useful to differentiate these patient populations.
Gait Accommodations
Gait deviations commonly seen with ITW include an increased anterior pelvic tilt,
mild knee hyperextension in late stance, and external foot progression (out-toeing)
(Hicks et al. 1988; Westberry et al. 2008; Stott et al. 2004; McMulkin et al. 2006).
These gait deviations are accommodations for the abnormal kinematics and kinetics
seen at the ankle. It has been postulated that increased PF in stance due to toe
walking may cause the center of body mass to shift forward over the base of support,
6 K. Davies et al.
increasing the anterior pelvic tilt (McMulkin et al. 2016). The mid-stance knee
hyperextension commonly observed in toe walkers results from limited ankle DF
in stance. Restricted motion at the ankle limits forward tibial progression during the
second rocker resulting in knee hyperextension. The limited forward tibial progres-
sion with continued forward femoral progression over the tibia may further contrib-
ute to stance phase knee hyperextension in ITW (Hicks et al. 1988).
Limited ankle DF with maximum knee extension leads to further gait accommo-
dation with increased external foot progression to shorten the lever arm and avoid the
forward tibial progression typically occurring during the second rocker (Hicks et al.
1988; McMulkin et al. 2016). It is unknown whether the increased external foot
progression angle is due to increased hip external rotation, increased external torsion
of the tibia, or a combination of both. It has been suggested that this increase in
external foot progression is due to increased gastrocnemius tightness; however, static
clinical ankle DF ROM is not correlated with peak ankle DF in stance (Stott et al.
2004; McMulkin et al. 2016). Five-year postsurgical intervention follow-up data
(n = 8) also showed an increase in external foot progression angle; however, the
overall mean was not significant (McMulkin et al. 2016). Surgical procedures in this
follow-up study were either a gastrocnemius/soleus recession or tendo-Achilles
lengthening (TAL). It is currently unclear whether increased external foot progres-
sion in ITW is a true gait accommodation for gastrocnemius tightness or a habitual
pattern.
Pedobarography
Pedobarography is used to demonstrate abnormal foot loading and center of pressure

progression. Bowen et al. (1998) and Alvarez et al. (2008) have described loading
patterns in the feet of typically developing children. Restricting the assessment to the
dynamic phase of stance and dividing the foot into medial and lateral forefoot,
midfoot, and hindfoot simplifies the evaluation of foot pressures to six segments.
Consideration of pressure location, timing, and time spent in each of the six
segments provides descriptive information of the loading experience of each foot.
In children who exclusively walk with a toe-toe pattern, the pedobarography report is
entirely forefoot loading with reasonably symmetrical medial and lateral pressure
distribution (Fig. 2). Less severe toe walkers can be identified by minimal pressure
on the heel segments and reduced time spent on the heel when compared to normal
(Fig. 3). These toe walkers will quickly unload the hindfoot, equivalent to an early
heel rise, spending the majority of stance phase on the forefoot. Pressure at initial
contact in the most subtle toe walkers is seen simultaneously on the fore, midfoot,
and hindfoot, or a reversal of loading where the loading is initiated in the forefoot but
rapidly transmitted to the hindfoot and then back to the forefoot. Lastly, accommo-
dation to toe walking can be seen as increased external foot progression resulting in
increased medial forefoot pressures on pedobarography.
Fig. 2 Plantar pressure

profile of severe toe walker
demonstrating forefoot
loading
Fig. 3 Plantar pressure

profile of less severe toe
walker demonstrating
minimal pressure on heel
segments
8 K. Davies et al.
Electromyography
Dynamic electromyography (EMG) can be applied by surface or indwelling elec-

trodes to assess muscle activity and contraction patterns throughout the gait cycle
and is useful for detecting neuromuscular abnormalities. EMG data in children with
ITW demonstrates abnormal co-contraction and out-of-phase muscle activity with
premature firing of gastrocnemius in swing as well as low-amplitude tibialis anterior
firing throughout stance and swing (Griffin et al. 1977). Brunt et al. (2004) showed
that early swing phase gastrocnemius activity and early and restricted duration of
tibialis anterior activity commonly result in flatfoot or forefoot weight bearing at
initial contact. Children with ITW, equinus deformities, and cerebral palsy all
demonstrated comparable EMG findings, differentiating them from a matched
control group of children with typical gait walking on their toes voluntarily (Kalen
et al. 1986). Therefore, EMG is not necessarily useful for differentiating between
causes of toe walking. Additionally, EMG studies have shown large variability and
may not be conclusive in providing objective diagnoses due to inconsistent findings
(Papariello and Skinner 1985; Kalen et al. 1986).
Temporal Spatial Characteristics
Few studies describe temporal spatial characteristics in ITW. McMulkin et al. (2006)
found the velocity in 14 children with ITW (mean age 8.9 years, range
5.6–12.6 years) was comparable to a normative database of 44 healthy subjects;
however, the toe walkers had a significant increase in cadence and diminished stride
length. In contrast, Westberry et al. (2008) studied 51 children with ITW (mean age
9.3 years, range 6.0–18.0 years) and found diminished velocity secondary to a
decrease in stride length and decreased cadence when compared to typically devel-
oping children.
Severity Classification of Idiopathic Toe Walking
A classification framework has been developed based on a cohort of 133 children

with a history of toe walking to quantitatively describe ITW, differentiate it from
other pathology, and delineate severity gradation (Alvarez et al. 2007). In this
classification framework, toe walkers are classified into three severity groups
(mild, moderate, or severe) based on the absence or presence of a first rocker, an
early third rocker, and an early predominant internal ankle moment. These criteria
were developed from preliminary work addressing gait adaptations of the contralat-
eral limb in children with hemiplegic cerebral palsy (Sawatzky et al. 1999). As
described previously, Alvarez et al. (2007) refer to the early predominant ankle
moment as the AM1, to delineate the peak PF moment in early stance from the peak
PF moment typically seen in late stance (AM2) (Fig. 1). Mild (Type 1) toe walking is
defined by presence of a first rocker, typical third rocker, and typical ankle PF
moment. Refer to Case Study No.1 in Case Examples. In moderate (Type 2) toe
walkers, there may or may not be a first rocker and typical third rocker. Kinetics
distinguish moderate toe walkers by the presence or absence of an early ankle
moment (AM1) which is lower than the normally timed but diminished second
moment (AM2). Refer to Case Study No. 2 in Case Examples. Severe (Type 3) toe
walking demonstrates no first rocker, premature third rocker timing, and an early
predominant ankle moment (AM1 greater than AM2). Refer to Case Study No. 3 in
Case Examples. Alvarez et al. (2007) did not find clinically significant differences in
knee position between the three levels of toe walking severity.
Clinical Perspective
Idiopathic toe walking in children causes considerable consternation with parents,

teachers, and coaches. Initially the concern is in regard to a potentially undiagnosed
condition in the child. Given the multitude of possible underlying diagnoses that
present in part with equinus gait, the toe walker cannot be set aside as idiopathic until
proven otherwise. This becomes increasingly important as the child ages into
adolescence. Children who have toe walked since initiation of walking, presenting
with a normal birth history, normal lower extremity tone, no weakness, and no
known significant past medical history, are commonly idiopathic toe walkers.
However, a child who started walking with a heel-toe gait and later developed a
toe-toe pattern is a far more concerning scenario and requires the appropriate
assessments and consultations with orthopedic surgery and neurology.
Once a diagnosis of idiopathic toe walking is made, concern from the family is
often alleviated. Considering the severity of toe walking in the clinical setting is
important to determine if further assessment by 3D gait analysis is necessary.
Observational gait analysis and ROM assessment may delineate which children
require further gait evaluation. With respect to visual gait observation – can the
child produce a normal heel strike without knee hyperextension or excessive external
rotation? Can the child walk exclusively on their heels elevating the midfoot and
forefoot off the ground? What degree of motion does the child have at the ankle? Do
they have passive ankle DF greater than 10 with knee extension? Le Cras et al.
(2011) recommend referral to the appropriate specialist if a child is unable to achieve
10 of passive ankle DF ROM with knee extension in spite of previous therapeutic
intervention. A child who cannot heel walk and has ROM at the ankle below 5 PF is
at one end of the spectrum, and at the other end is the child who has the ability to
walk on their heels and has ROM in excess of 10 DF with the knee extended. It is
often the moderate toe walker in between these two ends of the spectrum who needs
further delineation by instrumented gait analysis. The intervention for a child who is
either a severe toe walker or a moderate toe walker with significant compensations at
other joint levels is worth exploring.
There are limited long-term studies evaluating interventions for ITW. Through
systematic review of the ITW literature, precursory evidence from a limited number
of randomized controlled trials and a large number of non-randomized retrospective
10 K. Davies et al.
cohort studies suggest that serial casting and surgical treatment have short-term
improvements in passive ankle DF, gait kinematics and kinetics, and toe walking
(van Bemmel et al. 2014; van Kuijk et al. 2014). Despite the fact that passive ankle
DF improved more in those treated surgically versus by casting, there were no
significant differences between treatment type with respect to reports of persistent
toe walking (van Bemmel et al. 2014; van Kuijk et al. 2014). A more recent cohort
study, without a concurrent control group, evaluated gait outcomes in 8 subjects
5 years postoperatively and found significant improvements in pelvic tilt, mean peak
ankle DF during stance and swing, internal ankle moments, and ankle power
(McMulkin et al. 2016). Reports from adult foot and ankle orthopedic surgeons
describing reconstructive or salvage orthopedic procedures including a gastrocne-
mius recession or open TAL are also limited (Cychosz et al. 2015).
Although persistent concerns of parents and patients regarding clumsiness and
bullying around severe toe walkers exist, it is difficult to determine if these com-
plaints are sufficient to consider them as indications for intervention. Previous gait
studies of children with idiopathic clubfeet have shown that a TAL accounts for 27%
loss of gastrocnemius/soleus muscle strength and subsequent 23% decrease in PF
power (Karol et al. 1997). Given the long-term sequelae of untreated equinus
deformity is unknown in otherwise typically developing children, a TAL may be
an excessive intervention, particularly in the growing child. However, children with
severe equinus have limited potential to generate forward propulsive energy through
dynamic PF at the ankle due to loss of ankle ROM. Power generation at the ankle is
the product of the mass moment of inertia and angular acceleration of the anatomical
segment. Therefore, normal power generation requires normal PF power (ankle
moment) and normal ankle ROM (angular acceleration). In a child or adolescent
with severe equinus, a TAL may be warranted to restore normal ankle ROM and
improve power – yet one must be cautious not to overlengthen the tendo-Achilles.
An overlengthened tendo-Achilles will result in loss of PF strength and a decreased
internal ankle moment, leading to decreased power generation at the ankle.
The premise of being a toe walker is still unknown, but it is an engrained gait
behavior pattern. To attempt elimination of the pattern, treatment must take advan-
tage of neural plasticity theory and give the child the opportunity to relearn how to
walk. First, given a physical abnormality exists, in terms of a tight gastrocnemius/
soleus muscle complex, treatment needs to achieve a correction such that the
necessary ROM is possible. Secondly, retraining the gait pattern by preventing
recurrent equinus is pivotal. Maintenance of normal heel to toe loading and properly
timed and loaded rockers is crucial. Achievement of correction can be done with
serial casting, casting following Botulinum Toxin, casting following a gastrocne-
mius/soleus recession, casting following a percutaneous TAL, or casting following
an open TAL. A randomized controlled trial by Engström et al. (2013), however, has
shown that the addition of Botulinum Toxin type A prior to casting did not improve
passive ankle DF ROM, ankle DF strength, or gait outcomes of cast-only treatment
at 1 year. Further clinical care guidelines for conservative management of ITW are
outlined by Le Cras et al. (2011). Care must be taken to expose the child to the least
possible risks for complications and emotional trauma. Maintenance is achieved
with the use of articulated ankle foot orthoses that allow DF and block PF. Initially
this is used to retrain the child’s gait pattern, and then the orthoses can be weaned.
Reassessment using instrumented gait analysis is important, and in the experience of
the authors rewarding, as often the abnormalities approximate normal studies 1 year
posttreatment.
Case Examples
Case Study No. 1: Mild (Type 1) Idiopathic Toe Walking (Fig. 4)
This is a 6-year-old male with ITW. On physical examination he has mild gastroc-

nemius tightness with 10 and 5 ankle DF in knee extension on the right and left
side, respectively. On video observation, he demonstrates a variable weight accep-
tance pattern with occasional forefoot weight acceptance and inconsistent early heel
rise bilaterally. Mild external foot progression angle is also noted. Ankle kinematics
are nearly normal (Fig. 4a). There is a decreased, but present, first rocker on the right,
with normal second rocker and third rocker formation bilaterally. There is normal
ankle DF in swing bilaterally. Internal ankle moments demonstrate a normal AM2
bilaterally without evidence of an AM1 (Fig. 4b). Sagittal ankle joint powers are
normal on the left and mildly decreased on the right (Fig. 4c). This child is
accommodating with a mild external foot progression angle that is still within normal
range (Fig. 4d). The foot-floor graph demonstrates the angle of the foot with respect
to the floor during stance phase. As defined by the authors, normal heel rise occurs
after the contralateral leg passes, indicated by the vertical lines in the graphs
(Fig. 4e). This child has a normal foot-floor angle with no evidence of early heel
rise. At the knee, this child demonstrates near normal kinematics with peak knee
extension seen at initial contact (Fig. 4f). However, there is intermittent knee
hyperextension in late stance on the left, which is a sign of adjacent level gait
accommodation for a tight gastrocnemius. This case is an excellent example of the
variability in gait pattern that is commonly seen in children with ITW. In some trials
this child demonstrates normal gait kinematics and kinetics, while in other trials
there is evidence of a tendency for early heel rise and the associated gait accommo-
dations discussed earlier, including increased external foot progression angle and
knee hyperextension in mid- to late stance phase.
Case Study No. 2: Moderate (Type 2) Idiopathic Toe Walking (Fig. 5)
This is a 6-year-old male with ITW. On physical examination he has moderate

gastrocnemius tightness with 0 and 5 ankle DF in knee extension on the right
and left side, respectively. On video observation, he demonstrates a variable weight
acceptance pattern with early heel rise bilaterally. Mild external foot progression
angle is present, right greater than left. Instrumented 3D gait analysis confirmed
these findings. Ankle kinematics show an absent first rocker, abbreviated second
12 K. Davies et al.
a Ankle Joint Angles (deg) (Right) Ankle Joint Angles (deg) (Left)
14.5 13.9
Plant-Dors
Plant-Dors
–5.3 –5.3
–25.1 –24.5
0.0 50.0 100.0 0.0 50.0 100.0
b Ankle Joint Moment (Nm/kg) (Right) Ankle Joint Moment (Nm/kg) (Left)
1.5 1.4
Dors-Plant
Dors-Plant
0.5 0.5
–0.4 –0.5
0.0 50.0 100.0 0.0 50.0 100.0
c Sagittal Joint Powers (Watts/kg) (Right) Sagittal Joint Powers (Watts/kg) (Left)
3.5 3.1
Ankle (Abs-Gen)
Ankle (Abs-Gen)
1.3 1.1
–0.9 –0.8
0.0 50.0 100.0 0.0 50.0 100.0
d Foot Progression e Foot-Floor Angle (Rep)

70.8
Toes Up-Down
2.6
Ext-Int
–14.4 23.0
–31.5 –24.9
0.0 50.0 100.0 0.0 50.0 100.0
% Gait Cycle % Stance Phase
f Knee Joint Angles (deg) Knee Joint Angles (deg)

65.7 62.7
Ext-Flex
Ext-Flex
29.9 25.6
–6.0 –11.5
0.0 50.0 100.0 0.0 50.0 100.0
Fig. 4 Case study no. 1: mild (Type 1) idiopathic toe walker. (a) Ankle kinematics. (b) Internal
ankle moments (kinetics). (c) Ankle sagittal joint power (kinetics). (d) Representative foot progres-
sion angle. (e) Representative foot-floor angle. (f) Knee kinematics. The gray bands represent
a Ankle Angle (Rep) b Foot-Floor Angle (Rep)

14.5 73.0
Toes Up-Down
Plant-Dors
–8.7 24.1
–31.9 –24.9
0.0 50.0 100.0 0.0 50.0 100.0
c Ankle Moment (Rep) d 3.5

Ankle Power (Rep)
1.5
Abs-Gen
Flex-Ext
1.2
0.5
–0.5 –1.2
0.0 50.0 100.0 0.0 50.0 100.0
e Knee Flexion - Extension (Rep) f Foot Progression

65.7
2.6
Ext-Flex
Ext-Int
29.9 –21.2
–6.0 –45.0
0.0 50.0 100.0 0.0 50.0 100.0
g EMG Gastrocnemius (Right)
0.25
Gastrocnemius
0.13
0.00
0.0 50.0 100.0
% Gait Cycle
Fig. 5 Case study no. 2: moderate (Type 2) idiopathic toe walker. (a) Representative ankle
kinematics. (b) Representative foot-floor angle. (c) Representative internal ankle moments (kinet-
ics). (d) Representative ankle sagittal joint power (kinetics). (e) Representative knee kinematics. (f)
Representative foot progression angle. (g) Example of electromyography. The gray band represents
rocker, and normal third ankle rocker formation. There is decreased ankle DF in
swing bilaterally (Fig. 5a). Early heel rise is seen bilaterally with increased foot-floor
angle initiated prior to 30% of the gait cycle (Fig. 5b). Internal ankle moments
demonstrate the presence of an AM1 bilaterally, which are of equal or lesser value
than AM2 (Fig. 5c). Sagittal ankle joint powers demonstrate early power generation
corresponding to the early heel rise and decreased terminal stance phase power
generation (Fig. 5d). Gait accommodations seen in this child include knee hyperex-
tension in late stance (Fig. 5e) and external foot progression angle (Fig. 5f). EMG
shows preparatory gastrocnemius activation in late swing phase contributing to the
14 K. Davies et al.
preparatory positioning of the foot in equinus, in addition to activation throughout

stance (Fig. 5g).
Case Study No. 3: Severe (Type 3) Idiopathic Toe Walking (Fig. 6)
This is a 10-year-old male with severe ITW. On physical examination he has

gastrocnemius tightness with 10 and 20 ankle DF in knee extension on the
right and left side, respectively. On video observation, he demonstrates a toe-toe
weight acceptance pattern with brief heel contact and early heel rise bilaterally.
There is an external foot progression angle bilaterally. Instrumented 3D gait analysis
confirmed these findings. Ankle kinematics show an absent first and second rocker
and decreased third rocker bilaterally. He is in ankle PF throughout stance phase and
has decreased ankle DF in swing (Fig. 6a). The foot-floor angle also demonstrates PF
throughout stance and early heel rise bilaterally occurring in early stance phase
(Fig. 6b). Internal ankle moments demonstrate a predominant AM1 that is greater
than AM2 bilaterally (Fig. 6c). Sagittal ankle joint powers show early power
generation corresponding to the early heel rise and decreased terminal stance
phase power generation (Fig. 6d). Gait accommodations seen in this child include
knee hyperextension in late stance (Fig. 6e) and external foot progression angle, left
greater than right (Fig. 6f). EMG shows preparatory gastrocnemius activation in late
swing phase and activation throughout stance corresponding with early heel rise and
terminal stance push-off (Fig. 6g). In comparison to the previous example of Type
2 ITW, this child shows increased severity with respect to his gastrocnemius
contracture, predominant first ankle moment, and more severe external foot progres-
sion angle.
Summary
Idiopathic toe walking is a condition of childhood characterized by a bilateral toe-toe

gait pattern of unknown cause. Persistent toe walking is considered abnormal in
typically developing children after 2–3 years of age and is a diagnosis of exclusion.
Other neurologic and orthopedic conditions such as cerebral palsy, muscular dys-
trophy, peripheral neuropathy, and spinal cord abnormalities must be excluded prior
to obtaining a diagnosis of ITW. The role of clinical gait analysis in children with
ITW is to identify the condition and describe the necessary accommodations and
impairments affecting their gait. It can be used to classify the severity of their
condition and help identify children most likely to benefit from treatment. Children
with a long-standing history of ITW typically present with decreased passive ankle
DF ROM. Ankle kinematics often show a variable pattern with absent first rocker,
early peak ankle DF in stance, and early heel rise occurring prior to 30% of the gait
cycle. Swing phase ankle kinematics show normal ankle DF in early swing with
decreased DF in late swing and poor ankle pre-positioning for initial foot contact.
Ankle kinetics show an early internal ankle moment (AM1) and a diminished late
Ankle Angle (Rep) Foot-Floor Angle (Rep)

a b
14.5 72.8
Toes Up-Down
Plant-Dors
–9.6 24.0
–33.7 –24.9
0.0 50.0 100.0 0.0 50.0 100.0
c Ankle Moment (Rep) d Ankle Power (Rep)
1.6 3.5
Abs-Gen
Flex-Ext
0.5 0.4
–0.5 –2.7
0.0 50.0 100.0 0.0 50.0 100.0
e Knee Flexion - Extension (Rep)

f Foot Progression
65.7
2.6
Ext-Flex
Ext-Int
29.2 –19.9
–7.2 –42.5
0.0 50.0 100.0
0.0 50.0 100.0
% Gait Cycle
% Gait Cycle
g EMG Gastrocnemius (Left)
0.25
Gastrocnemius
0.13
0.00
0.0 50.0 100.0
% Gait Cycle
Fig. 6 Case study no. 3: severe (Type 3) toe walker. (a) Representative ankle kinematics.
(b) Representative foot-floor angle. (c) Representative internal ankle moments (kinetics). (d)
Representative ankle sagittal joint power (kinetics). (e) Representative knee kinematics. (f) Repre-
sentative foot progression angle. (g) Example of electromyography. The gray band represents
internal ankle moment (AM2); in more severe cases, the early ankle moment (AM1)
is greater than the late ankle moment (AM2). Pedobarography demonstrates
decreased hindfoot pressure and medialization of the foot center of pressure second-
ary to an increased external foot progression angle. EMG is found to be similar in
patients with cerebral palsy and other causes of toe walking, presenting with
abnormal co-contraction and out-of-phase muscle activity with premature firing of
gastrocnemius in swing as well as low-amplitude tibialis anterior firing throughout
stance and swing. Gait accommodations seen with ITW often include knee hyper-
extension in late stance and increased external foot progression angle.
16 K. Davies et al.
Three-dimensional gait analysis is useful for identifying the child with moderate
toe walking who has multilevel gait accommodations not easily appreciated with
observational gait assessment. Treatment options to address the gastrocnemius
contracture include serial casting, casting with Botulinum Toxin, or casting follow-
ing gastrocnemius recession or percutaneous/open TAL. After the initial interven-
tion, it is important to follow immediately with ankle-foot orthoses for gait retraining
and maintenance of ankle ROM in the growing child.
Cross-References

▶ Ankle Foot Orthotics and Their Influence on Gait
Applications
Analysis
▶ Surface EMG
References
Accardo P, Barrow W (2015) Toe walking in autism: further observations. J Child Neurol 30
(5):606–609
Accardo P, Morrow J, Heaney M et al (1992) Toe walking and language development. Clin Pediatr
(Phila) 31:158–160
Alvarez C, De Vera M, Beauchamp R et al (2007) Classification of idiopathic toe walking based on
gait analysis: development and application of the ITW severity classification. Gait Posture
26:428–435
Alvarez C, De Vera M, Chhina H et al (2008) Normative data for the dynamic pedobarographic
profiles of children. Gait Posture 28:309–315
Baker R (2013) Measuring walking: a handbook of clinical gait analysis. MacKeith Press, London
Barrow W, Jaworski M, Accardo P (2011) Persistent toe walking in autism. J Child Neurol
26:619–621
Bovens A, van Baak M, Vrencken J et al (1990) Variability and reliability of joint measurements.
Am J Sports Med 18(1):58–63
Bowen T, Miller F, Castagno P et al (1998) A method of dynamic foot-pressure measurement for the
evaluation of pediatric orthopaedic foot deformities. J Pediatr Orthop 18(6):789–793
Brunt D, Woo R, Kim H et al (2004) Effect of botulinum toxin type A on gait of children who are
idiopathic toe-walkers. J Surg Orthop Adv 13(3):149–155
Burnett C, Johnson E (1971) Development of gait in childhood II. Dev Med Child Neurol 13
(2):207–215
Clark E, Sweeney J, Yocum A et al (2010) Effects of motor control intervention for children with
idiopathic toe walking: a 5-case series. Pediatr Phys Ther 22:417–426
Crenna P, Redrizzi E, Andreucci E et al (2005) The heel-contact gait pattern of habitual toe walkers.
Cychosz C, Phisitkul P, Belatti D et al (2015) Gastrocnemius recession for foot and ankle conditions
in adults: evidence-based recommendations. Foot Ankle Surg 21(2):77–85
DiGiovanni C, Kuo R, Tejwani N et al (2002) Isolated gastrocnemius tightness. J Bone Joint Surg
84(6):962–970
Eastwood D, Dennett X, Shield L et al (1997) Muscle abnormalities in idiopathic toe-walkers. J
Pediatr Orthop Part B 6:215–218
Eastwood D, Memelaus M, Dickens D et al (2000) Idiopathic toe-walking: does treatment alter the
natural history? J Pediatr Orthop 9:47–49
Engelbert R, Gorter J, Uiterwaal C et al (2011) Idiopathic toe-walking in children, adolescents and
young adults: a matter of local or generalised stiffness? BMC Musculoskelet Disord 12:61
Engström P, Tedroff K (2012) The prevalence and course of idiopathic toe-walking in 5-year-old
children. Pediatrics 130(2):279–284
Engström P, Bartonek A, Tedroff K et al (2013) Botulinum toxin A does not improve the results of
cast treatment for idiopathic toe-walking. A randomized controlled trial. J Bone Joint Surg Am
95:400–407
Fox A, Deakin S, Pettigrew G et al (2006) Serial casting in the treatment of idiopathic toe-walkers
and review of the literature. Acta Orthop Belg 72:722–730
Furrer F, Deonna T (1982) Persistent toe-walking in children: a comprehensive clinical study of
28 cases. Helv Paediatr Acta 37:301–316
Gage J (2004) A qualitative description of normal gait. In: Gage J (ed) The treatment of gait
problems in cerebral palsy. Mac Keith Press, London, pp 42–70
Griffin P, Wheelhouse W, Shiavi R et al (1977) Habitual toe-walkers. a clinical and electromyo-
graphic gait analysis. J Bone Joint Surg Am 49(4):97–101
Grimston S, Nigg B, Hanley D et al (1993) Differences in ankle joint complex range of motion as a
function of age. Foot Ankle 14(4):215–222
Hall J, Salter R, Bhalla S (1967) Congenital short tendo calcaneus. J Bone Joint Surg (Br) 49
(4):695–697
Hemo Y, Macdessi S, Pierce R et al (2006) Outcome of patients after achilles tendon lengthening for
treatment of idiopathic toe walking. J Pediatr Orthop 26(3):336–340
Hicks R, Durinick N, Gage J (1988) Differentiation of idiopathic toe-walking and cerebral palsy. J
Hill R (1995) Ankle equinus. Prevalence and linkage to common foot pathology. J Am Podiatr Med
Assoc 85(6):295–300
Hirsch G, Wagner B (2004) The natural history of idiopathic toe-walking: a long-term follow-up of
fourteen conservatively treated children. Acta Pediatr 93:196–199
Kalen V, Adler N, Bleck E (1986) Electromyography of idiopathic toe walking. J Pediatr Orthop
6:31–33
Karol L, Concha M, Johnston C (1997) Gait analysis and muscle strength in children with surgically
treated clubfeet. J Pediatr Orthop 17(6):790–795
Kelly I, Jenkinson A, Stephens M et al (1997) The kinematic patterns of toe-walkers. J Pediatr
Orthop 17(4):478–480
Le Cras S, Bouck J, Brausch S et al (2011) Cincinnati Children’s Hospital Medical Center:
evidence-based clinical care guideline for management of idiopathic toe walking. Guideline
040, pp 1–17. Available via http://www.cincinnatichildrens.org/service/j/anderson-center/evi
dence-based-care/occupational-therapy-physical-therapy/. Accessed 1 Aug 2016
McMulkin M, Baird G, Caskey P et al (2006) Comprehensive outcomes of surgically treated
idiopathic toe walkers. J Pediatr Orthop 26(5):606–611
McMulkin M, Gordon A, Tompkins B et al (2016) Long term gait outcomes of surgically treated
idiopathic toe walkers. Gait Posture 44:216–220
18 K. Davies et al.
Ming X, Brimacombe M, Wagner G (2007) Prevalence of motor impairment in autism spectrum

disorders. Brain and Development 29:565–570
Papariello S, Skinner S (1985) Dynamic electromyography analysis of habitual toe-walkers. J
Perry J (1985) Normal and pathologic gait. In: Bunch W (ed) Atlas of orthotics, 2nd edn.
C.V. Mosby, St. Louis, pp 76–111
Perry J (1992) Gait analysis: normal and pathological function. Slack Incorporated, Thorofare
Pierz K, Õunpuu S (2013) Toe walking: how to know who to worry about. Paper presented at the
67th American Academy of Cerebral Palsy and Developmental Medicine Meeting, Milwaukee,
16–19 October 2013
Pomarino D, Ramirez Llamas J, Pomarino A (2016) Idiopathic toe walking tests and family
predisposition. Foot Ankle Spec 20(10):1–6
Sawatzky B, Alvarez C, Beauchamp R et al (1999) Adaptations of gait on the contralateral limb in
children with spastic hemiplegia. Paper presented at the 17th Congress of the International
Society for Biomechanics, Calgary, 8–13 August 1999
Shulman L, Sala D, Chu M et al (1997) Developmental implications of idiopathic toe walking. J
Pediatr 130(4):541–546
Sobel E, Caselli M, Velez Z (1997) Effect of persistent toe walking on ankle equinus. Analysis of
60 idiopathic toe walkers. J Am Podiatr Med Assoc 87(1):17–22
Stott N, Walt S, Lobb G et al (2004) Treatment for idiopathic toe walking: results at skeletal
maturity. J Pediatr Orthop 24(1):63–69
Stricker S, Angulo J (1998) Idiopathic toe walking: a comparison of treatment methods. J Pediatr
Orthop 18(3):289–293
Sutherland D, Olsen R, Cooper L et al (1980) The development of mature gait. J Bone Joint Surg
Am 62(3):336–353
Tabrizi P, McIntyre W, Quesnel M et al (2000) Limited dorsiflexion predisposes to injuries of the
ankle in children. J Bone Joint Surg 82-B(8):1103–1106
van Bemmel A, van de Graaf V, van den Bekerom M et al (2014) Outcome after conservative and
operative treatment of children with idiopathic toe walking: a systematic review of the literature.
Musculoskelet Surg 98:87–93
Van Kuijk A, Kosters R, Vugts M et al (2014) Treatment for idiopathic toe walking: a systematic
review of the literature. J Rehabil Med 46:945–957
Walker J (1991) Musculoskeletal development: a review. Phys Ther 71(12):878–889
Westberry D, Davids J, Davis R et al (2008) Idiopathic toe walking: a kinematic and kinetic profile.
Williams C, Tinley P, Curtin M et al (2012) Vibration perception thresholds in children with
idiopathic toe walking gait. J Child Neurol 27(8):1017–1021
Williams C, Michalitsis J, Murphy A et al (2013) Do external stimuli impact the gait of children
with idiopathic toe walking? A study protocol for a within-subject randomized control trial.
BMJ Open 3(3):e0002389
World Health Organization (2007) International classification of functioning, disability and health:
children and youth version: ICF-CY. World Health Organization, Geneva
Gait Disorders in Persons After Stroke
Johanna Jonsdottir and Maurizio Ferrarin
Abstract
Muscle weakness and motor control deficits associated with stroke result in fairly
typical hemiparetic gait. This chapter will focus on persons with hemiparesis after
stroke in the post-acute and chronic stage, their major movement disorders, and
how these affect locomotion. Differences in spatiotemporal, kinematic, and
kinetic parameters that commonly characterize hemiparetic gait in the subacute
and chronic post-stroke periods will be reviewed in reference to characteristics of
speed-matched gait of nondisabled persons.
Keywords
Gait • Stroke • Gait analysis • Kinematics • Kinetics • Coactivation • Hemiparetic
gait • Movement disorders • Speed-matched gait • Muscles • Weakness
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Importance of Gait Velocity in Hemiparetic Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Characteristics of Hemiparetic Gait Independent of Gait Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Kinetics of Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
J. Jonsdottir (*)
LaRiCE, Department of Neurorehabilitation, IRCCS Fondazione Don Carlo Gnocchi Onlus, Milan,
Italy
e-mail: jjonsdottir@dongnocchi.it
M. Ferrarin
Biomedical Technology Department, IRCCS Fondazione Don Carlo Gnocchi Onlus, Milan, MI,
Italy
e-mail: mferrarin@dongnocchi.it

DOI 10.1007/978-3-319-30808-1_61-1
2 J. Jonsdottir and M. Ferrarin
Neuromuscular Factors Affecting Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Voluntary Changes in Gait Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Introduction
Stroke is the most frequent cause of disability in industrialized countries, with

hemiparesis the most common motor impairment affecting approximately 65% of
stroke victims (WHO 2003). Stroke typically affects the upper motor neuron path-
ways, and hemiparesis following stroke is due to a unilateral disruption of
descending neural pathways originating in the cortex or brain stem. The symptoms
seen after stroke depend on the localization of the damage, but mostly a mixed
pattern of muscle weakness and overactivity occurs, and in the period following
stroke, recovery and compensatory strategies develop in the central and peripheral
systems (Sheean and McGuire 2009).
Persons after stroke may present different levels of motor and cognitive disorders
depending on several factors, such as severity of stroke, time after stroke, and age
(Teasell et al. 2005). Three bands/categories of stroke patients were proposed by
Garraway (1985) based on severity of symptoms. An upper band of persons showing
minimal deficits tend to recover well with minimal movement disorders that can be
treated in a community or outpatient setting. A middle band of persons that have
suffered moderately severe strokes, while presenting various motor and cognitive
problems, recover markedly in most areas although remaining dependent on reha-
bilitation. Most of them become walkers and are discharged to the community after a
period of rehabilitation. The lower band of persons are those who suffered a severe
stroke and remained severely paretic, often with serious medical comorbidities
which add to their stroke-related disability. This group most often does not achieve
independent ambulation outside the house.
The main focus in this chapter will be on mobility disorders in the above-defined
upper-to-middle band of persons after stroke that are ambulatory with or without
assistance.
One of the characteristics of stroke is the functional change that occurs from the
first hours after stroke to days and sometimes years after stroke. The damaged brain
goes through recovery processes typical of the immediate post-stroke period of
approximately 48 h, and then in the months following stroke, there is continued
recovery through reorganization that occurs spontaneously and in response to
therapy, as well as due to compensatory strategies developed to deal with daily
activities. Even in the chronic stage after stroke, there is continued possibility of
further functional improvements both through recovery and compensation (Teasell
et al. 2005).
This characteristic functional change in persons after stroke is reflected also in
longitudinal improvements in kinematic and kinetic gait parameters although rarely
do they return to pre-stroke values. Six months after stroke about 75–85% of persons
Gait Disorders in Persons After Stroke 3
with resultant motor and sensorimotor impairment will have recovered the ability to
walk at least 45 m without physical assistance (Patel et al. 2000).
Mobility is achieved through interaction of various systems, of which the central
nervous system and the peripheral neuromuscular and sensory systems play an
important role. Standing and walking require (i) the retaining of the supporting of
the body against gravity in order to prevent a fall and (ii) a propulsion force in order
to move (Clark 2015). Problems at the neural, sensory, and/or muscular level, as well
as problems with coordination among the systems, are all possible causes of gait
deficits in persons after stroke. The resultant movement disorders, such as spasticity,
ataxia, sensory, or proprioceptive deficits that are common in persons after stroke
and evident on neurologic examination, all affect motor control and consequently
gait performance (Snijders et al. 2007; Clark 2015).
The majority of persons that have suffered a stroke show decreased gait velocity
and abnormal gait kinematics and kinetics. Their gait is characterized by spatial and
temporal asymmetries, with poor selective motor control, reduced weight bearing on
the paretic limb, and reduced propulsive force generated by the paretic limb (Chen
et al. 2005; Balasubramanian et al. 2007; Allen et al. 2011; Balaban and Tok 2014).
State of the Art
Importance of Gait Velocity in Hemiparetic Gait
Gait velocity is an important indicator of locomotor function and quality of life after
stroke. Preferred (self-selected) velocity, which is the gait speed a person exhibits
spontaneously when walking at ease, is considered an indicator of overall gait
performance. It is commonly used to monitor locomotor performance and to eval-
uate the effects of rehabilitation in persons with neurological disorders (Perry et al.
1995). Elderly healthy individuals have a preferred speed of between 1.1 and 1.3 m/s
(Turnbull et al. 1995; Lusardi et al. 2003) but can easily change their speed from
slow to running. In general, walking speeds greater than 0.8 m/s indicate that a
person can independently move around in the external environment (Perry et al.
1995; Lord et al. 2004). Being independent community walkers makes it more likely
that a person can participate in community activities, including working. Gait speed
of persons after stroke differs markedly from their healthy counterparts in that they
tend to walk much slower, which considerably influences their participation in
activities of daily life. Only about 40% of persons who recover walking abilities
after stroke achieve community-walking velocities, while half of them will remain at
gait speeds less than 0.5 m/s (Goldie et al. 2001; Macko et al. 2001; Lamontagne
et al. 2007). Post-stroke persons are also generally more limited in their choice of
gait speed; however, when asked some of them are able to increase their gait speed
considerably or at least up to 0.2–0.3 m/s more than their spontaneous speed (Bayat
et al. 2005; Jonsdottir et al. 2009). This indicates the availability of additional
functional resources that are not commonly utilized in daily walking, maybe due
to a need to maintain balance, or divide attention between walking and various

environmental requests.
Along with reduced speed, studies of kinematic and kinetic characteristics of
main features of hemiparetic gait of persons post-stroke through gait analysis have
evidenced reduced cadence, stride length, and joint angular excursion, as well as
asymmetries between the affected and non-affected leg in most gait parameters.
Further, a reduction in force output from the affected leg and a general increased
mechanical energetic cost have been evidenced (Roerdink and Beek 2011; Chen
et al. 2005; Mulroy et al. 2003; Jonsdottir et al. 2009).
Characteristics of Hemiparetic Gait Independent of Gait Speed
Walking velocity, however, tends to correlate with many of the measured temporal
and spatial parameters of the gait cycle, and many of the gait deviation characteris-
tics of hemiparetic gait are consistent with slower walking in nondisabled individ-
uals. This underlines the importance of characterizing gait of persons post-stroke
independent of gait speed (Chen et al. 2005; Jonsdottir et al. 2009; Wagenaar and
Beek 1992; Rinaldi and Monaco 2013). This can be achieved by comparing the
various gait parameters of the gait phases with nondisabled persons walking at
similar speeds.
Chen and colleagues (2005) studied differences in kinematic and kinetic param-
eters of speed-matched gait in five hemiparetic persons post-stroke and nondisabled
controls walking on treadmill. They identified a large set of gait differences consis-
tent with compensatory strategies related to impaired swing initiation in the affected
limb and shortened single limb support on the affected limb. The impaired swing
initiation was associated with inadequate propulsion of the leg during pre-swing,
increased percentage of swing time of the affected leg, as well as reduced knee
flexion at toe-off and mid-swing. This led the persons with stroke to compensate
with leg circumduction or pelvis hiking during the swing phase. The shortened
support time on the paretic limb instead was related to exaggerated propulsion of
the non-affected limb during pre-swing to shorten its swing time.
Other gait deficits independent of gait speed included asymmetric step length and
increased step width. However, when speed-matched, spatiotemporal parameters
cadence, stride time and stride length were similar between gait of persons post-
stroke and nondisabled persons and so were ankle plantar flexion angles at toe-off in
both limbs.
These findings of similar spatiotemporal parameters at matched speed are cor-
roborated with findings from another study comparing the gait of 39 hemiparetic
persons post-stroke with the gait of speed-matched healthy controls (Jonsdottir et al.
2009). The persons, all in the chronic phase after stroke and moderately affected,
walked at their preferred speed and then as fast as possible for gait analysis. Their
spatiotemporal parameters were compared with velocity dependent profiles of non-
disabled persons. Algorithms were used to classify the resulting gait parameters as
reduced, normal, or increased (relative to two standard deviations from the mean of
the nondisabled sample) at preferred and fast speed. At preferred speed cadence and
stride length of the persons post-stroke tended to be similar to those of the
non-disabled persons walking at matched speed, only about one quarter of the
sample had increased cadence and/or reduced stride length, while the rest were
within normal limits. During fast speed half of the sample had increased cadence
and was within normal limits, while stride length remained similar with three
quarters of the sample being within normal limits (Jonsdottir et al. 2009). Rinaldi
and Monaco (2013) demonstrated similar findings from a group of persons post-
stroke that walked on treadmill. Their gait parameters were compared to those of
nondisabled subjects walking at controlled matched speed, resulting in a reduction of
differences in spatiotemporal parameters between the persons post-stroke and the
nondisabled persons.
Kinetics of Gait
Preferred gait speed is related to stride length, and increases in speed are associated
with increased stride length and/or cadence. However, this always necessitates an
associated production of mechanical work at the distal and proximal joints (Olney
and Richards 1996; Jonsdottir et al. 2009). Persons post-stroke can, depending on
their functional status, distribute this work production between the distal and prox-
imal joints in different ways. Kinetic analysis of gait includes the measurement of
ground reaction forces through which joints moment and power can be determined.
Through knowledge of kinetics, it is possible to gain insight into the pathogenesis of
walking impairment post-stroke. In both nondisabled and hemiparetic gait, the
muscle groups that mainly contribute to the generation of energy for forward
propulsion are the hip extensors in early stance, the plantar flexors at push-off, and
the hip flexors at pre-swing (Olney et al. 1994; Parvatanemi et al. 2007). In
nondisabled persons, faster walking speeds are associated with larger power gener-
ation bursts from the same muscle groups with the ankle plantar flexors providing
about 75% of the force contribution (Winter 1990). In hemiparetic gait, the magni-
tude of power generation at both the hip and ankle is reduced with patterns of muscle
activation, tending to differ in both timing and amplitude from normal values when
both walk at self-selected speed (Winter 1990). However, when the joint power
production of post-stroke persons is compared with that of healthy controls walking
at matched speed, the most abnormal parameter is the power generation at the ankle,
while power generation at the hip appears to be mostly in the range of that of healthy
subjects (Jonsdottir et al. 2009; Hsu et al. 2003). It is likely that post-stroke persons
compensate for poor power generation at the ankle by increasing the generation of
power at the hip. According to Hsu et al. (2003) , the strength of the hip flexor
muscles and the knee extensor muscles of the hemiparetic limb is the most important
factor determining comfortable or fast walking speed. However, also plantar flexor
muscles affected walking speed. When post-stroke persons are asked to walk at
maximum, or near maximum, gait velocity capacity, poor distal power generation
appears to be compensated for either by increasing power generation proximally or
by increasing the ankle power generation of the non-affected limb (Nadeau et al.
1999; Hsaio et al. 2016; Awad et al. 2015).
These findings were corroborated by findings from Jonsdottir and colleagues
(2009). When gait parameters of post-stroke persons walking at their preferred speed
were compared to those of nondisabled persons walking at matched speeds, the most
obvious difference was in positive work at the ankle with 64% of the hemiparetic
sample showing significantly lower values than their healthy counterparts (see
Fig. 1a from Jonsdottir et al. 2009). This difference became even larger at faster
speed when over 80% of the hemiparetic sample produced significantly less work at
the ankle. Instead, hip flexor positive work of post-stroke subjects was comparable to
that of nondisabled subjects with 89% and 75% of the subjects demonstrating normal
values during self-selected and fast gait, respectively (see Fig. 1b from Jonsdottir
et al. 2009).
Neuromuscular Factors Affecting Gait
Alterations in the normal activation patterns of typical hemiparetic gait include

increased, early, delayed, and prolonged activation of muscles in the lower limb
during gait activities as well as coactivation of agonist and antagonist muscles (Perry
et al. 1995; Sheffler and Chae 2015).
Muscle coactivation or timing problems at the knee and/or ankle joints are
frequent in persons after stroke. Even in the less affected limb, excessive
coactivation has been reported during both double support phases of gait. This
general increase in coactivation in persons after stroke has been implicated to be
an adaptive behavior to changed neuromuscular dynamics, namely, reduced plantar
flexor strength that may be a limit on safe weight transfer during the double support
phases (Lamontagne et al. 2000; Den Otter et al. 2007; Rosa et al. 2014). In order to
compensate for the ankle plantar flexor weakness, the persons with stroke appear to
increase coactivation.
At the knee level, a characteristic of hemiparetic gait is a genu recurvatum or
excessive knee extension in midstance caused by exaggerated activation or contrac-
ture of the plantar flexors and/or the quadriceps, hamstring, and/or quadriceps
weakness. It has been demonstrated that when there is hyperextension of the affected
leg during midstance, the plantar flexor produces inadequate power at terminal
stance (Cooper et al. 2012).
At the ankle level, foot drop is quite a common problem in hemiparetic gait and
has been associated with insufficient strength of dorsiflexor muscles and/or
coactivation of plantar flexors. Foot drop is noticeable during swing and may be
the only manifestation of hemiparetic gait in less severely impaired persons after
stroke. In more impaired post-stroke persons, this may become evident at initial foot
contact, so that the foot touches the ground in a flat position or even that the forefoot
touches the ground first. The foot drop appears to be due to prolonged
gastrocnemius-soleus activation that then leads to inadequate dorsiflexion during
swing (Sheffler and Chae 2015). Another common problem during hemiparetic gait
Fig. 1 Plots of ankle positive work (a) and hip positive work (b) vs. normalized gait velocity;
individual patients’ values are reported at preferred velocity (circle) and max speed (triangle). The
normal profile and its variability (_1 S.D. range as dash-dotted line) are also shown. The external
dotted lines mark the bounds (_1.96 S.D.) for statistical analysis (Figure from Jonsdottir et al. 2009)
is equinovarus; this occurs when foot drop is associated with excessive plantar
flexion and inversion. Prolonged activation of the tibialis muscles leads to initial
contact occurring with the lateral border of the foot; weight bearing is thus shifted
laterally and can lead to instability at the ankle and increased risk of falling (Sheffler
and Chae 2015).
Coactivation is particularly evident at the ankle during stance and at the knee
during weight acceptance. It is likely that this increased muscle coactivation seen
during post-stroke gait is a compensatory strategy to compensate for muscle weak-
ness and postural instability. This is consistent with the suggestions of Martino et al.
(2015) that the nervous system copes with unstable conditions, such as those
associated to several pathologies, by prolonging the duration of basic muscle activity
patterns. In support of this are the findings of Kitatani and colleagues that investi-
gated the gait of a group of 44 ambulatory post-stroke persons and found that
impaired balance ability and decreased ankle plantar flexor strength on the affected
side predicted coactivation of the ankle muscles (Kitatani et al. 2016).
This increase in ankle muscle coactivation during gait after stroke might thus be a
compensatory strategy for impaired balance and ankle muscle weakness, as well as a
sign of the disturbed motor control following a lesion to the neuromotor system
(Kitatani et al. 2016). The plantar flexors appear to have a unique role during
nondisabled gait, in that they are the only muscle group that can regulate angular
momentum throughout the gait cycle and also appear essential in maintaining
dynamic balance during walking (Neptune and McGowan 2011). Interesting find-
ings by Honeine et al. (2013, 2014) highlighted the role of the triceps surae in
nondisabled gait. Their results indicated that triceps activity during gait set the
kinematics and kinetics of gait through counteracting the force of gravity. In the
first study (2013), they demonstrated that modulation of triceps surae in both
amplitude and duration allows the central nervous system to adapt step length and
cadence. By setting the amplitude and duration of triceps surae activation, the
inherent disequilibrium during gait is controlled, and consequently gait velocity is
set. When subjects walked faster, both the braking action and triceps activity
increased. In a second study, Honeine et al. (2014) demonstrated that the surge of
triceps activity in faster walking is tied to increased requirement of the center of mass
vertical braking action rather than an increase in propulsion. The central nervous
system thus modulated step length and cadence through the control of triceps activity
of the stance leg in the form of a braking action of the fall on the contralateral
advancing leg. The authors speculated that in pathologies where interaction between
brain, central pattern generators, motoneurons and sensory feedback is affected, such
as after a stroke, the step length becomes inconsistent and asymmetric. If the
controlled variable during changing of speed in healthy gait is the activity of triceps
to break the fall on the advancing leg, it is likely that in most post-stroke persons, this
functional resource is not readily available, and thus their neuromotor system has to
readapt its kinetics and kinematics to impose speed changes.
Awad and colleagues (2015) further investigated the importance of ankle propul-
sion during longer distance walking. Forty four individuals walked over ground
during gait analysis and did the 6 minutes walking test. The authors demonstrated
that the paretic limb’s ability to generate propulsion was the determining factor in
long distance walking function. The biomechanical variables during stance phase
propulsion and trailing limb angle were related to distance walked, while swing
phase parameters, paretic ankle dorsiflexion and knee flexion, and symmetry param-
eters, step length and swing time, did not influence long distance walking function.
Voluntary Changes in Gait Speed
When it comes to persons post-stroke, a determinant characteristic of their gait is a

deficit in muscle activation at the ankle; however, it appears that the capacity to
modulate the timing of the force production may be mostly preserved. When asked
to increase gait velocity from self-selected to fast, a typical strategy of persons after
stroke is the more-than-normal advance of the ankle power onset of the affected leg,
while there is little or no increase in positive work of the plantar flexors (Jonsdottir
et al. 2009). As mentioned above, during gait this deficit in propulsion capacity can
be compensated by more proximal propulsive mechanisms to achieve faster gait
velocity (Neptune and McGowan 2011; Hsu et al. 2003; Beaman et al. 2010).
According to Hsu et al. (2003) the strength of the hip flexor muscles and the knee
extensor muscles of the hemiparetic limb are the most important factors determining
comfortable or fast walking speed. Further, it has been demonstrated that the
nonparetic plantar flexor muscles are among the primary mechanisms in speed
modulation (Hsaio et al. 2016).
Summary
The majority of persons with mild-to-moderate sequelae post-stroke have reduced

gait velocity and abnormal gait kinematics and kinetics that can influence their
quality of life, including reduced stride length, asymmetric step lengths, and long
distance walking. Many of their gait deficits may be due to lack of torque-generating
capacities of the leg muscles and incoordination of the neuromuscular system, as
well as general instability. However, it is evident from the literature that comparison
of gait characteristics of persons that have had stroke should be made with non-
disabled persons walking at matched velocities, since many gait parameters are
sensitive to speed. Some apparent abnormalities in hemiparetic gait disappear
when speed is controlled for (Chen et al. 2005; Jonsdottir et al. 2009; Wagenaar
and Beek 1992; Rinaldi and Monaco 2013). A determinant characteristic of the
hemiparetic gait is a deficit in propulsion capacity at the affected ankle; however, it
appears that the capacity to modulate the timing of the force production may be
mostly preserved. During gait this deficit in propulsion capacity can be compensated
for by more proximal propulsive mechanisms or by an increase in propulsion of the
plantar flexors on the contralateral side in order to maintain a self-selected gait
velocity or to achieve faster gait velocity (Jonsdottir et al. 2009; Nadeau et al.
1999; Hsaio et al. 2016; Awad et al. 2015).
References
Allen JL, Kautz SA, Neptune RR (2011) Step length asymmetry is representative of compensatory
mechanisms used in post-stroke hemiparetic walking. Gait Posture 33(4):538–543. doi:10.1016/
j.gaitpost.2011.01.004. Epub 2011 Feb 11
Awad LN, Binder-Macleod SA, Pohlig RT, Reisman D (2015) Paretic propulsion and trailing limb
angle are key determinants of long-distance walking function after stroke. Neurorehabil Neural
Repair 29(6):499–508. doi:10.1177/1545968314554625. Epub 2014 Nov 10
Balaban B, Tok F (2014) Gait disturbances in patients with stroke. PM R 6(7):635–642.
doi:10.1016/j.pmrj.2013.12.017. Epub 2014 Jan 19
Balasubramanian CK, Bowden MG, Neptune RR, Kautz SA (2007) Relationship between step
length asymmetry and walking performance in subjects with chronic hemiparesis. Arch Phys
Med Rehabil 88(1):43–49
Beaman CB, Peterson CL, Neptune RR, Kautz SA (2010) Differences in self-selected and fastest-
comfortable walking in poststroke hemiparetic persons. Gait Posture 31(3):311–316
Bayat R, Barbeau H, Lamontagne A (2005) Speed and temporal-distance adaptations during
treadmill and overground walking following stroke. Neurorehabil Neural Repair 19(2):115–124
Chen G, Patten C, Kothari DH, Zajac FE (2005) Gait differences between individuals with post-
stroke hemiparesis and non-disabled controls at matched speeds. Gait Posture 22(1):51–56
Clark DJ (2015) Automaticity of walking: functional significance, mechanisms, measurement and
rehabilitation strategies. Front Hum Neurosci 9:246
Cooper A, Alghamdi GA, Alghamdi MA et al (2012) The relationship of lower limb muscle
strength and knee joint hyperextension during the stance phase of gait in hemiparetic stroke
patients. Physiother Res Int 17:150–156
Den Otter AR, Geurts ACH, Mulder T, Duysens J (2007) Abnormalities in the temporal patterning
of lower extremity muscle activity in hemiparetic gait. Gait Posture 25(3):342–352.
doi:10.1016/j.gaitpost.2006.04.007
Garraway M (1985) Stroke rehabilitation units: concepts, evaluation, and unresolved issues. Stroke
16(2):178–181
Goldie PA, Matyas TA, Evans OM (2001) Gait after stroke: initial deficit and changes in temporal
patterns for each gait phase. Arch Phys Med Rehabil 82(8):1057–1065
Honeine JL, Schieppati M, Gagey O, Do MC (2013) The functional role of the triceps surae muscle
during human locomotion. PLoS One 8(1):e52943. doi:10.1371/journal.pone.0052943. Epub
2013 Jan 16
Honeine J, Schieppati M, Gagey O, Do M (2014) By counteracting gravity, triceps surae sets both
kinematics and kinetics of gait. Phys Rep 2:e00229. doi:10.1002/phy2.229
Hsiao H, Awad LN, Palmer JA, Higginson JS, Binder-Macleod SA (2016) Contribution of paretic
and nonparetic limb peak propulsive forces to changes in walking speed in individuals post-
stroke. Neurorehabil Neural Repair 30(8):743–752. doi:10.1177/1545968315624780. Epub
2015 Dec 31
Hsu AL, Tang PF, Jan MH (2003) Analysis of impairments influencing gait velocity and asymmetry
of hemiplegic patients after mild to moderate stroke. Arch Phys Med Rehabil 84(8):1185–1193
Jonsdottir J, Recalcati M, Rabuffetti M, Casiraghi A, Boccardi S, Ferrarin M (2009) Functional
resources to increase gait speed in people with stroke: strategies adopted compared to healthy
controls. Gait Posture 29(3):355–359
Kitatani R, Ohata K, Sato S, Watanabe A, Hashiguchi Y, Yamakami N, Sakuma K, Yamada S
(2016) Ankle muscle coactivation and its relationship with ankle joint kinematics and kinetics
during gait in hemiplegic patients after stroke. Somatosens Mot Res 33(2):79–85. 2016 May 18
Lamontagne A, Fung J, McFadyen BJ, Faubert J (2007) Modulation of walking speed by changing
optic flow in persons with stroke. J NeuroEng Rehabil 4:22. doi:10.1186/1743-0003-4-22
Lamontagne A, Malouin F, Richards CL (2000) Contribution of passive stiffness to ankle
plantarflexor moment during gait after stroke. Arch Phys Med Rehabil 81(3):351–358.
PMID:10724082
Lord SE, McPherson K, McNaughton HK, Rochester L, Weatherall M (2004) Community ambu-
lation after stroke: how important and obtainable is it and what measures appear predictive?
Lusardi MM, Pellecchia GL, Schulman M (2003) Functional performance in community living
older adults. J Geriatr Phys Ther 26(3):14–22
Macko RF, Smith GV, Dobrovolny CL, Sorkin JD, Goldberg AP, Silver KH (2001) Treadmill
training improves fitness reserve in chronic stroke patients. Arch Phys Med Rehabil 82:879–884
Martino G, Ivanenko YP, d’Avella A, Serrao M, Ranavolo A, Draicchio F, Cappellini G, Casali C,
Lacquaniti F (2015) Neuromuscular adjustments of gait associated with unstable conditions.
Neurobiologia 114(5):2867–2882. doi:10.1152/jn.00029.2015. Epub 2015 Sep 16
Mulroy S, Gronley J, Weiss W, Newsam C, Perry J (2003) Use of cluster analysis for gait pattern
classification of patients in the early and late recovery phases following stroke. Gait Posture 18
(1):114–125
Nadeau S, Gravel D, Arsenault AB, Bourbonnais D (1999) Plantarflexor weakness as a limiting
factor of gait speed in stroke subjects and the compensating role of hip flexors. Clin Biomech
(Bristol, Avon) 14(2):125–135
Neptune RR, McGowan CP (2011) Muscle contributions to whole-body sagittal plane angular
momentum during walking. J Biomech 44(1):6–12. doi:10.1016/j.jbiomech.2010.08.015. Epub
2010 Sep 15
Olney SJ, Griffin MP, McBride ID (1994) Temporal, kinematic, and kinetic variables related to gait
speed in subjects with hemiplegia: a regression approach. Phys Ther 74(9):872–885
4(2):136–148
Parvatanemi K, Olney SJ, Brouwer B (2007) Changes in muscle group work associated with
changes in gait speed of persons with stroke. Clin Biomech 22:813–820
Patel AT, Duncan PW, Lai SM, Studenski S (2000) The relation between impairments and
functional outcomes poststroke. Arch Phys Med Rehabil 81(10):1357–1363. PMID:11030501
Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE (2010) Changes in gait symmetry and
velocity after stroke: a cross-sectional study from weeks to years after stroke. Neurorehabil
Neural Repair 24(9):783–790
Perry J, Garrett M, Gronley JK, Mulroy SJ (1995; Jun) Classification of walking handicap in the
stroke population. Stroke 26(6):982–989
Rinaldi LA, Monaco V (2013) Spatio-temporal parameters and intralimb coordination patterns
describing hemiparetic locomotion at controlled speed. J NeuroEng Rehabil 10(1):53
Roerdink M1, Beek PJ (2011) Understanding inconsistent step-length asymmetries across hemi-
plegic stroke patients: impairments and compensatory gait. Neurorehabil Neural Repair 25
(3):253–258. doi:10.1177/1545968310380687. Epub 2010 Nov 1
Rosa MC, Marques A, Demain S, Metcalf CD (2014) Lower limb co-contraction during walking in
subjects with stroke: a systematic review. J Electromyogr Kinesiol 24(1):1–10
Sheean G, McGuire JR (2009) Spastic hypertonia and movement disorders: pathophysiology,
clinical presentation, and quantification. PM R 1(9):827–833. doi:10.1016/j.pmrj.2009.08.002
Sheffler LR, Chae J (2015) Hemiparetic Gait. Phys Med Rehabil Clin N Am 26(4):611–623
Snijders AH, van de Warrenburg BP, Giladi N, Bloem BR (2007) Neurological gait disorders in
elderly people: clinical approach and classification. Lancet Neurol 6:63–74
Teasell R, Bayona NA, Bitensky J (2005) Plasticity and reorganization of the brain post stroke. Top
Stroke Rehabil 12:11–26. doi:10.1310/6aum-etyw-q8xv-8xac
Turnbull GI, Charteris J, Wall JC (1995) A comparison of the range of walking speeds between
normal and hemiplegic subjects. Scand J Rehabil Med 27(3):175–182
Wagenaar RC, Beek WJ (1992) Hemiplegic gait: a kinematic analysis using walking speed as a
basis. J Biomech 25(9):1007–1015
WHO (2003) World health report. World Health Organization, Geneva
Hereditary Motor Sensory Neuropathy:
Understanding Function Using Motion
Analysis
Sylvia Õunpuu and Kristan Pierz
Abstract
Hereditary motor sensory neuropathies, or Charcot-Marie-Tooth disease, repre-
sent a heterogeneous group of inherited neuropathies that are characterized by
progressive wasting and resulting weakness of the distal muscles in the legs and
arms. Lower extremities are typically initially effected and, as result, impact
ambulation. At present there is no curative treatment available; therefore, treat-
ment of gait issues is often sought to help with ambulation and activities of daily
living. Computerized motion analysis techniques have improved our understand-
ing of the various presentations of hereditary neuropathies and can assist in
making optimal treatment decisions to improve gait. These presentations include
three distinct ankle variations: excessive equinus (toe walking), cavo-varus (lat-
eral border weight bearing), and flail foot (heel weight bearing) patterns. As each
patient presents differently in terms of deformity specifics and severity, a detailed
analysis that describes ankle/foot function during gait in terms of foot pressures,
muscle activity, kinematics, and kinetics along with clinical examination infor-
mation such as muscle strength and passive range of motion is very beneficial.
Assessment of treatment outcomes from bracing to orthopedic surgery as well as
disease progression which also varies person to person is necessary to develop
evidence-based treatment indications and goals. Motion analysis can play a very
important role in the assessment of inherited neuropathies on both an individual
patient basis and in research with the ultimate goal of improving treatment
outcomes.
S. Õunpuu (*) • K. Pierz

Center for Motion Analysis, Division of Orthopaedics, Connecticut Children’s Medical Center,
Farmington, CT, USA
e-mail: Sounpuu@ccmckids.org; soupuuu@connecticutchildrens.org;
kpierz@connecticutchildrens.org

DOI 10.1007/978-3-319-30808-1_62-1
2 S. Õunpuu and K. Pierz
Keywords
Hereditary neuropathies • Charcot-Marie-Tooth • Clinical gait analysis • Peak
dorsiflexion • Joint kinematics • Joint kinetics • Ankle
Abbreviations
AFO Ankle foot orthosis
CMT Charcot-Marie-Tooth
EMG Electromyography
PLS Posterior leaf spring orthosis
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Gait Function and Treatment Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Clinical Examination Findings in CMT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gait Findings: As Defined by Motion Analysis Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Management: Evaluation of Treatment Outcomes Using Motion Analysis . . . . . . . . . . . . . . . . . 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Introduction
Hereditary motor sensory neuropathies, or Charcot-Marie-Tooth disease (CMT),

represent a heterogeneous group of inherited neuropathies that are characterized by
progressive weakness and resulting wasting of the distal muscles in the extremities
(legs and arms). This group of neuropathies was described in detail with the tools
available at the time in 1886 by neurologists Charcot, Marie, and Tooth who defined
them as peroneal muscular atrophies of familial origin (Charcot 1886; Tooth 1886).
These neuropathies are length dependent, meaning the longest nerves in the body are
affected, and therefore, the lower extremities are typically initially affected in the
course of disease progression. In terms of function, the resulting muscle weakness
results in foot deformity such as flatfoot or high arch and gait issues such as ankle
instability, clumsiness, and slow running. During standing and walking, ankle
support is often needed and provided by orthoses, and in some cases walking
becomes so difficult over time that wheelchair mobility is required. At present
there is no curative treatment available, so treatment of gait issues can help provide
more functional ambulation and improve/facilitate participation and activities of
daily living. As with any complex movement disorder, CMT presents differently
in each patient in terms of deformity, severity, and rate of disease progression.
Comprehensive motion analysis techniques can provide objective documentation
of (a) the pathomechanics, (b) disease progression, and (c) treatment outcomes. The
goal of this chapter is to highlight what we have learned about the pathomechanics of
gait in persons with CMT using comprehensive motion analysis techniques. The
application of motion analysis is in its infancy for this gait pathology. The ultimate
Hereditary Motor Sensory Neuropathy: Understanding Function Using Motion. . . 3
goal is to improve treatment outcomes and to understand prognosis for future

function near the time of diagnosis through a better understanding of gait
pathomechanics. Although progress has been made, there is much more to be done.
State of the Art
The application of motion analysis in this clinical setting is relatively new in

comparison to other gait disorders such as cerebral palsy and spina bifida, so our
understanding of gait function in persons with CMT limited. Because CMT is a rare
disease, the clinical application of motion analysis in this patient population is not as
common, and research applications for the most part have focused on small diverse
groups of patients that span a wide age range. However, over time this body of
literature has grown, and clinicians have gained more understanding of the
pathomechanics of gait. The presentation of CMT in classical textbooks generally
describes a common set of impairments (weakness and contracture) that result in a
specific gait pattern including excessive equinus in swing and increased hip flexion,
pelvic hiking, and circumduction to clear the foot (Holmes and Hansen 1993).
Clinical experience shows that many patients with CMT do not exhibit excessive
equinus in swing and the associated proximal compensations. Also cavo-varus foot
deformity is noted as common (Smith 2002); however, many patients with CMT
have flail feet with pes valgus. It is clear from previous research that incorporates
comprehensive motion analysis assessments that there are substantial variations in
presentation and severity of impairments (Burns et al. 2005, 2006; Garcia et al. 1998;
Vinci and Perelli 2002; Vinci et al. 2006) that result in a variety of gait pathologies
(Don et al. 2007; Kuruvilla et al. 2000; Newman et al. 2007; Ramdharry et al. 2009).
As a result, this latter group of authors divided CMT patients into groups based upon
gait pathology. In many cases gait features often overlapped, that is, multiple
features were present in a single patient. These studies mostly focused on adults
with CMT, and as a result there is very little understanding of CMT gait pathology
evolution during childhood when the impairments typically present. However, there
are a few exceptions. The first signs of CMT were evaluated by Burns et al. 2009 and
include issues with ability to toe walk and run and clumsy gait (Burns et al. 2009).
Ferrarin (Ferrarin et al. 2012) divided gait pathology into three categories in a group
of children and adolescents using motion analysis outcomes: (a) pseudo normal,
(b) drop foot only, and (c) foot drop and push-off deficiency. In a group of similarly
aged patients, Ounpuu (Ounpuu et al. 2013)focused on peak dorsiflexion in terminal
stance as a way to differentiate gait in persons with CMT as this particular gait
parameter is relevant to treatment strategy. In this study, patients were placed into
three groups: (a) less than typical peak dorsiflexion, (b) typical peak dorsiflexion,
and (c) increased peak dorsiflexion. Patients with CMT, however, rarely present with
one issue at the ankle, so in groups (a) and (c), there was also simultaneous increased
equinus, and in groups (b) and (c), there was delayed peak dorsiflexion in stance.
There is, however, still work needed to further refine gait classifications in these
patients, understand disease progression and implications of asymmetry and
ultimately link gait issues to phenotype. Initial reports on CMT often depicted large
asymmetries, however, it has been recently reported that asymmetry is not that
common in CMT in terms of a wide variety of outcome measures including gait
parameters (Burns et al. 2012).
In that there is such variation in presentation, treatments need to be patient
specific and based on a comprehensive patient examination including motion anal-
ysis especially at the level of the ankle and foot (Jani-Acsadi et al. 2015). There is,
however, very little objective documentation of treatment outcomes in terms of gait
function using comprehensive motion analysis techniques. As more centers incor-
porate comprehensive motion measurement techniques as part of the standard of
care, this will become possible.
Gait Function and Treatment Options
This section will include a description of the following for persons with CMT: (a) the
current understanding of gait pathology in terms of gait analysis parameters and
associated clinical examination findings and (b) the current standard of treatment for
gait-related issues including therapies, orthoses, and orthopedic surgery. Example
patient data will be incorporated to illustrate common findings.
Clinical Examination Findings in CMT
Clinical examination findings which are collected using standard protocols for the
assessment of strength, passive range of motion, and bony deformity are integral to
the interpretation of motion analysis data so need to be a part of a full assessment of
gait pathology. Clinical findings help to determine the primary impairments that lead
to gait pathology. Persons with CMT typically have impairments that include
weakness, contracture, and bony deformity which are further complicated by the
disease progression (increased problems over time) as well as issues with reduced
sensation and reflex response which also need to be assessed. For the context of this
chapter with the focus on gait analysis, we will briefly discuss the primary impair-
ments that impact gait in persons with CMT.
The first manifestations of disease for most persons with CMT are distal limb
weakness and muscle atrophy. Sensory loss and absent reflexes are also present
(Thomas 1999). Comprehensive research on strength testing methods and compar-
ison between controls and persons with CMT1A in terms of strength and passive
range of motion findings have been completed by Burns and colleagues (Burns et al.
2005, 2009; Rose et al. 2010). Their detailed studies have led to some primary
findings in children with CMT and cavus foot deformity which include reduced
strength in all of the muscles of the feet. Initial deficits in young children in
comparison with their age-matched peers start with weakness of the plantar flexors,
dorsiflexors, and evertors of the ankle (Rose et al. 2010). Young children with CMT
also have greater inversion-to-eversion and plantar flexion-to-dorsiflexion strength
ratios and a high correlate between dorsiflexion range of motion and foot and ankle
strength. Those with CMT also have significantly less passive ankle dorsiflexion
range of motion. Strength assessments in adult persons with CMT have also
documented weakness in the ankle dorsiflexors, plantar flexors, and evertors (Don
et al. 2007; Newman et al. 2007). Strength and range of motion deficits result in gait
pathology and should be linked to the interpretation of gait data to determine causes
of gait pathology. This has been done in the majority of research involving motion
analysis and CMT. Longitudinal data that would help to explain the pathogenesis
and progression of this disease in terms of gait decline over time would be extremely
helpful and is not yet available.
Gait Findings: As Defined by Motion Analysis Parameters
Motion analysis has led to a more comprehensive understanding of the

pathomechanics of gait in persons with CMT. The following section will describe
CMT in the context of the following components of a comprehensive motion
analysis: gait kinematics and kinetics, electromyography, and pedobarography.
Each component provides a unique source of information not available using
standard assessments and when combined provide a comprehensive view of the
pathomechanics of gait in this population. Motion analysis also provides a unique
opportunity to assess gait function objectively over time and to evaluate treatment
outcomes such as orthoses and orthopedic surgery.
Kinematic and Kinetic Findings in CMT

As mentioned above, the majority of the previous research focused on CMT and gait
includes joint and segment kinematic and joint kinetic outcomes. These data provide
an opportunity to better understand the pathomechanics of gait in CMT and help to
make links between the primary pathology and compensations noted at proximal
joints. It is clear from these works that CMT with its heterogeneous genetic makeup
also has a heterogeneous impairment and associated functional presentation. Ankle
function during gait in CMT is impacted depending on the impairments of weakness
and contracture in the associated musculature. Therefore, presentation can be dif-
ferentiated by the degree of peak dorsiflexion in terminal stance (which is impacted
by both ankle plantar flexor range of motion and strength) and divided into three
groups: (a) increased and delayed peak dorsiflexion, (b) delayed peak dorsiflexion,
and (c) increased plantar flexion (Ounpuu et al. 2013). Similarly, ankle dorsiflexion
in mid-swing and at initial contact is often impacted by ankle dorsiflexor strength but
is also impacted by plantar flexor tightness that does not “allow” the anterior tibialis
to effectively dorsiflex the ankle in a non-weight-bearing position (swing phase). So
CMT commonly presents with excessive equinus in mid-swing and initial contact.
Cavus foot deformity is also common but not well documented using current motion
analysis methods. As it impacts the available length of the ankle plantar flexors, it is
an important consideration for the overall function of the foot and ankle.
Generally, the foot and ankle in CMT fall into three groups of impairments.
Plantar-Dorsiflexion
40
Dor
deg
Pla
-40
Ankle Moment
2.0
Ext
Nm/kg
Flx
-1.0
Ankle Power
3.0
W/kg
-2.0
25% 50% 75%
a b Gait Cycle c
Fig. 1 Example of the flail foot during (a) relaxed standing with significant weakness of the ankle
musculature as evident in the (b) ankle sagittal plane kinematic, moment and power (three gait
cycles) during stance and swing phases and in (c) foot pressures for the right foot. Increased and
delayed peak dorsiflexion, reduced peak ankle plantar flexor moment and power generation in
terminal stance and reduced pressures under the distal foot with a limited length of center of
pressure path are evidence of ankle plantar flexor weakness. Increased plantar flexion in swing is
evidence of ankle dorsiflexor weakness
• The flail foot with significant weakness in all ankle muscles that leads to increased
and delayed peak dorsiflexion in stance and equinus in swing with medial and
lateral instability resulting in pes planovalgus deformity over time (Fig. 1a–c).
Although claw toes are common in earlier phases of the disease, in this group, the
toe flexors are typically weak with no toe contact during stance.
• The cavus foot deformity with lateral weight bearing during stance and delayed
peak dorsiflexion in terminal stance consistent with ankle plantar flexor weakness
and typically normal or only minimal increased equinus in swing (Fig. 2a–c). The
cavus deformity limits available plantar flexor length which when at its end range
may mask ankle plantar flexor weakness.
• Equinus ankle deformity with increased equinus in stance and swing due to lack
of plantar flexor range of motion (assessment of shank vs. plantar aspect of the
foot) (Fig. 3a–c). This lack of passive range of motion may be a result of a
combination of cavus and/or limited plantar flexor range of motion. The cavus
40
Dor
deg
Pla
-40
2.0
Ankle Moment
Ext
Nm/kg
Flx
-1.0
Ankle Moment
3.0
Gen
W/kg
Abs
-2.0
25% 50% 75%
a b Gait Cycle c
Fig. 2 Example of the cavo-varus foot during (a) relaxed standing with some evidence of plantar
flexor weakness in the (b) ankle sagittal plane kinematic, moment and power (three gait cycles)
during stance phase and evidence of lateral weight bearing in (c) foot pressures for the right foot.
Weakness of the ankle plantar flexors is most evident in the delayed and increased peak ankle
dorsiflexion in stance, however, this is not yet manifested at the joint kinetic level
deformity limits available plantar flexor length and may mask the presence of
ankle plantar flexor weakness.
The compensatory proximal gait findings depend on the ankle impairment. For
those patients with increased equinus in swing and dorsiflexion in stance, compensa-
tory steppage gait is common with increased hip flexion and in some cases
circumduction to aid in clearance (Fig. 4). For those with increased ankle dorsiflexion
in stance, increased knee and in some cases hip flexion occur. Because the kinematic
and kinetic presentations vary from patient to patient, the treatment options need to be
specific to the patient and are discussed in the Management section below.
Electromyographic Findings in CMT

Muscle activity for those muscles that are used to support the ankle during gait in
persons with CMT is often abnormal and is manifested in multiple ways including:
(a) reduced activity, (b) no activity, and (c) atypical recruitment patterns. Lack of
muscle activity during the appropriate phases of the gait cycle can result in ankle
40
Dor
deg
Pla
-40
Ankle Moment
2.0
Ext
Nm/kg
Flx
-1.0
Ankle Power
3.0
Gen
W/kg
Abs
-2.0
25% 50% 75%
a b Gait Cycle c
Fig. 3 Example of the equinus ankle during (a) relaxed standing with inability to dorsiflex above
neutral and increased equinus documented in the (b) ankle sagittal plane kinematic, moment and
power (three gait cycles) during stance and swing phases and in (c) foot pressures with the right
forefoot and toes only in contact with the ground instance. Increased plantar flexion in stance and
swing is evidence of plantar flexor tightness
lateral instability if the peroneals are involved, increased equinus in swing if the
anterior tibial group is involved, and increased peak dorsiflexion in stance if the
plantar flexors are involved. Abnormal muscle contractile patterns such as single
motor unit recruitment and fasciculations (repetitive single motor unit contractions at
rest) are also characteristic of demyelinated muscle that can be noted on the surface
EMG signal. Dynamic electromyography (EMG) techniques can provide insight into
which muscles are active during which phases of gait and assist in understanding
potential for muscles to provide dynamic support when transferred. As in the varied
kinematic presentation, one would expect that the muscle function for this patient
group is also different from person to person as muscle activity is the driver of motion
and stability. So EMG patterns for key ankle muscle groups can show minimal issues
or provide evidence to explain why the ankle joint is unstable (Fig. 5a–c).
Pedobarography (Foot Pressures) Findings in CMT

The analysis of foot pressures during gait provides an opportunity to understand
where the peak pressures are under the foot and the path of the center of pressure
during the stance phase of gait. High mean and peak pressures and pressure-time
a b
Pelvic Tilt Pelvic Obliquity
40 20
Ant Up
deg
Pos Dn
-20 -20
Hip Flexion-Extension Hip Ab-Adduction

60 20
Flx Add
deg
Ext Abd
-20 -20
25% 50% 75%
Knee Flexion-Extension Gait Cycle

80
Flx
Ext
-20
40
Dor
deg
Pla
-40 25% 50% 75%

Gait Cycle
Fig. 4 Compensations for excessive equinus in swing typically include (a) increased hip flexion in
the sagittal plane (left column second plot) and (b) circumduction and pelvic hiking in the coronal
plane (right column)
a) Right Side EMG

5
Right Gastrocnemius 5
Right Gastrocnemius* 5
Right Gastrocnemius**
V V V
-5 -5 -5
Right Anterior Tibialis Right Anterior Tibialis* Right Anterior Tibialis**
5 5 5
V V V
-5 -5 -5
5
Right Peroneus Longus 5
Right Peroneus Longus* 5
Right Peroneus Longus**
V V V
-5 -5 -5
Right Peroneus Brevis Right Peroneus Brevis* Right Peroneus Brevis**
5 5 5
V V V
-5 -5 -5
b) Right Side EMG

5
Right Gastrocnemius 5
Right Gastrocnemius* 5
Right Gastrocnemius**
V V V
-5 -5 -5
Right Anterior Tibialis Right Anterior Tibialis* Right Anterior Tibialis**
5 5 5
V V V
-5 -5 -5
5
Right Peroneus Longus 5
Right Peroneus Longus* 5
Right Peroneus Longus**
V V V
-5 -5 -5
Right Peroneus Brevis Right Peroneus Brevis* Right Peroneus Brevis**
5 5 5
V V V
-5 -5 -5
c) Left Side EMG

5
Left Gastrocnemius 5
Left Gastrocnemius* 5
Left Gastrocnemius**
V V V
-5 -5 -5
Left Anterior Tibialis Left Anterior Tibialis* Left Anterior Tibialis**
5 5 5
V V V
-5 -5 -5
5
Left Peroneus Longus 5
Left Peroneus Longus* 5
Left Peroneus Longus**
V V V
-5 -5 -5
Left Peroneus Brevis Left Peroneus Brevis* Left Peroneus Brevis**
5 5 5
V V V
-5 -5 -5
Fig. 5 Example EMG data for three different patients with CMT show heterogeneity of this disease
in terms of the primary muscles at the ankle with (a) minimal atypical findings except for the
gastrocnemius which shows delayed onset and minimal activity consistent with the common finding
of plantar flexor weakness and increased dorsiflexion in terminal stance, (b) minimal activity other
than fasciculations of peroneus longus and (c) no EMG findings consistent with the flail foot
Fig. 6 Example foot pressure plots highlight common patterns found in persons with CMT
including (a) cavus deformity with no toe contact, (b) increased lateral weight bearing pressure,
(c) inability to weight bear over the distal foot in any capacity with all pressures focused under the
heel in comparison to the (d) typically developing reference. In very young patients who cannot
cooperate with a clinical exam to evaluate strength, the foot pressure plot is an excellent tool to
understand plantar flexor strength
integrals have been measured with pedobarography in persons with CMT; however,
their relationship to foot pain is not clear (Burns et al. 2005; Crosbie et al. 2008). As
with other measurements, there is a larger variety of presentations in terms of foot
pressures in persons with CMT (Fig. 6). This assessment is of particular interest in
children where skin changes may not yet provide adequate information to under-
stand the impact of cavus and adductus deformities. When used in conjunction with
radiographs of the foot, the impact of anatomical abnormalities can be better
understood. Foot pressures also provide an assessment of ankle plantar flexor and
toe flexor strength. When toe flexor strength is compromised, the ability of the last
portion of push-off is compromised, and no toe contact is made with the floor
(Fig. 6a, b). When ankle plantar flexors are compromised, there is reduced ability
to weight bear over the distal foot. Making this assessment by observing the foot/
ankle is limited as while the flail foot may be in contact with the ground the center of
pressure may not be able to move over the distal aspect of the foot (Fig. 6c). The foot
pressure plot in some cases may be the only way to obtain adequate information
regarding plantar flexor strength when a patient is too young or unable to understand
directions in a strength assessment. Pre- versus postsurgical intervention to the foot
can also be assessed and allow for a critical examination of the impact of bony and/or
soft tissue surgery outcomes during gait.
Management: Evaluation of Treatment Outcomes Using Motion

Analysis
It is clear from the previous research and clinical experience that persons with CMT
have a wide variety of presentations and therefore require different treatment strat-
egies to provide optimum outcomes. Motion analysis is an excellent tool to assist in
better understanding the pathomechanics with which to make more informed treat-
ment and provide an opportunity to objectively evaluate treatment outcomes. If
motion analysis is incorporated as the standard of care, it will be possible someday
to provide improved care through evidenced-based medical practice. The following
is a discussion of how motion analysis can be used to better define treatment and
understand treatment outcomes.
Physical Therapy
Physical therapy is a common component of treatment for patients with CMT with a
focus on strengthening, maintaining range of motion, and balance training with the
goal of maintaining mobility. It has been shown that strength training can benefit
measures of muscle strength and other outcome measures such as walking velocity
(Burns et al. 2009) as well as activities of daily living in adults with CMT (Chetlin
et al. 2004); however, there is limited current knowledge on how strength training
impacts gait function in terms of joint kinematics and kinetics. Understanding
therapy outcomes at the joint level will help to explain why some therapies may
be more successful than others. For example, understanding if strengthening of the
plantar flexors improved ankle sagittal plane kinematics and kinetics would help to
understand why there may be benefits in walking velocity. Motion analysis could
also help with targeting specific therapies to provide the most functional benefits.
Orthoses
Orthoses are very effective treatment modalities for support of the ankle when
muscle weakness is a primary clinical finding. In persons with CMT, ankle instabil-
ity is a very common finding resulting from weakness of the ankle plantar flexors and
the medial and lateral stabilizers (Mandarakas et al. 2013). Also, clearance problems
may be present as a result of weakness of the anterior tibialis and long toe extensors.
Motion analysis techniques can provide insight into the impact of orthoses on ankle
function that includes not only kinematic function of the ankle but also kinetic
function that can explain in some cases why patients are non-compliant with their
orthoses. Walking velocity has been shown to improve with the application of
orthoses (Phillips et al. 2011); however, the assessment of this valuable outcome
measure on its own does not provide information about the impact of the orthosis at
the joint level which is required to understand why an orthosis improves walking
velocity and ultimately indications for orthosis specific design. Ramdharry
(Ramdharry et al. 2012) incorporated comprehensive motion analysis and
documented reduced (improved) excessive dorsiflexion in terminal stance and
plantar flexion in swing with the application of an appropriately molded ankle foot
orthosis (AFO). However, it is clear that in patients with CMT, one AFO design does
Plantar-Dorsiflexion Plantar-Dorsiflexion
40 40
Dor Dor
deg deg
Pla Pla
-40 -40
Ankle Moment Ankle Moment
2.0 2.0
Ext Ext
Nm/kg Nm/kg
Flx Flx
-1.0 -1.0
Ankle Power Ankle Power
3.0 3.0
W/kg W/kg
-2.0 2.0
25% 50% 75% 25% 50% 75%
a Gait Cycle b Gait Cycle c
Fig. 7 Comparison of ankle sagittal plane kinematic, moment and power for (a) barefoot and (b) solid
AFO walking for the right side in a youth with CMT. The solid AFO (c) can provide improved stability
at the ankle in the sagittal plane by restricting excessive peak dorsiflexion in terminal stance, however,
this is done at the expense peak ankle power generation which shows a decrease in the solid AFO. It is
likely that this explains why some persons with CMT do not like wearing their solid AFOs especially
when there are greater requirements for ankle push off such as in running
not suit all due to the variation in presentation and degree of severity of pathology
and the variety of orthosis design and purpose. Although the solid AFO design may
provide excellent support for excessive ankle dorsiflexion in terminal stance due to
weakness of the plantar flexors, in some cases it may overly restrict ankle movement
and thus reduce the ankle power generation (Fig. 7a, b). For those persons with CMT
who are more functional, this restriction in motion will have a negative impact on
push-off especially during running. The posterior leaf spring orthosis (PLS) which
has less support through trim lines posterior to the malleoli may result in less than
adequate support in some patients but function as a solid AFO in others depending
on brace stiffness, patient body weight, and plantar flexor strength which may be
decreasing over time due to the disease progression.
We have recently reviewed AFO outcomes in children and adolescents with CMT
and have shown that different AFOs (solid AFO vs. PLS), although intended to have
differences in terms of function based upon their design, do not always operate as
proposed. In many cases, the PLS design functions as a solid AFO and restricts most
ankle motion, and in others the PLS does not provide adequate support and allows
a b
25 30
Barefoot
Barefoot
PLS AFO
Solid AFO 25
20
Peak Dorsiflexion (deg)
20
15
15
10
10
5 5
0 0
1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 6 7 8 9 10
Number of Sides Number of Sides
Fig. 8 Comparison peak ankle dorsiflexion during stance for barefoot versus AFO for the (a) solid
AFO and (b) PLS AFO. The red line indicates typically developing peak dorsiflexion in stance. The
large variation in values for peak dorsiflexion in terminal stance in both brace designs confirms the
heterogeneous nature of CMT in terms of gait outcomes. In some cases, the brace design provided
improvement and in others made no change and in other resulted in a worse outcome
ongoing excessive ankle dorsiflexion. As for the solid AFO design, in some cases, it
functions like a PLS and does not provide adequate support (Fig. 8a, b). It is also clear
that hinged AFO designs, while supporting the ankle in swing and reducing excessive
plantar flexion, do not provide adequate support in stance (Fig. 9a, b). Understanding
orthosis function in persons with CMT using motion analysis has allowed us to clarify
why some orthosis designs are more effective than others and to highlight the
importance of understanding patient impairment and gait function during barefoot
walking that needs to be supplemented with the AFO. Additional research is needed
in this area to better match orthosis design with patient impairment. As well, there are
technical considerations when using motion analysis techniques in the assessment of
orthoses including marker placement and documentation of consistent ankle sagittal
plane angles in both barefoot and orthosis conditions (Ounpuu 1996).
Orthopedic Surgery
Orthopedic surgery may be required when patients with CMT have foot deformity and
pain that limits walking and performing activities of daily living. Surgery to address
the wide variety of issues varies from individual muscle procedures (lengthenings or
transfers) to complex combinations of soft tissue and bony interventions. The majority
of research evaluating surgical outcomes is based upon clinical examination findings
alone and, in some cases, only postoperative assessments. Motion analysis provides
an excellent opportunity to assist in surgical decision-making and to evaluate surgical
outcomes objectively. This is possible in those clinical settings where the standard of
care includes comprehensive motion analysis techniques both pre- and post-surgery.
These assessments should include an evaluation of clinical impairments (muscle
weakness, contracture, and bony deformity) which is integrated with gait findings
documented with dynamic EMG, joint and segment kinematics, joint kinetics, and
40 40
Dor Dor
deg deg
Pla Pla
-40 -40
2.0 2.0
Ext Ext
Nm/kg Nm/kg
Flx Flx
-1.0 -1.0
3.0 3.0
Gen Gen
W/kg W/kg
Abs Abs
-2.0 -2.0
25% 50% 75% 25% 50% 75%
a Gait Cycle b Gait Cycle c
Fig. 9 Comparison of ankle sagittal plane kinematic, moment and power for (a) barefoot and (b)
hinged AFO walking for the right side in a youth with CMT. Although the hinged AFO (c) can
provide medial/lateral stability there is no restriction on peak ankle dorsiflexion in stance and
therefore there is a continued reduction in peak ankle plantar flexor moment in stance and
improvement in stability is not improved in the sagittal plane
foot pressure data. This approach allows for the most informed treatment decisions
and objective evaluation of outcomes when the patient has adequately recovered. This
approach will allow new knowledge on a patient-by-patient basis and ultimately
through research when this protocol is followed over time. Pairing treatment
approaches with clear and specific indications at the level of impairment and function
is needed. When interpreting postoperative data, it is important to also consider
changes that might be due to the natural history of this progressive disorder.
Motion analysis has provided some insight into treatment indications from both
routine clinical use and research applications, but at this time, there is only one example
of comprehensive motion analysis outcomes used to assess treatment in this patient
population. It has been shown by Dreher (Dreher et al. 2014) that the transfer of the
posterior tibialis tendon can reduce excessive equinus in swing which can lead to
clearance issues and inappropriate prepositioning at initial contact. However, they
also found reduced peak ankle plantar flexion at push-off. Joint kinetic data would
provide additional insight into this finding related to plantar flexor function which is
also confounded by the possibility of increased weakness over time. Orthopedic
treatment outcomes have also been assessed using pedobarography with positive
outcomes in terms of foot pressures; however, these measures do not correlate well with
other aspects of gait (Metaxiotis et al. 2000). A study by Ward (Ward et al. 2008)
evaluated 25 patients with CMT who underwent treatment for cavus foot including
dorsiflexion osteotomy of the first metatarsal, transfer of the peroneus longus to the
peroneus brevis, plantar fascia release, transfer of the extensor hallucis longus to the
neck of the first metatarsal, and in selected cases transfer of the tibialis anterior tendon to
the lateral cuneiform. Temporal and stride parameters were analyzed and revealed that
those patients that had undergone the anterior tibialis tendon transfer spent less time in
double-support stance phase. Additional outcomes of ankle kinematics and kinetics in
the above study would have provided relevant information as to the causes of changes
in temporal and stride parameters to further clarify treatment indications. This would
have allowed confirmation of the surgical goals of addressing current deformity and
ultimately long-term recurrence with objective measures of ankle function. Although
these studies provide interesting findings, comprehensive motion analysis that includes
both kinematic and kinetic outcomes would be very useful.
Individual patient cases where motion analysis has been incorporated as the
standard of care can also provide important knowledge especially if this methodol-
ogy is followed over the long term. This is particularly important in rare diseases
when it takes time to gather a large data base of patients to study. For example,
motion analysis data for an individual patient has provided some insight into the
possible impact of the plantar fascia release for the correction of cavus deformity. In
a patient who is in excessive equinus in stance (toe walker) and swing due to limited
dorsiflexion range of motion (plantar aspect of the foot in relation to the tibia), a
plantar fascia release may provide sufficient reorientation of the foot anatomically to
allow for increased passive dorsiflexion range to eliminate the excessive equinus in
stance and swing (Fig. 10a, b). In a patient who has normal peak ankle dorsiflexion
in terminal stance but a significant cavus deformity, a plantar fascia release to correct
for the cavus and varus position may provide reorientation of the foot anatomically
to allow for excessive ankle passive dorsiflexion range of motion and excessive
dorsiflexion in terminal stance. This may lead secondarily to increased knee flexion
in stance (Fig. 11). Unmasked ankle plantar flexor weakness may in part play a role
in the downside of this procedure which may still be relevant to help with foot pain.
Appreciating the complex relationship between maximum dorsiflexion range of
motion, plantar flexor strength, patient body weight, and the extent of cavus defor-
mity is all more possible with motion analysis. Learning from these individual cases
is an important step toward the goal of improving treatment outcomes.
Future Directions
The application of comprehensive motion analysis techniques for clinical evaluation

as part of the standard of care and in research settings for persons with CMT is
relatively new compared to other complex gait pathologies such as cerebral palsy. As
a result there is still much to learn and new developments needed to optimize motion
measurement for this pathology. Future effort should focus on the following:
a b
40 40
Dor Dor
deg deg
Pla Pla
-40 -40
2.0 2.0
Ext Ext
Nm/kg Nm/kg
Flx Flx
-1.0 -1.0
3.0 3.0
Gen
W/kg W/kg
Abs
-2.0 -2.0
25% 50% 75% 25% 50% 75%
Gait Cycle Gait Cycle
Fig. 10 Comparison of the sagittal plane ankle kinematics, moments and powers for (a) pre and
(b) several years post orthopaedic lengthening of the plantar fascia in an child with CMT. Three gait
cycles on the right side are plotted for each condition. Improvements in ankle sagittal plane motion
in stance and swing and ankle moment modulation due to a heel initial contact without reduction in
ankle power generation are noted. Objective documentation of the outcome of this procedure using
motion analysis can help to clarify surgical indications
(a) Application of a more comprehensive foot model is needed to better understand

the complex relationship between cavus deformity, plantar flexor strength, and
passive ankle dorsiflexion range of motion. The foot model needs to include
measurement of the extent of cavus and may need to incorporate radiographic
information to allow for the most accurate assessment. Understanding how the
extent of cavus deformity impacts available plantar flexor length and masks
weakness is critical for treatment decision-making at the ankle/foot in these
patients.
Fig. 11 Comparison of the Knee Flexion-Extension

knee and ankle sagittal plane 80
kinematics for pre (dashed)
versus post (solid)
orthopaedic surgical outcome Flx
for plantar fascia release.
Increased peak ankle
dorsiflexion and associated
knee flexion may be due to
multiple causes: (a) ongoing
disease process of increasing Ext
weakness of the plantar
flexors over time and/or
(b) unmasking of existing
-20
ankle plantar flexor weakness
through increased dorsiflexion Plantar-Dorsiflexion
range of motion 40
Dor
deg
Pla
-40
25% 50% 75%
Gait Cycle
(b) The presentation of CMT is heterogeneous with varying clinical findings,

severity, as well as disease progression. Establishing if there is any link between
impairments and associated gait function with phenotype will provide a better
ability to determine prognosis for future function at the time of diagnosis and
lead to treatment guidelines based upon phenotype.
(c) CMT is a progressive disease that typically results in increasing weakness and
associated gait issues over time. Disease progression, however, is patient depen-
dent and not a lot is known about expectations for decline for a given individual
with CMT. Long-term natural progression studies are needed which hopefully
can be linked eventually to phenotype. Achieving this goal is difficult as disease
progression in many is not rapid so long-term studies are needed.
(d) Treatment of gait issues for this patient population focuses on improving or
maintaining gait function and reducing foot/ankle pain. There is very limited
objective documentation of orthopedic and other treatment outcomes such as
bracing in terms of comprehensive motion analysis. Systematic reviews of
surgical outcomes are needed to assist in identifying specific treatment
indications and expectations. This research is further complicated by disease

progression which needs to be taken into account.
(e) Finally, establishing a diagnosis of CMT is difficult, and in the initial phases, it is
often confused with other diagnoses such as idiopathic toe walking to cavo-varus
deformity. Initial steps include a detailed clinical assessment and should also
include a family history. If there is a suspected CMT diagnosis, often genetic
testing and nerve conduction tests are recommended to confirm a diagnosis and
phenotype. These later tests are often not completed due to expense, and nerve
conduction tests can be painful. Therefore, establishing functional biomarkers
related to movement would be helpful in the initial stages of this disease to help
correctly diagnosis and therefore treat appropriately common initial findings of
cavo-varus foot deformity in some and toe walking in others. These biomarkers
may include any or a combination of the following: ankle kinematic or kinetic
variables, EMG signal analysis, and muscle impedance assessments. Additional
investigation is needed to determine if these or other biomarkers can be
established.
Cross-References

References
Burns J, Crosbie J, Hunt A, Ouvrier R (2005) The effect of pes cavus on foot pain and plantar
pressure. Clin Biomech (Bristol, Avon) 20:877–882
Burns J, Crosbie J, Ouvrier R, Hunt A (2006) Effective orthotic therapy for the painful cavus foot: a
randomized controlled trial. J Am Pediatr Med Assoc 96:205–211
Burns J, Raymond J, Ouvrier R (2009) Feasibility of foot and ankle strength training in childhood
Charcot-Marie-Tooth disease. Neuromuscul Disord 19:818–821
Burns J, Ouvrier R, Estilow T, Shy R, Laura M, Eichinger K, Muntoni F, Reilly MM, Pareyson D,
Acsadi G, Shy ME, Finkel RS (2012) Symmetry of foot alignment and ankle flexibility in
paediatric Charcot-Marie-Tooth disease. Clin Biomech (Bristol, Avon) 27:744–747
Charcot JM (1886) Sue une forme particulaire d'atrophie musculaire progressive souvent familial
debutant par let pieds et les jambes et atteingnant plus tard les mains. Rev Med Paris 6:97–138
Chetlin RD, Gutmann L, Tarnopolsky M, Ullrich IH, Yeater RA (2004) Resistance training
effectiveness in patients with Charcot-Marie-Tooth disease: recommendations for exercise
prescription. Arch Phys Med Rehabil 85:1217–1223
Crosbie J, Burns J, Ouvrier RA (2008) Pressure characteristics in painful pes cavus feet resulting
from Charcot-Marie-Tooth disease. Gait Posture 28:545–551
Don R, Serrao M, Vinci P, Ranavolo A, Cacchio A, Ioppolo F, Paoloni M, Procaccianti R,
Frascarelli F, De Santis F, Pierelli F, Frascarelli M, Santilli V (2007) Foot drop and plantar
flexion failure determine different gait strategies in Charcot-Marie-Tooth patients. Clin Biomech
(Bristol, Avon) 22:905–916
Dreher T, Wolf SI, Heitzmann D, Fremd C, Klotz MC, Wenz W (2014) Tibialis posterior tendon
transfer corrects the foot drop component of cavovarus foot deformity in Charcot-Marie-Tooth
disease. J Bone Joint Surg Am 96:456–462
Ferrarin M, Bovi G, Rabuffetti M, Mazzoleni P, Montesano A, Pagliano E, Marchi A, Magro A,
Marchesi C, Pareyson D, Moroni I (2012) Gait pattern classification in children with Charcot-
Marie-Tooth disease type 1A. Gait Posture 35:131–137
Garcia A, Combarros O, Calleja J, Berciano J (1998) Charcot-Marie-Tooth disease type 1A with
17p duplication in infancy and early childhood: a longitudinal clinical and electrophysiologic
study. Neurology 50:1061–1067
Holmes JR, Hansen ST Jr (1993) Foot and ankle manifestations of Charcot-Marie-Tooth disease.
Foot Ankle 14:476–486
Jani-Acsadi A, Ounpuu S, Pierz K, Acsadi G (2015) Pediatric Charcot-Marie-Tooth disease. Pediatr
Clin N Am 62:767–786
Kuruvilla A, Costa JL, Wright RB, Yoder DM, Andriacchi TP (2000) Characterization of gait
parameters in patients with Charcot-Marie-Tooth disease. Neurol India 48:49–55
Mandarakas M, Hiller CE, Rose KJ, Burns J (2013) Measuring ankle instability in pediatric
Charcot-Marie-Tooth Disease. J Child Neurol 28:1456–1462
Metaxiotis D, Accles W, Pappas A, Doederlein L (2000) Dynamic pedobarography (DPB) in
operative management of cavovarus foot deformity. Foot Ankle Int 21:935–947
Newman CJ, Walsh M, O’sullivan R, Jenkinson A, Bennett D, Lynch B, O’brien T (2007) The
characteristics of gait in Charcot-Marie-Tooth disease types I and II. Gait Posture 26:120–127
Ounpuu S (1996) An evaluation of the posterior leaf spring orthosis using joint kinematics and
kinetics. J Gerontol A Biol Sci Med Sci 16:378–384
Ounpuu S, Garibay E, Solomito M, Bell K, Pierz K, Thomson J, Acsadi G, Deluca P (2013) A
comprehensive evaluation of the variation in ankle function during gait in children and youth
with Charcot-Marie-Tooth disease. Gait Posture 38:900–906
Phillips MF, Robertson Z, Killen B, White B (2011) A pilot stud of a crossover trial with
randomized use of ankle-foot orthoses for people with Charcot-Marie-Tooth disease. Clin
Rehabil 26:534–544
Ramdharry GM, Day BL, Reilly MM, Marsden JF (2009) Hip flexor fatigue limits walking in
Charcot-Marie-Tooth disease. Muscle Nerve 40:103–111
Ramdharry GM, Day BL, Reilly MM, Marsden JF (2012) Foot drop splints improve proximal as
well as distal leg control during gait in Charcot-Marie-Tooth disease. Muscle Nerve 46:512–519
Rose KJ, Burns J, North KN (2010) Factors associated with foot and ankle strength in healthy
preschool-age children and age-matched cases of Charcot-Marie-Tooth disease type 1A. J Child
Neurol 25:463–468
Smith BG (2002) Hereditary sensory motor neuropathies. In: Fitzgerald RH, Kaufer H, Malkani AL
(eds) Orthopaedics. Mosby, Missouri
Thomas PK (1999) Overview of Charcot-Marie-Tooth disease type 1A. Ann N Y Acad Sci 883:1–5
Tooth HH (1886) The peroneal type of progressive muscular atrophy. MD thesis, University of
Cambridge
Vinci P, Perelli SL (2002) Footdrop, foot rotation, and plantarflexor failure in Charcot-Marie-Tooth
disease. Arch Phys Med Rehabil 83:513–516
Vinci P, Serrao M, Pierelli F, Sandrini G, Santilli V (2006) Lower limb manual muscle testing in the
early stages of Charcot-Marie-Tooth disease type 1A. Funct Neurol 21:159–163
Ward CM, Dolan LA, Bennett DL, Morcuende JA, Cooper RR (2008) Long-term results of
reconstruction for treatment of a flexible cavovarus foot in Charcot-Marie-Tooth disease. J
Bone Joint Surg Am 90:2631–2642
Motor Patterns Recognition in Parkinson’s
Disease
Pierpaolo Sorrentino, Valeria Agosti, and Giuseppe Sorrentino
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Objective Assessment of the Motor Cardinal Symptoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Tremor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Rigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Bradykinesia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Integrated Motor Functions: Gait and Postural Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Technological Approaches to Gait and Posture Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Gait in PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Posture in PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Abstract
Parkinson’s disease (PD) is characterized clinically by main motor symptoms
such as tremor at rest, rigidity, and bradykinesia that affect movements, including
gait and postural adjustments. The diagnosis is based on the clinical recognition
of these symptoms with the consequent high interrater variability. In order to
perform an objective and early diagnosis, approaches that overcome the limita-
tions inherent to clinical examination are needed. In the present work, we will
describe several classical technological approaches, such as 3D motion analysis,
P. Sorrentino
Department of Engineering, University of Naples Parthenope, Naples, Italy
e-mail: ppsorrentino@gmail.com; ppsorrentino@yahoo.com
V. Agosti (*) • G. Sorrentino
Department of Motor Sciences and Wellness, University of Naples Parthenope, Naples, Italy
Institute Hermitage-Capodimonte, Naples, Italy
e-mail: valeria.agosti@uniparthenope.it; giuseppe.sorrentino@uniparthenope.it

DOI 10.1007/978-3-319-30808-1_64-1
2 P. Sorrentino et al.
to achieve an objective evaluation of the cardinal motor symptoms in

PD. Furthermore, we will take into account the attempts to identify pathological
patterns of integrated, more complex functions such as gait and posture. Finally,
as future directions, we will discuss the machine learning approaches in the
individuation of specific gait patterns in PD.
Keywords
Parkinson’s disease • Movement pattern • Gait analysis • Machine learning •
Clinical scales • Gait disorders • Postural instability
Introduction
Idiopathic Parkinson’s disease (PD) is a progressive multisystem neurodegenerative

disease with a prevalence in industrialized countries estimated at 0.3 % of the entire
population and about 1 % in people over 60 years of age (de Lau and Breteler 2006).
Neuropathologically, PD is characterized by the degeneration of dopaminergic
nigrostriatal neurons with deposits of insoluble polymers of alpha synuclein forming
cytoplasmic inclusions called Lewy bodies.
The diagnosis of PD is based on the clinical observation. The clinical symptoms
are usually dominated by the motor disturbances. The main motor feature to
diagnose PD is the slowness of voluntary movements; one additional symptom,
either muscular rigidity, resting tremor, or postural instability, should be present. A
number of nonmotor symptoms may also be present. However, nonmotor features
have extremely low specificity, limiting their application in clinical diagnosis.
The sole clinical evaluation may be source of possible interrater variability and
mistakes. In order to perform an objective diagnosis, approaches that overcome the
limitation inherent to clinical examination are needed.
In this work, firstly we will discuss the methodological issues to take into account
in order to use properly the available technological approaches. Furthermore, we will
consider the attempts to identify pathological patterns of more complex functions
that require the integration of many motor and cognitive skills, such as gait and
posture. Finally, as future directions, we will discuss the machine learning
approaches in the individuation of specific gait patterns in PD.
State of the Art
In 1817, James Parkinson, in the article An Essay on the Shaking Palsy, first
described the main clinical manifestations of the disease that now bears his name.
In the work, based entirely on the personal visual observations of six patients,
Parkinson stated “Involuntary tremulous motion, with lessened muscular power, in
parts not in action and even when supported; with a propensity to bend the trunk
forward, and to pass from a walking to a running pace: the senses and intellects being
Motor Patterns Recognition in Parkinson’s Disease 3
uninjured” (Walshe et al. 1961). As of today, except for a few new approaches (e.g.,
DaTscan) that have only a supportive value, the diagnosis of PD remains based on
the clinical observation. The disease is usually diagnosed by the first motor symp-
toms according to the criteria from the UK PD Brain bank (Hughes et al. 1992). The
cardinal motor feature that physicians look at when diagnosing PD is the slowness of
initiation of voluntary movements with progressive reduction in speed and ampli-
tude of repetitive actions (bradykinesia); one additional symptom, either resting
tremor, muscular rigidity, or postural instability, should be present (Hughes et al.
1992). However, differently from Parkinson’s opinion (the senses and intellects
being uninjured), there are a number of nonmotor features that can be detected
clinically or anamnestically earlier then the motor ones. Interestingly, nonmotor
symptoms might show a much earlier onset (Schrag et al. 2015). However, nonmotor
features have extremely low specificity, limiting their application in clinical practice.
In the attempt to overcome the limitations linked to the subjective observation,
clinical scales have been adopted, the Unified Parkinson’s Disease Rating Scale
(UPDRS) part III being the most widely used (Gelb et al. 1999). Although currently
considered the gold standard, clinical scales are flown on two different levels. Firstly,
interrater variability is a source of possible mistakes and confounders (Shulman et al.
2016). On a more profound level, the detectability of clinical features depends on
both the magnitude of the clinical features and the strength that the rater can feel. A
proposed relationship between these two quantities can be estimated according to the
Webener-Fetchner law (Dehaene 2003; Nieder and Miller 2003), whereby the
intensity of the stimulus and the minimum perceivable variation would be related
by a logarithmic relationship. This implies that a logarithmic variation of the
stimulus would relate to an additive progression of the perception of the stimulus.
Hence, in order to perform early diagnosis and stratification of patients, there is
the necessity to have an objective and precise approach that overcomes the limitation
inherent to clinical examination.
Objective Assessment of the Motor Cardinal Symptoms
Tremor
Tremor is usually defined as an involuntary, rhythmic, oscillatory movement of a

body part (Daroff et al. 2016). Clinically, tremors are classically divided into rest,
postural, and action tremors, according to its semiologic features. Typically, PD
patients show rest tremor, characterized by the involvement of agonist and antago-
nist muscles, with a frequency between 4 and 6 Hz. Since the rating of tremor does
not relate with the ratings of bradykinesia and rigidity (Martinez-Martin et al. 1994),
it has been hypothesized that different pathophysiological mechanisms underlie
these symptoms. Indeed, patients with tremorigen form of PD tend to have a milder
course of the disease as compared to the ones with a rigid onset (Obeso et al. 2010).
Hence, careful classification might yield direct information on specific central
structures. Typically, PD shows the pattern of a central tremor, whose frequency is
function of neither the limb properties (i.e., joint inertia, stiffness of the soft tissues)
nor of the peripheral nervous system (loop–reflex time) (Elble and Deuschl 2011).
Classically, the tremor is analyzed in order to help diagnosis, and its rating is part
of various clinical scores used to monitor disease progression and to modify the
therapy accordingly. In section III of the UPDRS, several items evaluate the tremor
and its main characteristics. However, according to the above mentioned
Webener–Fatchner law, the clinical evaluation of tremor is inherently bounded to
the rater’s perception of the frequency and amplitude, and several attempts have been
performed to achieve an objective evaluation of tremor.
Classically used instruments are electromyography (EMG), accelerometers, gyro-
scopes, flexible angular sensors and goniometers, video and optoelectronic devices,
force sensors, and wearable orthosis.
First off, it has to be noticed how the choice of the device should be in accordance
with the features that we want to capture and that the interpretation of the data should
take into account the influence of the device on the measurements. For instance,
devices that restrain the range of movement of the joint hinder the interpretation of
amplitude while allowing for the simulation of isometric conditions. In this case, it
has been showed that the frequency of the tremor is not influenced by the constraints,
while the amplitude is. This would point to the idea that frequency is more directly
related to the features of the tremor generators, while amplitude might by affected by
the mechanics of the limb.
The typical characteristic of the tremor signal that are studied in PD are:
(a) frequency spectrum (by Fourier transform of the time series) (Timmer et al.
1996); (b) peak frequency (Gao 2004); (c) temporal fluctuation; (d) root mean square
of linear acceleration; (e) approximate entropy; (f) shape of signal distribution;
(g) wavelet coefficient; and (h) higher order statistics parameters (Thanawattano
et al. 2015).
Rigidity
The term rigidity defines an increase in muscle tone involving both agonist and
antagonist muscles of the same joint (Daroff et al. 2016). In PD rigidity is one of the
main clinical signs that concurs to impair the ability to move about freely. Although
the underlying mechanisms are poorly understood, rigidity is a precise indicator of
the residual motor function, and its detection and monitoring appears crucial since it
is responsive to dopaminergic treatments.
The part III of the UPDRS includes the rigidity evaluation. The neurologist
performs manual maneuvers of passive flex-extension of the upper and lower
limbs as well as of the trunk to estimate, subjectively, the rate of resistance or
stiffness. However, similarly (may be more) to the tremor evaluation, interrater
variability is a source of possible mistakes.
To date, to assess muscle tone, surface EMG has been widely used, classically
performed during dynamic flex/extension of the limb (Mortimer and Webster 1979;
Xia et al. 2011). The methods to evaluate EMG signals are usually based on the
analysis of the amplitude and spectrum (Nieminen and Takala 1996). However,
given the inherent nonlinear nature of the EMG signals, great effort has been put into
the fractal-based dimensionality analysis as well as nonlinear time series analysis
(Swie et al. 2005; Meigal et al. 2013). However, surface EMG signals are strongly
influenced by the characteristics of the soft tissues. Indeed, the correlation between
EMG characteristics and UPDRS-based evaluation of rigidity has shown not to be as
strong as expected. In order to overcome these limitations, many studies have
focused on purely biomechanical measures, such as peak torque, elastic coefficient,
and mechanical impedance. Despite these techniques yielded improved correlations
as compared to EMG-based analysis, they do not allow the distinction between the
rigidity induced by neurologically mediated mechanisms from that derived by
altered mechanical properties of the joints (Endo et al. 2009). Interestingly, Park
et al. (2011) proposed a model that might allow the distinction between these two
components of rigidity.
Bradykinesia
Bradykinesia is defined as slowness of movement, often accompanied by other

features such as hypokinesia (reduced amplitude of movement) (Daroff et al.
2016). Bradykinesia is a very complex sign, whose pathophysiological mechanisms
are only partially understood. Furthermore, the patients present a general impover-
ishment of the movements including hypomimia, micrographia, and speech alter-
ations (Berardelli et al. 2001). Concomitant causes that might lead to bradykinesia
include: muscle weakness, rigidity, tremor, movement variability, and bradyphrenia.
The severity of bradykinesia is classically measured with different tasks involving
rapid and voluntary movements. In UPDRS III, the tasks for the upper limb include
the finger tapping (patient taps thumb with index finger in rapid succession), the
hand movement (patient opens and closes hands in rapid succession), and rapid
alternating movement of hand (prono-supination of the hands). The tasks for the
lower limbs include leg agility (patient taps heals on the ground in rapid succession
picking up entire leg).
In order to improve the reliability of such estimates, several studies investigated
the use of accelerometers (Yokoe et al. 2009; Stamatakis et al. 2013) or magnetic
coils (Kandori et al. 2004; Shima et al. 2008) in measuring finger tapping. Yokoe
et al. (2009) analyzed 14 parameters of finger tapping and used a statistical approach
called principal component analysis (PCA) to reduced them into three components,
of which the first (velocity and amplitude related) and the third (rhythm related)
could differentiate between patients and controls. Kandori et al. (2004) used mag-
netic coil to analyze the finger taping. After filtering the signal, they found out that
the average amplitude of each waveform decreased proportionally to the Hoehn and
Yahr staging system (Hoehn and Yahr 1967).
However, both these clinical and instrumented measurement do not capture the
complex and multifactorial nature of this clinical feature. Furthermore, these inves-
tigations are to be performed in advanced clinical settings and in a strictly controlled
environment, hence they do not provide information on the movement of the patient
in daily activity. A possible solution to bridge this gap might be the new technolog-
ical applications such as wearable sensors tools, which allow us to check the patient
daily movements. Focusing on two experimental tools and conditions, Salarian et al.
(2007) experimented the use of gyroscopes in quantifying bradykinesia. In particu-
lar, they used a self-made micro-gyroscope unit, easy to wear and to use, to detect
parameters of the movement and their correlation with the UPDRS III scores. The
author showed that bradykinesia, as detected by the analysis of the entire recorded
period, related with the UPDRS III bradykinesia subscore. Another interesting and
more recent study analyzed the joint movement in the upper limb (finger and
forearm) but also in the lower limb (ankle) by using a gyro-sensors system (Kim
et al. 2015). In this study, 14 features were derived from the sensors and a stepwise
multiple linear regression analysis was used to develop a model able to predict the
bradykinesia score. Interestingly, the regression models were better than any single
variable in predicting the clinical scores.
Integrated Motor Functions: Gait and Postural Stability
According to Jacksonian’s model (Walshe et al. 1961), the nervous system is

organized in three levels of increasing complexity. In this light, movements such
as gait and postural adjustments may be considered as the result of the integration of
the spinal cord (lower level), basal ganglia and motor cortex (middle level), and
prefrontal cortex (higher level). Hence, from the functional stand point, gait and
balance are the results of the integration of different mechanisms ranging from the
automatic reflexes (spinal cord) to the cognitive functions (prefrontal cortex). Con-
sequently, the recognition of motion patterns is a challenging issue since gait and
posture can be influenced by the (mis)functioning of any of those structures.
The gait describes human locomotion, or, according to what stated above, the way
of walking. Since each individual has distinctive gait patterns it is useful to divide
them into gait cycles, defined as the interval from the initial placement of the
supporting heel on the ground to the moment when the same heel touches the ground
again. Classically, the gait cycle is further divided into eight phases (Perry et al.
2010). This division is useful since each phase has a specific functional meaning, and
this helps the clinical interpretation of the patterns. Hence, by evaluating individual
gait patterns, we can determine specific weaknesses and/or tailor rehabilitation
programs. Furthermore, when the gait is affected by injury or disease, it can show
some features/patterns that are common to all affected individuals (Perry et al. 2010;
Levine et al. 2012).
The postural stability emerges from activation and integration of multiple
sensory, motor, and cognitive systems. The complex patterns of coordinated
activity that are needed to achieve postural stability are normally referred to as
postural control (Massion 1992; Cappa et al. 2008). In other words, postural control
could be defined as the ability to maintain stability of the body and its segments.
Postural control is crucial to maintain the body’s orientation in the space, the body’s
center of mass (CoM) over the base of support (BoS), and to stabilize the position
of the head.
From the neuromechanical perspective, the central nervous system interprets the
inputs from the various subsystems and organizes appropriate responses based on
previous and current information (Enoka 2015). These responses could be either
compensatory or anticipatory. The formers occur as reactions to external forces that
displace the body’s CoM, and the latters anticipate internally generated destabilizing
forces such as those induced, for instance, by raising arms or bending forward.
Català et al. (2016) using a motion analysis system evaluated the dynamic stability
control during gait in PD patients. They estimated the adaptability and the postural
strategies to maintain balance during locomotion. Interestingly, they found that the
compensatory responses (called “reactive” by the authors) are specifically impaired
in PD, as opposed to anticipatory one. In order to explain the strict relationship
between gait and postural stability, it is important to highlight that one of the three
main tasks of the gait cycle is “to ensure the advancement maintaining dynamic
stability.” Indeed, in order to achieve balance during static or dynamic tasks, a
perfect integration between CoM position and velocity is required (Jian et al.
1993). This ability is often lost in neurological diseases (i.e., PD), predisposing to
tripping and falling during walking (Ashburn et al. 2001).
Technological Approaches to Gait and Posture Assessment
Gait parameters can be described as kinetic and/or kinematic data. On the one hand,
the kinetic data is useful to describe the forces acting on the joints, the moments
produced by the muscles crossing the joints, and the energy required during gait. The
kinetic data usually results from a force plate, providing information about the
components of the ground reaction force (GRF) that are in turn proportional to the
magnitude and direction of the load applied to one or both feet when the body
touches the ground. On the other hand, kinematic data helps us to study the timing
and the magnitude of gait. In fact, kinematic data provides information about the
phases of gait (spatiotemporal data) and/or about the three components of the
articular range of motion (RoM). Kinematic data are usually derived from video
analysis (two-dimensional data) or motion analysis (three-dimensional data, 3DMA)
or even from special sensorized mats.
On a practical standpoint, we can divide the devices that are eligible to study the
gait in those that are to be used in specialized settings (i.e., laboratory) and those that
can be easily used in a day to day context. The information that can be obtained in a
controlled setting are better in quality and quantity, while the data from the living
environment of the patients can provide valuable insight on the impact of the
impairment (or the therapy) on the daily life of the patient.
Integrated motion analysis systems allow to collect simultaneously kinematic,
muscular, and kinetic data from stereophotogrammetric cameras, EMG, and force
plate, respectively. Morris et al. (1999, 2005) by means of a motion analysis
approach described the spatial and temporal dysfunctions in PD gait that is
characterized by decreased velocity, shorter stride, increased cadence, stride length

and stride time variability, altered heel strike, toe-off, and arm swing. Multiple
studies on kinematic changes in PD showed decreased ankle RoM in pre-swing,
decreased hip extension in pre-swing, and reduced power at ankle push off and hip
pull off (Sofuwa et al. 2005; Švehlík et al. 2009; Roiz et al. 2010). 3DMA is a
powerful approach in validating rehabilitation strategies as well. For instance, a
significant improvement of both spatiotemporal (Vitale et al. 2012) and kinematic
parameters (Agosti et al. 2016) has been demonstrated in PD patients following a
global postural reeducation program.
Despite the fact that the 3DMA remains the gold standard in the motor evaluation
of PD (Sale et al. 2013), this technique has been criticized since it might induce
biases due to the laboratory setting. In the experience of the authors, PD specific
alterations, such as fatigue or altered perspiration (which makes it hard to attach the
markers), may specifically bias the analysis of these patients.
To overcome these problems, further laboratory instruments have been developed
such as GAITRite® (McDonough et al. 2001). GAITRite® is a portable sensorized mat
built for automated measurement of spatiotemporal gait parameters in daily ambula-
tory use. Nelson et al. (2002) used this tool to assess footfall patterns and selected gait
characteristics in early PD. Recently, with the advent of new technological solutions,
the use of wearable sensorized tools is gaining ground not only in the outpatient use
but also for home monitoring. GAITRite® was used to validate the measurement of
these wearable systems. Indeed, Lin et al. (2016) proposed a new and useful portable
monocular image system to track and analyze parkinsonian gait patterns using a
centroid tracking algorithm (a system which can be easily used in locations with
budget and space limitations). Lord et al. (2008) tested the validity of a five acceler-
ometer portable system (the Vitaport Activity Monitor® (VAM)) in assessing postural
transitions and dynamic mobility during daily life activities in PD patients.
In order to extend the analysis to the total body, Tzallas et al. (2014) presented a
new wearable multi-sensor monitor unit composed by tri-axial accelerometers and
gyroscopes connected wirelessly. This kind of device is a step toward a
multiparameter approach to the monitoring of PD that might help to personalize
the treatment and medication schedules according to comprehensive data generated
in the daily life of the patient.
Static stability has been classically investigated by a postural clinical evaluation
or by a posturographic platform (Bloem et al. 2016). The UPDRS motor score has a
task called “postural stability” whereby response to sudden posterior displacement is
evaluated. Finally, dynamic postural stability is usually evaluated by means of a
complex test, the Time Up and Go Test (TUG), which requires the ability of standing
up from a chair, walking along 3 m, turning, and finally sitting down. An
instrumented version (iTUG) of this scale was recently proposed on early-mild PD
subjects using a single tri-axial accelerometer worn on the lower back (Palmerini
et al. 2013). In the last few years, there is an increasing body of evidence, confirmed
by systematic review (Hubble et al. 2015), that wearable sensors are useful systems
not only in evaluating gait patterns but are also excellent screening tools for balance
and postural instability.
Gait in PD
Parkinsonian gait has a clearly identifiable pattern when the neurological involve-
ment is advanced, including features such as festination and freezing of gait (FoG)
(Grabli et al. 2012). The former expresses “the tendency to move forward with
increasingly rapid, but ever smaller steps, associated with the Centre of gravity
(CoG) falling forward over the stepping feet” (Bloem et al. 2004). The latter defines
“the brief, episodic absence or marked reduction of forward progression of the feet
despite the intention to walk” (Giladi and Nieuwboer 2008). The link between
festination and FoG is still debated and is a topic of growing interest to understand
gait impairment mechanisms in PD (Nutt et al. 2011).
Further typical characteristics that might affect gait include a range of disorders
globally known as axial motor impairments (such as lateral trunk flexion, aka PISA
syndrome (PS), and camptocormia) (Doherty et al. 2011) as well as cognitive decline
(Ebersbach et al. 2013). However, early detection of typical patterns, when the
alterations are subtle, is a challenging issue. The picture becomes even more
complex considering that PD involves compensatory mechanisms to maintain an
effective gait (i.e., nonautomatic functioning can initially compensate for the
misfunctioning of normally automated processes) (Ebersbach et al. 2013). Interest-
ingly, some insight can be gained from the study of gait in carriers of genetic
mutations causing PD. These patients carry in their genome a mutation that causes
PD, yet being clinically healthy. These patients showed higher variability of the
stride only during a challenging condition (either during walking at high speed or
during a dual task). These results are intriguing since they might show that chal-
lenging tasks might unmask deficits by preventing the use of further cognitive
resources (compensative mechanisms) (Mirelman et al. 2011). In this line of think-
ing, patients with Gaucher disease type 1, a condition that is classically
nonneurological yet carrying a higher risk of developing PD, showed reduced
amplitude of the RoMs of the lower limbs (Sorrentino et al. 2016), again suggesting
that alterations may be present before clinical onset.
Bonora et al. (2015) investigated gait initiation in PD. This is especially interest-
ing, since the most common form of FoG, known as “start hesitation,” happens when
the patient wishes to start walking. Furthermore, from a biomechanical point of view,
the postural adjustments that anticipate the initiation of walking are fundamental to
achieve stability. Using a system based on inertial sensors and a force plate, the
authors showed that PD patients had reduced medio-lateral CoP displacement and
concluded that the PD patients might have more difficulties in adjusting the imbal-
ance phase during gait initiation.
PS syndrome is defined as a “marked lateral tilt of the trunk, typically reducible
by passive mobilization or when the patient lies down” (Doherty et al. 2011). A
recent pilot observational cross-sectional study (Geroin et al. 2015), using both
posturography and gait analysis data, estimated the impact of PS in PD patients.
They showed that in PD patients with PS, as compared to PD patients without PS and
age matched healthy controls, there was significantly greater body sway velocity in
the antero-posterior and medial-lateral directions.
Finally, mild cognitive impairment (MCI), that can be present in PD patients, may
have an effect on gait patterns. For instance, comparable groups of PD patients with
or without MCI showed different gait patterns, such as reduced step length and
swing time. Furthermore, the worsening of gait parameters was more pronounced in
PD patients with MCI than in patients without MCI (Amboni et al. 2012). A recent
cross-sectional study (Kelly et al. 2015) demonstrated the association of specific
motor dysfunctions with cognitive decline in a large PD cohort. The authors showed
that the severity of motor symptoms of postural instability is linked to dysfunction in
specific cognitive domains (executive function, memory, visuospatial function, and
phonemic fluency).
Posture in PD
Postural instability, a condition that is poorly responsive to levodopa, is a risk factor

in predisposing PD patients to falls (Grabli et al. 2012). There are a number of
clinical functional balance tests to evaluate postural stability, i.e., the Tinetti scale,
the functional reach test, one leg stance test, the postural balance task on the UPDRS
III, etc. (Mancini and Horak 2010). Clinical balance tests (especially when com-
bined) allow a good prediction of fallers and nonfallers in the PD population (Jacobs
et al. 2006) although with margin for improvement.
Often the technological approach is useful to determine the underlying causes of
the balance loss. However, when measuring in a clinical setting, it is important to
bear in mind that consensus has been reached on some standard requirements
(Scoppa et al. 2013). Static posturographic assessment showed greater postural
imbalance in PD patients as compared to healthy controls (Doná et al. 2015). Further
improvement can be obtained using dynamic posturography, which is the
posturographic assessment during postural perturbations. It was found that in PD
the forward bending induced by backward perturbation is impaired. Interestingly,
after repetitive perturbation, the postural reaction of the PD patients improved as
much as that of the healthy controls. This has implications for rehabilitation and is an
example of the further insight achievable using dynamic posturography (Visser et al.
2010). Lastly, using an inertial–sensor based system, Rocchi et al. (2013) computed
postural measures to classify PD subtypes. Furthermore, they showed that postural
displacement measures might capture different compensative mechanisms.
Future Directions
As we have mentioned, PD is a very heterogeneous disease, and in real world

scenarios patients are likely to face a number of symptoms. Furthermore, the
presence of a given set of symptoms has important implication for therapeutic
purposes (i.e., patients with rigid form of PD will have a heavier disease load as
well as a poorer and shorter response to drug therapy). Hence, a variety of statistical
techniques, that might be classified as machine learning, have been used in order to
spot patterns of symptoms that might lead to diagnosis.
Machine learning approaches include a bunch of techniques roughly involving
the following steps:
1. Define the relevant features (either using prior knowledge or using data driven
algorithms)
2. Select an appropriate algorithm to identify statistically consistent patterns among
the selected features in a part of the population in analysis (training set)
3. Use the training set to identify classes (or to allocate elements in previously
defined classes)
4. Test the performance by allocating new elements (test set) that the algorithm has
never met before in accordance to the classification criteria defined on the training
set
It appears evident how such techniques can be used to exploit typical signs and
symptoms of the disease in order to classify a subject, for instance, as affected or not
affected. The most widely used devices to obtain data are accelerometer, gyroscopes
and, less often, EMG recording. Once the signal has been obtained, it is important to
choose how to analyze it. Various kind of parameters have been used to analyze time
series, such as spectral features, skewness, curtosis, fractal features, entropy, and
information theory–based metrics. Furthermore, a great effort has been put into the
selection of the most effective algorithms in order to classify the features. Attempts
have been made to apply neural networks, hidden Markov models, support vector
machines, and decision trees.
Daliri (2012) applied force-sensitive resistors (corresponding roughly to the force
under the foot) to distinguish patients with Huntington’s disease, PD, and
amyotrophic lateral sclerosis. Classical parameters used in gait analysis were
retrieved, such as stride, swing, stance, and double support intervals. The features
extracted of these parameters were the minimum, maximum, average, and the
standard deviation. The most informative features were selected using a genetic
algorithm, and then they were used in a support vector machine in order to classify
the subjects. Various kernels were used, and the best results were obtained with the
radial basis function kernel. Sensitivity, specificity, and accuracy were around 90 %
for all the diagnosis. In a subsequent study (Daliri 2013), the difference between the
force measured in eight sensors placed underneath each foot was analyzed using a
short time Fourier transform. Subsequently, a support vector machine was applied
(with a kernel formed by the chi-square distance between the histogram of the
selected features), obtaining an accuracy of roughly 91 %.
Lee and Lim (2012) used a database of features of the gait to implement a
procedure to identify PD patients. The authors tried three different preprocessing
approaches to obtain characteristic of the signals that might capture known features
of the parkinsonian gait (i.e., shuffling gait). Then, using a wavelet-based feature
extraction, they selected 40 features and used them as an input in a neural network
with weighted fuzzy membership functions (NEWFM). The results varied according
to the preprocessing methods, with an accuracy of around 75 % for the three of them.
Tahir and Manap (2012) compared artificial neural network and support vector
machine in distinguishing gait pattern during self-selected speed walking, finding
comparable results for the two methods. Yunfeng and Krishnan (2010) estimated the
probability density functions (PDFs) of stride intervals. They found that the gait
variability was increased in PD patients. The stride interval parameters were used in
a support vector machine with polynomial kernel, and the results validated with the
leave–one–out approach. The classification rate had an area under the receiving
operator curve (AUC) of 0.952. Wahid et al. (2015) argued that true characteristics of
the gait pattern might be modified by other factors affecting gait, such as age, height,
body mass, and gender. To try to address this issue, they normalized gait parameters
before using machine learning approaches. In this case, the random forest yielded the
highest accuracy (92.6 %) as compared to support vector machine (80.4 %) and
kernel Fisher discriminant (86.2 %).
Bächlin et al. (2009) used a set of accelerometers on shank, thigh, and waist in
order to build a wearable device to identify FoG. They obtained a sensitivity of 73.1 %
and a specificity of 81.6 %.
Djuric-Jovicic et al. (2014) applied to the data obtained from an inertial sensor an
algorithm technique to automatically detect and classify episodes of FoG. Tripoliti
et al. (2013) explored how to combine a set of six accelerometers and two gyro-
scopes in order to classify episodes of FoG. The authors tried different algorithms
(Naïve Bayes, Random Forests, Decision Trees, and Random Tree) and concluded
that the best results were obtained applying a random forest to all the sensors (98.6
AUC). Interestingly, Cole et al. (2011) combined several accelerometers (forearm,
thigh, and shin) and one EMG channel (shin) and set out to identify episodes of FoG.
They applied a dynamic neural network and, when tested on experimental data from
PD patients, the algorithm showed a specificity of 97 % and sensitivity of 83 %.
All in all, it is possible to notice how machine learning techniques are able to
capture features of the gait and exploit them to either diagnose subjects or classify
symptoms. In the opinion of the authors, it comes to no surprise that very good
results are achieved with random forests, given the fact that these algorithms are
particularly prone to capture heavy nonlinearities such as those that are likely to arise
between features of gait.
Future direction might involve improved techniques of feature selection and/or
better classification algorithms as well as the integration of more patient character-
istics (anamnestic, clinical, imaging, gait analysis) into one classification algorithm.
This will mimic always more the clinical process that goes on in the mind of every
physician faced with a hard clinical diagnosis.
References
Agosti V, Vitale C, Avella D et al (2016) Effects of Global Postural Reeducation on gait kinematics
in parkinsonian patients: a pilot randomized three-dimensional motion analysis study. Neurol
Sci 37:515–522. doi:10.1007/s10072-015-2433-5
Amboni M, Barone P, Iuppariello L et al (2012) Gait patterns in parkinsonian patients with or

without mild cognitive impairment. Mov Disord 27:1536–1543. doi:10.1002/mds.25165
Ashburn A, Stack E, Pickering RM, Ward CD (2001) A community-dwelling sample of people with
Parkinson’s disease: characteristics of fallers and non-fallers. Age Ageing 30:47–52.
doi:10.1093/ageing/30.1.47
Bächlin M, Roggen D, Tröster G et al (2009) Potentials of enhanced context awareness in wearable
assistants for Parkinson’s disease patients with the freezing of gait syndrome. In: Proceedings –
international symposium on wearable computers, ISWC, IEEE, 4–7 Sep - Linz, Austria, pp
123–130
Berardelli A, Rothwell JC, Thompson PD, Hallett M (2001) Pathophysiology of bradykinesia in
Parkinson’s disease. Brain 124:2131–2146. doi:10.1093/brain/124.11.2131
Bloem BR, Hausdorff JM, Visser JE, Giladi N (2004) Falls and freezing of gait in Parkinson’s
disease: a review of two interconnected, episodic phenomena. Mov Disord 19:871–884.
doi:10.1002/mds.20115
Bloem BR, Marinus J, Almeida Q et al (2016) Measurement instruments to assess posture, gait, and
balance in Parkinson’s disease: critique and recommendations. Mov Disord. doi:10.1002/
mds.26572
Bonora G, Carpinella I, Cattaneo D et al (2015) A new instrumented method for the evaluation of
gait initiation and step climbing based on inertial sensors: a pilot application in Parkinson’s
disease. J Neuroeng Rehabil 12:45. doi:10.1186/s12984-015-0038-0
Cappa P, Patanè F, Rossi S et al (2008) Effect of changing visual condition and frequency of
horizontal oscillations on postural balance of standing healthy subjects. Gait Posture
28:615–626. doi:10.1016/j.gaitpost.2008.04.013
Catalá MM, Woitalla D, Arampatzis A (2016) Reactive but not predictive locomotor adaptability is
impaired in young Parkinson’s disease patients. Gait Posture 48:177–182. doi:10.1016/j.
Cole BT, Roy SH, Nawab SH (2011) Detecting freezing-of-gait during unscripted and
unconstrained activity. In: Proceedings of the annual international conference of the IEEE
Engineering in Medicine and Biology Society, EMBS, IEEE, 30 Aug - 03 Sep - Boston, MA,
USA, pp 5649–5652
Daliri MR (2012) Automatic diagnosis of neuro-degenerative diseases using gait dynamics. Meas J
Int Meas Confed 45:1729–1734. doi:10.1016/j.measurement.2012.04.013
Daliri MR (2013) Chi-square distance kernel of the gaits for the diagnosis of Parkinson’s disease.
Biomed Signal Process Control 8:66–70. doi:10.1016/j.bspc.2012.04.007
Daroff RB, Jankovic J, Mazziotta JC et al (2016) Bradley’s neurology in clinical practice, vols 1–2,
7th edn. Elsevier, London
de Lau LM, Breteler MM (2006) Epidemiology of Parkinson’s disease. Lancet Neurol 5:525–535.
doi:10.1016/S1474-4422(06)70471-9
Dehaene S (2003) The neural basis of the c law: a logarithmic mental number line. Trends Cogn Sci
7:145–147. doi:10.1016/S1364-6613(03)00055-X
Djuric-Jovicic MD, Jovicic NS, Radovanovic SM et al (2014) Automatic identification and
classification of freezing of gait episodes in Parkinson’s disease patients. IEEE Trans Neural
Syst Rehabil Eng 22:685–694. doi:10.1109/TNSRE.2013.2287241
Doherty KM, van de Warrenburg BP, Peralta MC et al (2011) Postural deformities in Parkinson’s
disease. Lancet Neurol 10:538–549
Doná F, Aquino CC, Gazzola JM et al (2015) Changes in postural control in patients with
Parkinson’s disease: a posturographic study. Physiotherapy. doi:10.1016/j.physio.2015.08.009
Ebersbach G, Moreau C, Gandor F et al (2013) Clinical syndromes: Parkinsonian gait. Mov Disord
28:1552–1559. doi:10.1002/mds.25675
Elble R, Deuschl G (2011) Milestones in tremor research. Mov Disord 26:1096–1105. doi:10.1002/
mds.23579
Endo T, Okuno R, Yokoe M et al (2009) A novel method for systematic analysis of rigidity in
Parkinson’s disease. Mov Disord 24:2218–2224. doi:10.1002/mds.22752
Enoka RM (2015) Neuromechanics of human movement. Human Kinetics, Champaign

Gao JB (2004) Analysis of amplitude and frequency variations of essential and Parkinsonian
tremors. Med Biol Eng Comput 42:345–349. doi:10.1007/BF02344710
Gelb DJ, Oliver E, Gilman S et al (1999) Diagnostic criteria for Parkinson disease. Arch Neurol
56:33. doi:10.1001/archneur.56.1.33
Geroin C, Smania N, Schena F et al (2015) Does the Pisa syndrome affect postural control, balance,
and gait in patients with Parkinson’s disease? An observational cross-sectional study. Parkin-
sonism Relat Disord 21:736–741. doi:10.1016/j.parkreldis.2015.04.020
Giladi N, Nieuwboer A (2008) Understanding and treating freezing of gait in parkinsonism,
proposed working definition, and setting the stage. Mov Disord 23:S423–S425. doi:10.1002/
mds.21927
Grabli D, Karachi C, Welter M et al (2012) Normal and pathological gait: what we learn from
Parkinson’s disease. J Neurol Neurosurg Psychiatry 83:979–985. doi:10.1136/jnnp-2012-
302263
Hoehn MM, Yahr MD (1967) Parkinsonism : onset, progression, and mortality. Neurology
17:427–442. doi:10.1212/WNL.17.5.427
Hubble RP, Naughton GA, Silburn PA, Cole MH (2015) Wearable sensor use for assessing standing
balance and walking stability in people with Parkinson’s disease: a systematic review. PLoS One
10:1–22. doi:10.1371/journal.pone.0123705
Hughes AJ, Daniel SE, Kilford L, Lees AJ (1992) Accuracy of clinical diagnosis of idiopathic
Parkinson’s disease: a clinico-pathological study of 100 cases. J Neurol Neurosurg Psychiatry
55:181–184. doi:10.1136/jnnp.55.3.181
Jacobs JV, Horak FB, Tran VK, Nutt JG (2006) Multiple balance tests improve the assessment of
postural stability in subjects with Parkinson’s disease. J Neurol Neurosurg Psychiatry
77:322–326. doi:10.1136/jnnp.2005.068742
Jian Y, Winter D, Ishac M, Gilchrist L (1993) Trajectory of the body COG and COP during
initiation and termination of gait. Gait Posture 1:9–22. doi:10.1016/0966-6362(93)90038-3
Kandori A, Yokoe M, Sakoda S et al (2004) Quantitative magnetic detection of finger movements in
patients with Parkinson’s disease. Neurosci Res 49:253–260. doi:10.1016/j.neures.2004.03.004
Kelly VE, Johnson CO, McGough EL et al (2015) Association of cognitive domains with postural
instability/gait disturbance in Parkinson’s disease. Parkinsonism Relat Disord 21:692–697.
doi:10.1016/j.parkreldis.2015.04.002
Kim J-W, Kwon Y, Yun J-S et al (2015) Regression models for the quantification of Parkinsonian
bradykinesia. Biomed Mater Eng 26(Suppl 1):S2249–S2258. doi:10.3233/BME-151531
Lee S-H, Lim JS (2012) Parkinson’s disease classification using gait characteristics and wavelet-
based feature extraction. Expert Syst Appl 39:7338–7344. doi:10.1016/j.eswa.2012.01.084
Levine D, Richards J, Whittle M, Whittle M (2012) Whittle’s gait analysis. Churchill Livingstone/
Elsevier, Edinburgh
Lin S-H, Chen S-W, Lo Y-C et al (2016) Quantitative measurement of Parkinsonian gait from
walking in monocular image sequences using a centroid tracking algorithm. Med Biol Eng
Comput 54:485–496. doi:10.1007/s11517-015-1335-2
Lord S, Rochester L, Baker K, Nieuwboer A (2008) Concurrent validity of accelerometry to
measure gait in Parkinsons disease. Gait Posture 27:357–359. doi:10.1016/j.
Mancini M, Horak FB (2010) The relevance of clinical balance assessment tools to differentiate
balance deficits. Eur J Phys Rehabil Med 46:239–248
Martinez-Martin P, Gil-Nagel A, Gracia LM et al (1994) Unified Parkinson’s disease rating scale
characteristics and structure. Mov Disord 9:76–83. doi:10.1002/mds.870090112
Massion J (1992) Movement, posture and equilibrium: interaction and coordination. Prog
Neurobiol 38:35–56. doi:10.1016/0301-0082(92)90034-C
McDonough AL, Batavia M, Chen FC et al (2001) The validity and reliability of the GAITRite
system’s measurements: a preliminary evaluation. Arch Phys Med Rehabil 82:419–425.
doi:10.1053/apmr.2001.19778
Meigal AY, Rissanen SM, Tarvainen MP et al (2013) Non-linear EMG parameters for differential
and early diagnostics of Parkinson’s disease. Front Neurol. doi:10.3389/fneur.2013.00135
Mirelman A, Gurevich T, Giladi N et al (2011) Gait alterations in healthy carriers of the LRRK2
G2019S mutation. Ann Neurol 69:193–197. doi:10.1002/ana.22165
Morris ME, McGinley J, Huxham F et al (1999) Constraints on the kinetic, kinematic and
spatiotemporal parameters of gait in Parkinson’s disease. Hum Mov Sci 18:461–483
Morris M, Iansek R, McGinley J et al (2005) Three-dimensional gait biomechanics in Parkinson’s
disease: evidence for a centrally mediated amplitude regulation disorder. Mov Disord 20:40–50.
doi:10.1002/mds.20278
Mortimer JA, Webster DD (1979) Evidence for a quantitative association between EMG stretch
responses and Parkinsonian rigidity. Brain Res 162:169–173. doi:10.1016/0006-8993(79)
90768-6
Nelson AJ, Zwick D, Brody S et al (2002) The validity of the GaitRite and the functional
ambulation performance scoring system in the analysis of Parkinson gait. NeuroRehabilitation
17:255–262
Nieder A, Miller EK (2003) Coding of cognitive magnitude: compressed scaling of numerical
information in the primate prefrontal cortex. Neuron 37:149–157. doi:10.1016/S0896-6273(02)
01144-3
Nieminen H, Takala EP (1996) Evidence of deterministic chaos in the myoelectric signal.
Electromyogr Clin Neurophysiol 36:49–58
Nutt JG, Bloem BR, Giladi N et al (2011) Freezing of gait: moving forward on a mysterious clinical
phenomenon. Lancet Neurol 10:734–744. doi:10.1016/S1474-4422(11)70143-0
Obeso JA, Rodriguez-Oroz MC, Goetz CG (2010) Missing pieces in the Parkinson’s disease puzzle.
Nat Med 16:653–661. doi:10.1038/nm.2165
Palmerini L, Mellone S, Avanzolini G et al (2013) Quantification of motor impairment in
Parkinson’s disease using an instrumented timed up and go test. IEEE Trans Neural Syst Rehabil
Eng 21:664–673. doi:10.1109/tnsre.2012.2236577
Park BK, Kwon Y, Kim JW et al (2011) Analysis of viscoelastic properties of wrist joint for
quantification of parkinsonian rigidity. IEEE Trans Neural Syst Rehabil Eng 19:167–176.
doi:10.1109/TNSRE.2010.2091149
Perry J, Burnfield JM, Cabico LM (2010) Gait analysis: normal and pathological function. Second
Edition. SLACK Incorporated. Thorofare, NJ, USA
Rocchi L, Palmerini L, Weiss A et al (2013) Balance testing with inertial sensors in patients with
Parkinson’s disease: assessment of motor subtypes. IEEE Trans Neural Syst Rehabil Eng
22:1064–1071. doi:10.1109/TNSRE.2013.2292496
Roiz RDM, Cacho EWA, Pazinatto MM et al (2010) Gait analysis comparing Parkinson’s disease
with healthy elderly subjects. Arq Neuropsiquiatr 68:81–86. doi:10.1590/S0004-
282X2010000100018
Salarian A, Russmann H, Wider C et al (2007) Quantification of tremor and bradykinesia in
Parkinson’s disease using a novel ambulatory monitoring system. IEEE Trans Biomed Eng
54:313–322
Sale P, De Pandis MF, Vimercati SL et al (2013) The relation between Parkinson’s disease and
ageing. Comparison of the gait patterns of young Parkinson’s disease subjects with healthy
elderly subjects. Eur J Phys Rehabil Med 49:161–167
Schrag A, Horsfall L, Walters K et al (2015) Prediagnostic presentations of Parkinson’s disease in
primary care: a case–control study. Lancet Neurol 14:57–64. doi:10.1016/S1474-4422(14)
70287-X
Scoppa F, Capra R, Gallamini M, Shiffer R (2013) Clinical stabilometry standardization. Basic
definitions – acquisition interval – sampling frequency. Gait Posture 37:290–292. doi:10.1016/j.
Shima K, Tsuji T, Kan E et al (2008) Measurement and evaluation of finger tapping movements
using magnetic sensors. Conf Proc IEEE Eng Med Biol Soc 1–8:5628–5631. doi:10.1109/
IEMBS.2008.4650490
Shulman LM, Armstrong M, Ellis T et al (2016) Disability rating scales in Parkinson’s disease:
critique and recommendations. Mov Disord. doi:10.1002/mds.26649
Sofuwa O, Nieuwboer A, Desloovere K et al (2005) Quantitative gait analysis in Parkinson’s
disease: comparison with a healthy control group. Arch Phys Med Rehabil 86:1007–1013.
doi:10.1016/j.apmr.2004.08.012
Sorrentino P, Barbato A, Del Gaudio L et al (2016) Impaired gait kinematics in type 1 Gaucher’s
disease. J Parkinsons Dis 6:191–195. doi:10.3233/JPD-150660
Stamatakis J, Ambroise J, Crémers J et al (2013) Finger tapping clinimetric score prediction in
Parkinson’ s disease using low-cost accelerometers. Comput Intell Neurosci. doi:10.1155/2013/
717853
Švehlík M, Zwick EB, Steinwender G et al (2009) Gait analysis in patients with Parkinson’s disease
off dopaminergic therapy. Arch Phys Med Rehabil 90:1880–1886. doi:10.1016/j.
apmr.2009.06.017
Swie YW, Sakamoto K, Shimizu Y (2005) Chaotic analysis of electromyography signal at low back
and lower limb muscles during forward bending posture. Electromyogr Clin Neurophysiol
45:329–342
Tahir NM, Manap HH (2012) Parkinson disease gait classification based on machine learning
approach. J Appl Sci 12:180–185. doi:10.3923/jas.2012.180.185
Thanawattano C, Pongthornseri R, Anan C et al (2015) Temporal fluctuations of tremor signals
from inertial sensor: a preliminary study in differentiating Parkinson’s disease from essential
tremor. Biomed Eng Online 14:101. doi:10.1186/s12938-015-0098-1
Timmer J, Lauk M, Deuschl G (1996) Quantitative analysis of tremor time series. Electroen-
cephalogr Clin Neurophysiol 101:461–468. doi:10.1016/0924-980X(96)94658-5
Tripoliti EE, Tzallas AT, Tsipouras MG et al (2013) Automatic detection of freezing of gait events in
patients with Parkinson’s disease. Comput Methods Programs Biomed 110:12–26. doi:10.1016/
j.cmpb.2012.10.016
Tzallas AT, Tsipouras MG, Rigas G et al (2014) PERFORM: a system for monitoring, assessment
and management of patients with Parkinson’s disease. Sensors (Basel) 14:21329–21357.
doi:10.3390/s141121329
Visser JE, Oude Nijhuis LB, Janssen L et al (2010) Dynamic posturography in Parkinson’s disease:
diagnostic utility of the “first trial effect”. Neuroscience 168:387–394. doi:10.1016/j.
Vitale C, Agosti V, Avella D et al (2012) Effect of global postural rehabilitation program on
spatiotemporal gait parameters of parkinsonian patients: a three-dimensional motion analysis
study. Neurol Sci 33:1337–1343. doi:10.1007/s10072-012-1202-y
Wahid F, Begg RK, Hass CJ et al (2015) Classification of Parkinson’s disease gait using spatial-
temporal gait features. IEEE J Biomed Health Inform 19:1794–1802. doi:10.1109/
JBHI.2015.2450232
Walshe FMR, Dejerine J, Ferrier D et al (1961) Contributions of John Hughlings Jackson to
neurology. Arch Neurol 5:119–131. doi:10.1001/archneur.1961.00450140001001
Xia R, Powell D, Zev Rymer W et al (2011) Differentiation between the contributions of shortening
reaction and stretch-induced inhibition to rigidity in Parkinson’s disease. Exp Brain Res
209:609–618. doi:10.1007/s00221-011-2594-2
Yokoe M, Okuno R, Hamasaki T et al (2009) Opening velocity, a novel parameter, for finger tapping
test in patients with Parkinson’s disease. Parkinsonism Relat Disord 15:440–444. doi:10.1016/j.
parkreldis.2008.11.003
Yunfeng W, Krishnan S (2010) Statistical analysis of gait rhythm in patients with Parkinson’s
disease. IEEE Trans Neural Syst Rehabil Eng 18:150–158. doi:10.1109/tnsre.2009.2033062
Gait and Multiple Sclerosis
James McLoughlin
Abstract
Gait analysis technology has led to an increased understanding of the changes in
mobility that contribute to increasing disability in people with multiple sclerosis
(MS). At all levels of the disease spectrum, changes in joint angles, moments, and
muscle activation have been identified, which has lead to the investigation of key
impairments that limit functional mobility in people with MS. Comprehensive
three-dimensional gait analysis has characterized gait changes that worsen with
increasing disability and deteriorate with walking-induced fatigue. This important
ability to measure movement performance is resulting in a greater appreciation of
the link between movement impairment during gait and its impact on daily
functions. This knowledge will hopefully increase our ability to target specific
aspects of gait with individualized physical therapy strategies.
Keywords
Multiple sclerosis • Gait • Mobility limitation • Disability evaluation •
Rehabilitation
Introduction
Multiple sclerosis (MS) is a chronic degenerative condition that affects two million
people worldwide (Trisolini et al. 2010) and is characterized by inflammation,
demyelination, and axonal degeneration in the central nervous system. Symptoms
of MS are heterogeneous and may include a number of problems associated with
central nervous system dysfunction, such as optic neuritis, fatigue, weakness, spas-
ticity, ataxia, tremor, numbness, paresthesia, bladder and bowel dysfunction, sexual
J. McLoughlin (*)
Flinders University, Adelaide, Australia
e-mail: james.mcloughlin@flinders.edu.au

DOI 10.1007/978-3-319-30808-1_65-1
2 J. McLoughlin
dysfunction, speech difficulties, swallowing problems, and cognitive and emotional

disturbances (Kurtzke 1983). People with MS often report difficulty with mobility as
one of the most disabling aspects of living with the condition (Heesen et al. 2008;
LaRocca 2011) with a significant impact on quality of life (Hemmett et al. 2004).
Mobility-related problems in MS contribute to a considerable socioeconomic burden
on society in terms of employment and healthcare utilization (Pike et al. 2012).
Almost all people with MS will have some difficulty in walking due to a combina-
tion of symptoms, including poor balance (Sosnoff et al. 2011b), weakness (Yahia
et al. 2011), spasticity (Sosnoff et al. 2011a; Sosnoff et al. 2010), and sensory loss
(Cameron et al. 2008; Cattaneo and Jonsdottir 2009).
State of the Art
Movement analysis technology has allowed both researchers and clinicians to further
characterize some of the gait impairments that occur in people with MS. This
information is increasing our understanding of disease impact and progression
while enhancing our ability to target and measure therapeutic interventions aimed
at improving safe mobility for people with MS. A number of studies have charac-
terized the spatiotemporal differences in gait in people with MS compared to healthy
controls. These studies have typically shown reductions in speed, cadence, and
double limb support time in people with MS (Givon et al. 2009; Pilutti et al. 2012;
Remelius et al. 2012; Sosnoff et al. 2012; Thoumie et al. 2005). Even in those people
with MS who are fully ambulant with very minimal clinical disability, spatiotempo-
ral gait changes are still present, demonstrating reductions in speed, cadence, step
length, and increased step width and double limb support time (Sosnoff et al. 2012).
These parameters are known to deteriorate further with higher disability levels as
measured by the Expanded Disability Status Scale (EDSS) (Lizrova Preiningerova
et al. 2015).
Comprehensive Gait Analysis in People with Multiple Sclerosis
Two- and three-dimensional motion capture enables a more in-depth analysis of gait
impairments and compensations that occur in people with MS and provides insight
into disease progression and physical disability. Kinematic (movement) and kinetic
(forces) gait parameters in people with MS typically show reductions in joint range
of motion and moments (Kelleher et al. 2010a; Huisinga et al. 2012; Lizama et al.
2016). Reductions in power at the hip, knee, and ankle show a relationship with
reduced levels of clinical disability as measured by the EDSS (Huisinga et al. 2012).
Similarly, Kelleher et al. (2010a) also demonstrated that hip, knee, and ankle angles
and propulsive force are reduced further in those people with lower ambulation
levels (Kelleher et al. 2010a). Hip flexor moment is reduced as due to lower extensor
and flexor powers at the end of stance and early swing phase (Huisinga et al. 2012).
In early stance people with MS demonstrate a reduced knee extensor moment
Gait and Multiple Sclerosis 3
(Huisinga et al. 2012). This reduction in knee control has been shown to worsen with
reduced mobility levels and may well be due to deterioration in a number of
mechanisms, including knee extensor strength, coordination (Güner et al. 2015),
and possibly hamstring spasticity limiting knee extension range in late swing phase
(Kelleher et al. 2010a). Push-off power at the ankle, which provides most of the
propulsive energy for walking (McGibbon 2003), has been shown to be reduced in
people with MS (Huisinga et al. 2012). Reduced plantar flexion strength may be the
major culprit, as reduced plantar flexion strength has been shown to be a major
contributor to reduced walking capacity in people with MS (Wagner et al. 2014).
Kinematic analysis of people with MS with very mild physical disability has found
early reductions in ankle range of motion (Benedetti et al. 1999; Martin et al. 2006).
Electromyographic (EMG) recordings also can provide useful information about
individual muscle activity and timing. EMG recordings from gastrocnemius and
tibialis anterior muscles during gait indicate muscle cocontraction in these muscles,
leading to reduced range of motion and increased stiffness at the ankle. This is
possibly a stabilization ankle strategy (Benedetti et al. 1999; Kelleher et al. 2010a).
This pattern has also been detected in those people with MS without any clinical
signs of pyramidal dysfunction (Martin et al. 2006).
Gait and MS Impairments
Sensory tracts are often affected by demyelinating lesions in MS (Zackowski et al.

2009) resulting in reductions in important proprioceptive sensory information which
is known to affect standing balance in people with MS (Cattaneo and Jonsdottir
2009; Cameron et al. 2008; Citaker et al. 2011); however, less is known on its effect
on gait. Proprioceptive loss may well increase both flexor and extensor muscular
compensations in order to maintain gait speed (Thoumie and Mevellec 2002;
Rougier et al. 2007).
Cerebellar lesions resulting in a deterioration in coordination are also common in
people with MS (Kutzelnigg et al. 2007) and appear to be a major contributor to poor
balance and increasing disability (McLoughlin et al. 2015). Not surprisingly cere-
bellar pathology has been shown to relate to reduced balance (Prosperini et al. 2011;
2013a; b) and falls (Sosnoff et al. 2011). Cerebellar ataxia can contribute to some
gait deficits such as increased step width (Givon et al. 2009). In people with MS, gait
asymmetry and increased step width have been associated with increased risk of falls
(Kasser et al. 2011).
The overall influence of cerebellar ataxia on spatiotemporal gait deficits may be
less when compared with other upper motor neuron (UMN) or pyramidal weakness
(Thoumie et al. 2005; Kalron and Givon 2016). The UMN syndrome is characterized
by varying degrees of velocity-dependent increases in muscle tone and weakness,
both of which can alter the gait pattern observed in people with MS. Oversimplistic
clinical spasticity scales used to measure increases in passive muscle tone are known
to have major limitations and only show a relationship with changes in posture and
balance rather than gait (Sosnoff et al. 2010, 2011). However, increases in muscle
4 J. McLoughlin
tone can restrict active movement and may well contribute to reductions in hip and
knee range of motion in the gait cycle (Pau et al. 2015). Reduced strength is an
important part of the UMN syndrome with weakness of the knee extensors and ankle
plantar flexor linked to reduced mobility levels (Güner et al. 2015; Wagner et al.
2014). Despite a large heterogeneity and enormous mixed overlap of sensory,
cerebellar, and pyramidal signs often present in people with MS, weakness stands
out as a key impairment that contributes to many of the spatiotemporal, kinetic, and
kinematic gait changes (Kalron and Givon 2016; Thoumie et al. 2005; Thoumie and
Mevellec 2002). Overall, poor function at the ankle appears to be a major contributor
to slow walking speeds in people with MS (Lizama et al. 2016) and may therefore
present as a key therapeutic target for orthotic and physical therapies.
Gait and Fatigue
The 1998 Multiple Sclerosis of Clinical Practice Guidelines define fatigue in MS as a

“subjective lack of physical and/or mental energy that is perceived by the individual
or caregiver to interfere with usual and desired activities” (for Clinical Practice
Guidelines and Others 1998). People with MS often report fatigue as one of their
most common and disabling symptoms (Bakshi 2003). Fatigue is known to impact
heavily on employment levels (Hadjimichael et al. 2008; Smith and Arnett 2005)
and is linked to reduced quality of life (Amato et al. 2001) in people with
MS. Fatigue can occur in all MS subtypes, although is more pronounced in those
people with a less stable natural history, accumulating impairment and reduced
mobility levels (Hadjimichael et al. 2008). Some studies provide evidence for the
link between motor fatigue and gait. Rate of foot taps and maximal force generation
has been shown to be associated with walking speed and cadence in people with MS
(Ng et al. 2004). Changes in corticospinal activity following walking-induced
fatigue have also been demonstrated in a small sample of people with MS; however,
walking distance and fatigue levels were not recorded (Schubert et al. 1998).
Mobility in MS can be objectively assessed using a number of measures including
timed or distance measures of walking performance and is currently the gold
standard in assessing accumulating disability levels in those people with MS with
moderate to severe disease (Kurtzke 1983). Common clinical measures of mobility
in people with MS include timed walks over 10 meters or 25 feet (Kieseier and
Pozzilli 2012); however, more demanding tests of walking over 2 min (Gijbels et al.
2011) or 6 min (Gijbels et al. 2011; Motl et al. 2012b) may be more representative of
functional mobility (Savci et al. 2005).
Distance over 6 min of walking has been shown to correlate with fatigue on the
physical subscale of the Modified Fatigue Impact Scale (Fisk et al. 1994). Together,
mobility and fatigue have been recognized as important measures of functional
capacity (Hutchinson et al. 2009) and major contributors to the physical impact of
MS (Kehoe et al. 2014). In addition, there is potential in utilizing measures of
mobility to also objectively measure motor fatigue (Burschka et al. 2012; Dobkin
2008; Phan-Ba et al. 2012). Walking measures that increase the energy expenditure
are important when considering that energy cost with walking correlates with
walking disability (Motl et al. 2010) and is associated with gait impairments, daily
activity, and fatigue (Motl et al. 2012a).Preliminary evidence suggests that fatigue is
associated with kinetic gait changes in people with MS, with reduced ankle power at
push-off, reduced knee power absorption, and reduced peak knee extensor moments
shown to be associated with increased levels of self-reported fatigue (Huisinga et al.
2011). One study of fourteen people with MS compared walking in the morning
versus the afternoon when fatigue levels were significantly higher, yet found no
change in spatiotemporal gait parameters (Morris et al. 2002).
The six-minute walk test (6MWT) is a common mobility-related assessment that
has been used as a measure of functional ambulation capacity (Savci et al. 2005).
Recent research indicates that while the 6MWT can measure both walking and
fitness in people with MS, it relates more strongly to walking performance rather
than aerobic or muscular fitness (Sandroff et al. 2015). Performance of the 6MWT
has been shown to produce marked increased levels in perceived fatigue (Barr et al.
2014). People with MS demonstrate reduced distance with the 6MWT, compared to
healthy controls, which correlates with total and physical aspects of fatigue as
measured by the Modified Fatigue Impact Scale (MFIS) (Goldman et al. 2008).
6MWT distance has also been associated with limitations in activities of daily living
in people with MS (Karpatkin and Rzetelny 2012), which suggests that 6MWT
performance may be an accurate measure of the capacity to undertake “real-world”
daily activities. While there is no standardized method of measuring the 6MWT in
people with MS, 6MWT protocols involving more frequent 180-degree turns have
been shown to be more physically demanding when compared to square walkway
protocols (Sandroff et al. 2014). It has become clear that both mobility and fatigue in
MS are inextricably linked and should therefore be considered together. In 2007, a
multidisciplinary consensus conference with the Consortium of MS Centres in the
USA recommended that rehabilitation professionals need to improve their under-
standing of gait and fatigue outcome measures used in MS rehabilitation (Hutchin-
son et al. 2009). Therefore, the investigation of activity-dependent fatigability in
lower limb muscles following the 6MWT in people with MS remains an important
research and clinical question.
Further evidence suggests that following a period of walking, detrimental effects
on spatiotemporal gait parameters are evident in people with MS with slow walking
speed (below 0.82 m/s) (Feys et al. 2013). Only a few pilot studies have investigated
the direct effect of walking-induced fatigue on kinematic and kinetic parameters of
gait. A pilot study of walking at preferred walking speed to the point of exhaustion
on a treadmill reported some increases in kinematic variability in selected patients
(Sehle et al. 2014), whereas another study measuring gait variability found no
kinematic or kinetic variability following a fatiguing walk in study of 20 people
with MS (Crenshaw et al. 2006). For both of these studies, walking distance was not
reported. McLoughlin et al. was the first controlled study to identify a number of
fatigue-induced changes to kinematics and kinetic gait mechanics in a group of
people with MS with mild to moderate disability (McLoughlin et al. 2016). Com-
pared with healthy controls, the MS group had significantly reduced total 6MWT
6 J. McLoughlin
distance with a trend toward slowing down over the last 3 min. This reduced speed in
the second half of the test supports the results of other studies that have found
reduced speed in the last 1–5 min of the 6MWT (Dalgas et al. 2014; Gijbels et al.
2011). This gait slowing appears to coincide with steady-state aerobic metabolism,
as Motl et al. (2012b) demonstrated in physiological measurements during the final
3 min of the 6MWT (Motl et al. 2012b). These findings suggest that a “deceleration
index” may be a useful measure of walking capacity in people with MS (Phan-Ba
et al. 2012). Another study showed that “decline” in walking speed, rather than
“mean” walking speed over the 6MWT, had a stronger association with subjective
fatigue (Burschka et al. 2012). While there has been some support for a shortened
two-minute walk test to asses walking ability in MS (Gijbels et al. 2011), gait studies
suggest a shortening of the 6MWT may result in the loss of important clinical
information when assessing functional mobility and the relevant changes caused
by walking-induced fatigue (McLoughlin et al. 2016). The finding of no change in
spatiotemporal parameters following the 6MWT is consistent with a previous study,
which only found reduced cadence following a 6MWT in a less able group of people
with MS (Feys et al. 2013). It appears that moderately disabled people with MS are
able to maintain their pre-walk spatiotemporal characteristics following the 6MWT,
but kinematic and kinetic gait analysis reveals changes that are likely to be move-
ment compensations in response to the walking-induced fatigue (McLoughlin et al.
2016). In particular, joint kinetic analyses show increases in moments and power
absorption at the hip, knee, and ankle during stance phase – signs of fatigue-induced
weakness (McLoughlin et al. 2014b) which are likely to further increase the energy
cost of walking (Kuo et al. 2005). Increases in hip extensor moment in the less
affected side suggests some capacity to increase the positive compensatory output to
try and maintain speed, while an increase in hip flexor moment in the more affected
side may be a compensation to assist toe clearance with increased fatigue
(McLoughlin et al. 2016). People with MS are likely to increase the underlying
central corticomotor activity needed to produce these positive changes in joint
kinetics, which could contribute further to perceived fatigue levels (Morgante et al.
2011; Thickbroom et al. 2006, 2008; ). A deeper understanding of the relationship
between fatigue and mobility may provide opportunities to develop treatment-
specific strategies for both fatigue and mobility in people with MS.
Some Targets for Therapeutic Intervention
One potential target for therapeutic intervention to improve walking capacity in

people with MS is assisting ankle dorsiflexion during gait. Neurophysiological
studies have shown that ankle dorsiflexor muscles are particularly susceptible to
motor fatigue (Ng et al. 2004; Thickbroom et al. 2008) and may alter gait in people
with MS (Martin et al. 2006; Benedetti et al. 1999). Kinematic analysis has demon-
strated reduced dorsiflexion in the weaker leg at initial contact phase of gait
following a fatiguing six-minute walk (McLoughlin et al. 2016). The consequences
of reduced dorsiflexion can be serious as difficulties in toe clearance in the swing
phase of gait can lead to trips, which are reported as the most common reasons for
falling by people with MS (Matsuda et al. 2011).
Ankle dorsiflexion can be assisted with commonly prescribed ankle-foot orthosis
(AFOs) that comprise rigid or nonrigid articulated braces that support the foot from
underneath to prevent foot drop. To date there is no evidence that AFOs can improve
mobility in people with MS (Sheffler et al. 2008). A review of the effect of these
devices found that rigid AFOs can compromise balance, while more flexible,
nonrigid AFOs can facilitate dynamic balance (Ramstrand and Ramstrand 2010).
In a crossover study involving people with MS, rigid and nonrigid AFOs improved
static balance, but rigid AFOs impaired dynamic balance when walking (Cattaneo
et al. 2002).One of the limitations of AFOs is that they limit not only the ankle and
knee range of motion, but reduce the proprioceptive input via the sole of the foot,
which may further exacerbate preexisting sensory deficits that limit balance ability
(Cameron et al. 2008; Citaker et al. 2011).
Another option to assist ankle dorsiflexion is the use of functional electrical
stimulation (FES) to the pretibial muscles, which has been shown to assist ankle
dorsiflexion during gait in people with MS (Scott et al. 2013). This form of FES has
been shown to have an acute effect of speed and reduced physiological cost of
walking during a five-minute walk around a 10-meter elliptical course (Paul et al.
2008). A small study, including only two people with MS with unilateral foot drop
who used FES for 10 weeks, found improvements in walking speed and reduced
physiological cost of walking over a 10-meter course (Taylor et al. 1999). Scheffler
et al. found some improvement with stair ascent and descent in people with MS, with
9 out of the 11 participants preferring FES over an AFO and all preferring FES over
no device (Sheffler et al. 2009). A more recent study examined the training effect of
12 weeks of FES use in nine people with MS and found improvement in ankle
dorsiflexion in gait, increased speed over 10-meter walk test, and increased speed
and reduced perceived exertion over the two-minute walk test (van der Linden et al.
2014). This suggests that assisting ankle dorsiflexion may have benefits in terms of
walking speed and reducing physiological cost of walking in people with
MS. Another benefit of FES to the ankle dorsiflexors may be the potential to
“retrain” the gait pattern via cortical neuroplasticity motor learning (Everaert et al.
2010). Prolonged use of FES for 8 weeks has shown some training effect to ankle
and knee kinetics in stance phase in people with MS (Barr et al. 2016), which
suggests that FES has potential as an adjunct to gait retraining therapy. From a
practical viewpoint, however, FES devices are expensive and not easily accessible at
present, which remains a significant barrier to their use for people with MS
worldwide.
An alternative (and perhaps more cost-effective) method of addressing the issues
of restricted dynamic balance and physiological cost of walking may involve
assisting ankle dorsiflexion with a dorsiflexion assist orthosis (DAO) (McLoughlin
et al. 2014). The DAO dynamically assists ankle dorsiflexion with an elastic strap
attached to the shoe that can assist pulling the foot toward dorsiflexion. This is a very
simple and inexpensive device compared with custom-made AFOs and FES devices,
and unlike AFOs the DAO does not interfere with the sensory interaction of the foot
8 J. McLoughlin
with the sole of the shoe. Maintaining and even enhancing the sensory feedback
within the shoe remains an attractive target in people with MS, with textured (Dixon
et al. 2014; Kelleher et al. 2010b) or contoured insoles (Ramdharry et al. 2006)
showing promise for improving mobility and balance. The DAO is therefore an
attractive option to assist dorsiflexion without interfering with sensory feedback via
the sole of the foot. To date, this type of dorsiflexion assist orthosis (DAO) has not
been evaluated with prolonged use in people with MS. When considering the issue
of motor fatigue with MS, the known detrimental effects of AFOs on standing
balance and the improved physiological cost of walking with the use of FES devices,
both the DAO and FES, should be investigated in terms of reducing the physiolog-
ical cost of walking and mitigating any effects of walking-induced fatigue on balance
and lower limb muscle strength with prolonged walking. These devices may com-
pliment individualized neurological physiotherapy focusing on ankle control and
mobility, which has shown promise in improving gait and balance in MS (Davies
et al. 2015).
Conclusion
Gait analysis has provided useful information about the many sensorimotor impair-
ments that contribute to declining mobility in people with multiple sclerosis. Spa-
tiotemporal, kinematic, and kinetic measures have been able to capture the changes
associated with accumulating disability over time. This type of research has identi-
fied useful therapeutic targets that have the potential to improve walking pattern and
efficiency. It is hoped that this will lead to more meaningful improvements in
functional mobility and associated fatigue-induced impairments. Future gait analysis
will hopefully allow for more relevant subgrouping of gait presentations in MS and
lead to specific interventions that are tailor-made for each individual.
Cross-References

▶ Activity Monitoring in Orthopaedic Patients
▶ Brain Computer Interface Assisted Gait
▶ Concussion Assessment During Gait
▶ Conventional Gait Model – Success and Limitations
▶ Effects of Knee Replacement on Gait
▶ Effects of Total Hip Replacement on Gait
▶ EMG Activity in Gait -the Influence of Motor Disorders

▶ Functional Capacity Evaluation in Rehabilitation
Applications
▶ Gait and Multiple Sclerosis
▶ Gait Changes in Skeletal Dysplasia
▶ Gait Rehabilitation with Exoskeletons
▶ Gait with Incomplete Spinal Cord Injury
▶ Idiopathic Toe Walking
▶ Impact of Scoliosis on Gait
▶ Implications of Incline Walking in Prosthetics
Analysis
▶ Kinematic Foot Models for Instrumented Gait Analyis
▶ Muscle Structure in Cerebral Palsy
▶ Muscular Dystrophy and Duchenne Gait
▶ Natural History of Cerebral Palsy and Outcome Assessment
▶ Oxygen Consumption in Adult Central Nervous System Disability
▶ Oxygen Consumption in Cerebral Palsy
▶ Pattern Recognition in Parkinson
▶ Slips and Falls
▶ Surface EMG
▶ Talipes Equinocavus and Hereditary Motor Sensory Neuropathy
▶ The Importance of Foot Pressure in Diabetes
10 J. McLoughlin

▶ Torsional Deformities on Cerebral Palsy
▶ Trunk and Spine Models for Instrumented Gait Analyis
References
Amato MP et al (2001) Quality of life in multiple sclerosis: the impact of depression, fatigue and
disability. Mult Scler 7:340–344
Bakshi R (2003) Fatigue associated with multiple sclerosis: diagnosis, impact and management.
Mult Scler 9:219–227
Barr C et al (2014) Walking for six minutes increases both simple reaction time and stepping
reaction time in moderately disabled people with Multiple Sclerosis. Mult Scler Relat Disord
3(4):457–462
Barr CJ et al (2016) Orthotic and therapeutic effect of functional electrical stimulation on fatigue
induced gait patterns in people with multiple sclerosis. Disabil Rehabil 0(0):1–13
Benedetti MG et al (1999) Gait abnormalities in minimally impaired multiple sclerosis patients.
Burschka JM et al (2012) An exploration of impaired walking dynamics and fatigue in multiple
sclerosis. BMC Neurol 12:161
Cameron MH et al (2008) Imbalance in multiple sclerosis: a result of slowed spinal somatosensory
conduction. Somatosens Mot Res 25:113–122
Cattaneo D, Jonsdottir J (2009) Sensory impairments in quiet standing in subjects with multiple
sclerosis. Mult Scler 15:59–67
Cattaneo D et al (2002) Do static or dynamic AFOs improve balance? Clin Rehabil 16:894–899
Citaker S et al (2011) Relationship between foot sensation and standing balance in patients with
multiple sclerosis. Gait Posture 34:275–278
Crenshaw SJ et al (2006) Gait variability in people with multiple sclerosis. Mult Scler 12:613–619
Dalgas U et al (2014) Aerobic intensity and pacing pattern during the six-minute walk test in
patients with multiple sclerosis. J Rehabil Med 46:59–66
Davies BL et al (2015) Neurorehabilitation strategies focusing on ankle control improve mobility
and posture in persons with multiple sclerosis. J Neurol Phys Ther 39(4):225–232
Dixon J et al (2014) Effect of textured insoles on balance and gait in people with multiple sclerosis:
an exploratory trial. Physiotherapy 100(2):142–149
Dobkin BH (2008) Fatigue versus activity-dependent fatigability in patients with central or periph-
eral motor impairments. Neurorehabil Neural Repair 22:105–110
Everaert DG et al (2010) Does functional electrical stimulation for foot drop strengthen
corticospinal connections? Neurorehabil Neural Repair 24:168–177
Feys P et al (2013) Spatio-temporal gait parameters change differently according to speed instruc-
tions and walking history in MS patients with different ambulatory dysfunction. Mult Scler
Relat Disord 2(3):238–246. doi: 10.1016/j.msard.2013.01.004
Fisk JD et al (1994) Measuring the functional impact of fatigue: initial validation of the fatigue
impact scale. Clinical Infect Dis 18(Suppl 1):S79–S83
For Clinical Practice Guidelines, M.S.C. & Others (1998) Fatigue and multiple sclerosis: evidence-
based management strategies for fatigue in multiple sclerosis: clinical practice guidelines. The
Council. The Consortium of Multiple Sclerosis Centers, in conjunction with the Paralyzed
Veterans of America (PVA). http://www.mscare.org/?page=practice_guidelines
Gijbels D, Eijnde B, Feys P (2011) Comparison of the 2- and 6-minute walk test in multiple
Givon U, Zeilig G, Achiron A (2009) Gait analysis in multiple sclerosis: characterization of

temporal-spatial parameters using GAITRite functional ambulation system. Gait Posture
29:138–142
Goldman MD, Marrie RA, Cohen JA (2008) Evaluation of the six-minute walk in multiple sclerosis
subjects and healthy controls. Mult Scler 14:383–390
Güner S et al (2015) Knee muscle strength in multiple sclerosis: relationship with gait character-
istics. J Phys Ther Sci 27(3):809–813
Hadjimichael O, Vollmer T, Oleen-Burkey M (2008) Fatigue characteristics in multiple sclerosis:
the North American Research Committee on Multiple Sclerosis (NARCOMS) survey. Health
Qual Life Outcomes 6:100
Heesen C et al (2008) Patient perception of bodily functions in multiple sclerosis: gait and visual
function are the most valuable. Mult Scler 14:988–991
Hemmett L et al (2004) What drives quality of life in multiple sclerosis? QJM 97:671–676
Huisinga JM et al (2011) Is there a relationship between fatigue questionnaires and gait mechanics
in persons with multiple sclerosis? Arch Phys Med Rehabil 92:1594–1601
Huisinga JM et al (2012) Gait mechanics are different between healthy controls and patients with
multiple. J Appl Biomech 2012 Human Kinetics, p 2
Hutchinson B et al (2009) Toward a consensus on rehabilitation outcomes in MS: gait and fatigue.
Int J MS Care 11:67–78
Kalron A, Givon U (2016) Gait characteristics according to pyramidal, sensory and cerebellar
EDSS subcategories in people with multiple sclerosis. J Neurol. Available at: http://dx.doi.org/
10.1007/s00415-016-8200-6
Karpatkin H, Rzetelny A (2012) Effect of a single bout of intermittent versus continuous walking on
perceptions of fatigue in people with multiple sclerosis. Int J Care 14:124–131
Kasser SL et al (2011) A prospective evaluation of balance, gait, and strength to predict falling in
women with multiple sclerosis. Arch Phys Med Rehabil. Available at: http://www.ncbi.nlm.nih.
gov/pubmed/21840497
Kehoe M et al (2014) Predictors of the physical impact of multiple sclerosis following community-
based, exercise trial. Multiple Scler 10:1352458514549395
Kelleher KJ, Spence W et al (2010a) The characterisation of gait patterns of people with multiple
sclerosis. Disabil Rehabil 32:1242–1250
Kelleher KJ, Spence WD et al (2010b) The effect of textured insoles on gait patterns of people with
multiple sclerosis. Gait Posture 32:67–71
Kieseier BC, Pozzilli C (2012) Assessing walking disability in multiple sclerosis. Mult Scler J
18:914–924
Kuo AD, Donelan JM, Ruina A (2005) Energetic consequences of walking like an inverted
pendulum: step-to-step transitions. Exerc Sport Sci Rev 33:88–97
Kurtzke JF (1983) Rating neurologic impairment in multiple sclerosis: an expanded disability status
scale (EDSS). Neurology 33:1444–1452
Kutzelnigg A et al (2007) Widespread demyelination in the cerebellar cortex in multiple sclerosis.
Brain Pathol 17:38–44
LaRocca NG (2011) Impact of walking impairment in multiple sclerosis. Patient 4:189–201
Lizama LEC et al (2016) The use of laboratory gait analysis for understanding gait deterioration in
people with multiple sclerosis. Mult Scler J 22(14):1768–1776
Lizrova Preiningerova J et al (2015) Spatial and temporal characteristics of gait as outcome
measures in multiple sclerosis (EDSS 0 to 6.5). J Neuroeng Rehabil 12(1):1–7
Martin CL et al (2006) Gait and balance impairment in early multiple sclerosis in the absence of
clinical disability. Mult Scler 12:620–628
Matsuda PN et al (2011) Falls in multiple sclerosis. PM & R J Inj funct Rehabil 3:624–632, quiz 632
McGibbon CA (2003) Toward a better understanding of gait changes with age and disablement:
neuromuscular adaptation. Exerc Sport Sci Rev 31(2):102–108
McLoughlin JV et al (2014) Dorsiflexion assist orthosis reduces the physiological cost and mitigates
deterioration in strength and balance associated with walking in people with multiple sclerosis.
12 J. McLoughlin
Archives of physical medicine and rehabilitation. Available at: http://www.sciencedirect.com/

science/article/pii/S0003999314010703
McLoughlin JV et al (2014b) Six minutes of walking leads to reduced lower limb strength and
increased postural sway in people with multiple sclerosis. NeuroRehabilitation 35(3):503–508
McLoughlin J et al (2015) Association of postural sway with disability status and cerebellar
dysfunction in people with multiple sclerosis: a preliminary study. Int J MS Care 17(3):146–151
McLoughlin JV et al (2016) Fatigue induced changes to kinematic and kinetic gait parameters
following six minutes of walking in people with multiple sclerosis. Disabil Rehabil 38
(6):535–543
Morgante F et al (2011) Is central fatigue in multiple sclerosis a disorder of movement preparation?
J Neurol 258:263–272
Morris ME et al (2002) Changes in gait and fatigue from morning to afternoon in people with
multiple sclerosis. J Neurol Neurosurg Psychiatry 72:361–365
Motl RW et al (2010) Multiple sclerosis walking Scale-12 and oxygen cost of walking. Gait Posture
31:506–510
Motl RW, Sandroff BM et al (2012a) Energy cost of walking and its association with gait
parameters, daily activity, and fatigue in persons with mild multiple sclerosis. Neurorehabil
Neural Repair 26:1015–1021
Motl RW, Suh Y et al (2012b) Evidence for the different physiological significance of the 6-and
2-minute walk tests in multiple sclerosis. BMC Neurol 12:6
Ng AV et al (2004) Functional relationships of central and peripheral muscle alterations in multiple
sclerosis. Muscle Nerve 29:843–852
Pau M et al (2015) Effect of spasticity on kinematics of gait and muscular activation in people with
multiple sclerosis. J Neurol Sci 358(1–2):339–344
Paul L et al (2008) The effect of functional electrical stimulation on the physiological cost of gait in
people with multiple sclerosis. Mult Scler 14:954–961
Phan-Ba R et al (2012) Motor fatigue measurement by distance-induced slow down of walking
speed in multiple sclerosis. PloS One 7:e34744
Pike J et al (2012) Social and economic burden of walking and mobility problems in multiple
sclerosis. BMC Neurol 12:94
Pilutti LA et al (2012) Further validation of multiple sclerosis walking scale-12 scores based on
spatiotemporal gait parameters. Arch Phys Med Rehabil
Prosperini L et al (2011) The relationship between infratentorial lesions, balance deficit and
accidental falls in multiple sclerosis. J Neurol Sci 304:55–60
Prosperini L, Petsas N et al (2013a) Balance deficit with opened or closed eyes reveals involvement
of different structures of the central nervous system in multiple sclerosis. Mult Scler J
20(1):81–90. doi: 10.1177/1352458513490546
Prosperini L, Sbardella E et al (2013a) Multiple sclerosis: white and gray matter damage associated
with balance deficit detected at static posturography. Radiology 268:181–189
Ramdharry GM et al (2006) De-stabilizing and training effects of foot orthoses in multiple sclerosis.
Ramstrand N, Ramstrand S (2010) AAOP state-of-the-science evidence report: the effect of ankle-
foot orthoses on balance—a systematic review. J Prosthet Orthot 22:P4–P23
Remelius JG et al (2012) Gait impairments in persons with multiple sclerosis across preferred and fixed
walking speeds. Arch Phys Med Rehabil 93(9):1637–1642. doi: 10.1016/j.apmr.2012.02.019
Rougier P et al (2007) What compensatory motor strategies do patients with multiple sclerosis
develop for balance control? Rev Neurol 163:1054–1064
Sandroff BM et al (2014) Comparing two conditions of administering the six-minute walk test in
people with multiple sclerosis. Int J MS Care 16:48–54
Sandroff BM, Pilutti LA, Motl RW (2015) Does the six-minute walk test measure walking
performance or physical fitness in persons with multiple sclerosis? NeuroRehabilitation 37
(1):149–155
Savci S et al (2005) Six-minute walk distance as a measure of functional exercise capacity in

multiple sclerosis. Disabil Rehabil 27:1365–1371
Schubert M et al (1998) Walking and fatigue in multiple sclerosis: the role of the corticospinal
system. Muscle Nerve 21:1068–1070
Scott SM et al (2013) Quantification of gait kinematics and walking ability of people with multiple
sclerosis who are new users of functional electrical stimulation. J Rehabil Med 45:364–369
Sehle A et al (2014) Objective assessment of motor fatigue in multiple sclerosis: the Fatigue index
Kliniken Schmieder (FKS). J Neurol 261(9):1752–1762
Sheffler LR et al (2008) Functional effect of an ankle foot orthosis on gait in multiple sclerosis: a
pilot study. Am J Phys Med Rehabil 87:26
Sheffler LR et al (2009) Neuroprosthetic effect of peroneal nerve stimulation in multiple sclerosis: a
preliminary study. Arch Phys Med Rehabil 90:362–365
Smith MM, Arnett PA (2005) Factors related to employment status changes in individuals with
multiple sclerosis. Mult Scler 11:602–609
Sosnoff JJ, Shin S, Motl RW (2010) Multiple sclerosis and postural control: the role of spasticity.
Sosnoff JJ et al (2011a) Influence of spasticity on mobility and balance in persons with multiple
sclerosis. J Neurol Phys Ther 35:129–132
Sosnoff JJ et al (2011b) Mobility, balance and falls in persons with multiple sclerosis. PloS One 6:
e28021
Sosnoff JJ, Sandroff BM, Motl RW (2012) Quantifying gait abnormalities in persons with multiple
sclerosis with minimal disability. Gait Posture 36(1):154–156. doi: 10.1016/j.gaitpost.2011.11.027
Taylor PN et al (1999) Clinical use of the Odstock dropped foot stimulator: its effect on the speed
and effort of walking. Arch Phys Med Rehabil 80:1577–1583
Thickbroom GW et al (2006) Central motor drive and perception of effort during fatigue in multiple
sclerosis. J Neurol 253:1048–1053
Thickbroom GW et al (2008) Enhanced corticomotor excitability with dynamic fatiguing exercise
of the lower limb in multiple sclerosis. J Neurol 255:1001–1005
Thoumie P, Mevellec E (2002) Relation between walking speed and muscle strength is affected by
somatosensory loss in multiple sclerosis. J Neurol Neurosurg Psychiatry 73:313–315
Thoumie P et al (2005) Motor determinants of gait in 100 ambulatory patients with multiple
Trisolini M et al (2010) Global economic impact of multiple sclerosis. Multiple Sclerosis Interna-
tional Federation London, United Kingdom [Online]
van der Linden ML et al (2014) Habitual functional electrical stimulation therapy improves gait
kinematics and walking performance, but not patient-reported functional outcomes, of people
with multiple sclerosis who present with foot-drop. PloS One 9:e103368
Wagner JM et al (2014) Plantarflexor weakness negatively impacts walking in persons with multiple
sclerosis more than plantarflexor spasticity. Arch Phys Med Rehabil 95(7):1358–1365. doi:
10.1016/j.apmr.2014.01.030
Yahia A et al (2011) Relationship between muscular strength, gait and postural parameters in
multiple sclerosis. Ann Phys Rehabil Med 54:144–155
Zackowski KM et al (2009) Sensorimotor dysfunction in multiple sclerosis and column-specific
magnetization transfer-imaging abnormalities in the spinal cord. Brain J Neurol 132:1200–1209
Brain-Computer Interfaces for Motor
Rehabilitation
Rüdiger Rupp
Abstract
Injuries of the central nervous system such as stroke or spinal cord injury are a
major cause for severe motor impairments. The inability to activate muscles
voluntarily and the associated loss of ambulatory or manipulation skills constitute
a substantial handicap in the patients’ life. This results in a tremendously reduced
quality of life and represents a severe barrier for social and professional integra-
tion. Therefore, rehabilitation aims at achieving the greatest amount of autonomy
in everyday life by application of restorative or compensatory therapeutic
approaches. While restorative strategies are based on principles of motor learning
and aim at the recovery of the original function, compensatory therapies are
applied in cases where recovery is unlikely to happen and are often based on
the use of assistive technology.
Brain-computer interfaces (BCIs) are an emerging technology that measure
brain activities and translate them into control signals for a variety of assistive
devices. BCIs may contribute to both rehabilitative and compensatory therapeutic
strategies. While most of the BCI-controlled assistive technology such as com-
munication devices, robot arms, or neuroprostheses based on functional electrical
stimulation focus on the compensation of a lost function, there are only a few,
however, very promising examples on the successful use of rehabilitative BCIs
for restoration of grasping or walking.
Although assistive and rehabilitative BCIs seem to be a valuable component of
motor rehabilitation programs, more end user studies are needed to reveal the full
potential of BCIs for better participation and improved quality of life.
R. Rupp (*)
Spinal Cord Injury Center – Experimental Neurorehabilitation, Heidelberg University Hospital,
Heidelberg, Germany
e-mail: ruediger.rupp@med.uni-heidelberg.de

DOI 10.1007/978-3-319-30808-1_67-1
2 R. Rupp
Keywords
Brain-computer interface • Brain-machine interface • Motor rehabilitation • Res-
toration • Recovery • Compensation • Robotic arm • Neuroprosthesis • Functional
electrical stimulation • Grasp • Exoskeleton • Ambulation
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Noninvasive Brain-Computer Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Invasive Brain-Computer Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Applications of BCIs in Motor Rehabilitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Brain-Computer Interfaces for Control of Assistive Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Brain-Computer Interfaces for Movement Restoration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Introduction
Among the consequences of injuries or diseases of the central nervous system (CNS)
are impairments of motor functions such as grasping, trunk stability, or walking. The
limited ability to voluntarily initiate movements up to the complete loss of motor
functions represents a major barrier for leading an independent, self-determined life
and results in a substantial decrease of the perceived quality of life. People with
restrictions of motor functions are often not fully integrated in the society resulting in
a relevant economic burden on the social system.
Stroke is one of the most prevalent neurological conditions worldwide and one of
the leading causes of motor impairment (Warlow et al. 2008). Strokes can be
classified into ischemic and hemorrhagic. Ischemic strokes, which form the majority
of strokes, are caused by an interruption of the blood supply of parts of the brain,
while hemorrhagic strokes result from the rupture of a blood vessel and the associ-
ated bleeding. In Europe every year 1.1 Mio. first strokes occur, from which 70%
survive the first year, 80% with a hemiparesis/-plegia, i.e., impaired to no upper and
lower extremity function on one side of the body (Truelsen et al. 1997). Estimates
suggest that ~80% of all stroke survivors with upper limb motor deficits do not fully
regain the function of the affected limb (Hendricks et al. 2002). Moreover, the
economic burden of stroke is high and is likely to increase in the future (Feigin
et al. 2009).
In Europe, an estimated number of 330,000 people are affected by the conse-
quences of spinal cord injury (SCI) with 11,000 new injuries per year (Ouzký 2002;
van den Berg et al. 2010). Approximately 50% of the SCI population are tetraplegic
due to injuries of the cervical spinal cord with paralyses of the lower as well as the
upper extremities. The bilateral loss of the grasp function severely limits the affected
individuals’ ability to live independently (Anderson 2004; Snoek et al. 2004) and
retain gainful employment postinjury (NSCISC 2015). In industrial countries, there
Brain-Computer Interfaces for Motor Rehabilitation 3
is trend toward a higher percentage of people with incomplete SCI and with high
cervical injuries (McCaughey et al. 2016). Additionally, in the western world, the
percentage of nontraumatic SCI due to spinal infections, ischemia, or degenerative
spine diseases increases over the last decades resulting in more incomplete injuries
with preserved sensory and/or motor functions below the level of injury (Exner
2004).
The aim of rehabilitation of neurological patients is to provide them with as much
autonomy as possible. A strong focus is put on the ability to perform activities of
daily living such as dressing/undressing, personal hygiene, eating, computer opera-
tion, or mobility in indoor and outdoor environments. In this context, motor reha-
bilitation is based on two fundamental therapeutic principles, namely, restoration and
compensation/substitution. Restoration or recovery stands for the reacquisition of
elemental motor patterns by motor learning. In the absence of reacquisition, func-
tional improvements are achieved by adaptation of remaining (compensation) or
integration of alternative (substitution) motor elements.
The term recovery of motor performance is defined as the restoration of elemental
motor pattern present prior to CNS injury meaning that a given task is performed
using the same end effectors and joints in the same movement patterns typically used
by able-bodied individuals. Although the neurobiological basis for motor recovery is
not yet fully understood, it is generally accepted that the structural and functional
reorganization of neural networks at different levels of the CNS largely contributes
to it. This ability of the CNS for reorganization is called neuroplasticity.
In chronic stroke, the present view of brain reorganization is that overuse of the
unimpaired contralesional and underuse of the ipsilesional hemisphere leads to
increased inhibition of the ipsilesional hemisphere by the contralesional hemisphere.
This inhibition is thought to block excitatory reorganization of the intact ipsilesional
areas and block recovery of the affected motor system (Ward and Cohen 2004). This
hypothesis is supported by the positive effects of constraint-induced movement
therapy (CIMT) in chronic stroke (Mark and Taub 2004). In this therapy, physical
restraint of the healthy limb for an extended period of time forces the patient to use
the paretic limb and increases excitatory neural activity in the lesioned hemisphere.
Therefore, modern stroke rehabilitation aims at assisting or inducing the reorgani-
zation of neural circuits to achieve motor recovery.
Motor compensation is defined as the appearance of new motor patterns resulting
from the adaptation of remaining motor elements or substitution, meaning that func-
tions are taken over, replaced, or substituted by technical aids or assistive devices.
The decision, whether a restorative or compensatory/substitutional therapeutic
strategy is applied, depends in general on several factors, but time after injury and
severity of the injury are the most important ones (Fig. 1). The potential for
spontaneous motor recovery is highest during the first 6 months after SCI or stroke
(Langhorne et al. 2011; Curt et al. 2004). In patients with initially severe impair-
ments or complete SCI, the potential for recovery is much lower than in patients with
moderate impairments or motor incomplete lesions. The resulting commonly
accepted therapeutic guideline states that in less severe and more acute CNS injuries,
the therapeutic focus is on restoration of the original function. This does not mean
4 R. Rupp
Fig. 1 Therapeutic strategy in relation to severity and time after CNS lesion
that intensive restorative therapies do not lead to any functional improvement in the
chronic stage, but the degree of the improvement is likely much lower than in the
subacute stage.
Over the last two decades, the use of technology in the rehabilitation of motor
impairments has constantly increased both in restorative therapies based on motor
learning approaches as well as for compensation or substitution of permanently
restricted or lost motor functions. In a restorative setting, body weight-supported
treadmill training with or without the help of motorized gait orthoses helps to
achieve a high number of task repetitions – one of the key principles of motor
learning – and represent an established therapy for improving ambulatory functions
(Dietz and Fouad 2014). Passive or active arm support systems help patients to
accomplish specific upper extremity motor tasks and thereby improving their motor
skills (Prochazka 2015).
If lower extremity function is lost, wheelchairs provide some degree of mobility.
In recent times, active mobile exoskeletons allow individuals with no or almost no
residual muscle functions in the lower extremities but stable trunk to ambulate at
least over short distances (Rupp et al. 2015b). Grasp neuroprostheses on the basis of
functional electrical stimulation (FES) alone or in combination with an active
orthosis or robot arms with several degrees of freedom compensate for the loss of
upper extremity function (Rupp et al. 2015a).
Recent advances in neurotechnology have led to the development of brain-
computer interfaces (BCIs, also named brain-machine interfaces (BMIs)) that
might become an important component of new technological strategies that strive
to overcome severe CNS-injury-related motor impairments. First, BCIs may bypass
nonfunctional corticospinal pathways to allow for brain control of FES devices or

high-dimensional robotic arms to assist in daily life activities (assistive BCI).
Second, BCIs could facilitate CNS and in particular brain neuroplasticity (see
chapter “▶ Brain and Behavioural Bases of Guitar Learning in Novices”), thus
aiming at movement restoration by enhancing motor learning and motor recovery.
If a BCI is used in the context of restoration of motor functions, it is often referred to
as rehabilitative BCI, although the term “restorative BCI” would be more precise
(Soekadar et al. 2015; Chaudhary et al. 2016).
State of the Art
BCIs are technical systems that provide a direct connection between the human brain
and a computer (Wolpaw et al. 2002). These systems are able to detect thought-
modulated changes in electrophysiological brain activity and transform the changes
into control signals (Fig. 2).
A BCI system consists of five sequential components: (1) signal acquisition,
(2) feature extraction, (3) feature translation, and (4) classification output, which
interfaces to an output device, and is providing (5) feedback to the user. These
components are controlled by an operating protocol that defines the onset and timing
of operation, the details of signal processing, the nature of the device commands, and
the oversight of performance (Shih et al. 2012). Although all implementations of
BCIs build upon the same basic components, they differ substantially in regard to the
degree of invasiveness, their complexity in terms of hard- and software components,
the underlying physiological mechanisms, and their basic mode of operation (cue
based, synchronous vs. asynchronous).
Noninvasive Brain-Computer Interfaces
Noninvasive BCIs obviously represent the first choice in end-user applications due
to their ease of application. Although a variety of different data acquisition methods
can be used for setup of a BCI, e.g., near infrared spectroscopy (NIRS), functional
magnetic resonance imaging (fMRI), and magnetoencephalography (MEG), most of
the practically used noninvasive systems rely on brain signals that are recorded by
electrodes on the scalp (electroencephalogram, EEG). EEG-based BCI systems are
based on relatively inexpensive, small equipment, which is commercially available,
and therefore offer the possibility of everyday use in the clinical rehabilitation
environment and at end users’ homes.
Depending on the residual capabilities of the end user and the requirements of the
application in terms of intuitiveness, training time, or speed, a variety of EEG signals
have been used as noninvasive measures of brain activity. The most common are
signals such as event-related potentials (ERPs), steady-state responses mostly visual
evoked potentials (SSVEP), and frequency oscillations particularly of sensorimotor
rhythms (SMRs). For rehabilitative BCIs, mainly the motor-imagery- or motor-
6 R. Rupp
Fig. 2 General framework of brain-computer interfaces for motor rehabilitation: Invasive BCI
approaches (right) include the measurement of local field potentials (LFPs) and electro-
corticography (ECoG). Noninvasive BCI approaches (left) are mainly based on EEG. Brain signals
are processed to extract features relevant to the aim of the BCI and then classified using a
translational algorithm to construct a control signal that drives the BCI. BCIs can be classified as
assistive to substitute for a loss of motor functions or as rehabilitative to enhance neural recovery
execution-based modulation of SMRs registered with electrodes over the motor

cortex is used.
Depending on the underlying signal, EEG-based BCIs can be categorized into
endogenous, asynchronous and exogenous, synchronous systems. Synchronous
BCIs depend on the electrophysiological activity evoked by external stimuli and
do therefore not require intensive training. The most common synchronous BCI is
Table 1 A overview of the most common practical types of EEG-based assistive BCIs together
with their minimal number of electrodes, a qualitative estimation of typical training times, and their
typical accuracy and bit rate
Parameter
Minimal (typical) Population with 90–100% Typical rate
number of Training (below 80%) accuracy without of decisions/
BCI electrodes time training min
SMR 4 (10) + 1 Weeks to 6% (81%) 4 Bits/min
(2-class) reference + months
1 ground
P300 3 (9) + 1 reference Minutes 73% (11%) 10 Bits/min
+ 1 ground to <1 h
SSVEP 6 + 1 reference + Minutes 87% (4%) 12 Bits/min
1 ground to <1 h
based on the P300 event-related potentials. Although systems based on steady-state

evoked potentials such as SSVEPs combine components of asynchronous and
synchronous approaches, the introduction of cues improves their accuracy. Asyn-
chronous BCIs depend on the users’ ability to voluntary modulate their electrophys-
iological activity such as the EEG amplitude in a specific frequency band. In
asynchronous BCIs, the time point for changes of the control signals is not pre-
defined by the system, but the user is free to initiate decisions at any time. Examples
for this class of BCIs are SMR-based systems, for which users train to modulate the
power of specific EEG frequency bands. In rehabilitative BCIs, this training forms
the basis to initiate neuroplastic changes in the brain, while in assistive BCIs, it
represents a necessary evil to achieve robust control. Depending on the brain signals
used for operation, BCIs greatly vary in regard to the minimal and typically used
number of electrodes, training times, accuracies, and typical information transfer
rates (for overview see Table 1). In any case, at the current state of the art, all
noninvasive assistive BCIs have a significantly lower performance than manual
input devices like keyboards, computer or head mouses, or joysticks. Therefore an
assistive BCI represents only an alternative user interface in end users, who are
generally not able to operate such traditional user interfaces, cannot use these
devices over a prolonged period of time, or need an additional control input.
BCIs Based on Event-Related Potentials

ERP-based BCIs make use of the fact that specific neural activity is triggered by and
involved in the processing of specific events. These systems are exclusive assistive
BCIs and are implemented with an oddball paradigm, wherein a rare target (oddball
event) is presented within frequent non-target events. They usually exploit an
endogenous ERP component, known as P300, as input signal. The P300 is a positive
deflection in the EEG occurring 200–500 ms after the presentation of a rare visual,
auditory, or somatosensory stimulus and is a reliable, easy to detect ERP. By
focusing attention on the rare target, e.g., by keeping a mental count of its occur-
rence, the P300 amplitude can be increased, and therefore its detection and classi-
fication improves (Kleih et al. 2011). In individuals with stroke or SCI, eye gaze is
8 R. Rupp
preserved, and thus a visual rather than an auditory oddball paradigm is the preferred
choice. The information transfer rate and accuracy is substantially higher, and
perceived workload is much lower in visual P300-based BCIs than in other
ERP-based BCIs. The big advantage of P300 compared to SMR-based BCIs is
that they can be operated with almost no training time in 99% of the general
population. Typical applications are virtual keyboards for electronic communication
or environmental control systems.
BCIs Based on Steady-State Evoked Potentials

Steady-state evoked potentials (SSEPs) are stable oscillations that can be elicited by
rapid repetitive (usually >6 repetitions per sec) visual, auditory, or sensory stimuli.
The most common type of SSEP-based BCIs is the SSVEP-based BCI, where screen
objects flickering at different frequencies are visually presented to users. Focusing
their attention to the intended stimulus elicits enhanced SSVEP responses at the
corresponding frequency, which can be detected, classified, and translated into
control commands. SSVEP-based BCIs have the advantages of a high information
transfer rate, practically no training time, and they work in almost every user.
SSVEP-based assistive BCIs are the preferred choice in individuals with stroke or
SCI and unimpaired visual function, because the information transfer rate of BCIs
based on auditory steady-state responses is tenfold lower than of SSVEP-based
systems. Typical applications are the same than in ERP-based assistive BCIs.
BCIs Based on Sensorimotor Rhythms

The type of EEG-based BCIs that can be used as rehabilitative as well as assistive
BCIs exploits the modulation of SMRs. These rhythms are oscillations in the EEG
occurring in the alpha/mu (8–12 Hz) and/or beta (18–26 Hz) bands and can be
recorded over the primary sensorimotor areas on the scalp. While the mu rhythm is
more pronounced in the EEG during “idling” (event-related synchronization, ERS),
its amplitude typically decreases (event-related desynchonization, ERD) during
actual movement and similarly during mental rehearsal of movements (motor imag-
ery, MI) (Neuper et al. 2005; Pfurtscheller and Lopes da Silva 1999). It has been
shown that able-bodied people can learn to modulate the SMR amplitude by
practicing MIs of simple movements, e.g., hand/foot movements (Kaiser et al.
2014; Toppi et al. 2014). This process occurs in a closed loop, meaning that the
system recognizes the SMR amplitude changes evoked by MI and these changes are
instantaneously fed back to the users. This neurofeedback procedure together with
mutual human-machine adaptation normally enables BCI users after some weeks of
training to control their SMR activity and use these modulations to control output
devices in an asynchronous, self-paced manner (Cincotti et al. 2008; Pineda et al.
2003). With SMR-based assistive BCIs, a sophisticated level of control, e.g., a three-
dimensional cursor control, can be achieved (McFarland et al. 2010).
While assistive BCIs based on the registration of P300 and SSVEPs can be
operated by a vast majority of users, SMR-BCIs cannot. In the SMR-based assistive
BCI approach, in up to one third of the non-motor-impaired participants, the BCI is
unable to detect classifiable, task-related EEG patterns. Consequently, these subjects
cannot quickly be provided with a BCI-controlled application or need at least a

substantial amount of training to reach a sufficient level of performance. The causes
for this inability to control a BCI (other synonyms are BCI inefficiency, BCI aptitude,
BCI illiteracy) have not yet been satisfactorily described. Most of the experiments on
the gain in performance with training in SMR-BCIs were performed in able-bodied
subjects and it is still unclear, in how far these results can be generalized to persons
with CNS injuries. In a single case study, in which an individual with a lesion of the
upper cervical spinal cord was provided with a BCI-controlled upper extremity
neuroprosthesis, no training effects occurred over a training time of more than
6 months. Even after 415 MI-BCI runs, the end user’s average performance did not
show any trend toward improvement, but remained at about 70% with large day-to-
day variances (Rohm et al. 2013). This confounding factor represents among others a
challenge for successful use of SMR-BCIs (Rupp 2014).
For a typical 2-class SMR-BCI, different paradigms of MIs such as one hand
vs. feet or left vs. right hand are used either in a switch-based (“brain switch”)
fashion by introduction of a threshold or in an analog manner by directly connecting
the classifier output to the output device. An often underestimated problem in
practical applications of BCIs and in particular of SMR-based BCIs is the detection
of a non-intention condition, during which a user does not want to send any
command (zero class). This so called zero-class problem is often handled in brain
switch implementations by defining one MI class as the resting class or to use long
MIs to pause or reactivate the system (Rupp 2014). However, this approach is not
appropriate for all applications, which renders the zero-class problem as one of the
major limiting factors for practical use of SMR-BCIs.
MI-BCIs offer further possibilities in the context of neurorehabilitation of patients
with CNS injuries that go beyond the traditional use for control of assistive devices.
After a CNS injury, substantial functional brain reorganization occurs that plays a
critical role for functional recovery and may have pathological consequences
(Nardone et al. 2013). The basis for a therapeutic use of BCIs is formed by the
fact that the central nervous system shows a lifelong ability for neural plasticity,
which can be enhanced after a trauma or injury by task-specific training (Dietz and
Fouad 2014). The key elements for an effective neurorehabilitative training based on
motor learning are voluntarily triggered movement intentions and a synchronized
sensory and proprioceptive feedback of the limbs’ motor actions. BCIs hold promise
to enable the detection of intended movements, e.g., the hand, even in patients with
no preserved movements, making them an ideal tool for closed-loop neuroreh-
abilitative therapies when used in combination with grasping and reaching
neuroprosthesis (Jackson and Zimmermann 2012; Rupp et al. 2013; Savic et al.
2014). Additionally, by practicing feedback-controlled MI of paralyzed limbs, the
integrity of cortical neuronal connections may be preserved or neurological recovery
of motor function may be even enhanced (Kaiser et al. 2014).
Hybrid BCIs
A novel development in particular noninvasive BCI research is the introduction of
the hybrid BCI (hBCI) concept (Müller-Putz et al. 2015). An hBCI consists of a
10 R. Rupp
combination of multiple BCIs or a BCI with other input devices. These input devices
may be based on the registration of biosignals other than brain signals, such as
myoelectric activities. The hBCI concept helps to overcome limitations inherent to a
singular BCI system, e.g., false-positive, unintended decisions, or the zero-class
problem. In fact, the second input signal can be effectively used to indicate an
“idling” state or to introduce a context-specific correction mechanism. Using this
approach, a user can generate a single command signal either by fusing different
input signals or by simply selecting one of them (Müller-Putz et al. 2011). In the
latter case, the input signals can be dynamically routed based on their reliability, i.e.,
continuously monitoring the quality, and the input channel with the most stable
signal will then be selected (Kreilinger et al. 2011). In the case of signal fusion, each
of the input signals contributes to the overall command signal with a dedicated
weighting factor. These factors are generally not static but can be dynamically
adjusted in accordance with their reliability, which is quantified by appropriate
quality measures. The hBCI is fully compliant with the user-centered design concept
(ISO 2010). The key message of this approach is that the technology has to be
adapted to the individual users’ ability and needs and not vice versa. Combining
BCIs with established user interfaces may allow more end users to control assistive
technology or may simplify the use of existing devices. However, this extension of
the target population comes with the drawback that longer preparation times are
needed for setup of the additional components of the hBCI. From the users’
perspective, it is important to carefully evaluate the design of the hBCI’s control
scheme and not to cause additional mental workload. Control schemes based on a
sequential control task of the different input signals are – at least at the beginning of
the training – superior to those, for which a user must control different input signals
simultaneously (Rupp et al. 2014). With practice, users might learn to perform
multiple tasks, thereby making full use of the hBCI approach.
In any case, the hBCI concept helps to overcome limitations inherent to a singular
BCI system, e.g., false-positive, unintended decisions, or the zero-class problem. In
fact the second input signal can be effectively used to indicate an “idling” state or to
introduce a context-specific correction mechanism. The use of brain signals recorded
by the BCI itself is a very elegant way of setting up an hBCI. An example is the use
of error potentials to detect unintended BCI decisions and correct them (Chavarriaga
et al. 2014).
Invasive Brain-Computer Interfaces
Although noninvasive EEG-based BCIs are easily available and were successfully
applied in individuals with CNS injuries, they face several problems in particular for
long-term control of assistive devices: While temporal resolution is generally good,
spatial resolution of noninvasive BCIs is rather poor due to the summed recording of
signals from large cell populations (up to one million neurons per EEG electrode).
The poor selectivity is the reason why with noninvasive BCIs, an intuitive,
simultaneous control of many degrees of freedom of e.g., a robot arm, is hardly

possible. The signal-to-noise ratio of EEG-based systems and also the temporal
resolution are limited because of the low-pass characteristics of the bony structures
of the skull. Noninvasive BCIs are prone to artifacts caused by movements of
electrodes and caps and crosstalk from muscles, e.g., during eye movements. In
most of the BCI systems, electrode contact gel is needed, which results in a relatively
time-consuming montage and needs to be washed out from the hair of the end users
after use.
Most of the disadvantages of noninvasive BCIs can be overcome by invasive
BCIs, where electrodes for recording of brain activity are implanted intracranially.
Although it is known that many cortical areas contribute to the planning and
execution of movements, the primary motor cortex (M1) is assumed to be the area
most involved in motor actions. The site of implantation can vary from extradural to
intraparenchymal, depending on the technology of the electrodes. In general, inva-
sive electrodes are directly placed onto or in the cortex, which allows the recording
of multiple forms of electrical potentials like high-frequency oscillations, local field
potentials (LFPs), and spikes from single or multiple neurons. LFPs are promising
sources for real-time control of multidimensional assistive devices such as robot
arms, because they seem to be more robust and the high-gamma LFP recordings may
contain more information about executed or imagined movements than recordings
from single neurons.
BCIs Based on Electrocorticographic Recordings

The least invasive BCI is based on the recording of the electrocorticogram (ECoG).
The electrodes typically consist of platinum discs with a diameter of 2–3 mm
embedded in a silicon or polymer sheet. These sheets are placed epi- or subdurally
but never penetrate the brain tissue and therefore have a low risk for damage of
neuronal tissue. With ECoG- und microECoG-based BCIs, local electrical field
changes can be recorded with far better spatial resolution than with noninvasive
BCIs. Like in EEG-based assistive BCIs, the modulation of cortical activity patterns
is robust and stable once they have been established by training. It has been recently
shown in an individual with high-level SCI that a 3D-cursor control can be achieved
within a training period of approx. 20 days (Wang et al. 2013). In his case, the grid
was removed according to the study protocol after 28 days, which unfortunately
makes it impossible to evaluate the BCI performance and the stability of the brain
signals over a longer period of time. The few observations that were made with long-
term implantations of ECoG electrode grids indicate that the limited fluid exchange
and the mechanical irritation of the brain tissue by the rigid electrode grid lead to
signs of chronic inflammation. However, this may be prevented by the future
development of very thin, flexible, and porous grids. For minimization of the risk
of infections including wound infections, meningitis, and osteomyelitis due to the
transcutaneous wires, a fully implantable ECoG grid would be highly desirable but is
not yet available for human use. Surgery-related and noninfectious complications
among them cerebrospinal fluid leaks, neurological injuries, and subdural/deep
12 R. Rupp
hemorrhage represent a general problem in invasive BCIs, which need to be clearly

communicated to end users prior to implantation (Blabe et al. 2015).
BCIs Based on Intracortical Recordings

The spatial resolution of ECoG recordings is still not sufficient for decoding of
imagery of fine motor skills, e.g., movements of single fingers. Higher spatial
resolution can only be achieved with intracortical recordings. For this purpose,
high-density, multi-channel electrodes were developed which are implanted into
the neocortex either epipial or intraparenchymal and record single-cell activities
through cone electrodes or microelectrode arrays (MEA). Most of the research
carried out on BCIs with intracortical recordings is based on the Utah array MEA,
which is commercially available from Blackrock Microsystems (Salt Lake City, UT,
USA). This array consists of 100 needle electrodes, which are placed on a rectan-
gular carrier with a dimension of 4 4 mm. The electrode array is wired with a
transcutaneous connector plug, which is fixed by screws with the skullcap and
transmits the signals to an amplifier mounted on top of the head. Special care
needs to be taken of the area around the connector to avoid local infections and
more severe adverse events such as meningitis. For implantation of the MEA, special
equipment in the form of a pneumatically driven insertion device and well-trained
surgeons are needed to minimize brain tissue damage. If applied properly, the
damage on brain tissue seems to be minor with negligible functional consequences
(Bouton et al. 2016; Flesher et al. 2016).
As with every invasive procedure, the implantation of electrodes bears some
surgical risks. Long-term complications may occur such as cerebrospinal fluid leakage
or infections due to the transcutaneous wires. In any case, a fully implanted recording
device needs to be developed to increase the usability and lower the risk for infections.
The ultrafine structures of the electrode tips are not long-term stable, and after 5 years,
only a few electrodes may be left for proper recording of neuronal activity (Hochberg
et al. 2012). The loss of signals is mainly caused by degradation of electrode materials,
gliosis, or chronic inflammation. This may be overcome by coating of the electrodes
with extracellular matrix or with short adherent proteins to minimize inflammatory
responses and support cell proliferation and differentiation (Gunasekera et al. 2015).
From a mechanical viewpoint, these small electrode arrays tend to sink into the
nonrigid brain tissue. The long-term stability and biocompatibility are key prerequi-
sites for routine clinical use of MEAs, because their removal causes additional damage
to the brain tissue adhering to them. Further, it is not clear whether with a new implant
the same level of performance will be reached as with the explanted device.
Although BCIs based on intracortical recordings can be operated by the end user
with almost no training time, daily tuning of the movement decoding algorithms
might be necessary (Bouton et al. 2016). Only one third of the electrodes used for
control of the robot on one day contribute to its operation on the next day (Collinger
et al. 2013). At the current state of the art, this tuning procedure needs the support of
technically qualified personnel and therefore automatic tuning algorithms need to be
introduced to allow for the autonomous operation of a BCI with intracortical
electrodes by the caregivers of end users at their homes.
Applications of BCIs in Motor Rehabilitation
In motor rehabilitation, BCIs are mostly used as assistive BCIs for direct brain
control of assistive devices in patients with limited preserved motor functions.
However, over the last few years rehabilitative BCIs have gained interest for
fostering neuroplasticity and functional recovery (van Dokkum et al. 2015). While
most of the applications of BCIs focus on the rehabilitation of upper extremity
function, in recent years some efforts were also undertaken to use BCIs for gait
rehabilitation.
Brain-Computer Interfaces for Control of Assistive Devices
The first applications of BCIs described in the early 1990s were EEG-based com-
munication systems for patients with locked-in syndrome not able to speak (Kübler
et al. 1999; Wolpaw and McFarland 1994). Therefore, there is a history of almost
two decades with the use of brain signals for assistive BCIs. While the first goal was
to allow communication, over time other BCI applications in particular for motor
rehabilitation such as the control of grasp neuroprosthesis based on functional
electrical stimulation, multi-degree-of-freedom robot arms to replace a lost upper
extremity function, and in recent time exoskeletons for substitution of the loss of
lower extremity functions have moved into the focus. Most of these assistive systems
are applied in individuals with complete spinal cord injury to compensate for the
associated permanent loss of motor and sensory function below the level of injury,
while in stroke survivors mainly efforts in the direction of motor recovery are
undertaken.
BCIs for Control of Upper Extremity Neuroprostheses

Neuroprostheses based on functional electrical stimulation (FES) represent the only
possibility for at least partial restoration of permanently restricted or lost functions in
case of missing surgical options. Over the last three decades, FES systems with
different levels of complexity were developed, and some of them were successfully
transferred into clinical routine. FES systems deliver short current impulses eliciting
physiological action potentials on efferent nerves, which cause contractions of the
innervated yet paralyzed muscles of the arm. On this basis, FES artificially compen-
sates for the loss of voluntary muscle control. The simplest form of a neuroprosthesis
is based on the application of multiple surface electrodes. With a small number of
only seven surface electrodes placed on the forearm, two grasp patterns, namely, a
key grasp and power grasp, can be restored (Rupp et al. 2015a). With implantable
systems, the main disadvantages of noninvasive systems may be overcome, which
include limited selectivity and stability of grasp patterns, simpler handling, and long-
term stability of stimulation parameter.
Through the last decade, it has become obvious that the user interface of all
current FES devices is not optimal in the sense of natural control. In the case of
individuals with a high, complete SCI and the associated severe disabilities, not
14 R. Rupp
enough residual functions are preserved for control either by movements or by

underlying electrical muscle activation signals of a non-paralyzed body part. This
has been a major limitation in the development of a reaching neuroprostheses for
individuals with a loss not only of hand and finger but also of elbow and shoulder
function.
In 2003, a pioneering work showed for the first time that a MI-BCI control of a
neuroprosthesis based on surface electrodes is feasible (Pfurtscheller et al. 2003). In
this single case study, the restoration of a lateral grasp was achieved in a tetraplegic
subject, who suffers from a chronic SCI with completely missing hand and finger
function. The end user was able to move through a predefined sequence of grasp
phases by imagination of foot movements detected by an SMR-BCI with 100%
accuracy (Fig. 3). He reached this performance level already prior to the experiment
by some weeks of training with the MI-BCI and has maintained it for almost a
decade by regular continuation of the training (Enzinger et al. 2008).
A second feasibility experiment has been performed, in which a short-term
BCI-training was applied in another individual with tetraplegia and also missing
hand and finger function. This subject was using an implanted Freehand system for
several years. After 3 days of training, the end user with chronic SCI was able to
control the grasp sequence of the implanted neuroprosthesis with a moderate but
sufficient performance (Müller-Putz et al. 2005). In contrast to the first single case
study, in which nonintuitive imaginations of foot movements were used as brain
switch, a more intuitive control scheme was implemented, in which an imagination
of left-hand movements triggered changes in the grasp pattern of the right hand.
In these first attempts, the BCI was rather used as a substitute for the traditional
neuroprosthesis control interface than as an extension. With the introduction of
FES-hybrid orthoses, it becomes more important to increase the number of indepen-
dent control signals. With the recent implementation of the hBCI framework, it
became feasible to use a combination of input signals rather than BCI alone. In a first
single case study, a combination of a MI-BCI and an analog shoulder position sensor
is proposed (Rupp et al. 2015a). By upward/downward movements of the shoulder,
the user can control the degree of elbow flexion/extension or of hand opening/
closing. The routing of the analog signal from the shoulder position sensor to the
control of the elbow or the hand and the access to a pause state is determined by a
digital signal provided by the MI-BCI (Fig. 4). With a short imagination of a hand
movement, the user switches from hand to elbow control or vice versa. A longer
activation leads to a pause state with stimulation turned off or reactivates the system
from the pause state. With this setup a highly paralyzed end user, who had no
preserved voluntary elbow, hand and finger movements, was able to perform several
activities of daily living, among them eating a pretzel stick, signing a document, and
eating an ice cone (Fig. 5), which he was not able to perform without the
neuroprosthesis.
Recently, also an invasive BCI consisting of a Utah MEA in the primary motor
cortex was used to control a noninvasive grasp neuroprostheses based on a multi-
channel FES array electrode on the forearm (Bouton et al. 2016). With the help of
this system, the subject with a C5/C6 spinal cord injury and no voluntary finger and
Brain-Computer Interfaces for Motor Rehabilitation
Fig. 3 Subfigures (a, c): bandpass filtered (15–19 Hz) EEG channels C3 (right hand) and Cz (feet) recorded during the first 3-foot movement imaginations
(i1, i2, and i3) of one complete grasp sequence. The vertical lines indicate the start-on of the individual grasp phases. Subfigures (b, d): bandpower time courses
(15–19 and 20–60 Hz) of EEG channels C3 and Cz averaged over the five grasp phases with movement start at second 0 (amplitudes in arbitrary units).
Subfigure (e): A complete sequence of switches through the five grasp phases “hand open – finger flexed – key grip – hand open – stimulation off”
15
16 R. Rupp
Fig. 4 Subfigure (a): overview of the hBCI control of the upper extremity neuroprosthesis.
Subfigure (b): control scheme of the neuroprosthesis – a short imagination of a movement of the
right hand switches between shoulder and elbow control. A long imagination switches to or
reactivates the system from a pause state (Rupp et al. 2015a)
Fig. 5 Sequence of pictures showing a user with SCI at the level of C4 eating an ice cream cone.
The user starts in the hand control mode, lifts his left shoulder to open his right hand (a). After
successfully grasping the ice cream cone by depression of his left shoulder (b) the user initiates a
BCI command by imagination of a right hand movement to switch from hand to elbow control and
lifts his left shoulder to flex his right elbow (c). Now, the user licks the ice cream (d). Finally, the
user lowers his left shoulder to extend the elbow, puts the cone in its original place, switches back to
hand mode via BCI (e) and releases the cone by lifting his left shoulder (f) (Rupp et al. 2015a)
18 R. Rupp
thumb movements was able to generate six different movements and perform some
tasks of daily living such as pouring water from a bottle into a glass. Although these
experiments are highly interesting from a research standpoint, it must be emphasized
that in this subject, enough residual movements were preserved for control of the
neuroprosthesis. In his case, the additional benefit of using an invasive BCI in
respect to the risk of causing an additional impairment of preserved sensation of
the thumb and partial movements of his wrist must be critically verified. Further-
more, it is not clear, if in this setup the BCI control is cue based since a visual
feedback of the imagined grasp pattern is provided on a computer screen before the
hand movements are actually generated by FES.
BCIs for Control of Robotic Arms

In individuals with very high spinal cord injury, a total paralysis of the upper
extremities might be present. In those individuals, functional restoration of grasping
and reaching by FES is hard to achieve. The reasons for this are rapid fatigue of
continuously stimulated muscles needed to maintain a stable joint position such as
the elbow, insufficient ability to selectively activate single finger muscles, real-time
measurement of joint kinematics for closed-loop control, and the non-linear input-
output relationship of the neuromuscular FES activation. In these cases, robotic arms
with known dynamics and kinematics, and technical sensors to precisely determine
the position of the end point in space are from a technical perspective a much easier
to handle alternative to provide the end user with some degree of autonomy.
Invasive and noninvasive BCI systems have been used to enable neural control of
robotic arms (Hochberg et al. 2012; Collinger et al. 2013; Wang et al. 2013; Kim
et al. 2015). However, the few approaches that used noninvasive systems provided
limited control with a low success rate of hitting the intended target. In contrast,
BCIs with implantable electrodes allowed patients to simultaneously control many
degrees of freedom of a robotic arm, enabling end users to perform more complex
and functional movements. In the most recent studies involving individuals with
SCI, two intracortical MEAs (Utah Array) were implanted in the motor cortex about
14 mm apart (Collinger et al. 2013) and in another study in the posterior parietal
cortex (Aflalo et al. 2015), respectively. With these implants, it was shown in a
woman with high tetraplegia and no preserved movements in both arms that a very
sophisticated real-time control of a robot arm with seven and later with ten degrees of
freedom (three-dimensional translation, three-dimensional orientation, four-
dimensional grasping and hand shaping) was possible with very short amount of
training of a few days (Collinger et al. 2013; Wodlinger et al. 2015). This level of
simultaneous control of multiple degrees of freedom almost immediately after the
implantation has not been achieved with noninvasive MI-SMR-based BCIs so far.
From a neuroscientific viewpoint, it is highly interesting that the algorithm for
decoding of the movement intention is initially trained with the cortical activities
during observation of predefined movements of the robot arm (Hiremath et al. 2015).
One problem that still has to be solved is the maintaining of a stable position in
space. Here, intelligent shared control strategies with semiautonomous operation of
the robot arm may help to improve the performance (Downey et al. 2016). Another
problem that needs to be overcome before everyday use is the need for daily tuning
of the classification algorithms due to high inter-day variations in neuronal activation
patterns of the same movement task.
BCIs for Control of Robotic Lower Extremity Exoskeletons

Motorized lower extremity exoskeletons are an emerging technology, which have
become commercially available for clinical and personal use over the last 5 years,
some of them with approval from American and European regulation authorities
such as the Food and Drug Administration (FDA) (Contreras-Vidal et al. 2016).
Although devices from different manufacturers vary substantially in their technical
specifications and intended field of application, their common goal is to compensate
for a lost standing and walking function (see chapter “▶ Gait Rehabilitation with
Exoskeletons”). Potential users of exoskeletons need to fulfill many prerequisites,
e.g., sufficient trunk stability or minor spasticity/joint contractures, to successfully
use these systems in everyday life. Therefore, only a small number of end users will
likely be able to successfully operate such devices. Nevertheless, BCIs hold the
promise to serve as an intuitive user interface. In contrast to the need for fine motor
control in robotic arms or upper extremity neuroprosthesis, the demands are much
lower for sufficient control of an exoskeleton for walking. Exoskeletons with a
comprehensive set of sensors are capable to generate predefined movement trajec-
tories and to adapt them to changing environments autonomously, thereby relieving
the end user of the burden to precisely control the timing and the pattern of the
movements. This enables the implementation of shared-control schemes, where the
user switches between different modes such as sit-to-stand, walking, turning, and the
selection of walking or stair climbing/descending mode (Contreras-Vidal and
Grossman 2013). First implementations of an assistive BCI for control of a self-
stabilizing exoskeleton show the feasibility of such an approach (Contreras-Vidal
et al. 2012). Recently, also the feasibility of a real-time control of a lower extremity
FES system for ambulation by a BCI-based brain switch activated by foot MI was
shown in a single person with sensory incomplete but motor complete thoracic SCI
(King et al. 2015).
Brain-Computer Interfaces for Movement Restoration
In noninvasive BCIs based on the modulation of SMRs by imagination of move-

ments, some sort of training is necessary to achieve a sufficient degree of modulation
of the power in certain EEG frequency bands, which can then be used to control an
assistive device. The higher the degree of change of the power in a user-specific
frequency band, the more reliable is the control signal. To achieve a high accuracy of
the detection of MI and herewith a high level of robustness, normally training over
20 or more sessions is necessary. In a rehabilitative application, this BCI training is
not just a means to acquire good control of the system to efficiently send a specific
command to the outer world, but the training itself and its effects on brain motor
circuits are the final objective of the BCI application. There is growing evidence
20 R. Rupp
from case reports (Broetz et al. 2010; Daly et al. 2009) and small clinical trials (Buch
et al. 2008; Muralidharan et al. 2011; Prasad et al. 2010) that this training improves
motor function of stroke survivors in their daily activities.
The application of BCI technology in motor rehabilitation either alone or com-
bined with FES or exoskeletons offers an encouraging means for enhancing the
activity-dependent neuroplasticity of the CNS and to guide the spontaneous plastic
changes occurring in the CNS after injury. Although there is need for a better
understanding of the complex rules underlying this form of plasticity, two different
strategies for the application of BCI in stroke rehabilitation are currently under
investigation. The first strategy foresees the use of BCIs to train patients to produce
more normal brain activity. The hypothesis behind this approach is that more normal
brain activity reflects more normal brain function, and more normal brain function
results in an improvement of motor control. The second strategy is to use BCIs to
detect the movement intent of a user, to control an assistive device that executes this
movement, and thereby to generate external and internal sensory and proprioceptive
feedback to the CNS that matches the movement intention (Daly and Wolpaw 2008;
Ethier et al. 2015). By this approach, the physiological loop is restored, and the
desired corticospinal and intraspinal connections are strengthened promoting a better
functional outcome. This is in particular important in the rehabilitation of upper limb
function in stroke patients, in whom well-established rehabilitation strategies such as
constraint-induced movement therapy or bilateral arm training are not possible due
to a lack of residual motor abilities. This is unfortunately the case in 30–50% of all
stroke survivors, leaving them with few rehabilitative options (Langhorne et al.
2009). Here, the use of BCI systems relying on motor related brain activity could
offer a valuable tool to support training and practice even in the absence of residual
motor activities. For obvious reasons, noninvasive BCIs based on the recording of
either EEG, MEG, or NIRS are the preferred choice in motor rehabilitation
applications.
BCIs for Restoration of Grasping and Reaching

Although the restoration of walking represents the primary focus of rehabilitative
efforts in subacute stroke survivors, every improvement of upper extremity function
is beneficial in particular in patients with severe motor impairments. Patients with
hemiplegia after stroke generally exhibit limited potential for recovery in response to
conventional treatment strategies. In particular in this patient group, there is an
urgent need for better therapeutic approaches to achieve at least some level of
autonomy. In individuals with cervical SCI and the associated bilateral restrictions
of grasping, the restoration of upper extremity function has highest priority (Ander-
son 2004; Snoek et al. 2004).
Many studies have shown that mental practice can reduce motor impairments and
improve functional recovery of the upper limb in stroke patients. Motor imagery is
often applied in conventional therapy as a strategy to access the motor system after
damage resulting from stroke (Nilsen et al. 2010). One of the greatest restrictions in
the systematic application of this strategy is the impossibility to objectively monitor
the patients’ adherence to the therapist’s instructions and the lack of information on
the correctness of the performed mental task. From this perspective, BCI technology
might provide the therapist with an instrument capable to objectively monitor MI.
However, as outlined in section “State of the Art” of this chapter, there are
manifold signals used for BCI control and the question arises, which signal should
be used in rehabilitative BCIs. Additionally, it is crucial that the BCI training is
conducted in such a way that the MI task required to operate the system is performed
in a congruent, ecological setting. With respect to this, the feedback provided to the
patient during the exercise is of utmost relevance in order to keep him focused on the
required task (Prasad et al. 2010). The congruence of the feedback with the imagined
movement is of utmost importance, if the feedback is not only a visual perception of
movement, but is accompanied by a real movement of the patient’s limb, actuated
either by a means of a robotic device or by FES of the hand muscles. Although a final
conclusion on the most effective BCI signals and optimal feedback modalities
cannot be drawn and the literature is not fully consistent, there are at least some
indicators that some implementations of rehabilitative BCIs are more effective than
others.
In a first feasibility study, eight chronic stroke survivors with no residual finger
function in their affected hand underwent a MEG-based BCI training during which
they were asked to modulate their SMRs by performing the MI of their affected hand
in order to operate a mechanical orthosis that passively flexed or extended their
fingers (Buch et al. 2008). The MEG features that best discriminated the imagery
tasks from the rest condition were chosen as control features regardless of their
location on the scalp (either collected from the lesioned or the intact hemisphere).
This approach in the selection of control features is taken from established BCI
applications for communication and control, in which the best discriminative fea-
tures is chosen in order to achieve the highest accuracy (Wolpaw et al. 2002).
Although the training resulted in successful BCI control in six of eight patients,
clinical scales used to rate hand function showed no significant improvement after
training.
Other authors have been more selective with regard to the selection of control
features, based on the assumption that in motor recovery applications the source of
the signal adopted for BCI training must be as close as possible to normal activity.
Thus, features were selected comparing the EEG activity generated from MI of the
affected hand to that generated from MI of the unaffected limb (Daly et al. 2009). A
case report of a chronic stroke patient undergoing a BCI-controlled FES training
proved feasibility and showed promising results of such an integrated approach
(Daly et al. 2009).
In other studies, the control signal was collected from the ipsilesional hemisphere
only (contralateral to the imagined movement of the affected hand) (Ang et al. 2010;
Broetz et al. 2010; Caria et al. 2011). There is evidence that adaptive changes
occurring after stroke may result in an increased recruitment of contralesional
motor areas overcoming the more normal responsiveness of brain areas contralateral
to the limb involved in the motor task (Dimyan and Cohen 2011). Therefore,
collecting and reinforcing the signal from the affected hemisphere has the objective
of contrasting this “taking over” of the unaffected hemisphere.
22 R. Rupp
Fig. 6 In the prototype setting, the patient is seated with his hands resting on a desk, with an
adjustable forearm orthosis that provides support. The hands are covered by a flat screen monitor, on
which the cue and feedback for the patient as well as a visual representation of the patient’s hands
(“virtual hands”) are provided by a custom software. During the session, the therapist is allowed to
monitor the patient’s mental “activity” continuously through instant brain-computer interface
feedback, displayed on a dedicated screen. The degree of desynchronization of selected elec-
trodes/frequencies determines the vertical velocity of the cursor on the therapist’s screen; once
the cursor reaches a target in the upper part of the screen, the virtual hand performs the imagined
movement (feedback to patients in successful trials). The therapist is also allowed to monitor the
patient’s extent of muscle relaxation based on the electromyographic signal, recorded from the hand
and forearm muscles and displayed on a screen (Pichiorri et al. 2015)
Recently, the efficacy of a BCI-monitored ipsilesional MI practice as add-on

intervention to usual rehabilitation care was evaluated in a randomized controlled
pilot study in subacute stroke patients (Pichiorri et al. 2015). Twenty-eight hospital-
ized subacute stroke patients with severe upper extremity motor deficits were
randomized into two intervention groups: 1-month BCI-supported MI training
(BCI group, n = 14, Fugl-Meyer Assessment (FMA) arm section = 23.4 17.3
(max. 54)) and 1-month MI training without BCI support (control group; n = 14,
FMA = 24.2 18.2). The BCI-supported training was based on the MI only of the
affected hand to control the movement (either opening or closing) of a visual
representation of the own affected hand. To provide this visual representation, a
computer screen showing a virtual hand was placed above the own affected hand,
giving the patient the illusion of movements of his own hand (Fig. 6).
Functional and neurophysiological assessments were performed before and after
the interventions, including evaluation of the upper limbs by FMA and analysis of
oscillatory activity and connectivity at rest, based on high-density EEG recordings.
Better functional outcome was observed in the BCI group including a significantly
higher probability of achieving a clinically relevant (the minimal clinically important
difference of the FMA arm section in subacute stroke survivors is nine to ten
(Arya et al. 2011)) increase in the FMA arms section score (post-therapy FMA arm
section of BCI vs. control group = 44 34.7 vs. 19.8 19.8). Post-BCI training
changes in EEG sensorimotor power spectra (i.e., stronger desynchronization in the
alpha and beta bands) occurred with greater involvement of the ipsilesional hemi-
sphere in response to MI of the paralyzed trained hand. Also, FMA improvements
correlated with the changes (i.e., post-training increase) at rest in ipsilesional
intrahemispheric connectivity in the same bands. These study results demonstrate
impressively the rehabilitative potential of BCI technology in assisting MI practice
and providing feedback of correct brain activation.
While the above study used a realistic visual feedback, the outcome of a
BCI-based therapy may be further improved by providing somatosensory feedback.
The results of a randomized controlled trial with 12 chronic stroke patients, who
received either visual or somatosensory feedback, indicate that provision of somato-
sensory feedback by a motor-driven orthosis is more effective in restoration of upper
extremity function than visual feedback alone (Ono et al. 2014). Later, in a double-
blinded, controlled study, patients with chronic stroke and severe upper limb impair-
ment (mean FMA arm section of all patients = 12.15 8.8) markedly improved as a
result of proprioceptive BCI training (Ramos-Murguialday et al. 2013). Over 20 ses-
sions, patients learned to control a robotic hand exoskeleton for flexing and
extending the fingers of their paretic limb by decreasing the power of the SMR in
the ipsilesional motor cortex. They were instructed to change their brain rhythm by
attempting to move their paralyzed arm, even if no movement was possible. Patients
received additional behavioral physiotherapy to facilitate the generalization of
improvements and received online proprioceptive feedback of brain oscillations;
half of the participants received contingent reward feedback, whereas the other half
received feedback of random brain activity. Marked improvements in motor function
were seen in the patients that received contingent reward feedback but not in the
group that received random feedback (group difference in FMA arm sec-
tion = 3.41 0.56). Furthermore, a consistent pattern of brain reorganization and
connectivity changes was seen in the patients who improved but not in the controls.
The behavioral benefits remained stable during a 6-month follow-up period. This
study showed that use of rehabilitative BCIs can be effective in motor rehabilitation
of chronic stroke patients and demonstrated that cortical and subcortical reorgani-
zation (including functional and structural connectivity) in stroke patients without
residual movement is a consequence of BCI use.
While the number of clinical trials investigating the efficacy of rehabilitative
BCIs in stroke patients is low, the number of studies investigating the effect of
BCI-based rehabilitative upper extremity training in more than one individual with
cervical spinal cord injury is even lower. In a recent pilot study, seven patients
received 20 sessions of BCI-controlled FES, while five patients received the same
number of sessions of passive FES for both hands. The results show that patients in
both groups initially had intense ERD during movement attempt that was not
restricted to the sensorimotor cortex. Following the treatment, ERD cortical activity
restored toward the activity in able-bodied people in the BCI-FES group only,
remaining wide-spread in the FES group. Likewise, SSEPs returned in three patients
24 R. Rupp
in the BCI-FES group, having no changes in the FES group. The range of motion of
the wrist improved in both groups. Muscle strength significantly improved for both
hands in BCI-FES group, while in the FES group, a significant improvement was
noticed for right hand flexor muscles only. Although the results were obtained in a
small number of patients, the study indicates that combined BCI-FES therapy results
in better neurological recovery and better improvement of muscle strength than FES
alone (Osuagwu et al. 2016).
Although the number of studies on efficacy of a BCI-based restorative therapy is
low, the results from the first trials suggest that BCI might lead to a better functional
outcome compared to treatment as usual. From the current literature in stroke motor
rehabilitation, it can be concluded that BCIs should focus on the modulation of SMR
of the affected hemisphere. Feedback should be provided as somatosensory feed-
back, but also congruent visual feedback seems to be effective.
BCIs for Restoration of Walking

While at least the feasibility of MI-BCI-based motor rehabilitation regimes of the
upper extremity has been shown over the last 5 years, the use of rehabilitative BCIs
for restoration of walking is still in its infancy. Most of the investigations so far are
conducting an offline analysis of EEG signals recorded in stroke survivors during
walking and non-walking conditions (He et al. 2014; Do et al. 2013; Sburlea et al.
2015). However, the offline classification results cannot fully transfer to single-trial
conditions, and therefore the performance in real-time BCI control of lower extrem-
ity assistive devices is largely unknown. Results from studies performing single-trial
analysis of EEG data in able-bodied subjects show only moderate true-positive rates
in the detection of the intention of gait initiation (Jiang et al. 2015).
The main open question in using rehabilitative BCIs for control of lower extrem-
ity exoskeletons is in how far the movement intent can be reliably classified. It has
been shown in able-bodied subjects that the classification results are better when
walking is executed than during imagined walking movements (Severens et al.
2015). Additionally, the contribution of sensory feedback on the classification results
is not yet clear. It is highly interesting to know if the classification results are in the
same range in patients with sensory impairments such as patients with SCI or stroke
than in non-impaired subjects.
Thus, more studies involving a larger number of neurological patients are needed
to demonstrate the clinical usefulness of a rehabilitative BCI for restoration of
walking.
Conclusion
BCIs are an evolving technology that contributes to the motor rehabilitation of

people after CNS injuries in mainly two ways: First, they allow for a control of
assistive devices for substitution of a lost motor function mainly in people with high
spinal cord injury, in whom not enough residual movements are preserved for
operation of traditional user interfaces. Second, they can be used in a rehabilitative

setting in subacute patients to support motor recovery.
Currently available BCI systems differ substantially in regard to the degree of
invasiveness, the complexity of hard- and software components, the basic modes of
operation, and the underlying physiological mechanisms. In recent years, noninva-
sive EEG-based BCIs have reached a level of maturity sufficient to be used at end
users’ homes. Systems based on the MI-initiated modulation of SMRs are the
preferred choice for self-paced control of grasp neuroprostheses and exoskeletons
for ambulation. With the recent introduction of hybrid BCIs combining a BCI with
another BCI or with traditional control interfaces such as joysticks or eye trackers,
the group of potential BCI users is expected to increase.
However, even after substantial time of training, not every user might be able to
reliably control an SMR-based BCI, and high-performing users are only able to
generate a few decisions per minute. Users of noninvasive BCIs complain most
about the time-consuming setup of the system in particular of the gel electrodes. This
might be overcome in the future by recently introduced wireless BCI systems based
on dry or semi-dry electrodes (Lee et al. 2013).
Although a huge progress has been made throughout the last decade in noninva-
sive BCIs, mainly digital control signals in form of brain switches can be generated.
This is not sufficient for real-time operation of complex assistive devices such as
robot arms. An intuitive, simultaneous control of the multiple degrees of freedom of
arm and grasp neuroprostheses or multi-joint robot arms in real-time is at the current
state of the art only possible with invasive BCIs based on ECoG recordings or better
with high-density intracortical electrode arrays. While invasive systems provide a
better spatial and – to some parts – temporal resolution and a higher signal-to-noise
ratio, they are associated with surgical risks and have only a limited long-term
stability. Furthermore, daily tuning of the movement decoder algorithms by technical
experts is necessary. These disadvantages need to be clearly communicated to end
users, who at the current state of the art seem to prefer noninvasive solutions (Blabe
et al. 2015).
This may change in the future, because it has been recently shown that intracortical
electrode arrays cannot only be used for the detection of the movement intention but
additionally provide sensory feedback by stimulation of cortical sensory neurons
(Flesher et al. 2016). It is known from prosthetics that providing sensory feedback
to end users will likely enhance the embodiment of the assistive device and the
performance of control. However, also in this application, the general problem of
limited long-term stability of the intracortical implants needs to be solved.
In any case, invasive and noninvasive BCIs are companions rather than compet-
itors, because some end users will not accept a surgery in their unaffected brain and
others are not willing to undergo a substantial period of training for only a few
degrees of control. In conclusion, BCIs are a promising technology for control of
assistive devices in particular for individuals with high cervical SCI. However, more
clinical trials involving a larger number of end users in real need are needed to obtain
objective information about the users’ satisfaction and increase of quality of life. In
invasive BCIs, there is an urgent need to evaluate the long-term risk/benefit ratio.
26 R. Rupp
BCIs for detection of motor intentions offer further possibilities in the context of
rehabilitation of patients with subacute SCI or stroke that go beyond the control of
assistive device. After a CNS injury, substantial reorganization of neural networks
occurs in the brain and spinal cord that plays a critical role for motor recovery. With
BCIs, the neural plasticity of the CNS may be guided in particular in patients with
severe impairments and no preserved motor functions, which do not profit from
traditional motor rehabilitation methods. The key elements for an effective neuroreh-
abilitative training based on motor learning are voluntarily triggered movement
intentions and a synchronized sensory and proprioceptive feedback of the limbs’
motor actions. BCIs hold promise to enable the detection of intended movements
independent from the user’s residual motor functions making them an ideal tool for
closed-loop neurorehabilitative therapies. By practicing feedback-controlled MI of
paralyzed limbs, the integrity of cortical neuronal connections may be preserved,
negative side effects such as neuropathic pain might be prevented (Hassan et al.
2015), and neurological recovery of motor function may be enhanced. The combina-
tion of BCIs with brain activity modulating therapies such as transcranial direct current
stimulation might further enhance the outcome of rehabilitative BCIs (Ang et al.
2015). In any case, more clinical studies involving stroke patients and individuals
with SCI are needed to determine optimal therapy parameters, training doses, and the
possibility for seamless integration of BCI tools into the rehabilitative setting.
In summary, over the relatively short period of two decades, clinical research in
BCIs has provided promising strategies to improve the prospects for patients with
otherwise debilitating neurological disorders that are currently difficult or impossible
to treat.
References
Aflalo T, Kellis S, Klaes C, Lee B, Shi Y, Pejsa K, Shanfield K, Hayes-Jackson S, Aisen M, Heck C,
Liu C, Andersen RA (2015) Neurophysiology. Decoding motor imagery from the posterior
parietal cortex of a tetraplegic human. Science 348(6237):906–910. doi:10.1126/science.
aaa5417
Anderson KD (2004) Targeting recovery: priorities of the spinal cord-injured population.
J Neurotrauma 21(10):1371–1383
Ang KK, Guan C, Chua KS, Ang BT, Kuah C, Wang C, Phua KS, Chin ZY, Zhang H (2010)
Clinical study of neurorehabilitation in stroke using EEG-based motor imagery brain-computer
interface with robotic feedback. Conference proceedings: Annual International Conference of
the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and
Biology Society Conference 2010:5549–5552. doi:10.1109/IEMBS.2010.5626782
Ang KK, Guan C, Phua KS, Wang C, Zhao L, Teo WP, Chen C, Ng YS, Chew E (2015) Facilitating
effects of transcranial direct current stimulation on motor imagery brain-computer interface with
robotic feedback for stroke rehabilitation. Arch Phys Med Rehabil 96(3 Suppl):S79–S87.
doi:10.1016/j.apmr.2014.08.008
Arya KN, Verma R, Garg RK (2011) Estimating the minimal clinically important difference of an upper
extremity recovery measure in subacute stroke patients. Top Stroke Rehabil 18(Suppl 1):599–610.
doi:10.1310/tsr18s01-599
Blabe CH, Gilja V, Chestek CA, Shenoy KV, Anderson KD, Henderson JM (2015) Assessment of
brain-machine interfaces from the perspective of people with paralysis. J Neural Eng 12
(4):043002. doi:10.1088/1741-2560/12/4/043002
Bouton CE, Shaikhouni A, Annetta NV, Bockbrader MA, Friedenberg DA, Nielson DM, Sharma G,
Sederberg PB, Glenn BC, Mysiw WJ, Morgan AG, Deogaonkar M, Rezai AR (2016) Restoring
cortical control of functional movement in a human with quadriplegia. Nature 533
(7602):247–250. doi:10.1038/nature17435
Broetz D, Braun C, Weber C, Soekadar SR, Caria A, Birbaumer N (2010) Combination of brain-
computer interface training and goal-directed physical therapy in chronic stroke: a case report.
Neurorehabil Neural Repair 24(7):674–679. doi:10.1177/1545968310368683
Buch E, Weber C, Cohen LG, Braun C, Dimyan MA, Ard T, Mellinger J, Caria A, Soekadar S,
Fourkas A, Birbaumer N (2008) Think to move: a neuromagnetic brain-computer interface (BCI)
system for chronic stroke. Stroke 39(3):910–917. doi:10.1161/STROKEAHA.107.505313
Caria A, Weber C, Brotz D, Ramos A, Ticini LF, Gharabaghi A, Braun C, Birbaumer N (2011)
Chronic stroke recovery after combined BCI training and physiotherapy: a case report. Psycho-
physiology 48(4):578–582. doi:10.1111/j.1469-8986.2010.01117.xPSYP1117
Chaudhary U, Birbaumer N, Ramos-Murguialday A (2016) Brain-computer interfaces for commu-
nication and rehabilitation. Nat Rev Neurol 12(9):513–525. doi:10.1038/nrneurol.2016.113
Chavarriaga R, Sobolewski A, Millan Jdel R (2014) Errare machinale est: the use of error-related
potentials in brain-machine interfaces. Front Neurosci 8:208. doi:10.3389/fnins.2014.00208
Cincotti F, Mattia D, Aloise F, Bufalari S, Schalk G, Oriolo G, Cherubini A, Marciani MG, Babiloni
F (2008) Non-invasive brain-computer interface system: towards its application as assistive
technology. Brain Res Bull 75(6):796–803. doi:10.1016/j.brainresbull.2008.01.007
Collinger JL, Wodlinger B, Downey JE, Wang W, Tyler-Kabara EC, Weber DJ, McMorland AJ,
Velliste M, Boninger ML, Schwartz AB (2013) High-performance neuroprosthetic control by an
individual with tetraplegia. Lancet 381(9866):557–564. doi:10.1016/S0140-6736(12)61816-9
Contreras-Vidal JL, Grossman RG (2013) NeuroRex: a clinical neural interface roadmap for
EEG-based brain machine interfaces to a lower body robotic exoskeleton. Conference pro-
ceedings: Annual International Conference of the IEEE Engineering in Medicine and Biology
Society IEEE Engineering in Medicine and Biology Society Conference 2013:1579–1582.
doi:10.1109/EMBC.2013.6609816
Contreras-Vidal J, Presacco A, Agashe H, Paek A (2012) Restoration of whole body movement:
toward a noninvasive brain-machine interface system. IEEE Pulse 3(1):34–37. doi:10.1109/
MPUL.2011.2175635
Contreras-Vidal JL, AB N, Brantley J, Cruz-Garza JG, He Y, Manley Q, Nakagome S, Nathan K,
Tan SH, Zhu F, Pons JL (2016) Powered exoskeletons for bipedal locomotion after spinal cord
injury. J Neural Eng 13(3):031001. doi:10.1088/1741-2560/13/3/031001
Curt A, Schwab ME, Dietz V (2004) Providing the clinical basis for new interventional therapies:
refined diagnosis and assessment of recovery after spinal cord injury. Spinal Cord 42(1):1–6.
doi:10.1038/sj.sc.3101558
Daly JJ, Wolpaw JR (2008) Brain-computer interfaces in neurological rehabilitation. Lancet Neurol
7(11):1032–1043 . doi:10.1016/S1474-4422(08)70223-0S1474-4422(08)70223-0 [pii]
Daly JJ, Cheng R, Rogers J, Litinas K, Hrovat K, Dohring M (2009) Feasibility of a new application of
noninvasive brain computer interface (BCI): a case study of training for recovery of volitional motor
control after stroke. J Neurol Phys Ther 33(4):203–211. doi:10.1097/NPT.0b013e3181c1fc0b.
01253086-200912000-00007 [pii]
Dietz V, Fouad K (2014) Restoration of sensorimotor functions after spinal cord injury. Brain J
Neurol 137(Pt 3):654–667. doi:10.1093/brain/awt262
Dimyan MA, Cohen LG (2011) Neuroplasticity in the context of motor rehabilitation after stroke.
Nat Rev Neurol 7(2):76–85. doi:10.1038/nrneurol.2010.200
Do AH, Wang PT, King CE, Chun SN, Nenadic Z (2013) Brain-computer interface controlled
robotic gait orthosis. J Neuroeng Rehabil 10:111. doi:10.1186/1743-0003-10-111
28 R. Rupp
Downey JE, Weiss JM, Muelling K, Venkatraman A, Valois JS, Hebert M, Bagnell JA, Schwartz
AB, Collinger JL (2016) Blending of brain-machine interface and vision-guided autonomous
robotics improves neuroprosthetic arm performance during grasping. J Neuroeng Rehabil 13:28.
doi:10.1186/s12984-016-0134-9
Enzinger C, Ropele S, Fazekas F, Loitfelder M, Gorani F, Seifert T, Reiter G, Neuper C,
Pfurtscheller G, Müller-Putz GR (2008) Brain motor system function in a patient with complete
spinal cord injury following extensive brain-computer interface training. Exp Brain Res 190
(2):215–223. doi:10.1007/s00221-008-1465-y
Ethier C, Gallego JA, Miller LE (2015) Brain-controlled neuromuscular stimulation to drive neural
plasticity and functional recovery. Curr Opin Neurobiol 33:95–102. doi:10.1016/j.conb.2015.03.007
Exner G (2004) The working group “paraplegia” of the federation of commercial professional
associations in Germany. Facts, figures and prognoses. Trauma Berufskrankh 6:147–151
Feigin VL, Lawes CM, Bennett DA, Barker-Collo SL, Parag V (2009) Worldwide stroke incidence
and early case fatality reported in 56 population-based studies: a systematic review. Lancet
Neurol 8(4):355–369. doi:10.1016/S1474-4422(09)70025-0
Flesher SN, Collinger JL, Foldes ST, Weiss JM, Downey JE, Tyler-Kabara EC, Bensmaia SJ,
Schwartz AB, Boninger ML, Gaunt RA (2016) Intracortical microstimulation of human
somatosensory cortex. Sci Transl Med 8(361):361ra141. doi:10.1126/scitranslmed.aaf8083
Gunasekera B, Saxena T, Bellamkonda R, Karumbaiah L (2015) Intracortical recording interfaces:
current challenges to chronic recording function. ACS Chem Neurosci 6(1):68–83. doi:10.1021/
cn5002864
Hassan MA, Fraser M, Conway BA, Allan DB, Vuckovic A (2015) The mechanism of
neurofeedback training for treatment of central neuropathic pain in paraplegia: a pilot study.
BMC Neurol 15:200. doi:10.1186/s12883-015-0445-7
He Y, Nathan K, Venkatakrishnan A, Rovekamp R, Beck C, Ozdemir R, Francisco GE, Contreras-
Vidal JL (2014) An integrated neuro-robotic interface for stroke rehabilitation using the NASA
X1 powered lower limb exoskeleton. Conference proceedings: Annual International Conference
of the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and
Biology Society Conference 2014:3985–3988. doi:10.1109/EMBC.2014.6944497
Hendricks HT, van Limbeek J, Geurts AC, Zwarts MJ (2002) Motor recovery after stroke: a
systematic review of the literature. Arch Phys Med Rehabil 83(11):1629–1637
Hiremath SV, Chen W, Wang W, Foldes S, Yang Y, Tyler-Kabara EC, Collinger JL, Boninger ML
(2015) Brain computer interface learning for systems based on electrocorticography and
intracortical microelectrode arrays. Front Integr Neurosci 9:40. doi:10.3389/fnint.2015.00040
Hochberg LR, Bacher D, Jarosiewicz B, Masse NY, Simeral JD, Vogel J, Haddadin S, Liu J, Cash
SS, van der Smagt P, Donoghue JP (2012) Reach and grasp by people with tetraplegia using a
neurally controlled robotic arm. Nature 485(7398):372–375. doi:10.1038/nature11076
ISO (2010) ISO 9241:2010 Ergonomics of human-system interaction Part 210: Human-centred
design for interactive systems, Geneva
Jackson A, Zimmermann JB (2012) Neural interfaces for the brain and spinal cord – restoring motor
function. Nat Rev Neurol 8(12):690–699. doi:10.1038/nrneurol.2012.219
Jiang N, Gizzi L, Mrachacz-Kersting N, Dremstrup K, Farina D (2015) A brain-computer interface
for single-trial detection of gait initiation from movement related cortical potentials. Clin
Neurophysiol 126(1):154–159. doi:10.1016/j.clinph.2014.05.003
Kaiser V, Bauernfeind G, Kreilinger A, Kaufmann T, Kübler A, Neuper C, Müller-Putz GR (2014)
Cortical effects of user training in a motor imagery based brain-computer interface measured by
fNIRS and EEG. NeuroImage 85(Pt 1):432–444. doi:10.1016/j.neuroimage.2013.04.097
Kim YJ, Park SW, Yeom HG, Bang MS, Kim JS, Chung CK, Kim S (2015) A study on a robot arm
driven by three-dimensional trajectories predicted from non-invasive neural signals. Biomed
Eng Online 14:81. doi:10.1186/s12938-015-0075-8
King CE, Wang PT, McCrimmon CM, Chou CC, Do AH, Nenadic Z (2015) The feasibility of a
brain-computer interface functional electrical stimulation system for the restoration of over-
ground walking after paraplegia. J Neuroeng Rehabil 12:80. doi:10.1186/s12984-015-0068-7
Kleih SC, Kaufmann T, Zickler C, Halder S, Leotta F, Cincotti F, Aloise F, Riccio A, Herbert C,
Mattia D, Kübler A (2011) Out of the frying pan into the fire – the P300-based BCI faces real-
world challenges. Prog Brain Res 194:27–46. doi:10.1016/B978-0-444-53815-4.00019-4
Kreilinger A, Kaiser V, Breitwieser C, Williamson J, Neuper C, Müller-Putz GR (2011) Switching
between manual control and brain-computer interface using long term and short term quality
measures. Front Neurosci 5:147. doi:10.3389/fnins.2011.00147
Kübler A, Kotchoubey B, Hinterberger T, Ghanayim N, Perelmouter J, Schauer M, Fritsch C,
Taub E, Birbaumer N (1999) The thought translation device: a neurophysiological approach to
communication in total motor paralysis. Exp Brain Res 124(2):223–232
Langhorne P, Coupar F, Pollock A (2009) Motor recovery after stroke: a systematic review. Lancet
Neurol 8(8):741–754. doi:10.1016/S1474-4422(09)70150-4
Langhorne P, Bernhardt J, Kwakkel G (2011) Stroke rehabilitation. Lancet 377(9778):1693–1702.
doi:10.1016/S0140-6736(11)60325-5
Lee S, Shin S, Woo S, Kim K, Lee H-N (2013) Review of wireless brain-computer interface
systems. In: Fazel-Rezai R (ed) Brain-computer interface systems – recent progress and future
prospects. doi:10.5772/56436
Mark VW, Taub E (2004) Constraint-induced movement therapy for chronic stroke hemiparesis and
other disabilities. Restor Neurol Neurosci 22(3–5):317–336
McCaughey EJ, Purcell M, McLean AN, Fraser MH, Bewick A, Borotkanics RJ, Allan DB (2016)
Changing demographics of spinal cord injury over a 20-year period: a longitudinal population-
based study in Scotland. Spinal Cord 54(4):270–276. doi:10.1038/sc.2015.167
McFarland DJ, Sarnacki WA, Wolpaw JR (2010) Electroencephalographic (EEG) control of three-
dimensional movement. J Neural Eng 7(3):036007. doi:10.1088/1741-2560/7/3/036007
Müller-Putz GR, Scherer R, Pfurtscheller G, Rupp R (2005) EEG-based neuroprosthesis control: a
step towards clinical practice. Neurosci Lett 382(1–2):169–174. doi:10.1016/j.neulet.2005.03.021
Müller-Putz GR, Breitwieser C, Cincotti F, Leeb R, Schreuder M, Leotta F, Tavella M, Bianchi L,
Kreilinger A, Ramsay A, Rohm M, Sagebaum M, Tonin L, Neuper C, Millan JR (2011) Tools
for brain-computer interaction: a general concept for a hybrid BCI. Front Neuroinform 5:30.
doi:10.3389/fninf.2011.00030
Müller-Putz G, Leeb R, Tangermann M, Hohne J, Kubler A, Cincotti F, Mattia D, Rupp R, Müller KR,
Millan JD (2015) Towards noninvasive hybrid brain-computer interfaces: framework, practice,
clinical application, and beyond. P Ieee 103(6):926–943. doi:10.1109/Jproc.2015.2411333
Muralidharan A, Chae J, Taylor DM (2011) Extracting attempted hand movements from EEGs in
people with complete hand paralysis following stroke. Front Neurosci 5:39. doi:10.3389/
fnins.2011.00039
Nardone R, Holler Y, Brigo F, Seidl M, Christova M, Bergmann J, Golaszewski S, Trinka E (2013)
Functional brain reorganization after spinal cord injury: systematic review of animal and human
studies. Brain Res 1504:58–73. doi:10.1016/j.brainres.2012.12.034
Neuper C, Scherer R, Reiner M, Pfurtscheller G (2005) Imagery of motor actions: differential
effects of kinesthetic and visual-motor mode of imagery in single-trial EEG. Brain Res Cogn
Brain Res 25(3):668–677. doi:10.1016/j.cogbrainres.2005.08.014
Nilsen DM, Gillen G, Gordon AM (2010) Use of mental practice to improve upper-limb recovery
after stroke: a systematic review. Am J Occup Ther 64(5):695–708
NSCISC (2015) The 2015 annual statistical report for the model spinal cord injury care systems.
National SCI Statistical Center. Available online: https://www.nscisc.uab.edu/PublicDocuments/
reports/pdf/2015%20NSCISC%20Annual%20Statistical%20Report%20Complete%20Public%
20Version.pdf, Last date of access: 10th of November 2016
Ono T, Shindo K, Kawashima K, Ota N, Ito M, Ota T, Mukaino M, Fujiwara T, Kimura A, Liu M,
Ushiba J (2014) Brain-computer interface with somatosensory feedback improves functional
recovery from severe hemiplegia due to chronic stroke. Front Neuroeng 7:19. doi:10.3389/
fneng.2014.00019
30 R. Rupp
Osuagwu BC, Wallace L, Fraser M, Vuckovic A (2016) Rehabilitation of hand in subacute

tetraplegic patients based on brain computer interface and functional electrical stimulation: a
randomised pilot study. J Neural Eng 13(6):065002. doi:10.1088/1741-2560/13/6/065002
Ouzký M (2002) Towards concerted efforts for treating and curing spinal cord injury, Report of the
Council of Europe Doc. No. 9401. https://assembly.coe.int/nw/xml/XRef/X2H-Xref-
ViewHTML.asp?FileID=9680&lang=en. Accessed Doc.9401
Pfurtscheller G, Lopes da Silva FH (1999) Event-related EEG/MEG synchronization and
desynchronization: basic principles. Clin Neurophysiol: Off J Int Fed Clin Neurophys 110
(11):1842–1857
Pfurtscheller G, Müller GR, Pfurtscheller J, Gerner HJ, Rupp R (2003) ‘Thought’ – control of
functional electrical stimulation to restore hand grasp in a patient with tetraplegia. Neurosci Lett
351(1):33–36
Pichiorri F, Morone G, Petti M, Toppi J, Pisotta I, Molinari M, Paolucci S, Inghilleri M, Astolfi L,
Cincotti F, Mattia D (2015) Brain-computer interface boosts motor imagery practice during
stroke recovery. Ann Neurol 77(5):851–865. doi:10.1002/ana.24390
Pineda JA, Silverman DS, Vankov A, Hestenes J (2003) Learning to control brain rhythms: making
a brain-computer interface possible. IEEE Trans Neural Syst Rehab Eng: Publ IEEE Eng Med
Biol Soc 11(2):181–184. doi:10.1109/TNSRE.2003.814445
Prasad G, Herman P, Coyle D, McDonough S, Crosbie J (2010) Applying a brain-computer
interface to support motor imagery practice in people with stroke for upper limb recovery: a
feasibility study. J Neuroeng Rehabil 7:60. doi:10.1186/1743-0003-7-60
Prochazka A (2015) Technology to enhance arm and hand function. In: Dietz V, Ward N (eds)
Oxford textbook of neurorehabilitation. Oxford University Press, Oxford. doi:10.1093/med/
9780199673711.003.0031_update_001
Ramos-Murguialday A, Broetz D, Rea M, Laer L, Yilmaz O, Brasil FL, Liberati G, Curado MR,
Garcia-Cossio E, Vyziotis A, Cho W, Agostini M, Soares E, Soekadar S, Caria A, Cohen LG,
Birbaumer N (2013) Brain-machine interface in chronic stroke rehabilitation: a controlled study.
Ann Neurol 74(1):100–108. doi:10.1002/ana.23879
Rohm M, Schneiders M, Müller C, Kreilinger A, Kaiser V, Müller-Putz GR, Rupp R (2013) Hybrid
brain-computer interfaces and hybrid neuroprostheses for restoration of upper limb functions in
individuals with high-level spinal cord injury. Artif Intell Med 59(2):133–142. doi:10.1016/j.
artmed.2013.07.004
Rupp R (2014) Challenges in clinical applications of brain computer interfaces in individuals with
spinal cord injury. Front Neuroeng 7:38. doi:10.3389/fneng.2014.00038
Rupp R, Rohm M, Schneiders M, Weidner N, Kaiser V, Kreilinger A, Müller-Putz GR (2013)
Think2grasp – BCI-controlled neuroprosthesis for the upper extremity. Biomed Tech.
doi:10.1515/bmt-2013-4440
Rupp R, Rohm M, Schneiders M, Kreilinger A, Müller-Putz GR (2015a) Functional rehabilitation
of the paralyzed upper extremity after spinal cord injury by Noninvasive Hybrid
Neuroprostheses. P Ieee 103(6):954–968. doi:10.1109/Jproc.2015.2395253
Rupp R, Rohm M, Schneiders M (2014) Brain-computer interfaces for control of upper extremity
neuroprostheses in individuals with high spinal cord injury, In: Naik G (ed) Emerging Theory
and Practice in Neuroprosthetics, IGIGlobal, Hershey, 237–264
Rupp R, Schließmann D, Schuld C, Weidner N (2015b) Technology to enhance locomotor function.
In: Dietz V, Ward N (eds) Oxford textbook of neurorehabilitation. Oxford University Press,
Oxford. doi:10.1093/med/9780199673711.003.0032_update_001
Savic AM, Malesevic NM, Popovic MB (2014) Feasibility of a hybrid brain-computer interface for
advanced functional electrical therapy. TheScientificWorldJOURNAL 2014:797128.
doi:10.1155/2014/797128
Sburlea AI, Montesano L, Cano de la Cuerda R, Alguacil Diego IM, Miangolarra-Page JC,
Minguez J (2015) Detecting intention to walk in stroke patients from pre-movement EEG
correlates. J Neuroeng Rehabil 12:113. doi:10.1186/s12984-015-0087-4
Severens M, Perusquia-Hernandez M, Nienhuis B, Farquhar J, Duysens J (2015) Using actual and

imagined walking related desynchronization features in a BCI. IEEE Trans Neural Syst Rehabil
Eng: Publ IEEE Eng Med Biol Soc 23(5):877–886. doi:10.1109/TNSRE.2014.2371391
Shih JJ, Krusienski DJ, Wolpaw JR (2012) Brain-computer interfaces in medicine. Mayo Clin Proc
87(3):268–279. doi:10.1016/j.mayocp.2011.12.008
Snoek GJ, IJ MJ, Hermens HJ, Maxwell D, Biering-Sorensen F (2004) Survey of the needs of
patients with spinal cord injury: impact and priority for improvement in hand function in
tetraplegics. Spinal Cord 42(9):526–532. doi:10.1038/sj.sc.31016383101638
Soekadar SR, Birbaumer N, Slutzky MW, Cohen LG (2015) Brain-machine interfaces in neuroreh-
abilitation of stroke. Neurobiol Dis 83:172–179. doi:10.1016/j.nbd.2014.11.025
Toppi J, Risetti M, Quitadamo LR, Petti M, Bianchi L, Salinari S, Babiloni F, Cincotti F, Mattia D,
Astolfi L (2014) Investigating the effects of a sensorimotor rhythm-based BCI training on the
cortical activity elicited by mental imagery. J Neural Eng 11(3):035010. doi:10.1088/1741-
2560/11/3/035010
Truelsen T, Prescott E, Gronbaek M, Schnohr P, Boysen G (1997) Trends in stroke incidence. The
Copenhagen City Heart Study. Stroke 28(10):1903–1907
van den Berg ME, Castellote JM, Mahillo-Fernandez I, de Pedro-Cuesta J (2010) Incidence of
spinal cord injury worldwide: a systematic review. Neuroepidemiology 34(3):184–192 .
doi:10.1159/000279335discussion 192
van Dokkum LE, Ward T, Laffont I (2015) Brain computer interfaces for neurorehabilitation – its
current status as a rehabilitation strategy post-stroke. Ann Phys Rehabil Med 58(1):3–8.
doi:10.1016/j.rehab.2014.09.016
Wang W, Collinger JL, Degenhart AD, Tyler-Kabara EC, Schwartz AB, Moran DW, Weber DJ,
Wodlinger B, Vinjamuri RK, Ashmore RC, Kelly JW, Boninger ML (2013) An electro-
corticographic brain interface in an individual with tetraplegia. PLoS One 8(2):e55344.
Ward NS, Cohen LG (2004) Mechanisms underlying recovery of motor function after stroke. Arch
Neurol 61(12):1844–1848. doi:10.1001/archneur.61.12.1844
Warlow C, van Gijn J, Dennis M, Wardlaw J, Bamford J, Sandercock P, Rinkel G, Langhorne P,
Sudlow C, Rothwell P (2008) Stroke: practical management, 3rd edn. Blackwell Publishing,
Oxford
Wodlinger B, Downey JE, Tyler-Kabara EC, Schwartz AB, Boninger ML, Collinger JL (2015)
Ten-dimensional anthropomorphic arm control in a human brain-machine interface: difficulties,
solutions, and limitations. J Neural Eng 12(1):016011. doi:10.1088/1741-2560/12/1/016011
Wolpaw JR, McFarland DJ (1994) Multichannel EEG-based brain-computer communication.
Electroencephalogr Clin Neurophysiol 90(6):444–449
Wolpaw JR, Birbaumer N, McFarland DJ, Pfurtscheller G, Vaughan TM (2002) Brain-computer
interfaces for communication and control. Clin Neurophysiol 113(6):767–791. doi:
S1388245702000573 [pii]
Impact of Scoliosis on Gait
Elizabeth A. Rapp and Peter G. Gabos
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Structural Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Spatiotemporal Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Step Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Time in Stance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Cadence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Body Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Arms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Spine and Trunk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Pelvis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Distal Kinematics: Hip, Knee, and Ankle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Lower Extremity Asymmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Vertical Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Horizontal Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Moments and Powers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Balance and Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Muscle Activation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Relationship of Gait Parameters to Curve Severity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Response to Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Response to Bracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Response to Surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Summary and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
E.A. Rapp (*)

University of Delaware, Newark, DE, USA
e-mail: lizrapp@udel.edu
P.G. Gabos (*)
Nemours/A.I. DuPont Hospital for Children, Wilmington, DE, USA
e-mail: peter.gabos@nemours.org

DOI 10.1007/978-3-319-30808-1_68-1
2 E.A. Rapp and P.G. Gabos
Abstract
Scoliosis is one of the most common orthopedic disorders in children and
adolescents. The idiopathic classification has been studied at length in hopes of
identifying factors contributing to the origin and progression of the disease. Gait
analysis is frequently employed to analyze the balance and movement abnormal-
ities associated with the disorder. While the majority of gait studies in scoliosis
note some deviations from normal gait, specific conclusions are often based on
weak or inconsistent evidence. The most widely reported findings include
restricted motion of the pelvis and hip and an asymmetrical rotation of the
trunk as well as general asymmetry between limbs. Additionally, energy cost
and muscle activation are higher during scoliotic gait than in normal walking.
These differences seem to improve with both orthotic and surgical treatment,
although postoperative adolescents with idiopathic scoliosis still maintain a
higher-energy cost of walking than their typically developing peers. Ultimately,
the relationship between gait abnormalities and the origin or progression of the
scoliotic curve remains unclear. The idea of a neurological dysfunction that
contributes to both the spinal deformity and the gait deviation is predominantly
rooted in theory. Still, future research into motor control and somatosensory
function during gait may provide more insight into a neurological influence in
the scoliosis population.
Keywords
Gait • Idiopathic scoliosis • Spinal deformity • Scoliotic gait • Kinematics •
Asymmetry • Kinetics • Dynamic balance • Proprioception • Muscle activity •
Curve severity • Bracing • Spinal fusion • Energy cost
Introduction
Scoliosis is a widespread orthopedic condition and is the most prevalent type of

spinal deformity in children and adolescents (Konieczny et al. 2013). The large
majority of cases are idiopathic (scoliosis for which there is no obvious attributed
cause), and the remainder are non-idiopathic, including those due to congenital
abnormalities and those secondary to a neuromuscular disorder. Many of the under-
lying diseases that lead to non-idiopathic scoliosis can affect the musculoskeletal
system and impair all aspects of mobility, including gait. Due to the confounding
influences of these diseases, this chapter will focus only on the most prevalent
classification (idiopathic scoliosis) in order to discuss the isolated contribution of
the scoliotic deformity to gait patterns. This chapter will describe how scoliosis
induces structural changes throughout the body, present evidence of gait patterns in
patients with scoliosis, and discuss theories of how gait abnormalities relate to the
pathoetiology of scoliosis.
Impact of Scoliosis on Gait 3
State of the Art
Despite the prevalence of idiopathic scoliosis, the factors contributing to the devel-
opment of the disease are poorly understood. The current body of evidence offers
little insight as to factors that contribute to the origin or progression of the deformity.
The impact of scoliosis on gait parameters has been studied at length in an attempt to
identify structural or neurological trends. The majority of studies note some devia-
tions from normal gait, but weak or inconsistent results limit the substantiation of
distinct conclusions. Consequently, these gait findings have limited application for
diagnosis or treatment. Results from several studies regarding balance and postural
control hint toward a potential neuromuscular influence. Future research will likely
explore this direction, with the goal of identifying neurological factors in scoliosis
that contribute to both the structural deformity and deviations from normal gait.
Background
Literature suggests the incidence of adolescent idiopathic scoliosis to be between

0.47 % and 5.2 % of children between the ages of 10 and 18. Idiopathic scoliosis is
more common in females than males, with this gender bias increasing substantially
in more severe curves (Konieczny et al. 2013).
Curvature is classified as thoracic, thoracolumbar, lumbar, or double major
(Fig. 1), depending on the vertebral location of the apex of the curve.
Fig. 1 Curvature in scoliosis: (a) thoracic, (b) thoracolumbar, (c) lumbar, or (d) double major
(Reproduced from Grivas et al. 2010 with permissions granted under the terms of the Creative
Commons Attribution License: http://creativecommons.org/licenses/by/2.0)
Thoracic curves are most common and are typically convex right, while lumbar
curves are typically convex left. Thoracolumbar curves are also common, and their
location and convexity can vary. Double or “S” curves can be the result of actual
structural deformity or develop as compensatory spinal balance.
Adolescents with idiopathic scoliosis complain of struggles with appearance,
self-esteem, and in some cases, pain and functional deficits (Mayo et al. 1994).
Curve progression is common without treatment: between 20 % and 40 % more
likely with no treatment as opposed to even a conservative intervention such as
bracing (Ascani et al. 1986). Adolescents with curves greater than 50 are at risk for
curve progression throughout adult life. Curves with this degree of severity are
typically recommended for surgical correction (Asher and Burton 2006).
Patients do not commonly complain of difficulty walking; however, there is a
clinically observed stiffness and asymmetry seen in scoliotic gait (Mahaudens
et al. 2009). The majority of studies investigating gait in scoliosis have noted
some abnormality in gait parameters, and thus impaired gait control appears to be
moderately associated with idiopathic scoliosis (Schlösser et al. 2014). Nevertheless,
there are conflicting findings with respect to some kinematic deviations, as well as
opposing theories on the mechanism of these abnormalities. It is widely debated
whether the observed gait pathology is a manifestation of the underlying deformity
or whether it may be responsible for the causation or progression of the disease.
Structural Differences
The three-dimensional spinal deformity in scoliosis creates a trunk distortion that

alters the lateral position of the center of mass within the body (Fig. 2).
A laterally displaced center of mass can in turn influence the direction of ground
reaction forces, requiring atypical and asymmetric joint moments to maintain sta-
bility. Additionally, a deformity in the lumbar spine can alter pelvis orientation,
inducing subsequent structural and alignment abnormalities that extend through the
lower extremities. Scoliosis is associated with a multiplanar rotation deformity of the
pelvis. In the transverse plane, the pelvis rotates to the right (clockwise). In anterior/
posterior radiographs, this results in the impression of the right ilium being wider
than the left, i.e., a left to right ilium width ratio of less than one (Gum et al. 2007;
Fig. 3).
This pelvic torsion is the same rotational direction as a typical scoliotic thoracic
curve and the opposite rotational direction of a typical lumbar curve. Furthermore,
radiographic evidence also demonstrates increased pelvic and sacral obliquity with
the segments commonly tilting toward the apex of the lumbar curve (Mahaudens
et al. 2005; Schwender and Denis 2000). Both of these findings suggest that the
pelvic orientation deformities are a compensatory response to the scoliotic curvature.
Compensatory spinal curves are common in the coronal plane, and the radiographic
evidence above indicates this response is neither limited to the spine nor to any one
plane. All of the compensations are thought to be an attempt to retain balance.
Fig. 2 Drawing of right

thoracic scoliosis. Center of
mass shift can be noted by the
gravitational vector skewed
toward the right pelvis
(Courtesy of University of
Delaware Biomechanics and
Movement Science
Laboratories)
The deformity can extend all the way to the femur, with some patients showing
increased femoral neck-shaft angles on both sides (Saji et al. 1995). While this result
could be an extension of the postural stability compensations at the spine and pelvis,
the bilateral occurrence of the effect leads some to suggest a neurological influence.
Regardless of origin, these structural deformities can hinder the mechanism of
walking. In normal gait, the trunk, pelvis, and limbs move in synchronized patterns
to generate propulsion and maintain stability (Perry et al. 2010). Irregular segment
rotation or alignment can disrupt coordination during gait, inducing compensatory
changes in walking patterns and imposing excessive demands on muscles. Still,
evidence and theory are mixed as to whether these structural changes are the cause of
gait abnormalities or merely a side effect of a greater governing disorder.
Spatiotemporal Parameters
Spatiotemporal parameters can provide a global measure of gait performance. In

scoliotic gait, these measures are similar to those in normal walking, with a few
marginal differences. In general, adolescents with idiopathic scoliosis do not show
any deficits in walking speed. Still, while their overall speed is no different than their
healthy peers, patients do exhibit differences in some of the components controlling
the speed of ambulation (Schizas et al. 1998; Yang et al. 2013; Prince et al. 2010).
Fig. 3 Standing position

posterior-anterior radiograph
of a patient with a right
thoracic scoliosis illustrating
the location of the hemi-pelvis
landmarks, inferior ilium at
the sacroiliac joint, and the
anterior superior iliac spine
and the lines necessary to
measure the left and right
hemi-pelvis coronal plane
widths (Reproduced from
Gum et al. 2007 with
permission of Springer). Note
the left/right (L/R) ilium width
ratio of less than one,
indicating a right-rotated
pelvis
Step Length
Irrespective of curve type, adolescents with idiopathic scoliosis have been shown to
walk with a reduced step length (Kramers-de Quervain et al. 2004; Mahaudens
et al. 2009; Syczewska et al. 2012; Mallau et al. 2007). This difference is generally
small, but significant. Furthermore, while step length in normal gait is symmetrical
across both sides, patients with scoliosis demonstrate bilateral differences: usually,
the right step length is slightly shorter than the left (Yang et al. 2013). Recalling the
prevalence of the right-rotated thoracic curve, right-rotated pelvis, and left-rotated
lumbar curve, one might theorize that step length is related to the structural defor-
mity. Surprisingly, there is no evidence of association between those observations –
step length abnormalities occur in both left and right curves of all degrees of severity.
Thus, any relationship remains theoretical.
Time in Stance
Similar to step length, time in stance is marginally reduced in scoliosis (Mahaudens

et al. 2009). The main difference from normal walking is an asymmetry between
sides, with longer time in stance phase on the right side, in line with the presentation
of longer step length on the left (Yang et al. 2013).
Cadence
In most cases, adolescents with idiopathic scoliosis walk with similar cadence to
typically developing adolescents (Mahaudens et al. 2009; Yang et al. 2013). Patients
with thoracolumbar curves, however, have been shown to walk with a slower
cadence, theorized to be an attempt to increase postural stability (Chen
et al. 1998). Still, differences in cadence and step length do not produce a clinically
observable effect in patients, as evidenced by normal performance in walking speed.
Body Kinematics
Kinematic analysis of gait in scoliosis can offer insight as to whether structural

abnormalities translate to pathological motion. Beginning with the arms, the follow-
ing section examines the motion of the limbs, trunk, and pelvis during the gait cycle
and discusses how the scoliotic deformity might contribute to the observed patterns.
Arms
A primary observation in scoliotic gait is an asymmetrical arm swing: patients walk

with greater flexion and extension of both the shoulder and elbow on one side
compared to the other (Kramers-de Quervain et al. 2004). While one might reason-
ably assume that the skewed arm motion occurs as a consequence of an irregularly
shaped torso, this phenomenon is not seen on the same side across patients and does
not appear to be related to the side or severity of the curve deformity (Kramers-de
Quervain et al. 2004). In addition to arm swing asymmetry, frontal plane arm motion
is slightly reduced in scoliosis (Mahaudens et al. 2009). However, while the unequal
arm swing pattern is clinically detectable, the frontal plane differences are minimal.
Motion is reduced by only a few degrees compared to healthy adolescents, which
would not likely be noticeable to the naked eye.
Spine and Trunk
Along with arm asymmetry, irregular motion of the trunk in the transverse plane is
also a frequent observation during scoliotic walking. Adolescents with idiopathic
scoliosis – particularly those with a thoracic curve – show asymmetric rotational
progression of the trunk during the gait cycle. In normal gait, both shoulders
alternately advance throughout the cycle, accompanied by a corresponding rotation
of the trunk. In scoliosis, particularly patients with a right convex thoracic curve, the
Fig. 4 Relative motion of the 40° Trunk Versus Pelvis

trunk versus the pelvis,
reflecting torsional deformity 20°
(Adapted from Kramers-de
Quervain et al. 2004 with
0°
permission of Springer)
-20°
-40°
0 20 40 60 80 100% Cycle
right shoulder remains more advanced relative to the line of progression. The
asymmetrical trunk rotation can be seen as a torsion of the trunk relative to the
pelvis or head. Throughout the gait cycle, patients maintain head orientation in the
line of progression and rotate the pelvis side to side symmetrically, as seen in normal
walking. Consequently, the right shoulder biasing toward the midline produces a
right-forward rotation of the trunk throughout the gait cycle (Kramers-de Quervain
et al. 2004; Fig. 4).
One study involving patients with thoracolumbar and lumbar curves did not
observe this asymmetrical trunk and pelvis pattern, and these authors suggested
the torsion may only be a consequence of a thoracic deformity (Mahaudens
et al. 2009). However, in general, the coordination of trunk and pelvis motion
appears to be atypical and also more variable in scoliosis than observed in normal
walking (Park et al. 2015).
Pelvis
Motion of the pelvis during gait is reduced in scoliosis (Chen et al. 1998;
Mahaudens et al. 2009; Park et al. 2015). Clinically, this presents as an
observed “stiffness” during gait, and biomechanical analyses show that pelvic
range of motion is restricted in all three planes. It is generally believed that
these altered mechanics derive from the structural abnormalities of the pelvis
and further theorized that the restricted range of motion is a direct consequence
of the spinal deformity (Mahaudens et al. 2009; Kramers-de Quervain
et al. 2004).
Distal Kinematics: Hip, Knee, and Ankle
Hip motion is also slightly reduced in all three planes (Mahaudens et al. 2009);
however, it is unknown whether this detail is related to restricted pelvic motion or the
aforementioned femoral neck-shaft deformities. It is more commonly speculated that
restrictions in hip motion may be an attempt at postural control and part of an overall
strategy to maintain stability by limiting lower extremity range of motion in walking.
The kinematics of more distal segments are more variable in scoliosis, and there is
not a strong consensus on whether knee and ankle motion deviate from normal
patterns. There is some evidence of reduced knee motion within the sagittal plane;
however, the mechanism and rationale for this effect are unclear. This phenomenon
may be an extended consequence of the structural pelvic or femoral deformity, or it
may be a stability strategy: an attempt to restrict range of motion. Nevertheless,
reduced knee motion is not a consistent finding and does not typically bear consid-
eration in the analysis of scoliotic gait.
Lower Extremity Asymmetry
With the exception of arm swing and asymmetrical trunk rotation, evidence on the
asymmetry of lower extremity kinematics is mixed. Most studies fail to find a
statistically significant side-to-side asymmetry at specific lower extremity joints.
Additionally, there does not appear to be any association between any observed
kinematic asymmetry and the convex and concave sides of the curve (Mahaudens
et al. 2009; Kramers-de Quervain et al. 2004). However, in scoliosis, frontal and
transverse plane motion of opposite limbs during walking has been shown to be less
correlated than in healthy adolescents (Yang et al. 2013). This suggests that asym-
metry is not due to one particular joint, but rather the collective motion of all
segments. Again, it is unclear whether this occurs due to an altered position of the
center of mass or as part of a greater postural control strategy. The asymmetries, as
well as the kinematic deviations, appear unrelated to the side and severity of the
curve, challenging the hypothesis that gait abnormalities are an isolated result of the
structural deformity. Still, theories of neurological dysfunction warrant further
investigation, particularly in the arena of postural control and sensorimotor
performance.
Kinetics
The observations of altered kinematics during gait in scoliosis suggest

corresponding alterations in the joint forces controlling movement. Kinetic analyses
reveal abnormalities in muscle moments, as well as magnitudes and patterns of
ground reaction forces. Patients demonstrate absolute deviations from healthy
parameters as well as some bilateral asymmetries.
Vertical Forces
Most research agrees that vertical ground reaction forces during gait in scoliosis are
similar to those of healthy adolescents. Patients with scoliosis show normal peak and
average ground reaction forces during the gait cycle. Additionally, there is no
evidence of asymmetry of the vertical ground reaction forces of opposite limbs.
Table 1 Correlation CC
coefficient (CC) of the
Control Scoliosis P
ground reaction force
(GRF) (Reproduced from FX 0.087 (0.02) 0.75 (0.05) 0.039
Yang et al. 2013 with FY 0.98 (0.00) 0.97 (0.00) ns
permission of Springer) FZ 0.99 (0.00) 0.99 (0.00) ns
Values in brackets represent the standard error. P value is the t test
result. FX, FY, and FZ represent the GRF in the medial/lateral, anterior/
posterior, and vertical forces, respectively. ns no significance
Still, while the force magnitudes are similar to those seen in healthy walking, there is
some evidence of pattern differences. In one study that focused solely on vertical
ground reaction forces, the majority of adolescents with scoliosis showed asymmet-
rical loading and unloading rates (Schizas et al. 1998). Previous data established that
bilateral vertical ground reaction force differences in healthy walking were within
4 % (Herzog et al. 1989), and the differences present within the scoliotic subjects
exceeded this threshold. Nevertheless, these differences did not appear to be related
to the severity or location of the scoliotic curve.
Horizontal Forces
In contrast to the vertical forces, which mostly resemble normal walking, the
horizontal ground reaction forces are notably different in scoliosis. Patients show a
significant asymmetry in the anterior/posterior (Park et al. 2016) and medial/lateral
(Giakas et al. 1996; Yang et al. 2013) component of the ground reaction force, as
evidenced by lower correlations between right and left sides than the correlations
observed in healthy subjects (Table 1).
This finding aligns with the kinematic abnormalities seen during the gait
cycle. Frontal and transverse plane asymmetry as seen in the trunk and pelvis
deviates the center of mass from the midline of the body, inducing the observed
bilateral differences in the medial/lateral component of the ground reaction
force.
Moments and Powers
Adolescents with idiopathic scoliosis appear to have similar peak joint moments and
powers to their healthy peers; however, the bilateral differences in the medial/lateral
ground reaction forces are accompanied by an asymmetry in the free moment. The
free moment, or the reaction of the foot’s force acting about a vertical axis, is an
indicator of torsional loading. This parameter is asymmetrical in scoliosis: the right
side of patients with a right thoracic curve demonstrates a bias toward an external
rotation moment (Kramers-de Quervain et al. 2004). This is the same direction of the
structural pelvic torsion and the opposite direction of the asymmetrical trunk rotation
seen in the kinematics. An asymmetrical free moment is yet another finding that
supports the idea of impaired control of motion about the vertical axis of the body.
Balance and Stability
The kinematic and kinetic abnormalities in scoliosis all seem to suggest that stability
may be compromised during walking. Standing and dynamic balance are of great
interest in the scoliotic population, given the alterations in center of mass position
resulting from the structural deformity and corresponding restrictions (i.e., stiffness)
in joint motion.
Measures of standing balance and proprioception in scoliosis demonstrate deficits
in these parameters. Patients with scoliosis show differences between limbs in their
abilities to reproduce a static knee angle, although the side of the limb with greater
difficulty does not appear to be correlated with the side or severity of the spinal curve
(Barrack et al. 1984). Additionally, in a standing postural control task, adolescents
with idiopathic scoliosis exhibit significantly more sway in all directions (Chen
et al. 1998).
Dynamic stability, however, does not appear to differ from healthy adolescents
(Mallau et al. 2007). As previously mentioned, walking speed is not slower in
scoliosis. Furthermore, when confronted with a more difficult ambulation task,
such as walking on a line or on a beam, adolescents with scoliosis utilize the same
strategy as healthy walkers: they slow down. There is no difference in the rate or
magnitude of this reduction of speed. Additionally, patients with scoliosis and
healthy adolescents react similarly to an increase in difficulty of a dynamic balancing
task by increasing horizontal angular dispersions, i.e., medial/lateral and anterior/
posterior limb motion.
Muscle Activation
The uncertainty regarding a neurological influence on observed joint motion in

scoliosis has prompted further investigation into motor function during gait. Abnor-
malities in proprioception and postural sway suggest impaired control, and the
presence of altered kinetics and kinematics are sometimes theorized to be the result
of neuromuscular dysfunction. For idiopathic scoliosis, where there is no known
underlying cause of the deformity, analysis of muscle activation patterns may
provide insight into the origin or progression of the scoliosis pathology.
Adolescents with idiopathic scoliosis demonstrate increased duration of muscle
activity throughout the gait cycle. The gluteus medius, quadratus lumborum,
semitendinosus, and erector spinae muscles have been shown to be active for an
additional 10 % of the gait cycle on average, when compared to healthy walking
(Mahaudens et al. 2009; Fig. 5).
This phenomenon is observed for muscles on both the convex and concave sides
(Mahaudens et al. 2005, 2009, 2010). Furthermore, the increased demand on the
Erector Spinae Quadratus lumborum
Healthy Subject
Cobb angle < 20°
Cobb angle [20-40°]
Cobb angle > 40°
Gluteus medius Rectus femoris
Healthy Subject
Cobb angle < 20°
Cobb angle > 40°
0 20 40 60 80 100
Stride (%)
Semitendinosus
Healthy Subject
Cobb angle < 20°
Cobb angle > 40°
0 20 40 60 80 100
Stride (%)
Fig. 5 Typical trace of electromyographic activity of quadratus lumborum, erector spinae, gluteus
medius, rectus femoris, and semitendinosus for a scoliosis patient from each scoliosis group
compared to a normal subject, expressed as a function of normalized stride (expressed in %). The
horizontal black bars represent the phasic activity of the muscles for the right side of the normal
subject and convex side of patients. The horizontal gray bars represent the phasic activity of the
muscles for the right side of the normal subject and concave side of patients (Reproduced from
Mahaudens et al. 2009 with permission of Springer)
muscles raises the energetic cost of walking in scoliosis and decreases the efficiency
of gait, as measured by total muscle work divided by the net energy cost (Mahaudens
et al. 2010).
It is unknown whether this prolonged activation is a result of neurological
dysfunction or is simply a compensatory mechanism to maintain stability during
walking. If the latter, the abnormalities in muscle activity would be considered a
result of the deformity rather than part of its cause. However, if this were the case, the
duration of muscle activity would likely be related to the severity of the scoliotic
curvature, and the effect would be exacerbated in more extreme curves. Instead, the
prolonged activation is observed similarly in all levels of curve severity, even those
with only minor scoliosis (Mahaudens et al. 2009). This finding reduces the likeli-
hood that the gait pathology is entirely a consequence of the spine and pelvis
deformities, again suggesting a neuromuscular contribution.
Relationship of Gait Parameters to Curve Severity
Many studies have investigated the relationship of the gait pathology to the severity
and type of scoliotic curve. With patients walking at a fixed speed, Mahaudens
et al. found no significant relationship of any of the aforementioned gait abnormal-
ities to the degree of curvature; however, they did observe trends suggesting
transverse pelvic motion decreases with greater curve severity (2009). In another
large study of patients with thoracolumbar curves, where patients were permitted to
walk at a self-selected speed, several gait variables did appear to be related to curve
severity. Knee flexion at initial contact increased with the degree of deformity, while
knee range of motion decreased. Additionally, reductions in cadence and pelvic
range of motion were more pronounced in patients with more severe curves
(Syczewska et al. 2012).
Evidence of the relationship between kinetic abnormalities and curve severity is
mixed as well. Schizas et al. determined there was no association between ground
reaction force asymmetry and degree of spinal deformity (1998). However, in a
study that considered multiple types of curvature, Park et al. established relation-
ships between ground reaction force asymmetry and the severity of spinal curvature
and pelvic tilt (2016).
Amid conflicting findings, this relationship of the gait pathology to curve severity
continues to be of great interest for clinicians. If the gait pathology is related to curve
severity, then correction of the deformity, such as surgical or therapeutic interven-
tions, may influence gait outcomes.
Response to Treatment
In theory, it seems likely that interventions for scoliosis would affect performance in
walking, though one might argue these effects could be either beneficial or delete-
rious. The most common treatments, orthotic bracing and surgical fusion of the
spine, are partially aimed at restoring structural symmetry. However, both bracing
and surgery impose a rigidity on the torso, which could exacerbate the preexisting
stiffness during gait. The following section discusses the effect of both types of
treatment on motion of the spine and distal segments during walking.
Fig. 6 Common bracing

options for adolescents with
idiopathic scoliosis: Boston
brace (left) and Wilmington
brace (right) (Courtesy of
Dr. Peter Gabos, Nemours/
Alfred I. duPont Hospital for
Children, Wilmington, DE)
Response to Bracing
Bracing treatment is a common nonsurgical intervention for idiopathic scoliosis, but

rarely results in any curve improvement. The treatment regimen typically requires
the patient to wear a customized orthosis between 12 and 20 h a day. Various types of
braces exist, some spanning the cervical spine to the sacrum, some shorter, some
rigid, and some more flexible (Fig. 6).
In the short term, both rigid and flexible braces appear to reduce motion of the hip
and pelvis during gait (Wong et al. 2008). In contrast, after long-term orthotic
treatment (6 months) and a substantial period between removing the brace and gait
testing, treatment effects show increased pelvis and hip motion in the frontal plane
(Mahaudens et al. 2014). As previously established, untreated scoliosis patients
demonstrate restricted pelvis and hip motion compared to their healthy peers, and
thus, an increase in these parameters represents progress toward a more normal
pattern of walking.
Additionally, a decrease in the abnormally high duration of activity of the erector
spinae throughout the gait cycle is observed following long-term orthotic treatment
(Mahaudens et al. 2014). In untreated patients, the excessive activity is theorized to
provide a stiffening effect for balance, which may no longer be necessary following
prolonged brace-wearing. Still, the energy cost of walking, which is elevated in
scoliosis, does not appear significantly reduced after long-term bracing treatment.
Response to Surgery
Surgical treatment in scoliosis is typically reserved for severe curves (those exceed-
ing 50 ) (Weinstein et al. 2003). The most common technique, spinal fusion,
involves the insertion of metal screws into vertebral pedicles and attachment of a
rod spanning the length of the curve. This results in an immediate straightening and
Fig. 7 Example of a 14-year-old girl with a combined right thoracic and left lumbar (Lenke
2C-[R]) curve operated on with an all pedicle screw construct. (a) Preoperative and (b) last
follow-up frontal radiographs; (c) preoperative and (d) last follow-up sagittal radiographs (Courtesy
of Dr. Peter Gabos, Nemours/Alfred I. duPont Hospital for Children, Wilmington, DE)
de-rotation of the spine (Fig. 7) with patients typically returning to full activity by
6 months (Lehman et al. 2015).
Fusion surgery imposes a restriction of spinal range of motion in all three planes,
the extent of which depends on the number of vertebrae involved in the fusion
(Danielsson et al. 2006; Engsberg et al. 2003). While the reduction in spinal range of
motion could theoretically exacerbate the stiffness observed in the gait of untreated
scoliosis patients, it is believed that the structural correction can reduce energy
demands, thereby increasing muscle efficiency.
Results vary by the type of scoliotic curve. Surgical correction of thoracic curves
seems to have little effect on most gait variables. The main result is a reduction of
transverse plane shoulder motion (Mahaudens et al. 2010), which essentially corrects
the asymmetrical forward rotation of the trunk and shoulders described by Kramers-
de Quervain et al. (2004).
For thoracolumbar and lumbar curves, frontal plane pelvis and hip motion
increases postoperatively. These results are similar to the long-term effects of
bracing and represent a normalization of the motion of these segments during gait
(Mahaudens et al. 2010). Additionally, there is a reduction in lateral center of mass
displacement (i.e., sway) during walking, demonstrating potential evidence of better
dynamic stability postsurgery (Paul et al. 2014).
Fusion surgery does not appear to effect significant changes in muscular work or
muscle activation timing. There may be a trend toward reduction of energy cost;
however, the differences are not significant. Even postsurgery, adolescents with
Fig. 8 Total energy cost. The 4

mean (vertical bar chart)
SD (vertical bars) are drawn
Energy cost (J kg-1 m-1)

in presurgery condition (white
bar) and in postsurgery
condition (gray-lined bar).
The black bar represents the
mean of norms (Adapted from 2
Mahaudens et al. 2010 with
permission of Springer)
0
AIS pre surgery AIS post surgery Norms
idiopathic scoliosis still walk with increased energy cost when compared to their
healthy peers (Mahaudens et al. 2010; Fig. 8).
Overall, while some abnormalities remain, treatment for scoliosis generally
results in modified kinematics that better resemble motion in healthy walking,
specifically increased motion of the hip and pelvis. Additionally, bracing treatment
reduces excessive muscle activity, and surgical treatment appears to slightly improve
efficiency of walking by reducing overall energy cost.
Summary and Conclusion
Gait performance in scoliosis has been heavily researched, with most investigations
reporting some abnormalities compared to healthy walking. Primary observations
include reduced trunk and pelvic motion within the frontal and transverse planes,
often reported clinically as a “stiffness” in ambulation. Various reports of asymmetry
exist, the most consistent finding an uneven progression of the trunk in the transverse
plane, with excessive forward rotation of the right shoulder throughout the gait
cycle. There is some evidence of postural control deficits and asymmetries in lower
limb kinematics and joint moments; however, these results vary throughout the
literature. Muscle activity appears increased in duration throughout the gait cycle,
and energy cost of walking is higher when compared to healthy adolescents.
Treatment for scoliosis, both orthotic and surgical, has a positive effect on gait
variables. Shoulder, pelvis, and hip motion improve toward a more normal pattern,
and center of mass displacement is reduced. Furthermore, while overall levels are
still higher than normal, treatment also alleviates some of the energy cost of walking
for patients with scoliosis.
Despite the extensive research, conclusions about how gait abnormalities relate to
the origin or progression of scoliosis remain vague and often conflicted. Impaired
gait does appear to be associated with idiopathic scoliosis. Still, the ideas that the gait
pathology contributes to curve progression and that impaired gait and the spinal
deformity are both secondary to some underlying neurological disorder are still
largely based in theory. Continued research into motor control and somatosensory
function may provide more insight into neurological relationships between the
deformity and gait performance. In the meantime, gait analysis can still provide a
valuable assessment of global function and evaluation of therapeutic and surgical
outcomes in scoliosis.
References
Ascani E et al (1986) Natural history of untreated idiopathic scoliosis after skeletal maturity. Spine
11(8):784–789
Asher MA, Burton DC (2006) Adolescent idiopathic scoliosis: natural history and long term
treatment effects. Scoliosis [Online] 1. Available at: http://www.ncbi.nlm.nih.gov/pmc/arti
cles/PMC1475645/. Accessed 12 Feb 2015
Barrack RL et al (1984) Proprioception in idiopathic scoliosis. Spine (Phila Pa 1976) 9(7):681–685
Chen P-Q et al (1998) The postural stability control and gait pattern of idiopathic scoliosis
adolescents. Clin Biomech 13(1):S52–S58
Danielsson AJ, Romberg K, Nachemson AL (2006) Spinal range of motion, muscle endurance, and
back pain and function at least 20 years after fusion or brace treatment for adolescent idiopathic
scoliosis: a case-control study. Spine (Phila Pa 1976) 31(3):275–283
Engsberg JR et al (2003) Prospective comparison of gait and trunk range of motion in adolescents
with idiopathic thoracic scoliosis undergoing anterior or posterior spinal fusion. Spine (Phila Pa
1976) 28(17):1993–2000
Giakas G et al (1996) Comparison of gait patterns between healthy and scoliotic patients using time
and frequency domain analysis of ground reaction forces. Spine (Phila Pa 1976) 21
(19):2235–2242
Grivas TB et al (2010) Brace technology thematic series: the dynamic derotation brace. Scoliosis
[Online] 1. Available at: http://www.ncbi.nlm.nih.gov/pubmed/20858270. Accessed 18 Feb
2015
Gum JL et al (2007) Transverse plane pelvic rotation in adolescent idiopathic scoliosis: primary or
compensatory? Eur Spine J 16(10):1579–1586
Herzog W et al (1989) Asymmetries in ground reaction force patterns in normal human gait. Med
Sci Sports Exerc 21(1):110–114
Konieczny MR, Senyurt H, Krauspe R (2013) Epidemiology of adolescent idiopathic scoliosis. J
Child Orthop 7:3–9
Kramers-de Quervain IA et al (2004) Gait analysis in patients with idiopathic scoliosis. Eur Spine J
13(5):449–456
Lehman RA et al (2015) Return to sports after surgery to correct adolescent idiopathic scoliosis: a
survey of the Spinal Deformity Study Group. Spine J 15(5):951–958
Mahaudens P, Thonnard JL, Detrembleur C (2005) Influence of structural pelvic disorders during
standing and walking in adolescents with idiopathic scoliosis. Spine J 5(4):427–433
Mahaudens P et al (2009) Gait in adolescent idiopathic scoliosis: kinematics and electromyographic
analysis. Eur Spine J 18(4):512–521
Mahaudens P et al (2010) Gait in thoracolumbar/lumbar adolescent idiopathic scoliosis: effect of
surgery on gait mechanisms. Eur Spine J 19(7):1179–1188
Mahaudens P et al (2014) Effect of long-term orthotic treatment on gait biomechanics in adolescent
idiopathic scoliosis. Spine J 14(8):1510–1519
Mallau S et al (2007) Locomotor skills and balance strategies in adolescents idiopathic scoliosis.
Spine (Phila Pa 1976) 32(1):E14–E22
Mayo NE et al (1994) The Ste-Justine adolescent idiopathic scoliosis cohort study. Part III: back
pain. Spine (Phila Pa 1976) 19(14):1573–1581
Park HJ et al (2015) Analysis of coordination between thoracic and pelvic kinematic movements
during gait in adolescents with idiopathic scoliosis. Eur Spine J 25:385–393
Park YS et al (2016) Association of spinal deformity and pelvic tilt with gait asymmetry in
adolescent idiopathic scoliosis patients: investigation of ground reaction force. Clin Biomech
36:52–57
Paul JC et al (2014) Gait stability improvement after fusion surgery for adolescent idiopathic
scoliosis is influenced by corrective measures in coronal and sagittal planes. Gait Posture 40
(4):510–515
Perry J, Burnfield JM, Cabico LM (2010) Gait analysis: normal and pathological function, 2nd edn.
SLACK, Thorofare
Prince F et al (2010) Comparison of locomotor pattern between idiopathic scoliosis patients and
control subjects. Scoliosis [Online] 1. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/
PMC2938665/. Accessed 1 Mar 2015
Saji M, Upadhyay S, Leong J (1995) Increased femoral neck-shaft angles in adolescent idiopathic
scoliosis. Spine (Phila Pa 1976) 20(3):303–311
Schizas CG et al (1998) Gait asymmetries in patients with idiopathic scoliosis using vertical forces
measurement only. Eur Spine J 7(2):95–98
Schlösser TPC et al (2014) How “idiopathic” is adolescent idiopathic scoliosis? A systematic
review on associated abnormalities. PLoS One [Online] 9(5). Available at: http://www.ncbi.
nlm.nih.gov/pmc/articles/PMC4018432/. Accessed 14 Jan 2015
Schwender JD, Denis F (2000) Coronal plane imbalance in adolescent idiopathic scoliosis with left
lumbar curves exceeding 40 degrees: the role of the lumbosacral hemicurve. Spine (Phila Pa
1976) 25(18):2358–2363
Syczewska M et al (2012) Influence of the structural deformity of the spine on the gait pathology in
scoliotic patients. Gait Posture 35(2):209–213
Weinstein SL et al (2003) Health and function of patients with untreated idiopathic scoliosis: a
50-year natural history study. JAMA 289(5):559–567
Wong MS et al (2008) The effect of rigid versus flexible spinal orthosis on the gait pattern of
patients with adolescent idiopathic scoliosis. Gait Posture 27(2):189–195
Yang JH, Suh SW et al (2013) Asymmetrical gait in adolescents with idiopathic scoliosis. Eur Spine
J 22(11):2407–2413
Concussion Assessment During Gait
Robert D. Catena and Kasee J. Hildenbrand
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
What is a Concussion? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Current Clinical Considerations in Concussion Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Steady-State Gait Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Functional Gait Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Dual-Task Gait Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
The acute signs and symptoms (SS) of a concussion can vary widely between
individuals. Clinicians currently use a variety of measures to diagnosis and
manage both physical and cognitive SS associated with concussion. Balance is
typically assessed using quick sideline measures in sports; however, researchers
have found through more thorough assessments of dynamic balance during gait
that SS may persist beyond those detected through typical assessment techniques.
An appropriate gait assessment of concussion must be adequately complex to
distinguish persistent balance deficits, but not so complex that healthy individuals
would be challenged to maintain balance. A steady-state gait assessment may
indicate conservative gait adaptations but will seldom yield distinct signs of
continued dysfunction following concussion. Obstacle avoidance tasks demon-
strate conservative gait adaptations long after other SS have resolved. Concussion
R.D. Catena (*)

Gait and Posture Biomechanics Lab, Washington State University, Pullman, WA, USA
e-mail: robert.catena@wsu.edu
K.J. Hildenbrand
Athletic Training Program, Washington State University, Pullman, WA, USA
e-mail: khildenbrand@wsu.edu

DOI 10.1007/978-3-319-30808-1_69-1
2 R.D. Catena and K.J. Hildenbrand
typically results in balance deficits in divided attention dual-task paradigms even

after a return to normal daily activities. Refinements of gait paradigms to be more
specific and clinically useful define future advances in concussion assessment
during gait.
Keywords
Concussion assessment • Balance • Steady-state gait • Obstacle avoidance • Dual
task • Balance deficits
State of the Art
The state of concussion management and research has exploded over the last decade,
with many different disciples using their unique lens to examine the issue. Clinicians
may take the patient-centered perspective that focuses on specific symptoms, daily
activities, and quality of life. Researchers use broad group comparisons to determine
causation, correlations, diagnosis, and rehabilitative techniques. Each clinical and
research discipline currently has its own criteria, which can make cross-discipline
comparisons difficult. Historically, gait researchers have found significant differ-
ences when examining subjects with “mild” concussions versus “severe” concus-
sion, while clinical professionals have now moved away from mild and severe
dichotomization. Comparison of research between groups and between older and
newer research is difficult with different groups occasionally changing the criteria by
which they grade severity, measure symptoms, treat, and clear for activity. Gait
analysis has the benefit of detecting changes from a concussion that will be present
after other criteria have been used to return individuals to participation. Analysis of
gait also allows treatment for patients to improve overall balance, movement effi-
ciency, and overall quality of life. Currently many gait assessment techniques
involve expensive and complex equipment, but strides are being made to use more
widely available measures that can be implemented in a diagnostic or rehabilitative
manner.
What is a Concussion?
The Zurich Consensus Statement defines concussion as “a complex pathophysio-

logical process affecting the brain, induced by traumatic biomechanical forces”
(McCrory et al. 2013). The consensus statement goes further to indicate several
common features that incorporate clinical, pathological, and biomechanical injury
constructs, which can be helpful in discussing the nature of the concussion:
1. Concussion may be caused either by a direct blow to the head, face, or neck or a
blow elsewhere on the body with an “impulsive” force transmitted to the head.
2. Concussion typically results in the rapid onset of short-lived impairment of
neurologic function that resolves spontaneously.
Concussion Assessment During Gait 3
3. Concussion may result in neuropathological changes, but the acute clinical

symptoms largely reflect a functional disturbance rather than a structural injury.
4. Concussion results in a graded set of clinical symptoms that may or may not
involve loss of consciousness. Resolution of the clinical and cognitive symptoms
typically follows a sequential course. In a small percentage of cases, however,
post-concussive symptoms may be prolonged.
5. No abnormality on standard structural neuroimaging studies is seen in
concussion.
The acute clinical symptoms of a concussion vary widely between individuals,

complicating a diagnosis. This variation between individuals is one of the most
difficult problems clinicians and researchers face. Concussions differ from other
injuries because often the symptoms are vague and no clear indicators exist to
determine if an injury occurred. Common indicators for a concussion are physical
signs (loss of consciousness), behavioral changes (irritability), cognitive impairment
(slowed reaction time), or sleep disturbances (drowsiness) (McCrory et al. 2009).
Concussion as defined above, along with the common symptoms, is the information
medical professionals must rely on to diagnose and guide treatment.
Immediately after a head impact occurs, there is a release of the excitatory amino
acid neurotransmitters glutamine and aspartate (Grady 2010). These molecules lead
to a loss of cell membrane integrity in the brain, which increases the concentration of
sodium ions and decreases the concentration of potassium ions within cells. These
changes in ion concentration affect the brain cells’ pH and lead to an increase in
calcium ion concentration. These disturbances in the ion concentration lead to
cellular damage resulting in the death of the affected brain cells (Grady 2010).
Upon cell death, these brain cells release cytokines. Cytokines are responsible for
the body’s inflammatory response, and they upregulate inflammation upon release
within cells. This increase in inflammation is what is observed in the brain following
a concussion injury and is responsible for added damage to the affected brain region
(Grady 2010).
The brain’s response to a concussion can be thought of in two separate parts.
Initially, there is cellular damage resulting from the changes in ion concentration,
such as sodium and potassium. These changes are responsible for the acute symp-
toms of concussion such as headaches and dizziness (Grady 2010). The second part
is the inflammatory response caused by the release of cytokines within brain cells
after a concussion. The timeline of this inflammation is the reason concussion
symptoms often worsen in the 6–24 h post-injury (Grady 2010).
The concussions that cause these chemical reactions are commonly referred to as
mild traumatic brain injuries (mTBIs). Mild traumatic brain injury is technically not
considered synonymous with concussion by many experts, but mTBI is often seen in
the literature as a replacement for concussion. Research on concussions and mTBI
has rapidly increased over the last decade due to specific findings that concussions
may lead to degenerative brain diseases and cognitive impairments later in life.
Sports-related concussions (SRC) within an athletic population have been esti-
mated between 1.1 and 1.9 million annually, while estimates range from 22.5 % to
52.7 % of concussions likely not reported to a healthcare professional (Bryan et al.

2016). Evaluating for the incidence of concussions has an added challenge, as many
athletes choose not to report their symptoms due to the restrictions from activity
when suffering from a diagnosed concussion. Challenges in determining specific
incidence rates of concussions also occur due to changes in the previous definition of
SRC and difficulty in establishing appropriate reporting of SRC to coaches and
healthcare professionals. Theye and Mueller (2004) state 20 % of head injuries
(>300,000) are sports-related concussions. A study has shown that emergency
department (ED) visits for SRC have risen over 200 % from 1997 to 2007, indicat-
ing a considerable increase in incidence (Schatz and Moser 2011). This growth can
be attributed to both an increase in participation in sports and progression in
knowledge of signs and symptoms of the condition, resulting in improved reporting
of the condition (Register-Mihalik, et al. 2013). Another factor in the increase in
reporting is the growth in awareness of concussions in the media, resulting in greater
appreciation of its severity by the healthcare community.
Current Clinical Considerations in Concussion Management
Certain licensed medical professionals such as athletic trainers in North America and
medical doctors can both diagnose and determine qualifications for return to activity
for those who have been afflicted with head trauma. One of the most widely used
systems to diagnose concussions is the SCAT3 or Sport Concussion Assessment
Tool – 3rd Edition (McCrory et al. 2013). The SCAT3 includes a system of questions
that the afflicted person must answer. The person’s score at the end of the test is then
compared to a baseline score or normative values if a baseline was not assessed. The
baseline assessment should be conducted before the athlete experiences a concus-
sion, since some athletes may already experience some symptoms naturally, such as
having poor balance. The fewer the number of symptoms the athlete has, the lower
the overall score on the SCAT3. Returning an athlete to activity requires a multifac-
eted approach with several measures of symptoms, cognitive function, balance, and
ocular function all returning to baseline or “normed” values.
The SCAT3 evaluates subjective symptoms as well, such as mood and nausea.
Cognitive information is included, such as the current month, day, and year. The
SCAT3 evaluates the athlete’s memory of sequences of words and numbers. Balance
is also evaluated using the BESS test, or Balance Error Scoring System, where the
athlete must balance on both feet, one foot, and then tandem stance.
Another cognitive assessment tool commonly used to evaluate concussions is
ImPACT or Immediate Post-Concussion Assessment and Cognitive Testing
(Covassin et al. 2009). Many universities and some high schools pay to have access
to this system, since it can track athletes’ records. Athletes should take the ImPACT
as a baseline before the season begins to get a score, which is compared to normative
ranges. Then, after an athlete sustains a concussion and his or her symptoms have
subsided, the athlete retakes the ImPACT in order to gain an objective score of his or
her cognitive function.
Even with these diagnostic tools, clinicians are expected to follow a strict
progression when returning afflicted athletes to sport participation (McCrory et al.
2013). Return-to-play progression may not begin until the athlete’s symptoms have
returned to baseline and subsided. At this point, an athlete can begin to move through
stages of light exercise, then progress to sport-specific activities, noncontact training,
and integration into practice, and finally return to play. However, 24 h must pass
between each stage, and if the athlete experiences any reoccurrence of symptoms, he
or she must return to the previous stage.
The extreme care taken with traumatic brain injuries may seem extensive, but
these strict regimens are necessary. If athletes return to sport participation with
unresolved symptoms, the results can be irreversible, like with second impact
syndrome (Cantu 1998). This syndrome occurs when an individual suffering post-
concussive symptoms participates in activity and receives a second head trauma. At
this point, major brain swelling occurs and death is a possible outcome with a 50 %
mortality rate.
The cautious approach to returning an individual to activity is also used to
minimize lingering symptoms past when normal resolution of symptoms should
occur. Some medical professionals labeled this lingering of symptoms as post-
concussion syndrome, though other professionals argue with the diagnosis. The
important point for this discussion is that balance is often a symptom that can have
lasting issues after other symptoms have resolved. Typically balance is clinically
assessed quickly using a modified version of the BESS test or Romberg (standing on
one leg with arms out to a T and eyes closed). Gait can be used to assess balance as
well, but is not part of a typical concussion management program, especially within
high school or college athletic programs. Specialized clinics for patients who have
longer-lasting symptoms may use gait to detect differences beyond simple balance
tests, and gait analysis can sometimes detect differences well beyond when an athlete
may have returned to participation.
Steady-State Gait Assessment
The most common form of gait assessment is a steady-state gait analysis. Gait
requires a series of coordinating interjoint and interlimb movements to achieve
movement of the center of mass while also balancing that center of mass appropri-
ately to avoid a fall. Strength, reaction time, and coordination highlight some of the
important motor components that are required during both of these two tasks in gait.
Since balance incorporates many neuromuscular and neuropsychological compo-
nents that can be affected by brain injury (Fig. 1), balance assessment is one of the
most commonly performed and effective motor assessments following concussions.
Balance deficits following concussion were widely recognized in the 1990s when
about half of all surveyed concussed individuals (of all severity levels) reported
dysfunction due to their injury in a 5 year post-TBI survey and physician assessment
study appraising balance impairment (Hillier et al. 1997). This, along with some
Fig. 1 A conceptual model of the postural control system (Reproduced from Maki and McIlroy
1996)
high-profile cases of concussions to children and famous athletes, paved the way for
concussion balance research through the next decade.
As indicated in Fig. 1 (Maki and McIlroy 1996), balance control can be modu-
lated by a number of different factors. Concussion affects balance at the central
nervous system level but can be measured through gait assessment of the mechanical
output. Research findings are mixed for effects at any particular sensory system in
this balance pathway. Some research indicates no particular deficit in any one
sensory input to balance following traumatic brain injury, but still a level of
imbalance corresponding with injury severity (Mrazik et al. 2000). Others suggest
vestibular dysfunction resulting from concussion (Aligene and Lin 2013; Alsalaheen
et al. 2010; Corwin et al. 2015; Fife and Kalra 2015; Murray et al. 2014). While
sensory organization tests are a typical method for isolating the effect of vestibular
dysfunction on balance, dysfunction can be evident through, and must be accounted
for, in clinical tests involving balance such as a gait assessment. When performing
gait, a kinematic asymmetry may be most easily detectable with unilateral vestibular
dysfunction. Others have used more intricate measures of center of mass motion to
detect dysfunction (Deshpande and Patla 2005).
Stability, posture, and balance are sometimes considered synonymous. For our
purposes, we will differentiate these terms to clearly define the research from here
on. “Posture” is an instantaneous pose of the body and the many joint positions that
create that pose. “Balance” is the instantaneous measure of the propensity to fall
through measures of the body center of mass with respect to a fulcrum (center of
pressure) or base of support (area contained within the feet). “Stability” represents
the consistency of a repeated cyclical action, and in the case of postural stability, that
is the consistency of body motions at the joint level or of the whole body through a
measure of the center of mass or more commonly measured center of pressure over a
time period. These are referred to as “nonlinear analyses.” The interpretation of
consistency versus randomness in movements in healthy human motion versus
deficient motion is being explored by a number of research groups, and so the
clinical applications of these nonlinear measurements are yet to be fully understood.
We suggest that the reader consults additional sources such as Cavanaugh et al.
(2005) about how nonlinear measurement techniques can provide clinically relevant
information particular to different populations.
Standing balance tests represent some of the earliest motor tests conducted in a
concussed population to detect persistent balance deficits. More severe forms of
brain injury will manifest symptoms in basic standing tests. Balance tests have
indicated increased postural sway following concussions compared to healthy bal-
ance in clinical (Geurts et al. 1996; McCrea et al. 2003) and functional testing
(Zhang et al. 2002). These tests will typically only detect immediate symptoms;
however, individuals that have a history of concussive events can present persistent
short-term (Gao et al. 2011; Quatman-Yates et al. 2015) and long-term
(De Beaumont et al. 2011, 2013; Sosnoff et al. 2011) postural stability irregularities
compared to healthy individuals. It typically takes more demanding motor tests to
detect any balance symptoms from more mildly concussed individuals.
In conducting a gait assessment, it is proper to first consider your goal. One goal
may be to simply detect lingering motor (or other) performance symptoms that
indicate that the concussed individual is not completely healthy (De Beaumont
et al. 2011). Alternatively, your goal may be to detect particular lingering symptoms
that could affect performance in a functional activity. Some forms of gait assessment
may satisfy both of these goals but with reduced power. The more mildly concussed
your group or individual is, the more precise you need your gait assessment to
be. Secondly, you must develop an appropriate gait assessment considering the
sensitivity and specificity needed to distinguish lingering symptoms for a particular
level of concussion severity and duration since the concussive event. Since the
persistence of symptoms is in question, the appropriate gait assessment must be
robust enough to examine a range of likely symptomology. It includes an appropriate
range of difficulties and considers how performance may be influenced by assess-
ment duration, assessment time, and assessment environment.
Balance is a measure of the center of mass with respect to the base of support.
Most tests of balance in a clinical setting are either standing balance, which can be
measured accurately without a direct measure of the center of mass, or involve costly
tools that can better predict the center of mass motion. The challenge with a balance
measure during gait is that a simpler measure, like how center of pressure is typically
used in standing tests, doesn’t as accurately predict center of mass motion in this
more dynamic task. The benefit of being able to measure balance during gait is that it
provides the best indication of likelihood of a fall or other loss-of-balance injuries.
Subsequent injury following concussions can be in the form of physical injury to the
body (Brooks et al. 2016) and the potential for additional concussive events that
compound the deleterious effects (CDC 1997).
Balance in gait additionally highlights different components of balance in Fig. 1

and, to an amplified level, compared to standing balance tests. Volitional movements
and feedforward corrections are more prevalent in gait than they are in standing;
thus, there is more reliance on the cerebral cortex, the very component concussion
affects. Coordination of the musculoskeletal linkage and processing visual and
somatosensory information are also more changing in gait than in standing. Gait is
a more unbalancing task than standing due to these factors and therefore may be a
more appropriate test of balance for mildly concussed individuals or individuals
further into their rehabilitation from a severe concussion.
Even though balance is a direct measure of the center of mass with respect to the
base of support, there is no perfect way to measure center of mass location in the
human body during dynamic tasks. With a full-body motion capture system and
previously established anthropometric models of segment inertial parameters, you
can estimate the center of mass location changes during gait. Use of anthropometry
to aid in determining the body center of mass doesn’t quite provide perfectly
accurate information; however, the young adult concussed population (less so for
adolescent concussed individuals) has plenty of anthropometry studies to validate
their use, and there are methods by which you can optimize the data to better fit a
specific individual (Pavol et al. 2002). A motion capture system can also track the
base of support so that measures of the center of mass with respect to the base of
support can measure balance.
Without a motion capture system for identifying center of mass location, force
plates can be used to estimate the center of mass motion. The projection of the center
of mass onto the ground (center of gravity) can be used for several measures of
balance and measured by double integration of force plate data (Zatsiorsky and King
1998). The problem with using the center of gravity is that it doesn’t account for
three-dimensional motion of the center of mass, which can affect balance. Center of
pressure alone (collected by force plate or pressure mat) has also been used to
estimate gait balance. The center of gravity is encompassed and controlled by the
center of pressure and ground reaction force during steady state gait, so center of
pressure will always overestimate the actual motion of the center mass. Outside of
this level of equipment sophistication, spatiotemporal measures of movement have
been used in the past as measures of balance in gait but are several levels away from
being true measures of balance.
Research into balance assessment during gait following concussions began from
several groups in the early 2000s. Mayo Clinic researchers used clinical assessments,
standing posturography (with the sensory organization test), and gait analysis to
compare assessment techniques following traumatic brain injury (TBI) (Basford
et al. 2003; Kaufman et al. 2006). They demonstrated a high correlation between
physical impairments described by the TBI participants and reduced scores on the
sensory organization test designed to isolate vestibular input. Physical disability
scores also correlated with reduced center of mass motion in the anterior/posterior
direction. Functional disability index correlated with increased center of mass
motion in the mediolateral direction. TBI participants also showed reduced gait
A/P motion and increased gait M/L motion compared to healthy controls. This
analysis indicates the significance of a gait analysis in assessments of perceived

impairments in a more severe TBI group.
For the more mildly concussed adult, a simple gait test doesn’t clearly discern
persistent balance deficits. Within 48 h of a concussive event, adults will display a
slowed gait, but no indications of balance deficits (Parker et al. 2005). By a week
post-concussion, single-task gait is indistinguishable to healthy adults just like
typical neuropsychological tests (Parker et al. 2006). Concussed adolescents how-
ever are more likely to present with balance deficits in single-task gait (Howell et al.
2013b). It is important to note that high-impact athletes that may frequently encoun-
ter “subconcussive” blows to the head could present gait balance deficits even
without a medical diagnosis of concussion (Parker et al. 2008). To accurately detect
persistent balance symptoms in a clinical setting, it is important to consider baseline
performance of your patient. In research, baseline information is ideal, but at
minimum this previous research indicates that athletic participation is important to
consider in group comparisons.
Functional Gait Assessment
Physical obstacles that must be negotiated during daily gait include curbs, stairs,
unstable surfaces, traffic, and a variety of other obstacles. Our ability to reorient our
attention to these particular obstacles factors into gait performance (Catena et al.
2009b). Concussed individuals demonstrate deficits in spatially orienting attention in
both auditory and visual tasks (Breton et al. 1991; Cremona-Meteyard et al. 1992;
Daffner et al. 2000; Halterman et al. 2006). In this process, we must disengage, shift,
and reengage attention (Posner 1980) through unique neuronal pathways. Broad
posterior parietal lobe damage has been linked to disengagement of attention (Posner
et al. 1984). The superior parietal gyrus is linked to shifting attention (Vandenberghe
et al. 2001). The intraparietal sulcus is involved in shifting and refocusing attention
(Yantis et al. 2002) but also the superior colliculus and lateral pulvinar when
distractions are present (Posner and Petersen 1990).
Obstacle avoidance tasks are a typical functional activity added to gait assessment
to increase the balance complexity and make the task more indicative of everyday
hazards to injury. The complexity of the obstacle crossing task can be modulated to
the expected ability of your population by making the task more or less physically
demanding (Chou et al. 2004) or more perceptually demanding (Baker and Cinelli
2014). Compared to clinical assessment techniques of balance, such as the Berg
Balance Test, an obstacle crossing task was better in distinguishing TBI individuals
from healthy controls with slower gait velocities, increased obstacle clearance, and
decreased stride lengths indicating the TBI group adopted a cautious gait during
obstacle crossing (McFadyen et al. 2003). Others have shown similar results from an
obstacle crossing task (Fait et al. 2013; Martini et al. 2011; Vallee et al. 2006).
Measures of whole body center of mass motion of TBI patients during the crossing
of several different obstacle heights show that participants with TBI and healthy
controls have similar gait patterns during unobstructed walking, indicating that
normal level walking may not be as sensitive in detecting long-term changes in

dynamic balance (Chou et al. 2004). On the other hand, obstructed walking resulted
in slowed gait velocities and shorter stride lengths (indicating more cautious gait)
and increased mediolateral swaying motion (indicating a lack of balance control)
2 years after a TBI (Chou et al. 2004).
The balance effects of an obstacle crossing during gait are essentially equivalent
between more mildly concussed individuals within 48 h of injury and healthy
individuals (Catena et al. 2007a), as both groups tend to be taxed with the challenge
of maintaining balance during obstacle crossing depending on the obstacle height.
However, the potential for a trip is higher after a recent concussion (indicated by
lower foot clearances and higher trip rates) for individuals that also have deficits in
spatially orienting attention (Catena et al. 2009b). Similar to long-term performance
in TBI individuals, mildly concussed individuals adopt a more conservative obstacle
crossing strategy (indicated both in balance and obstacle clearance measures) as
concussion symptoms subside several weeks following injury compared to their 48 h
performance (Catena et al. 2009a) and compared to healthy individuals (Catena et al.
2009a; Sambasivan et al. 2015). On the other hand, balance deficits are more likely
to be elicited by gait tasks that require a cognitive reaction to a suddenly presented
perturbation (Powers et al. 2014).
Dual-Task Gait Assessment
While broad neuropsychological tests have become the standard method for
assessing persistent concussive symptoms for a clinical return-to-activity decision
(Resch et al. 2013), it is important to note that widely used neuropsychological
clinical tests don’t measure all cognitive deficits to the same degree (Choe and Giza
2015) and broad tests of cognition don’t always present the same findings, as there
are specific tests that focus on particular cognitive components (Keightley et al.
2009). Nevertheless, there is the potential to refine current cognitive testing to
become an even better measure of persistent symptoms. In doing so, the chance
that a patient is involved in another deleterious subsequent concussive event is
reduced. One area to improve current testing is to focus on enhancing testing
methodologies that correlate with (to potentially predict) motor performance as
concussion-induced motor deficiencies could result in subsequent injuries (Brooks
et al. 2016; Herman et al. 2015).
Unlike many neurological pathologies, concussions don’t present any consis-
tently localized cognitive symptoms. Instead, axonal injury is diffuse, and so are
cognitive symptoms. Cognitive processing distribution throughout the brain obvi-
ously leads to questions about the probability of any one component affected by a
single biomechanical force with an intricate direction, magnitude, and point of
application. Cognitive deficits could include (but not necessarily limited to or
necessarily include in specific individuals) executive dysfunction, slower reaction
times, decreased focus, reduced working memory, reduced attention capacity, and
inability to shift attention.
Gait is not an automated task in which no attentional resources are needed.

Decreased attention capacity has been eluded to as a major determinant in reduced
gait performance following a concussion (Catena et al. 2011). No matter the general
theory of divided attention to which you prescribe, there is an abundance of evidence
to suggest an interaction between gait performance and cognition through attention
in healthy individuals (de Bruin and Schmidt 2010; Hegeman et al. 2012; Lajoie
et al. 1993; Szturm et al. 2013). Accomplishing a dual task, with reasonable success
in both simultaneously performed tasks, is even less automatic when challenged by a
deficit that can affect performance of either task (Brown et al. 1999; Vaportzis et al.
2015; Yardley et al. 2001). Concussion, directly affecting cognition and neurophys-
iology, challenges an individual to complete both simple and more complex dual-
task scenarios (Bernstein 2002; De Monte et al. 2005; Tapper et al. 2016; Vilkki et al.
1996). Dual-task gait research has provided even more scientific evidence of divided
attention deficits.
Executive control over cognitive processes allows individuals to achieve goals by
planning, focusing, and coordinating actions. Executive dysfunction has been con-
sistently reported as deficient following concussion (Hart et al. 2005; Howell et al.
2013a; Moore et al. 2016; Serino et al. 2006; Tapper et al. 2016). In particular,
sustained attention [primarily controlled in the right frontal areas (Posner and
Petersen 1990; Sturm et al. 1999; Wang et al. 2005; Wilkins et al. 1987)] on gait
performance is important for populations at risk of fall, injury, or re-injury due to a
fall. This is particularly important in gait when balance has been compromised.
Sustained attention does not seem to be deficient shortly after concussion when
balance is compromised (Halterman et al. 2006; van Donkelaar et al. 2005, 2006),
but there is some research that indicates a positive relationship between concussion
and lapses in attention long after a reported concussion occurred (Killgore et al.
2016; Pontifex et al. 2012).
In sustaining attention, executive control involves resolving conflicting informa-
tion. Concussed individuals experience conflict resolution deficits up to a month
post-injury (Chan 2002; Chan et al. 2003; Halterman et al. 2006; Larson et al. 2011;
Moore et al. 2014). The anterior cingulate cortex seems to be primarily responsible
for conflict resolution in such tasks (Posner and Rothbart 1998; Swick and Jovanovic
2002) and more specifically the mid-dorsal region (Swick and Jovanovic 2002). The
dorsal prefrontal cortex has also been linked to the actual selection response in more
difficult tasks (MacDonald et al. 2000).
Including a cognitive component to balance tests is one alternative to provide
increased task complexity to tease out milder symptoms or concussion symptoms
over longer duration. Distractions clearly play a role when assessing balance deficits
following concussion (Rahn et al. 2015). Cognitive performance is correlated to
balance performance following concussion (Alsalaheen et al. 2016). And through
attention, cognitive and balance performance interacts to diminish the performance
of either, or both, following a concussion (Catena et al. 2007a). Immediate balance
deficits are commonly observed in a dual-task paradigm (Sosnoff et al. 2008).
Month-long dual-task balance deficits are occasionally evidenced in the literature
as well (Dorman et al. 2015). Cognition may also interact with other motor
components, or even specific motor components used in balance, following concus-

sion (Brown et al. 2015).
Gait with a simultaneously performed cognitive task (dual task) has become the
functional paradigm of interest in the concussion research over the last decade.
These types of paradigms have been described as most similar to real-world condi-
tions (Cock et al. 2003; Weerdesteyn et al. 2003) when we are often performing
cognitive processing along with gait. In performing a dual-task analysis, the cogni-
tive task can be modulated to involve or exclude particular sensory and cognitive
tracts along with modulation of task complexity to fit your population. This is on top
of the modulation that can be made to the gait task as described for obstacle crossing
above. As such, there are a wide variety of paradigms described throughout the
research literature similar to the wide variety of dual-task situations faced daily.
Consider that concussion is a diffuse injury that can affect multiple areas of the brain,
when picking the correct dual-task paradigm.
Differences between TBI patients and healthy individuals are mixed when com-
bining obstacle avoidance with a cognitive secondary task compared to just single-
task obstacle crossing (Chiu et al. 2013; Martini et al. 2011; McFadyen et al. 2009;
Vallee et al. 2006). Cognitive tasks may interfere with motor performance through
peripheral sensory distraction or through central nervous system attention division in
dual-task paradigms. Peripheral sensory distractions, for example, when visual
cognition tracts are tasked simultaneously with an inherently visual motor task like
obstacle crossing, typically want to be avoided as they are a challenge regardless of
injury or injury severity (Fait et al. 2013). Along with picking the correct mode of
cognitive task, it is important to consider the correct amplitude of cognitive com-
plexity. While a visual cognition and obstacle crossing task may present as too
complex for even healthy individuals, some tasks like a simple reaction time test
during gait may present as too simple for mild concussion patients (Catena et al.
2007b).
A continuous choice mental task has typically shown to be best at discriminating
between symptomatic and asymptomatic individuals. A dual-task paradigm that
includes steady-state gait and a variety of continuous mental tasks results in few
changes for healthy individuals but reduced performance on cognitive tasks and
increased spatial-temporal gait variability for severe TBI patients, similar to indi-
viduals after stroke and subarachnoid hemorrhage (Haggard et al. 2000). Following
mild TBI, spatial-temporal variables only seem to be sensitive to conservative
control of dual-task gait immediate after injury (Parker et al. 2005) and don’t seem
to be sensitive enough to detect any lingering single-task gait performance deficits
several weeks after concussion (Howell et al. 2013b; Parker et al. 2006; Sambasivan
et al. 2015). However, continued balance deficits measured by center of mass motion
can still be detected several weeks following a concussion using a dual-task para-
digm (Parker et al. 2006). There is evidence to suggest that balance deficits are even
more apparent following an adolescent concussion (Howell et al. 2015a) and can be
prolonged by a return to activity too soon (Howell et al. 2015b, c; Parker et al. 2008).
Future Directions
Published research is skewed toward positive results. It is not clear how likely it is
for a concussed individual to present with deficits in a gait assessment. There is
immediate need for epidemiological research to uncover how likely concussions are
to cause gait deficits. This research could also account for other standard assessment
techniques and, in doing so, better inform clinicians about a comprehensive assess-
ment for concussion management.
Ideally clinicians could accurately measure all cognitive and motor deficits
following a concussion both quickly and cost-effectively. Unlike cognitive testing,
gait assessment is neither quick nor cheap. Individual-specific results are required in
a clinical setting, so baseline information and multiple testing post-concussion are
crucial for diagnostic comparisons. Future research needs to advance gait assessment
techniques so that they can be implemented in a quick and cost-effective manner.
Mobile technology, through inertial measurement units (IMUs) or force platforms,
could provide a route as technology advances for more cost-effective precise mea-
surement advances. Research is also exploring augmented reality, which may refine
our ability to provide more realistic dual-task scenarios during gait in the lab or in the
clinic.
Simultaneously, researchers need to direct efforts toward correlations of attention
components to gait performance to refine dual-task paradigms. Concussions are a
neurophysiological phenomenon currently tested via cognitive assessment, but
motor deficits are often as crucial to a return to normal activity as cognitive deficits
as they result in subsequent injuries (Brooks et al. 2016; Herman et al. 2015). If
research advances in predicting gait performance from cognitive assessment results,
more costly and time-consuming motor tests may be avoided altogether, but as of
now, the perfect cognitive test to make this motor performance prediction has yet to
be created.
A team approach is crucial in both developing new measurement techniques and
refining current techniques. The partnership between a clinician and a gait researcher
in writing this chapter highlighted to us the importance for both to be equally
involved in future collaborative research into concussion assessment during gait.
One such difficulty is the use of terminology between the fields and roles. Clinicians
have abandoned the idea of “grading” concussions or labeling them as mild versus
severe, while researchers designing experiments may continue to divide subjects into
those suffering from “mild” and “severe” concussions. This difference in terminol-
ogy can make translation of research results to clinical practice more difficult. Work
must continue with both sides collaborating together in a common language with the
patients’ health and well-being as the focus.
References
Aligene K, Lin E (2013) Vestibular and balance treatment of the concussed athlete. NeuroReh-
abilitation 32:543–553. doi:10.3233/nre-130876
Alsalaheen BA et al (2010) Vestibular rehabilitation for dizziness and balance disorders after
concussion. J Neuro Phys Ther 34:87–93. doi:10.1097/NPT.0b013e3181dde568
Alsalaheen BA, Whitney SL, Marchetti GF, Furman JM, Kontos AP, Collins MW, Sparto PJ (2016)
Relationship between cognitive assessment and balance measures in adolescents referred for
vestibular physical therapy after concussion. Clin J Sport Med 26:46–52. doi:10.1097/
jsm.0000000000000185
Baker CS, Cinelli ME (2014) Visuomotor deficits during locomotion in previously concussed
athletes 30 or more days following return to play. Physiol Reports 2 doi: 10.14814/phy2.12252
Basford JR et al (2003) An assessment of gait and balance deficits after traumatic brain injury. Arch
Bernstein DM (2002) Information processing difficulty long after self-reported concussion. J Int
Neuropsychol Soc 8:673–682
Breton F, Pincemaille Y, Tarriere C, Renault B (1991) Event-related potential assessment of
attention and the orienting reaction in boxers before and after a fight. Biol Psychol 31:57–71
Brooks MA, Peterson K, Biese K, Sanfilippo J, Heiderscheit BC, Bell DR (2016) Concussion
increases odds of sustaining a lower extremity musculoskeletal injury after return to play among
collegiate athletes. Am J Sports Med 44:742–747. doi:10.1177/0363546515622387
Brown LA, Shumway-Cook A, Woollacott MH (1999) Attentional demands and postural recovery:
the effects of aging. J Gerontol 54A:M165–M171
Brown JA, Dalecki M, Hughes C, Macpherson AK, Sergio LE (2015) Cognitive-motor integration
deficits in young adult athletes following concussion. BMC Sports Sci Med Rehabil 7:25.
doi:10.1186/s13102-015-0019-4
Bryan MA, Rowhani-Rahbar A, Comstock RD et al (2016) Sports-and recreation-related concus-
sions in US youth. Pediatrics 138(1), e20154635
Cantu RC (1998) Second-impact syndrome. Clin Sport Med 17(1):37–44
Catena RD, van Donkelaar P, Chou LS (2007a) Altered balance control following concussion is
better detected with an attention test during gait. Gait Posture 25:406–411
Catena RD, van Donkelaar P, Chou LS (2007b) Cognitive task effects on gait stability following
concussion. Exp Brain Res 176:23–31. doi:10.1007/s00221-006-0596-2
Catena RD, van Donkelaar P, Chou L-S (2009a) Different gait tasks distinguish immediate vs. long-
term effects of concussion on balance control. J Neuroeng Rehabil 6:25–25
Catena RD, van Donkelaar P, Halterman CI, Chou LS (2009b) Spatial orientation of attention and
obstacle avoidance following concussion. Exp Brain Res 194:67–77
Catena RD, van Donkelaar P, Chou LS (2011) The effects of attention capacity on dynamic balance
control following concussion. J Neuroeng Rehabil 8:8. doi:10.1186/1743-0003-8-8
Cavanaugh JT, Guskiewicz KM, Stergiou N (2005) A nonlinear dynamic approach for evaluating
postural control: new directions for the management of sport-related cerebral concussion. Sports
Med 35:935–950
CDC (1997) Sports-related recurrent brain injuries – United States. Morb Mortal Wkly Rep
46:224–227
Chan RC (2002) Attentional deficits in patients with persisting postconcussive complaints: a
general deficit or specific component deficit? J Clin Exp Neuropsychol 24:1081–1093
Chan RC, Hoosain R, Lee TM, Fan YW, Fong D (2003) Are there sub-types of attentional deficits in
patients with persisting post-concussive symptoms? A cluster analytical study. Brain Injury
17:131–148
Chiu SL, Osternig L, Chou LS (2013) Concussion induces gait inter-joint coordination variability
under conditions of divided attention and obstacle crossing. Gait Posture 38:717–722.
doi:10.1016/j.gaitpost.2013.03.010
Choe MC, Giza CC (2015) Diagnosis and management of acute concussion. Semin Neurol
35:29–41. doi:10.1055/s-0035-1544243
Chou LS, Kaufman KR, Walker-Rabatin AE, Brey RH, Basford JR (2004) Dynamic instability
during obstacle crossing following traumatic brain injury. Gait Posture 20:245–254.
doi:10.1016/j.gaitpost.2003.09.007
Cock J, Fordham C, Cockburn J, Haggard P (2003) Who knows best? Awareness of divided
attention difficulty in a neurological rehabilitation setting. Brain Inj 17:561–574
Corwin DJ, Wiebe DJ, Zonfrillo MR, Grady MF, Robinson RL, Goodman AM, Master CL (2015)
Vestibular deficits following youth concussion. J Pediatr 166:1221–1225. doi:10.1016/j.
jpeds.2015.01.039
Covassin T, Elbin RJ III, Stiller-Ostrowski JL et al (2009) Immediate post-concussion assessment
and cognitive testing (ImPACT) practices of sports medicine professionals. J Athl Train 44
(6):639–644
Cremona-Meteyard SL, Clark CR, Wright MJ, Geffen GM (1992) Covert orientation of visual
attention after closed head injury. Neuropsychologia 30:123–132
Daffner KR et al (2000) The central role of the prefrontal cortex in directing attention to novel
events. Brain 123:927–939
De Beaumont L et al (2011) Persistent motor system abnormalities in formerly concussed. J Athl
Train 46:234–240
De Beaumont L, Tremblay S, Henry LC, Poirier J, Lassonde M, Theoret H (2013) Motor system
alterations in retired former athletes: the role of aging and concussion history. BMC Neurol
13:109. doi:10.1186/1471-2377-13-109
de Bruin ED, Schmidt A (2010) Walking behaviour of healthy elderly: attention should be paid.
Behavior Brain Funct 6:59. doi:10.1186/1744-9081-6-59
De Monte VE, Geffen GM, May CR, McFarland K, Heath P, Neralic M (2005) The acute effects of
mild traumatic brain injury on finger tapping with and without word repetition. J Clin Exp
Neuropsychol 27:224–239. doi:10.1080/13803390490515766
Deshpande N, Patla AE (2005) Dynamic visual-vestibular integration during goal directed human
locomotion. Exp Brain Res 166:237–247. doi:10.1007/s00221-005-2364-0
Dorman JC, Valentine VD, Munce TA, Tjarks BJ, Thompson PA, Bergeron MF (2015) Tracking
postural stability of young concussion patients using dual-task interference. J Sci Med Sport
18:2–7. doi:10.1016/j.jsams.2013.11.010
Fait P, Swaine B, Cantin JF, Leblond J, McFadyen BJ (2013) Altered integrated locomotor and
cognitive function in elite athletes 30 days postconcussion: a preliminary study. J Head Trauma
Rehabil 28:293–301. doi:10.1097/HTR.0b013e3182407ace
Fife TD, Kalra D (2015) Persistent vertigo and dizziness after mild traumatic brain injury. Ann N Y
Acad Sci 1343:97–105. doi:10.1111/nyas.12678
Gao J, Hu J, Buckley T, White K, Hass C (2011) Shannon and Renyi entropies to classify effects of
mild traumatic brain injury on postural sway. PLoS One 6, e24446. doi:10.1371/journal.
pone.0024446
Geurts AC, Ribbers GM, Knoop JA, Limbeek JV (1996) Identification of static and dynamic
postural instability following traumatic brain injury. Arch Phys Med Rehabil 77:639–644
Grady M (2010) Concussion in the adolescent athlete. Curr Probl Pediatr Adolesc Health Care 40
(7):154–169
Haggard P, Cockburn J, Cock J, Fordham C, Wade D (2000) Interference between gait and
cognitive tasks in a rehabilitating neurological population. J Neurol Neurosurg Psychiatry
69:479–486
Halterman CI, Langan J, Drew A, Rodriguez E, Osternig LR, Chou LS, van Donkelaar P (2006)
Tracking the recovery of visuospatial attention deficits in mild traumatic brain injury. Brain
129:747–753
Hart T, Whyte J, Kim J, Vaccaro M (2005) Executive function and self-awareness of “real-world”
behavior and attention deficits following traumatic brain injury. J Head Trauma Rehabil
20:333–347
Hegeman J, Weerdesteyn V, van den Bemt B, Nienhuis B, van Limbeek J, Duysens J (2012) Dual-
tasking interferes with obstacle avoidance reactions in healthy seniors. Gait Posture
36:236–240. doi:10.1016/j.gaitpost.2012.02.024
Herman DC, Zaremski JL, Vincent HK, Vincent KR (2015) Effect of neurocognition and concus-
sion on musculoskeletal injury risk. Curr Sports Med Rep 14:194–199. doi:10.1249/
jsr.0000000000000157
Hillier SL, Sharpe MH, Metzer J (1997) Outcomes 5 years post-traumatic brain injury (with further
reference to neurophysical impairment and disability). Brain Inj 11:661–675
Howell D, Osternig L, Van Donkelaar P, Mayr U, Chou LS (2013a) Effects of concussion on
attention and executive function in adolescents. Med Sci Sports Exerc 45:1030–1037.
doi:10.1249/MSS.0b013e3182814595
Howell DR, Osternig LR, Chou LS (2013b) Dual-task effect on gait balance control in adolescents
with concussion. Arch Phys Med Rehabil 94:1513–1520. doi:10.1016/j.apmr.2013.04.015
Howell DR, Osternig LR, Chou LS, Chou LS (2015a) Adolescents demonstrate greater gait balance
control deficits after concussion than young adults. Am J Sports Med 43:625–632. doi:10.1177/
0363546514560994
Howell DR, Osternig LR, Chou LS (2015b) Return to activity after concussion affects dual-task gait
balance control recovery. Med Sci Sports Exerc 47:673–680. doi:10.1249/
mss.0000000000000462
Howell DR, Osternig LR, Christie AD, Chou LS (2015c) Return to physical activity timing and
dual-task gait stability are associated 2 months following concussion. J Head Trauma Rehabil 31
(4):262–268. doi:10.1097/htr.0000000000000176
Kaufman KR, Brey RH, Chou LS, Rabatin A, Brown AW, Basford JR (2006) Comparison of
subjective and objective measurements of balance disorders following traumatic brain injury.
Med Eng Phys 28:234–239. doi:10.1016/j.medengphy.2005.05.005
Keightley M et al (2009) Paediatric sports-related mild traumatic brain injury. BMJ Case Rep.
doi:10.1136/bcr.06.2008.0148
Killgore WD, Singh P, Kipman M, Pisner D, Fridman A, Weber M (2016) Gray matter volume and
executive functioning correlate with time since injury following mild traumatic brain injury.
Neurosci Lett 612:238–244. doi:10.1016/j.neulet.2015.12.033
Lajoie Y, Teasdale N, Bard C, Fleury M (1993) Attentional demands for static and dynamic
equilibrium. Exp Brain Res 97:139–144
Larson MJ, Farrer TJ, Clayson PE (2011) Cognitive control in mild traumatic brain injury: conflict
monitoring and conflict adaptation. Int J Psychophysiol 82:69–78. doi:10.1016/j.
ijpsycho.2011.02.018
MacDonald AW 3rd, Cohen JD, Stenger VA, Carter CS (2000) Dissociating the role of the
dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science
288:1835–1838
Maki BE, McIlroy WE (1996) Postural control in the older adult. Clin Geriatr Med 12:635–658
Martini DN, Sabin MJ, DePesa SA, Leal EW, Negrete TN, Sosnoff JJ, Broglio SP (2011) The
chronic effects of concussion on gait. Arch Phys Med Rehabil 92:585–589. doi:10.1016/j.
apmr.2010.11.029
McCrea M et al (2003) Acute effects and recovery time following concussion in collegiate football
players: the NCAA Concussion Study. JAMA 290:2556–2563
McCrory P, Meeuwisse W, Johnston K (2009) Consensus statement on concussion in sport 3rd
international 3rd international conference on concussion in sport held in Zurich, November
2008. Clin J Sport Med 19(3):185–200
McCrory P, Meeuwisse WH, Aubry M et al (2013) Consensus statement on concussion in sport: the
4th international conference on concussion in sport held in Zurich, November 2012. Brit J of
Sport Med 47(5):250–258
McFadyen BJ, Swaine B, Dumas D, Durand A (2003) Residual effects of a traumatic brain injury
on locomotor capacity: a first study of spatiotemporal patterns during unobstructed and
obstructed walking. J Head Trauma Rehabil 18:512–525
McFadyen BJ, Cantin JF, Swaine B, Duchesneau G, Doyon J, Dumas D, Fait P (2009) Modality-
specific, multitask locomotor deficits persist despite good recovery after a traumatic brain injury.
Arch Phys Med Rehabil 90:1596–1606. doi:10.1016/j.apmr.2009.03.010
Moore RD, Hillman CH, Broglio SP (2014) The persistent influence of concussive injuries on
cognitive control and neuroelectric function. J Athl Train 49:24–35. doi:10.4085/1062-6050-
49.1.01
Moore DR, Pindus DM, Raine LB, Drollette ES, Scudder MR, Ellemberg D, Hillman CH (2016)
The persistent influence of concussion on attention, executive control and neuroelectric function
in preadolescent children. Int J Psychophysiol 99:85–95. doi:10.1016/j.ijpsycho.2015.11.010
Mrazik M, Ferrara MS, Peterson CL, Elliott RE, Courson RW, Clanton MD, Hynd GW (2000)
Injury severity and neuropsychological and balance outcomes of four college athletes. Brain Inj
14:921–931
Murray NG, Ambati VN, Contreras MM, Salvatore AP, Reed-Jones RJ (2014) Assessment of
oculomotor control and balance post-concussion: a preliminary study for a novel approach to
concussion management. Brain Inj 28:496–503. doi:10.3109/02699052.2014.887144
Parker TM, Osternig LR, Lee HJ, Donkelaar P, Chou LS (2005) The effect of divided attention on
gait stability following concussion. Clin Biomech (Bristol, Avon) 20:389–395. doi:10.1016/j.
clinbiomech.2004.12.004
Parker TM, Osternig LR, Van Donkelaar P, Chou LS (2006) Gait stability following concussion.
Med Sci Sports Exerc 38:1032–1040. doi:10.1249/01.mss.0000222828.56982.a4
Parker TM, Osternig LR, van Donkelaar P, Chou LS (2008) Balance control during gait in athletes
and non-athletes following concussion. Med Eng Phys 30:959–967. doi:10.1016/j.
medengphy.2007.12.006
Pavol MJ, Owings TM, Grabiner MD (2002) Body segment inertial parameter estimation for the
general population of older adults. J Biomech 35:707–712
Pontifex MB, Broglio SP, Drollette ES, Scudder MR, Johnson CR, O'Connor PM, Hillman CH
(2012) The relation of mild traumatic brain injury to chronic lapses of attention. Res Q Exerc
Sport 83:553–559. doi:10.1080/02701367.2012.10599252
Posner MI (1980) Orienting of attention. Q J Exp Psychol 32:3–25
Posner MI, Petersen SE (1990) The attention system of the human brain. Annu Rev Neurosci
13:25–42
Posner MI, Rothbart MK (1998) Attention, self-regulation and consciousness. Phil Trans R Soc
Lond B Biol Sci 353:1915–1927
Posner MI, Walker JA, Friedrich FJ, Rafal RD (1984) Effects of parietal injury on covert orienting
of attention. J Neurosci 4:1863–1874
Powers KC, Kalmar JM, Cinelli ME (2014) Dynamic stability and steering control following a
sport-induced concussion. Gait Posture 39:728–732. doi:10.1016/j.gaitpost.2013.10.005
Quatman-Yates CC, Bonnette S, Hugentobler JA, Mede B, Kiefer AW, Kurowski BG, Riley MA
(2015) Postconcussion postural sway variability changes in youth: the benefit of structural
variability analyses. Pediatr Phys Ther 27:316–327. doi:10.1097/pep.0000000000000193
Rahn C, Munkasy BA, Barry Joyner A, Buckley TA (2015) Sideline performance of the balance
error scoring system during a live sporting event clinical. J Sport Med 25:248–253. doi:10.1097/
jsm.0000000000000141
Register-Mihalik JK, Guskiewicz KM, McLeod TC et al (2013) Knowledge, attitude, and
concussion-reporting behaviors among high school athletes: a preliminary study. J Athl Train
48(5):645–653
Resch J et al (2013) ImPact test-retest reliability: reliably unreliable? J Athl Train 48:506–511.
doi:10.4085/1062-6050-48.3.09
Sambasivan K, Grilli L, Gagnon L (2015) Balance and mobility in clinically recovered children and
adolescents after a mild traumatic brain injury. J Pediatr Rehabil Med 8:335–344. doi:10.3233/
prm-150351
Schatz P, Moser R (2011) Current issues in pediatric sports concussion. Clin Neuropsychol 25
(6):1042–1057
Serino A, Ciaramelli E, Di Santantonio A, Malagù S, Servadei F, Làdavas E (2006) Central
executive system impairment in traumatic brain injury. Brain Inj 20:23–32
Sosnoff JJ, Broglio SP, Ferrara MS (2008) Cognitive and motor function are associated following
mild traumatic brain injury. Exp Brain Res 187:563–571. doi:10.1007/s00221-008-1324-x
Sosnoff JJ, Broglio SP, Shin S, Ferrara MS (2011) Previous mild traumatic brain injury and
postural-control dynamics. J Athl Train 46:85–91. doi:10.4085/1062-6050-46.1.85
Sturm W et al (1999) Functional anatomy of intrinsic alertness: evidence for a fronto-parietal-
thalamic-brainstem network in the right hemisphere. Neuropsychologia 37:797–805
Swick D, Jovanovic J (2002) Anterior cingulate cortex and the stroop task: neuropsychological
evidence for topographic specificity. Neuropsychologia 40:1240–1253
Szturm T, Maharjan P, Marotta JJ, Shay B, Shrestha S, Sakhalkar V (2013) The interacting effect of
cognitive and motor task demands on performance of gait, balance and cognition in young
adults. Gait Posture 38:596–602. doi:10.1016/j.gaitpost.2013.02.004
Tapper A, Gonzalez D, Roy E, Niechwiej-Szwedo E (2016) Executive function deficits in team
sport athletes with a history of concussion revealed by a visual-auditory dual task paradigm.
J Sports Sci 1–10 doi:10.1080/02640414.2016.1161214
Theye F, Mueller KA (2004) “Heads up”: concussions in high school sports. Clin Med Res
2(3):165–171
Vallee M, McFadyen BJ, Swaine B, Doyon J, Cantin JF, Dumas D (2006) Effects of environmental
demands on locomotion after traumatic brain injury. Arch Phys Med Rehabil 87:806–813.
doi:10.1016/j.apmr.2006.02.031
van Donkelaar P, Langan J, Rodriguez E, Drew A, Halterman C, Osternig LR, Chou L-S (2005)
Attentional deficits in concussion. Brain Inj 19:1031–1039
van Donkelaar P, Osternig LR, Chou L-S (2006) Attentional and biomechanical deficits interact
after mild traumatic brain injury. Exerc Sport Sci Rev 34:77–82
Vandenberghe R, Gitelman DR, Parrish TB, Mesulam MM (2001) Functional specificity of superior
parietal mediation of spatial shifting. Neuroimage 14:661–673
Vaportzis E, Georgiou-Karistianis N, Churchyard A, Stout JC (2015) Effects of task difficulty
during dual-task circle tracing in Huntington's disease. J Neurol 262:268–276. doi:10.1007/
s00415-014-7563-9
Vilkki J, Virtanen S, Surma-Aho O, Servo A (1996) Dual task performance after focal cerebral
lesions and closed head injuries. Neuropsychologia 34:1051–1056
Wang J, Rao H, Wetmore GS, Furlan PM, Korczykowski M, Dinges DF, Detre JA (2005) Perfusion
functional MRI reveals cerebral blood flow pattern under psychological stress. Proc Natl Acad
Sci U S A 102:17804–17809
Weerdesteyn V, Schillings AM, van Galen GP, Duysens J (2003) Distraction affects the perfor-
mance of obstacle avoidance during walking. J Mot Behav 35:53–63
Wilkins AJ, Shallice T, McCarthy R (1987) Frontal lesions and sustained attention. Neuropsy-
chologia 25:359–365
Yantis S, Schwarzbach J, Serences JT, Carlson RL, Steinmetz MA, Pekar JJ, Courtney SM (2002)
Transient neural activity in human parietal cortex during spatial attention shifts. Nat Neurosci
5:995–1002
Yardley L, Gardner M, Bronstein A, Davies R, Buckwell D, Luxon L (2001) Interference between
postural control and mental task performance in patients with vestibular disorder and healthy
controls. J Neurol Neurosurg Psychiatry 71:48–52
Zatsiorsky VM, King DL (1998) An algorithm for determining gravity line location from
posturographic recordings. J Biomech 31:161–164
Zhang L, Abreu BC, Gonzales V, Huddleston N, Ottenbacher KJ (2002) The effect of predictable
and unpredictable motor tasks on postural control after traumatic brain injury. NeuroReh-
abilitation 17:225–230
Functional Dystonias
Jessica Pruente and Deborah Gaebler-Spira
Abstract
Dystonia is a movement disorder characterized by involuntary muscle contrac-
tions resulting in twisting movements and abnormal postures. This movement
disorder can cause significant impairments during functional tasks including gait,
mobility, and reaching. Dystonia must be distinguished from other hypertonic
movement disorders, spasticity, and rigidity, to guide treatment and management
options. Several clinical measurement scales have been developed to identify
dystonia and rate its severity; these can be easily adapted for use in motion
analysis labs. Additionally, the use of motion analysis kinetics, kinematics, and
surface EMG has increased in use for monitoring dystonia. This chapter will
discuss the common etiologies of dystonia, clinical scales used for diagnosis and
efficacy of treatments, and the role of instrumented gait analysis, kinetics, and
kinematics in the evaluation of dystonia.
Keywords
Barry-Albright Dystonia Scale • Co-contraction • Dystonia • Gait analysis •
Hyperkinetic • Hypertonia asessment tool • Involuntary movements • Kinemat-
ics • Motion analysis • Movement disorder • Overflow muscle activation • Spasms
Electronic supplementary material: Supplementary material is available in the online version of

this chapter at ▶ 10.1007/978-3-319-30808-1_70-1. Videos can also be accessed at http://www.
springerimages.com/videos/.
J. Pruente (*) • D. Gaebler-Spira

Shirley Ryan Ability Lab, Chicago, IL, USA
e-mail: jessica.pruente@gmail.com; dgaebler@sralab.org

DOI 10.1007/978-3-319-30808-1_70-1
2 J. Pruente and D. Gaebler-Spira
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Pathology and Functional Effects of Dystonia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Etiologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Clinical Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Kinetics/Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Summarizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Introduction
Dystonia is a movement disorder characterized by involuntary muscle contractions

resulting in twisting movements and abnormal postures. Dystonia is often initiated
or worsened by volitional activity and is associated with overflow muscle activation.
Dystonia is highly variable but can impair a person’s function in walking, hand
manipulation, and speech due to reduced quality and speed of voluntary movement
(van der Kamp et al. 1989; Agostino et al. 1992; Inzelberg et al. 1995; Curra et al.
2000; Gregori et al. 2008).
Utilization of motion analysis for gait or arm function is particularly helpful
because dystonia is characterized as a movement disorder and also impacts muscle
tone (Sanger 2015). The motion analysis laboratory captures data from various
perspectives; this fosters integration of a methodical clinical examination, video
review of gait, kinetics and kinematics, motion trajectories, muscle activation pat-
terns, and co-contractions.
In this chapter, we will review the definition of dystonia, the clinical tests
discriminating dystonia from spasticity, severity rating scales for dystonia, etiolo-
gies, the concepts of co-contraction and dynamic motor control, as well as the kinetic
and kinematic characteristics of functional dystonias. Though dystonia can be easily
and commonly recognized during the review of motion analysis, little literature
exists to provide a conclusive approach to reporting dystonia during motion analysis.
Recommendations for inclusion of clinical scales, close observation of the video,
as well as known gait characteristics of dystonia will be discussed for consideration.
Motion analysis provides a unique quantifiable understanding of both motor control
and the impact of dystonia on function.
State of the Art
Information obtained through motion analysis provides insights into the pathology
and functional effects of dystonia. In children with cerebral palsy, the use of motion
analysis is frequent for planning interventions based on gait deviations or functional
improvements for hand use. Though no published motion analysis criteria for
Functional Dystonias 3
dystonia exists, common findings reported between children and adults include a
high variability in step length and base of support. Surface electromyography data
likewise suggests diagnosis of dystonia through co-contraction, overflow muscle
activity, and increased muscle activity during volitional tasks. Further advancements
in the use of formal motion analysis are needed to improve confidence in diagnosis
and treatment, either surgical or medical, for dystonia.
Pathology and Functional Effects of Dystonia
Definitions
Dystonic syndromes are some of the more commonly observed movement disorders
with an estimated prevalence of 2–50 per million in early-onset dystonia and
30–7,320 per million in late-onset dystonias (Carecchio et al. 2015). Secondary
dystonia has been increasingly recognized in children with cerebral palsy. Since
proper treatment of dystonia is available, it is crucial to be able to identify the
movement disorder and the confounding hypertonic muscle abnormalities frequently
coexisting in children with cerebral palsy.
Dystonia is defined as a movement disorder in which involuntary sustained or
intermittent muscle contractions cause twisting and repetitive movements, abnormal
postures, or both. Dystonia falls into the category of hypertonic and hyperkinetic
movement disorders, see Fig. 1. It is frequently worsened with voluntary activity and
may be reduced or absent when at rest. Various organization taxonomies have been
proposed to categorize movement disorders. One such schema breaks disorders into
those of tone, inhibition, execution, and planning of movements (Sanger 2003).
Dystonia is therefore both a disorder of tone and of execution.
Dystonia can have many clinical expressions; this includes dystonic spasms,
tremor, repetitive movements, abnormal fixed postures, and hypertonia. An impor-
tant distinction is delineating the difference between dystonia, spasticity, and rigid-
ity. Spasticity refers to hypertonia in which resistance to movement increases with
speed of stretch and varies with direction of movement. Rigidity is resistance to
movement that does not depend upon movement speed or direction. Dystonia is
more complex to define and includes resistance to joint movements that do not
depend on speed, co-contraction of agonist, and antagonist muscle groups and is
worsened by voluntary activity. Overflow movements may also suggest the presence
of dystonia.
Dystonia quite often presents in similar patterns despite a wide range of diagnoses
and etiologies. One example is neck or back extension, variable scoliosis or kypho-
sis, ulnar wrist deviation and flexion, and finger flexion or extension (Sanger 2004).
In the lower extremity, dystonic posturing often includes knee extension,
plantarflexion, and inversion of the foot. Common dystonic syndromes include
hand cramps, blepharospasm, torticollis, opisthotonus, and more generalized
dystonias involving multiple extremities and the torso.
Hypertonic Hyperkinetic Negative
Spasticity Chorea Weakness
Dystonia Dystonia Selective Motor Control
Rigidity Athetosis Ataxia
Myoclonus Dyspraxia
Tremor
Bradykinesia
Tics
Balance
Stereotyples
Fig. 1 Positive and negative symptoms of hypertonia. Dystonia is associated with both hypertonic
and hyperkinetic movements
Treatment and management of these various hyperkinetic and hypertonic move-

ment disorders is very much dependent upon specific type. Certain medications,
such as trihexyphenidyl, and surgical procedures, such as deep brain stimulation,
have better success rates with dystonia.
Etiologies
Early-onset dystonias refer to presentation prior to age 26. The dystonic movement
disorders can be further divided into subsets of inherited and acquired dystonias.
DYT1, also known as Oppenheim’s dystonia, is the most common of the inherited
early-onset dystonias, with a worldwide frequency of 1:160,000 cases (Carecchio
et al. 2015; de Carvalho Aguiar and Ozelius 2002). A GAG deletion in the TOR1A
gene was identified as the cause of DYT1 dystonia. DYT1 is an autosomal dominant
trait with reduced penetrance of about 30%. Clinical presentation typically occurs by
age 12 with involvement of a single extremity; this generalizes to the remainder of
extremities and the trunk in about 50% of cases within a few years. Treatment
remains symptomatic with oral medications and in select refractory cases with
deep brain stimulation. DYT 2 is an autosomal recessive trait of unknown genetic
etiology. It is an early-onset dystonia; DYT6 dystonia is also an autosomal dominant
inherited form of dystonia. This involves a variety of different mutations in the
THAP1 gene. DYT6 dystonia also presents in childhood through adolescence.
Clinical presentation typically involves oromandibular, craniocervical, or laryngeal

dystonias.
The second major subset of dystonias are secondary or acquired dystonias. The
most common etiology is related to acquired brain injuries. Dystonia is second only
to post-traumatic tremor in the movement disorder sequelae of severe traumatic brain
injury (Krauss and Jankovic 2002; Sanger 2015). It is thought to be related to
involvement of the basal ganglia, caudate, and putamen, and there is some evidence
to suggest cerebellar and thalamic involvement (Skogseid 2014). Traumatic and
hypoxic brain injuries can lead to acquired dystonias in adults. The PAID syndrome,
or paroxysmal autonomic instability and dystonia syndrome, occurs after hypoxic
injuries, such as those sustained in cardiac arrest. In children, the most common
cause of acquired dystonia is cerebral palsy, and in fact, dyskinetic cerebral palsy is
the second largest CP type (Monbaliu et al. 2016). Dystonia may be related to
hypoxic ischemic injuries or prematurity in these cases. Other etiologies include
autoimmune disorders such as anti-N-methyl-D-aspartate receptor (NMDAR)
encephalitis and autoimmune basal ganglia encephalitis (van Egmond et al. 2015).
Dystonia can also be induced by a variety of drugs and toxins including levodopa,
dopamine antagonists, selective serotonin reuptake inhibitors, buspirone, cocaine,
monoamine oxidase inhibitors, carbon monoxide, manganese, and cyanide (Phukan
et al. 2011). Inborn errors of metabolism may also cause dystonic movement
disorders; if not treated early, this can be found in organic acidurias, glut-1 defi-
ciency, and lysosomal storage disorders.
A third category of dystonia refers to dystonia-plus syndromes. Dystonia-plus
syndromes are dystonic syndromes that also include other neurological complaints
most commonly parkinsonism or myoclonus. Examples of this include dopa-
responsive dystonia, rapid-onset dystonia parkinsonism, and myoclonus dystonia
syndrome. These probably represent just 5% of childhood-onset dystonias.
Clinical Scales
Appropriate diagnosis of dystonia can be difficult to achieve owing to the presence

of coexisting movement disorders and spasticity in many of those affected. There are
several clinical scales in use today to aid in the measurement of dystonia. Clinical
scales were first utilized for primary dystonias. The Burke Fahn Marsden (BFM)
scale, published in 1985, was the first utilized scale for rating generalized dystonia,
hemidystonia, and segmental dystonia (Krystkowiak et al. 2007). The BFM dystonia
scale evaluates the presence of dystonia in nine body regions, eyes, mouth, neck
trunk, and right and left arms and legs. This scale identifies both provoking factors
and severity factors in each region and includes a separate disability rating. Scores
range from 0 to 120, with higher scores indicating more severe dystonia. This scale
has demonstrated good inter- and intra-rater reliability. This scale remains in use in
both clinical and research setting to track dystonia over time and in response to
treatments.
As secondary dystonias and available treatments have evolved, the need to

discriminate dystonia from other hypertonic syndromes led to the development of
additional measurement scales. The Barry–Albright Dystonia Scale was developed
to improve reliability and measurement of secondary dystonias (Barry et al. 1999;
Pavone et al. 2013). Assessment of dystonia in this population can be more difficult
secondary to presence of coexisting brain injury or cognitive impairment. This scale
consists of assessing secondary dystonia in eight regions, eyes, mouth, neck, trunk,
and the four extremities. Scores range from 0 to 32, and higher scores indicate more
severe dystonia. Finally, the Unified Dystonia Rating Scale measures severity and
duration of dystonia in 14 body regions, eyes and upper face, lower face, jaw and
tongue, larynx, neck, trunk, shoulder/proximal arm, distal arm/hand, proximal leg,
and distal leg/foot. This scale has high internal consistency and inter-rater reliability
in primary dystonias (Goetz et al. 2008; Monbaliu et al. 2010).
In children, assessment of movement disorders can be more difficult owing to
difficulty with exam instructions. Several scales have been developed to address this
group. The hypertonia assessment tool or HAT is one such scale. This scale was
developed to help differentiate the different types of hypertonia; dystonia, spasticity,
and rigidity (Jethwa et al. 2010; Pavone et al. 2013). The HAT is a seven-item
clinical assessment tool designed for children aged 4–19. It involves a binary rating
scale for three presentations of dystonia, two of spasticity and two of rigidity. This
scale has fair inter-rater reliability and moderate test-retest reliability for dystonia.
The Movement Disorder-Childhood Rating Scale was likewise designed for clinical
evaluation of movement disorders in ages 4–19 (Battini et al. 2015). The movement
disorder portion of this scale assesses dystonia at rest and during functional tasks in
the eye/orbital region, face, neck, perioral, trunk, upper extremities, and lower
extremities. The childhood rating scale includes assessment of motor function
(including head control, sitting position, standing position, walking, reaching,
grasping, and handwriting), oral/verbal function, self-care, and attention/alertness.
This has been demonstrated to have high inter-rater reliability and high internal
consistency.
Dystonia can be difficult to measure accurately given its variable presentation,
dynamic changes, and coexistence with other movement disorders. Newer therapeu-
tic options including intrathecal baclofen and deep brain stimulation have increased
the widespread utilization of these scales in order to ensure appropriate patient
selection and to monitor response to treatment. New scales continue to be developed
in response to the need for improved measurement of dystonia in response to
treatments. The dyskinesia impairment scale was more recently developed and
includes two subscales, for dystonia and choreoathetosis, and evaluates both dura-
tion and amplitude (Elegast Monbaliu et al. 2012). This was demonstrated to have
good inter-rater reliability and internal consistency in initial studies (Fig. 2).
HYPERTONIA ASSESSMENT TOOL (HAT)
SCORE TYPE OF
SCORING GUIDELINES 0=negative HYPERTONIA
HAT ITEM (0=negative or 1=positive) 1=positive
(circle score)
1. Increased involuntary 0= No involuntary movements or postures
0
movements/postures of the observed
designated limb with tactile stimulus DYSTONIA
1= Involuntary movements or postures observed 1
of a distal body part
2. Increased involuntary 0= No involuntary movements or postures 0
movements/postures with purposeful observed
movements of a distal body part 1 DYSTONIA
1= Involuntary movements or postures observed
0= No increased resistance noticed during fast 0
3. Velocity dependent resistance to stretch compared to slow stretch
stretch 1= Increased resistance noticed during fast 1 SPASTICITY
stretch compared to slow stretch
0= No spastic catch noted 0
4. Presence of a spastic catch
1= Spastic catch noted SPASTICITY
1
0= Equal resistance not noted with bi-directional
5. Equal resistance to passive stretch 0
movement
during bi-directional movement of a
1= Equal resistance noted with bi-directional RIGIDITY
joint 1
movement
0= No increased tone noted with purposeful 0
6. Increased tone with movement of
movement
a distal body part DYSTONIA
1= Greater tone noted with purposeful movement 1
0= Limb returns (partially or fully) to original 0
7. Maintenance of limb position position
after passive movement RIGIDITY
1= Limb remains in final position of stretch 1
Fig. 2 The hypertonia assessment tool (Fehlings et al. 2010)
Kinetics/Kinematics
While clinical scales can aid the clinician in diagnosis of dystonia, objective
measurement using gait labs, kinetics, and kinematics also has an important role in
dystonia assessment and determining therapeutic effect. Animal models were first
used to characterize gait changes associated with dystonia. In a rat model of
dystonia, decreased walking speed, increased hind limb spread, and increased step
length ratio variability were consistent with dystonia. Of these, step length ratio
variability was the most sensitive for detecting dystonia (Chaniary et al. 2009).
Further data in rat models of dystonia and ataxia also demonstrate coactivation of
muscle and similar changes to gait parameters (Scholle et al. 2010).
Clinical studies in pediatrics are somewhat limited with the largest amount of data
pertaining to upper extremity dystonias. Kinematics data have been collected in
children with cerebral palsy during reach and grasp activities. Those with dystonia
have slower movements during reaching and decreased coordination of movement.
Different kinematics were obtained in the three CP subtypes, spastic, dystonic, and
mixed, and may be useful in distinguishing between movement disorders (Butler
et al. 2010; Kukke et al. 2015). The kinematic dystonia measure collects kinematic
data during an upper extremity finger tapping task has been demonstrated to
correlate with the Barry–Albright Dystonia Scale and may improve quantitative
assessment of dystonia (Kawamura et al. 2012). A rest-tap test involving repeated
tapping in one limb during assessment of the contralateral limb for dystonia
Fig. 3 Dystonia kinematics. Displacement of the shoulder, elbow, and wrist joints over time during
the rest-tap paradigm. (a) represents a control subject. (b) represents a cerebral palsy subject with
low dystonia. (c) represents a cerebral palsy subject with high dystonia (Figure reproduced with
permission from Pediatric Neurology, Can Spasticity and Dystonia Be Independently Measured in
Cerebral Palsy (Gordon et al. 2006))
demonstrates a different kinematic pattern compared to controls. This is character-

ized by overflow in the contralateral limb and increased joint excursion, see Fig. 3
(Gordon et al. 2006).
Gait analysis in children that includes surface EMG suggests that co-contraction
and increased resistance to external motion and slow velocities are present with
dystonia (Lebiedowska et al. 2004). Dystonia presents in EMG data as an increased
number of muscles responding during volitional activity but did not respond at rest
or during quick stretch. This is in contrast to spasticity, which demonstrated brief
bursts of EMG activity during quick stretch, but low activity levels during rest or
volitional activity. A specific electromyographic protocol suggests that lower
extremity assessment during rest, quick stretch, and five volitional tasks can detect
different muscle activation patterns in spasticity and dystonia and might provide
objective data for diagnosis (Beattie et al. 2016). Please see the example videos for
two demonstrations of dystonic gait (online only).
Figure 4 demonstrates EMG tracings during motion analysis that can help to
distinguish dystonia from spasticity (Reference: personal communication with
Dr. Jules Becher). Dystonia can be inferred by careful analysis of EMG patterns as
well as by kinematic evidence of posturing of extremities during motion analysis.
With respect to the electromyographic patterns, typically in dystonia, the raw or
rectified tracing displays a peak and valley pattern or an inconsistent pattern of
activation. This is in contrast to the muscle patterns of spasticity that are more
consistent with the amplitude of the electromyographic signal remaining constant
(see Fig. 4b).
In adults, head and neck kinematics is useful for accurate description of severity
of cervical dystonia as a baseline for treatment effects. Fastrack allows the extraction
of kinematic information (i.e., posture, angular range of motion, movement times,
angular velocity) about head deviations (Galardi et al. 2003; Jordan et al. 2000). As
a
L Anterior Tibialis
2000
mV
–2000
L Gastrocnemius
2000
mV
–2000
L Rectus Femoris
2000
mV
–2000
b
L Anterior Tibialis
1000
mV
–1000
L Gastrocnemius
1000
mV
–1000
L Rectus Femoris
1000
mV
–1000
Fig. 4 (continued)
c
L Anterior Tiabialis
1000
mV
–1000
L Gastrocnemius
1000
mV
–1000
L Rectus Femoris
1000
mV
–1000
Fig. 4 EMG patterns obtained during motion analysis using surface EMG pads. (a) tracing shows
the anterior tibialis, gastrocnemius, and rectus femoris in a child with mild dystonia. (b) tracing
shows the same muscles in a more typical pattern for spasticity. (c) tracing shows an example of
coexisting spasticity and dystonia
in dystonic gait, head and neck asymmetry or the symmetry index is increased for
rotation and lateral flexion to those with dystonia compared to those unaffected
(Boccagni et al. 2008).
The use of lower extremity motion analysis for adults with dystonia has been used
to evaluate and document effects of treatment for disorders characterized by dysto-
nia. In a patient with dopamine-responsive dystonia, 3D motion analysis accurately
documented changes in gait with reduction of dystonia by medication. When the
dystonia was reduced, the gait pattern demonstrated an increase in the walking
speed, explained by a significant increase in step frequency and length. With
improvement of dystonia, the asymmetry decreased, as did the step width. The
gait analysis allowed clinicians to quantify the effects of dystonia on gait (Rebour
et al. 2015).
MAC for an adult patient with DYT-1 before and after DBS determined which
involuntary movements were related to postural instability and gait disturbance.
Neck and trunk markers add value and allow discrimination of the cervical dystonic
posture, on balance or gait. Prior to DBS, posture and gait were asymmetrical and
unstable. Functional body balance was controlled by changes of symptoms, with
partial corrections of neck and spinal alignments in a static posture. The patient was
better able to maintain the stability of center of mass and center of pressure. The neck
angles remained abnormal with specific motions during gait compared to the spine
while maintaining improved gait. Functional improvements of gait were captured by
gait parameters including increasing of cadence (step rate) and walking speed,
increased step length, reduction of a wide base, and extension of single support
time and symmetry (Nakao 2011). These two case reports illustrate the power of
motion analysis in capturing the effects of dystonia on gait rather than clinical
measures of speed and distance tests.
Conclusion
Motion analysis assists with planning surgical decisions and in establishing energy
cost of walking. When dystonia is present on motion analysis data, this information
informs the surgeon and could be useful with surgical decision-making. Whereas
predictable outcomes are likely with spastic gait patterns, the child with dystonia will
have more variation in surgical outcomes. Thus far, the published use of motion
analysis in adults most often documents the treatment effects of various
interventions.
The physical therapy evaluation at the time of the motion analysis typically
includes ROM, strength, and estimate of spasticity. By including the HAT and a
severity rating of dystonia the clinical picture can then be associated with the
biomechanics of gait. The video review portion of gait analysis is critical as dystonia
will be apparent during gait with atypical trunk, arm, and hand postures identified. A
severity rating can then be validated from the PT evaluation. Identifying dystonia by
motion analysis can theoretically quantify and validate severity rating of dystonia
(Sanger et al. 2010). Including the HAT and severity rating as a routine part of the
clinical motion analysis is a good first step in validating the typical findings of
asymmetry, variability, and EMG firing patterns reported in children with dystonia.
With dystonia in the motion analysis lab, common findings reported in children
and adults include a high variability in step length and base of support. Surface
electromyography data likewise suggests diagnosis of dystonia through
co-contraction, overflow muscle activity, and increased muscle activity during
volitional tasks. Formal motion analysis has the potential to accurately quantify
the motion deviation and determine changes of movement trajectories following
treatment. Motion analysis for clinical and research practice promises quantifiable
insight into the neural mechanisms of hypertonia and hyperkinetic syndromes during
functional tasks.
Summarizing
• Dystonia is a common hypertonic and hyperkinetic movement disorder that can

have profound impact on gait and upper extremity function.
• Diagnosis of dystonia relies upon visual observations of gait and functional tasks,
dystonia severity rating scales, and regimented physical examination.
• Distinguishing dystonia from other movement disorders, especially spasticity,
informs treatment and management decisions.
• Characteristic changes in motion analysis associated with dystonia include vari-
able step length and base of support, muscle co-contraction, overflow muscle
activity during functional tasks, and increased muscle activity during volitional
tasks.
• Motion analysis laboratories may play more of a role in the future in diagnosing
dystonia, assessing treatment effects, and in surgical/treatment planning.
References
Agostino R, Berardelli A, Formica A, Accornero N, Manfredi, M (1992). Sequential arm
movements in patients with Parkinson’s disease, Huntington’s disease and dystonia. Brain 115
(Pt 5):1481–1495
Barry MJ, Van Swearingen JM, Albright AL (1999) Reliability and responsiveness of the Barry-
Albright dystonia scale. Dev Med Child Neurol. 41(6), 404–411
Battini R, Olivieri I, Di Pietro R, Casarano M, Sgandurra G, Romeo DM, Cioni G (2015)
Movement disorder-childhood rating scale: a sensitive tool to evaluate movement disorders.
Pediatr Neurol 53(1):73–77. doi:10.1016/j.pediatrneurol.2015.02.014
Beattie C, Gormley M, Wervey R, Wendorf H (2016) An electromyographic protocol that distin-
guishes spasticity from dystonia. J Pediatr Rehabil Med 9(2):125–132. doi:10.3233/PRM-
160373
Boccagni C, Carpaneto J, Micera S, Bagnato S, Galardi G (2008) Motion analysis in cervical
dystonia. Neurol Sci 29(6):375–381. doi:10.1007/s10072-008-1033-z
Butler EE, Ladd AL, Lamont LE, Rose J (2010) Temporal-spatial parameters of the upper limb
during a Reach & Grasp Cycle for children. Gait Posture 32(3):301–306. doi:10.1016/j.
Carecchio M, Zorzi G, Nardocci N (2015) Inherited isolated dystonia in children. J Pediatr Neurol
13(04):174–179. doi:10.1055/s-0035-1558863
Chaniary KD, Baron MS, Rice AC, Wetzel PA, Ramakrishnan V, Shapiro SM (2009) Quantification
of gait in dystonic Gunn rats. J Neurosci Methods 180(2):273–277. doi:10.1016/j.
jneumeth.2009.03.023
Curra A, Berardelli A, Agostino R, Giovannelli M, Koch G, Manfredi M (2000). Movement cueing
and motor execution in patients with dystonia: a kinematic study. Mov Disord 15(1):103–112
de Carvalho Aguiar PM, Ozelius LJ (2002) Classification and genetics of dystonia. Lancet Neurol 1
(5):316–325. doi:10.1016/s1474-4422(02)00137-0
Fehlings, D., Switzer, L, Jethwa, A, Mink, J, Macarthur, C, Knights, S, & Fehlings, T. (2010).
Hypertonia assessment tool: user manual p.1–10
Galardi G, Micera S, Carpaneto J, Scolari S, Gambini M, Dario P (2003) Automated assessment of
cervical dystonia. Mov Disord 18(11):1358–1367. doi:10.1002/mds.10506
Goetz CG, Nutt JG, Stebbins GT (2008) The unified dyskinesia rating scale: presentation and
clinimetric profile. Mov Disord 23(16):2398–2403. doi:10.1002/mds.22341
Gordon LM, Keller JL, Stashinko EE, Hoon AH, Bastian AJ (2006) Can spasticity and dystonia be
independently measured in cerebral palsy? Pediatr Neurol 35(6):375–381. doi:10.1016/j.
pediatrneurol.2006.06.015
Gregori B, Agostino R, Bologna M, Dinapoli L, Colosimo C, Accornero N, Berardelli A (2008).
Fast voluntary neck movements in patients with cervical dystonia: a kinematic study before and
after therapy with botulinum toxin type A. Clin Neurophysiol. 119(2):273–280. doi:10.1016/j.
clinph.2007.10.007
Inzelberg R, Flash T, Schechtman E, Korczyn AD (1995). Kinematic properties of upper limb
trajectories in idiopathic torsion dystonia. J Neurol Neurosurg Psychiatry. 58(3):312–319
Jethwa A, Mink J, Macarthur C, Knights S, Fehlings T, Fehlings D (2010) Development of the
hypertonia assessment tool (HAT): a discriminative tool for hypertonia in children. Dev Med
Child Neurol 52(5):e83–e87
Jordan K, Dziedzic K, Jones PW, Ong BN, Dawes PT (2000) The reliability of the three-
dimensional FASTRAK measurement system in measuring cervical spine and shoulder range
of motion in healthy subjects. Rheumatology (Oxford) 39(4):382–388
Kawamura A, Klejman S, Fehlings D (2012) Reliability and validity of the kinematic dystonia
measure for children with upper extremity dystonia. J Child Neurol 27(7):907–913.
doi:10.1177/0883073812443086
Krauss JK, Jankovic J (2002) Head injury and posttraumatic movement disorders. Neurosurgery 50
(5):927–939.
Krystkowiak P, du Montcel ST, Vercueil L, Houeto JL, Lagrange C, Cornu P, ... Group S (2007)
Reliability of the Burke-Fahn-Marsden scale in a multicenter trial for dystonia. Mov Disord 22
(5):685–689. doi:10.1002/mds.21392
Kukke S, Curatalo L, de Campos A, Hallett M, Alter K, Damiano D (2015) Coordination of reach-
to-grasp kinematics in individuals with childhood-onset dystonia due to hemiplegic cerebral
palsy. IEEE Trans Neural Syst Rehabil Eng. doi:10.1109/tnsre.2015.2458293
Lebiedowska MK, Gaebler-Spira D, Burns RS, Fisk JR (2004) Biomechanic characteristics of
patients with spastic and dystonic hypertonia in cerebral palsy. Arch Phys Med Rehabil 85
(6):875–880
Monbaliu E, Ortibus E, Roelens F, Desloovere K, Deklerck J, Prinzie P, ... Feys H (2010) Rating
scales for dystonia in cerebral palsy: reliability and validity. Dev Med Child Neurol 52
(6):570–575. doi:10.1111/j.1469-8749.2009.03581.x
Monbaliu E, Ortibus ELS, de Cat JOS, Dan B, Heyrman L, Prinzie P, ... Feys H (2012) The
dyskinesia impairment scale: a new instrument to measure dystonia and choreoathetosis in
dyskinetic cerebral palsy. Dev Med Child Neurol 54(3):278–283. doi:10.1111/j.1469-
8749.2011.04209.x
Monbaliu E, de Cock P, Ortibus E, Heyrman L, Klingels K, Feys H (2016) Clinical patterns of
dystonia and choreoathetosis in participants with dyskinetic cerebral palsy. Dev Med Child
Neurol. 58(2):138–144. doi:10.1111/dmcn.12846
Nakao S (2011) Gait and posture assessments of a patient treated with deep brain stimulation in
dystonia using three-dimensional motion analysis systems. J Med Investig 58:264–272
Pavone L, Burton J, Gaebler-Spira D (2013) Dystonia in childhood: clinical and objective measures
and functional implications. J Child Neurol 28(3):340–350. doi:10.1177/0883073812444312
Phukan J, Albanese A, Gasser T, Warner T (2011) Primary dystonia and dystonia-plus syndromes:
clinical characteristics, diagnosis, and pathogenesis. Lancet Neurol 10(12):1074–1085.
doi:10.1016/s1474-4422(11)70232-0
Rebour R, Delporte L, Revol P, Arsenault L, Mizuno K, Broussolle E, ... Rossetti Y (2015) Dopa-
responsive dystonia and gait analysis: a case study of levodopa therapeutic effects. Brain Dev 37
(6):643–650. doi:10.1016/j.braindev.2014.09.005
Sanger TD (2003) Pediatric movement disorders. Curr Opin Neurol 16(4):529–535. doi:10.1097/
01.wco.0000084233.82329.0e
Sanger TD (2004) Toward a definition of childhood dystonia. Curr Opin Pediatr 16:623–627
Sanger T (2015) Movement disorders in Cerebral Palsy. J Pediatr Neurol 13(04):198–207.
doi:10.1055/s-0035-1558866
Sanger TD, Chen D, Fehlings DL, Hallett M, Lang AE, Mink JW, ... Valero-Cuevas F (2010)
Definition and classification of hyperkinetic movements in childhood. Mov Disord 25
(11):1538–1549. doi:10.1002/mds.23088
Scholle HC, Jinnah HA, Arnold D, Biedermann FH, Faenger B, Grassme R, ... Schumann NP
(2010) Kinematic and electromyographic tools for characterizing movement disorders in mice.
Mov Disord 25(3):265–274. doi:10.1002/mds.22933
Skogseid IM (2014) Dystonia – new advances in classification, genetics, pathophysiology and
treatment. Acta Neurol Scand Suppl 198:13–19. doi:10.1111/ane.12231
van der Kamp W, Berardelli A, Rothwell JC, Thompson PD, Day BL, Marsden CD (1989). Rapid
elbow movements in patients with torsion dystonia. J Neurol Neurosurg Psychiatry. 52
(9):1043–1049
van Egmond ME, Kuiper A, Eggink H, Sinke RJ, Brouwer OF, Verschuuren-Bemelmans CC, ... de
Koning TJ (2015) Dystonia in children and adolescents: a systematic review and a new
diagnostic algorithm. J Neurol Neurosurg Psychiatry 86(7):774–781. doi:10.1136/jnnp-2014-
309106
Functional Effects of Ankle Sprain
Ilona M. Punt and Lara Allet
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Symptoms and Functional Deficits Related to an Ankle Sprain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Risk Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Epidemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Health-Care Costs Related to Ankle Sprain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Assessments Needed for Proper Decision Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Clinical Exam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Specific Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Functional Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Patient Reported Outcome Measures (PROMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Treatment Modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
Ankle sprain is one of the most common sports-related injuries and can lead to
recurrences and chronic ankle instability (CAI). In the acute phase, ankle sprain
I.M. Punt (*)

Department of Epidemiology, Maastricht University, CAPHRI, Maastricht, The Netherlands
Department of Physical Therapy, University of Applied Sciences of Western Switzerland, Carouge,
Switzerland
e-mail: ilona.punt@maastrichtuniversity.nl
L. Allet
Department of Physical Therapy, University of Applied Sciences of Western Switzerland, Carouge,
Switzerland
Department of Community Medicine, Geneva University Hospitals and University of Geneva,
Geneva, Switzerland
e-mail: Lara.Allet@hcuge.ch

DOI 10.1007/978-3-319-30808-1_72-1
2 I.M. Punt and L. Allet
patients experience mostly pain, limited ankle mobility, and reduced ankle muscle
strength. CAI patients have a history of their ankle “giving way” and/or “feeling
unstable,” after at least one significant ankle sprain. They continue to suffer from
pain and impaired performance during functional tasks. Both acute ankle sprains
and CAI have a negative influence on daily life activities such as walking, sports-
related activities such as jump landings, as well as on patients’ perception of
health and function. Functional deficits should be carefully assessed for appro-
priate clinical decision making and to propose the most suitable, individualized
(physiotherapeutic) intervention. Acute ankle sprains are first treated according to
the rest, ice, compression, and elevation (RICE) protocol. Nonsteroidal anti-
inflammatory drugs may also be recommended for pain management. A short
period of immobilization by means of a lower leg cast can facilitate rapid decrease
in pain and swelling. Afterward, functional exercise therapy is recommended. In
the case of CAI, patients should wear external ankle support during sporting
activities to reduce the risk of recurring sprains and undergo exercise therapy
including balance and muscle strengthening exercises. New technologies could
be implemented in future rehabilitation programs in order to offer athletes greater
flexibility in terms of training time and more varied, sports-related, exercises
at home.
Keywords
Ankle sprain • Chronic ankle instability • Clinical exam • Gait • Balance • Jump •
Patient reported outcome measures • Treatment
State of the Art
Definition
Ankle sprain is defined as a partial or complete tear of the ligaments of the ankle due
to sudden stretching. The most common mechanism causing lateral ankle sprain is
excessive and explosive inversion and some degree of plantar flexion of the rear-foot
on the tibia (Balduini and Tetzlaff 1982) during gait, cutting maneuvers during
sports, jump landings, or stepping off an uneven surface (Bullock-Saxton et al.
1994; Hertel 2008; Wikstrom et al. 2006). In particular, athletes playing indoor/
court sports (i.e., basketball, volleyball, tennis), field-based sports (i.e., soccer), or
long-distance running have an increased risk of ankle sprain injuries (Doherty et al.
2014b; Nery et al. 2016).
After an ankle sprain injury, the lateral ligaments of the ankle are the most
frequently injured, in particular the anterior talofibular ligament (ATFL), followed
by injuries to the calcaneal fibular ligament (CFL) (Martin et al. 2013). The severity
of an ankle sprain can be graded as follows:
Functional Effects of Ankle Sprain 3
• Grade I: mild damage to the fibers of the ligament without ligamentous laxity of
the affected joint
• Grade II: partial tear of the ligament with abnormal laxity of the ankle joint
• Grade III: complete rupture of the ankle ligament (Birrer et al. 1999)
Assessing the grade of an ankle sprain is important to make the appropriate

decision about future treatment strategy. The more severe the grade, the more time
the patient will need to fully recover.
Symptoms and Functional Deficits Related to an Ankle Sprain
Ankle sprain patients experience in the acute phase mostly pain, limited ankle
mobility, and reduced ankle muscle strength. These symptoms negatively influence
daily life activities such as gait, balance performance, and sports-related activities
such as jump landings (Aiken et al. 2008; Hertel 2000; Rose et al. 2000).
Persons with chronic ankle instability (CAI) have a history of their ankle “giving
way” and/or “feeling unstable” after at least one significant ankle sprain that was
associated with inflammatory response. These individuals experience pain and
demonstrate impaired performance during functional tasks (Mcgovern and Martin
2016). CAI patients also find activities like (single-leg) balance performance, gait,
and sports-related activities difficult to perform.
Risk Factors
Risk factors and mechanisms which potentially contribute to recurrent ankle sprains
include altered intrinsic body functions such as decreased proprioception in the ankle
ligaments, muscle weakness, limited range of motion of the ankle joint, and extrinsic
factors such as inappropriate footwear (Van Rijn et al. 2008; Mckeon and Hertel 2008a).
A history of ankle sprain is in itself a risk factor for a re-sprain and may lead to
mechanical or functional instability resulting in CAI (Van Rijn et al. 2008; Hertel 2002).
Epidemiology
Lateral ankle sprain injury is the most common sports-related acute injury and occurs
predominantly in persons aged 15–19 years (Fong et al. 2007; Hootman et al. 2007;
Waterman et al. 2010). The incidence rate is 11.55 per 1000 exposures (Doherty et al.
2014b). Despite various treatment modalities, persons with a history of ankle sprain
are known to present higher risk for re-spraining their ankle and developing
mechanical or functional instability, resulting in chronic ankle instability (CAI)
(Van Rijn et al. 2008; Hertel 2002). Up to 34% of patients report recurrent ankle
sprains during the first year after the initial injury (Van Rijn et al. 2008). Up to 74%
of all ankle sprain patients also experience residual symptoms such as pain, swelling,
peroneal muscle weakness, or neuromuscular dysfunctions (Hertel 2000), all of

which make patients susceptible for further injury and negatively influence activities
of daily life (ADL) and sport performance.
Health-Care Costs Related to Ankle Sprain
Ankle sprains lead to high direct and indirect health-care costs (Nazarenko et al.
2013; Verhagen et al. 2005). In the United States (US), the estimated costs are
between 318 and 914 US dollars per acute ankle sprain (Nazarenko et al. 2013). In
the Netherlands, the costs are estimated to be 360 euros per sprain (Verhagen et al.
2005). Another Dutch study calculated direct health-care costs of patients visiting
the emergency department after a ligamentous ankle injury. These costs are esti-
mated to be 684 euros per injury (De Boer et al. 2014). Costs increase with patients’
age, in particular for ambulance care, home care and rehabilitation. Since the
introduction of new guidelines in the Netherlands, patients with minor injuries are
able to visit the general practitioner 24 h a day, 7 days a week. Consequently, fewer
ankle sprain patients visited the emergency department (De Boer et al. 2014).
Assessments Needed for Proper Decision Making
Whenever a patient visits a medical doctor or physical therapist after suffering an

ankle injury, the physician and/or therapist inquires about the accident occurrence
and performs a clinical exam. In particular, the physician and/or physical therapist
inquire about perceived pain intensity and if the patient has a history of “giving way”
and/or “feelings of instability” which might indicate CAI. The degree of swelling is
measured, and the remaining range of motion (ROM) and muscle strength evaluated.
Physicians/therapists then check for functional deficits such as deficits in balance,
gait, and sports-related tasks (e.g., jump landings). Patient reported outcome mea-
sures (PROMs), assessing patients’ perception of their health and function, complete
the examination.
Clinical Exam
A clinical exam serves to develop an individualized evidence-based rehabilitation

plan that supports recovery while decreasing the risk of reinjury (Mcgovern and
Martin 2016). During the clinical exam, the physician or physical therapist inquires
about pain intensity during rest and activity, observes the patient in the standing and
lying down positions, and screens for swelling, hematoma, bruising, and deformity.
The physician or therapist also checks postural issues such as overpronation of the
foot or difficulty in putting the foot on the floor. The entire ankle joint is then
palpated to assess skin temperature (which can increase due to acute inflammation)
and swelling and ascertain whether the ligaments are painful to touch. Then the
Fig. 1 The figure-of-eight

method for measuring ankle
edema
therapist measures the degree of edema, range of motion (ROM), and muscle
strength and checks for functional deficits.
Swelling
The gold standard to measure edema is the water displacement method (Mawdsley
et al. 2000; Mckay et al. 2001), but this method may be too time-consuming for
efficient clinical use. However, Mawdsley et al. (2000) and Watson et al. (2008)
showed recently that the figure-of-eight method is valid when assessing ankle edema
in a clinical setting and that the inter-rater reliability of this method is excellent
(ICC > 0.99, SEM of 0.2 cm). The therapist wraps a tape measure around
standardized anatomical landmarks near the ankle, and the distance provides a
circumferential estimate of volume (Fig. 1).
Range of Motion (ROM)

Ankle mobility can be assessed actively and passively in the classical way using a
goniometer. However, measuring dorsiflexion in standing simulates the ROM required
for functional tasks. This is particularly relevant because the torques applied to the
ankle during weight bearing are clearly greater than in non-weight-bearing conditions
and the resulting measurement may be more indicative of the range available for
functional activities (Bennell et al. 1998; Bohannon et al. 1989). To measure the ankle
dorsiflexion range during weight bearing, the participant stands on an apparatus
consisting of a horizontal footplate attached to a vertical board. Participants align the
big toe and heel of the test leg over a line marked along the center of the footplate.
Participants are instructed not to lift the test heel, which is checked by the examiner
who gently palpates for lifting while the participant moves his knee forward into a
lunge position until the patella touches the midline of the vertical board (Bennell et al.
1998; Bohannon et al. 1989). To prevent forward movement of the big toe as the knee
moves forward over the foot, a block is placed in front of the big toe. The distance
(in cm) from the vertical board to the big toe is measured.
Muscle Strength
Classical manual muscle testing in a sitting position is indicated to test strength in
dorsiflexion, inversion, inversion with dorsiflexion, eversion, and eversion with
plantar flexion. However, using a handheld dynamometer could further improve
the measuring of muscle strength.
The strength of the gastrocnemius and soleus can also be tested in a standing
position. To test the gastrocnemius, the patient stands on the test limb with the knee
extended. Patients may place one or two fingers on a table to assist with balance. The
patient actively raises his heel from the floor 20 consecutive times without resting
through full range of plantar flexion. To test the soleus the same procedure is used;
only the patient stands on the test limb with his knee slightly flexed (Hislop et al.
2013; Spink et al. 2010).
Specific Tests
After these classical tests, specific ankle tests are used to assess the integrity of the
ligaments or the damage sustained.
• Anterior drawer test: used to assess the ATFL. The patient sits with his knee
flexed in order to relax the gastrocnemius and soleus muscles. The ankle is in 10
plantar flexion. The heel is held and forcefully pulled forward with one hand,
while the other hand applies proximal counterpressure (Fig. 2). The test is
positive when the injured ankle shows severe anterior subluxation compared to
the noninjured ankle (Welck et al. 2015).
• Talar tilt test: used to test the CFL. The patient is positioned with his knee flexed.
The heel is grasped and the talus tilted into varus (Fig. 3). A normal degree of tilt
is 0–23 . The injured side is compared to the noninjured ankle (Welck et al.
2015).
Fig. 2: Anterior drawer test

Fig. 3: Talar tilt test
• External rotation test: used to assess the syndesmosis. The leg is stabilized
proximally to the ankle joint while grasping the plantar aspect of the foot and
rotating the foot externally. The test is positive when painful (Alonso et al. 1998).
• Squeeze tests: also used to assess the syndesmosis. The fibula and tibia are
compressed at midcalf. The test is positive when it elicits pain (Alonso et al.
1998; Welck et al. 2015).
Functional Tests
Physical therapists should use functional tests to assess how the patient moves and
how the ankle injury affects balance, walking, and jumping. The severity of the ankle
injury dictates which tests are selected.
Balance Performance
Balance deficits have been found to be present after acute ankle sprains (Mckeon and
Hertel 2008a) for up to 1 week both in the injured ankle and in the noninjured ankle
(Evans et al. 2004). Balance performance can be assessed using clinical tests or
laboratory analyses using force platforms.
A frequently used dynamic clinical test is the Star Excursion Balance Test
(SEBT), a series of single-limb squats using the non-stance limb to reach maximally
to touch a point along one of eight designated lines on the ground (Fig. 4) (Gribble
et al. 2012). The SEBT showed that acute ankle sprain patients have a shorter
anterior reach/leg length compared to healthy controls (Pourkazemi et al. 2016;
Akbari et al. 2006).
Force plate data are most often characterized by the analysis of the trajectory of
the center of pressure (COP). Parameters that are frequently chosen to assess balance
performance are COP range, length, and speed. The review of Mckeon and Hertel
(2008a) showed that COP range, length, and speed of the injured ankle were
increased after an acute lateral ankle sprain compared to healthy controls.
Anterior
Anteriorlateral Anteromedial
Lateral Medial
Posterolateral Posteromedial
Posterior
Fig. 4: Reach direction for left ankle stance of the Star Excursion Balance Test (SEBT). Directions
are labeled based on the reach direction from the stance limb
Although patients significantly improve postural control (e.g., COP range, length,
as well as speed) during the first 4 weeks after an ankle sprain (Evans et al. 2004;
Hertel et al. 2001), they frequently present residual functional deficits (muscle
strength, mobility) and impaired postural control after this period (Hertel et al.
2001; Holme et al. 1999). Genthon et al. (2010) showed that ankle sprain patients
present asymmetric balance in bipedal stance during the first 10 days after injury.
From day 10 to day 30, bipedal balance improved and returned to normal after
30 days (Genthon et al. 2010). However, deficits may become more evident while
balancing on one leg (Mckeon and Hertel 2008a; Rozzi et al. 1999; Wikstrom et al.
2009a). The study of Hertel et al. (2001), for example, showed that during a single-
leg balance test, COP length and speed were greater in the injured ankle compared to
the noninjured ankle for up to 4 weeks after injury. However, both parameters
significantly improved 4 weeks after injury compared to the day after the injury
(Hertel et al. 2001).
Using the noninjured leg as a reference is not recommended to estimate residual
deficit because it is assumed that central neural changes after unilateral lateral ankle
trauma affect motor control of both extremities (Holme et al. 1999).
The meanings differ as regard to the balance impairments for CAI patients.
Dynamic clinical tests (i.e., SEBT) in CAI patient showed that presenting deficits
in ankle dorsiflexion ROM leads to difficulties with the anterior reach direction of
the SEBT (Basnett et al. 2013; Munn et al. 2010; Arnold et al. 2009). Meta-analyses
studying force plate data concluded that ankle instability leads to impaired balance
performance (Arnold et al. 2009; Munn et al. 2010). These meta-analyses further
stated that it remains unclear whether these differences in balance preexisted or were
the consequence of the ankle injury (Arnold et al. 2009). Furthermore, no definitive
conclusion could be made based on a systematic review from Mckeon and Hertel
(2008a). They compared the COP performance achieved with the injured ankle of
CAI patients with the COP performance of healthy controls or the one achieved with
the uninjured ankle (Mckeon and Hertel 2008a).
Gait Performance
Temporal-Spatial
Acute ankle sprain patients demonstrated disturbed gait parameters such as slower
gait speed, shorter step length, shorter single support time, as well as disturbed
symmetry for single support time (Crosbie et al. 1999; Punt et al. 2015). For
example, healthy persons demonstrate on average a walking speed of
1.29 0.17 m/s, while ankle sprain patients only walk 1.12 0.25 m/s 4 weeks
after the initial injury (Punt et al. 2015). Punt et al. (2015) showed that decreased
walking speed was correlated to increased pain levels and deficits in dorsiflexion
muscle strength measured with a handheld dynamometer.
In contrast, CAI patients show similar gait speed compared to healthy
age-matched controls (Monaghan et al. 2006).
Kinematics
For normal gait, which is one of the most frequent activities, a minimum ankle
dorsiflexion of 10 has been shown to be necessary (Riener et al. 2002). In acute ankle
sprain patients, Crosbie et al. (1999) showed that the degree of maximum passive
dorsiflexion of the ankle measured was correlated with gait speed, step length, and
symmetry for single support time. However, Punt et al. (2015) found no difference
between ankle sprain patients and healthy controls for maximum dorsiflexion during
the stance phase of gait while walking at a self-selected walking speed. In contrast,
Punt et al. (2015) found that the maximum plantar flexion was reduced on the injured
side of patients ( 14.2 7.9) compared to healthy controls ( 18.7 8.0). Doherty
et al. (2015a) described similar findings when comparing the injured ankle with the
noninjured side. In addition, the timing of maximum plantar flexion was delayed at
the injured ankle compared with that of the healthy group (Punt et al. 2015).
Furthermore acute ankle sprain patients demonstrated increased ankle inversion
with a greater inversion moment (Delahunt et al. 2006; Doherty et al. 2015a;
Monaghan et al. 2006).
While walking with a similar gait speed, CAI patients showed more ankle
inversion compared to healthy controls (Monaghan et al. 2006; Delahunt et al.
2006). In addition, CAI patients inverted the ankle at a rate of 0.5 rad/s around
heel strike, while healthy controls slowly everted their ankle at a rate of 0.1 rad/s
(Monaghan et al. 2006).
Kinetics
Ankle sprain patients demonstrated lower maximum concentric dorsiflexion power
compared with the healthy controls. They also demonstrated lower maximum
eccentric plantar flexion power compared with the healthy subjects. Furthermore,
the maximum moment was lower in the ankle sprain group compared with the
healthy group (Punt et al. 2015). These findings indicate that patients with an
ankle sprain use a more conservative and secure gait pattern, characterized by
slow self-selected walking speed. This might have been even more marked if
patients had been tested with stricter requirements, such as faster walking speed,
running, or irregular walking surfaces.
CAI patients showed an evertor moment directly after heel strike, whereas
healthy controls showed an inventor moment while walking at a similar gait
speed. Joint power also differed between these two groups after heel strike. CAI
patients showed concentric power generation, while healthy controls showed eccen-
tric power generation (Monaghan et al. 2006).
Sports-Related Tasks
Previous studies showed that ankle sprain patients displayed reduced ankle plantar
flexion while performing a bipedal drop vertical jump 2 weeks, but also 6 months,
after the initial injury, compared to healthy controls (Doherty et al. 2014a, 2015b).
Similar findings were found when patients performed a single-leg jump 4 weeks
after the injury compared to a control group (Allet et al. 2016). Thus, perhaps the
altered movement pattern of ankle sprain patients represents a security strategy to
avoid recurrences. Increased dorsiflexion of the ankle brings the joint into a more
closed-packed position that protects the lateral ligament complex more efficiently.
This evasive movement avoids the typical ankle sprain injury mechanism, a combi-
nation of inversion and plantar flexion of the ankle (Balduini and Tetzlaff 1982).
Nevertheless, abnormal foot positioning at initial contact might lead to faulty
neuromuscular preprogramming of ankle joint movement, thereby, in the long
term, contributing to the development of CAI (Hertel 2008). Alternatively,
dorsiflexion could be a precaution behavior of ankle sprain patients. Ankle sprain
patients may not jump as high as healthy persons, and therefore their toes would not
be able to clear the floor.
Studies including CAI patients showed divergent results ranging from increased
ankle dorsiflexion (Caulfield and Garrett 2002) to decreased dorsiflexion when
patients performed a single-leg drop landing (Ashtonmiller et al. 1996). Studies
investigating neuromuscular control mechanisms in CAI patients also reported
reduced activation of the peroneus muscle before initial contact during a single-leg
drop landing compared to healthy subjects (Caulfield et al. 2004; Delahunt et al.
2006).
CAI patients, whose landing phase while running and stop-jump maneuvers was
evaluated, showed a more inverted ankle, reduced muscle co-contraction, and
decreased dynamic stiffness in the ankle joint during landing phase compared to
healthy controls (Lin et al. 2011).
Patient Reported Outcome Measures (PROMs)
Patient reported outcome measures (PROMs) are instruments assessing patients’

perception of their health and function. Ankle PROMs typically include questions
addressing pain, mobility, function, and quality of life. PROMs are important but
should supplement, rather than replace, existing measures of quality and perfor-
mance. Several validated PROMs are available for foot and ankle disorders
(Eechaute et al. 2007; Weel et al. 2015; Martin and Irrgang 2007). The most
commonly used PROMs related to foot and ankle disorders are the following:
• Ankle Joint Functional Assessment Tool (AJFAT): it contains five impairment

items (pain, stiffness, stability, strength, rolling over), four activity-related items
(walking on uneven ground, cutting when running, jogging, and descending
stairs), and one overall quality item. The maximal total score of the AJFAT is
40 points, and the minimum is 0 points (Rozzi et al. 1999).
• Functional Ankle Ability Measure (FAAM): this self-report questionnaire con-
sists of a 21-item, activities of daily living (ADL) subscale and an 8-item sports-
related subscale. The final score ranges from 0 to 100 for ADL as well as sports,
higher scores indicating higher levels of function (Martin et al. 2005).
• Foot and Ankle Disability Index (FADI): this 34-item questionnaire is divided
into two subscales: the FADI and the FADI Sport. The FADI contains 4 pain-
related items and 22 activity-related items. The FADI Sport contains eight
activity-related items. The scores of the FADI and FADI Sport are then trans-
formed into percentages (Hale and Hertel 2005).
• Functional Ankle Outcome Score (FAOS): this 42-item questionnaire is divided
into five subscales, pain, other symptoms, ADL, sport and recreation function,
and foot and ankle-related quality of life. Final scores are then transformed into
ratings from 0 to 100 (worst to best score) (Roos et al. 2001).
According to Eechaute et al. (2007), the FADI and the FAAM are considered as
the most appropriate self-report tools to quantify functional disabilities in patients
with CAI. CAI patients demonstrate worse FADI and FADI Sport scores compared
to healthy controls (Wikstrom et al. 2009b).
Treatment Modalities
Appropriate management of lateral ankle sprain is vital for a successful recovery.

Acute ankle sprains can be managed initially according to the rest, ice, compression,
and elevation (RICE) protocol, and the use of nonsteroidal anti-inflammatory drugs
may be recommended for pain management (Kerkhoffs and Van Dijk 2013). A short
period (<2 weeks) of immobilization in, for example, a lower leg cast, can facilitate
rapid decrease in pain and swelling, thereby speeding up the recovery of normal
ROM (Kerkhoffs et al. 2001; Lamb et al. 2009). Afterward, functional exercise
therapy, such as physical therapy, is recommended (Fong et al. 2009; Kemler et al.
2011; Kerkhoffs et al. 2012; Petersen et al. 2013). Therapists should develop an
individualized intervention plan depending on the severity of the injury, their clinical
exam, and on the results of functional tests.
The rehabilitation plan should be adapted to the needs of the patients. For
example, when a patient recovers normal active ROM, muscle strength, and propri-
oceptive control of the ankle through a strength and balance training program
(Holme et al. 1999), (physical) therapists should incorporate (cutting) sports exer-
cises, such as tennis, basketball, or football, in order to return to sport.
CAI patients should wear external ankle support during sporting activities to
reduce the risk of recurrent ankle sprains (Aaltonen et al. 2007), and exercise therapy
should include balance and muscle strengthening exercises (Mckeon and Hertel
2008b; Wikstrom et al. 2009a).
Future Directions
It is important for clinicians involved in the treatment and guidance of ankle sprain
patients (e.g., general practitioners, physical therapists, rehabilitation physicians,
sports physicians) to understand if an alternative exercise therapy tool like the Wii
Fit™ does add value to standard physical therapy exercise rehabilitation protocols in
ankle sprain patients and can be considered as a standard or an additional treatment
modality.
Future research should assess the advantages of incorporating new technologies,
that aim to provide a more attractive and individualized treatment plan for acute
lateral ankle sprain or CAI patients, in rehabilitation programs (Mcgovern and
Martin 2016). New technologies such as virtual reality exercise programs (i.e.,
Nintendo Wii Fit™) may enrich standard physical therapy exercise protocols for
both acute ankle sprain and CAI patients (Punt et al. 2016; Kim and Heo 2015).
Compared to conventional physical therapy, virtual reality exercises offer patients
greater flexibility in terms of training time and more varied, sports-related exercises.
It is thought that such innovative treatment modalities stimulate patients’ interest and
motivation. Usage of available virtual reality games such as the Nintendo Wii Fit™,
which is a relatively low-cost intervention, might also contribute to reducing health-
care costs (Warburton et al. 2007). According to Mcgovern and Martin (2016), future
research in the field of ankle sprain and CAI should furthermore assess the effect of
treatment directed at upper extremity postural control, the benefit of fibular posi-
tioning, and the effect of trigger point dry needling. The rationale for introducing
upper extremity postural control exercises is that individuals with CAI are thought to
suffer from deficits in central sensorimotor integration and processing (Springer et al.
2015). Fibular positioning might be corrected by taping (Someeh et al. 2015), and
dry needling has been proposed to treat trigger points and their negative effects
(Mcgovern and Martin 2016; Salom-Moreno et al. 2015).
Besides clinical outcomes such as swelling, mobility and strength, and functional
tests (i.e., balance, gait, and jump performance), outcomes such as return to sport for
athletes should be taken into consideration. Risk of reinjury, which is currently as
high as 34% (Van Rijn et al. 2008), should also be considered in the future.
Furthermore, young patients, in particular, should be included in studies comparing
different treatments as the peak incidence of ankle sprain is between 15 and 19 years
of age (Waterman et al. 2010; Bridgman et al. 2003). Finally, after studying the
clinical effectiveness and efficiency of different treatments, cost-effectiveness ana-
lyses should assess and thus potentially improve the performance of different
exercise programs.
References
Aaltonen S, Karjalainen H, Heinonen A, Parkkari J, Kujala UM (2007) Prevention of sports
injuries: systematic review of randomized controlled trials. Arch Intern Med 167:1585–1592
Aiken AB, Pelland L, Brison R, Pickett W, Brouwer B (2008) Short-term natural recovery of ankle
sprains following discharge from emergency departments. J Orthop Sports Phys Ther
38:566–571
Akbari M, Karimi H, Farahini H, Faghihzadeh S (2006) Balance problems after unilateral lateral
ankle sprains. J Rehabil Res Dev 43:819–824
Allet L, Zumstein F, Eichelberger P, Armand S, Punt IM (2016) Neuromuscular control mechanisms
during single-leg jump landing in sub-acute ankle sprain patients: a case control study. PM R.
2016 Jul 25. pii: S1934-1482(16)30260-X. doi: 10.1016/j.pmrj.2016.07.006. [Epub ahead of
print]
Alonso A, Khoury L, Adams R (1998) Clinical tests for ankle syndesmosis injury: reliability and
prediction of return to function. J Orthop Sports Phys Ther 27:276–284
Arnold BL, De La Motte S, Linens S, Ross SE (2009) Ankle instability is associated with balance
impairments: a meta-analysis. Med Sci Sports Exerc 41:1048–1062
Ashtonmiller JA, Ottaviani RA, Hutchinson C, Wojtys EM (1996) What best protects the inverted
weightbearing ankle against further inversion? Evertor muscle strength compares favorably with
shoe height, athletic tape, and three orthoses. Am J Sports Med 24:800–809
Balduini FC, Tetzlaff J (1982) Historical perspectives on injuries of the ligaments of the ankle. Clin
Sports Med 1:3–12
Basnett CR, Hanish MJ, Wheeler TJ, Miriovsky DJ, Danielson EL, Barr JB, Grindstaff TL (2013)
Ankle dorsiflexion range of motion influences dynamic balance in individuals with chronic
ankle instability. Int J Sports Phys Ther 8:121–128
Bennell KL, Talbot RC, Wajswelner H, Techovanich W, Kelly DH, Hall AJ (1998) Intra-rater and
inter-rater reliability of a weight-bearing lunge measure of ankle dorsiflexion. Aust J Physiother
44:175–180
Birrer RB, Fani-Salek MH, Totten VY, Herman LM, Politi V (1999) Managing ankle injuries in the
emergency department. J Emerg Med 17:651–660
Bohannon RW, Tiberio D, Zito M (1989) Selected measures of ankle dorsiflexion range of motion-
differences and intercorrelations. Foot Ankle 10:99–103
Bridgman SA, Clement D, Downing A, Walley G, Phair I, Maffulli N (2003) Population based
epidemiology of ankle sprains attending accident and emergency units in the West Midlands of
England, and a survey of UK practice for severe ankle sprains. Emerg Med J 20:508–510
Bullock-Saxton JE, Janda V, Bullock MI (1994) The influence of ankle sprain injury on muscle
activation during hip extension. Int J Sports Med 15:330–334
Caulfield BM, Garrett M (2002) Functional instability of the ankle: differences in patterns of ankle
and knee movement prior to and post landing in a single leg jump. Int J Sports Med 23:64–68
Caulfield B, Crammond T, O’sullivan A, Reynolds S, Ward T (2004) Altered ankle-muscle
activation during jump landing in participants with functional instability of the ankle joint. J
Sport Rehabil 13:189–200
Crosbie J, Green T, Refshauge K (1999) Effects of reduced ankle dorsiflexion following lateral
ligament sprain on temporal and spatial gait parameters. Gait Posture 9:167–172
De Boer AS, Schepers T, Panneman MJM, Van Beeck EF, Van Lieshout EMM (2014) Health care
consumption and costs due to foot and ankle injuries in the Netherlands, 1986-2010. BMC
Delahunt E, Monaghan K, Caulfield B (2006) Altered neuromuscular control and ankle joint
kinematics during walking in subjects with functional instability of the ankle joint. Am J Sports
Med 34:1970–1976
Doherty C, Bleakley C, Hertel J, Sweeney K, Caulfield B, Ryan J, Delahunt E (2014a) Lower
extremity coordination and symmetry patterns during a drop vertical jump task following acute
ankle sprain. Hum Mov Sci 38:34–46
Doherty C, Delahunt E, Caulfield B, Hertel J, Ryan J, Bleakley C (2014b) The incidence and
prevalence of ankle sprain injury: a systematic review and meta-analysis of prospective epide-
miological studies. Sports Med 44:123–140
Doherty C, Bleakley C, Hertel J, Caulfield B, Ryan J, Delahunt E (2015a) Lower extremity function
during gait in participants with first time acute lateral ankle sprain compared to controls. J
Doherty C, Bleakley C, Hertel J, Caulfield B, Ryan J, Sweeney K, Patterson MR, Delahunt E
(2015b) Coordination and symmetry patterns during the drop vertical jump, 6-months after first-
time lateral ankle sprain. J Orthop Res 33:1537–1544
Eechaute C, Vaes P, Van Aerschot L, Asman S, Duquet W (2007) The clinimetric qualities of
patient-assessed instruments for measuring chronic ankle instability: a systematic review. BMC
Evans T, Hertel J, Sebastianelli W (2004) Bilateral deficits in postural control following lateral
ankle sprain. Foot Ankle Int 25:833–839
Fong DT, Hong Y, Chan LK, Yung PS, Chan KM (2007) A systematic review on ankle injury and
ankle sprain in sports. Sports Med 37:73–94
Fong DT, Chan YY, Mok KM, Yung P, Chan KM (2009) Understanding acute ankle ligamentous
sprain injury in sports. Sports Med Arthrosc Rehabil Ther Technol 1:14
Genthon N, Bouvat E, Banihachemi JJ, Bergeau J, Abdellaoui A, Rougier PR (2010) Lateral ankle
sprain alters postural control in bipedal stance – part 1: restoration over the 30 days following
the injury. Scand J Med Sci Sports 20:247–254
Gribble PA, Hertel J, Plisky P (2012) Using the star excursion balance test to assess dynamic
postural-control deficits and outcomes in lower extremity injury: a literature and systematic
review. J Athl Train 47:339–357
Hale SA, Hertel J (2005) Reliability and sensitivity of the foot and ankle disability index in subjects
with chronic ankle instability. J Athl Train 40:35–40
Hertel J (2000) Functional instability following lateral ankle sprain. Sports Med 29:361–371
Hertel J (2002) Functional anatomy, pathomechanics, and pathophysiology of lateral ankle insta-
bility. J Athl Train 37:364–375
Hertel J (2008) Sensorimotor deficits with ankle sprains and chronic ankle instability. Clin Sports
Med 27:353–370 vii
Hertel J, Buckley WE, Denegar CR (2001) Serial testing of postural control after acute lateral ankle
sprain. J Athl Train 36:363–368
Hislop H, Avers D, Brown M (2013) Daniels and Worthingham’s muscle testing: techniques of
manual examination and performance testing. Elsevier Health Sciences: Missouri
Holme E, Magnusson SP, Becher K, Bieler T, Aagaard P, Kjaer M (1999) The effect of supervised
rehabilitation on strength, postural sway, position sense and re-injury risk after acute ankle
ligament sprain. Scand J Med Sci Sports 9:104–109
Hootman JM, Dick R, Agel J (2007) Epidemiology of collegiate injuries for 15 sports: summary
and recommendations for injury prevention initiatives. J Athl Train 42:311–319
Kemler E, Van De Port I, Backx F, Van Dijk CN (2011) A systematic review on the treatment of
acute ankle sprain brace versus other functional treatment types. Sports Med 41:185–197
Kerkhoffs GM, Van Dijk CN (2013) Acute lateral ankle ligament ruptures in the athlete: the role of
surgery. Foot Ankle Clin 18:215–218
Kerkhoffs GM, Rowe BH, Assendelft WJ, Kelly KD, Struijs PA, Van Dijk CN (2001)
Immobilisation for acute ankle sprain. A systematic review. Arch Orthop Trauma Surg
121:462–471
Kerkhoffs GM, Van Den Bekerom M, Elders LA, Van Beek PA, Hullegie WA, Bloemers GM, De
Heus EM, Loogman MC, Rosenbrand KC, Kuipers T, Hoogstraten JW, Dekker R, Ten Duis HJ,
Van Dijk CN, Van Tulder MW, Van Der Wees PJ, De Bie RA (2012) Diagnosis, treatment and
prevention of ankle sprains: an evidence-based clinical guideline. Br J Sports Med 46:854–860
Kim KJ, Heo M (2015) Effects of virtual reality programs on balance in functional ankle instability.
J Phys Ther Sci 27:3097–3101
Lamb SE, Marsh JL, Hutton JL, Nakash R, Cooke MW, Collaborative Ankle Support T (2009)
Mechanical supports for acute, severe ankle sprain: a pragmatic, multicentre, randomised
controlled trial. Lancet 373:575–581
Lin CF, Chen CY, Lin CW (2011) Dynamic ankle control in athletes with ankle instability during
sports maneuvers. Am J Sports Med 39:2007–2015
Martin RL, Irrgang JJ (2007) A survey of self-reported outcome instruments for the foot and ankle. J
Orthop Sports Phys Ther 37:72–84
Martin RL, Irrgang JJ, Burdett RG, Conti SF, Van Swearingen JM (2005) Evidence of validity for
the foot and ankle ability measure (FAAM). Foot Ankle Int 26:968–983
Martin RL, Davenport TE, Paulseth S, Wukich DK, Godges JJ, Orthopaedic Section American
Physical Therapy Association (2013) Ankle stability and movement coordination impairments:
ankle ligament sprains. J Orthop Sports Phys Ther 43:A1–40
Mawdsley RH, Hoy DK, Erwin PM (2000) Criterion-related validity of the figure-of-eight method
of measuring ankle edema. J Orthop Sports Phys Ther 30:149–153
Mcgovern RP, Martin RL (2016) Managing ankle ligament sprains and tears: current opinion. Open
Access J Sports Med 7:33–42
Mckay GD, Goldie PA, Payne WR, Oakes BW (2001) Ankle injuries in basketball: injury rate and
risk factors. Br J Sports Med 35:103–108
Mckeon PO, Hertel J (2008a) Systematic review of postural control and lateral ankle instability, part
I: can deficits be detected with instrumented testing. J Athl Train 43:293–304
Mckeon PO, Hertel J (2008b) Systematic review of postural control and lateral ankle instability, part
II: is balance training clinically effective? J Athl Train 43:305–315
Monaghan K, Delahunt E, Caulfield B (2006) Ankle function during gait in patients with chronic
ankle instability compared to controls. Clin Biomech (Bristol, Avon) 21:168–174
Munn J, Sullivan SJ, Schneiders AG (2010) Evidence of sensorimotor deficits in functional ankle
instability: a systematic review with meta-analysis. J Sci Med Sport 13:2–12
Nazarenko A, Beltran LS, Bencardino JT (2013) Imaging evaluation of traumatic ligamentous
injuries of the ankle and foot. Radiol Clin N Am 51:455–478
Nery C, Raduan F, Baumfeld D (2016) Foot and ankle injuries in professional soccer players:
diagnosis, treatment, and expectations. Foot Ankle Clin 21:391–403
Petersen W, Rembitzki IV, Koppenburg AG, Ellermann A, Liebau C, Bruggemann GP, Best R
(2013) Treatment of acute ankle ligament injuries: a systematic review. Arch Orthop Trauma
Surg 133:1129–1141
Pourkazemi F, Hiller C, Raymond J, Black D, Nightingale E, Refshauge K (2016) Using balance
tests to discriminate between participants with a recent index lateral ankle sprain and healthy
control participants: a cross-sectional study. J Athl Train 51:213–222
Punt IM, Ziltener JL, Laidet M, Armand S, Allet L (2015) Gait and physical impairments in patients
with acute ankle sprains who did not receive physical therapy. PM R 7:34–41
Punt IM, Ziltener JL, Monnin D, Allet L (2016) Wii Fit exercise therapy for the rehabilitation of
ankle sprains: its effect compared with physical therapy or no functional exercises at all. Scand J
Med Sci Sports 26:816–823
Riener R, Rabuffetti M, Frigo C (2002) Stair ascent and descent at different inclinations. Gait
Posture 15:32–44
Roos EM, Brandsson S, Karlsson J (2001) Validation of the foot and ankle outcome score for ankle
ligament reconstruction. Foot Ankle Int 22:788–794
Rose A, Lee RJ, Williams RM, Thomson LC, Forsyth A (2000) Functional instability in
non-contact ankle ligament injuries. Br J Sports Med 34:352–358
Rozzi SL, Lephart SM, Sterner R, Kuligowski L (1999) Balance training for persons with
functionally unstable ankles. J Orthop Sports Phys Ther 29:478–486
Salom-Moreno J, Ayuso-Casado B, Tamaral-Costa B, Sanchez-Mila Z, Fernandez-De-Las-Penas C,
Alburquerque-Sendin F (2015) Trigger point dry needling and proprioceptive exercises for the
management of chronic ankle instability: a randomized clinical trial. Evid Based Complement
Alternat Med 2015:1–9
Someeh M, Norasteh AA, Daneshmandi H, Asadi A (2015) Immediate effects of Mulligan’s fibular
repositioning taping on postural control in athletes with and without chronic ankle instability.
Phys Ther Sport 16:135–139
Spink MJ, Fotoohabadi MR, Menz HB (2010) Foot and ankle strength assessment using hand-held
dynamometry: reliability and age-related differences. Gerontology 56:525–532
Springer S, Gottlieb U, Moran U, Verhovsky G, Yanovich R (2015) The correlation between
postural control and upper limb position sense in people with chronic ankle instability. J Foot
Ankle Res 8:23
Van Rijn RM, Van Os AG, Bernsen RM, Luijsterburg PA, Koes BW, Bierma-Zeinstra SM (2008)
What is the clinical course of acute ankle sprains? A systematic literature review. Am J Med
121:324–331
Verhagen EA, Van Tulder M, Van Der Beek AJ, Bouter LM, Van Mechelen W (2005) An economic
evaluation of a proprioceptive balance board training programme for the prevention of ankle
sprains in volleyball. Br J Sports Med 39:111–115
Warburton DE, Bredin SS, Horita LT, Zbogar D, Scott JM, Esch BT, Rhodes RE (2007) The health
benefits of interactive video game exercise. Appl Physiol Nutr Metab 32:655–663
Waterman BR, Owens BD, Davey S, Zacchilli MA, Belmont PJ (2010) The epidemiology of ankle
sprains in the United States. J Bone Joint Surg Am 92a:2279–2284
Watson CP, Boland RA, Refshauge KM (2008) Measurement reliability of swelling in the acute
ankle sprain. Foot Ankle J 1:4
Weel H, Zwiers R, Sierevelt IN, Haverkamp D, Van Dijk CN, Kerkhoffs GM (2015) Dutch-
language patient-reported outcome measures for foot and ankle injuries; a systematic review.
Ned Tijdschr Geneeskd 159:A8831
Welck M, Rafferty M, Eltz S, Al-Nammari SS, Eseonu KC (2015) Management of ankle injuries.
BMJ 351:h6698
Wikstrom EA, Tillman MD, Chmielewski TL, Borsa PA (2006) Measurement and evaluation of
dynamic joint stability of the knee and ankle after injury. Sports Med 36:393–410
Wikstrom EA, Naik S, Lodha N, Cauraugh JH (2009a) Balance capabilities after lateral ankle
trauma and intervention: a meta-analysis. Med Sci Sports Exerc 41:1287–1295
Wikstrom EA, Tillman MD, Chmielewski TL, Cauraugh JH, Naugle KE, Borsa PA (2009b) Self-
assessed disability and functional performance in individuals with and without ankle instability:
a case control study. J Orthop Sports Phys Ther 39:458–467
Gait Changes in Skeletal Dysplasia
William G. Mackenzie and Oussama Abousamra
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Gait Changes in Achondroplasia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Gait Changes in Diastrophic Dysplasia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Gait Changes in Morquio Syndrome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gait Changes in Spondyloepiphyseal Dysplasia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Abstract
Skeletal dysplasias form a group of conditions that affect the musculoskeletal
system resulting in gait changes. These conditions share some common gait
patterns, such as short stride length, low gait velocity, and increased forward
pelvic tilt. Increased pelvic rotation is also a common mechanism, used to
increase the step length. However, specific gait deviations are found in each
condition. Knee axis deviations in the coronal plane are well-described clinical
features in these conditions. While bilateral knee varus is common in achondro-
plasia, bilateral knee valgus is a shared finding between diastrophic dysplasia and
Morquio syndrome. Sagittal kinematic deviations in stance show normal knee
flexion in achondroplasia and increased knee flexion values in diastrophic dys-
plasia and Morquio syndrome.
Due to the different, multilevel gait deviations that follow the multilevel
deformity found in skeletal dysplasia, instrumented, three-dimensional, gait anal-
ysis becomes a helpful tool to evaluate the individual aspects of the deformity and
to approach the patient with an accordingly customized management plan.
W.G. Mackenzie (*) • O. Abousamra (*)

Nemours Alfred I. duPont Hospital for Children, Wilmington, DE, USA
e-mail: William.Mackenzie@nemours.org; oussama.abousamra@gmail.com

DOI 10.1007/978-3-319-30808-1_73-1
2 W.G. Mackenzie and O. Abousamra
Keywords
Skeletal dysplasia • Achondroplasia • Diastrophic dysplasia • Morquio syndrome •
Spondyloepiphyseal dysplasia
Introduction
Multiple conditions have been described under the term “skeletal dysplasia.” The
pathophysiology of each condition has its effect on the anatomy and function of the
musculoskeletal system. Since the affected body level, and the severity of the
disease, can be different between these conditions, the total impact on the individ-
ual’s gait can be different as well. As in any other, gait disturbing, disorders, gait
evaluation allows the managing team to address the deformity more accurately and
tailor the surgical intervention accordingly.
Gait studies in skeletal dysplasia have been scarce in the literature. Published
studies have addressed the three-dimensional dynamic lower extremity defor-
mities of children with different skeletal dysplasia conditions. Comparisons
with dynamic measurements of typically developing children have been
reported. This chapter reviews the available data that present gait changes in
achondroplasia, diastrophic dysplasia, Morquio syndrome, and spondyloe-
piphyseal dysplasia.
Gait Changes in Achondroplasia
Achondroplasia is the most common skeletal dysplasia. It is caused by a mutation of

the gene encoding the fibroblast growth factor receptor 3 (FGFR3), and it is inherited
as an autosomal dominant disorder. The extremities are affected more than the trunk,
and long bones are affected more in the proximal portions of the extremities (Shirley
and Ain 2009).
Children with achondroplasia have shorter stride length and lower gait velocity
when compared with typically developing children (stride length (achondroplasia,
67 18 cm; normal, 106 13 cm) and gait velocity (achondroplasia, 80 19 cm/
s; normal, 122 18 cm/s)) (Inan et al. 2006). In order to increase the step length,
children with achondroplasia show higher rotation at the pelvis level (21 10 ).
Increased anterior pelvic tilt (21 6 ) and limited hip extension ( 10 9 ) are
also noted and might be due to the increased lordotic curve seen in achondroplasia
(Kopits 1988).
In the coronal plane, varus malalignment of the knee (19 13 ) has been
reported in children with achondroplasia (Inan et al. 2006). When knee varus
measurements were compared between gait analysis and radiographs of the same
patients, poor correlation was found with less varus noted on the radiographs
(16 9 ). This has been related to the pattern of weight bearing since the kinematic
data were obtained in stance when the child was bearing weight on one leg, while the
radiographs were taken with the child bearing weight on both legs (Inan et al. 2006).
Gait Changes in Skeletal Dysplasia 3
In the sagittal plane, knee flexion has been found normal across the gait cycle;
however, radiographs of the same patients showed genu recurvatum (23.5 13.2 )
(Inan et al. 2006). This finding has been explained by the normal neuromuscular
function in these children, which can prevent the recurvatum during walking.
Conversely, reduced knee extension in midstance was also reported in adults with
achondroplasia (mean age of 24.5 6.1 years), and this extension limitation
remained after limb lengthening despite having a good range of motion (van der
Meulen et al. 2008). In another group of adolescents with achondroplasia, knee
hyperextension range was seen displaying relaxed hamstrings; however, there was
still limited knee extension in midstance, which might be due to excessive ankle
dorsiflexion (Egginton et al. 2006).
Elevated internal valgus moments at the knee have been found compared with the
general population and suggested as a good predictor of deformity progression and
long-term degenerative disease (Inan et al. 2006).
While tibial torsion kinematic measurements have showed a good correlation
with the clinical measurements, using the angle between the transmalleolar axis and
the long axis of the thigh, variable measurements have been reported between
patients and even between limbs of the same patient. This variability necessitates a
careful assessment in case tibial torsion is to be surgically addressed (Inan
et al. 2006).
Foot varus has been noted on foot pressure analysis. However, full passive
flexibility was reported (Inan et al. 2006). Reduced plantar flexion at push off and
increased dorsiflexion during swing were reported (Egginton et al. 2006) and
suggested as a mechanism to facilitate foot clearance of the relatively long foot
compared to the short leg (Egginton et al. 2006). In a different group of adults with
achondroplasia who had limb lengthening, sagittal plane kinematics at the ankle were
within one standard deviation of normal (van der Meulen et al. 2008). This finding
suggests improved gait following limb lengthening (van der Meulen et al. 2008).
Gait Changes in Diastrophic Dysplasia
Diastrophic dysplasia is a rare form of skeletal dysplasia, inherited as an autosomal

recessive disorder. The genetic defect is in chromosome 5q that encodes diastrophic
dysplasia sulfate transfer protein (Bayhan et al. 2015). The typical clinical picture is
short stature with short trunk and short limbs (Poussa et al. 1991).
Similar to children with achondroplasia, children with diastrophic dysplasia have
shorter stride length and lower gait velocity when compared with typically devel-
oping children (Bayhan et al. 2015) (stride length (diastrophic dysplasia,
52 16 cm; normal, 98 11 cm) and gait velocity (diastrophic dysplasia,
61 26 cm/s; normal, 115 6 cm/s)). Increased pelvic rotation (12.1 8.14
forward and 12.2 8.3 trailing), increased forward pelvic tilt (26.6 5.7 ), and
increased forward trunk tilt (14.9 12.2 ) have also been reported for these chil-
dren along with increased hip flexion and limited hip abduction (Bayhan et al. 2015).
In the coronal plane, knee valgus malalignment (12.9 8.6 ) has been noted;
however, these measurements did not correlate with radiographic knee valgus
measurements (11.7 13 ), and they moderately correlated with the clinical valgus
measurements (13 7 ) (Bayhan et al. 2015).
In the sagittal plane, increased knee flexion has been reported with smaller range
of motion for the knee flexion extension movement (lower delta knee motion:
24.9 6.9 ) (Bayhan et al. 2015).
In this group of children, two subgroups were reported according to whether they
have patellar dislocation or not. Average hip flexion and average knee flexion were
both elevated in the subgroup with patellar dislocation (Bayhan et al. 2015). There-
fore, the early management of the extensor mechanism malalignment, in addition to
the valgus malalignment, was proposed as a possible effective treatment to slow
deformity progression.
Gait Changes in Morquio Syndrome
Morquio syndrome, or mucopolysaccharidosis IV, is a systemic lysosomal storage

disorder (Dhawale et al. 2013). It is an autosomal recessive disorder caused by a
deficiency of the lysosomal enzyme N-acetylgalactosamine-6-sulfate sulfatase
(Tomatsu et al. 2011). The clinical picture in this syndrome includes a short trunk
in addition to multiple limb deformities (Tomatsu et al. 2011).
Children with Morquio syndrome also have shorter stride length and lower gait
velocity when compared with typically developing children (stride length (Morquio,
72.3 7.9 cm; normal, 114.5 6.8 cm) and gait velocity (Morquio,
67.2 17.3 cm/s; normal, 126.3 14.7 cm/s)) (Dhawale et al. 2013). Increased
forward pelvic tilt (19.3 2.9 ) and increased forward trunk tilt (5.9 8.9 ) have
also been reported findings in these children. In addition, increased hip flexion and
limited hip abduction have been described (Dhawale et al. 2013).
In the coronal plane, knee valgus has been reported (22 11.2 ), and in this
group of children, strong correlation was noted between kinematic knee valgus and
radiographic knee valgus (24.7 9.6 ) (Dhawale et al. 2013). In the sagittal plane,
increased knee flexion has been reported (19.3 15.5 ).
Elevated internal varus moments at the knee have been found in response to knee
valgus (Dhawale et al. 2013).
Limited ankle dorsiflexion has been described compared with typically develop-
ing children (Dhawale et al. 2013), and pedobarography showed normal foot
pressures (Table 1).
Gait Changes in Spondyloepiphyseal Dysplasia
Spondyloepiphyseal dysplasia (SED) is a condition that primarily affects the verte-

brae and the proximal epiphyses of the long bones (Anderson et al. 1990). The
genetic defect has been localized on the COL2A1 gene coding for type II collagen
Gait Changes in Skeletal Dysplasia 5
Table 1 Measurements of means standard deviations in every condition in addition to the

percentages of the measurements taken from the typically developing children
Measurements Achondroplasia Diastrophic dysplasia Morquio syndrome
Stride length (cm) 67 18 (63 %) 52 16 (53 %) 72.3 7.9 (63 %)
Gait velocity (cm/s) 80 19 (66 %) 61 26 (53 %) 67.2 17.3 (53 %)
Anterior pelvic tilt 21 6 (162 %) 26.6 5.7 (193 %) 19.3 2.9 (127 %)
Knee varus/valgus 19 13 varus 12.9 8.6 valgus 22 11.2 valgus
(Anderson et al. 1990). The clinical findings include short trunk and short limbs with
multiple deformities in the spine, chest, and hips.
The authors have unpublished data for a group of children with SED congenita
who had instrumented gait analysis before and after they underwent proximal
femoral valgus osteotomy to address their coxa vara. The preoperative gait evalua-
tion showed short stride length (70 18 cm) and reduced gait velocity
(87 23 cm/s). Forward pelvic tilt was increased (34 9 ) as well as hip flexion
(23 13 ).
None of the parameters showed significant changes after the valgus osteotomy
except forward pelvic tilt that reduced from (34 9 ) to (25 10 ). Gait deviation
index (GDI) showed a statistically significant improvement (from 52 15 to
58 15); however, this change was within the normal, ten points, standard devia-
tion of GDI.
Summary
In skeletal dysplasia, specifically in the conditions discussed in this chapter,

different gait changes are noted. Overall, people with skeletal dysplasia walk
with shorter steps and slower speed than the general population. Increased
forward pelvic tilt has been a common finding. Different malalignment patterns
can be found at the knee level. Instrumented gait analysis can serve as a helpful
tool to evaluate the multilevel malalignment profile in children with skeletal
dysplasia and to customize the treatment based on their functional needs.
References
Anderson IJ, Goldberg RB, Marion RW et al (1990) Spondyloepiphyseal dysplasia congenita:
genetic linkage to type II collagen (COL2AI). Am J Hum Genet 46(5):896–901
Bayhan IA, Er MS, Nishnianidze T et al (2015) Gait pattern and lower extremity alignment in
children with diastrophic dysplasia. J Pediatr Orthop. Available from: 10.1097/
BPO.0000000000000530
Dhawale AA, Church C, Henley J et al (2013) Gait pattern and lower extremity alignment in
children with Morquio syndrome. J Pediatr Orthop B 22(1):59–62
Egginton R, Newman C, Walsh M et al (2006) Kinematic characteristics of Achondroplasia. Gait
Posture. Elsevier 24(2):S249–S250
Inan M, Thacker M, Church C et al (2006) Dynamic lower extremity alignment in children with
achondroplasia. J Pediatr Orthop 26(4):526–529
Kopits SE (1988) Thoracolumbar kyphosis and lumbosacral hyperlordosis in achondroplastic
children. Basic Life Sci 48:241–255
Poussa M, Merikanto J, Ryöppy S et al (1991) The spine in diastrophic dysplasia. Spine 16
(8):881–887
Shirley ED, Ain MC (2009) Achondroplasia: manifestations and treatment. J Am Acad Orthop Surg
17(4):231–241
Tomatsu S, Montano AM, Oikawa H et al (2011) Mucopolysaccharidosis type IVA (Morquio a
disease): clinical review and current treatment: a special review. Curr Pharm Biotechnol 12
(6):931–945
van der Meulen J, Dickens W, Burton M et al (2008) O052 Gait characteristics of achondroplasia
following lower limb-lengthening. Gait Posture 28:S36–S37, Elsevier
Prosthetic Foot Principles and Their
Influence on Gait
Andrew Hansen and Felix Starker
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Studying Prosthetic Feet Using Standard Gait Analysis Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Clinical Comparisons of Existing Prosthetic Feet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Principles of Prosthetic Feet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Stiffness/Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Roll-Over Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Active Push-Off in Late Stance Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Toe Clearance During Swing Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
This chapter describes mechanical properties of prosthetic feet and their effects on
level gait of persons with amputations. These descriptions focus on recent
literature and knowledge gained from controlled studies of different properties
including stiffness/flexibility, damping, roll-over characteristics, active push-off
A. Hansen (*)
Minneapolis VA Health Care System, Minneapolis, MN, USA
University of Minnesota, Minneapolis, MN, USA
e-mail: Andrew.Hansen2@va.gov
F. Starker
Biomechatronic Systems, Fraunhofer Institute for Manufacturing Engineering and Automation,
Stuttgart, Germany
e-mail: felix.starker@ipa.fraunhofer.de
# Springer International Publishing AG (outside the USA) 2016 1

DOI 10.1007/978-3-319-30808-1_74-1
2 A. Hansen and F. Starker
in late stance phase, and toe clearance during swing phase. The chapter also
discusses future directions in prosthetic foot research, including the need for both
amputee-independent measurements combined with clinical trials. Specific
approaches to future studies of prosthetic feet are proposed that could further
our knowledge base and ultimately lead to improved prescription of prosthetic
feet for persons with amputations.
Keywords
Prosthetic • Stiffness • Flexibility • Damping • Heel • Toe • Mechanical proper-
ties • Push-off • Toe clearance • Roll-over shape • Research • Amputation
Introduction
Commercially available prosthetic feet are quite simple when compared to the
amazingly complex human ankle-foot systems they attempt to replace (chapter
“▶ Kinematic Foot Models for Instrumented Gait Analyis”). The human ankle-
foot system is composed of numerous bones, muscles, and connective tissue. Most
muscles actuating movements of the ankle-foot system operate across multiple
joints and are controlled by the human’s complex neural control system. Attempts
have been made to directly use the complex anatomical bone structures
from cadaveric feet as the endoskeletons of prosthetic feet (Kabra 1990; Kabra
and Narayanan 1991). The cadaveric bony structures were set within a rubber
covering providing elastic deformation; however, the control of these structures
by a neuromuscular system was not attempted and does not seem feasible in the near
future.
The physiologic human ankle-foot system is used for numerous tasks including
standing, walking, and running. For these tasks, the system is known to adapt into
different functional states. For example, the ankle-foot system is thought to act
primarily as a spring structure for running and as a rolling system for forward
walking (Cavagna et al. 1964). Also, recent work suggests the ankle-foot system
acts to create a flat base of support for standing balance (Hansen and Wang 2010).
The system is highly adaptive to changing variables associated with these tasks. For
example, walking and running occur over a variety of speeds and on changing terrain
(e.g., slopes, rocks, grass, stairs). A full examination of all tasks and their associated
variables is beyond the scope of this chapter. Instead, this chapter will focus on
current principles of prosthetic feet and their effects on level walking. As will be
explained, more research is needed to gain a more complete understanding of this
topic. This chapter provides the authors’ viewpoints based on a brief review of
relevant literature in this area and years of experience conducting research and
development related to prosthetic feet and their influence on the gait of persons
using them in lower-limb prostheses.
Prosthetic Foot Principles and Their Influence on Gait 3
State of the Art
Studying Prosthetic Feet Using Standard Gait Analysis Methods
The most common approach to studying prosthetic feet is to utilize gait analysis
methods that were developed for the study of physiologic ankle-foot systems
(chapter “▶ Variations of Marker-Sets and Models for Standard Gait Analysis”).
With respect to lower-limb prostheses, this requires placement of motion analysis
markers to approximate joint centers of rotation (chapter “▶ Next Generation
Models Using Optimized Joint Center Location”). For single-axis knee joints, this
marker placement is straightforward since the axis of rotation is obvious (chapter
“▶ The Influence of Prosthetic Knee Joints on Gait”). However, for prosthetic feet it
is rather difficult to determine the appropriate placement of markers. Despite the
findings that most prosthetic feet have measured centers of rotation that are signif-
icantly different than the anatomical human ankle (Rusaw and Ramstrand 2010) and
that move during walking (Sawers and Hahn 2011), most investigators use a marker
placement on the prosthetic foot that is similar to the anatomical ankle for simplicity.
Most prosthetic feet are also lighter than anatomical feet, which should be kept in
mind when analyzing kinetic data from gait analyses. Problems associated with
utilizing standard gait analysis techniques for prosthetic feet have been described in
the prosthetic foot literature for a number of years (e.g., Geil et al. 2000). However,
these techniques continue to be used due to their widespread availability and because
their results, although somewhat flawed, allow more direct comparison with ana-
tomical systems.
Clinical Comparisons of Existing Prosthetic Feet
Early studies of prosthetic feet were mainly clinical comparisons of different com-
mercially available prosthetic feet, with the goal of identifying criteria for prescrip-
tion (see Hofstad et al. 2004 and Hafner 2005 for reviews on this topic). Most studies
involved repeated measures trials of several different prosthetic foot types followed
by gait analyses to try and determine meaningful differences. The two reviews cited
above found that evidence from these studies was insufficient to establish criteria for
prescription of prosthetic feet. Some of the studies found increased prosthetic side
dorsiflexion and decreased sound limb loading when subjects used prosthetic feet
like the Flex-Foot compared with the SACH foot. These findings led to hypotheses
of causality between these features. However, these hypotheses are not logical when
taken to extremes (Hansen 2005). For example, an extremely soft prosthetic foot
should provide large amounts of prosthetic side dorsiflexion, but not allow appre-
ciable forward progression of the center of pressure. In effect, such a foot would be
similar to a peg leg and would likely lead to a noticeable limp and increased sound
limb loading. More recent work in the last decade has tried to focus on prosthetic
foot properties (principles) and their effects on gait of persons with lower-limb
amputations. Much has been learned from these recent studies, but much is still
unknown.
Principles of Prosthetic Feet
There are several recent studies that have investigated various principles of pros-
thetic feet and their effects on human walking. The bulk of these studies can be
grouped into examination of the following properties:
1. Stiffness/flexibility
2. Damping
3. Roll-over characteristics
4. Active push-off in late stance phase
5. Toe clearance during swing phase
Many recent studies have tried to focus in on specific prosthetic foot principles
and their effects on walking, by either using experimental ankle-foot systems or by
varying parameters of commercially available prostheses. Both approaches have
value at increasing our general knowledge base surrounding prosthetic feet. The
following sections overview some of the studies aimed at improving our under-
standing of prosthetic foot principles and their effects on gait.
Stiffness/Flexibility
A number of recent studies have tried to determine the effects of different prosthetic
ankle-foot stiffness or flexibility on walking. These studies have focused either on
varying/adding ankle stiffness or varying the stiffness or flexibility of prosthetic
keels.
A few recent studies suggest that adding flexible prosthetic ankles to existing
commercial prosthetic feet increases the measured ankle range of motion, leading to
increased ankle power and propulsive impulses in late stance phase of walking
(Zmitrewicz et al. 2006; Su et al. 2010; Ventura et al. 2011). Major et al. (2014)
used an experimental ankle-foot system to also find that more compliant ankle
stiffness led to increased dorsiflexion in late stance, providing an easier tibial
progression and improved gait performance. Ventura et al. (2011) found that exper-
imental ankles attached to commercial prosthetic feet led to decreased activity of
muscles associated with body propulsion, but increased activity of muscles associ-
ated with body support. More compliant ankles led to larger changes in muscle
activity, but yielded lower energy storage and return compared with the stiffer ankles
in their study. These studies collectively suggest that there is an optimal ankle
stiffness that may improve gait performance by increasing energy storage and return
yet minimize effects of muscle activity for body support. The work of Zelik et al.
(2011) supports this idea, finding that an intermediate stiffness in an experimental

prosthetic ankle-foot system provided the lowest metabolic energy expenditure.
However, the stiffness in their system also contributed to heel mechanics, which
may not translate to more general prosthetic ankle-foot systems. More research is
needed to examine ankle stiffness and its effects on human walking.
Some recent studies have systematically altered the keel stiffness or flexibility of
prosthetic feet to determine their effects on the user. Klodd et al. (2010a, b) used an
experimental foot design to show that prosthetic feet with excessive forefoot
flexibility lead to increased contralateral limb loading, most likely due to dramat-
ically shortened effective keel lengths and a “drop-off” effect during late stance
phase on the prosthesis. Although prosthetic feet in Klodd et al.’s study did not
produce significant changes in oxygen cost, there was a strong preference against
prosthetic feet that were overly flexible. Fey et al. (2011) used experimental
prosthetic feet made using additive manufacturing to find that more compliant
prosthetic feet lead to increased dorsiflexion in late stance, increased energy storage
during midstance, and increased energy release during late stance. However, the
increase also required increased activity of muscles responsible for body support
(i.e., quadriceps). Fey et al.’s study also showed that prosthetic feet that were stiffer
than nominal led to increased loading to the contralateral limb, while the more
compliant ankle did not change the contralateral limb loading. Close examination of
the compliant system used by Fey suggests strongly that the effective foot length
was much less reduced from nominal, when compared with the most compliant
keels used in Klodd’s study. Collectively, the results of these studies suggest that
compliance can be beneficial for prosthesis users with strong quadriceps and good
control of these muscles. Prosthetic keels that are too compliant or too stiff lead to
increase loading on the intact limb, which may be detrimental over time. Later work
by Fey et al. (2012) utilized a forward dynamics model to find that stiffness should
be altered in different regions of the prosthetic forefoot to minimize metabolic cost
and intact knee loading. Future human studies should attempt to validate these
models to also determine further gains that can be made with passive mechanical
prosthetic feet.
Damping
The damping properties of prosthetic feet have been estimated by numerous inves-
tigators using mechanical tests. One test to assess energy loss due to damping is
loading and unloading the prosthesis in a materials testing machine and assessing the
amount of hysteresis in the load versus deflection curve (Van Jaarsveld et al. 1990).
Others have assessed damping losses by measuring responses at different loading
rates and fitting to spring-damper models (Geil 2002). Damped oscillations of
prosthetic feet when reacting with body weight loads have also been examined as
a way of estimating natural frequency and damping (Lehmann et al. 1993; Knox
1996; Major et al. 2011). Some early testing attempted to link mechanical properties
such as natural frequency and damping with human testing results (e.g., Lehmann
et al. 1993). However, the results of these tests are difficult to decipher given the use
of prosthetic feet with many different properties.
More recently, prosthetic ankles have been commercialized that have dampers in
series with spring elements over a range of ankle motion (e.g., Endolite Echelon).
One initial study comparing the Echelon to a very similar foot without the damper
(Endolite Esprit) showed that the damping in the system led to a smoother, more
consistent progression of the center of pressure during walking (De Asha et al.
2013). On the other hand, a recent study of experimental feet suggests that high
damping in a prosthetic foot may lead to increased loading on the contralateral limb
(Jin et al. 2016). A major difference in these studies relates to the prosthetic ankle-
foot systems tested. The Echelon allows damping over a range of motion that is then
blocked at 3 degrees of dorsiflexion, followed by increased storage of energy in a
spring mechanism. The experimental foot in the latter study (Jin et al. 2016)
incorporated large pieces of high damping foam that likely provided a much different
response, and the feet were tested by able-bodied persons using a prosthesis simu-
lator. Nonetheless, the findings of these two studies suggest that damping over a
small range of motion, and potentially more in the plantarflexion range of motion,
may be beneficial for smoother rollover, but damping in the forefoot region of the
prosthetic foot may reduce the energy return in late stance and potentially lead to
increased contralateral loading. More research is needed to understand specific
benefits and drawbacks of damping in prosthetic feet on human walking.
Roll-Over Characteristics
A relatively recent approach for describing prosthetic ankle-foot systems is the roll-
over shape, which is the effective shape that the system conforms to during the
stepping phase of walking (Hansen et al. 2000). The roll-over shape is currently
measured in a motion laboratory directly from walking trials of persons with lower-
limb amputations or from simulated rollover of prosthetic feet using mechanical
apparatus (Hansen et al. 2000; Curtze et al. 2009; Major et al. 2011). An advantage
of the roll-over shape approach is that it provides a simple outcome from complex
systems with different inherent stiffness and damping properties, providing value for
design, alignment, and evaluation of prosthetic ankle-foot systems (Hansen and
Childress 2010). It also provides information on design of prosthetic ankle-foot
systems (e.g., roll-over shape radius and effective foot length) as well as the
alignment (i.e., positioning of the roll-over shape with respect to the residual limb
socket). An inherent drawback is that the details involved in creating the roll-over
shape are missed, and late stance energy return properties are not considered. These
details are likely important to the prosthesis user’s function and comfort when using
different prosthetic ankle-foot systems. Nonetheless, a few recent studies altering
roll-over shape properties have been useful at increasing our understanding of
prosthetic ankle-foot function and effects on the prosthesis user’s gait.
Hansen et al. (2006) used an experimental prosthetic foot to systematically
examine the effects of reducing the effective foot (keel) length, by creating an overly
flexible joint at different regions of the forefoot. As the effective foot length was
reduced, the prosthesis user experienced a “drop-off” effect on the prosthetic side
and increased loading on the contralateral limb. The asymmetry in limb loading was
more pronounced for faster walking speeds. As mentioned earlier, Klodd et al.
(2010a) also found that prosthetic feet with excessive flexibility led to reductions
in the effective foot length and were also associated with the “drop-off” effect. The
addition of ankle units to commercial prosthetic feet has been found to significantly
decrease the roll-over shape radius when used by persons with bilateral transtibial
amputations (Gard et al. 2011) yet not significantly decrease the effective foot
length.
A key recent study by Adamczyk and Kuo (2013) suggests that effective foot
length is much more impactful on mechanical work associated with step-to-step
transitions and the energetic cost of walking compared with roll-over shape radius.
Their findings seem to corroborate many past studies in that measured ankle motion
can be increased using more compliant ankles or keel structures. These changes
require more muscle activity for body support, but also lead to increased energy
storage and return. These changes do not appear to have large effects on contralateral
limb loading unless the compliance is so high that it leads to significant reductions in
effective foot length.
Active Push-Off in Late Stance Phase
A key difference between most prosthetic feet and the able-bodied ankle-foot system
is the lack of active push-off in late stance phase, particularly for fast walking speeds.
Most prosthetic feet have materials that store energy from midstance to opposite
initial contact and then release energy during unloading. Some newer prototype
systems have been developed that store energy in early stance phase and release it
later in stance phase (Segal et al. 2012; Morgenroth et al. 2011). A prototype ankle-
foot system that provided controlled energy storage and return (CESR) was devel-
oped that would store energy in a spring from initial stance phase and return the
energy during unloading of the prosthetic foot. Segal et al. (2012) found that the
CESR system demonstrated increased energy storage during early stance phase and
increased push-off power and work in late stance phase compared to a conventional
and a previously prescribed prosthetic foot. However, the net metabolic cost for
CESR was unchanged compared to the conventional foot and was increased com-
pared to the previously prescribed foot, perhaps due to increased weight of the
prototype or other factors. Morgenroth et al. (2011) found that increased push-off
in the CESR foot led to reductions in contralateral limb loading, suggesting that
push-off in prostheses may have benefits for the remaining limb (in persons with
unilateral amputations). Ankle-foot prototypes have been developed that “find the
surface” in early stance, set an equilibrium point at foot flat, and then store energy
from a plantarflexed position (Williams et al. 2009; Nickel et al. 2014). These
systems have been shown to automatically adapt their alignment to different slopes
of walking and, in theory, should reduce metabolic cost of walking by storing and
releasing more energy during level walking. However, metabolic testing of these
systems has not yet been studied. Although these systems start to examine late stance
push-off, they are essentially passive mechanical systems that use clever mechanical
approaches and control to try and improve over the current state of the art. More
direct approaches have been used to provide active push-off in late stance phase of
walking. A recent prosthesis emulator has been developed to allow study of active
push-off, and one commercially available system that provides active push-off has
also been studied.
Caputo and Collins (2014) used able-bodied persons walking with a prosthesis
emulator to examine the effects of push-off work in the late stance phase of walking.
They found that increases in push-off work led to reduced metabolic cost of walking
and reduction in hip work during swing phase initiation, suggesting that push-off
from the ankle assisted in swing phase of the simulated prosthetic leg. Using a
similar prosthesis emulator, Kim and Collins (2015) found that adjusting the push-
off work to stabilize step-to-step differences in walking led to reduction in metabolic
cost of walking compared with constant or destabilizing push-off work. The data
from prosthesis emulator studies is valuable, and further studies using these types of
systems should provide an improved understanding of prosthetic foot function and
corresponding effects on the prosthesis user.
The BiOM ankle is an ankle-foot system with a powered ankle joint that provides
active push-off during the late stance phase of walking. Studies of the first clinical
release of this system showed that the system provided increased ankle range of
motion, but was outputting more ankle power than an able-bodied ankle (Ferris et al.
2012). Compensatory movements at proximal joints were increased with the
powered limb compared with the energy storage and release prosthetic feet, but
patients had a preference for the powered ankle. Other studies of this powered ankle
showed that some contralateral limb loading variables were reduced with the
powered ankle compared with energy storage and return prostheses (Russell
Esposito and Wilken 2014; Hill and Herr 2013). Herr and Grabowski (2012)
found that use of the powered ankle increased preferred walking speed and reduced
metabolic cost to levels that were similar to able-bodied persons. Most likely the
design of this system was changing during these various studies, leading to some
variation in results. However, these studies collectively suggest that active push-off
in late stance phase can reduce metabolic cost, can increase walking speed, and may
reduce some loading conditions to the contralateral limb. More research is needed to
examine specific push-off approaches and their effects on the gait of prosthesis users.
Also more work is needed to reduce the cost of these types of systems to make them
affordable to more prosthesis users.
Toe Clearance During Swing Phase
One concern when walking with a prosthetic ankle-foot system is toe clearance
during swing phase. Inadequate toe clearance can lead to stumbles and sometimes
falls with the prosthesis, a known problem in persons with lower-limb amputations
Table 1 Summary of prosthetic foot principles and their effects on gait (see text for details)
Stiffness/flexibility There is likely an optimal level of stiffness/flexibility in prosthetic
ankles and keels that allows measured ankle dorsiflexion, improved
tibial progression, and increased energy storage and return. However,
if the stiffness is so low (flexibility so high) that the effective foot
length of the foot is dramatically shortened, the user will experience a
“drop-off” effect, leading to a limp and/or increased contralateral
limb loading. Lower stiffness feet are also associated with increased
activity of the muscles that provide body support in late stance
(quadriceps), so different stiffness levels may be needed for
prosthesis users with varying levels of quadriceps strength and
control
Damping Some damping in the prosthesis may smooth the forward progression
of the center of pressure of the ground reaction force during walking.
However, excessive damping in the forefoot area reduces energy
return from the prosthesis and leads to increased contralateral limb
loading
Roll-over characteristics Roll-over shapes with short arc lengths (shorter effective foot
lengths) are associated with the “drop-off” effect, leading to a limp
and/or increased contralateral limb loading. This effect is magnified
at higher walking speeds. There may be an optimal effective foot
length and roll-over shape radius for walking, and variations from the
optimal effective foot length have more dramatic effects on walking
economy than changes in roll-over shape radius
Active push-off in late Late stance active push-off in a powered prosthesis can lead to
stance phase walking speed and metabolic costs in amputees that are similar to
those in able-bodied persons. Active push-off may also lead to
reductions in contralateral limb loading
Toe clearance during Prosthesis users who walk with reduced toe clearance are more likely
swing phase to stumble in the following year. Ankle dorsiflexion in a prosthesis is
an effective means for reducing toe clearance in swing phase and
may lead to reduced stumbles and falls
(Miller et al. 2001). Sensinger et al. (2013) found that ankle dorsiflexion during
swing phase had a much larger effect on toe clearance compared with use of a four-
bar prosthetic knee joint. Some commercially available prosthesis systems on the
market end up in dorsiflexion as a function of their design. For example, the Echelon
ankle allows 3 degrees of dorsiflexion movement in a series damper. When this
damper bottoms out in late stance phase, the ankle stores and releases energy but
maintains the 3 degrees of dorsiflexion during swing phase. Johnson et al. (2014)
found that minimum toe clearance was increased with the Echelon foot compared
with the prosthesis users’ regular prosthetic feet. Other systems exist that actively
dorsiflex during early swing phase for toe clearance (e.g., Ossur Proprio). A recent
study suggests that prosthesis users with lower toe clearance are more likely to
stumble in the following year (Rosenblatt et al. 2016). Future prospective studies are
needed to examine interventions aimed at increasing toe clearance (e.g., training,
prosthesis designs) on reductions in stumbles and falls in lower-limb prosthesis users
(Table 1).
Future Directions
Prosthetic ankle-foot system research has improved over recent years and we have
gained significant understanding as a result. However, there is much to learn as many
of the principles overviewed in this chapter are interrelated (e.g., stiffness/flexibility
and roll-over shape radius). Several investigators have noted the importance of
coupling mechanical testing and human subjects testing to improve our understand-
ing of prosthetic foot principles and their effects on gait (e.g., Lehmann et al. 1993;
Knox 1996; Hansen 2005; Major et al. 2011; Postema et al. 1997). Major et al.
(2011) proposed an Amputee Independent Prosthesis Properties (AIPP) approach to
measure variables of interest in a prosthetic foot loading machine, independent of the
prosthesis user. However, the system used to make the measurements and similar
prosthetic foot loading systems (Knox 1996; Hansen et al. 2000) may not provide the
best approach since each prosthetic foot needs to be tested at several discrete angles
and manual setup between these angles is time consuming.
A potentially promising approach for future research in prosthetic feet includes
the following steps:
Create a working group of experts to establish a set of core measurable mechan-
ical properties of prosthetic ankle-foot systems – Based on the prosthetic foot
research to date, a possible set of properties might include heel stiffness and energy
stored during initial loading, effective foot length and roll-over shape during roll-
over, torsional stiffness properties from measured ankle torque versus ankle angle
curves (Pitkin 1995; Klute et al. 2001) during rollover, and energy returned during
unloading. Note that the energy returned during unloading would naturally be
increased for systems with active push-off in late stance phase. Working groups
already exist in the industry (AOPA’S Prosthetic Foot Project, ISO Tc168 WG3) to
discuss both durability testing and ways to mechanically characterize prosthetic feet.
These efforts provide a wonderful starting point for a future unified mechanical
description of prosthetic ankle-foot systems.
Formulate methods for capturing mechanical properties of prosthetic ankle-foot
systems – The same working groups should agree on testing methods for measure-
ment of the prosthetic foot mechanical properties. A particularly interesting
approach is to add sensors to current durability testing machines to capture func-
tional data. For example, the ISO 22675 (DIN EN ISO 22675:2006) testing approach
utilizes a heel-to-toe rolling loading and unloading scheme that closely simulates
humans walking with prostheses (Fig. 1). Current efforts are underway to measure
important mechanical properties of prosthetic feet prior to durability testing in the
same machines (ISO/TS 16955:2016-06). This approach would remove the need for
expensive and time-consuming testing after a deeper understanding of prosthetic
foot principles and their effects on prosthesis users is developed. Current test
schemes include uniaxial loading in different static angles (Fig. 2) but also torsional
and complete dynamic roll-over (heel-to-toe) tests with complex load profiles
(Fig. 3). With the addition of 6 degree-of-freedom load cells, the test machines can
provide a quick view of a prosthetic foot’s mechanical behavior. Static tests provide
some insight into some specific foot mechanics such as energy storage and return as
Fig. 1 ISO 22675/ISO TS 16955 test machine with a series of load sensors during dynamic
rollover of a prosthetic test sample
Fig. 2 Loading and unloading tests at different static angles. This approach is similar to that
proposed by Major et al. (2011) and could similarly capture Amputee Independent Prosthesis
Properties (Adapted from ISO 16955 (ISO/TS 16955))
Fig. 3 Loading and

unloading tests with dynamic
angular and load changes.
Several machines currently
exist to provide this type of
loading to prosthetic feet for
durability testing. When
coupled with the appropriate
sensors, these systems can
provide important mechanical
properties of prosthetic feet
(Adapted from ISO 16955
(ISO/TS 16955))
well as effective lever arm length, moment generation, and stiffness. In combination
with the dynamic tests that are performed under a complete rollover with a defined
range of angles and load profiles, more user-relevant information can be calculated
such as the roll-over shape and the total effective foot length. A dynamic loading
over an inclined surface in the frontal plane reveals more information about multi-
axial behavior by mimicking an everyday life situation. Furthermore applying a
torque moment helps to understand more about the torsional behavior that a pros-
thesis wearer would feel when walking around a curve. Most important and crucial
for all these mechanical tests is a defined, reproducible, and realistic foot setup
(alignment) and methods for load application to the prosthetic foot (e.g., posterior
shift, moment compensation).
Capture the range of mechanical properties in current commercially available
prosthetic ankle-foot systems – This step involves collecting data on most or all of
the current commercially available prosthetic ankle-foot systems and new systems as
they enter the prosthetics market. If this is coupled with testing that is already
occurring for durability, it seems logical that it could be gathered more easily over
time. Data should be stored and be available for public to avoid too much influence
from the industry. Although current tests described within ISO 16955 (ISO/TS
16955) or the AOPA prosthetic foot project show a suitable way to describe the
mechanical behavior of passive prosthetic feet, some electronically controlled or
propelled prosthetic feet demand a more complex sensory input information such as
shank angle with respect to gravity and several forward-directed steps to study more
complex algorithms and their potential benefits for the user. Suitable methods for this
testing could involve loading using a six-axis industrial robot with direct force or
hardware in the loop control (Dennerlein et al. 2014).
Conduct studies of mechanical properties and their effects on prosthesis users –
Knowing the range of mechanical properties from the last step will provide infor-
mation necessary to start to understand how clinically relevant variations in these
properties affect the user’s gait. One approach that would allow quick testing of
variations of prosthetic foot properties is the use of prosthesis emulators (Caputo and
Collins 2014; Kim and Collins 2015). These tests would indicate which properties
have the largest effects on gait and thus the properties that require further study in
clinical trials. Prosthetic feet with different known mechanical properties could then
be tested in randomized double-blind studies, similar to the recent study by Raschke
et al. (2015). At the end of the study, a removal of blinding would allow investigators
to pair mechanical properties of prosthetic feet with clinical trial outcomes. Other
large outcome measure databases may take in both clinical outcomes (e.g., on
mobility and balance confidence) with mechanical properties of prosthetic ankle-
foot systems. Over time, large data sets would be established that could be analyzed
to see associations between mechanics of prosthetic ankle-foot systems and the
prosthesis user’s corresponding outcomes. These future research studies should
also include qualitative feedback from prosthesis users to identify issues that are
not being captured by gait analysis systems, but that are important to practical use of
the prosthesis.
Much work is still needed to further our understanding of prosthetic foot princi-
ples and their effects on gait of the prosthesis user. An improved understanding will
help to guide clinical prescription of components in the future and will lend insight
needed to further improve prosthesis designs. Lastly, outcome data will assist
clinicians in justifying their choice of prosthetic ankle-foot systems to their cus-
tomers and third party payers.
References
Adamczyk PG, Kuo AD (2013) Mechanical and energetic consequences of rolling foot shape in
human walking. J Exp Biol 216:2722–2731
Caputo JM, Collins SH (2014) Prosthetic ankle push-off work reduces metabolic rate but not
collision work in non-amputee walking. Sci Rep 4(7213):1–9
Cavagna GA, Saibene FP, Margaria R (1964) Mechanical work in running. J Appl Physiol 19
(2):249–256
Curtze C, Hof AL, van Keeken HG, Halbertsma JPK, Postema K, Otten B (2009) Comparative roll-
over analysis of prosthetic feet. J Biomech 42:1746–1753
De Asha AR, Johnson L, Munjal R, Kulkarni J, Buckley JG (2013) Attenuation of centre-of-
pressure trajectory fluctuations under the prosthetic foot when using an articulating hydraulic
ankle attachment compared to a fixed attachment. Clin Biomech 28:218–224
Dennerlein F, Blab F, Starker F, Schneider U (2014) Simulation of the prosthetic gait with a six-axis
robot OT World 2014
DIN EN ISO 22675:2006 (2006) Prosthetics – testing of ankle-foot devices and foot units –
requirements and test methods
Ferris AE, Aldridge JM, Rabago CA, Wilken JM (2012) Evaluation of a powered ankle-foot
prosthetic system during walking. Arch Phys Med Rehabil 93:1911–1918
Fey NP, Klute GK, Neptune RR (2011) The influence of energy storage and return foot stiffness on
walking mechanics and muscle activity in below-knee amputees. Clin Biomech 26:1025–1032
Fey NP, Klute GK, Neptune RR (2012) Optimization of prosthetic foot stiffness to reduce metabolic
cost and intact knee loading during below-knee amputee walking: a theoretical study. J Biomech
Eng 134:111005 . 1-10
Gard SA, Su PF, Lipschutz RD, Hansen AH (2011) Effect of prosthetic ankle units on roll-over
shape characteristics during walking in persons with bilateral transtibial amputations. J Rehabil
Res Dev 48(9):1037–1048
Geil MD (2002) An iterative method for viscoelastic modeling of prosthetic feet. J Biomech
35:1405–1410
Geil MD, Parnianpour M, Quesada P, Berme N, Simon S (2000) Comparison of methods for the
calculation of energy storage and return in a dynamic elastic response prosthesis. J Biomech
33:1745–1750
Hafner BJ (2005) Clinical prescription and use of prosthetic foot and ankle mechanisms: a review of
the literature. J Prosthet Orthot 17(4S):5–11
Hansen AH (2005) Scientific methods to determine functional performance of prosthetic ankle-foot
systems. J Prosthet Orthot 17(4S):23–29
Hansen AH, Childress DS (2010) Investigations of roll-over shape: implications for design,
alignment, and evaluation of ankle-foot prostheses and orthoses. Disabil Rehabil 32
(26):2201–2209
Hansen AH, Wang CC (2010) Effective rocker shapes used by able-bodied persons for walking and
fore-aft swaying: implications for design of ankle-foot prostheses. Gait Posture 32:181–184
Hansen AH, Childress DS, Knox EH (2000) Prosthetic foot roll-over shapes with implications for
alignment of trans-tibial prostheses. Prosthetics Orthot Int 24:205–215
Hansen AH, Meier MR, Sessoms PH, Childress DS (2006) The effects of prosthetic foot roll-over
shape arc length on gait of trans-tibial prosthesis users. Prosthetics Orthot Int 30(3):286–299
Herr HM, Grabowski AM (2012) Bionic ankle-foot prosthesis normalizes walking gait for persons
with leg amputation. Proc Royal Soc 279:457–464
Hill D, Herr H (2013) Effects of a powered ankle-foot prosthesis on kinetic loading of the
contralateral limb: a case series. 2013 I.E. International Conference on Rehabilitation Robotics,
June 24–26, Seattle, Washington
Hofstad C, Linde H, Limbeek J, Postema K (2004) Prescription of prosthetic ankle-foot mecha-
nisms after lower limb amputation. Cochrane Database Syst Rev (1):CD003978
ISO/TS 16955:2016-06 (2016) Prosthetics – quantification of physical parameters of ankle foot
devices and foot units
Jin L, Adamczyk PG, Roland M, Hahn ME (2016) The effect of high- and low-damping prosthetic
foot structures on knee loading in the uninvolved limb across different walking speeds. J Appl
Biomech 32:233–240
Johnson L, De Asha AR, Munjal R, Kulkarni J, Buckley JG (2014) Toe clearance when walking in
people with unilateral transtibial amputation: effects of passive hydraulic ankle. J Rehabil Res
Dev 51(3):429–438
Kabra SG (1990) Articulated cadaveric bones as a structural endoskeleton in an ankle-foot
prosthesis: a preliminary report. J Rehabil Res Dev 27(1):43–52
Kabra SG, Narayanan R (1991) Ankle-foot prosthesis with articulated human bone endoskeleton:
force-deflection and fatigue study. J Rehabil Res Dev 28(3):13–22
Kim M, Collins SH (2015) Once-per-step control of ankle-foot prosthesis push-off work reduces
effort associated with balance during walking. J Neuroeng Rehabil 12(43):1–13
Klodd E, Hansen A, Fatone S, Edwards M (2010a) Effects of prosthetic foot forefoot flexibility on
gait of unilateral transtibial prosthesis users. J Rehabil Res Dev 47(9):899–910
Klodd E, Hansen A, Fatone S, Edwards M (2010b) Effects of prosthetic foot forefoot flexibility on
oxygen cost and subjective preference rankings of unilateral transtibial prosthesis users.
J Rehabil Res Dev 47(6):543–552
Klute GK, Kallfelz CF, Czerniecki JM (2001) Mechanical properties of prosthetic limbs: adapting
to the patient. J Rehabil Res Dev 38(3):299–307
Knox E (1996) The role of prosthetic feet in walking. PhD thesis, Department of Biomedical
Engineering, Northwestern University, Evanston
Lehmann JF, Price R, Boswell-Bessette S, Dralle A, Questad K (1993) Comprehensive analysis of
dynamic elastic response feet: Seattle Ankle/Lite foot versus SACH foot. Arch Phys Med
Rehabil 74:853–861
Major MJ, Twiste M, Kenney LPJ, Howard D (2011) Amputee independent prosthesis properties –
a new model for description and measurement. J Biomech 44:2572–2575
Major MJ, Twiste M, Kenney LPH, Howard D (2014) The effects of prosthetic ankle stiffness on
ankle and knee kinematics, prosthetic limb loading, and net metabolic cost of trans-tibial
amputee gait. Clin Biomech 29:98–104
Miller WC, Deathe AB, Speechley M, Koval J (2001) The influence of falling, fear of falling, and
balance confidence on prosthetic mobility and social activity among individuals with a lower
extremity amputation. Arch Phys Med Rehabil 82(9):1238–1244
Morgenroth DC, Segal AD, Zelik KE, Czerniecki JM, Klute GK, Adamczyk PG, Orendurff MS,
Hahn ME, Collins SH, Kuo AD (2011) The effect of prosthetic foot push-off on mechanical
loading associated with knee osteoarthritis in lower extremity amputees. Gait Posture
34:502–507
Nickel E, Sensinger J, Hansen A (2014) Passive prosthetic ankle-foot mechanism for automatic
adaptation to sloped surfaces. J Rehabil Res Dev 51(5):803–814
Pitkin MR (1995) Mechanical outcomes of a rolling-joint prosthetic foot and its performance in the
dorsiflexion phase of transtibial amputee gait. J Prosthet Orthot 7(4):114–123
Postema K, Hermens HJ, de Vries J, Koopman HF, Eisma WH (1997) Energy storage and release of
prosthetic feet. Part 1: biomechanical analysis related to user benefits. Prosthetics Orthot Int 21
(1):17–27
Raschke SU, Orendurff MS, Mattie JL, Kenyon DEA, Jones OY, Moe D, Winder L, Wong AS,
Moreno-Hernandez A, Highsmith MJ, Sanderson DJ, Kobayashi T (2015) Biomechanical
characteristics, patient preference and activity level with different prosthetic feet: a randomized
double blind trial with laboratory and community testing. J Biomech 48:146–152
Rosenblatt NJ, Bauer A, Grabiner MD (2016) Relating minimum toe clearance to prospective, self-
reported, trip-related stumbles in the community. Prosthetics Orthot Int Epub ahead of print.
https://www.ncbi.nlm.nih.gov/pubmed/27280640
Rusaw D, Ramstrand N (2010) Sagittal plane position of the functional joint centre of prosthetic
foot/ankle mechanisms. Clin Biomech 25:713–720
Russell Esposito E, Wilken JM (2014) Biomechanical risk factors for knee osteoarthritis when
using passive and powered ankle-foot prostheses. Clin Biomech 29:1186–1192
Sawers A, Hahn ME (2011) Trajectory of the center of rotation in non-articulated energy storage
and return prosthetic feet. J Biomech 44:1673–1677
Segal AD, Zelik KE, Klute GK, Morgenroth DC, Hahn ME, Orendurff MS, Adamczyk PG, Collins
SH, Kuo AD, Czerniecki JM (2012) The effects of a controlled energy storage and return
prototype prosthetic foot on transtibial amputee ambulation. Hum Mov Sci 31(4):918–931
Sensinger JW, Intawachirarat N, Gard SA (2013) Contribution of prosthetic knee and ankle
mechanisms to swing-phase foot clearance. IEEE Trans Neural Syst Rehabil Eng 21(1):74–80
Su PF, Gard SA, Lipschutz RD, Kuiken TA (2010) The effects of increased prosthetic ankle motions
on the gait of persons with bilateral transtibial amputations. Am J Phys Med Rehabil 89
(1):34–47
Van Jaarsveld HWL, Grootenboer HJ, De Vries J, Koopman HFJM (1990) Stiffness and hysteresis
properties of some prosthetic feet. Prosthetics Orthot Int 14:117–124
Ventura JD, Klute GK, Neptune RR (2011) The effect of prosthetic ankle energy storage and return
properties on muscle activity in below-knee amputee walking. Gait Posture 33:220–226
Williams RJ, Hansen AH, Gard SA (2009) Prosthetic ankle-foot mechanism capable of automatic
adaptation to the walking surface. J Biomech Eng 131:034002-1-7
Zelik KE, Collins SH, Adamczyk PG, Segal AD, Klute GK, Morgenroth DC, Hahn ME, Orendurff
MS, Czerniecki JM, Kuo AD (2011) Systematic variation of prosthetic foot spring affects
center-of-mass mechanics and metabolic cost during walking. IEEE Trans Neural Syst Rehabil
Eng 19(4):411–419
Zmitrewicz RJ, Neptune RR, Walden JG, Rogers WE, Bosker GW (2006) The effect of foot and
ankle prosthetic components on braking and propulsive impulses during transtibial amputee
gait. Arch Phys Med Rehabil 87:1334–1339
The Influence of Prosthetic Knee Joints
on Gait
Steven A. Gard
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Manual Locking Knee Joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Single-Axis Knees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Polycentric Knees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Swing-Phase Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Stance Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Microprocessor-Controlled Knee Joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Effect of Prosthetic Knee Designs on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Temporal-Spatial Measures of Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Kinematics and Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Energy Expenditure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Subjective Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Abstract
Transfemoral prostheses are intended to restore function and cosmesis in persons
with limb loss. Lower-limb prostheses are generally intended to restore ambula-
tion, a periodic activity in which the primary concern for the device is for the
provision of natural, efficient movement. Of all the elements affecting
S.A. Gard (*)

Department of Physical Medicine & Rehabilitation, Feinberg School of Medicine, Northwestern
University Prosthetics-Orthotics Center (NUPOC), Chicago, IL, USA
Department of Physical Medicine & Rehabilitation, Feinberg School of Medicine, Northwestern
University, Chicago, IL, USA
Jesse Brown VA Medical Center, Department of Veterans Affairs, Chicago, IL, USA
e-mail: sgard@northwestern.edu

DOI 10.1007/978-3-319-30808-1_75-1
2 S.A. Gard
locomotion, those most amenable to change relate to the device; thus, in the
description of walking patterns, emphasis should be placed on prosthetic design,
alignment, and fit. The designs of commercially available prosthetic knee units
are generally biomimetic in nature, and their functions are fundamentally simi-
lar—the prosthetic knee must provide stability during stance phase to ensure that
the user is safely supported on their prosthesis, and it must flex during swing
phase to shorten the prosthesis and allow the user to advance the limb. However,
different prosthetic knee designs offer different features above and beyond these
minimum requirements, and selection of a particular knee joint depends upon an
individual’s functional needs and abilities.
Keywords
Prosthesis • Prosthetics • Amputation • Lower limb • Knee joint
Introduction
The anatomical knee is a complex, polycentric joint characterized by a large range of

motion in the sagittal plane and limited mobility in the coronal and transverse planes.
The sagittal plane motion is used for progression in stance phase and limb clearance
and advancement in swing (Perry and Burnfield 2010). The knee joint undergoes two
cycles of flexion with each step (Fig. 1), larger in the swing phase and smaller during
stance, with each motion providing important but distinctly different functions.
Swing-phase knee flexion is initiated during preswing and is generally on the
order of 60–70 . Flexion of the knee during swing is necessary to shorten the leg
in order to achieve sufficient foot clearance as the leg is brought from behind to in
front of the body to execute the step (Moosabhoy and Gard 2006). Stance-phase
knee flexion commences at initial contact, increases through loading response phase,
and peaks at about contralateral toe-off, and then the knee extends through mid-
stance. At freely selected speeds, the stance-phase knee flexion wave has a magni-
tude of approximately 10–20 . Stance-phase knee flexion is the primary contributor
to shock absorption during able-bodied walking (Gard and Childress 1999; Kaufman
and Sutherland 2006; Perry and Burnfield 2010). Thus, the knee joint influences the
function of the entire leg during both swing and stance.
Limb prostheses are intended to restore function and cosmesis in persons with
limb loss. The designs of commercially available prosthetic knee units are generally
biomimetic in nature, and their functions are fundamentally similar. At a minimum,
the prosthetic knee must provide stability during stance phase to ensure that the user
is safely supported on their prosthesis, and it must flex during swing phase to shorten
the prosthesis and allow the user to advance the limb. However, different prosthetic
knee designs offer different features above and beyond these minimum require-
ments, and selection of a particular knee joint depends upon an individual’s func-
tional needs and abilities. The ideal prosthetic knee unit should mimic the anatomical
knee joint’s function during gait (Michael 1994) and thus should:
The Influence of Prosthetic Knee Joints on Gait 3
Fig. 1 Kinematics of the anatomical knee joint during able-bodied walking. The knee normally
undergoes two waves of flexion during the gait cycle, with the primary rotation occurring during
swing phase and a smaller amount of flexion occurring during stance
• Be just stable enough to accept weight in early stance phase

• Absorb shock and allow smooth forward progression of the body via controlled
knee flexion during the loading response phase
• Support the body on a bent knee during midstance
• Begin flexion during single-limb weight-bearing late in terminal stance phase
• Respond instantly in the swing phase to a faster pace and to variable cadences,
continuing the smooth, energy-efficient forward progression of the body
These desired features have been implemented to varying extents in the different
knee designs, resulting in nonexclusive categories of prosthetic knee functions. The
various functions provided by prosthetic knee units affect gait in different ways.
There are a number of different gait parameters that can be used to characterize
human walking. Temporal-spatial parameters, such as walking speed, cadence, and
step length, are useful because they provide gross indications of overall walking
ability, and many of them can be made with simple clinical methods or tools. Other
measures, such as joint kinematics and kinetics, ground reaction forces, and energy
expenditure, provide a more comprehensive assessment of an individual’s walking
ability, though acquiring these data generally require more sophisticated, expensive
equipment that may not be practical for use in a clinical setting. Nonetheless, review
of these data is useful for clinicians because they increase understanding about how
different prosthetic componentries affect walking performance. Finally, it is useful to
consider the perceptions of the prosthesis user, since this feedback can provide
tremendous insight into how the prosthesis affects stability, balance, control, and
comfort during walking. Therefore, the purpose of this communication is to review
the different categories of prosthetic knee designs and report how they affect the
4 S.A. Gard
walking performance of transfemoral prosthesis users. This document is not

intended to be a comprehensive review on this topic, but instead provide an
overview of some of the more typical gait outcomes that are associated with the
different prosthetic knee designs.
State of the Art
Manual Locking Knee Joints
Prosthetic knees are generally classified according to their designs and the features
they provide (Michael 1994, 1999, 2004; Oberg and Kamwendo 1988; Schuch
1992; Shurr and Michael 2002). The ultimate in stance stability is achieved with a
manual locking knee joint (Michael 2004) (Fig. 2). The knee is unlocked for sitting
when the user pulls on a lever or cable. The manual locking knee is considered a last
resort by prosthetists, and they are fitted only if the prosthetist feels that the patient
would be at great risk of falling with an articulating knee joint. The manual locking
knee joint is locked in a fully extended position during the entire gait cycle, forcing
the user to have a pronounced limp or lurch. An individual walking with a manual
locking knee joint will be required to utilize compensatory actions such as hip
hiking, circumduction, or vaulting in order to achieve sufficient foot clearance
during prosthetic swing phase, which reduces gait aesthetics and increases energy
expenditure (Michael 1994). Manual locking knee joints are typically prescribed for
minimal ambulators and can be useful for standing and transfers. These devices are
intended for persons who simply cannot walk with any other type of knee joint due
to poor control or very weak hip musculature.
Fig. 2 Manual locking knee

joint
Single-Axis Knees
Single-axis knees—sometimes called “constant friction” knees—are the most basic

of prosthetic knee designs (Fig. 3). They are simple hinge mechanisms that allow for
free knee flexion, with a friction adjustment that is set by the prosthetist according to
the swing-phase requirements of the patient. The lack of swing-phase control permits
only a single fixed cadence, and the lack of stance control requires that the user rely
on their own muscle power to some extent in order to stabilize the knee joint during
standing and walking. The primary advantages of single-axis knees include low cost,
light weight, reliability, and relative durability because of their simplicity. Single-
axis knees are prescribed for children, who tend to be very energetic and outgrow
their prostheses quickly, and for individuals who tend to walk at the same speed all of
the time. Because of their simplicity and reliability, single-axis knees are also a good
choice for persons who do not live close to a prosthetic facility and cannot come in
regularly for maintenance of their prosthesis.
Polycentric Knees
Polycentric knee units (Fig. 4) are characterized by multiple axes of rotation that
produce complex knee motion, having an instantaneous center of rotation that is
determined by the knee flexion angle. Four-bar linkage knees are the most common
type of polycentric knees (Michael 1994, 2004), but there are also designs that
incorporate five, six, and even seven linkages. The knee units can be designed to
produce a desired path of the center of rotation, or “centrode,” which affects stance-
phase stability and ease of swing flexion. These knee units offer several distinct
advantages over single-axis knees. Stability in polycentric knee units is inherent to
the design, and some units can support body weight even while partially flexed.
They can offer both excellent stance stability and ease of swing-phase knee flexion.
Fig. 3 Single-axis knee

6 S.A. Gard
Fig. 4 Polycentric knees
There are polycentric knee units that are designed to offer improved cosmesis by
enabling the instant center of rotation to be placed near to the knee joint’s anatomical
position, a feature that is especially beneficial for individuals with relatively long
residual limbs that do not permit a single-axis knee unit to be physically placed at the
level of the knee’s anatomical axis of rotation (Greene 1983). Disadvantages of
polycentric knees include their increased weight, higher cost, and greater need for
maintenance compared with single-axis knees.
Swing-Phase Control
Some prosthetic knee units offer “swing-phase control” (Fig. 5), which simulates the
normal dampening action of the anatomical knee’s flexors and extensors during the
swing phase of able-bodied gait. This feature enables prosthetic users to walk with a
variable cadence and with a greater range of walking speeds (Godfrey et al. 1975;
Michael 2004; Murphy 1964; Radcliffe 1977). Swing-phase control is provided by a
fluid-filled cylinder, either pneumatic or hydraulic, that changes damping resistance
in proportion to the velocity of knee flexion/extension. This enables the knee unit to
automatically adjust during swing phase so there is greater resistance to knee motion
at faster speeds and less resistance at slower speeds, producing a smoother gait with
near-normal gait parameters. Furthermore, swing-phase control may encourage a
more normal pattern of hip motion since the user will be less inclined to compensate
for prosthetic knee deficiencies that might occur with changing walking speeds. For
this reason, prosthetic knees with swing-phase control are sometimes referred to as
being “cadence responsive knees” since they can accommodate to a range of
different cadences (corresponding to different walking speeds) (Godfrey et al.
1975). Pneumatic devices have been found to be best suited to low-active to
moderately active individuals, while hydraulic mechanisms are suitable for a broad
Fig. 5 Single-axis knee with

swing and stance control
range of activity levels. Disadvantages of swing-phase control mechanisms include

increased weight, higher cost, and greater maintenance.
Stance Control
Prosthetic knee units with “stance control” utilize mechanisms that increase resis-
tance to knee motion during weight-bearing (Michael 1994, 2004). They provide
additional safety and stability during stance phase, particularly if the user loads a
partially flexed knee while they are walking. Stance control mechanisms are fre-
quently used in persons undergoing initial gait training, in individuals who have
limited control over their knee joint during stance phase due to weakened hip
musculature, and in limited ambulators who are unable to initiate or control pros-
thetic swing phase. A significant drawback to some stance control knees is that they
must be completely unloaded before knee flexion can occur. This can make it
difficult for the user with a unilateral amputation to sit down unless body weight is
completely shifted to the sound leg. Because of this issue, stance control mecha-
nisms may be impractical for some bilateral transfemoral amputees, though a single
unit could feasibly be used on only one side. Users of stance control devices may
also find it difficult to initiate swing phase on the prosthesis while it is partially
8 S.A. Gard
loaded, which may encourage them to utilize compensatory strategies and thus
increase the energy that must be expended to walk. Another disadvantage of these
knees is that they require periodic maintenance of the brake or resistance
mechanism.
Microprocessor-Controlled Knee Joints
The newest development in prosthetic knee designs is microprocessor-controlled

devices (Michael 1999, 2004). Microprocessor-controlled knees (MCKs) do not
utilize motors to power joint rotation, but they adjust damping in the knee to improve
swing- and/or stance-phase control as appropriate for the user during gait (Fig. 6).
Transducers in the knee units, such as strain gauges and goniometers, monitor
loading and the state of the knee during gait, and the microprocessor adjusts the
knee’s performance accordingly through feedback to operate mechanical and
hydraulic systems. Some of the newer microprocessor-controlled knees currently
Fig. 6 Microprocessor-
controlled knee
on the market utilize adaptive control to automatically adjust to a user’s gait over
time, so less setup and fine-tuning of the device is required by the prosthetist.
Emphasis, too, is being placed on developing knees that are less costly, less complex,
and more dependable than earlier microprocessor-controlled designs. The primary
disadvantage of these devices at present is their cost to the user or insurance provider.
Some clinicians and consumers may be under the impression that the most expensive
prosthetic component provides the maximum benefit, but this is not necessarily true.
There are some very simple and relatively low-cost mechanical prosthetic knee
components that provide significant functional restoration. Nonetheless, the appli-
cation of microprocessor control in lower-limb components is expected to increase
steadily in the future (Michael 2004).
Effect of Prosthetic Knee Designs on Gait
Fundamentally, the prosthetist’s goal in fitting a lower-limb amputee is to restore the

ability to perform everyday activities in an easy, natural, and comfortable manner.
The prosthesis user’s basic requirements of the prosthetic limb are comfort, function,
and appearance, the latter embracing both cosmetic appearance and appearance in
use (Radcliffe 1970b). The goals of the practitioner and the patient are similar, but
prosthesis users may not be able to fully or accurately articulate what they perceive
when they stand and walk, making it difficult for the prosthetist to make all of the
necessary adjustments. Augmenting the human body with a prosthesis markedly
affects the individual’s mode of travel, and the task of the clinician should be to
recognize optimal gait with a given device so that departures from the standard can
be identified, their causes determined, and, wherever possible, corrected (Edelstein
1990). Of all the elements affecting locomotion, those most amenable to change
relate to the device; thus, in the description of walking patterns, emphasis should be
placed on prosthetic design, alignment, and fit.
Prosthetic alignment—the process that is utilized for spatial positioning of the
prosthetic components with respect to each other and to the residual anatomy of the
patient—is an important consideration for the proper functioning of a prosthetic knee
unit during walking (Jacobs 1988; Peizer and Gardner 1972; Radcliffe 1977). The
principles of prosthetic alignment are perhaps best understood with regard to single-
axis knees. For a transfemoral prosthesis (Fig. 7) the prosthetic foot must initially be
aligned (i.e., positioned and oriented) to the prosthetic socket, and then the prosthetic
knee unit is positioned relative to a vertical alignment line (also called the “weight
line” or “load line”) based upon landmarks on the socket and foot (Fig. 8). For
mechanical knees, stability is achieved by aligning the knee axis in the sagittal plane
so that the line of action of the load borne by the prosthesis passes anterior to the
knee axis and forces the prosthetic knee against a mechanical stop in the fully
extended position. Prosthetic alignment determines the relative stability of the user
on the prosthesis while standing and during the stance phase of walking and the
relative ease with which the user is able to walk with the prosthetic limb. Positioning
10 S.A. Gard
Fig. 7 Transfemoral
prosthesis
Fig. 8 Trochanter-knee-
ankle (TKA) alignment line
for transfemoral prosthesis
the prosthetic knee’s axis of rotation more posterior to the alignment line provides
greater stance stability, but breaking the knee in late stance in preparation for swing
will be more difficult. Conversely, aligning the prosthetic knee so its axis of rotation
is closer to the alignment line will make the knee easier to flex in late stance phase,
but stability in stance phase may be challenged. Thus, there is a balance between
prosthesis stability and controllability that must be achieved by the prosthetist when
aligning the prosthetic knee joint. The principles of alignment provide guidelines for
fitting a wide variety of users, but it is up to the prosthetist to fine-tune the alignment
to best suit the needs and capabilities of the patient (Jacobs 1988).
Temporal-Spatial Measures of Gait
Temporal-spatial measures of prosthetic gait are appealing because many of them

can easily be made in the clinic with relatively simple tools. Freely selected walking
speed is often used as an overall indicator of the quality of gait, and it is well known
that the walking speed of prosthesis users generally decreases with more proximal
levels of amputation (Waters et al. 1976). Therefore, it is quite natural to assume that
an improved prosthesis design should enable users to walk at faster freely selected
speeds. Additionally, since walking speed is the product of step length and cadence,
these parameters would also be expected to improve with the utilization of more
functional components. Studies of transfemoral amputees walking with single-axis
knees generally indicate that the prosthesis users walk at slower freely selected
speeds than able-bodied individuals (0.9–1.1 m/sec compared to 1.4–1.5 m/sec)
and slower fast self-selected speeds (1.3–1.4 m/sec vs 2.0–2.1 m/sec) (Cappozzo
et al. 1976; Jaegers et al. 1995; James and Oberg 1973; Murray et al. 1980). The
primary reason for the slower walking speed appears to be associated with the
cadence, which is about 85 steps/min for the transfemoral prosthesis users compared
with 113 steps/min for able-bodied individuals (Jaegers et al. 1995; James and Oberg
1973; Murray et al. 1980). Step lengths are generally asymmetrical for unilateral
transfemoral prosthesis users, with the prosthetic step length being about 68–72 cm
and the sound step length being about 62–64 cm for unilateral amputees at freely
selected speeds of walking (James and Oberg 1973; Murray et al. 1980). The stance
phase is generally longer on the sound side of unilateral transfemoral prosthesis users
(Jaegers et al. 1995; James and Oberg 1973; Zuniga et al. 1972), attributed to a
greater feeling of security and stability while load bearing on the sound leg.
Similarly, the swing phase of the prosthetic limb is generally longer than that of
the sound leg (Jaegers et al. 1995; James and Oberg 1973; Zuniga et al. 1972).
Finally, step widths are typically wider among transfemoral prosthesis users, mea-
suring about 15–20 cm at freely selected speeds compared to about 7–8 cm for able-
bodied individuals (Jaegers et al. 1995; James and Oberg 1973; Murray et al. 1980;
Radcliffe 1970a). Transfemoral prosthesis users having amputation due to peripheral
vascular disease or diabetes generally have a poorer walking performance than
individuals with a traumatic transfemoral amputation.
12 S.A. Gard
The responsiveness of prosthetic knees with swing-phase control allows prosthe-

sis users to walk at faster freely selected and self-selected speeds (Boonstra et al.
1996; Godfrey et al. 1975; Murray et al. 1983), primarily attributable to a higher
cadence than a longer step length. However, step lengths between the sound and
prosthetic sides are generally more symmetric with knees that provide swing-phase
control (Murray et al. 1983). Step width is reportedly unchanged in comparisons
between knees with swing-phase control and single-axis knees, as are stance and
swing-phase durations.
Microprocessor-controlled knees (MCKs) have been reported to increase self-
selected walking speeds on both level and uneven terrain compared with walking
performance with mechanical knees (Chin et al. 1999; Kahle et al. 2008; Orendurff
et al. 2006; Perry et al. 2004; Segal et al. 2006). This presumably can be attributed to
the ability of the MCKs to fine-tune swing-phase response to permit a more
appropriate cadence for a particular walking speed. MCKs may also improve step
length symmetry (Kirker et al. 1996; Segal et al. 2006). However, when comparing
performance of users walking with MCKs and mechanical knees having swing
and/or stance control, there are also a number of studies that report no difference
in the typical temporal-spatial measures (Jepson et al. 2008; Johansson et al. 2005).
Studies involving step activity monitors that record the number of steps taken
during a specified period of time indicate that there is no difference between users of
MCKs and mechanical knees (Hafner et al. 2007; Klute et al. 2006). This is an
interesting finding, since it might reasonably be expected that prosthesis users would
significantly increase their activity if they are provided a more functional compo-
nent. However, it could be that prosthesis users, like able-bodied individuals, have a
set activity schedule that does not vary much from 1 day to the next. Therefore,
prosthesis users walk as much as they need to in order to get around their house,
travel to work, go to the store, etc. and that the fundamental level of activity is
unchanged with new componentry. That is not to say that it would not be easier and
more efficient for them to ambulate with a more functional component, but it may
not be possible to determine this from step activity monitor data.
Some MCKs are able to compensate and adapt to the demands of walking on
stairs, slopes, and irregular terrain (Buckley et al. 1997; Chin et al. 1999; Heller et al.
2000; Kirker et al. 1996; Taylor et al. 1996). Therefore, the use of an obstacle course
for assessing performance of prosthesis users could help determine improved func-
tionality of one knee prosthetic design compared to another. Protocols are typically
designed so that subjects are timed as they navigate the obstacle course, and
presumably the component that provides superior function will enable a shorter
overall traverse time. The course will typically incorporate multiple turns, steps up
and down, uneven terrains, slaloms around obstacles, and different compliant sur-
faces to represent an environment that individuals are likely to encounter in the real
world (Meier et al. 2012). A primary concern, however, is how to standardize the
obstacle course to create a repeatable test that can be duplicated and validated by
other investigators. Nonetheless, results thus far indicate that an MCK increases
functional mobility and ease of performance on obstacle courses compared to a
mechanical knee (Hafner et al. 2007; Seymour et al. 2007).
Kinematics and Kinetics
Gait kinematics and kinetics are generally similar between transfemoral prosthesis
users at freely selected speeds of walking regardless of the different prosthetic knee
designs. Of particular concern are the sagittal plane motions of the prosthetic side hip
and knee joints, since it is the residual limb hip that helps to stabilize the prosthetic
knee during stance phase and controls the passive knee motion during swing phase.
Rapid flexion of the hip joint is required to create the swing-phase knee flexion to
shorten the prosthetic leg and provide sufficient foot clearance. Specifically, the
distance between the hip joint center and the toe must be shortened to swing the leg
forward from behind the body, and the primary means for achieving this shortening
is through knee flexion. Simulations have demonstrated that the hip-toe distance
increases in length during the first 20 of knee flexion, shortens to standing length at
about 40 of flexion, and does not actually begin shortening for foot clearance until
knee flexion exceeds 40 (Gard et al. 1996). Therefore, a maximum swing-phase
knee flexion of less than about 40 actually hinders foot clearance instead of
facilitating it. Foot clearance for able-bodied ambulators walking at freely selected
speeds has been reported to be about 1.3 cm on average (Winter 1992). However,
transfemoral amputees seem to prefer an increased margin of safety for prosthetic
foot clearance even if they can achieve sufficient knee flexion during swing (Murray
et al. 1980). The amount of prosthetic swing-phase foot clearance perceived by users
may be influenced by the inability of the prosthetic foot to dorsiflex during swing
phase combined with some degree of longitudinal displacement of the prosthetic
socket on the residual limb due to the pull of gravity (Cappozzo et al. 1976).
Therefore, it is not uncommon for transfemoral prosthesis users to utilize vaulting,
circumduction, and/or hip hiking during walking to further increase swing-phase
foot clearance beyond that provided by prosthetic knee flexion. Lateral trunk
bending over the prosthesis during single-support phase is often employed, too,
particularly for users with short residual limbs and wider step widths (Jaegers et al.
1995; James and Oberg 1973; Murray et al. 1983; Murray et al. 1980). All prosthetic
knees (except for manual locking knees) can be set up to provide sufficient swing-
phase knee flexion at freely selected speeds of walking, though some of the more
sophisticated designs enable broader ranges of walking speeds with more fine-tuned
damping resistance. Stance-phase knee flexion, on the other hand, is generally not
provided by prosthetic knee units (James and Oberg 1973; Murray et al. 1980),
except for a few that are specifically designed to provide this feature. Therefore, the
prosthetic knee typically remains in full extension during the stance phase of
walking. Whereas the hip joint of the able-bodied ambulator generally maintains a
constant flexed posture during early stance phase due to the accommodation of
stance-phase knee flexion, the hip joint of the prosthetic side generally extends
rapidly upon initial contact in order to stabilize the knee and advance the body
forward (Jaegers et al. 1995; Murray et al. 1983).
When the frictional resistance of single-axis knees is appropriately tuned by the
prosthetist, transfemoral prosthesis users can walk at their freely selected speeds with
reasonably good swing-phase knee flexion that approximates that of able-bodied
14 S.A. Gard
ambulators in magnitude and timing. However, single-axis knees can only be set up
for a single walking speed (or cadence) because the friction resistance to knee flexion
is constant and does not vary with rotational velocity (Michael 1994). Therefore, a
transfemoral prosthesis user walking at faster speeds will typically generate too
much knee flexion during swing phase or create insufficient knee flexion at slower
walking speeds (James and Oberg 1973; Murray et al. 1983). Hence, single-axis
knees are generally prescribed for individuals who only walk at a single speed.
The muscles acting about the residual limb joint are the only power source
available to the prosthesis user to manipulate and control the prosthesis during the
various phases of the gait cycle (Murphy 1964; Yang et al. 1991). Therefore, it is
useful to consider how the hip joint moments on the prosthetic side are affected by
changes in prosthetic knee design. During able-bodied walking, the internal hip joint
moment is extensor during the first portion of stance phase as the body is elevated
upward and proceeds forward onto the supporting leg. As the body surpasses the
supporting leg, the internal hip moment becomes flexor to decelerate the forward fall
of the body in preparation for contralateral initial contact. Generally, the stance-
phase hip moment for unilateral transfemoral prosthesis users follows a similar
pattern as that of able-bodied individuals. However, the prosthetic alignment of a
single-axis knee unit, affecting both stability and controllability of the prosthesis,
highly influences the user’s hip moment of their residual limb during walking.
Studies have demonstrated that a more unstable prosthetic knee alignment (i.e.,
posterior alignment) produces higher peak hip extensor moments in early stance
phase, while more stable knee alignment (i.e., anterior alignment) can reduce the
peak hip extensor moment (Koehler-McNicholas et al. 2016; Schmalz et al. 2002). It
appears that prosthesis users make these accommodations at their hip joints in order
to maintain the prosthetic knee in extension during stance phase, but particularly
during loading response phase when the ground reaction force vector is oriented so
that it passes close to the knee’s axis of rotation. Furthermore, prosthetic alignment
that decreases knee stability in stance is often accompanied by greater anterior pelvic
rotation and increased trunk flexion (Koehler-McNicholas et al. 2016) in an effort to
manipulate the ground reaction force vector orientation relative to the prosthetic
knee’s axis of rotation.
Prosthetic knees with swing-phase control allow for a broader range of walking
speeds because they incorporate damping having a flexion resistance that varies with
the speed (or cadence) of walking. That is, greater resistance is produced at faster
walking speeds, and less resistance is generated at slow walking speeds, so the
magnitude of swing-phase knee flexion remains roughly the same at any speed
(Murray et al. 1983).
Polycentric knee units have also been shown to offer increased foot clearance
during swing phase compared with a single-axis knee (Gard et al. 1996; Greene
1983; Sensinger et al. 2013), which may decrease the likelihood of the user stubbing
their toe or tripping during prosthetic swing. Four-bar linkage knees may also allow
for a closer leg-length equality in unilateral transfemoral amputees (Gard et al.
Fig. 9 Polycentric knee

designs that provide stance-
phase knee flexion
1996). The length of the transfemoral prosthesis is often fabricated by the prosthetist
to be slightly shorter than the length of the sound leg to aid clearance during swing
phase, but because of the greater floor clearance afforded by four-bar knees, the
prosthetic length can be made closer to the full length. Furthermore, the additional
foot clearance provided by polycentric knees may offer particular advantages for
persons with bilateral amputations, enabling them to reduce or eliminate compen-
satory actions for swing-phase toe clearance.
Some polycentric prosthetic knee units provide stance-phase knee flexion (Fig. 9)
and permit controlled flexion during early stance phase when shock absorption is
required (Blumentritt et al. 1997; Sutherland et al. 1997). These knee units are an
attractive option for transfemoral amputees because they are designed to replicate an
anatomical mechanism for shock absorption that occurs in able-bodied gait. The
stance-phase knee flexion is generally limited to about 15 and occurs against a
rubber bumper, which serves to absorb energy and reduce the impact forces associ-
ated with load transfer while providing a positive flexion stop. After flexing in early
stance phase, the knees fully extend by the middle of stance phase, restoring the
prosthesis to standing length. These knee units are multi-linkage designs that have
separate mechanisms that provide for swing-phase and stance-phase knee flexion.
Stability in stance phase is assured by their inherent designs, so users do not need to
16 S.A. Gard
be concerned about the knee collapsing under them as they load it. Unfortunately,
some users find the stance flexion feature to be disconcerting, and many adopt
walking patterns that prevent the stance-phase flexion from occurring (Koehler
et al. 2004), thus eliminating any potential shock absorption benefits. Stance-flexion
knees should probably be prescribed early in an amputee’s rehabilitation before they
become accustomed to walking with other types of knee units that require the
utilization of a strong hip extensor moment for stabilization in stance. Experienced
prosthetic users who are fitted with one of these devices may have to undergo
extensive gait retraining to learn how to use the knee unit properly. Stance-phase
knee flexion units are best utilized for moderately to very active users.
Microprocessor-controlled knees (MCKs) constantly monitor knee joint forces
and moments during gait and fine-tune the joint’s resistance to motion accordingly.
Some studies have reported significant differences in hip, knee, and/or ankle kine-
matics and kinetics as a result between users of MCKs and mechanical knees
(Johansson et al. 2005; Kaufman et al. 2007). Swing-phase knee flexion is reportedly
improved with MCKs compared to mechanical knees (Herr and Wilkenfeld 2003),
owing to their ability to fine-tune damping to more appropriate values. Presumably,
this ability to provide appropriate damping would be more evident at slower and
faster walking speeds since greater tuning would be required. The ability to achieve
stance-phase knee flexion is often reported as a key benefit of MCKs (Herr and
Wilkenfeld 2003; Kaufman et al. 2007; Schmalz et al. 2014; Willingham et al. 2004).
However, it is not known at this time how the stance-phase knee flexion provided by
MCKs affects gait or whether it actually improves shock absorption during the
loading response phase.
As with mechanical knee units, prosthetic alignment is crucial to the proper
functioning of MCKs. One study analyzed the prosthetic alignment of MCK users
and reported that most of the prostheses were aligned so that they were too stable
(Willingham et al. 2004). Visual gait analyses supported this result, identifying gait
deviations in the majority of subjects consistent with walking on an overly stable
prosthesis. This finding is significant because it suggests that some MCK users may
not be deriving the full benefit afforded by this sophisticated technology if prosthet-
ists are not aligning the prosthesis appropriately. Prosthetists should be aware that
stance stability needs to be controlled by the MCK electronics and not by conven-
tional prosthetic alignment techniques.
Most investigations of subjects walking with MCKs involve assessments on level
ground, while limited analyses have been performed on subjects walking on uneven
terrain. A study of subjects descending stairs with an MCK indicated differences in
hip, knee, and ankle joint kinematics and kinetics between the prosthetic and sound
limbs and was also distinct from those of able-bodied control subjects (Schmalz et al.
2007). However, since no other prosthetic knees were analyzed, interpretation of the
data was difficult, and it could not be determined if the MCK actually provided any
functional advantage while descending stairs. More of these types of evaluations are
required to better understand the benefits afforded by more functional components
while users perform routine and typical activities of daily living.
Energy Expenditure
Energy expenditure is a gross measure of walking that appears to reflect the function
imparted by the prosthesis to the user. In many cases, the measurement of energy
expenditure is sufficiently sensitive to changes in the designs of prosthetic compo-
nents. Modern gait assessment laboratories generally determine energy expenditure
by indirect calorimetry based upon a person’s oxygen consumption during steady-
state walking. Energy expenditure is generally normalized to a subject’s body mass
and expressed as either energy rate (i.e., energy expended per unit time (ml O2/
kgmin)) or energy cost (i.e., energy expended per unit distance (ml O2/kgm)).
Differentiation between these two measures of energy expenditure is crucial for
understanding the effect of different prosthetic designs on gait efficiency.
The energy expenditure of walking at freely selected speeds by lower-limb
amputees increases with more proximal levels of amputation (Waters and Mulroy
1999; Waters et al. 1976; Waters and Yakura 1989). Transfemoral prosthesis users
walking at freely selected speeds with single-axis knees typically have an oxygen
cost that is 25–35 % higher than able-bodied individuals (Jaegers et al. 1993), with
the prosthesis users having an energy cost of about 0.20–0.30 ml/kgm (Waters and
Mulroy 1999; Waters et al. 1976) compared to able-bodied ambulators having about
0.15 ml/ kgm. Furthermore, vascular transfemoral amputees typically expend more
energy (0.28 ml/kgm) than traumatic amputees (0.20 ml/kgm) (Waters and Yakura
1989). Interestingly, the rate of energy expenditure of amputees walking at their
freely selected speeds is about 12.6–12.9 ml/kgsec, which is comparable to that of
able-bodied ambulators (Jaegers et al. 1993; James 1973; Waters et al. 1976). Thus,
it appears that amputees modify their gait so that their rate of energy expenditure is
on par with that of able-bodied ambulators. The way that amputees accomplish this
is by reducing their freely selected walking speed. Surprisingly, transfemoral ampu-
tees reportedly expend the same amount of energy whether they walk with their
knees locked or unlocked (Isakov et al. 1985; Traugh et al. 1975).
Results reported from energy expenditure studies involving MCKs are quite
variable. Some studies have reported small but significant reductions in energy
expenditure when subjects walk with MCKs compared to mechanical knees (Buck-
ley et al. 1997; Johansson et al. 2005; Perry et al. 2004; Schmalz et al. 2002;
Seymour et al. 2007; Taylor et al. 1996). However, other studies similarly report
small decreases in energy expenditure of subjects walking with MCKs compared to
that on mechanical knees, but the differences are not statistically significant (Datta
et al. 2005; Kaufman et al. 2008; Kirker et al. 1996; Orendurff et al. 2006). It should
be pointed out that, in spite of some slight reductions in energy expenditure of users
walking with MCKs, the energy consumption is still considerably greater than that of
able-bodied individuals. When differences in energy expenditure are small and
borderline statistically significant, the clinical significance of the finding should be
called into question. However, over the course of a day, even small differences in
energy expenditure could produce a perceptible change to the user.
Studies that have utilized physiological cost index (PCI) or heart rate to estimate
energy expenditure generally report little or no difference when comparing MCKs to
18 S.A. Gard
mechanical knees (Jepson et al. 2008; Perry et al. 2004; Seymour et al. 2007).
Results are typically highly variable with no distinct trend one way or the other.
The problem, in this case, does not appear to be with the intervention, but with using
PCI and/or heart rate to estimate energy expenditure. Because of their inconsistent
results, investigators should be discouraged from using these parameters to estimate
energy expenditure.
Subjective Measures
At this time, subjective measures of user perception and quality of life provide the
strongest evidence for demonstrating efficacy of MCKs. No doubt, these types of
measures capture the anecdotal evidence often reported by prosthetists and prosthe-
sis users. One of the primary advantages of these types of measures is that they can
easily be administered in the clinic and used to document patients’ progress over
time. However, many more studies need to be conducted to validate previous
findings. Particularly, users often report decreased frequency of stumbles and falls
(Hafner et al. 2007; Kahle et al. 2008; Orendurff et al. 2006).
Results from the prosthesis evaluation questionnaire (PEQ) generally indicate
that MCKs improve performance and increase satisfaction among prosthesis users
(Hafner et al. 2007; Kahle et al. 2008; Kaufman et al. 2008; Orendurff et al. 2006).
Specifically, users have indicated improved ambulation, reduce frustration,
decreased sounds, and increased utility with MCKs. Subjectively, most subjects
report a strong preference for MCKs (Kahle et al. 2008; Kirker et al. 1996).
However, the lack of blinding in previous studies may have unduly influenced
subjects’ opinions, since there could have been perceived benefit from being fitted
with a more advanced technological device.
As an indication of improved quality of life and ambulation ability, the Medicare
Functional Classification Level (MFCL) classification has reportedly been increased
from K2 to K3 for some MCK users (Kahle et al. 2008; Theeven et al. 2011). The
mechanism for this improvement is not known, but it may be related to the increased
stability and improved confidence of users while walking with an MCK (Segal et al.
2006). Nonetheless, it suggests that functional performance may be improved with
the use of MCKs, warranting further study.
One of the long-claimed benefits of MCKs is that they reduce cognitive burden
(Bellmann et al. 2010); that is, users no longer have to concentrate as intently on
walking or other activities involving the prosthesis since the computer control adapts
and monitors stability automatically. However, finding a quantitative measure in
support of this assertion is difficult. Many users exhibit greater confidence during
walking due to the consistency of the microprocessor control, and there is prelim-
inary evidence that indicates the cognitive demand is reduced so the wearer is able to
walk naturally without having to think about using the prosthesis (Heller et al. 2000).
However, objective measures of cognitive performance have demonstrated no sta-
tistically significant difference in subjects walking with MCKs and mechanical
knees (Hafner et al. 2007; Heller et al. 2000; Williams et al. 2006). Conversely,
however, subjective assessment indicates that MCKs require less cognitive attention
than mechanical knees (Williams et al. 2006). It appears that the difference in
cognitive burden is slight, but obviously noticeable to users, suggesting that more
sensitive measures or more demanding tasks will be required to provide the neces-
sary objective, quantitative evidence.
Conclusion/Summary
Limb prostheses are intended to restore function and cosmesis in persons with limb
loss. The complexity and functionality of prosthetic components have advanced
significantly as technology has improved, and knowledge about human movement
has increased, but a state-of-the-art artificial limb is still a relatively poor substitute
for an anatomical one. Lower-limb prostheses are generally intended to restore
ambulation, a periodic activity in which the primary concern for the device is for
the provision of natural, efficient movement. They typically consist of passive
components, though over the past decade, there have been several new develop-
ments in prosthetic knee mechanisms that incorporate microprocessors in order to
improve swing-phase characteristics and provide greater stability during stance
phase.
One of the primary criticisms of previous studies involving prosthetic knees
relates to experimental design and the number of independent variables. If investi-
gators want to determine how one knee design compares to another, then all other
aspects of the prosthesis must be appropriately controlled. The same prosthetic foot
must be used when comparing the different knees, while the socket and suspension
systems should be the same. Ideally, the only independent variable would be knee
type. Then, if a difference in the measured parameters is noted, it can be directly
attributed to the experimental intervention. However, this is often not the case.
Because some MCKs require they be fitted with only a few specific prosthetic
feet, researchers have had to make compromises in their study designs. Additionally,
since many studies have utilized subjects that are long-term users of mechanical
knees who are recruited as they begin to transition to a MCK, the subjects are often
prescribed new prosthetic sockets and suspension systems at the same time. There-
fore, at the completion of the study, the investigators have, in reality, compared an
old prosthesis to a brand new one, with the prosthetic knee being only one of several
significant differences.
Many practitioners may mistakenly believe that fitting a client with a new
prosthetic component and sending them out the door is sufficient for improving
their quality of life. However, too little attention appears to have been directed to
training (or retraining) the subjects to walk with the new component in order to both
break bad gait habits and to enable them to derive full benefit from capabilities
afforded by the modified prosthesis. Additionally, some of the results presented here
suggest that prosthesis users should undergo regular physical conditioning to further
improve their walking performance (Chin et al. 2006). A good start would be to train
individuals who have used their prosthetic knee designs for several years to
20 S.A. Gard
determine if prosthetic performance can be significantly improved through technique

and confidence building.
When comparing different prosthetic knee designs, more studies should be
conducted that investigate prosthesis users ambulating on surfaces other than flat,
smooth level ground and that are typically encountered during daily living (Hafner
et al. 2007; Johansson et al. 2005). Much of the anecdotal evidence reported by users
in favor of MCKs seems to stem from their use on uneven terrain (Hafner et al.
2007). A more useful approach would be to test subjects’ prosthetic performance
during distinct, well-defined activities such as walking up- and downstairs, on
ramps, on different compliant surfaces, etc. Instead of merely monitoring traverse
times, investigators should attempt to characterize and understand the biomechanics
of the movement and compare them among different components. It could be that,
fundamentally, all components perform the same unless significant functional
improvement is provided.
The programming of the MCKs deserves further attention. One study noted that
some previous users of mechanical hydraulic knees wanted their MCKs to be set up
so they would have a similar function as their passive counterpart (Orendurff et al.
2006). If this were done, would this negate the advantages of the MCKs? The
difference in the responsiveness and control of the MCK is typically not accounted
for in studies comparing them to mechanical knees, and yet this could help explain
the tremendous variability in the results, particularly those pertaining to energy
expenditure.
Further consideration should be given to the mechanisms that could potentially
affect energy expenditure in prosthesis users beyond that of the prosthetic compo-
nent. Should the incorporation of a novel prosthetic knee unit be expected to have a
significant effect on the energy required to walk? After all, energy expenditure
during stance phase would not necessarily be affected by the prosthetic knee joint
since the knee typically remains fully extended during this time. Even for those
subjects who achieve some degree of stance-phase knee flexion (which has not been
adequately demonstrated to date), how would this necessarily reduce energy expen-
diture? It probably would not if it is providing some restoration of shock absorption
during walking. What about swing phase? Since both MCKs and mechanical knees
are set to provide optimal performance at freely selected speeds, the damping of the
knees during swing would presumably be similar, and therefore the energy expen-
diture should be the same. However, at extremely slow and fast speeds, there may be
greater differences in the performance of the MCKs and mechanical knees, which
may be evident in energy expenditure measures assuming the MCKs are able to fine-
tune damping to more appropriate values compared to passive hydraulic mecha-
nisms. It has been suggested that the slight reduction in energy expenditure relates
not to prosthetic knee performance, but to the confidence users instill in the MCKs.
Because users can trust the MCKs not to buckle under load while they walk, they
may use less muscle activation during stance phase required to ensure stabilization of
mechanical knees and therefore reduce metabolic energy requirements. Evidence
suggests that slight differences exist in the energy expenditure of users when they
walk with MCKs and mechanical knees, but it is currently not known whether this
relates to the prosthetic knee performance, to user control, or to other potential

mechanisms that have yet to be identified.
There also needs to be recognition that not all prosthesis users will like walking
with a MCK. Some of the reasons for rejection of an MCK that have been cited in
previous studies include: satisfaction with their current mechanical knees, weight,
having to recharge the units, decreased cosmesis, less voluntary control, and cost
(Jepson et al. 2008; Kahle et al. 2008). It is important to document these concerns so
that technology can be further improved.
Cross-References

Analysis
References
Bellmann M, Schmalz T, Blumentritt S (2010) Comparative biomechanical analysis of current
microprocessor-controlled prosthetic knee joints. Arch Phys Med Rehabil 91(4):644–652.
doi:10.1016/j.apmr.2009.12.014
Blumentritt S, Scherer HW, Wellershaus U, Michael JW (1997) Design principles, biomechanical
data and clinical experience with a polycentric knee offering controlled stance phase knee
flexion: a preliminary report. J Prosthet Orthot 9(1):18–24
Boonstra AM, Schrama JM, Eisma WH, Hof AL, Fidler V (1996) Gait analysis of transfemoral
amputee patients using prostheses with two different knee joints. Arch Phys Med Rehabil 77
(5):515–520
Buckley JG, Spence WD, Solomonidis SE (1997) Energy cost of walking: comparison of "intelli-
gent prosthesis" with conventional mechanism. Arch Phys Med Rehabil 78(3):330–333
Cappozzo A, Figura F, Leo T, Marchetti M (1976) Biomechanical evaluation of above-knee
prosthetics. In: Komi PV (ed) Biomechanics V-A. University Park Press, Baltimore, pp 366–372
Chin T, Machida K, Sawamura S, Shiba R, Oyabu H, Nagakura Y et al (2006) Comparison of
different microprocessor controlled knee joints on the energy consumption during walking in
trans-femoral amputees: intelligent knee prosthesis (IP) versus C-leg. Prosthet Orthot Int 30
(1):73–80
Chin T, Sawamura S, Fujita H, Nakajima S, Ojima I, Oyabu H et al (1999) The efficacy of
physiological cost index (PCI) measurement of a subject walking with an Intelligent Prosthesis.
Prosthet Orthot Int 23(1):45–49
Datta D, Heller B, Howitt J (2005) A comparative evaluation of oxygen consumption and gait
pattern in amputees using Intelligent Prostheses and conventionally damped knee swing-phase
control. Clin Rehabil 19:398–403
Edelstein JE (1990) Prosthetic and orthotic gait. In: Smidt GL (ed) Gait in Rehabilitation. Churchill
Livingston Inc., New York, pp 281–300
Gard SA, Childress DS (1999) The influence of stance-phase knee flexion on the vertical displace-
ment of the trunk during normal walking. Arch Phys Med Rehabil 80(1):26–32
22 S.A. Gard
Gard SA, Childress DS, Uellendahl JE (1996) The influence of four-bar linkage knees on prosthetic
swing-phase foot clearance. J Prosthet Orthot 8(2):34–40
Godfrey CM, Jousse AT, Brett R, Butler JF (1975) A comparison of some gait characteristics with
six knee joints. Orthot Prosthet 29(3):33–38
Greene MP (1983) Four bar linkage knee analysis. Orthot Prosthet 37(1):15–24
Hafner BJ, Willingham LL, Buell NC, Allyn KJ, Smith DG (2007) Evaluation of function,
performance, and preference as transfemoral amputees transition from mechanical to micropro-
cessor control of the prosthetic knee. Arch Phys Med Rehabil 88(2):207–217
Heller BW, Datta D, Howitt J (2000) A pilot study comparing the cognitive demand of walking for
transfemoral amputees using the Intelligent Prosthesis with that using conventionally damped
knees. Clin Rehabil 14(5):518–522
Herr H, Wilkenfeld A (2003) User-adaptive control of a magnetorheological prosthetic knee. Ind
Robot 30(1):42–55
Isakov E, Susak Z, Becker E (1985) Energy expenditure and cardiac response in above-knee
amputees while using prostheses with open and locked knee mechanisms. Scand J Rehabil
Med Suppl 12:108–111
Jacobs NA (1988) Chap 16: Biomechanics of above-knee prostheses. In: Murdoch G, Donovan RG
(eds) Amputation surgery & lower limb prosthetics. Blackwell Scientific Publications, Oxford,
pp 130–139
Jaegers SM, Arendzen JH, de Jongh HJ (1995) Prosthetic gait of unilateral transfemoral amputees: a
kinematic study. Arch Phys Med Rehabil 76(8):736–743
Jaegers SM, Vos LD, Rispens P, Hof AL (1993) The relationship between comfortable and most
metabolically efficient walking speed in persons with unilateral above-knee amputation. Arch
James U (1973) Oxygen uptake and heart rate during prosthetic walking in healthy male unilateral
above-knee amputees. Scand J Rehabil Med 5(2):71–80
James U, Oberg K (1973) Prosthetic gait pattern in unilateral above-knee amputees. Scand J Rehabil
Med 5:35–50
Jepson F, Datta D, Harris I, Heller B, Howitt J, McLean J (2008) A comparative evaluation of the
Adaptive knee and Catech knee joints: a preliminary study. Prosthet Orthot Int 32(1):84–92
Johansson JL, Sherrill DM, Riley PO, Bonato P, Herr H (2005) A clinical comparison of variable-
damping and mechanically passive prosthetic knee devices. Am J Phys Med Rehabil 84
(8):563–575
Kahle JT, Highsmith MJ, Hubbard SL (2008) Comparison of nonmicroprocessor knee mechanism
versus C-Leg on Prosthesis Evaluation Questionnaire, stumbles, falls, walking tests, stair
descent, and knee preference. J Rehabil Res Dev 45(1):1–14
Kaufman KR, Levine JA, Brey RH, Iverson BK, McCrady SK, Padgett DJ, Joyner MJ (2007) Gait
and balance of transfemoral amputees using passive mechanical and microprocessor-controlled
prosthetic knees. Gait Posture 26(4):489–493
Kaufman KR, Levine JA, Brey RH, McCrady SK, Padgett DJ, Joyner MJ (2008) Energy expen-
diture and activity of transfemoral amputees using mechanical and microprocessor-controlled
prosthetic knees. Arch Phys Med Rehabil 89(7):1380–1385
Kaufman KR, Sutherland DH (2006) Chapter 3. Kinematics of Normal Human Walking. In:
Gamble JRJ (ed) Human walking, 3rd edn. Lippincott Williams & Wilkins, Philadelphia, pp
33–51
Kirker S, Keymer S, Talbot J, Lachmann S (1996) An assessment of the Intelligent Knee prosthesis.
Clin Rehabil 10:267–273
Klute GK, Berge JS, Orendurff MS, Williams RM, Czerniecki JM (2006) Prosthetic intervention
effects on activity of lower-extremity amputees. Arch Phys Med Rehabil 87(5):717–722
Koehler-McNicholas SR, Lipschutz RD, Gard SA (2016) The biomechanical response of persons
with transfemoral amputation to variations in prosthetic knee alignment during level walking. J
Rehabil Res Dev, (In press).
Koehler SR, Gard SA, Meier MR, Cassar M, Lipschutz R (2004) Stance-phase knee flexion in
persons with unilateral transfemoral amputations walking on an Otto Bock 3R60 EBS Knee: a
preliminary report. Paper presented at the 9th Annual Meeting of the Gait and Clinical
Movement Analysis Society (GCMAS), Lexington, KY.
Meier MR, Hansen AH, Gard SA, McFadyen AK (2012) Obstacle course: users’ maneuverability
and movement efficiency when using Otto Bock C-Leg, Otto Bock 3R60, and CaTech SNS
prosthetic knee joints. J Rehabil Res Dev 49(4):583–596
Michael JW (1994) Prosthetic knee mechanisms. Phys Med Rehabil State Art Rev 8(1):147–164
Michael JW (1999) Modern prosthetic knee mechanisms. Clin Orthop Relat Res 361:39–47
Michael JW (2004) Chap 33: Prosthetic Suspensions and Components. In: Michael JW, Bowker JH,
Smith DG (eds) Atlas of amputations and limb deficiencies: surgical, prosthetic, and rehabili-
tation principles, 3rd edn. American Academy of Orthopaedic Surgeons, Rosemont, pp 409–427
Moosabhoy MA, Gard SA (2006) Methodology for determining the sensitivity of swing leg toe
clearance and leg length to swing leg joint angles during gait. Gait Posture 24(4):493–501
Murphy EF (1964) The swing phase of walking with above-knee prostheses. Bull Prosthet Res 10
(1):5–39
Murray MP, Mollinger LA, Sepic SB, Gardner GM, Linder MT (1983) Gait patterns in above-knee
amputee patients: hydraulic swing control vs constant-friction knee components. Arch Phys
Murray MP, Sepic SB, Gardner GM, Mollinger LA (1980) Gait patterns of above-knee amputees
using constant friction knee components. Bull Prosthet Res 17(2):35–45
Oberg KET, Kamwendo K (1988) Chap 19: knee components for the above-knee amputation. In:
Murdoch G, Donovan RG (eds) Amputation surgery & lower limb prosthetics. Blackwell
Scientific Publications, Oxford
Orendurff MS, Segal AD, Klute GK, McDowell ML, Pecoraro JA, Czerniecki JM (2006) Gait
efficiency using the C-Leg. J Rehabil Res Dev 43(2):239–246
Peizer E, Gardner HF (1972) Selection and application of knee mechanisms. Bull Prosthet Res 10
(18):90–158
Perry J, Burnfield JM (2010) Gait analysis: normal and pathological function, 2nd edn. SLACK Inc,
Thorofare
Perry J, Burnfield JM, Newsam CJ, Conley P (2004) Energy expenditure and gait characteristics of
a bilateral amputee walking with C-leg prostheses compared with stubby and conventional
articulating prostheses. Arch Phys Med Rehabil 85(10):1711–1717
Radcliffe CW (1970a) Biomechanics of above-knee prostheses. In: Murdoch G (ed) Prosthetic and
Orthotic Practice. Edward Arnold Ltd., London, pp 191–198
Radcliffe CW (1970b) Functional considerations in the fitting of above-knee prostheses. In: Wilson
AB (ed) Selected articles from artificial limbs. Robert E. Krieger Publishing Co. Inc., Hunting-
ton, pp 5–30
Radcliffe CW (1977) The Knud Jansen Lecture: above-knee prosthetics. Prosthet Orthot Int 1
(3):146–160
Schmalz T, Bellmann M, Proebsting E, Blumentritt S (2014) Effects of adaptation to a functionally
new prosthetic lower-limb component: results of biomechanical tests immediately after fitting
and after 3 months of use. J Prosthet Orthot 26(3):134–143
Schmalz T, Blumentritt S, Jarasch R (2002) Energy expenditure and biomechanical characteristics
of lower limb amputee gait: the influence of prosthetic alignment and different prosthetic
components. Gait Posture 16(3):255–263
Schmalz T, Blumentritt S, Marx B (2007) Biomechanical analysis of stair ambulation in lower limb
amputees. Gait Posture 25(2):267–278
Schuch CM (1992) Transfemoral amputation: Prosthetic management. In: Michael JBJ (ed) Atlas of
limb prosthetics. Mosby-Year Book, Inc., St. Louis, pp 509–533
Segal AD, Orendurff MS, Klute GK, McDowell ML, Pecoraro JA, Shofer J, Czerniecki JM (2006)
Kinematic and kinetic comparisons of transfemoral amputee gait using C-Leg and Mauch SNS
prosthetic knees. J Rehabil Res Dev 43(7):857. doi:10.1682/jrrd.2005.09.0147
24 S.A. Gard
Sensinger JW, Intawachirarat N, Gard SA (2013) Contribution of prosthetic knee and ankle
mechanisms to swing-phase foot clearance. IEEE Trans Neural Syst Rehabil Eng 21
(1):74–80. doi:10.1109/TNSRE.2012.2224885
Seymour R, Engbretson B, Kott K, Ordway N, Brooks G, Crannell J et al (2007) Comparison
between the C-leg microprocessor-controlled prosthetic knee and non-microprocessor control
prosthetic knees: a preliminary study of energy expenditure, obstacle course performance, and
quality of life survey. Prosthet Orthot Int 31(1):51–61
Shurr DG, Michael JW (2002) Prosthetics & orthotics, 2nd edn. Prentice Hall, Upper Saddle River
Sutherland JL, Sutherland DH, Kaufman KR, Teel M (1997) Gait comparison of two prosthetic
knee units. J Prosthet Orthot 9(4):168–173
Taylor MB, Clark E, Offord EA, Baxter C (1996) A comparison of energy expenditure by a high
level trans-femoral amputee using the Intelligent Prosthesis and conventionally damped pros-
thetic limbs. Prosthet Orthot Int 20(2):116–121
Theeven P, Hemmen B, Rings F, Meys G, Brink P, Smeets R, Seelen H (2011) Functional added
value of microprocessor-controlled knee joints in daily life performance of Medicare Functional
Classification Level-2 amputees. J Rehabil Med 43(10):906–915. doi:10.2340/16501977-0861
Traugh GH, Corcoran PJ, Reyes RL (1975) Energy expenditure of ambulation in patients with
above-knee amputations. Arch Phys Med Rehabil 56(2):67–71
Waters RL, Mulroy S (1999) The energy expenditure of normal and pathologic gait. Gait Posture 9
(3):207–231
Waters RL, Perry J, Antonelli D, Hislop H (1976) Energy cost of walking of amputees: the influence
of level of amputation. J Bone Joint Surg Am 58(1):42–46
Waters RL, Yakura JS (1989) The energy expenditure of normal and pathologic gait. Crit Rev Phys
Rehabil Med 1(3):183–209
Williams RM, Turner AP, Orendurff M, Segal AD, Klute GK, Pecoraro J, Czerniecki J (2006) Does
having a computerized prosthetic knee influence cognitive performance during amputee walk-
ing? Arch Phys Med Rehabil 87(7):989–994
Willingham LL, Buell NC, Allyn KJ, Hafner BJ, Smith DG (2004) Measurement of knee center
alignment trends in a national sample of established users of the Otto Bock C-Leg micropro-
cessor-controlled knee unit. J Prosthet Orthot 16(3):72–75
Winter DA (1992) Foot trajectory in human gait: a precise and multifactorial motor control task.
Phys Ther 72(1):45–53 discussion 54-46
Yang L, Solomonidis SE, Spence WD, Paul JP (1991) The influence of limb alignment on the gait of
above-knee amputees. J Biomech 24(11):981–997
Zuniga EN, Leavitt LA, Calvert JC, Canzoneri J, Peterson CR (1972) Gait patterns in above-knee
amputees. Arch Phys Med Rehabil 53(8):373–382
Influence of Prosthetic Socket Design
and Fitting on Gait
Jan Andrysek and Arezoo Eshraghi
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Prosthesis/Limb Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Socket Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Evaluation of Socket Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Effect of Socket Design and Fit on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Prosthetic Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Gait Measurement Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Prosthesis/Limb Interfaces and Gait/Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Abstract
This chapter deals with lower-limb prostheses and, more specifically, the rela-
tionship between socket design and the quality of gait and mobility. Prosthetic
sockets and suspension systems provide an intimate contact with the residual
limb, and their interaction is important for overall comfort, fit, and mobility
J. Andrysek
Holland Bloorview Kids Rehabilitation Hospital, Bloorview Research Institute, Toronto, ON,
Canada
Institute of Biomaterial and Biomedical Engineering, University of Toronto, Toronto, ON, Canada
e-mail: jandrysek@hollandbloorview.ca
A. Eshraghi (*)
Holland Bloorview Kids Rehabilitation Hospital, Bloorview Research Institute, Toronto, ON,
Canada
e-mail: arezoo@um.edu.my; aeshraghi@hollandbloorview.ca

DOI 10.1007/978-3-319-30808-1_76-2
2 J. Andrysek and A. Eshraghi
function. Even though technological advances have led to the enhancement of

socket/limb interfaces, prosthesis users still suffer from a variety of problems,
particularly related to the high and repetitive loading of the limb during gait. In
this chapter, socket designs are reviewed, along with clinical issues relating to
their use. This is followed by an examination of the effect of the socket/limb
interface on the quality of gait. The chapter concludes with future perspectives
and trends in prosthetic/limb interface approaches and technology.
Keywords
Prostheses • Amputation • Fit • Gait • Socket • Rehabilitation • Walking • Motion
analysis • Mobility • Residual limb • Transtibial • Transfemoral • Partial foot
Introduction
Lower-limb loss is mainly caused by trauma, diabetes, tumors, congenital limb

deficiency, and peripheral vascular disease (PVD) (Smith et al. 2004). According
to Smith et al. (2004), lower-limb amputations worldwide are primarily attributed
to PVD, which is frequently linked to diabetes mellitus. Furthermore, statistics
reveal an increasing number of diabetic patients worldwide. The rate of lower-
limb amputations in individuals with diabetes is 15 times higher than healthy people.
In the United States alone, 82% of all amputations occur because of PVD (Seymour
2002).
A prosthesis or artificial limb is the foremost element in the rehabilitation process
following an amputation. Prosthetic limbs for lower-limb amputees typically consist
of a socket, a soft liner as cushion at the skin-socket interface, and suspension system
to securely maintain the prosthetic limb in place, which itself is comprised of pylons,
connectors, and prosthetic foot and knee (in the case of a transfemoral prosthesis)
components. These components are assembled and aligned and may be cosmetically
finished to match the appearance of the sound limb.
Prosthetic components and systems have evolved tremendously in recent decades
to a level that affords individuals with amputations high-level mobility function and
the ability to do many of the things of their non-amputee counterparts. However,
despite these key technological and clinical advances in prosthetic care, numerous
problems persist among prosthetic users. Many problems are related to the residuum/
prosthesis interface, comprised of the socket, liners, and suspension components.
The socket and suspension play an imperative role, facilitating weight bearing on the
prosthetic limb, and also control to move the limb and to retain balance when
standing and walking. When these aspects are well managed, prosthetic rehabilita-
tion can lead to successful patient outcomes including safe, effective, efficient, and
comfortable prosthetic use during mobility.
Influence of Prosthetic Socket Design and Fitting on Gait 3
State of the Art
In this section, a comprehensive overview is provided relating to prosthetic interface

design focusing on the contemporary technologies and clinical practices applied
within the P&O industry. The section describes commercially available technolo-
gies, including various socket, liner, and suspension designs as well as the existing
evidence relating to their clinical use. As achieving an optimal fit between the
prosthesis and residual limb is one of the primary goals and also challenges, this
topic is thoroughly covered in the latter parts of this section. This includes the
techniques by which socket fit is measured and evaluated.
Prosthesis/Limb Interface
A prosthesis is attached to the residual limb and the rest of the body using certain
components. These, including the socket, liner, and suspension, come in close
contact with the soft tissue and skin. A prosthetic socket encompasses the residual
limb and connects other distal parts of the prosthesis to the individual. Suspension is
the means by which a socket is held in place over the residuum. In general,
suspension means can be categorized as internal (anatomic and atmospheric) or
external methods (straps and hinges). In addition to that, liners are soft inserts that
assist in suspension and at the same time provide added comfort within the typically
hard socket shell. Traditionally liners have been made of foam materials, while more
contemporary designs tend to utilize silicon gel materials.
Prosthetic sockets are fabricated from different materials and have various
designs based on the level of amputation and activity. Traditionally, sockets were
made of leather or wood, but in line with the advances in material science, new
thermoplastic, thermoset, and composite materials offer advantages including dura-
bility, flexibility, and high strength-to-weight ratios. Other aspects of socket designs
have also evolved, aimed at improving the fit, comfort, and weight-bearing capacity.
Some of these advancements are applicable broadly across amputee populations,
while others are uniquely applicable to a particular subgroup. A common differen-
tiation is based on the amputation level, with different socket designs being appro-
priate for transtibial and transfemoral amputations, as well as less commonly
presented amputation levels including partial foot and pelvic level amputations.
Conventional techniques for fabricating sockets involve taking plaster impres-
sions of the residuum to produce a positive mold which is rectified to facilitate the
fabrication of an optimal fitting socket. Computer-aided technology has also been
used in the fabrication of socket systems. The systems digitize the shape of the
residual limb using optical or laser scanners, for CNC fabrication of a positive mold.
These techniques are not presented in detail here, but the reader is instead referred to
other resources for this information (Smith et al. 2004).
For transtibial amputations, the patellar tendon-bearing (PTB) socket design has
been widely used around the world. These sockets were commonly fabricated of
thermoplastic or lamination material (Radcliffe et al. 1961). First introduced in 1959,
the concept was to provide an intimate fit with the residuum unlike previous designs
(Fergason and Smith 1999). In a PTB socket, the anterior socket wall covers the
distal third of the patella. An inward bar or counter that is located immediately below
the patella and at the center of the patellar ligament serves as a weight-bearing
surface and suspension means. The key weight-bearing area and pressure-tolerant
surface is the medial wall, dented at the medial tibial flare. A relief area for the distal
fibula and fibula head is formed on the lateral wall. The posterior wall applies force
in the anterior direction to keep the patellar tendon on the bar and terminates
proximally somewhat higher than the patellar bar.
A major innovation relating to prosthetic interfaces has been the development of
silicone suction suspension (Fillauer et al. 1989) and the Iceross (Baars and Geertzen
2005; Kristinsson 1993) sockets. Used for both transtibial and transfemoral ampu-
tations, these systems were characterized by total surface bearing (TSB), hydrostatic
loading, and improved techniques of suspension (Staats and Lundt 1987; Sewell
et al. 2000). These systems are designed to distribute the loads consistently over the
residual limb without any undercut or relief areas to reduce the peak pressure (Beil
and Street 2004). A gel or silicone liner (soft socket) that is surrounded by a hard
socket provides cushioning for the sensitive bony areas and dissipates the load
throughout the socket. The silicone liners are mainly suspended by pin and lock
mechanisms, seals around the liner circumference, or recently by magnetic systems
(Eshraghi et al. 2013a).
Suction suspension provides the most secure attachment to the residual limb. For
almost half a century, the suction socket was the main socket type for transfemoral
amputees. In the original method, the residual limb’s skin served as a gasket to seal
the socket brim against air penetration. The requirements are that the skin is pliable
and artifact free (e.g., deep scar). During the fitting, the circumference of the socket
should be reduced carefully to avoid blood flow restriction and ensure constant
friction between the skin and the socket. A valve is placed in the distal socket that
once opened allows air out of the socket as the residual limb enters the socket during
donning. Once all of the air exits the socket space, the valve is closed to create
atmospheric pressure.
A more recent evolution of the suction suspension, which has gained broad
clinical interest, is the vacuum-assisted socket or VASS. Both suction suspension
and elevated vacuum suspension use the difference in atmospheric pressure to attach
the residual limb to the socket (Fig. 1). The main difference is that suction systems
typically use a passive expulsion valve to create negative pressure once the limb
starts to move, while vacuum systems use an active pump to produce a negative
pressure difference regardless of the limb position. In the advanced versions of these
systems, the pressure is electronically monitored and automatically adjusted by the
onboard pump. The main advantages of elevated vacuum are believed to include
better accommodation to the residual limb’s volume fluctuation by promoting
normal fluid exchange, decreasing forces on the residual limb, and enhancing
Fig. 1 Transtibial prosthesis

with Harmony ® elevated
vacuum system by Otto Bock
(Ottobock HealthCare, North
America; Austin, TX USA.
http://www.ottobock.com)
proprioception and pistoning (Board et al. 2001; Klute et al. 2011; Gerschutz et al.
2010).
The elevated vacuum system has a pump to pull air out of the socket from
between the inner socket and the residual limb to keep the tissue against the socket
walls, hence inhibiting movement in all directions. Recently, Össur introduced a
sleeveless vacuum system, called the Unity (Fig. 2). Unlike the other available
systems, such as Harmony (Otto Bock), the system incorporates a Seal-In liner to
enhance prosthetic suspension. Basically, the system is claimed to provide the same
benefits of other vacuum-assisted systems, such as high level of suspension, security,
improved comfort and skin health, and better gait, without the restrictions that a
sleeve imposes on the amputee.
Evidence relating to the VASS shows improvements in residual limb health,
including a decrease in skin problems, such as sores and promotion of wound
healing (Traballesi et al. 2012). The blood supply is also believed to be improved
within the limb, because the socket acts as a vascular pump that promotes blood
circulation. Potential caveats of the system, partially related to improper use of the
device, can result if, for example, the user dons the socket incorrectly and then turns
on the pump causing the vacuum to pull the limb into the socket with such force that
it may create a blood blister or other soft tissue injury.
Fig. 2 Unity sleeveless vacuum system for transtibial and transfemoral prostheses by Össur ehf,
Reykjavik Iceland (http://www.ossur.com)
Transfemoral socket designs are quite different from the transtibial sockets,
mainly due to the difference in residual limb size, shape, tissue and skin coverage,
and bony prominences. Two main socket designs for the transfemoral prosthesis are
the quadrilateral socket and the ischial containment socket. These two designs are
differentiated by the proximal brim contours; in the quadrilateral socket, the poste-
rior brim lies parallel to the ground and has a wide seat (called the ischial seat), while
in the ischial containment, the ischium and the ischial ramus are inside the socket.
The ischial containment socket creates a bony lock that improves mediolateral
stability and gait (Fairley 2004). An evolution is the ischial-ramal containment
socket (MAS or Marlo Anatomical Socket). It was designed to contain the ischial
ramus and the ischial tuberosity, thus providing a bony lock within the socket for
improved stability. There is also evidence that this design can reduce energy cost of
walking in transfemoral amputees compared with the ischial containment socket
(Traballesi et al. 2011). The soft liners are similar to the transtibial designs and use
the same suspension types, such as pin and lock, seal, or magnet. Also, vacuum and
suction systems are used for both transfemoral and transtibial prostheses.
Variations of hard socket designs include reduced thickness of socket walls or the
use of more compliant materials, mainly aimed to provide freedom of movement by
increased muscle activity, reduce the weight of the prosthesis, provide better propri-
oception, improve perspiration, and enhance comfort. Examples of these design
variations include the Icelandic-Swedish-New York (ISNY) system and
Fig. 3 Left: Martin Bionics’ Bikini Socket. Right: Martin Bionics’ transfemoral Socket-Less
Socket (MartinBionics.com)
Scandinavian Flexible Socket (SFS). One commonly used design is a flexible

thermoplastic socket with a rigid or semirigid fenestrated frame. Kristinsson
described that “To label a socket as flexible I would say that you should be able to
deform it by your hands, and the material should not be elastic enough to stretch
under the loads it will be subjected to.” Yet, the frame design is very important as it
must support the flexible inner socket, and the combination of socket and frame must
be structurally stable and robust to counteract the reaction forces.
New socket designs are continually being introduced. For example, Martin
Bionics have introduced the Socket-less Socket™, utilizing a system of strategically
placed struts and cushioned support surfaces to contain and support the limb (Fig. 3).
The technology, currently available for knee disarticulation and transfemoral pros-
theses, claims a number of advantages over more conventional socket designs.
Rather than a static shape, the Socket-less technology is claimed to have greater
control and adjustability of the socket fit. Another design is for hip disarticulation
level of amputation, called “Bikini socket” (Fig. 3). With iliac crest stabilizers, this
socket is claimed to provide a more direct biomechanical link between the prosthesis
and user, resulting in greater control, and functional outcomes and comfort. With less
material, these designs are able to decrease overall weight of the prosthesis. Another
similar approach toward modular socket designs is the basis for the LIM Innovations
“Infinite” socket available for both transtibial and transfemoral amputations (Fig. 4).
The system includes a soft upper brim, specific adjustable features, and carbon fiber
strut structure. The design is customized to fit the amputee’s exact anatomy through
3D scanning, a software, and custom manufacturing methods.
Fig. 4 Infinite transfemoral

socket by LIM Innovations ®
(https://www.liminnovations.
com)
Most of the research literature pertains to transtibial, knee disarticulation, and

transfemoral amputations; however, a small body of work is focused on other
amputation levels. Partial foot amputations and those at the level of ankle joint,
such as Syme’s, may be less challenging in terms of socket design and suspension
compared with higher amputation levels as the lever arm is quite long and the
amputee is more stable during standing and walking. Some partial foot amputees
can ambulate over the amputation site for some daily activities, without a prosthesis.
Those that have lost parts of the foot, such as toes and/or metatarsal bones, may need
only minor changes to their shoes or insoles in the form of fillers, and thus, the term
socket might not apply to them. In general, given that most of the limb is intact and
unaltered in the case of partial amputation, gait limitations and alterations are less
pronounced than with higher amputation levels (Dillon and Barker 2008). Main areas
of focus relating to partial foot amputations have included the design of interfaces to
restore effective foot length and also instruments to monitor the in-socket pressure
distributions (Alvarez-Camacho et al. 2014; Dillon and Barker 2006).
Hip disarticulations and trans-pelvic amputations have also received limited
attention in the published literature, likely due to the fact that this level of amputation
is not very common. Generally, these levels of amputation have the highest rejection
rates and therefore lowest levels of prosthetic use (Smith et al. 2004), and one of the
primary reasons for this, as stated by amputees, is the uncomfortable socket. Not
only is it biomechanically more difficult to develop a well-fit socket, but most
prosthetists also have inadequate experience of fabricating and fitting sockets at
these less-frequently encountered levels of amputation. Gel liners are used for both
hip disarticulation and trans-pelvic levels to protect the soft tissues. The hip disar-
ticulation socket is usually suspended by locking onto above the iliac crest around
the waist.
The use of suction sockets has seen some success in this challenging population
with high-level amputations, providing greater comfort and fit, and resulting in
higher patient compliance and improved mobility (Zaffer et al. 1999). The Canadian
prosthesis design is still the most common design not only for hip disarticulations
but also trans-pelvic and very short transfemoral amputations (Fernández and
Formigo 2005). However, patient outcomes have not generally been very good.
Limited compliance with these systems has been found, largely due to intolerance of
socket fit and limited ability to ambulate. Challenges, in general, include poor gait
patterns, socket discomfort, loss of mobility, instability, and high energy consump-
tion (Ludwigs et al. 2010).
Socket Fit
Socket fit can be affected by numerous factors, relating to the patient’s condition,

rehabilitation process, and technical aspects. Factors include residual limb volume,
skin health, blood circulation, activity level, muscle activity, cause of amputation
(vascular or traumatic), level of amputation, method of amputation surgery, as well
as the amount of time since the surgery. Physiotherapy and gait training further
influence outcomes. On the technical side, socket design including the presence or
absence of soft liner or sleeve, suspension concept, and accessories all play a role, as
does the prosthetist’s skill in capturing an accurate impression (e.g., casting or
scanning) of the residual limb and appropriately modifying it to fabricate the socket.
The effect of volume fluctuations is among the most pervasive issues relating to
socket fit, because it can cause other problems, such as skin breakdown, pistoning,
increased shear forces, increased pressure on bony prominences, decreased comfort,
and gait deviations (Sanders and Fatone 2011). The main cause of volume change is
socket pressure on the soft tissue. To manage this, traditionally socks are added
during the day.
During the day, volume fluctuation can increase pressure in certain regions
because of the abatement of soft tissue within the socket, thus exposing bony areas
to the socket walls. This can lead to pain and possible tissue breakdown. Another
problem that arises from volume change or reduced socket fit is “pistoning,” in
which the limb has excessive movement in the socket, particularly during the
loading and unloading cycles of gait. This can lead to greater shear forces between
the socket and limb and higher peak pressures during the loading response phase of
gait (Beil and Street 2004). In addition to socket fit, the suspension system or
accessories, such as belts and straps, may contribute to issues related to pistoning.
Several changes in the socket design and suspension systems have been made in
an effort to address the aforementioned problems. Patellar tendon-bearing (PTB)
sockets were widely used for transtibial amputees for years. This socket design
exerts high loads on certain areas of the limb. While these areas are typically called
pressure tolerant, the users may struggle with reddened skin, calluses, and soft tissue
breakdown over the course of time. The peak pressures in the PTB socket can reach
300–400 kPa, whereas in a total contact socket or a TSB socket, pressure is reduced
to less than 200 kPa (Beil and Street 2004). A TSB, providing better distribution of
loads, can therefore aid in venous return and decreasing the skin reddening on the
distal residual limb.
In a transfemoral prosthesis, the current standard of care focuses on the proximal
aspect of the socket to gain a stable connection during stance phase. The brim (the
most proximal aspect of the socket) contains the ischial ramus and the associated
trim lines above the ischial tuberosity. The main purpose of this design is to provide a
stance stability mechanism by forming a bony lock with the pelvis in the coronal
plane (Kahle and Highsmith 2014), which can prevent the socket from lateral shift
and femoral abduction. Enhanced suspension also reduces motion within the socket
during swing (Klute et al. 2011). Vacuum-assisted suspension provides dynamic
suspension, either through an electronic or manual pump. Thus, recently the VASS
has become an effective option for transfemoral amputation.
In terms of suspension systems, pin and lock systems have advantages such as
ease of donning and enhanced suspension. Nevertheless, the silicone liner is apt to
squeeze the residual limb proximally and cause large distal suction during the swing
phase of gait, promoting tissue breakdown and skin problems at the distal end (Beil
and Street 2004). This problem is more pronounced in highly active users, leading to
a potential contraindication for the use of pin and lock systems in this demographic.
Overall, in any design, socket fit may be jeopardized due to volume loss during the
day. The elevated vacuum system is claimed to help maintain the limb volume as
constant as possible. However, current evidence is not adequate to fully support this.
Evaluation of Socket Fit
It is not easy to determine the socket fit objectively. In clinical practice, visual

inspections including the use of check sockets and subjective feedback of amputees
are the most commonly used methods to evaluate socket fit. Even though more
objective methods have been researched and developed, in general they have not
been advanced enough to be practically adopted in clinical practice. Important
factors to consider in the evaluation of socket fit include but are not limited to
in-socket pressure, skin condition, amount of pistoning, the presence or absence and
level of pain as well as skin and soft tissue problems, quality of mobility, and comfort
level.
In-socket pressure can be measured with different types of transducers (Laing
et al. 2011; Polliack et al. 2000; Sanders et al. 1998; Zhang et al. 1998); or estimated
using simulation techniques (Commean et al. 1997; Lin et al. 2004; Silver-Thorn
et al. 1996). Measuring in-socket pressures can provide important information in
case total surface bearing has been achieved or not. Friction within the prosthetic
socket has a twofold effect as it helps to retain the prosthesis on the residuum, but at
the same time, it may distort the soft tissue (Mak et al. 2001). If high friction occurs
at an interface, stress may be localized, and this can lead to the deformation of the
remaining tissue. Conversely, Zhang et al. found that lubricating the skin will
increase the interface pressure (Zhang et al. 1996). Liners and prosthetic sockets
may alter the pressure applied to the residual limb. Without understanding the
changes imposed on the soft tissue and skin by different socket designs and
suspension systems, it is impossible to effectively evaluate the overall prosthetic
fit. Moreover, prosthetic interface pressure is believed to be a determinant of the
amputee’s comfort (Dou et al. 2006; Jia et al. 2004; Sanders et al. 2006a; Sewell et al.
2000).
Another important measure of socket fit is pistoning (Commean et al. 1997;
Madsen et al. 2000; Sanders et al. 2006b). Pistoning is defined as the relative
movement, either between the liner and socket or the residual limb and bone, skin,
and liner/socket. Various measurement techniques have been used so far, including
the spiral computerized tomography (CT) (Madsen et al. 2000), Roentgenology
(Söderberg et al. 2003), ultrasound (Convery and Murray 2000), cineradiography,
radiography (Narita et al. 1997) as well as photoelectric sensors, and custom-made
transducers (Abu Osman et al. 2010; Sanders et al. 2006b). The most recent methods
use the 3D motion analysis system to quantify pistoning (Gholizadeh et al. 2011,
2012a, b). Several studies have measured pistoning during gait. Primarily, the
methods for pistoning measurement involve complicated settings and techniques,
mostly limited to the research laboratory. While it is important for clinicians to assess
socket fit, it still remains impractical to measure socket fit through the current
pistoning measurement approaches.
Volume accommodation strategies, such as socks, liquid- or air-filled bladders
and pads within the prosthetic socket, and adjustable sockets, are used to compensate
for volume loss (Sanders et al. 2015; Pirouzi et al. 2014). Sock plies can be worn
over the liners (e.g., pelite liners (PTB), pin/lock or Seal-In systems). Yet, this
method is not feasible for suction systems that require close contact between the
socket and skin for maintaining a vacuum. Some studies have explored methods of
measuring the socket ply thickness as an indirect measure of socket fit (Sanders et al.
2015). Moreover, reducing the socket size using air bladders or socks is said to
further promote fluid volume loss (Sanders et al. 2012, 2013). The physiological
processes that lead to volume loss are not well understood; however, a physical
analogy can be made to a sponge, where fluid is rapidly expelled when the amputee
is active and the residual limb is squeezed inside a tight socket (because of addition
of sock or bladder size). Yet, that fluid does not simply recover when the socket is
released (Sanders et al. 2013). To address this issue, a recent method called “socket
release” has been introduced to lessen volume change. In this technique, no material
is removed from or added to the socket, but the amputee temporarily doffs the
prosthesis to facilitate recovery of fluid volume by relieving interface stresses within
the socket immediately after a period of activity (Sanders et al. 2015). Removing the
prosthesis decreases interstitial fluid pressures, which allows the residual limb’s
blood vessels to expand, increasing blood volume within the limb.
Effect of Socket Design and Fit on Gait
Prosthetic Mobility
Prosthetic devices should be appropriate for the user’s medical and physical con-
dition, capabilities, and mobility goals. A prosthesis should also aim to provide
safe, efficient, and biomechanically appropriate gait. Hence, an important goal in
the rehabilitation of lower-limb amputees is to enable functional mobility, seen
as one of the most significant aspects of achieving independence (Sansam et al.
2009).
Amputees typically present with differences in their walking compared with
non-amputees, and minimizing these differences, or gait deviations, is a common
rehabilitation goal. Gait deviations are most often seen in the spatiotemporal,
kinematic, and kinetic measurements of gait, meaning that the placement and timing
of steps, movements of the parts of the lower limbs, and the loads passing through
them can all be altered as a result of prosthetic use.
Spatiotemporal parameters, being the simplest to measure, are also the most
commonly studied and reported. These most frequently include walking speed,
step length, cadence, and stride length (Sagawa et al. 2011). In general, amputees
walk more slowly, having lower cadences and shorter stride and steps lengths. The
said parameters also characterize a global predictor of gait (Montero-Odasso et al.
2004, 2005). Other spatiotemporal parameters for amputee gait analysis include step
time ratio and stance time.
In terms of kinematics, a number of differences in movements are commonly
observed. Pelvic range of motion is increased in the frontal plane in lower-limb
amputees compared with healthy individuals. At slow walking speed, a significant
difference was found in the pelvic range of motion (ROM) compared with able-
bodied subjects (Su et al. 2007). Lower-limb amputees lift the pelvis (hip hiking) on
the swing leg; this motion is frequently seen in unilateral amputees and viewed as a
compensatory mechanism for the incapability to dorsiflex the prosthetic ankle. Hip
hiking serves to increase foot clearance and decrease the possibility of tripping;
however, it can also increase metabolic energy and decrease gait efficiency (Sagawa
et al. 2011). Bilateral hip hiking has been observed in persons with bilateral
transtibial amputation, potentially resulting in higher energy expenditure than
found in unilateral amputees (Su et al. 2007). Other kinematic gait asymmetries
that have been reported include a higher range of motion at the prosthetic side hip
and knee compared with the sound limb in transtibial amputees during walking
(Bateni and Olney 2002).
While comparisons are often made between amputee and non-amputee gait, gait
parameters are also commonly assessed in terms of asymmetry or differences
between the two limbs or sides of the body. The gait patterns of a person with a
unilateral lower-limb amputation are typically much more asymmetrical compared
with that of healthy individuals, in terms of timing, joint angles, as well as loading
patterns (Bateni and Olney 2002; Isakov et al. 2000). In particular, the prosthetic side
step length tends to be longer than the sound limb because of the shorter stance time
on the prosthetic side (Bateni and Olney 2002; Board et al. 2001; Isakov et al. 2000;
McNealy and Gard 2008; Sanderson and Martin 1997; Andrysek et al. 2007). This
seems to be at least partially attributed to the tendency for unilateral amputees to use
their intact leg to compensate for the prosthesis’ deficiencies (Su et al. 2007).
However, longer time spent loading the intact limb, often also at higher loads
(Nolan and Lees 2000; Esposito et al. 2015), can be problematic toward the long-
term health of the sound limb (Pinzur et al. 1991).
During the loading phase, knee flexion acts as a shock absorber that is vital in
preventing wear and tear on lower-limb joints (Isakov et al. 2000). The magnitude of
stance-phase knee flexion for able-bodied individuals or the sound side of lower-
limb amputees is almost 15–18 during walking (Sagawa et al. 2011). For transtibial
amputees, prosthetic side knee flexion is typically only 9–12 (Isakov et al. 2000;
Powers et al. 1998; Su et al. 2007) and negative or absent for transfemoral amputees
(Segal et al. 2006). Knee flexion can be affected by many factors, such as the type of
socket (Isakov et al. 2000), barefoot or shoed walking (Han et al. 2003), and
rehabilitation protocol (Sjödahl et al. 2002).
As to the extent to which prosthetic rehabilitation should aim to minimize gait
deviations and asymmetries, this remains a bit of a point of controversy within the
clinical and research community. Given the physical and functional differences
between the intact and prosthetic legs, high symmetry levels in all gait parameters
may not be attainable, and trade-offs need to be considered, as well as the biome-
chanical limitations of prosthetic components. Several studies have found higher
reliance on the sound leg by increased loading and stance time, which was attributed
to ankle loss in transtibial amputees (Lemaire and Fisher 1994; Melzer et al. 2001).
Other studies, however, support the idea that amputees may not need to rely on the
intact leg because of the compensatory mechanisms adopted by the amputated leg
(Silverman et al. 2008). While the amputee-related literature is seeing an increas-
ingly greater use of gait symmetry measures (Winter and Sienko 1988), a greater
understanding may be needed as to the use of these metrics.
Numerous studies have aimed to understand the influence of different prosthetic
components on the quality of gait patterns. Some studies reported the positive effects
from certain prosthetic components. For instance, some prosthesis users stated more
confidence with their prosthetic foot and applied less force on the sound limb (Van
der Linden et al. 1999). The use of a proper socket and appropriate fitting enabled a
reduced degree of gait asymmetry by improved control of the prosthesis position
(Board et al. 2001). The use of different knee mechanisms has been shown to
influence a number of gait parameters, such as gait velocity (Furse et al. 2011) as
well as kinematic and kinetics (Thiele et al. 2014; Taheri and Karimi 2012; Kaufman
et al. 2012; Andrysek et al. 2007).
Gait Measurement Techniques
A variety of techniques are available for use depending on the gait measures of
interest, the accuracy of measurement required, and resources available. Many
Fig. 5 3D motion analysis with reflective markers (Image on left adopted from Russell et al. (2006))
clinicians rely on observational gait assessments, requiring minimal resources and

time to identify spatiotemporal and kinematic gait deviations. However, more
objective and accurate methods are typically incorporated when conducting
research. Most commonly utilized is the instrumented gait lab, comprised of optical
means for tracking movements of defined parts of the body in three dimensions
(Fig. 5). Such systems capture spatiotemporal and kinematic measurements of body
segments, and when combined with force plate data, kinetic measures, including
joint moments and powers, can also be derived. Many researchers have used 3D
motion analysis to investigate the gait parameters of transtibial amputees during
different activities using various prosthetic components (Bateni and Olney 2002;
Rusaw and Ramstrand 2010; Sanderson and Martin 1997).
Depending on the specific research objectives, gait laboratory measurements may
be augmented by other approaches. For example, where the interest exists to
measure the pressures on the plantar foot, or within the socket/limb interface, various
pressure transducers can be incorporated at these regions of the body. Inertial-based
measurement unit (IMU) similarly can provide important information about the
relative movements of the limbs, or gait events of features, such as heel strike and
toe off, or terminal swing impact (Boutaayamou et al. 2015; Furse et al. 2011).
Gait laboratories are of particular importance in the study of the influences of

socket and suspension. One example relates to the quantification of pistoning
(Eshraghi et al. 2012). Common methods of measuring pistoning movement of the
stump or the position of the bone include radiography (Narita et al. 1997; Söderberg
et al. 2003; Papaioannou et al. 2010), photographic methods (Gholizadeh et al.
2011), motion analysis techniques or marker-based methods (Childers and Siebert
2015; Gholizadeh et al. 2012b), spiral computerized tomography (Madsen et al.
2000), and photoelectric sensors (Sanders et al. 2006b).
Another important area of measurement is interfacial pressures. Transducers
embedded into the socket or thin sensor pads inserted between the skin and liner/
socket are typically used to measure pressure (Williams et al. 1992; Polliack et al.
2000). The pressures at the socket/limb interface can vary considerably among
individuals, as well as the location within the socket. Moreover, pressure levels are
influenced by the ambulation task or walking style, socket fit, residual limb shape,
and prosthetic alignment, as well as the socket, liner, and suspension accessories
used (Dou et al. 2006; Jia et al. 2004; Mak et al. 2001; Eshraghi et al. 2013b).
Prosthesis/Limb Interfaces and Gait/Mobility
A large body of research has aimed to assess the influence of various prosthetic
components on the gait of lower-limb amputees (Goujon et al. 2006; Schmalz et al.
2002; Torburn et al. 1995; Van der Linden et al. 1999). Some of the most commonly
and frequently studied components include foot-ankle devices and knee joint mech-
anisms. Despite their important role in gait and mobility, prosthetic sockets and
suspension systems have received more modest attention in terms of the research
literature.
Fit of the residuum inside the prosthetic socket and effectiveness of prosthetic
suspension (measured primarily in terms of pistoning, pressure distributions on the
limb) have been associated, albeit primarily indirectly, with changes in gait patterns
and energy consumption during ambulation (Baars and Geertzen 2005; Bateni and
Olney 2002; Isakov et al. 2000; Sanderson and Martin 1997). For example, the TSB
socket having more uniform contact pressure has been found to improve gait by
decreasing pistoning during walking (Hachisuka et al. 1998b; Kristinsson 1993;
Narita et al. 1997; Yigiter et al. 2002). In contrast, pin/lock suspension systems have
been found to cause pain and discomfort inside the prosthetic socket due to increased
pistoning, leading to long-term skin changes and pain (Beil and Street 2004).
A recent review of research literature reported the main studies that evaluated the
effect of transtibial socket design on gait variables (Safari and Meier 2015). The
findings suggest that prosthetic socket design, and in particular socket suspension,
has an effect on gait symmetry. In a PTB socket, a sleeve or a supracondylar wedge
can improve gait symmetry (Wirta et al. 1990). A TSB socket further improves gait
symmetry over the PTB socket (Yigiter et al. 2002), and the highest symmetry is
achieved with vacuum-assisted suction sockets (Board et al. 2001). On the other
hand, the type of suspension provided by the liner material (urethane vs silicone) and
the elastomeric liner (Seal-In vs sleeve suspension) showed no effect on gait

parameters (Åström and Stenström 2004).
There is evidence that the vacuum-assisted sockets can increase stance duration
symmetry and step length compared with the TSB sockets, possibly due to improved
proprioception and suspension (Safari and Meier 2015). These sockets potentially
control residual limb volume loss over long periods of use and promote wound
healing (Gerschutz et al. 2010), potentially by producing a negative interface
pressure during swing phase and small positive interface pressure during stance
phase, absorbing fluid into the residual limb, particularly in a more conical residuum
(Beil et al. 2002).
Stance time and single-limb support time were higher on the non-amputated side
with the PTB sockets and a SACH foot, while the prosthetic side exhibited higher
step length, step time, and swing time (Isakov et al. 2000). The knee angle at heel
strike was larger in the prosthetic limb than in the non-amputated limb. The
maximum swing knee flexion and knee flexion during loading response and at toe
off were not significantly different between the intact and prosthetic limbs, but
suspension type was not mentioned. It is however unclear from this study to what
extent these gait differences can be attributed to the prosthetic foot or socket.
In a single-subject case report, Rogers et al. (2008) compared gait parameters of a
participant wearing three different vacuum-assisted suction sockets: one conven-
tional socket and two with different wall-compliant features over the fibular head and
distal tibia (Rogers et al. 2008). The participant’s walking speed and cadence were
highest when using the socket with the most compliant features and associated with
lower inter-socket pressures by up to 50 percent. GRFs were, however, found to be
similar for all three sockets. While this pilot study offers interesting insights into the
influence of socket compliance on spatiotemporal gait parameters, the potential root
of this relationship remains unclear, as does the applicability of the findings to the
general amputee population.
Residuum discomfort may cause changes in gait parameters, because the ampu-
tee becomes reluctant to bear load over the prosthetic socket during walking. The
Seal-In suspension can relieve the distal end pressure by applying more load to the
proximal tissues of the residual limb. Both pin/lock and Seal-In liners control
pistoning; however, the Seal-In liner may limit pistoning more successfully
(Gholizadeh et al. 2012a, b). Recently, Bruneli et al. (2013) reported that the
Seal-in liner caused a reduction in the energy cost of walking compared with the
suction socket and sleeve, especially when walking on a slope. Furthermore,
amputees could walk faster with the Seal-In suspension system (Brunelli et al.
2013).
The direct effect of socket or interface properties on mobility variables has been
rarely studied. In one research, the thickness of prosthetic gel liners (a thin 3-mm
liner and a thick 9-mm liner) was reported to affect gait variables (Boutwell et al.
2012). The joint angles and ground reaction forces as well as speed of walking were
evaluated, but the discussion mainly focused on walking speed. No differences were
found in the magnitudes of prosthetic-side pelvic obliquity or stance-phase knee
flexion between liners nor in the peak fore-aft GRF during weight acceptance. Also,
the timing of the peak vertical GRF was unaffected during weight acceptance, but
the magnitude of this peak was significantly higher with the 9-mm liner. The self-
selected walking speed did not change. The interface between the residual limb and
socket may allow relative translations and rotations during activities such as walk-
ing, which may be increased by high compliance from the combined compression of
gel liner and residuum soft tissue. The thicker gel liner could potentially allow
greater relative motion between the residual limb and the socket, which subjects
may have perceived as less stable or secure (Boutwell et al. 2012).
In conclusion, many different related parameters have been reported to assess the
mobility performance of prosthetic components, including sockets and suspension
systems. However, the majority has not been explored in detail, and the diversity of
outcomes to designate the gait of lower-limb amputees remains a challenge, since
limited consensus among researchers and clinicians exists about the important
aspects of gait. Based on the literature, the analyses of gait in transtibial amputees
provide reliable and useful information about walking strategies. Nevertheless,
further studies are needed to evaluate the influence of prosthetic socket fit and design
on mobility more profoundly and for different levels of amputation.
Future Directions
It is well established that socket design and fit strongly impact the quality of mobility
among lower-limb prosthesis users. Developing better sockets and interfaces
requires understanding of the clinical problems that persist and need to be addressed.
Moreover, scientific and technological advancements are needed in developing the
tools that can aid in understanding of the biomechanics of amputee gait and mobility
and facilitating the means by which more effective interface designs can be achieved.
To solve the volume change and its effect on mobility, experimental approaches
include the use of bladders within the socket that are filled with fluid or air once the
limb loses volume (Sanders et al. 2006a; Greenwald et al. 2003; Ogawa et al. 2008).
However, challenges with such systems often lie in their complexity and increased
cost as well as maintenance requirements. Even though the VASS system has been
shown to have benefits for amputees, there is still gap in the literature, particularly
with regard to the kinetic, kinematic, and temporal characteristics of gait.
It is worth emphasizing here that the method of socket design and fabrication has
a tremendous effect on the fit. An important step in the fabrication process of sockets
is to accurately capture the shape of the residual limb. Digitization of the limb shape
is continually improving with advancements in affordable optics and software that
facilitate effective post-processing and rectification of the shape. While these
approaches and tools are becoming ever more viable, practical, and useful, signifi-
cant strides are also being made in the utilization of digitized data to directly
fabricate sockets. Additive manufacturing, including selective laser sintering (SLS)
and fused deposition modeling (FDM), is starting to play a role in the fabrication of
orthotics, and work is ongoing to extend application to prosthetic sockets as well.
Additive manufacturing presents a number of potential benefits, such as the ability to
Fig. 6 Direct attachment of

prosthesis to skeleton by an
intramedullary implant into
the femur (Van de Meent et al.
2013)
reproduce identical devices, utilization of complex geometries, and the use of

integrated compliant features to improve and fit and comfort of the prosthesis
(Faustini et al. 2005; Rogers et al. 2008). However, a number of challenges also
exist with the direct fabrication of sockets using additive manufacturing. While a
variety of materials can be applied for additive manufacturing, the use of composite
and reinforced materials that provide high strength-to-weight ratios has limitations.
Bonding between layers of the deposited material and the inability of reinforcing
material to flow between the layers make it difficult to achieve comparable strength
levels of socket structures that are made with conventional methods. While these
technical challenges are primarily associated with the FDM, limitation of the SLS
currently lies in a much higher fabrication cost, compared with the FDM as well as
conventional manual methods (Telfer et al. 2012). In summary, the use of additive
manufacturing for custom fabrication of prosthetic sockets is becoming more and
more of a reality; however, clinical, technical, and financial issues still need to be
addressed before these techniques can be made viable for mainstream use (Jin et al.
2015).
One procedure that has the potential to eliminate the need for conventional socket
systems altogether is bone anchorage (Fig. 6). Osseointegration of external prosthe-
ses was introduced in Sweden in 1990 and has been studied and developed for the
last 25 years (Hagberg and Brånemark 2009). A titanium implant provides the
anchorage by the formation of bony tissue around it without growth of fibrous tissue
at the bone-implant interface. The conventional socket and suspension accessories

are not needed anymore as the prosthesis is attached directly to the bone. There are
studies showing advantages of this method over the use of standard socket interfaces
(Van de Meent et al. 2013). Among these are improved gait (Frossard et al. 2010),
reduced skin issues, improved stability and comfort (Hagberg et al. 2005, 2008),
easier donning and doffing, excellent suspension, unrestricted hip movement, and
better body perception (Lundberg et al. 2011) and quality of life. Disadvantages
include long rehabilitation process, risk of infection, risk of fractures and loosening
of the implant, poor cosmesis of the permanent abutment, limits on high-impact
activities such as running or jumping, and the need for daily care of the abutment
skin area. Hence, currently the procedure is mainly recommended as an option for
transfemoral amputees that experience major problems with conventional sockets.
However, if the costs of care and postoperative management issues can be addressed,
including but not limited to infection and technical problems, osseointegration may
become a more mainstream part of care in near future.
The influence of prosthetic socket design and fit on gait and mobility is an
important topic, and as such it has been the sole focus of this chapter. Typically,
the primary goal of providing rehabilitative care and a prosthetic limb to an individ-
ual with a lower-limb amputation is to facilitate functional mobility. And, the means
and quality of the prosthesis as well as body interface are arguably the most
important prosthesis-related parts of this rehabilitation process. The published liter-
ature covered in this chapter provides evidence and new insights about the effects
that both conventional and more contemporary socket and suspension systems have
on aspects of gait as well as mobility function. A better understanding can be gained
from these studies to help inform clinicians and other rehabilitation professionals
toward implementing evidence-based practices and achieving better rehabilitation
care and patient outcomes. The clinical evidence is by no means perfect or complete,
but it does provide some initial clues about how different interface systems affect
biomechanics as well as gait and mobility function. Therefore, some guidance as to
how these systems might be most appropriately and effectively utilized as part of
clinical care is afforded. However, further research and evidence is needed, consid-
ering the variety of patient demographics, outcome measures, and the every-
emerging waves of new prosthetic technologies. It is hoped that readers of this
chapter have been provided with an adequate understanding of the topic, including
where the gaps in knowledge still lie, to better inform their works in the areas of
research, product development, education, or clinical care.
Cross-References


▶ Physics-Based Models for Human Gait Analysis
References
Abu Osman N, Spence W, Solomonidis S, Paul J, Weir A (2010) The patellar tendon bar! Is it a
necessary feature? Med Eng Phys 32(7):760–765
Alvarez-Camacho M, Urrusti J, Acero MC, Galván D-GC, Rodriguez-Reyes G, Mendoza-Cruz F
(2014) Device to assess in-socket pressure distribution for partial foot amputation. Rev Invest
Clin 66:S131–S141
Andrysek J, Redekop S, Naumann S (2007) Preliminary evaluation of an automatically stance-
phase controlled pediatric prosthetic knee joint using quantitative gait analysis. Arch Phys Med
Rehabil 88(4):464–470
Åström I, Stenström A (2004) Effect on gait and socket comfort in unilateral trans-tibial amputees
after exchange to a polyurethane concept. Prosthetics Orthot Int 28(1):28–36
Baars E, Geertzen J (2005) Literature review of the possible advantages of silicon liner socket use in
trans-tibial prostheses. Prosthetics Orthot Int 29(1):27–37
Bateni H, Olney SJ (2002) Kinematic and kinetic variations of below-knee amputee gait. JPO: J
Prosthet Orthot 14(1):2–10
Beil TL, Street GM (2004) Comparison of interface pressures with pin and suction suspension
systems. J Rehabil Res Dev 41(6):821–828
Beil TL, Street GM, Covey SJ (2002) Interface pressures during ambulation using suction and
vacuum-assisted prosthetic sockets. J Rehabil Res Dev 39(6):693–700
Board W, Street G, Caspers C (2001) A comparison of trans-tibial amputee suction and vacuum
socket conditions. Prosthetics Orthot Int 25(3):202–209
Boutaayamou M, Schwartz C, Stamatakis J, Denoël V, Maquet D, Forthomme B, Croisier J-L,
Macq B, Verly JG, Garraux G (2015) Development and validation of an accelerometer-based
method for quantifying gait events. Med Eng Phys 37(2):226–232
Boutwell E, Stine R, Hansen A, Tucker K, Gard S (2012) Effect of prosthetic gel liner thickness on
gait biomechanics and pressure distribution within the transtibial socket. J Rehabil Res Dev 49
(2):227–240
Brunelli S, Delussu AS, Paradisi F, Pellegrini R, Traballesi M (2013) A comparison between the
suction suspension system and the hypobaric Iceross Seal-In ® X5 in transtibial amputees.
Prosthetics Orthot Int 37(6):436–444
Childers WL, Siebert S (2015) Marker-based method to measure movement between the
residual limb and a transtibial prosthetic socket. Prosthetics Orthot Int 0309364615610660.
40(6):720–768
Commean PK, Smith KE, Vannier MW (1997) Lower extremity residual limb slippage within the
prosthesis. Arch Phys Med Rehabil 78(5):476–485
Convery P, Murray K (2000) Ultrasound study of the motion of the residual femur within a trans-
femoral socket during gait. Prosthetics Orthot Int 24(3):226–232
Dillon MP, Barker TM (2006) Can partial foot prostheses effectively restore foot length? Prosthetics
Orthot Int 30(1):17–23
Dillon MP, Barker TM (2008) Comparison of gait of persons with partial foot amputation
wearing prosthesis to matched control group: observational study. J Rehabil Res Dev 45
(9):1317
Dou P, Jia X, Suo S, Wang R, Zhang M (2006) Pressure distribution at the stump/socket interface in
transtibial amputees during walking on stairs, slope and non-flat road. Clin Biomech 21
(10):1067–1073. doi:10.1016/j.clinbiomech.2006.06.004
Eshraghi A, Abu Osman NA, Gholizadeh H, Karimi M, Ali S (2012) Pistoning assessment in lower
limb prosthetic sockets. Prosthetics Orthot Int 36(1):15–24
Eshraghi A, Abu Osman NA, Gholizadeh H, Ahmadian J, Rahmati B, Abas WABW (2013a)
Development and evaluation of new coupling system for lower limb prostheses with acoustic
alarm system. Scientific Reports 3. doi:10.1038/srep02270
Eshraghi A, Abu Osman NA, Gholizadeh H, Ali S, Saevarsson SK, Abas WABW (2013b) An
experimental study of the interface pressure profile during level walking of a new suspension
system for lower limb amputees. Clin Biomech 28(1):55–60. doi:10.1016/j.
Esposito ER, Whitehead JMA, Wilken JM (2015) Sound limb loading in individuals with unilateral
transfemoral amputation across a range of walking velocities. Clin Biomech 30(10):1049–1055
Fairley M (2004) MAS socket: a transfemoral revolution. O&P J 6. http://www.oandp.com/articles/
2004-06_03.asp
Faustini MC, Crawford RH, Neptune RR, Rogers WE, Bosker G (2005) Design and analysis of
orthogonally compliant features for local contact pressure relief in transtibial prostheses. J
Biomech Eng 127(6):946–951
Fergason J, Smith DG (1999) Socket considerations for the patient with a transtibial amputation.
Clin Orthop Relat Res 361:76–84
Fernández A, Formigo J (2005) Are Canadian prostheses used? A long-term experience. Prosthetics
Orthot Int 29(2):177–181
Fillauer CE, Pritham CH, Fillauer KD (1989) Evolution and development of the silicone suction
socket (3S) for below-knee prostheses. JPO: J Prosthet and Orthot 1(2):92–103
Frossard L, Hagberg K, Häggström E, Gow DL, Brånemark R, Pearcy M (2010) Functional
outcome of transfemoral amputees fitted with an osseointegrated fixation: temporal gait char-
acteristics. JPO: J Prosthet Orthot 22(1):11–20
Furse A, Cleghorn W, Andrysek J (2011) Improving the gait performance of nonfluid-based swing-
phase control mechanisms in transfemoral prostheses. IEEE Trans Biomed Eng 58
(8):2352–2359
Gerschutz MJ, Denune JA, Colvin JM, Schober G (2010) Elevated vacuum suspension influence on
lower limb amputee’s residual limb volume at different vacuum pressure settings. JPO: J
Gholizadeh H, Osman NAA, Lúvíksdóttir ÁG, Eshraghi A, Kamyab M, Abas WABW (2011) A
new approach for the pistoning measurement in transtibial prosthesis. Prosthetics Orthot Int 35
(4):360–364
Gholizadeh H, Abu Osman N, Kamyab M, Eshraghi A, Lúvíksdóttir A, Wan Abas WAB (2012a)
Clinical evaluation of two prosthetic suspension systems in a bilateral transtibial amputee. Am J
Gholizadeh H, Abu Osman NA, Eshraghi A, Ali S, S’varsson SK, WAB WA, Pirouzi GH (2012b)
Transtibial prosthetic suspension: less pistoning versus easy donning and doffing. J Rehabil Res
Dev 49(9):1321–1330
Goujon H, Bonnet X, Sautreuil P, Maurisset M, Darmon L, Fode P, Lavaste F (2006) A functional
evaluation of prosthetic foot kinematics during lower-limb amputee gait. Prosthetics Orthot Int
30(2):213–223
Greenwald RM, Dean RC, Board WJ (2003) Volume management: smart variable geometry socket
(SVGS) technology for lower-limb prostheses. JPO: J Prosthet Orthot 15(3):107–112
Hachisuka K, Dozono K, Ogata H, Ohmine S, Shitama H, Shinkoda K (1998b) Total surface
bearing below-knee prosthesis: advantages, disadvantages, and clinical implications. Arch Phys
Hagberg K, Brånemark R (2009) One hundred patients treated with osseointegrated transfemoral
amputation prostheses – rehabilitation perspective. J Rehabil Res Dev 46(3)
Hagberg K, Häggström E, Uden M, Brånemark R (2005) Socket versus bone-anchored trans-

femoral prostheses: hip range of motion and sitting comfort. Prosthetics Orthot Int 29
(2):153–163
Hagberg K, Brånemark R, Gunterberg B, Rydevik B (2008) Osseointegrated trans-femoral ampu-
tation prostheses: prospective results of general and condition-specific quality of life in
18 patients at 2-year follow-up. Prosthetics Orthot Int 32(1):29–41
Han TR, Chung SG, Shin HI (2003) Gait patterns of transtibial amputee patients walking indoors
barefoot. Am J Phys Med Rehabil 82(2):96–100
Isakov E, Keren O, Benjuya N (2000) Trans-tibial amputee gait: time-distance parameters and EMG
activity. Prosthetics Orthot Int 24(3):216–220
Jia X, Zhang M, Lee WC (2004) Load transfer mechanics between trans-tibial prosthetic socket and
residual limb—dynamic effects. J Biomech 37(9):1371–1377
Jin Y-a, Plott J, Chen R, Wensman J, Shih A (2015) Additive manufacturing of custom orthoses and
prostheses – a review. Procedia CIRP 36:199–204
Kahle JT, Highsmith MJ (2014) Transfemoral interfaces with vacuum assisted suspension compar-
ison of gait, balance, and subjective analysis: ischial containment versus brimless. Gait Posture
40(2):315–320
Kaufman KR, Frittoli S, Frigo CA (2012) Gait asymmetry of transfemoral amputees using
mechanical and microprocessor-controlled prosthetic knees. Clin Biomech 27(5):460–465
Klute GK, Berge JS, Biggs W, Pongnumkul S, Popovic Z, Curless B (2011) Vacuum-assisted socket
suspension compared with pin suspension for lower extremity amputees: effect on fit, activity,
and limb volume. Arch Phys Med Rehabil 92(10):1570–1575
Kristinsson Ö (1993) The ICEROSS concept: a discussion of a philosophy. Prosthetics Orthot Int 17
(1):49–55
Laing S, Lee PV, Goh JC (2011) Engineering a trans-tibial prosthetic socket for the lower limb
amputee. Ann Acad Med Singap 40(5):252
Lemaire ED, Fisher FR (1994) Osteoarthritis and elderly amputee gait. Arch Phys Med Rehabil 75
(10):1094–1099
Lin C-C, Chang C-H, Wu C-L, Chung K-C, Liao I (2004) Effects of liner stiffness for trans-tibial
prosthesis: a finite element contact model. Med Eng Phys 26(1):1–9
Ludwigs E, Bellmann M, Schmalz T, Blumentritt S (2010) Biomechanical differences between two
exoprosthetic hip joint systems during level walking. Prosthetics Orthot Int 34(4):449–460
Lundberg M, Hagberg K, Bullington J (2011) My prosthesis as a part of me: a qualitative analysis of
living with an osseointegrated prosthetic limb. Prosthetics Orthot Int 35(2):207–214
Madsen MT, Hailer J, Commean PK, Vannier MW (2000) A device for applying static loads to
prosthetic limbs of transtibial amputees during spiral CT examination. J Rehabil Res Dev 37
(4):383–387
Mak AF, Zhang M, Boone DA (2001) State-of-the-art research in lower-limb prosthetic
biomechanics-socket interface: a review. J Rehabil Res Dev 38(2):161–174
McNealy LL, Gard SA (2008) Effect of prosthetic ankle units on the gait of persons with bilateral
trans-femoral amputations. Prosthetics Orthot Int 32(1):111–126
Melzer I, Yekutiel M, Sukenik S (2001) Comparative study of osteoarthritis of the contralateral knee
joint of male amputees who do and do not play volleyball. J Rheumatol 28(1):169–172
Montero-Odasso M, Schapira M, Varela C, Pitteri C, Soriano ER, Kaplan R, Camera LA, Mayorga
L (2004) Gait velocity in senior people an easy test for detecting mobility impairment in
community elderly. J Nutr Health Aging 8(5):340–343
Montero-Odasso M, Schapira M, Soriano ER, Varela M, Kaplan R, Camera LA, Mayorga LM
(2005) Gait velocity as a single predictor of adverse events in healthy seniors aged 75 years and
older. J Gerontol Ser A Biol Med Sci 60(10):1304–1309
Narita H, Yokogushi K, Shi S, Kakizawa M, Nosaka T (1997) Suspension effect and dynamic
evaluation of the total surface bearing (TSB) trans-tibial prosthesis: a comparison with the
patellar tendon bearing (PTB) trans-tibial prosthesis. Prosthetics Orthot Int 21(3):175–178
Nolan L, Lees A (2000) The functional demands on the intact limb during walking for active trans-
femoral and trans-tibial amputees. Prosthetics Orthot Int 24(2):117–125
Ogawa A, Obinata G, Hase K, Dutta A, Nakagawa M (2008) Design of lower limb prosthesis with
contact pressure adjustment by MR fluid. In: Engineering in Medicine and Biology Society,
EMBS 2008. 30th Annual International Conference of the IEEE. IEEE, pp 330–333.
Papaioannou G, Mitrogiannis C, Nianios G, Fiedler G (2010) Assessment of amputee socket–stum-
p–residual bone kinematics during strenuous activities using Dynamic Roentgen Stereo-
grammetric Analysis. J Biomech 43(5):871–878
Pinzur M, Asselmeier M, Smith D (1991) Dynamic electromyography in active and limited walking
below-knee amputees. Orthopedics 14 (5):535–537; .Discussion 537–538
Pirouzi G, Abu Osman NA, Oshkour AA, Ali S, Gholizadeh H, Wan Abas WA (2014) Development
of an air pneumatic suspension system for transtibial prostheses. Sensors 14(9):16754–16765
Polliack A, Sieh R, Craig D, Landsberger S, McNeil D, Ayyappa E (2000) Scientific validation of
two commercial pressure sensor systems for prosthetic socket fit. Prosthetics Orthot Int 24
(1):63–73
Powers CM, Rao S, Perry J (1998) Knee kinetics in trans-tibial amputee gait. Gait Posture 8(1):1–7
Radcliffe CW, Foort J, Inman VT, Eberhart H (1961) The patellar-tendon-bearing below-knee
prosthesis. University of California, Biomechanics Laboratory
Rogers B, Bosker G, Faustini M, Walden G, Neptune RR, Crawford R (2008) Case report: variably
compliant transtibial prosthetic socket fabricated using solid freeform fabrication. JPO: J
Rusaw D, Ramstrand N (2010) Sagittal plane position of the functional joint centre of prosthetic
foot/ankle mechanisms. Clin Biomech 25 (7):713–720. doi:http://dx.doi.org/10.1016/j.
Russell KA, Palmieri RM, Zinder SM, Ingersoll CD (2006) Sex differences in valgus knee angle
during a single-leg drop jump. J Athl Train 41(2):166–171
Safari MR, Meier MR (2015) Systematic review of effects of current transtibial prosthetic socket
designs – part 2: quantitative outcomes. J Rehabil Res Dev 52(5):509–528
Sagawa Y, Turcot K, Armand S, Thevenon A, Vuillerme N, Watelain E (2011) Biomechanics and
physiological parameters during gait in lower-limb amputees: a systematic review. Gait Posture
33(4):511–526
Sanders JE, Fatone S (2011) Residual limb volume change: systematic review of measurement and
management. J Rehabil Res Dev 48(8):949
Sanders JE, Bell DM, Okumura RM, Dralle AJ (1998) Effects of alignment changes on stance phase
pressures and shear stresses on transtibial amputees: measurements from 13 transducer sites.
IEEE Trans Rehabil Eng 6(1):21–31
Sanders J, Jacobsen A, Fergason J (2006a) Effects of fluid insert volume changes on socket
pressures and shear stresses: Case studies from two trans-tibial amputee subjects. Prosthetics
Orthot Int 30(3):257–269
Sanders JE, Karchin A, Fergason JR, Sorenson EA (2006b) A noncontact sensor for measurement
of distal residual-limb position during walking. J Rehabil Res Dev 43(4):509
Sanders JE, Harrison DS, Allyn KJ, Myers TR, Ciol MA, Tsai EC (2012) How do sock ply changes
affect residual-limb fluid volume in people with transtibial amputation? J Rehabil Res Dev 49
(2):241–256. doi:10.1682/jrrd.2011.02.0022
Sanders JE, Cagle JC, Harrison DS, Myers TR, Allyn KJ (2013) How does adding and removing
liquid from socket bladders affect residual limb fluid volume? J Rehabil Res Dev 50(6):845
Sanders JE, Hartley TL, Phillips RH, Ciol MA, Hafner BJ, Allyn KJ, Harrison DS (2015) Does
temporary socket removal affect residual limb fluid volume of trans-tibial amputees? Prosthetics
Orthot Int 0309364614568413
Sanderson DJ, Martin PE (1997) Lower extremity kinematic and kinetic adaptations in unilateral
below-knee amputees during walking. Gait Posture 6(2):126–136
Sansam K, Neumann V, O’Connor R, Bhakta B (2009) Predicting walking ability following lower
limb amputation: a systematic review of the literature. J Rehabil Med 41(8):593–603

of lower limb amputee gait: The influence of prosthetic alignment and different prosthetic
components. Gait Posture 16 (3):255-263. doi:http://dx.doi.org/10.1016/S0966-6362(02)
00008-5
Segal AD, Orendurff MS, Klute GK, McDowell ML, Pecoraro JA, Shofer J, Czerniecki JM (2006)
Kinematic and kinetic comparisons of transfemoral amputee gait using C-Leg and Mauch SNS
prosthetic knees. J Rehabil Res Dev 43(7):857–870
Sewell P, Noroozi S, Vinney J, Andrews S (2000) Developments in the trans-tibial prosthetic socket
fitting process: a review of past and present research. Prosthetics Orthot Int 24(2):97–107
Seymour R (2002) Prosthetics and orthotics: lower limb and spinal. Lippincott Williams & Wilkins,
Philadelphia
Silverman AK, Fey NP, Portillo A, Walden JG, Bosker G, Neptune RR (2008) Compensatory
mechanisms in below-knee amputee gait in response to increasing steady-state walking speeds.
Silver-Thorn MB, Steege JW, Childress DS (1996) A review of prosthetic interface stress investi-
gations. J Rehabil Res Dev 33:253–266
Sjödahl C, Jarnlo G-B, Söderberg B, Persson B (2002) Kinematic and kinetic gait analysis in the
sagittal plane of trans-femoral amputees before and after special gait re-education. Prosthetics
Orthot Int 26(2):101–112
Smith DG, Michael JW, Bowker JH, Surgeons AAoO (2004) Atlas of amputations and limb
deficiencies: surgical, prosthetic, and rehabilitation principles. American Academy of Ortho-
paedic Surgeons, Rosemont
Söderberg B, Ryd L, Persson BM (2003) Roentgen stereophotogrammetric analysis of motion
between the bone and the socket in a transtibial amputation prosthesis: a case study. JPO: J
Staats TB, Lundt J (1987) The UCLA total surface bearing suction below-knee prosthesis. Clin
Su P-F, Gard SA, Lipschutz RD, Kuiken TA (2007) Gait characteristics of persons with bilateral
transtibial amputations. J Rehabil Res Dev 44(4):491–501
Taheri A, Karimi MT (2012) Evaluation of the gait performance of above-knee amputees while
walking with 3R20 and 3R15 knee joints. J Res Med Sci 17(3):258
Telfer S, Pallari J, Munguia J, Dalgarno K, McGeough M, Woodburn J (2012) Embracing additive
manufacture: implications for foot and ankle orthosis design. BMC Musculoskelet Disord 13
(1):1
Thiele J, Westebbe B, Bellmann M, Kraft M (2014) Designs and performance of microprocessor-
controlled knee joints. Biomed Te/Biomed Eng 59(1):65–77
Torburn L, Powers CM, Guiterrez R, Perry J (1995) Energy expenditure during ambulation in
dysvascular and traumatic below-knee amputees: a comparison of five prosthetic feet. J Rehabil
Res Dev 32:111–111
Traballesi M, Delussu AS, Averna T, Pellegrini R, Paradisi F, Brunelli S (2011) Energy cost of
walking in transfemoral amputees: comparison between Marlo Anatomical Socket and Ischial
Containment Socket. Gait Posture 34(2):270–274
Traballesi M, Delussu A, Fusco A, Iosa M, Averna T, Pellegrini R, Brunelli S (2012) Residual limb
wounds or ulcers heal in transtibial amputees using an active suction socket system. A
randomized controlled study. Eur J Phys Rehabil Med 48(4):613–623
Van de Meent H, Hopman MT, Frölke JP (2013) Walking ability and quality of life in subjects with
transfemoral amputation: a comparison of osseointegration with socket prostheses. Arch Phys
Med Rehabil 94(11):2174–2178
Van der Linden M, Solomonidis S, Spence W, Li N, Paul J (1999) A methodology for studying the
effects of various types of prosthetic feet on the biomechanics of trans-femoral amputee gait. J
Biomech 32(9):877–889
Williams R, Porter D, Roberts V, Regan J (1992) Triaxial force transducer for investigating stresses
at the stump/socket interface. Med Biol Eng Comput 30(1):89–96

(5):361–367
Wirta RW, Golbranson FL, Mason R, Calvo K (1990) Analysis of below-knee suspension systems:
effect on gait. J Rehabil Res Dev 27(4):385–396
Yigiter K, Sener G, Bayar K (2002) Comparison of the effects of patellar tendon bearing and total
surface bearing sockets on prosthetic fitting and rehabilitation. Prosthetics Orthot Int 26
(3):206–212
Zaffer SM, Braddom RL, Conti A, Goff J, Bokma D (1999) Total hip disarticulation prosthesis with
suction socket: report of two cases1. Am J Phys Med Rehabil 78(2):160–162
Zhang M, Turner-Smith A, Roberts V, Tanner A (1996) Frictional action at lower limb/prosthetic
socket interface. Med Eng Phys 18(3):207–214
Zhang M, Turner-Smith A, Tanner A, Roberts V (1998) Clinical investigation of the pressure and
shear stress on the trans-tibial stump with a prosthesis. Med Eng Phys 20(3):188–198
Functional Effects of Foot Orthoses
Christopher Nester
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Biomechanics of Foot Orthoses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Foot, Orthotic, and Footwear Load-Bearing Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Effect of Foot Orthoses on Plantar Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Transmission of Load Through Plantar Soft Tissues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Effect of Foot Orthoses on External Joint Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Effect of Foot Orthoses on Foot Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Effect of Foot Orthoses on Tissues Related to Symptoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Abstract
This chapter considers how foot orthoses affect the various aspects of foot
biomechanics, from the loads applied to the sole of the foot and how this load
is transmitted through plantar soft tissues, through to the effects on joint moments
and movement. The importance of understanding how footwear influences the
effect of foot orthoses and whether we know which aspects of foot biomechanics
change in relation to particular clinical cases is also covered.
Keywords
Orthotic insoles • Footwear interaction • Plantar pressure • Foot kinematics
C. Nester (*)
School of Health Sciences, University of Salford, Salford, UK
e-mail: C.J.Nester@salford.ac.uk

DOI 10.1007/978-3-319-30808-1_78-1
2 C. Nester
Introduction
Foot orthoses are an important intervention used in the clinical management of a

wide range of foot and lower limb disorders and are prescribed by a range of health
professionals. Foot orthoses are used in isolation and in conjunction with other
therapies but almost always concurrent with footwear advice due to their inextricable
relationship. Their key function is to allow the health professional to adjust how
loads are applied to the foot during a range of locomotion tasks. Since different
ambulatory tasks involve different loads being applied to the foot, in different
locations, at different rates, and in different sequences, it follows that the functional
effects of foot orthoses can be complex. In the last two decades, we have made some
effective progress in understanding their functional effects although a good deal
needs to be clarified. A key barrier to progress is our inability to measure the
behavior and properties of the internal structures associated with foot function.
The basic assumption that underpins the clinical use of foot orthoses is either that
the clinical symptoms have been caused by the loads applied to the foot, or that by
adjusting the loads the clinical symptoms will improve. Understanding the func-
tional effect on an orthosis is thus about how loads will be applied to the foot when
wearing orthoses and the consequences of adjusting these loads on the mechanical
performance of foot tissues during tasks such as walking and running. Modifying
foot loading has the potential to positively alter the mechanical stress associated with
clinical symptoms in foot skin (e.g., blisters or callus); the layers of fat, fascia, and
muscle under the foot (e.g., heel pain); and the bones and joints of the foot (midfoot
arthritic pain). Therefore, it is how foot orthoses affect tissues that is ultimately our
aim in exploring functional effects of foot orthoses.
State of the Art
While there are always minor changes to orthotic materials and manufacturing
techniques and thus orthotic products, the basics of foot orthosis design and perfor-
mance have been largely the same for several decades (Landorf et al. 2001). Foot
orthoses can vary significantly in their geometric design and material composition
from patient to patient. However, they have two main clinical functions, either to
(1) reduce pressure under plantar tissues vulnerable to tissue damage or (2) adjust the
external joint moments such that internal structures (e.g., plantar fascia) bear less
load. In the main orthoses for both these purposes have common geometric features,
such as a “cup” around the heel, a high contoured profile that contacts with the
medial and lateral arches of the foot, and a flat area under the forefoot (Fig. 1). This
allows the orthotic to contact all areas of the foot and thus loads to be adjusted
wherever it is clinically important to do so. To this basic geometric design, pieces of
wedged material can be added to the heel to adjust distribution of load under the
rearfoot (Bonnano et al. 2012), and elevations or cavities used in the forefoot to
increase/decrease loads respectively (Guldemond et al. 2007). The height of the
medial and lateral arches of the orthotic is also routinely adjusted. There are a
Functional Effects of Foot Orthoses 3
Fig. 1 For example, foot

orthotic, with deep heel cup,
contour to medial and lateral
foot arches, and flat forefoot.
The black material is medium
density EVA and the top cover
3 mm plastozote
plethora of materials which can be used in isolation or combination to vary the final
properties at different locations in the orthotic (Fauli et al. 2008; Lo et al. 2014). This
ranges from rigid plastics (e.g., polypropylene) to adjust heel and arch loads, through
to highly compliant open cell foams to offer cushioning under the forefoot (e.g.,
plastozote, Poron). State of the art in orthotic materials thus remains a range of
rubber- and plastic-based materials available in different thickness and densities,
allowing the creation of orthoses of various volumes (affecting the fit into footwear)
and stiffness (affecting the application of load to the foot).
An important and ongoing debate is the apparent “competition” between orthoses
made to the static shape of the patients’ foot, so-called custom orthoses, and those
made to a standard shape, so-called prefabricated orthoses (Landorf et al. 2006;
Davis et al. 2008; Redmond et al. 2009). Traditionally the former are believed to
offer a better solution in terms of control of foot biomechanics and thereafter clinical
outcomes. However, there is such a wide range of prefabricated orthotic shapes, and
in many material combinations, that with enough time and patience, a prefabricated
orthotic that is close in shape and properties to the customized foot orthotic equiv-
alent could probably be located for many patients.
Customization of orthotic design has traditionally focused almost exclusively on
the shape of the orthotic (Carroll et al. 2011), but choices in material alone made in a
standard orthotic shape might be equally effective. Furthermore, one might argue
that choosing one prefabricated orthosis over another to achieve a specific clinical
biomechanical objective for a specific patient is itself customization, because the
choices are specific to that patient. The underlying task is one of customizing the
interface between the foot and shoe and floor and thus customization of the loads
applied to the body. This should be the focus of clinical application of orthoses rather
than whether the device itself is the outcome of a mass or bespoke manufacturing
process.
What we have yet to resolve is how to quantitatively identify the right material
and geometry combination to create the required change in foot biomechanics that
leads to a specific clinical effect. Thus, the prescription process comprises iterations
of trial and error and is very much based on the preferences of individual
4 C. Nester
practitioners. There are of course guidelines to follow (e.g., Williams et al. 2011), but
these often advise of what the objective of the orthotic should be, such as to “off-load
the forefoot,” but fall short of describing precisely how that should be achieved for
specific patients (and therefore how it might be achieved to the same degree for
different patients with different feet). Consensus activities can be useful (e.g.,
Banwell et al. 2014), but these are a measure of practitioner behavior rather than
evidence-based design criteria. While a “cookbook” of orthotic practice is not
desirable because it may constrain patient-focused solutions and practice, a more
quantitative and systematic approach to orthotic prescription would undoubtedly
lead to better clinical outcomes. Some progress has been made to this end using
computational models of the foot and orthotic designs (Telfer et al. 2014a; Ghassemi
et al. 2015) and using plantar pressure data to define off-loading features under the
forefoot (Owings et al. 2008).
Biomechanics of Foot Orthoses
Foot, Orthotic, and Footwear Load-Bearing Surfaces
Rarely do objects with such complex and constantly changing geometry and material
properties come into contact with each other as when a shoe encases a foot and the
wearer walks. Load is applied to the outer sole of the shoe and propagates upward to
the top of the shoe sole, where after it passes through the orthotic to the plantar
surface of the foot. The response of the foot to plantar load creates an interaction
between the dorsal and medial/lateral surfaces of the foot and shoe upper, and a
reaction force from the upper holds the foot against the top of the orthotic. The
interaction between the shoe upper and medial, lateral, and dorsal sides of the foot is
thus constantly changing as the ground reaction forces under the shoe change.
Indeed we have yet to fully resolve precisely how shoe upper and foot interact
(Rupérez et al. 2009). For example, forces are applied in three directions at each
point of contact between foot and shoe upper, but the geometry and direction of
movement of both contact surfaces are almost constantly changing and changing in
different ways at different locations across the foot and the shoe. Due in part to this
complexity, but also because it bears greater loads, we have concentrated experi-
mental research on the interaction on the plantar surface, which has proved some-
what easier to measure and thus understand.
The surfaces of the orthotic are also complex and the interaction with the shoe still
somewhat ill understood. The orthotic has four functional surfaces; receives loads
from the sole via its base, from the shoe upper via its medial/lateral sides; and
transmits load to the foot through its upper surface. However, this upper surface itself
has complex geometry, “facing” directly upward to the foot in the middle of the heel
cup and forefoot, but changing perhaps as much as 90 across its surface, such as in
the formation of the walls of the heel cup, and somewhere closer to 20–40 in the
medial arch area.
Effect of Foot Orthoses on Plantar Pressure
At the core of foot orthotic practice is the assumption that by changing the geometry
or material of the orthotic, we can manipulate with a fairly high degree of certainty
the distribution of force applied to the foot. Assuming walking speeds and body
mass are constant, the forces that pass from the floor to the body remain constant, but
we can adjust how they are distributed across the interface between the ground and
body, i.e., the foot. Thus, the scientific basis of foot orthotic practice relies upon their
being evidence that choices in orthotic geometry or materials do lead to predictable
changes in plantar pressure distribution. This is one area of considerable research
and evidence, albeit there is always room for further work. Critically, it remains quite
rare for changes in plantar pressure to be investigated when small or singular
changes in orthotic design are changed incrementally (Bonnano et al. 2012). For
example, to understand how orthotic medial arch height or stiffness affects plantar
pressure, two or more orthoses of different arch height must be compared. However,
arch height should be the only design factor varying between the designs, i.e., arch
height should be the independent variable. If three orthoses with three arch heights
are made from three different materials, it is not arch height that is being investi-
gated, since any observed change in plantar pressure, or indeed the lack of change in
pressure, might well be due to the differences in material (e.g., Redmond et al. 2009).
Several authors have demonstrated that the use of a medial arch support structure
increases pressure under the medial arch of the foot, allowing the foot to bear load
where normally it would not (Redmond et al. 2009; Janin and Dupui 2009; Bonnano
et al. 2012; Bousie et al. 2013; Song et al. 2015). This load would otherwise be borne
by the heel and lateral midfoot in the main, though there is some removal of load
from the forefoot too when the medial arch of the foot bears load (though this effect
is lost as soon as the heel and therefore medial arch comes off the ground). With
increased load in the medial foot arch, there is a corresponding decrease in pressure
under the heel. This most likely occurs as the medial arch begins to bear load prior to
forefoot contact. Thus the heel might never experience the peak pressures it does
during walking without an orthotic comprising a medial arch support.
We can expect that changes in medial arch pressure are proportional to the
orthotic arch height, which might be most usefully described as a proportion of
foot length rather than in absolute terms (e.g., % of foot length, rather than stated in
mm). However, doubling the height of an orthotic arch support that is already in
contact with the medial foot arch will not necessary double plantar pressure. The
plantar soft tissues in the medial arch are easily compressed and become stiff under
even low loads, never mind those during gait (Sweeney et al. 2015). Once com-
pressed and stiff, further increases in surface loads will quickly elevate pressure
under the arch, and too much pressure and tissue compression are likely to prove
very uncomfortable, cause skin damage, and may elicit an avoidance strategy.
Precise details on the relationship between orthotic arch height and plantar pressure
changes (e.g., X% increase in orthotic arch height produces X% increase in plantar
pressure) eludes us in part because of the multiple mechanisms underlying the
relationship. It may also be too variable or too small a change to detect with some
6 C. Nester
plantar pressure measurement systems (Price et al. 2016). The pressures between the
foot and orthotic are the result of the dynamic response of the foot impacting the
orthotic, which is in effect occupying space it previously occupied, and a combina-
tion of mechanical and neurological mechanisms that are likely task and person (or at
least population) specific.
The weight-bearing function of the heel cup walls is perhaps often ignored since it
is not strictly part of the plantar surface. When loaded the plantar heel fat pad is
compressed in a vertical direction but is displaced sideward on its medial, posterior,
and lateral aspects. When walking in an orthosis with a sufficiently deep and stiff
heel cup, this sideward deformation is constrained by the heel cup and the fat pad
therefore compressed, becoming very stiff. This “bracing” effect of the fat pad
against the sidewalls of the heel cup reduces the load under the heel further and
maintains a thicker albeit most certainly a stiffer fat pad under the calcaneus
(Perhamre et al. 2012; Luo et al. 2011; Sweeney et al. 2015).
Under the forefoot, there is perhaps more extensive evidence of how changes in
orthotic geometry and materials affect plantar pressure (Bonanno et al. 2011; Spirka
et al. 2014; Lee et al. 2014; Najafi et al. 2014). The motivation is often the need to
off-load a vulnerable area under the forefoot or toes, perhaps an area of acute
inflammation due to overloading or inflammatory disease (e.g., rheumatoid arthritis;
Tenten-Diepenmaat et al. 2016) or risk of tissue damage (e.g., in diabetes; Ulbrecht
et al. 2014). The use of raised areas allows pressure to be increased in areas that can
tolerate it and thereby reduce pressure in areas more vulnerable. Removal of material
under the vulnerable areas can achieve similar effects since by reducing pressure in
one area, you are displacing it to adjacent areas. Compared to changes in pressure at
the medial arch and heel, these effects are relatively minor in terms of the conse-
quences on gait. This is because changes in pressure under the forefoot take place
during a period when the forefoot is the only weight-bearing area, and thus the forces
are redistributed, but the overall function of the forefoot remains largely the same.
By contrast, increase pressure under the medial arch fundamentally changes where
load acts on the foot and can create changes that affect the knee (Almonroeder et al.
2015) and potentially the overall performance of gait.
Transmission of Load Through Plantar Soft Tissues
Plantar pressure data tells us where pressure is applied under the foot and thus tells us
something of the plantar soft tissue structures that the load must subsequently pass
through. The skin is of course the first of these structures and is often ignored as part
of orthotic practice. However, its structure and properties can greatly affect how load
from an orthotic is transferred to the deeper tissues and indeed the orthotic design can
affect the loads it experiences.
Under the forefoot, for example, the presence of callus and other hyperkeratotic
changes will thicken and stiffen the plantar skin. The center of callus is significantly
(i.e., ~50%) less elastic (or much stiffer) than normal plantar metatarsal skin (Hashmi
et al. 2015). This is likely a consequence of its increased thickness, at least double
normal plantar skin, changes in the internal skin architecture, and water loss. Stiffer
structures transmit loads to deeper tissues more effectively than compliant skin, and
thus the fat, fascia, tendon, and bone tissues beneath the callus will receive different
loads than if the callus were not there. It follows that orthotics that further increase
the transmission of load to areas of stiff skin might increase risk of damage to deeper
tissue. Also, that off-loading strategies could mitigate against the effects of callused
skin, since less pressure may compensate for the stiffer tissues, bringing stress in
deeper tissues toward more tolerable levels.
At the heel the primary skin problems are dry and fissured skin. These deep
crevasses form when dry and thus less elastic heel skin is subjected to tensile forces
as the heel pad expands sideways under load. The crevasses will disrupt the normal
transfer of load to the deeper fat pad tissues, providing points of stress concentration
as the skin interfaces with deeper tissues. One orthotic strategy for this scenario
could involve a heel cup to constrain the expansion of the fat pad (Sweeney et al.
2015) and thus provide a period of reduced internal stress for the fissures. Alongside
proven approaches to skin hydration, management of skin conditions by adjusting
loads could be a useful avenue for orthotic therapy to pursue.
Deep to the skin of the heel and forefoot are layers of fat, combined with tendons
for the later, and there are muscles and fascia across almost all other plantar areas.
The soft tissues that load from the orthotic surface pass-through can vary greatly
under the heel, mid and forefoot, but also at each of these areas in a medial/lateral
direction. There is thus a great deal that soft tissue structures and properties can do to
affect how the pressure measured at the plantar surface is ultimately applied to the
bones of the foot. At one level the different soft tissues will function to filter the loads
in different ways due to natural variations in tissue thickness and properties and
changes in these during gait (Telfer et al. 2014b). At the same time, wherever there is
muscle, the tissues will have a dynamic stiffness, because periods of muscle con-
traction will increase muscle and tendon tissue stiffness. Fascia too, most notably
plantar fascia, will also vary in stiffness during gait, for example, as toes extended
and stretch the plantar fascia. Thus how the muscles and fascia affect transmission of
load from the internal surface of the skin to deeper soft and hard tissues will vary
during gait. That we are not able to understand the details of these mechanisms
reflects their complexity in varying so much between locations under the foot, and
people, and their inaccessibility for experimental study. However, it seems obvious
that these are important factors influencing how changes in external pressure due to
the foot orthoses eventually affect loads experienced by foot bones and joints, i.e.,
joint moments and movement.
Effect of Foot Orthoses on External Joint Moments
Changing joint moments is one of the most important ways by which foot orthoses
have their clinical effect, and yet measuring these changes remains perhaps the most
difficult area of study (Hsu et al. 2014). External moments applied to foot joints are
the forces which passive (ligaments and fascia and bone) and active (muscle) tissues
8 C. Nester
seek to oppose. As such the external moments are a strong determinant of the tissue
stresses that are associated with clinical pathology. Measurement of external
moments requires that both the pose of the foot segments and the external ground
reaction forces are measured appropriately. The latter are easily measured with
accuracy, whereas the former are approximations of different functional units into
which the foot can be divided (multi-segment foot models).
However, the moments must be calculated around a joint center, and this is
particularly problematic because the location of these centers is unknown and an
estimation at best. It is this joint center of rotation that determines the moment arms
that are used with the ground reaction force data to derive the joint moments. In some
human joints, the physical center of a joint is a reasonable approximation of the joint
center of rotation. At the hip, for example, the sphere-like shape of the femoral head
largely dictates that the joint center resides somewhere inside sphere (Sangeux et al.
2014). Such assumptions do not hold for the bones and joints of the foot since they
have more complex articular surfaces, and few can be assumed to have centers of
rotation that logically reside inside the bones. An exception might be the metatarsal
heads, whose curved articular surfaces strongly suggest the center of rotation is
within the metatarsal head. But what might we assume for the calcaneocuboid joint?
Or for the sub-talar joint with its three independent articular surfaces? Without a
well-defined and valid center of rotation, moment data can be largely erroneous and
subject to misinterpretation.
As a result of these methodological challenges, we are largely constrained to
reports of moments around a center of rotation positioned at the midpoint between
both malleoli, the so-called external ankle joint moment. The effect of rearfoot
wedges under the heel will be to increase the normal eversion moments at heel
strike in the case of a lateral heel wedges and to reduce the external eversion
moments in the case of a medial heel wedge (Nester et al. 2003). These occur due
to changes in the contact area beneath the heel and thus the position of the center of
pressure. While changes in ankle moment might also occur due to a change in the
direction of the ground reaction force, this direction is largely the result of the
direction and speed of the limb and thus application of body mass to the ground.
This is a fundamental characteristic of how each person walks and unlikely to be
perturbed significantly by a foot orthosis; indeed the body is likely programed to
compensate for any disruption of how it impacts the ground for reasons of safety and
efficiency. Instead, with an orthosis we influence the details of the contact area
between foot and ground, or shoe, and thus the subtlety of where the external forces
are applied. It is also for this reason that ground reaction forces are a poor measure of
foot orthotic effect – it is largely the distribution of forces that orthoses affect, not the
forces themselves.
Just as the aforementioned changes in pressure distribution medially/laterally
under the heel affect eversion moments at the ankle, so changes in the anterior/
posterior distribution of pressure between heel and midfoot due to a medial arch
support will likely affect sagittal plane moments. The shift of pressure from the heel
to the medial arch during contact phase will mean the center of pressure is advanced
forward faster than without a foot orthosis. As a consequence, the ground reaction
force vector may come in front of the ankle marginally earlier, encouraging the
external ankle plantarflexion moment in early stance to change to an external
dorsiflexion moment earlier. This effect is likely subtle and certainly not as marked
as the effect of forward displacement of the center of pressure achieved with rollover
footwear, which has an absent heel and thus moves the center of pressure at initial
contact toward the end of the heel, rather than at the back of the heel .
Foot orthoses are not well designed to influence transverse plane moments at the
ankle. This requires that loads are applied to the sides of the heel and the walls of the
orthosis heel cup are typically thin and flexible to accommodate footwear. Indeed it
is likely that the footwear upper material, lacing, and other footwear components
(such as heel counter) affect the transverse plane moments far more than a foot
orthosis might. This reemphasizes the need for the “function” and effect of orthoses
to be considered along with the footwear that they are housed in.
Since the effect of medial and lateral heel wedges at the ankle is quite marked, the
effect of this on frontal plane knee moments has long since been of interest (Barrios
et al. 2013; Lewison et al. 2016). In particular, the normal varus or adduction
moment at the knee is strongly associated with the risk of osteoarthritis and the
progression of the disease process. The knee adduction moment has been correlated
with medial knee pain and long-term outcomes from total knee replacement.
Increases in the knee adduction moment are thus contraindicated, and reduction in
the eversion ankle moment due to a medially wedged orthosis increases the adduc-
tion moment at the knee. Conversely, however, laterally wedged orthoses increase
the external ankle eversion moment and reduce the external knee adduction moment.
This has now been correlated with improvements in medial knee pain and function
during walking and stair climbing, and it is assumed there are tissue level improve-
ments that occur concurrently (such as reduction in bone marrow lesions and bone
edema) (Alshawabka et al. 2014; Jones et al. 2013a, b, 2015).
Given the changes in sagittal and transverse plane ankle moments are small, they
are likely to have even smaller effects at the knee; certainly they will be smaller than
our errors in measurement of joint moments and as such are currently considered to
be negligible.
Effect of Foot Orthoses on Foot Motion
Traditional clinical approaches to foot orthotic effect have focused largely on

changes in foot kinematics, and hence it is the area with greatest evidence. Mills
et al. (2010) and Cheung et al. (2011) provide the most comprehensive reviews and
useful pooling of data, and both indicate that the average effect of a foot orthosis on
peak rearfoot eversion is a little over 2 . This is very much a simplification because
the shape and material used in the various studies included in their meta-analyses
vary greatly. Interestingly these effects seem insensitive to whether the wearer has an
existing foot or lower limb compliant, since healthy research participants and
patients respond in the same way. Subsequent work has confirmed this (Telfer
et al. 2013). The size of change may appear small at 2 , but it should be remembered
10 C. Nester
this is the effect across only the sub-talar and ankle joints, since it relates to the
movement of the heel relative to the leg. This range of motion is typically 6–8 , and
so 2 is a reasonable proportion of the underlying movement taking place
(~20–30%). Furthermore, many of the soft tissue structures that are symptomatic
and result in clinical use of foot orthoses span many foot joints, and 2 change at
each joint would quickly accumulate to a significant change in distance between
tissue origin and insertion.
Traditional paradigms of foot orthoses prescription, or rather design, assume a
fairly linear relationship between the degrees of orthotic wedging (or posting) added
to an orthosis and the subsequent changes in kinematics. This assumes that the foot is
almost exclusively a mechanical system where specific changes in inputs (e.g.,
2 changes in the inclination of the supporting surface due to an orthotic) result in
corresponding changes in output (i.e., joint kinematics). However, a “dose response”
has been described whereby 2 of medial/lateral posting on an orthoses resulted in a
0.26 change in peak eversion (Telfer et al. 2013). The ratio of orthotic wedging to
kinematic change will vary depending upon the precise design of orthotic posting,
but this serves to indicate that it is not a 1:1 relationship.
A related issue is whether changes in rearfoot eversion, a surrogate measure of
foot pronation, are the correct foot biomechanical parameter to target. Indeed, the
concept of a “dose” is to impose a specific quantity to a patient, in this case a change
in the quantity of eversion experienced by the foot. But this implies an established
link between the changes targeted (i.e., reduction in eversion) and either the cause of
the clinical symptoms or the response in clinical symptoms. Neither has in fact been
established. This is however complicated by the fact that whenever a change in
eversion is reported, it is never the independent variable since changes in eversion
can only occur if changes in plantar pressure, joint moments, and so on also occur.
Thus changes in symptoms could be due to a wide range of other concurrent
biomechanical changes due to the orthoses; it is simply that the movement change
is the only one we can “see.”
Effect of Foot Orthoses on Tissues Related to Symptoms
For all the value that kinematic, pressure, moment, and so on data can add to our
understanding of the effect foot orthoses have on the human foot and ankle, patients
rarely if ever attend a clinic complaining of a “problem with my plantar pressure,”
joint moments, or suggesting that their kinematics are limiting their physical activity.
They report pain in specific soft or hard tissue and the subsequent limitations in
physical activity. The former are caused by mechanical stress at a cell and tissue level
and subsequent inflammation. The latter are what we hope to address for the patient.
It is therefore important to understand how foot orthoses affect stresses within
specific internal tissues that are associated with inflammation and clinical pain and
presentation. It would be ideal to understand these stresses at a cellular, tissue,
“organ,” and thereafter functional (e.g., ability to walk) level, but our understanding
is limited largely to the effect of foot orthoses at the tissue level.
Foot joint pain is common and occurs as a result of excessive stresses at specific
locations within bony tissues, alongside inflammation in the joint soft tissues. Foot
orthoses will alter the external joint moments such that distribution of stress between
joint contact surfaces is altered, increasing joint contact area, for example, or
reducing contact pressures at sites of cartilage damage. For example, external
dorsiflexion moments at the talonavicular joint will tend to dorsiflex the joint and
promote dorsal distribution of contact pressures between the joint surfaces. These are
associated with the growth of osteophytes dorsally in response to these localized
joint contact surface stresses. The foot orthosis reduces the external dorsiflexion
moment by applying load underneath the talar head and navicular. A change in joint
position might further assist because this too might alter contact area between the
two bones. There is some evidence of change in midfoot and first MPJ pain when
orthoses with a medial arch support and stiffened under the first metatarsal phalan-
geal joint are used, supporting this mechanical explanation (Menz et al. 2016;
Halstead et al. 2016). However, in terms of quantifying these changes, we rely on
computational models of joint contact pressures or results from cadaver work where
contract pressures can be directly measured.
Muscle and tendon pain and inflammation is likewise common especially in the
athletic population. Here too, understanding the behavior of these internal structures
and the effect of foot orthoses remains problematic. There is a considerable amount
of literature reporting electrical signals from specific lower limb muscles (based on
electromyography) that influence foot function, but not foot muscles when wearing
foot orthoses. There is evidence that muscles are affected (e.g., Murley et al. (2010)),
though this seems obvious in the face of changes in external moments at the rearfoot
and kinematic changes too, neither of which could occur without concurrent adap-
tation in muscle or tendon function. Foot orthoses designed to reduce rearfoot
eversion and foot pronation appear to reduce tibialis posterior activity in early
stance, when external ankle eversion moments are also reduced. Also, peroneus
longus seems to increase its activity in mid-stance. That these antagonistic muscles
appear to reduce/increase activity respectively makes sense since by working in
opposition, a change in one should be reflected in a change in the other.
However, measures of electrical activity in muscle are a very crude measure of
“muscle function”, apply only to the location of the EMG electrodes, and bear little
relation to the muscle forces nor type of contraction being produced. Furthermore,
both kinematic and EMG data are needed to reveal how the muscle–tendon unit
length is changing and whether this is due to length changes in either muscle, tendon,
or both. Thus, while there is a lot of data available, the lack of sensitivity in our
measures of dynamic muscle function means we have yet to identify the effects of
foot orthoses on muscle function.
12 C. Nester
Future Directions
While we have a general explanation for the functional effects of foot orthoses and
can draw upon mechanical and biological principles to fill in gaps where we are as
yet unable to provide quality data, there are inevitably questions that require
answering.
Perhaps one of the most important is simply whether we are targeting the correct
functional effects of orthoses. For many foot and ankle and, to lesser degree, lower
limb pathologies, we have yet to identify the precise biomechanical risk factor or risk
factors that are associated with causing each pathology. Thus, should we be trying to
change plantar pressure, joint moments, joint motion, and the timing or degree of
muscle contraction? Although some research might suggest that a specific foot type
is a risk factor for pathology, this is too crude an indicator, since we would like to
understand what aspect of the foot type is associated with the causation of symptoms
in specific structures. Thus, we do not know precisely what it is about foot biome-
chanics that we would like to change in order to prevent or reduce risk of specific
pathologies.
A related issue is that independent of the cause of any symptoms, we do not know
which biomechanical parameters are important to create meaningful changes in
clinical symptoms – orthoses just seem to work. Indeed, believing or demonstrating
that we can change a specific kinematic or pressure parameter in a laboratory study
does not mean that we do change that parameters in our patients’ lives. This assumes
they wear the orthoses, for example, and wear it enough, or during the right activities
to change the tissue stresses in sufficient volume or frequency that healing and repair
can occur. This reflects the fact that tissue damage is an accumulation of stress that
takes the tissue outside its biological range of function. This has as much to do with
physical activity and health behavior as it does biomechanics. Thus, blending
biomechanical effects of foot orthoses and physical activity data is an as yet
uncharted area of work. This will bring to life a more real-world picture of foot
biomechanics and foot orthotic effects.
Finally, even if we are to create a specific change in a foot biomechanical
parameter, which orthotic design feature (e.g., medial arch, medial wedge) or what
setting of that feature (in mm for arch height, or degrees for wedging) is required is
not known. However, we do know this will vary between individuals, but we do not
know why. Thus, prescription practices are trial and error, driven by what is tolerable
by patients or possible within constraints imposed by footwear choices, rather than
delivery of a specific dose of biomechanical change. The relationship between
underlying foot biomechanics, orthotic design decisions, and the clinical biome-
chanical objective remains somewhat elusive.
Cross-References


References
Almonroeder TG, Benson LC, O’Connor KM (2015) The effect of a prefabricated foot orthotic on
frontal plane joint mechanics in healthy runners. J Appl Biomech 31(3):149–158
Alshawabka AZ, Liu A, Tyson SF, Jones RK (2014) The use of a lateral wedge insole to reduce
knee loading when ascending and descending stairs in medial knee osteoarthritis patients. Clin
Biomech (Bristol, Avon) 29(6):650–656
Banwell HA, Mackintosh S, Thewlis D, Landorf KB (2014) Consensus-based recommendations of
Australian podiatrists for the prescription of foot orthoses for symptomatic flexible pes planus in
adults. J Foot Ankle Res 7(1):49
Barrios JA, Butler RJ, Crenshaw JR, Royer TD, Davis IS (2013) Mechanical effectiveness of lateral
foot wedging in medial knee osteoarthritis after 1 year of wear. J Orthop Res 31(5):659–664
Bonanno DR, Landorf KB, Menz HB (2011) Pressure-relieving properties of various shoe inserts in
older people with plantar heel pain. Gait Posture 33(3):385–389
Bonanno DR, Zhang CY, Farrugia RC, Bull MG, Raspovic AM, Bird AR, Landorf KB (2012) The
effect of different depths of medial heel skive on plantar pressures. J Foot Ankle Res 5:20
Bousie JA, Blanch P, McPoil TG, Vicenzino B (2013) Contoured in-shoe foot orthoses increase
mid-foot plantar contact area when compared with a flat insert during cycling. J Sci Med Sport
16(1):60–64
Carroll M, Annabell ME, Rome K (2011) Reliability of capturing foot parameters using digital
scanning and the neutral suspension casting technique. J Foot Ankle Res 4(1):9
Cheung RT, Chung RC, Ng GY (2011) Efficacies of different external controls for excessive foot
pronation: a meta-analysis. Br J Sports Med 45(9):743–751. Erratum in: Br J Sports Med 2012;
46(5):373.
Davis IS, Zifchock RA, Deleo AT (2008) A comparison of rearfoot motion control and comfort
between custom and semicustom foot orthotic devices. J Am Podiatr Med Assoc 98(5):394–403
Faulí AC, Andrés CL, Rosas NP, Fernández MJ, Parreño EM, Barceló CO (2008) Physical
evaluation of insole materials used to treat the diabetic foot. J Am Podiatr Med Assoc 98
(3):229–238
Ghassemi A, Mossayebi AR, Jamshidi N, Naemi R, Karimi MT (2015) Manufacturing and finite
element assessment of a novel pressure reducing insole for diabetic neuropathic patients.
Australas Phys Eng Sci Med 38(1):63–70
Guldemond NA, Leffers P, Schaper NC, Sanders AP, Nieman F, Willems P, Walenkamp GH (2007)
The effects of insole configurations on forefoot plantar pressure and walking convenience in
diabetic patients with neuropathic feet. Clin Biomech (Bristol, Avon) 22(1):81–87
Halstead J, Chapman GJ, Gray JC, Grainger AJ, Brown S, Wilkins RA, Roddy E, Helliwell PS,
Keenan AM, Redmond AC (2016) Foot orthoses in the treatment of symptomatic midfoot
osteoarthritis using clinical and biomechanical outcomes: a randomised feasibility study. Clin
Rheumatol 35(4):987–996
Hashmi F, Nester C, Wright C, Newton V, Lam S (2015) Characterising the biophysical properties
of normal and hyperkeratotic foot skin. J Foot Ankle Res 8:35
Hsu WH, Lewis CL, Monaghan GM, Saltzman E, Hamill J, Holt KG (2014) Orthoses posted in
both the forefoot and rearfoot reduce moments and angular impulses on lower extremity joints
during walking. J Biomech 47(11):2618–2625
Janin M, Dupui P (2009) The effects of unilateral medial arch support stimulation on plantar
pressure and center of pressure adjustment in young gymnasts. Neurosci Lett 461(3):245–248
14 C. Nester
Jones RK, Chapman GJ, Findlow AH, Forsythe L, Parkes MJ, Sultan J, Felson DT (2013a) A new
approach to prevention of knee osteoarthritis: reducing medial load in the contralateral knee. J
Rheumatol 40(3):309–315
Jones RK, Nester CJ, Richards JD, Kim WY, Johnson DS, Jari S, Laxton P, Tyson SF (2013b) A
comparison of the biomechanical effects of valgus knee braces and lateral wedged insoles in
patients with knee osteoarthritis. Gait Posture 37(3):368–372
Jones RK, Chapman GJ, Parkes MJ, Forsythe L, Felson DT (2015) The effect of different types of
insoles or shoe modifications on medial loading of the knee in persons with medial knee
osteoarthritis: a randomised trial. J Orthop Res 33(11):1646–1654
Landorf K, Keenan AM, Rushworth RL (2001) Foot orthosis prescription habits of Australian and
New Zealand podiatric physicians. J Am Podiatr Med Assoc 91(4):174–183
Landorf KB, Keenan AM, Herbert RD (2006) Effectiveness of foot orthoses to treat plantar fasciitis:
a randomized trial. Arch Intern Med 166(12):1305–1310
Lee PY, Landorf KB, Bonanno DR, Menz HB (2014) Comparison of the pressure-relieving
properties of various types of forefoot pads in older people with forefoot pain. J Foot Ankle
Res 7(1):18
Lewinson RT, Stefanyshyn DJ (2016) Prediction of knee joint moment changes during walking in
response to wedged insole interventions. Proc Inst Mech Eng H 230(4):335–342
Lo WT, Yick KL, Ng SP, Yip J (2014) New methods for evaluating physical and thermal comfort
properties of orthotic materials used in insoles for patients with diabetes. J Rehabil Res Dev 51
(2):311–324
Luo GM et al (2011) Finite element analysis of heel pad with insoles. J Biomech 44(8):1559–1565
Menz HB, Auhl M, Tan JM, Levinger P, Roddy E, Munteanu SE (2016) Biomechanical effects of
prefabricated foot orthoses and rocker-sole footwear in individuals with first meta-
tarsophalangeal joint osteoarthritis. Arthritis Care Res 68(5):603–611
Mills K, Blanch P, Chapman AR, McPoil TG, Vicenzino B (2010) Foot orthoses and gait: a
systematic review and meta-analysis of literature pertaining to potential mechanisms. Br J
Sports Med 44(14):1035–1046
Murley GS, Landorf KB, Menz HB (2010) Do foot orthoses change lower limb muscle activity in
flat-arched feet towards a pattern observed in normal-arched feet? Clin Biomech (Bristol, Avon)
25(7):728–736
Najafi B, Wrobel JS, Burns J (2014) Mechanism of orthotic therapy for the painful cavus foot
deformity. J Foot Ankle Res 7(1):2
Nester CJ, van der Linden ML, Bowker P (2003) Effect of foot orthoses on the kinematics and
kinetics of normal walking gait. Gait Posture 17(2):180–187
Owings TM, Woerner JL, Frampton JD, Cavanagh PR, Botek G (2008) Custom therapeutic insoles
based on both foot shape and plantar pressure measurement provide enhanced pressure relief.
Diabetes Care 31(5):839–844
Perhamre S, Lundin F, Klässbo M, Norlin R (2012) A heel cup improves the function of the heel pad
in Sever’s injury: effects on heel pad thickness, peak pressure and pain. Scand J Med Sci Sports
22(4):516–522
Price C, Parker D, Nester C (2016) Validity and repeatability of three in-shoe pressure measurement
systems. Gait Posture 46:69–74
Redmond AC, Landorf KB, Keenan AM (2009) Contoured, prefabricated foot orthoses demon-
strate comparable mechanical properties to contoured, customised foot orthoses: a plantar
pressure study. J Foot Ankle Res 2:20
Rupérez MJ, Alemany S, Monserrat C, Olaso J, Alcañíz M, González JC (2009) A study of the
viability of obtaining a generic animation of the foot while walking for the virtual testing of
footwear using dorsal pressures. J Biomech 42(13):2040–2046
in gait analysis? Gait Posture 40(1):20–25
Song Q, Xu K, Yu B, Zhang C, Sun W, Mao D (2015) Could insoles offload pressure? An

evaluation of the effects of arch-supported functional insoles on plantar pressure distribution
during race walking. Res Sports Med 23(3):278–288
Spirka TA, Erdemir A, Ewers Spaulding S, Yamane A, Telfer S, Cavanagh PR (2014) Simple finite
element models for use in the design of therapeutic footwear. J Biomech 47(12):2948–2955
Sweeney D, Nester C, Preece S, Mickle K (2015) Effect of antipronation foot orthosis geometry on
compression of heel and arch soft tissues. J Rehabil Res Dev 52(5):543–551
Telfer S, Abbott M, Steultjens MP, Woodburn J (2013) Dose-response effects of customised foot
orthoses on lower limb kinematics and kinetics in pronated foot type. J Biomech 46
(9):1489–1495
Telfer S, Erdemir A, Woodburn J, Cavanagh PR (2014a) What has finite element analysis taught us
about diabetic foot disease and its management? A systematic review. PLoS One 9(10):e109994
Telfer S, Woodburn J, Turner DE (2014b) Measurement of functional heel pad behaviour in-shoe
during gait using orthotic embedded ultrasonography. Gait Posture 39(1):328–332
Tenten-Diepenmaat M, Dekker J, Steenbergen M, Huybrechts E, Roorda LD, van
Schaardenburg D, Bus SA, van der Leeden M (2016) In-shoe plantar pressure measurements
for the evaluation and adaptation of foot orthoses in patients with rheumatoid arthritis: A proof
of concept study. Gait Posture 45:45–50
Ulbrecht JS, Hurley T, Mauger DT, Cavanagh PR (2014) Prevention of recurrent foot ulcers with
plantar pressure-based in-shoe orthoses: the CareFUL prevention multicenter randomized
controlled trial. Diabetes Care 37(7):1982–1989
Williams AE, Davies S, Graham A, Dagg A, Longrigg K, Lyons C, Bowen C, North West Clinical
Effectiveness Group for the Foot in Rheumatic Diseases (NWCEG) (2011) Guidelines for the
management of the foot health problems associated with rheumatoid arthritis. Musculoskeletal
Care 9(2):86–92
Functional Effects of Shoes
Benedicte Vanwanseele
Abstract
Shoes are an outer covering for the foot, consisting of a more or less stiff sole and
a lighter upper that does not extend above the ankle. Shoes have been worn
throughout history to protect the feet from physical trauma and environmental
extremes with the oldest known footwear being well-preserved leather shoes
dating from around 3500 BC. While protection is the primary purpose of shoes,
fashion and style considerations also influence shoe design with functional
consequences on the foot and lower limb often neglected or of secondary
consideration in shoe design and the purchase of shoes. All shoes have the
function to protect against physical trauma. For normal daily footwear that is
mainly against wounds while sport shoes have to protect not only the feet but also
the joint and musculoskeletal tissue against injuries (acute and overuse injuries).
In addition, in specific sports, they are also designed to improve performance
(e.g., lightweight spikes). Besides protection, a lot of research has focused on two
main functions of the shoe: shock attenuation and rearfoot pronation due to
the assumed interaction with the development of musculoskeletal injuries. This
chapter will discuss these functions in a broader framework, looking at the
function of shoes in relation to loading, foot movement, and performance.
Keywords
Footwear • Shock • Pronation • Motion
B. Vanwanseele (*)
Department of Movement Sciences, KU Leuven, Leuven, Belgium
Fontys University of Applied Sciences, Eindhoven, The Netherlands
e-mail: benedicte.vanwanseele@kuleuven.be

DOI 10.1007/978-3-319-30808-1_79-1
2 B. Vanwanseele
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Functional Effect of the Shoes: Shock Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Functional Effect of Shoe: Control of Foot Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Functional Effect of Shoe: Reduce Loading in Other Joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Functional Effect of Shoe: Improve Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Functional Effect of Shoe: Traction and Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Introduction
Shoes are an outer covering for the foot, consisting of a more or less stiff sole and a
lighter upper that does not extend above the ankle. Shoes have been worn throughout
history to protect the feet from physical trauma and environmental extremes with the
oldest known footwear being well-preserved leather shoes dating from around 3500
BC. While protection is the primary purpose of shoes, fashion and style consider-
ations also influence shoe design with functional consequences on the foot and lower
limb often neglected or of secondary consideration in shoe design and the purchase
of shoes. All shoes have the function to protect against physical trauma. For normal
daily footwear that is mainly against wounds while sport shoes have to protect the
feet but also the joint and musculoskeletal tissue against injuries (acute and overuse
injuries). In addition, in specific sports, they are also designed to improve perfor-
mance (e.g., lightweight spikes). Besides protection, a lot of research has focused on
two main functions of the shoe: shock attenuation and rearfoot pronation due to
the assumed interaction with the development of musculoskeletal injuries. This
chapter will discuss these functions in a broader framework, looking at the function
of shoes in relation to loading, foot movement, and performance.
State of the Art
Functional Effect of the Shoes: Shock Attenuation
The repetitive nature of weight-bearing activities such as walking and running has
been seen as a risk factor for overuse injury. With each ground contact, a rapid
deceleration occurs at the foot-ground interface which results in a sudden loading of
the musculoskeletal system. This loading of the lower leg over and over again with
forces up to eight times body weight on structures like the Achilles tendon or up to
11 times body weight on the patellofemoral joint (Novacheck 1998) is believed to
potentially overload these structures and therefore cause overuse injuries such as
Achilles tendinopathy, patellofemoral pain syndrome. It is for this reason that the
Functional Effects of Shoes 3
shock attenuating properties of shoes, particularly athletic shoes, have been a focus
of footwear research. It was hypothesized that the cushioning effect of the outer sole
of the footwear would dampen the collision forces caused by the foot-ground
interaction. In theory, more compliant cushioning system can be expected to reduce
peak impact forces as the outer sole would act as a spring.
The standard method to measure the shock-absorbing qualities of footwear is to
mechanically compress the shoe at the rearfoot and forefoot sections using a
durometer. The shoe can be given a rating of hardness and resistance to compression
using an appropriate hardness scale, most commonly the Shore A (used for normal
rubber) or the Asker C (used for soft rubber and insoles of differing materials scales)
with higher scores given to harder and more highly compression-resistant footwear.
However, this doesn’t represent the shock-absorbing effect during human movement
as it doesn’t account for the human interaction with footwear. It has been shown that
human movement is influenced by the hardness of the shoe, and this effect has been
used in the functional design of shoes. For example, to correct rearfoot over
pronation, different sole hardness are used medial compared to laterally. It is also
known that the foot strike pattern (rear foot strikers versus non-rear foot strikers) can
change dependent on the footwear. Changing from standard running shoes (with
good shock absorption) to minimalistic running shoes (with minimal shock absorp-
tion) can change runners with a rearfoot strike pattern to a non-rearfoot striker.
Changing to minimalistic shoes also increases the amount of plantar flexion at the
ankle joint at heel strike and toe-off in the runners that don’t change their strike
pattern (Fredericks et al. 2015). It remains unclear whether the knee joint angle
changes with different footwear as several studies gave opposite results. However
the step length is decrease when switching to minimalistic shoes. All of these results
show that the hardness of the shoe will influence the kinematics of movement, and
therefore effects such as shock attenuation need to be tested during movement.
The body is able to prepare the lower limbs for impact by adjusting joint stiffness
in relation to the perceived hardness of the landing surface. However when wearing
shoes, plantar sensory feedback and proprioception are limited which also might
influence the human movement. Shock attenuation during movement (dynamic
shock attenuation) can be assessed using a force platform extracting several param-
eters from the vertical ground reaction force. The impact peak and the loading rate
are the most commonly used parameter to estimate shock attenuation properties of
footwear. There have been some concerns that these parameters do represent more
the acceleration properties of the center of mass rather than the foot-ground interac-
tion. A more advanced data analysis technique has been suggested to overcome this
problem. One of these is using a fast Fourier transformation (FFT) to transform the
data into the frequency domain and then using an inverse FFT to return only the high
frequency to the time domain. Dynamic shock attenuating properties of footwear can
also be measured using in-sole pressure measurements which gives data on vertical
loading and pressure distribution on the sole of the foot inside the shoe. However,
results from these systems need to be interpreted with caution as the quality of the
data depends largely on the technical specifications of the sensors. If, for example,
the sensors have a high hysteresis and are not linear, then results could be
4 B. Vanwanseele
problematic. Another potential method which is commonly used to investigate

dynamic shock attenuation is the analysis of 3D acceleration on the tibia. The
major advantage of this method is that measurements can be done in an ecological
environment and during a longer period of time. However, similar to the pressure
insoles, the technical specification (range, sample frequency) will determine the
reliability of the data. Parameters that have been used are peak vertical acceleration;
however, this variable can be influenced by sensor placement (medial or lateral) as
well as the attachment of the sensor. Similar to the force plate data, these drawbacks
of the method can be minimized by looking at the acceleration signal in the
frequency domain instead of the time domain.
In a systematic review and meta-analysis of the shock attenuating properties of
shoes in comparison to barefoot (Fong Yan et al. 2013), shoes in general (athletic
shoes, Oxford shoes, rocker soled shoes and five-finger shoes) were found to
decrease tibial peak axial acceleration and plantar pressures during walking and
running. More specifically, during walking, shoes reduced plantar pressures and
axial tibial acceleration, but vertical ground reaction force and loading rate were not
affected. While during running, in general shoes reduced axial tibial acceleration and
vertical ground reaction force, while loading rate was not affected. In contrast,
athletic shoes increased peak vertical ground reaction force but decreased loading
rates during walking and running. The reduction in loading rate, plantar pressure,
and tibial acceleration by shoes is potentially beneficial as these variables have been
associated with increased rates of lower limb injury and foot pain (Davis et al. 2004;
Milner et al. 2006; Pohl et al. 2008; Zifchock et al. 2008; McLean et al. 2011; Burns
et al. 2005). However, prospective studies supporting the beneficial effects of
cushioned shoes on the prevention of lower limb injury are limited and at best
confusing.
Functional Effect of Shoe: Control of Foot Movement
The shoe will not only protect the feet and reduce shock but will also control or
influence the movement of the feet especially the medial arch and the rearfoot. In
general, footwear has a splinting effect on the movement of the different segment of
the foot. Shoes reduce the torsion motion of the forefoot relative to the rearfoot
during the loading phase of running and landing tasks compared to barefoot. This
reduction in torsion during the loading phase is then compensated by an increase in
rearfoot pronation. To allow natural foot torsion and consequently reduce the
compensatory rearfoot pronation, Stacoff and colleges suggested at the end of the
1980s to construct the forefoot and rearfoot section of the shoe independently.
Splinting Effect
The use of three-dimensional multisegment foot models in biomechanical research
enabled more research on the splinting effect of the shoes. It became clear that shoes
reduce midfoot motion during propulsion in all three planes of movement as well as
the motion at the first metatarsophalangeal joint (MTPJ). Midfoot sagittal plane
motion during propulsion was reduced by 26 during shod walking and by 17
during shod running. Shoes also reduce motion at the first MTPJ during propulsion
of walking by 25 and during running by 19 . Sagittal plane ankle motion during
propulsion increased during shod running by 7 (Wegener et al. 2015). Shoes do not
only have a splinting effect, especially on the midfoot plantarflexion but also reduce
the power contribution of the midfoot from 39 % in barefoot to 25 % in shoes during
propulsion of running. Surprisingly, the splinting effect of shoes is still present in
flexible shoes designed to complement children’s barefoot function and in adults
wearing flexible shoes. It has been hypothesized that the restriction in midfoot
motion maybe partially due to the inhibition of the windlass mechanism caused by
a reduction in the first MTPJ motion during shod gait. The reduction in midfoot
motion during propulsion of shod walking and running may also have implications
for gait and running efficiency. Midfoot motion during propulsion is partially
attributed to the return of stored elastic energy as part of the windlass mechanism
of the plantar aponeurosis. In addition to the release of stored elastic energy, the
tightening of the plantar aponeurosis during MTPJ dorsiflexion as part of the
windlass mechanism magnifies the force generated by the Achilles tendon. The
storage and return of elastic energy during the windlass mechanism are considered
to be a passive mechanism which is beneficial for gait efficiency. Therefore, reduc-
tions in metatarsal joint motion and alterations to the windlass mechanism during
shod gait may impair gait efficiency due to potential reductions in the magnification
of Achilles tendon force and reduced release of stored elastic energy within the
plantar fascia. However further research is required to investigate the effect of shoes
on gait efficiency.
The splinting effect of shoes on midfoot plantarflexion may have a number of
implications to foot function and injury. The reduction midfoot plantarflexion during
shod walking and running can be compensated by an increase in ankle plantarflexion
motion and power production. This increase in ankle sagittal plane motion during
shod walking and power generation during shod running may increase the work of
the Achilles tendon and triceps surae muscle complex. In addition, a reduction in
midfoot motion during shod gait potentially reduces the stimulus to foot muscula-
ture, as the use of flexible shoes has been shown to increase foot muscle strength.
Further analysis of the effect of shoes on the kinetics of the midfoot is required due to
the extensive kinematic midfoot changes during shod gait. Considering the impli-
cations of limited midfoot plantarflexion during shod gait, further investigation into
possible shoe modifications to improve the contribution of the midfoot to propulsion
is required.
Control Rearfoot Motion

One mainstream shoe technology which tries to reduce excessive rearfoot movement
during running is the “motion control” design. The mechanism of motion control
shoes was based on different deformation rates between lateral and medial midsoles
and wedging heel counter so as to control the relative rates of midfoot and rearfoot
motion. The efficacy of motion control shoe on controlling the foot pronation is still
not clear with the early studies clearly reporting a reduction in foot pronation and a
6 B. Vanwanseele
lengthening of the time to reach maximal pronation velocity, while other studies
reported no changes in foot pronation during running with motion control shoes. A
systematic review of Cheung et al. (2011) pooled the results of five studies and
concluded that motion control shoes do significantly reduce foot pronation on
average by 2 . As foot pronation and tibial rotation are closely linked, it can be
expected that any effect of the footwear on foot pronation will have an effect on tibial
rotation and therefore the knee joint. Therefore the effect of motion control shoes on
proximal joint (knee and hip) has also been examined in several studies. In contrary
to the expectation, no clear effects have been reported on the more proximal joints.
One of the explanations for this lack of results could be the intersubject variability in
the relationship between tibial and foot movements. Pronation of the foot is also
known as a mechanism to attenuate the shock at foot contact. Reducing the foot
pronation could therefore also increase impact. Pooled results showed that motion
control shoe reduce the impact loading by an average of 9 % of body weight.
Functional Effect of Shoe: Reduce Loading in Other Joints
As footwear can change the vertical ground reaction forces and the foot movement, it
also has the potential to change joint loading (joint moments and joint contact
forces). Although footwear is not primarily designed to change loading at the
joint, there has been research showing that alterations in the movement of the foot
can indirectly change the loading at different joints through the kinetic chain. The
first study by Bergman et al. (1995) using an instrumented prosthesis reported only a
minimal influence of footwear on hip joint force and bending moment. Shoes in
general did increase the joint force and the bending moment at the implant slightly,
but the torsional moment rose by up to 50 %. Only shoes with very hard soles were
clearly disadvantageous, while soft heels, soles, or insoles did not offer advantages
to hip joint loading. As this study found that shoes that were less fixed to the foot
such as clogs caused larger variation in hip joint loading, they suggested that gait
stability might play a larger role than the softness of the sole in increasing the joint
loading. Therefore it was suggested to be used as a criterion for the choice of
footwear. Smooth gait patterns with soft heel strikes could potentially reduce joint
loading during slow jogging. In a later study by the same group, the same results
were confirmed for the tibiofemoral joint, e.g., footwear tended to increase knee joint
loading slightly, with the dress shoe being the most unfavorable type of footwear. At
the early stance phase, all load components were increased by all shoe types. The
resultant force rose by 2–5 %, the internal adduction moment by 7–12 %, and the
forces on the medial compartment by 3–5 %. Significant reductions of the resultant
force were solely observed for the advanced running shoe ( 6 %) and the MBT
( 9 %) shoe at late stance. Also the medial compartment force was slightly yet
nonsignificantly reduced by 2–5 % with these two types of footwear. As shoes are
worn during most of the day, especially during ambulation, their influence on the
lower extremity joint loading is crucial to determine especially in population that are
at risk for degenerative changes to the joints. In patients with osteoarthritis, the use of
modern comfortable self-chosen walking shoes results in an approximately 14 %

increase in dynamic loading of the knees during ambulation, compared to barefoot
walking (Shakoor and Block 2006). Changes to the shoes to make them more flat
and flexible reduced the peak external knee adduction moment by 12 % compared
with the control shoe (Shakoor et al. 2008). All of the reduction in medial compart-
ment loading are quite small, and the clinical significance of these changes remains
to be determined. It is currently unknown if adjusted footwear in patients with knee
OA could reduce the disease progression. Future research is necessary to examine
which factors regarding the shoe design, such as heel height, arch support, or
flexibility, are most decisive for a reduction of knee joint loading. In sports, we
have recently shown that restricting the ankle rearfoot motion by means of high top
shoes increases the ankle plantar flexion moment and the peak knee internal rotation
moment compared to the standard condition (Vanwanseele et al. 2014). Although the
increases were significant, they are still small and the clinical and long-term effect
remains to be investigated.
Functional Effect of Shoe: Improve Performance
One of the theoretical functional advantages of a rigid shoe could be minimizing

energy loss at the level of the toes, improving performance. As discussed in the
paragraph about the splinting effect, this ignores the power contribution of the
midfoot during propulsion and the possible reduction in the midfoot power produc-
tion when shod. The performance in jumping tasks is worse when shod compared to
barefoot which might be explain by this reduction in midfoot power. Our own work
did not show the same results in children, where we found a trend for the conven-
tional sports shoes to improve horizontal jump performance. Potential apprehension
of a lack of shock attenuation and protection while barefoot may have inhibited the
distance jumped by children barefoot in the standing horizontal jump. However the
task was undertaken on a soft carpeted surface to reduce children’s potential
apprehension of landing on a firm surface. We did find an impairment in vertical
jump height in shoes which might be related with the additional weight of shoes or
the reduction in total power production due to a restriction in midfoot plantar flexion.
The contrasting nature of effect of shoes on vertical and horizontal jumps warrants
further examination utilizing kinetic analysis to identify the mechanisms behind
changes in performance between barefoot and shod conditions. When developing a
modified flexible shoe, we showed that children’s performance of physical tasks was
not deteriorated compared to an unmodified shoe. Increased midfoot plantarflexion
and power production in the modified flexible shoe did not alter children’s perfor-
mance of power-based activities of jumping and running agility compared to a
conventional children’s sports shoe. During a standing vertical jump, the increase
in ankle power in a conventional children’s sport’s shoe offsets the performance
benefit of an increase in midfoot power generated in the modified shoe. As men-
tioned before, the increase in ankle power in the stiffer conventional sports shoe
maybe due to the splinting of the metatarsophalangeal joint thereby reducing the
8 B. Vanwanseele
energy lost at the metatarsophalangeal joint compared to the more flexible modified
shoe. To obtain an increase in performance while shod, it appears that maximizing
midfoot power while simultaneously minimizing metatarsophalangeal joint power
absorption is required.
Functional Effect of Shoe: Traction and Friction
Another less commonly investigated function of footwear is providing traction and

friction. In certain sports, acceleration, deceleration, and changes in directions are
crucial. Footwear needs provide enough friction between the athlete and the floor to
perform these movements in a safe and efficient way. The slip and traction resistance
of footwear could be defined as the capability of the outersole to transfer horizontal
applied forces to the ground. Traction is determined by the interaction between
footwear properties (sole architecture) and playing surface. As too high slip resis-
tance will increase the risk of injury, there is a need for custom-made slip resistance
properties of the shoes depending on the functional activity and even the surface
played on. In different sports also different tractions are needed on different areas of
the shoe. Different traction types exist: static and dynamic traction depending on the
amount of movement between the outersole and the surface and translation and
rotation traction which is the resistance of the footwear with respect to translation
and rotation. Increased traction could have some advantages toward sports perfor-
mance such as quicker changes in directions and faster acceleration and decelera-
tions. On the contrary higher traction could also have some negative effects on joint
loading and therefore increase risk of injuries. An appropriate traction or slip
resistance will therefore be a careful balance between the two: high enough for
performance and low enough to avoid injuries. This is one of the reasons why
specific shoes are developed for netball and volleyball where rotation traction on
the forefoot are, for example, high, and therefore threads are introduced in this part
of the outersole.
Future Directions
The functional characteristics of the footwear have been well documented in isola-
tion as well as in a laboratory setting. However shoes are not just worn during a
couple of minutes or in a controlled environment, and it still not clear how shoes will
function throughout the day or during different tasks and on different surface that are
part of daily life activities. With the development of new mobile sensor technology,
we also might be able to test the effects of footwear long-term and in the ecological
environment of the athlete or patient. Certain footwears have been shown to acutely
influence foot motion, shock absorption, and joint loading; however the influence of
these changes on injury, performance, or disease progression has not been investi-
gated. There is a clear lack of prospective studies that demonstrated the clinical or
performance benefit of footwear.
Cross-References

Applications
References
Bergmann G, Kniggendorf H, Graichen F, Rohlmann A (1995) Influence of shoes and heel strike on
the hip joint. J Biomech 28(7):817–827
Burns J, Crosbie J, Hunt A, Ouvrier R (2005) The effect of pes cavus on foot pain and plantar
pressure. Clin Biomech 20:877–882. doi:10.1016/j.clinbiomech.2005.03.006
Cheung RTH, Wong MYM, Ng GYF, Cheung RTH, Wong MYM, Ng GYF (2011) Effects of
motion control footwear on running: a systematic review. J Sports Sci 29(12):1311–1219. http://
doi.org/10.1080/02640414.2011.591420
Davis I, Milner CE, Hamill J (2004) Does increased loading during running lead to tibial stress
fractures? A propective study. Med Sci Sport Exerc 36:S58. doi:10.1097/00005768-200405001-
00271
Fong Yan A, Sinclair PJ, Hiller C et al (2013) Impact attenuation during weight bearing activities in
barefoot vs. shod conditions: a systematic review. Gait Posture 38:175–186. doi:10.1016/j.
Fredericks W, Swank S, Teisberg M, Hampton B, Ridpath L, Hanna JB (2015) Lower extremity
biomechanical relationships with different speeds in traditional, minimalist, and barefoot foot-
wear. J Sports Sci Med 14:276–283, October 2014.
McLean SG, Oh YK, Palmer ML et al (2011) The relationship between anterior tibial acceleration,
tibial slope, and ACL strain during a simulated jump landing task. J Bone Joint Surg Am
93:1310–1317. doi:10.2106/JBJS.J.00259
Milner CE, Ferber R, Pollard CD et al (2006) Biomechanical factors associated with tibial stress
fracture in female runners. Med Sci Sports Exerc 38:323–328. doi:10.1249/01.
mss.0000183477.75808.92
Novacheck TF (1998) The biomechanics of running. Gait Posture 7:77–95
Pohl MB, Mullineaux DR, Milner CE et al (2008) Biomechanical predictors of retrospective tibial
stress fractures in runners. J Biomech 41:1160–1165. doi:10.1016/j.jbiomech.2008.02.001
Shakoor N, Block JA (2006) Walking barefoot decreases loading on the lower extremity joints in
knee osteoarthritis. Arthritis Rheum 54(9):2923–2927. http://doi.org/10.1002/art.22123
Shakoor N, Lidtke RH, Sengupta M, Fogg LF, Block JA (2008) Effects of specialized footwear on
joint loads in osteoarthritis of the knee. Arthritis Care Res 59(9):1214–1220. http://doi.org/10.
1002/art.24017
Wegener C, Greene A, Burns J, Hunt AE, Vanwanseele B, Smith RM (2015) Human movement
science in-shoe multi-segment foot kinematics of children during the propulsive phase of
walking and running. Hum Mov Sci 39:200–211. http://doi.org/10.1016/j.humov.2014.11.002
10 B. Vanwanseele
Vanwanseele B, Stuelcken M, Greene A, Smith R (2014) The effect of external ankle support on
knee and ankle joint movement and loading in netball players. J Sci Med Sport 17(5):511–515.
http://doi.org/http://dx.doi.org/10.1016/j.jsams.2013.07.009
Zifchock RA, Davis I, Higginson J et al (2008) Side-to-side differences in overuse running injury
susceptibility: a retrospective study. Hum Mov Sci 27:888–902. doi:10.1016/j.
humov.2008.03.007
Gait Rehabilitation with Exoskeletons
Stefano Federici, Fabio Meloni, and Marco Bracalenti
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Exoskeletons’ Development and Technical Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Technical Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
The Effectiveness of Powered, Active Lower Limb Exoskeletons in Gait Rehabilitation . . . . . 8
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Abstract
The exoskeleton is a robotics-assisted, powered device that enables paralyzed
patients to stand up and walk. This chapter examines the state of art concerning
the use of active, powered, and wearable lower limb exoskeletons for aiding and
rehabilitating paraplegic patients’ gait disorders resulting from serious central
nervous system lesions. A qualitative analysis of the literature review found that
the rehabilitative use of an exoskeleton is safe and practical, not physically
exhausting, and requires just a little cognitive effort. In addition, exoskeleton
use is easy to learn, increases mobility and functional abilities, and decreases the
risk of secondary injuries, restoring a gait pattern comparable to normal when
walking over ground. Nevertheless, the rehabilitative use of an exoskeleton has
some important limitations: the wearability criteria are too restrictive, the training
to use it autonomously at home is very complex, and the device is still extremely
expensive. A further limitation is the scarcity of experimental designs that
S. Federici (*) • F. Meloni • M. Bracalenti

Department of Philosophy, Social & Human Sciences and Education, University of Perugia,
Perugia, Italy
e-mail: stefano.federici@unipg.it; stefano.federici@gmail.com; fa.meloni@gmail.com; marco.
bracalenti90@gmail.com

DOI 10.1007/978-3-319-30808-1_80-1
2 S. Federici et al.
demonstrate the effectiveness of the exoskeleton compared to other rehabilitative

techniques and technologies.
Keywords
Powered active lower limb exoskeleton • Paraplegic patients • Gait disorders •
Central nervous system lesions • Gait rehabilitation
State of the Art
Gait disorders, classified in ICD-10 as “abnormalities of gait and mobility” (code

R26), involve a reduction in autonomy and in the ability to move independently. Gait
disorders can result from serious central nervous system (CNS) lesions due to, for
example, spinal cord injury (SCI), cerebrovascular accident (CVA), cerebral palsy,
and infectious diseases (Dickstein et al. 2014; Lee et al. 2014). Usually, the patient is
forced to rely on a wheelchair for mobility and often requires support from a
caregiver. Over time, patients may also develop secondary complications such as
hypertension, osteoporosis, and bedsores. These comorbidities severely limit the
individual’s ability to carry out activities of daily living, restrict social participation,
and affect the quality of life and mood (Suzuki et al. 2005).
A new neurorehabilitation therapy has been developed, different from the classi-
cal therapeutic techniques and based on robotic devices; it is referred to as
neurorobotic or neuroprosthetic training and includes devices such as the exoskel-
eton (Moreno et al. 2011). Spungen et al. (2013) maintain that robotics-assisted
powered exoskeletons represent a relatively new technology that has been shown to
be safe and effective in helping individuals with motor complete paraplegia to stand
and walk. An exoskeleton allows a paralyzed person to overcome environmental
barriers that preclude wheelchair use, such as stairs.
Initially, exoskeletons were used in gait disorder rehabilitation to assist with the
recovery of upper limb movement (Chaigneau et al. 2008). Researchers observed
that motor patterns of the hip, pelvis, and legs remained unchanged when users wore
a robotic gait-assisting exoskeleton (Kao et al. 2010; Lee et al. 2014) that reduced the
muscular effort required (Mooney et al. 2014). The exoskeletons implemented to
assist paraplegic patients with lower limb movements were passive ones (Rahman
et al. 2006); that is, the exoskeleton moves the patient’s body on a predefined
trajectory, regardless of what the patient is doing (Nef and Riener 2012). Subse-
quently, the first active exoskeletons were built (Quintero et al. 2012) that enabled
patients to move together with the robot in the desired direction (Nef and Riener
2012). These allowed patients with gait motor disorders not only to stand bolt
upright but also to move autonomously in their surroundings.
The exoskeleton has been used to assist patients with SCI by restoring their
functional abilities (Spungen et al. 2013) and to adapt physical activity in sport – as
in the Rome marathon of March 22, 2015, when two paraplegic patients ran 1 km
Gait Rehabilitation with Exoskeletons 3
wearing an exoskeleton (Fondazione Santa Lucia 2015) – as well as to enhance

strength and muscular endurance in military operations (Herr 2009). Unlike the
human skeleton, an exoskeleton supports the body weight externally, allowing the
user to follow through and to strengthen and improve the coordination of the
voluntary movements of the lower limbs. This technology has greater ecological
validity than other types of neurorobotic techniques – for example, robotic-assisted
gait training – as patients wearing an exoskeleton can walk and move autonomously
for a long period of time and on a wide range of walking surfaces. The range of
environments in which the exoskeleton can be used is also more extensive; it can be
used in the workplace or at home, as well as in a rehabilitation space, supporting
patients when performing functions critical to the activities of daily living.
Exoskeletons’ Development and Technical Data
Not all exoskeletons used to evaluate their effectiveness in gait rehabilitation are
currently commercialized. At present, only six main multinational companies pro-
duce and commercialize active lower limb exoskeleton suits; they are Ekso Bion-
ics™, Cyberdyne™, ReWalk Robotics™, Parker Hannifin Corporation™, Rex
Bionics™, and Honda™.
Below, we first present the development of the commercial exoskeletons and then
the main technical data of each of them.
Development
Ekso Bionics™ Pioneering the world of exoskeletons, Berkeley ExoWorks™ was

founded in 2005 when Kazerooni, Angold, and Harding partnered with members of
the Berkeley Robotics and Human Engineering Laboratory at the University of
California, Berkeley. The exoskeleton allows users to carry and even run with
weights of up to 200 lbs, over a variety of terrains. In addition, Berkeley Exo-
Works™ introduced the ExoClimber. It was designed to rapidly ascend stairs and
steep slopes. In 2007, Berkeley ExoWorks™ became Berkeley Bionics™. In 2008,
Berkeley Bionics™ unveiled their third-generation exoskeleton, the Human Univer-
sal Load Carrier (“HULC”). In 2010, Berkeley Bionics™ debuted what was then
called the “Exoskeleton Lower Extremity Gait System” (eLEGS), an intelligent,
bionic exoskeleton that actually allows wheelchair users to stand and walk over
ground; this was developed by Strausser. In 2011, Berkeley Bionics™ became Ekso
Bionics™, and, a year later, it produced “Ekso,” the first commercialized robotic
exoskeleton for use in rehabilitative and medical facilities.
Cyberdyne™ The first Cyberdyne exoskeleton prototype, named Hybrid Assistive

Limb (“HAL”) was proposed in 1989 by Sankai, a professor at Tsukuba University
(Japan). The third HAL prototype, developed in the early 2000s, was attached to a
computer. Its battery alone weighed nearly 22 kilograms (49 lb) and required two
helpers to put on, making it very impractical. By contrast, a later “HAL-5” model
weighed only 10 kilograms (22 lb); its battery and control computer were strapped
around the waist of the wearer. Cyberdyne™ began renting out the HAL suit for
medical purposes in 2008. In 2012, Cyberdyne™ was certified as ISO 13485 by
Underwriters Laboratories. In 2013, the HAL suit received a global safety certificate,
becoming the first powered exoskeleton to do so, as well as the European Confor-
mity Certificate that permitted its use for medical purposes in Europe as the first
medical treatment robot of its kind.
ReWalk Robotics™ ReWalk Robotics (originally “Argo Medical Technologies”) is

an Israeli medical device company that designs, develops, and commercializes
exoskeletons, allowing wheelchair-bound individuals to stand and walk again. In
2001, ReWalk Robotics created the ReWalk powered exoskeleton, approved by the
US Food and Drug Administration (FDA) in 2011. Currently, ReWalk designs are
intended for people with paraplegia. The system uses patented motion-sensing
technology, along with battery-powered motorized legs powering knee and hip
movements. The system is controlled by proprietary onboard computers and soft-
ware. The ReWalk system allows the user to sit, stand, walk, turn, and climb and
descend stairs. ReWalk provides two different operative systems: “personal,”
designed for everyday and all day use by individuals at home and in their commu-
nities, and “rehabilitation,” designed for training exercises and therapy in clinical
rehabilitation settings.
Parker Hannifin Corporation™ Vanderbilt University began testing its exoskele-

ton with paraplegics at a rehabilitation center in Atlanta, in 2010. In 2012, Hannifin
signed an exclusive licensing agreement with Vanderbilt University for the right to
develop and manufacture a commercial version of the exoskeleton, which it plans to
release under the name Indego. In 2014, Parker Hannifin entered into clinical trial
agreements for the exoskeleton with several major medical rehabilitation centers.
Rex Bionics™ The British engineers Richard Little and Robert Irving set about
designing Robotic Exoskeleton (REX) when Robert was diagnosed with multiple
sclerosis in 2003. In 2007, Rex Bionics completed the first prototype REX Personal.
The REX Rehab is designed for use in rehabilitation settings and can be adjusted to
fit any user in few minutes. REX Rehab allows users to experience the psychological
and physiological benefits of standing and walking again, as a part of a regular
rehabilitation program.
Honda™ In 1999, in the area of humanoid robot research, Honda developed their
first exoskeleton, the “Walking Assist Device.” In 2013, Honda began leasing the
Walking Assist Device to hospitals in Japan that provide rehabilitation training and
physical therapy in the area of walking, to monitor its use and verify the practicality
of the device. Finally, in 2015, this exoskeleton was officially commercialized in
Japan.
Technical Data
Ekso Bionics™ Ekso GT, the latest generation of Ekso Bionics’ exoskeletons{ XE
“exoskeleton” }, is a wearable bionic suit that enables individuals with any amount
of lower extremity weakness to stand up and walk over ground with a natural, full
weight bearing, reciprocal gait. Walking is achieved by shifting the user’s weight to
activate sensors in the device which initiate steps. Battery-powered motors drive the
legs, replacing deficient neuromuscular function. It provides a means for people with
even complete paralysis and minimal forearm strength to stand and walk. Moreover,
it helps patients to relearn proper step patterns and weight shifts using a functional
based platform and facilitates intensive step dosage over ground. Ekso can be used
by patients with various levels of paralysis or hemiparesis due to neurological
conditions such as CVA, SCI or disease, traumatic brain injury, or other illnesses.
Designed for utility and ease of use in a clinical setting, Ekso accommodates an
unprecedented spectrum of patients in motor ability.
The Ekso latest release has been designed for the needs of busy therapists treating
a wide range of patients in a single day. The suit is strapped over the user’s clothing
with easy adjustments, enabling transition between patients in as little as 5 min. Step
Generator software helps patients walk in their first session to quickly achieve work
on gait patterning or step dosage. Progressive step modes help patients to develop
their skills. It represents a tool to enforce proper biomechanical alignments and
symmetrical gait patterns over ground.
• Functioning: the Ekso exoskeleton adopts a human–computer interface based on

gesture, which exploits sensors, observes actions made to determine intentions,
and acts accordingly.
• Weight: 23 kg.
• Maximum Speed: 1.6 km/h.
• Battery Life: over 6 h.
• Movements: walking in a straight line, standing up from a sitting position,
remaining raised for a long period of time, and sitting down from a stand up
position.
• Crutches: necessary for this exoskeleton.
• Who Can Use: people who can move independently from a wheelchair to a chair
(height: 1.50 m/1.90 m; weight: not exceeding 100 kg).
Cyberdyne™ HAL is able to read bioelectric signals by only attaching the origi-
nally developed detectors on the surface of the wearer’s skin. By consolidating
various information, HAL recognizes the types of motions the wearer intends. In
accordance with the recognized motions, HAL controls its power units. This func-
tion enables HAL to assist the wearer’s movements as he or she intends and exerts
more power than he or she ordinarily exerts. Moreover, when HAL has appropriately
assisted the motions of “walking,” the feeling “I could walk!” is fed back to the
brain. By this means, the brain gradually becomes able to learn the way to emit the
necessary signals for “walking.” This leads to the important first step in walking by
the physically challenged person, without being assisted by HAL.
• Functioning: HAL is characterized by two control systems. For instance, when a

person attempts to walk, the brain sends electrical impulses to muscles. When
they arrive at the muscles, faint bioelectrical signals appear on the skin surface.
With the first of the two control systems, the Bio-Cybernetic Control System,
HAL assists the wearer with an intended movement. In the event that no good
bioelectrical signals are detectable due to some problems in the CNS or in the
muscles, HAL can be of use through the second of the two control systems, the
Robotic Autonomous Control System.
• Weight: 23 kg (whole body type) – 15 kg (legs-only type)
• Maximum speed: 1.6 km/h
• Battery life: 2 hours and 40 minutes
• Movements: standing up from a chair, walking, and going up- and downstairs
• Crutches: necessary for this exoskeleton
• Who can use: people with weakened muscles or people with disabilities caused by
stroke and/or SCI
ReWalk Robotics™ The ReWalk Personal System can be used at home, work, or
other locations. It functions outdoors and on different surfaces or terrains. The
ReWalk can facilitate patients in sitting, standing, turning and climbing, and
descending stairs. The key prerequisites for use include the ability to use hands
and shoulders, a healthy cardiovascular system, and a minimal bone density.
• Functioning: ReWalk control is based on motion sensors. Using sophisticated

algorithms, movements of the upper limbs are analyzed and used to trigger and
maintain the patterns of gait and other modes, such as going upstairs and moving
from a sitting to a standing position. In other words, ReWalk exoskeleton detects
movements of the upper limbs. The buttons on a remote control allows the user to
select various program settings and to choose the correct mode of movement:
walking, climbing stairs, sitting, getting up, etc.
• Weight: 18 kg.
• Maximum speed: 3 km/h.
• Battery life: 8 hours. It can be recharged during the night.
• Movements: standing, sitting, walking, going upstairs, and climbing and
descending slopes.
• Who can use: the system can accommodate a range of heights (160–190 cm) or
weights (up to 100 kg). The key prerequisites for use include the ability to use
hands and shoulders, a healthy cardiovascular system, and a minimal bone density.
Parker Hannifin Corporation™ Indego is a powered exoskeleton worn around the

waist and legs which allows individuals with SCI to stand and walk. At just 26 lbs,
Indego’s design has no exposed cables or upper body apparatus and does not require
bulky backpack-mounted components. Indego mirrors natural human movement,
leans forward to initiate standing or walking, and leans backward to stop and sit.
Moreover, Indego has a slim profile that is compatible with standard mobility aids
and can be worn while seated in a wheelchair. Furthermore, a single-hand strapping
and retention system allows Indego to be put on, taken off, and adjusted to fit without
assistance. Designed from the beginning for personal use, the features of Indego
make it well suited for use in the clinical setting and for an easy transition to use at
home.
• Functioning: with the Indego, patients with SCI or other motor problems strap
their lower bodies into a piece of equipment similar in appearance to leg braces.
Gyroscopes and accelerators anticipate a patient’s steps by subtle upper body
motion, similarly to how a Segway works. Then, the Indego moves in concert
with the patient’s leg to take a step. The wearer uses his or her muscles to do the
work; the Indego provides a little extra help. Sensors determine how much power
is needed, eventually decreasing as the patient grows stronger.
• Weight: 12.3 kg.
• Maximum speed: 3 km/h.
• Battery life: 4 h.
• Movements: standing, walking, and walking on hard surfaces (including ramps
and slopes).
• Who can use: Indego comes in interchangeable sizes and can accommodate
people from 155 cm to 193 cm in height and up to 113.5 kg in weight.
Rex Bionics™ Designed for people with mobility impairments, REX is completely
self-supporting and rapidly adjustable for each user. Rex Bionics is working with
physiotherapists to develop the practice of robot-assisted physiotherapy. In a session
of robot-assisted physiotherapy, REX lifts patients from a sitting position into a
robot-supported standing position, allowing them to take part in a set of supported
walking and stretching exercises designed by specialist physiotherapists.
• Functioning: REX is controlled by means of a joystick, not through sensors;

consequently no movement or function of nerves is needed to drive the
exoskeleton.
• Weight: 38 kg.
• Maximum speed: 1.8 km/h.
• Movements: standing, walking, moving sideway, turning around, going up- and
downstairs, and walking on hard surfaces (including ramps and slopes).
• Crutches: not necessary for this exoskeleton.
• Who can use: the system can accommodate a range of heights (142–195 cm),
weights (up to 100 kg), and hip width (up to 380 mm). The key prerequisites for
use include the ability to move autonomously and use a joystick.
Honda™ The Walking Assist Device features a function to influence the user to
achieve efficient walking based on the inverted pendulum model, which is a theory
of bipedal walking, and is designed as a device to be used in walking training. Three
training modes are available: (i) following, the exoskeleton influences the user’s
walking motion based on the walking pattern of the user; (ii) symmetric, based on
the walking pattern of the user, the device influences the user to achieve bilaterally
symmetric motions such as bending and extending both legs, and (iii) step, the
device influences the user’s steps repeatedly to recover the rocker functions which
enable the smooth shifting of weight.
• Functioning: the control computer activates motors based on data obtained from
hip angle sensors during walking, to improve the symmetry of the timing of each
leg lifting from the ground and extending forward and to promote a longer stride
for an easier walk.
• Weight: 2.7 kg, approximately.
• Maximum speed: n/a.
• Movements: indoor or outdoor (except when raining) on flat floor/ground.
• Crutches: not necessary for this exoskeleton.
• Who can use: the system can accommodate a range of size-adjustable frames,
making it possible for more people with various body sizes/types to use the
Walking Assist Device.
The Effectiveness of Powered, Active Lower Limb Exoskeletons

in Gait Rehabilitation
We conducted a systematic review aimed at examining the rehabilitative capacity of

current models of active, powered lower limb exoskeletons when used in paraplegic
patients with gait disorders resulting from CNS lesions caused by, for example, SCIs
or CVAs. Relevant publications were retrieved from searches of PubMed, EBSCO,
Web of Science, Google Scholar, Scopus, and ProQuest electronic databases. Table 1
summarizes the results of the reviewed 33 studies, published between 2001 and
2016, according to the nine data items (country, study design, exoskeleton, sample
inclusion criteria, sample size, sample type, measures, strengths, and weaknesses)
extracted from each reviewed study.
These studies, mainly in the area of medical and engineering research, were
designed to test the safety or effectiveness of a particular device in an experimental
Table 1 Summary of the 33 reviewed studies
Studies in
alphabetic Sample inclusion Sample Sample
order Country Study design Exoskeleton criteriaa size type Measuresb Strengths Weaknesses
Aach et al. Germany Uncontrolled HAL NA 4 SCI Observation The exoskeleton Absence of a
(2013) clinical trial measures: walking increases mobility randomized
distance, walking and functional clinical trial
speed, and abilities and design
walking time on decreases the risk comparing
treadmill of secondary exoskeleton
Outcome injuries use to
measures: conventional
10 MWT, TUG, gait training

and WISCI II
Aach et al. Germany Uncontrolled HAL 1. Traumatic SCI 8 SCI Clinical measures: The exoskeleton Absence of a
(2014) clinical trial with chronic Lower extremity increases mobility randomized
incomplete (ASIA circumferences and functional clinical trial
B/C/D) or Outcome abilities and design
complete paraplegia measures: decreases the risk comparing
(ASIA A) after 10 MWT, TUG, of secondary exoskeleton
lesions of the conus 6 MWT, WISCI II, injuries use to
medullaris/cauda LEMS, and conventional
equine with zones of Ashworth scale gait training
partial preservation
2. Motor
functions of hip and
knee extensor and
flexor muscle groups
to be able to trigger
the exoskeleton
(continued)
9
10
Table 1 (continued)
Studies in
Asselin et al. USA Uncontrolled ReWalk 1. 18–65 years of 8 SCI Observation 1. The Absence of a
(2015) clinical trial age; measures: oxygen exoskeleton randomized
2. Height uptake (VO2) and increases mobility clinical trial
152–193 cm heart rate (HR) and functional design
3. Weight < abilities and comparing
100 kg decreases the risk exoskeleton
4. At least of secondary use to
6 months after injury injuries conventional
5. Be 2. The gait training
nonambulatory exoskeleton use is
not physically
exhausting and
requires only a
little energetic
effort
Belforte et al. Italy Experimental Advanced NA 1 SCI Observation The device is Absence of a
(2001) trial reciprocating gait measures: ideal for the randomized
orthosis (ARGO) rotations and rehabilitation clinical trial
applied torques for stage, given the design
each joint; structure’s comparing
kinematic modularity and exoskeleton
magnitudes and the extremely use to
exchanged forces flexible means conventional
used to regulate gait training
gait
characteristics
S. Federici et al.
Benson et al. UK Uncontrolled ReWalk 1. 18–55 years of 5 SCI Outcome / 1. Absence
(2016) clinical trial age measures: 6 MWT, of a
2. Height TUG, 10 MWT, randomized
160–190 cm VAS, ISNCSCI, clinical trial
3. Weight < SCI-specific design
100 kg ADAPSS, and comparing
4. Male or ATD-PA exoskeleton
nonpregnant, use to
nonlactating female conventional
5. AIS (ASIA gait training
impairment scale) 2. User
grade A, B or & C, experience
according to the with an
ISNCSCI exoskeleton in
(International daily life

Standards for activities did
Neurological not meet
Classification of subjects’
SCI) expectations
6. Motor level of in terms of
injury from C7 to perceived
T12, according to benefits and
the ISNCSCI impact on
7. Onset spinal quality of life
cord injury/ 3. Presence
dysfunction at least of skin
1 before screening aberration
8. Able to stand or
maintain upright
position with or
without using a
standing device and
able to sit with hips
and knees ⩾90
11
flexion
(continued)
12
Table 1 (continued)
Studies in
Bishop et al. USA Uncontrolled Tibion bionic NA 1 SCI Outcome The rehabilitative Absence of a
(2012) clinical trial technologies measures: 6 MWT, use of the randomized
BBS, TUG, and exoskeleton to clinical trial
10 MWT restore gait design
disorders is safe comparing
and practical exoskeleton
use to
conventional
gait training
Bortole et al. Spain Experimental H2 NA 3 Stroke Outcome The rehabilitative Absence of a
(2015) trial measures: 6 MWT, use of the randomized
BBS, TUG, exoskeleton to clinical trial
Barthel index, and restore gait design
FM-LE disorders is safe comparing
Observation and practical exoskeleton
measures: walking use to
angular conventional
kinematics, gait training
interaction torques
and motor torques
for left and right
hip, and knee and
ankle joints
Esquenazi USA Experimental ReWalk 1. 18–55 years of 12 SCI Observation The exoskeleton 1. Absence
et al. (2012) trial age measures: distance allows restoration of a
Motor-complete walked in 6 min, of a gait pattern randomized
cervical (C7-8) and time spent to cover comparable to clinical trial
thoracic (T1-T12) 10 m design
S. Federici et al.
SCI Outcome normal over comparing
2. Male and measures: 6 MWT ground walking exoskeleton
nonpregnant, and 10 MWT use to
nonlactating woman conventional
3. At least gait training
6 months after injury
4. Regular use of a 2. Wearability
Reciprocating Gait criteria are too
Orthosis (RGO) or restrictive
KAFOs or able to
stand using a
standing device
Height 160–190 cm
5. Weight <
100 kg
Farris et al. USA Experimental Vanderbilt lower NA 1 SCI Outcome The exoskeleton Absence of a
(2014) trial limb measures: assists paraplegicrandomized
exoskeleton, 10 MWT, TUG, patients to clinical trial
knee–ankle–foot and 6 MWT perform gait design
orthoses activities faster comparing
(KAFOs) exoskeleton
use to
conventional
gait training
Farris et al. USA Experimental Vanderbilt lower NA 1 SCI Observation The exoskeleton Absence of a
(2011) trial limb exoskeleton measures: Hip and allows restoration randomized
knee joint angle of a gait pattern clinical trial
trajectories during comparable to design
walking and normal over comparing
average walking ground walking exoskeleton
speed use to
conventional
13
gait training
(continued)
14
Table 1 (continued)
Studies in
Farris et al. USA Experimental Vanderbilt lower NA 1 SCI Observation The exoskeleton Absence of a
(2012) trial limb exoskeleton measures: knee assists paraplegic randomized
joint torque patients to clinical trial
perform gait design
activities faster comparing
exoskeleton
use to
conventional
gait training
Hartigan USA Uncontrolled Indego 1. 18–55 years of 16 SCI Outcome The exoskeleton Absence of a
et al. (2015) clinical trial age; measures: increases mobility randomized
2. Complete or 10 MWT and and functional clinical trial
incomplete SCI, 6 MWT abilities and design
with injury levels decreases the risk comparing
ranging from L1 to of secondary exoskeleton
C5; injuries use to
3. Height conventional
155–191 cm; gait training
4. Weight <
113 kg;
5. AIS A, B, C, or
D
Ikehara et al. Japan Experimental Walking assist NA 2 NA Observation The device Absence of a
(2011) trial device measures: video reproduces the randomized
recording of power of kicking clinical trial
subjects’ walking motions at the design
ankle joints when comparing
exoskeleton
S. Federici et al.
controlled by the use to
hybrid system conventional
gait training
Kawamoto Japan Experimental HAL NA 1 Stroke Observation A larger knee Absence of a
et al. (2010) trial measures: knee angle is measured randomized
joint angle during leg flexion clinical trial
while participants design
wear the comparing
exoskeleton exoskeleton
use to
conventional
gait training
Kolakowsky- USA Uncontrolled Ekso NA 8 SCI Observation 1. The Absence of a

Hayner et al. clinical trial measures: Skin rehabilitative use randomized
(2013) evaluation; blood of the exoskeleton clinical trial
pressure; pain to restore gait design
level; spasticity, disorders is safe comparing
time, and level of and practical exoskeleton
assistance needed 2. The use to
to transfer into and exoskeleton conventional
on device; time increases mobility gait training
ambulating; and functional
assistive devices abilities and
used during decreases the risk
ambulation; step of secondary
length; and injuries
distance walked
(continued)
15
16
Table 1 (continued)
Studies in
Li et al. China Uncontrolled Bionic leg 1. Patients could 3 Stroke Outcome The use of the Absence of a
(2015) clinical trial actively move their measures: BBS, exoskeleton randomized
legs and walk more FM-LE, and improves the clinical trial
than 10 m without a electromyography participants’ gait design
walking aid (EMG) performance, comparing
2. No neurologic Observation muscle activation, exoskeleton
disorders, measures: video and walking use to
uncontrolled recording of speed conventional
hypertension, subjects’ walking gait training
cognitive deficits
(MMSE<23)
Mori et al. Japan Experimental ABLE system NA 1 NA Observation The subject Absence of a
(2006) trial measures: time succeeded in randomized
response of the standing up clinical trial
angles and electric wearing the design
currents of each exoskeleton comparing
joint exoskeleton
use to
conventional
gait training
Neuhaus USA Experimental MINA 1. Traumatic SCI 2 SCI Observation The exoskeleton Absence of a
et al. (2011) trial with chronic measures: use is not randomized
incomplete (ASIA conversation with physically clinical trial
B/C/D) or complete the user during exhausting and design
paraplegia (ASIA A) walking requires only a comparing
2. WISCI level little cognitive exoskeleton
9 or higher effort use to
conventional
S. Federici et al.
gait training
Nilsson et al. Sweden Uncontrolled HAL 1. Time since 8 Stroke Observation The rehabilitative 1. Absence
(2014) clinical trial stroke onset of < measures: walking use of the of a
7 weeks speed exoskeleton randomized
2. Able to sit on a Outcome restoring gait clinical trial
bench with/without measures: NIHSS, disorders is safe design
supervision at least FM-LE, BBS, and practical comparing
five minutes TUG, 10 MWT, exoskeleton
3. Unable to walk S-COVS, FAC, use to
independently FES(S), I, FIM, conventional
4. Sufficient and VAS gait training
postural control
5. Ability to 2. Wearability
understanding criteria are too
instruction restrictive
6. Body size
compatible with the
HAL suit
Quintero USA Experimental Vanderbilt lower NA 1 SCI Outcome The exoskeleton Absence of a
et al. (2012) trial limb exoskeleton measures: TUG provides randomized
appropriate user-
clinical trial
initiated control
design
of sitting, comparing
standing, and exoskeleton
walking use to
conventional
gait training
Raab et al. Germany Uncontrolled ReWalk NA 1 SCI Outcome The exoskeleton Absence of a
(2016) clinical trial measures: SF-36, increases mobility randomized
BBS, DGI and functional clinical trial
abilities and design
decreases the risk comparing
of secondary exoskeleton
injuries use to
conventional
17
gait training
(continued)
Table 1 (continued)
Studies in
18
Sanz- Spain Experimental ATLAS NA 1 NA Observation Experiments Absence of a
Merodio trial measures: joint validated a good randomized
et al. (2012) angles controlled clinical trial
performance in design
following the gait comparing
pattern given by exoskeleton
the parameterized use to
trajectory conventional
generator gait training
Sczesny- German Uncontrolled HAL NA 4 SCI Outcome Reduced Absence of a
Kaiser et al. clinical trial measures: fMRI somatosensory randomized
(2013) and EMG first area clinical trial
(S1) activation of design
the activated area comparing
in both exoskeleton
hemispheres after use to
tactile stimulation conventional
of the index finger gait training
Spungen USA Uncontrolled ReWalk 1. 18–65 years of 7 SCI Observation The exoskeleton 1. Absence
et al. (2013) clinical trial age measures: walking allows restoration of a
2. Motor- speed of gait pattern randomized
complete paraplegia comparable to clinical trial
(T1–T12) normal over design
3. Greater than ground walking comparing
6 months elapsed exoskeleton
since the SCI use to
4. Height conventional
160–190 cm gait training
5. Weigh <
100 kg 2. Wearability
S. Federici et al.
criteria are too

restrictive
Stein et al. USA Controlled Bionic leg 1. Single stroke 20 Stroke Outcome / Robotic
(2014) clinical trial (ischemic or measures: therapy for
hemorrhagic) 10 MWT, TUG, ambulatory
causing significant 6 MWT, 5XSST, stroke patients
leg weakness and Romberg’s test, with chronic
gait alterations at EFAP, BBS, and hemiparesis
least 6 months CAFE 40 and using a
before study entry robotic knee
2. Stroke brace resulted
confirmation in only modest
through CT or MRI functional
3. Independence benefits, in
in household comparison
ambulation with or with a group
without facilitator receiving only

exercise
intervention
Strausser and USA Experimental eLEGS NA 5 SCI Observation The exoskeleton Absence of a
Kazerooni trial measures: knee is easy to learn randomized
(2011) angle and time clinical trial
practicing design
comparing
exoskeleton
use to
conventional
gait training
Strausser USA Experimental Human NA 4 SCI Observation The exoskeleton Absence of a
et al. (2010) trial Universal Load measures: video increases mobility randomized
Carrier (HULC) recording of and functional clinical trial
subjects’ walking abilities and design
decreases the risk comparing
of secondary exoskeleton
injuries use to
conventional
19
gait training
(continued)
Table 1 (continued) 20
Studies in
Sylos-Labini Italy Experimental MindWalker 1. 18–45 years of 10 SCI Observation The exoskeleton 1. Absence
et al. (2014) trial exoskeleton age measures: joint allows restoration of a
2. Traumatic/non- angles and torques of gait pattern randomized
traumatic SCI Outcome comparable to clinical trial
3. At least measures: EMG normal over design
5 months after injury ground walking comparing
4. Complete exoskeleton
lesion below T7 use to
5. Inability to conventional
ambulate gait training
26 2. Wearability
criteria are too
restrictive
Talaty et al. USA Experimental ReWalk 1. Joint integrity/ 12 SCI Clinical measures: The exoskeleton Absence of a
(2013) trial absence of fractures isometric muscle provides randomized
that prevent walking strength fundamentally clinical trial
2. Standing alone Observation symmetric gait design
also with facilitator measures: video comparing
3. A complete recording of exoskeleton
neurological subjects’ walking use to
evaluation to assess and walking speed conventional
motor and Outcome gait training
physiological measures:
functioning EMG
4. Absence of
osteoporosis (BMD
> 2.5) at the right
limb femoral neck
and the L2 to L4
S. Federici et al.
spine
Tsukahara Japan Experimental HAL NA 1 SCI Observation The exoskeleton Absence of a
et al. (2009) trial measures: knee products a randomized
joint angle standing-up and a clinical trial
sitting motion design
support systems comparing
for completely exoskeleton
paraplegic use to
patients conventional
gait training
Tsukahara Japan Experimental HAL NA 1 SCI Observation The exoskeleton Absence of a
et al. (2010) Trial measures: products a randomized
knee joint angle standing-up and a clinical trial
sitting motion design

support systems comparing
for completely exoskeleton
paraplegic use to
patients conventional
gait training
Watanabe Japan Controlled HAL 1. Hemiparesis 22 Stroke Clinical measures:
et al. (2014) clinical trial resulting from isometric muscle
unilateral ischemic strength
or hemorrhagic Observation
stroke measures:
2. Time since maximum walking
stroke onset of < speed
6 months Outcome
measures: FAC,
TUG, 6 MWT,
SPPB, and FM-LE
Israel ReWalk 6 SCI
(continued)
21
22
Table 1 (continued)
Studies in
Zeilig et al. Experimental 1. 16–70 years of Observation The exoskeleton Absence of a
(2012)) trial age measure: distance increases randomized
2. Weight < walked in mobility and clinical trial
100 kg 6 minutes functional design
3. Height from Outcome abilities and comparing
155–200 cm measures: decreases the risk exoskeleton
4. Complete 10 MWT and of secondary use to
motor impairment TUG injuries conventional
C7–C8 or T1–T12 gait training
5. At least
6 months since
injury
6. Regular user of
a RGO or
therapeutic standing
frame
NA not available: in the study inclusion criteria were not reported
a
All the measures’ abbreviations are explained in the Table 2
b
See Table 2 for the measure reference details
S. Federici et al.
trial or to evaluate its rehabilitative effectiveness in an uncontrolled or controlled

clinical trial. Participants enrolled in experimental groups comprised patients with
CVA, SCI, and gait disorders without etiological data. They were recruited in a wide
range of countries, such as the USA, Japan, Germany, Sweden, Israel, Italy, and
Spain.
In the 33 studies included, three types of measures were used:
• Clinical measures, to define the subject’s functional profile before treatment –

such as isometric muscle strength (hip flexion and extension, knee flexion and
extension, trunk flexion and extension, ankle dorsi-/plantar flexion) and lower
extremity circumferences
• Outcome measures (standardized tests, neurophysiological, or neuroimaging
techniques), to evaluate the effectiveness of a treatment
• Observation measures, to observe the subject’s functional performance while
wearing the exoskeleton – such as knee joint angle or torque, a video recording
of a subject walking, hip and knee joint angle trajectories during walking, the time
practicing, and distance walked in a certain time period.
Table 2 provides details of the indicators and the studies in which they were used.
In the following subsections, we discuss 16 brands of exoskeletons experimen-
tally tested in the scientific literature, of which five have already been commercial-
ized (HAL, Ekso, ReWalk, Indego, and Walking Assist Device) and 11 are not yet
commercialized (Vanderbilt lower limb exoskeleton, Human Universal Load Carrier
(HULC), MindWalker exoskeleton, advanced reciprocating gait orthosis (ARGO),
eLEGS, X1 robotic exoskeleton (MINA), ATLAS, ABLE system, Tibion Bionic
Technologies, Bionic Leg, and H2 robotic exoskeleton).
Hybrid Assistive Limb (HAL) The HAL exoskeleton was tested with paraplegic
subjects with a SCI at the T10 level, complete or incomplete, or having hemiparesis
after a stroke (Aach et al. 2013, 2014; Kawamoto et al. 2010; Nilsson et al. 2014;
Sczesny-Kaiser et al. 2013; Tsukahara et al. 2009, 2010; Watanabe et al. 2014).
Clinical, observational, and outcome measures administered to investigate the

effectiveness of HAL varied. For example, Sczesny-Kaiser et al. (2013) used
functional magnetic resonance imaging and electromyography (EMG) to evaluate
cortical excitability and plastic changes after a three-month period of treadmill
training supported by HAL.
Taken together, the results of these studies guaranteed the system’s safety (Nils-
son et al. 2014) and its effectiveness. Paraplegic patients gained significant increases
in over ground walking functional abilities (Aach et al. 2013, 2014), and a larger
knee angle was measured during leg flexion (Kawamoto et al. 2010). In addition,
diagnostic imaging displayed an augmented paired pulse inhibition of somatosen-
sory evoked potentials in both hemispheres following median nerve stimulation at
the wrist. There was also a reduced somatosensory cortex activation of the activated
area in both hemispheres after tactile stimulation of the index finger (Sczesny-Kaiser
et al. 2013). Finally, even a gait training program with the single-leg version of HAL
could facilitate independent walking more efficiently than conventional gait training
(Watanabe et al. 2014).
ReWalk The ReWalk exoskeleton was tested with paraplegic patients with complete
SCIs, at the C7-T12 and the T1-T12 level.
By using the ReWalk, paraplegic patients were able to walk independently,

supervised by one person (Raab et al. 2016), and to achieve a level of walking
proficiency that was close to that needed for limited community ambulation in an
urban setting (Asselin et al. 2015; Benson et al. 2016; Esquenazi et al. 2012;
Spungen et al. 2013), for example, for a distance of 100 m (Zeilig et al. 2012),
with a fundamentally symmetrical gait (Talaty et al. 2013). Daily use of the exo-
skeleton seemed to increase activity energy expenditure, but this would be expected
to have positive cardiopulmonary and metabolic benefits. The level of effort required
to use the ReWalk exoskeleton system to ambulate appears to be acceptable and, as
such, could be envisioned as a device that people with SCI would use in their daily
lives (Asselin et al. 2015).
Moreover, quality of life, mobility, risk of falling, motor skills, and control
of bladder and bowel functions were improved after robot-assisted gait
training (Raab et al. 2016). Nonetheless, the presence of skin aberrations was
unexpectedly high, and the use of the exoskeleton generally did not meet subjects’
expectations in terms of perceived benefits and impact on quality of life (Benson
et al. 2016).
Vanderbilt Lower Limb Exoskeleton The Vanderbilt lower limb exoskeleton was
developed by a team of engineers of the Vanderbilt University, at Nashville in
Tennessee, chaired by H. A. Quintero (Farris et al. 2011, 2012, 2014; Quintero
et al. 2012). Their clinical trials involved paraplegic patients with motor and sensory
complete SCIs at the T10 level.
Findings showed the Vanderbilt lower limb exoskeleton system assisted paraple-
gic patients to perform gait activities faster (Farris et al. 2012, 2014), with knee and
hip joint amplitudes similar to those observed in non-SCI walking (Farris et al.
2011).
Human Universal Load Carrier (HULC) The HULC was developed by

H. Kazerooni and his team at Ekso Bionics in the USA. Given that the HULC is
designed to assist able-bodied individuals by powering knee movements only in
extension, just one study on its effectiveness met our inclusion criteria (Strausser
et al. 2010). In this study, double-acting hydraulic cylinders replaced the single
acting ones of the HULC, providing powered flexion and extension at the knee.
Likewise, bracing used in the exoskeleton was augmented to support a patient with
limited leg and torso muscle control. This clinical trial involved paraplegic patients
with a motor complete SCI at the T5-T10 level.
The study’s purpose was to discover whether the development of an exoskeleton

for medical use would facilitate an active life in paraplegic people, so reducing the
occurrence of secondary complications. The results confirmed that the HULC
exoskeleton, when readapted for medical use, is able to safely increase mobility
for those who are unable to walk unaided.
MindWalker Exoskeleton The MindWalker was developed by a European consor-

tium coordinated by M. Ilzkovitz and funded by the European Commission in 2009.
Only one study (Sylos-Labini et al. 2014) involved participants with an SCI at the
T7-L1 level.
The aim of this study was to quantify the level of muscle activity in a sample of
intact and injured patients while they walked with MindWalker. The measures used
were EMG activity, joint angles, and torques. The results showed that, in SCI
patients, EMG activity of the upper limb muscles was augmented, while activation
of the leg muscles was minimal. Contrary to expectations, however, in the neuro-
logically intact subjects, EMG activity of the leg muscles was similar, or even
greater, during exoskeleton-assisted walking compared to normal over ground
walking. In addition, significant variations in the EMG waveforms were found in
different walking conditions; the most variable pattern was observed in the ham-
string muscles. Overall, the results are consistent with a nonlinear reorganization of
the locomotor output when using the robotic stepping devices.
Advanced Reciprocating Gait Orthosis (ARGO) Only one study tested the ARGO,
produced by RSLSteeper, with paraplegic patients (Belforte et al. 2001). This
clinical trial involved one participant with a motor complete SCI at the T3 level,
enrolled in Italy. The design and construction of the ARGO, and experimental testing
to assist locomotion in paraplegic subjects when using the ARGO, were described.
Findings showed the device is ideal for the rehabilitation stage, given the structure’s
modularity and the extremely flexible means used to regulate gait characteristics.
Walking Assist Device The Walking Assist Device was tested by a team of engi-
neers led by T. Ikehara.
One of their studies (Ikehara et al. 2011) involved two participants with motor
paralysis, enrolled in Japan. The results of the experimental study showed that the
device could reproduce the power of kicking motions at the ankle joints when
controlled by the hybrid system.
Exoskeleton Lower Extremity Gait System (eLEGS) Only one clinical trial
(Strausser and Kazerooni 2011) was carried out on this system.
This study involved five participants with no leg motion due to SCI or ataxia,
recruited in the USA. The authors tested whether the eLEGS was intuitive and easy
to learn and use. The measures used were knee angle and time practicing. The results
26
Table 2 Summary of the measures adopted in the studies reviewed ordered by largest to smallest frequency in use
Measure Study where the measure was Measures’
Measure Abbreviation Frequency type administered references
Timed up and go test TUG 10 Outcome Aach et al. (2013, 2014), Bishop et al. Podsiadlo and
measures (2012), Bortole et al. (2015), Farris et al. Richardson (1991)
(2014), Nilsson et al. (2014), Quintero
et al. (2012), Stein et al. (2014), Watanabe
et al. (2014), Zeilig et al. (2012)
10-meter walk test 10 MWT 9 Outcome Aach et al. (2013, 2014), Bishop et al. Peters et al. (2014)
measures (2012), Esquenazi et al. (2012), Farris
et al. (2014), Hartigan et al. (2015),
Nilsson et al. (2014), Stein et al. (2014),
Zeilig et al. (2012)
6-minute walking test 6 MWT 8 Outcome Aach et al. (2014), Bishop et al. (2012), Reybrouck (2003)
measures Bortole et al. (2015), Esquenazi et al.
(2012), Farris et al. (2014), Hartigan et al.
(2015), Stein et al. (2014), Watanabe et al.
(2014)
Knee joint angle or torque / 7 Observation Bortole et al. (2015), Farris et al. (2011, /
measures 2012), Kawamoto et al. (2010), Strausser
and Kazerooni (2011), Tsukahara et al.
(2009, 2010)
Berg balance scale BBS 6 Outcome Bishop et al. (2012), Bortole et al. (2015), Downs et al.
measures Nilsson et al. (2014), Stein et al. (2014) (2013)
Li et al. (2015), Raab et al. (2016)
Walking speed / 6 Observation Aach et al. (2013), Farris et al. (2011), /
measures Nilsson et al. (2014), Spungen et al.
S. Federici et al.
(2013), Talaty et al. (2013), Watanabe
et al. (2014)
Distance walked in a certain time period / 4 Observation Aach et al. (2013), Esquenazi et al. /
measures (2012), Kolakowsky-Hayner et al. (2013),
Zeilig et al. (2012)
Electromyography EMG 4 Outcome Li et al. (2015), Sczesny-Kaiser et al. Mohseni Bandpei
measures (2013), Sylos-Labini et al. (2014), Talaty et al. (2014)
et al. (2013)
Fugl-Meyer assessment for lower FM-LE 4 Outcome Bortole et al. (2015), Li et al. (2015), Park and Choi
extremity measures Nilsson et al. (2014), Watanabe et al. (2014)
(2014)
Video recording of subject’s walking / 4 Observation Ikehara et al. (2011), Li et al. (2015), /
measures Strausser et al. (2010), Talaty et al. (2013)

Barthel index BI 2 Outcome Bortole et al. (2015), Nilsson et al. (2014) Cuesta-Vargas and
measures Perez-Cruzado
(2014)
Functional ambulation category FAC 2 Outcome Nilsson et al. (2014), Watanabe et al. Mehrholz et al.
measures (2014) (2007)
Isometric muscle strength (hip/knee/ / 2 Clinical Talaty et al. (2013), Watanabe et al. (2014) /
trunk flexion and extension, ankle measures
dorsi /plantar flexion)
Time practicing / 2 Observation Kolakowsky-Hayner et al. (2013), /
measures Strausser and Kazerooni (2011)
Walking index for SCI II WISCI II 2 Outcome Aach et al. (2013, 2014) Ditunno et al.
measures (2013)
Ashworth scale / 1 Outcome Aach et al. (2014) Pandyan et al.
measures (1999)
Assistive devices used during / 1 Observation Kolakowsky-Hayner et al. (2013) /
ambulation measures
Blood pressure / 1 Kolakowsky-Hayner et al. (2013) /
27
(continued)
28
Table 2 (continued)
Measure Study where the measure was Measures’
Measure Abbreviation Frequency type administered references
Observation
measures
California functional evaluation 40 CAFE 40 1 Outcome Stein et al. (2014) Fung et al. (1997)
measures
Clinical outcome variable scale, Swedish S-COVS 1 Outcome Nilsson et al. (2014) Andersson and
version measures Franzen (2015)
Conversation with the user during / 1 Observation Neuhaus et al. (2011) /
walking measures
Dynamic gait index DGI 1 Outcome Raab et al. (2016) Tinetti (1986)
measures
Emory functional ambulation profile EFAP 1 Outcome Stein et al. (2014) Wolf et al. (1999)
measures
Falls efficacy scale, Swedish version FES(S) 1 Outcome Nilsson et al. (2014) Hellstrom et al.
measures (2002)
Five times sit-to-stand test 5XSST 1 Outcome Stein et al. (2014) Whitney et al.
measures (2005)
Functional independence measure FIM 1 Outcome Nilsson et al. (2014) Saji et al. (2015)
measures
Functional magnetic resonance imaging fMRI 1 Outcome Sczesny-Kaiser et al. (2013) Buxton (2013)
measures
Heart rate HR 1 Observation Asselin et al. (2015) /
measures
Kinematic magnitudes and exchanged / 1 Observation Belforte et al. (2001) /
forces measures
Lower extremity circumference / 1 Clinical Aach et al. (2014) /
measures
S. Federici et al.
Lower extremity motor score LEMS 1 Outcome Aach et al. (2014) Shin et al. (2011)
measures
National Institutes of Health Stroke scale NIHSS 1 Outcome Nilsson et al. (2014) Yang et al. (2014)
measures
Oxygen uptake VO2 1 Observation Asselin et al. (2015) /
measures
Pain level / 1 Observation Kolakowsky-Hayner et al. (2013) /
measures
Romberg’s test / 1 Outcome Stein et al. (2014) Agrawal et al.
measures (2011)
Rotations and applied torques for each / 1 Observation Belforte et al. (2001) /
joint measures
Short form health survey SF-36 1 Outcome Raab et al. (2016) McHorney et al.
measures (1993)
Short physical performance battery SPPB 1 Outcome Watanabe et al. (2014) Stookey et al.
measures (2014)
Skin evaluation / 1 Observation Kolakowsky-Hayner et al. (2013) /
measures
Spasticity / 1 Observation Kolakowsky-Hayner et al. (2013) /
measures
Step length / 1 Observation Kolakowsky-Hayner et al. (2013) /
measures
Time and level of assistance needed to / 1 Observation Kolakowsky-Hayner et al. (2013) /
transfer into and on device measures
Time response of the angles and electric / 1 Observation Mori et al. (2006) /
currents of each joint measures
Visual analogue scale VAS 1 Outcome Nilsson et al. (2014) Reed and Van
measures Nostran (2014)
29
showed that the eLEGS’ human machine interface was easy to learn; all five subjects
were able to quickly learn how to use it. Indeed, the walking performance of the five
participants displayed an increased time between the heel off and the step, as
compared to the able-bodied user. However, this time was decreased in the experi-
enced user, while the user with no experience with the device had an average time of
0.859 sec, the experienced user was able to reduce the lag time to 0.590 s.
X1 Robotic Exoskeleton (MINA) The team of the Institute for Human and Machine
Cognition led by P. Neuhaus and the NASA Johnson Space Center jointly developed
MINA (Neuhaus et al. 2011).
The team carried out a clinical trial involving two participants with motor
complete SCIs at the T10 level, recruited in the USA. The recruitment inclusion
criteria stipulated an American Spinal Injury Association Impairment Scale of either
an A (complete) or B (incomplete) and a Walking Index for Spinal Cord Injury
(WISCI) of level 9 (ambulates with a walker, with braces, and no physical
assistance, 10 m). The paper presents a clinical and rehabilitative evaluation of the
MINA exoskeleton. The measures used qualitatively evaluated the cognitive effort
required to use MINA: researchers could talk with the subjects while they walked
with MINA. In addition, static standing balance stability was tested by having the
subjects catch and throw a ball while standing on both legs and using one crutch for
balance. Findings provided evidence that MINA currently facilitates paraplegics’
walking mobility at speeds of up to 0.20 m/sec. In addition, MINA is not physically
taxing and requires little cognitive effort, allowing the user to converse and maintain
eye contact while walking.
ATLAS The Center for Automation and Robotics in Spain developed the prototype
of the ATLAS exoskeleton to help a quadriplegic child to walk. Its development and
main features were tested by D. Sanz-Merodio and colleagues (Sanz-Merodio et al.
2012).
This clinical trial involved one participant, a Spanish girl aged 8 years, affected by
quadriplegia. Experiments validated a good controlled performance in following the
gait pattern given by the parameterized trajectory generator.
ABLE The ABLE system was designed and tested in Japan by Y. Mori and
collaborators at Ibaraki University, in the Department of Intelligent Systems Engi-
neering (Mori et al. 2006).
This clinical trial involved one participant with motor paralysis, recruited in
Japan. Mori and his team developed a standing style transfer system for a person
with disabled legs. They proposed a new motion technique and compared it to their
previous system (Mori et al. 2004). The measures used were the time response of the
angles of the joints and electric currents, corresponding with the torque of each joint.
The subject succeeded in standing up; a large arm force was needed in the beginning,
but was not needed afterward.
Tibion Bionic Technologies In 2013, Tibion Bionic Technologies was acquired by

AlterG that now produces the exoskeleton tested by L. Bishop of Columbia Univer-
sity in 2012 (Bishop et al. 2012). This clinical trial involved one participant with a
motor incomplete SCI at the C5-C6 level, enrolled in USA. Study outcomes
suggested that the use of this device, during a physical therapy program for an
individual with incomplete SCI, is practical and useful when used in addition to the
standard training.
Bionic Leg The Bionic Leg, produced by the Californian AlterG, is a powered knee
orthosis for patients with unilateral neurologic or orthopedic conditions, tested by
the team of the Department of Rehabilitation and Regenerative Medicine, Columbia
University College of Physicians and Surgeons, led by J. Stein (Li et al. 2015; Stein
et al. 2014).
These clinical trials involved hemiparesis patients after a CVA. Subjects were
required to be independent in household ambulation (with or without the use of
unilateral assistive devices and with or without the use of ankle–foot orthoses). The
study of Stein et al. (Li et al. 2015, Stein et al. 2014) was designed to test how the
Bionic leg restores mobility in stroke survivors in their living environments, while Li
et al. (2015) aimed to demonstrate the training effects of a 3-week robotic leg
orthosis and to investigate possible mechanisms of the sensorimotor alterations
and improvements by using gait analysis and EMG.
Outcomes suggested that robotic therapy for ambulatory stroke patients with
chronic hemiparesis and using a robotic knee brace resulted in only modest func-
tional benefits, in comparison with a group receiving only exercise intervention, for
example, without using the Bionic Leg (Stein et al. 2014). The 3-week training
period, using the wearable orthoses, improved the participants’ gait performance and
improved muscle activation and walking speed (Li et al. 2015).
Ekso One study tested the Ekso exoskeleton (Kolakowsky-Hayner et al. 2013). The
study involved participants with a motor complete SCI. The main inclusion criteria
for participants stipulated a body size compatible with the Ekso suit.
The main goal of the study was to evaluate the feasibility and safety of using Ekso
to aid ambulation in a group of individuals with SCI who had completed their initial
SCI rehabilitation. Secondarily, training effects with the progressive use of Ekso
were evaluated in terms of time tolerated, distance traveled, and assistance needed.
Outcomes suggest that Ekso is safe for those with a complete thoracic SCI in a
controlled environment, in the presence of experts. Ekso may eventually enhance
mobility in those without volitional lower extremity function. There appears to be a
training effect in the device.
Indego The Indego exoskeleton was tested by the team of Virginia C. Crawford
Research Institute, Shepherd Center, Atlanta, Georgia (Hartigan et al. 2015).
This study was conducted to evaluate mobility outcomes for 16 SCI subjects with
injury levels (ranging from C5 complete to L1 incomplete) after five gait training
sessions with a powered exoskeleton. The primary goal was to characterize the ease
of learning and usability of the system. Outcome measures of the study included the
10-meter walk test (10 MWT) and the 6-minute walk test (6 MWT). Results
highlighted that the average walking speed was 0.22 m/sec for persons with C5–6
motor complete tetraplegia, 0.26 m/sec for T1–T8 motor complete paraplegia, and
0.45 m/sec for T9–L1 paraplegia. Distances covered in 6 min averaged 64 m for
those with C5–C6, 74 m for T1–T8, and 121 m for T9–L1. Tetraplegic and
paraplegic patients learned to use the Indego exoskeleton quickly and could manage
a variety of surfaces. The walking speeds and distances achieved also indicated that
some individuals with paraplegia could quickly become limited community
ambulators using this system.
H2 Robotic Exoskeleton The Exo-H2 has been the result of many years of research
in the Grupo de Bioingeniería of the Spanish National Research Council (CSIC),
who has conceded an exclusive license to Technaid S.L. for the design, manufactur-
ing, and commercial exploitation of the system.
Bortole et al. (2015) evaluated the safety and usability of the H2 robotic exoskel-
eton for gait rehabilitation in three hemiparetic stroke patients across 4 weeks of
training per individual (approximately 12 sessions). Results showed that the training
was well tolerated and that H2 appears to be safe and easy to use. The system is
robust and safe when applied to assist a stroke patient performing an over ground
walking task.
Conclusions
In this chapter, we have reviewed the clinical effectiveness in rehabilitation of

various types of active, powered, and wearable lower limb exoskeletons used to
facilitate and rehabilitate paraplegic patients’ gait disorders resulting from serious
CNS lesions due to, for example, SCIs or CVAs.
The literature review revealed that the exoskeletons subjected to the highest
number of studies were the HAL (Aach et al. 2013, 2014; Kawamoto et al. 2010;
Nilsson et al. 2014; Sczesny-Kaiser et al. 2013; Tsukahara et al. 2009, 2010;
Watanabe et al. 2014), the ReWalk (Asselin et al. 2015; Benson et al. 2016;
Esquenazi et al. 2012; Raab et al. 2016; Spungen et al. 2013; Talaty et al. 2013;
Zeilig et al. 2012), and the Vanderbilt lower limb exoskeleton (Farris et al. 2011,
2012, 2014; Quintero et al. 2012).
By qualitatively analyzing the results for each type of exoskeleton, it was found
that the rehabilitative use of an exoskeleton to restore gait disorders is safe and
practical (Bishop et al. 2012; Bortole et al. 2015; Kolakowsky-Hayner et al. 2013;
Nilsson et al. 2014), is not physically exhausting, and requires only a little cognitive
(Neuhaus et al. 2011) or energetic (Asselin et al. 2015) effort. In addition, it is easy to
learn (Strausser and Kazerooni 2011), can increase mobility and functional abilities,
and decreases the risk of secondary injuries (Aach et al. 2013, 2014; Asselin et al.
2015; Hartigan et al. 2015; Kolakowsky-Hayner et al. 2013; Raab et al. 2016;
Strausser et al. 2010; Zeilig et al. 2012), as well as allowing restoration of a gait
pattern comparable to normal over ground walking (Esquenazi et al. 2012; Farris
et al. 2011; Spungen et al. 2013; Sylos-Labini et al. 2014). Among these positive
attributes, Benson et al. (2016) stressed a negative one, claiming that the use of
exoskeleton is characterized by the presence of a high rate of skin aberrations
In addition to the advantages pointed out by the literature review, we found that
the exoskeleton can be considered as an ecological device, replacing wheelchairs for
many hours at a time; it enables patients who cannot walk to regain a degree of
walking mobility and to retard the onset of a wide range of secondary disabilities
associated with the long-term use of wheelchairs. The exoskeleton can improve the
autonomy of the patient, who is enabled to walk independently, simply by wearing
it. No other rehabilitative or therapeutic techniques and technologies provide such an
extraordinary potential for autonomy.
Nevertheless, there are still some limitations in the rehabilitative use of an
exoskeleton. First, the wearability criteria are too restrictive; its use is limited to
people with specific values of height and weight (Esquenazi et al. 2012; Nilsson et al.
2014; Spungen et al. 2013; Sylos-Labini et al. 2014). Second, it requires very
complex and specialized training to use the exoskeleton autonomously at home.
Third, it is still an extremely expensive device, hardly covered by private or public
healthcare systems. For instance, the National Health Services in Europe generally
support the use of the exoskeleton for a rehabilitation program in specialized medical
centers, but never for a private individual’s use at home or in the workplace. A final
limitation is the scarcity of experimental designs based on evidence that demon-
strates the effectiveness of the exoskeleton compared to other rehabilitative tech-
niques and technologies. Only two studies that adopted a randomized clinical trial
design compared exoskeleton use to conventional gait training (Stein et al. 2014;
Watanabe et al. 2014); furthermore, the results of these two studies are contradictory.
Finally, user experience with an exoskeleton in daily life activities generally did not
meet subjects’ expectations in terms of perceived benefits and impact on quality of
life (Benson et al. 2016).
Cross-References
▶ Brain Computer Interface Assisted Gait

References
Aach M, Meindl R, Hayashi T, Lange I, Geßmann J, Sander A, Nicolas V, Schwenkreis P,
Tegenthoff M, Sankai Y, Schildhauer TA (2013) Exoskeletal neuro-rehabilitation in chronic
paraplegic patients – initial results. In: Pons JL, Torricelli D, Pajaro M (eds) Converging clinical
and engineering research on neurorehabilitation. Springer, Berlin, pp 233–236. doi:10.1007/
978-3-642-34546-3_99
Aach M, Cruciger O, Sczesny-Kaiser M, Hoffken O, Meindl RC, Tegenthoff M, Schwenkreis P,
Sankai Y, Schildhauer TA (2014) Voluntary driven exoskeleton as a new tool for rehabilitation
in chronic spinal cord injury: a pilot study. Spine J 14(12):2847–2853. doi:10.1016/J.
Spinee.2014.03.042
Agrawal Y, Carey JP, Hoffman HJ, Sklare DA, Schubert MC (2011) The modified Romberg balance
test: normative data in US adults. Otol Neurotol 32(8):1309–1311. doi:10.1097/
MAO.0b013e31822e5bee
Andersson P, Franzen E (2015) Effects of weight-shift training on walking ability, ambulation, and
weight distribution in individuals with chronic stroke: a pilot study. Top Stroke Rehabil [Epub
ahead of print]. doi:10.1179/1074935715Z.00000000052
Asselin P, Knezevic S, Kornfeld S, Cirnigliaro C, Agranova-Breyter I, Bauman WA, Spungen AM
(2015) Heart rate and oxygen demand of powered exoskeleton-assisted walking in persons with
paraplegia. J Rehabil Res Dev 52(2):147–158. doi:10.1682/JRRD.2014.02.0060
Belforte G, Gastaldi L, Sorli M (2001) Pneumatic active gait orthosis. Mechatronics 11(3):301–323.
doi:10.1016/S0957-4158(00)00017-9
Benson I, Hart K, Tussler D, van Middendorp JJ (2016) Lower-limb exoskeletons for individuals
with chronic spinal cord injury: findings from a feasibility study. Clin Rehabil 30(1):73–84.
doi:10.1177/0269215515575166
Bishop L, Stein J, Wong CK (2012) Robot-aided gait training in an individual with chronic spinal
cord injury: a case study. J Neurol Phys Ther 36(3):138–143. doi:10.1097/
NPT.0b013e3182624c87
Bortole M, Venkatakrishnan A, Zhu F, Moreno JC, Francisco GE, Pons JL, Contreras-Vidal JL
(2015) The H2 robotic exoskeleton for gait rehabilitation after stroke: early findings from a
clinical study. J Neuroeng Rehabil 12:54. doi:10.1186/s12984-015-0048-y
Buxton RB (2013) The physics of functional magnetic resonance imaging (fMRI). Rep Prog Phys
76(9):096601. doi:10.1088/0034-4885/76/9/096601
Chaigneau D, Arsicault M, Gazeau JP, Zeghloul S (2008) LMS robotic hand grasp and manipula-
tion planning (an isomorphic exoskeleton approach). Robotica 26(2):177–188. doi:10.1017/
S0263574707003736
Cuesta-Vargas AI, Perez-Cruzado D (2014) Relationship between Barthel index with physical tests
in adults with intellectual disabilities. SpringerPlus 3(543). doi:10.1186/2193-1801-3-543
Dickstein R, Levy S, Shefi S, Holtzman S, Peleg S, Vatine J-J (2014) Motor imagery group practice
for gait rehabilitation in individuals with post-stroke hemiparesis: a pilot study. NeuroReh-
abilitation 34(2):267–276. doi:10.3233/NRE-131035
Ditunno JFJ, Ditunno PL, Scivoletto G, Patrick M, Dijkers M, Barbeau H, Burns AS, Marino RJ,
Schmidt-Read M (2013) The walking index for spinal cord injury (WISCI/WISCI II): nature,
metric properties, use and misuse. Spinal Cord 51(5):346–355. doi:10.1038/sc.2013.9
Downs S, Marquez J, Chiarelli P (2013) The berg balance scale has high intra- and inter-rater
reliability but absolute reliability varies across the scale: a systematic review. J Physiother 59
(2):93–99. doi:10.1016/S1836-9553(13)70161-9
Esquenazi A, Talaty M, Packel A, Saulino M (2012) The ReWalk powered exoskeleton to restore
ambulatory function to individuals with thoracic-level motor-complete spinal cord injury. Am J
Phys Med Rehabil 91(11):911–921. doi:10.1097/PHM.0b013e318269d9a3
Farris RJ, Quintero HA, Goldfarb M (2011) Preliminary evaluation of a powered lower limb
orthosis to aid walking in paraplegic individuals. IEEE Trans Neural Syst Rehabil Eng 19
(6):652–659. doi:10.1109/TNSRE.2011.2163083
Farris RJ, Quintero HA, Goldfarb M (2012) Performance evaluation of a lower limb exoskeleton for
stair ascent and descent with paraplegia. In: 34th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society: EMBC 2012, San Diego, 28 Aug–1 Sep 2012. pp
1908–1911. doi:10.1109/EMBC.2012.6346326
Farris RJ, Quintero HA, Murray SA, Ha KH, Hartigan C, Goldfarb M (2014) A preliminary
assessment of legged mobility provided by a lower limb exoskeleton for persons with paraple-
gia. IEEE Trans Neural Syst Rehabil Eng 22(3):482–490. doi:10.1109/TNSRE.2013.2268320
Fondazione Santa Lucia (2015) Maratona di roma 2015, un esoscheletro hi-tech per tornare a
correre. http://www.hsantalucia.it/modules.php?name=News&file=article&sid=989.
Accessed 15 May 2015
Fung S, Byl N, Melnick M, Callahan P, Selinger A, Ishii K, Devins J, Fischer P, Torburn L, Andrade
C-K (1997) Functional outcomes: the development of a new instrument to monitor the effec-
tiveness of physical therapy. Eur J Phys Rehab Med 7(2):31–41
Hartigan C, Kandilakis C, Dalley S, Clausen M, Wilson E, Morrison S, Etheridge S, Farris R (2015)
Mobility outcomes following five training sessions with a powered exoskeleton. Top Spinal
Cord Inj Rehabil 21(2):93–99. doi:10.1310/sci2102-93
Hellstrom K, Lindmark B, Fugl-Meyer A (2002) The falls-efficacy scale, Swedish version: does it
reflect clinically meaningful changes after stroke? Disabil Rehabil 24(9):471–481. doi:10.1080/
09638280110105259
Herr H (2009) Exoskeletons and orthoses: classification, design challenges and future directions.
J Neuroeng Rehabil 6(1):1–9. doi:10.1186/1743-0003-6-21
Ikehara T, Nagamura K, Ushida T, Tanaka E, Saegusa S, Kojima S, Yuge L (2011) Development of
closed-fitting-type walking assistance device for legs and evaluation of muscle activity. In: IEEE
International Conference on Rehabilitation Robotics: ICORR 2011, Zurich, 29 Jun–1 Jul 2011.
pp 1–7. doi:10.1109/ICORR.2011.5975449
Kao P-C, Lewis CL, Ferris DP (2010) Invariant ankle moment patterns when walking with and
without a robotic ankle exoskeleton. J Biomech 43(2):203–209. doi:10.1016/j.
Kawamoto H, Taal S, Niniss H, Hayashi T, Kamibayashi K, Eguchi K, Sankai Y (2010) Voluntary
motion support control of robot suit HAL triggered by bioelectrical signal for hemiplegia. In:
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology
Society: EMBC 2010, Buenos Aires, 31 Aug–4 Sep 2010. pp 462–466. doi:10.1109/
IEMBS.2010.5626191
Kolakowsky-Hayner SA, Crew J, Moran S, Shah A (2013) Safety and feasibility of using the
EksoTM bionic exoskeleton to aid ambulation after spinal cord injury. J Spine S4(3):1–8.
doi:10.4172/2165-7939.S4-003
Lee HS, Song J, Min K, Choi Y-S, Kim S-M, Cho S-R, Kim M (2014) Short-term effects of
erythropoietin on neurodevelopment in infants with cerebral palsy: a pilot study. Brain Dev 36
(9):764–769. doi:10.1016/j.braindev.2013.11.002
Li L, Ding L, Chen N, Mao Y, Huang D, Li L (2015) Improved walking ability with wearable robot-
assisted training in patients suffering chronic stroke. Biomed Mater Eng 26:S329–S340.
doi:10.3233/bme-151320
McHorney CA, Ware JE, Raczek AE (1993) The MOS 36-item short-form health survey (SF-36):
II. Psychometric and clinical tests of validity in measuring physical and mental health con-
structs. Med Care 31(3):247–263
Mehrholz J, Wagner K, Rutte K, Meissner D, Pohl M (2007) Predictive validity and responsiveness
of the functional ambulation category in hemiparetic patients after stroke. Arch Phys Med
Rehabil 88(10):1314–1319. doi:10.1016/j.apmr.2007.06.764
Mohseni Bandpei MA, Rahmani N, Majdoleslam B, Abdollahi I, Ali SS, Ahmad A (2014)
Reliability of surface electromyography in the assessment of paraspinal muscle fatigue: an
updated systematic review. J Manipulative Physiol Ther 37(7):510–521. doi:10.1016/j.
jmpt.2014.05.006
Mooney LM, Rouse EJ, Herr HM (2014) Autonomous exoskeleton reduces metabolic cost of
human walking. J Neuroeng Rehabil 11(151):2–5. doi:10.1186/1743-0003-11-151
Moreno J, Ama A, Reyes-Guzmán A, Gil-Agudo Á, Ceres R, Pons J (2011) Neurorobotic and
hybrid management of lower limb motor disorders: a review. Med Biol Eng Comput 49
(10):1119–1130. doi:10.1007/s11517-011-0821-4
Mori Y, Takayama K, Zengo T, Nakamura T (2004) Development of straight style transfer
equipment for lower limbs disabled “able”. J Robot Mechatron 16(5):456–463
Mori Y, Okada J, Takayama K (2006) Development of a standing style transfer system “able” for
disabled lower limbs. IEEE/ASME Trans Mechatronics 11(4):372–380. doi:10.1109/
TMECH.2006.878558
Nef T, Riener R (2012) Three-dimensional multi-degree-of-freedom arm therapy robot (ARMin).
In: Dietz V, Nef T, Rymer WZ (eds) Neurorehabilitation technology. Springer, London, pp
141–157
Neuhaus PD, Noorden JH, Craig TJ, Torres T, Kirschbaum J, Pratt JE (2011) Design and evaluation
of mina: a robotic orthosis for paraplegics. In: IEEE International Conference on Rehabilitation
Robotics: ICORR 2011, Zurich, 29 Jun–1 Jul 2011. pp 1–8. doi:10.1109/ICORR.2011.5975468
Nilsson A, Vreede KS, Haglund V, Kawamoto H, Sankai Y, Borg J (2014) Gait training early after
stroke with a new exoskeleton – the hybrid assistive limb: a study of safety and feasibility.
J Neuroeng Rehabil 11(92):1–10. doi:10.1186/1743-0003-11-92
Pandyan AD, Johnson GR, Price CI, Curless RH, Barnes MP, Rodgers H (1999) A review of the
properties and limitations of the Ashworth and modified Ashworth scales as measures of
spasticity. Clin Rehabil 13(5):373–383
Park EY, Choi YI (2014) Psychometric properties of the lower extremity subscale of the Fugl-Myer
assessment for community-dwelling hemiplegic stroke patients. J Phys Ther Sci 26
(11):1775–1777. doi:10.1589/jpts.26.1775
Peters DM, Middleton A, Donley JW, Blanck EL, Fritz SL (2014) Concurrent validity of walking
speed values calculated via the GAITRite electronic walkway and 3 meter walk test in the
chronic stroke population. Physiother Theory Pract 30(3):183–188. doi:10.3109/
09593985.2013.845805
elderly persons. J Am Geriatr Soc 39(2):142–148
Quintero HA, Farris RJ, Goldfarb M (2012) A method for the autonomous control of lower limb
exoskeletons for persons with paraplegia. J Med Devices 6(4):1–6. doi:10.1115/1.4007181
Raab K, Krakow K, Tripp F, Jung M (2016) Effects of training with the ReWalk exoskeleton on
quality of life in incomplete spinal cord injury: a single case study. Spinal Cord Ser Cases
1(1):15025. doi:10.1038/scsandc.2015.25
Rahman T, Sample W, Jayakumar S, King MM, Wee JY, Seliktar R, Alexander M, Scavina M, Clark
A (2006) Passive exoskeletons for assisting limb movement. J Rehabil Res Dev 43(5):583–590.
doi:10.1682/JRRD.2005.04.0070
Reed MD, Van Nostran W (2014) Assessing pain intensity with the visual analog scale: a plea for
uniformity. J Clin Pharmacol 54(3):241–144. doi:10.1002/jcph.250
Reybrouck T (2003) Clinical usefulness and limitations of the 6-minute walk test in patients with
cardiovascular or pulmonary disease. Chest 123(2):325–327. doi:10.1378/chest.123.2.325
Saji N, Kimura K, Ohsaka G, Higashi Y, Teramoto Y, Usui M, Kita Y (2015) Functional indepen-
dence measure scores predict level of long-term care required by patients after stroke: a
multicenter retrospective cohort study. Disabil Rehabil 37(4):331–337. doi:10.3109/
09638288.2014.918195
Sanz-Merodio D, Cestari M, Arevalo JC, Garcia E (2012) A lower-limb exoskeleton for gait
assistance in quadriplegia. In: IEEE International Conference on Robotics and Biomimetics:
ROBIO 2012, Guangzhou, 11–14 Dec 2012. pp 122–127. doi:10.1109/ROBIO.2012.6490954
Sczesny-Kaiser M, Höffken O, Lissek S, Lenz M, Schlaffke L, Nicolas V, Meindl R, Aach M,
Sankai Y, Schildhauer TA, Tegenthoff M, Schwenkreis P (2013) Neurorehabilitation in chronic
paraplegic patients with the HAL® exoskeleton – preliminary electrophysiological and fMRI
data of a pilot study. In: Pons JL, Torricelli D, Pajaro M (eds) Converging clinical and
engineering research on neurorehabilitation. Springer, Berlin, pp 611–615. doi:10.1007/978-3-
642-34546-3_99
Shin JC, Yoo JH, Jung TH, Goo HR (2011) Comparison of lower extremity motor score parameters
for patients with motor incomplete spinal cord injury using gait parameters. Spinal Cord 49
(4):529–533. doi:10.1038/sc.2010.158
Spungen AM, Asselin P, Fineberg DB, Kornfeld SD, Harel NY (2013) Exoskeletal-assisted walking
for persons with motor-complete paraplegia. In: STO Human Factors and Medicine Panel
(HFM) Symposium, Milan, 15–17 Apr 2013
Stein J, Bishop L, Stein DJ, Wong CK (2014) Gait training with a robotic leg brace after stroke: a
randomized controlled pilot study. Am J Phys Med Rehabil 93(11):987–994. doi:10.1097/
PHM.0000000000000119
Stookey AD, Katzel LI, Steinbrenner G, Shaughnessy M, Ivey FM (2014) The short physical
performance battery as a predictor of functional capacity after stroke. J Stroke Cerebrovasc Dis
23(1):130–135. doi:10.1016/j.jstrokecerebrovasdis.2012.11.003
Strausser KA, Kazerooni H (2011) The development and testing of a human machine interface for a
mobile medical exoskeleton. In: IEEE/RSJ International Conference on Intelligent Robots and
Systems: IROS 2011, San Francisco, 25–30 Sep 2011. pp 4911–4916. doi:10.1109/
IROS.2011.6095025
Strausser KA, Swift TA, Zoss AB, Kazerooni H (2010) Prototype medical exoskeleton for para-
plegic mobility: first experimental results. In: ASME 2010 Dynamic Systems and Control
Conference: DSCC 2010, Cambridge, MA, 12–15 Sep 2010. ASME, pp 453–458.
doi:10.1115/DSCC2010-4261
Suzuki K, Kawamura Y, Hayashi T, Sakurai T, Hasegawa Y, Sankai Y (2005) Intention-based
walking support for paraplegia patient. In: IEEE International Conference on Systems, Man and
Cybernetics: SMC 2005, Waikoloa, 10–12 Oct 2005. pp 2707–2713 Vol. 2703. doi:10.1109/
ICSMC.2005.1571559
Sylos-Labini F, La Scaleia V, d’Avella A, Pisotta I, Tamburella F, Scivoletto G, Molinari M,
Wang S, Wang L, van Asseldonk E, van der Kooij H, Hoellinger T, Cheron G,
Thorsteinsson F, Ilzkovitz M, Gancet J, Hauffe R, Zanov F, Lacquaniti F, Ivanenko YP
(2014) EMG patterns during assisted walking in the exoskeleton. Front Hum Neurosci 8
(423):1–12. doi:10.3389/fnhum.2014.00423
Talaty M, Esquenazi A, Briceño JE (2013) Differentiating ability in users of the ReWalk™ powered
exoskeleton: an analysis of walking kinematics. In: IEEE International Conference on Rehabil-
itation Robotics: ICORR 2013 Seattle, 24–26 Jun 2013. pp 1–5. doi:10.1109/
ICORR.2013.6650469
Geriatr Soc 34(2):119–126. doi:10.1111/j.1532-5415.1986.tb05480.x
Tsukahara A, Hasegawa Y, Sankai Y (2009) Standing-up motion support for paraplegic patient with
robot suit hal. In: IEEE International Conference on Rehabilitation Robotics: ICORR 2009,
Kyoto, 23–26 Jun 2009. pp 211–217. doi:10.1109/ICORR.2009.5209567
Tsukahara A, Kawanishi R, Hasegawa Y, Sankai Y (2010) Sit-to-stand and stand-to-sit transfer
support for complete paraplegic patients with robot suit HAL. Adv Robotics 24(11):1615–1638.
doi:10.1163/016918610X512622
Watanabe H, Tanaka N, Inuta T, Saitou H, Yanagi H (2014) Locomotion improvement using a
hybrid assistive limb in recovery phase stroke patients: a randomized controlled pilot study.
Arch Phys Med Rehabil 95(11):2006–2012. doi:10.1016/J.Apmr.2014.07.002
Whitney SL, Wrisley DM, Marchetti GF, Gee MA, Redfern MS, Furman JM (2005) Clinical
measurement of sit-to-stand performance in people with balance disorders: validity of data for
the five-times-sit-to-stand test. Phys Ther 85(10):1034–1045
Wolf SL, Catlin PA, Gage K, Gurucharri K, Robertson R, Stephen K (1999) Establishing the
reliability and validity of measurements of walking time using the Emory functional ambulation
profile. Phys Ther 79(12):1122–1133
Yang N, Zhang B, Gao C (2014) The baseline NIHSS score in female and male patients and short-
time outcome: a study in young ischemic stroke. J Thromb Thrombolysis 37(4):565–570.
doi:10.1007/s11239-013-0986-9
Zeilig G, Weingarden H, Zwecker M, Dudkiewicz I, Bloch A, Esquenazi A (2012) Safety and
tolerance of the ReWalk™ exoskeleton suit for ambulation by people with complete spinal cord
injury: a pilot study. J Spinal Cord Med 35(2):96–101. doi:10.1179/2045772312Y.0000000003
Effects of Total Hip Arthroplasty on Gait
Swati Chopra and Kenton R. Kaufman
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
The State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Gait Assessment Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Hip Joint Gait Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Pathologic Versus Instrumented Hip Gait Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Abstract
The hip joint transfers upper body load to the lower limbs and plays an important
role in maintaining balance and stability during locomotion. Muscles around the
hip joint are also the major postural muscles in our body. A diseased hip joint
experiences diminished function affecting the activity of daily living gravely.
Furthermore, being a part of the closed kinetic chain, as the disease progresses,
problems originating at the hip joint start showing effects on the neighboring
joints including the pelvis, knee, ankle, and spine. The gold standard treatment for
end-stage hip disease is total hip arthroplasty (THA). From a clinical perspective,
total hip replacement is considered to be one of the most successful orthopedic
surgeries in terms of patient outcome, based on self-reported questionnaires for
health status. However, gait, being the major function of the hip joint, tends not to
be assessed objectively in the outcome analysis of THA surgery. In research
settings, THA has consistently shown a positive outcome with improved gait
parameters compared to preoperative analysis. Note that this recovery is still not
reported to be 100% in comparing to healthy controls and, as a result, objective
S. Chopra (*) • K.R. Kaufman (*)

Motion Analysis Laboratory, Mayo Clinic, Rochester, MN, USA
e-mail: Chopra.swati@mayo.edu; Kaufman.kenton@mayo.edu

DOI 10.1007/978-3-319-30808-1_81-1
2 S. Chopra and K.R. Kaufman
assessment data during the early postoperative period can also be used to help
design tailor-made rehabilitation protocols for individual patients, optimizing
outcome results. This chapter details gait alteration in hip arthritis and improve-
ment in gait following THA.
Keywords
Gait • Kinematics • Kinetics • Total hip arthroplasty • Electromyography
Introduction
Humans are known to have started bipedal locomotion prior to the complex brain
being developed. The human musculoskeletal system has evolved both anatomically
and biomechanically to provide optimal stability during locomotion. The hip joint is
one of the most heavily loaded joints taking the peak average joint contact force of
about 2.5 times of the body weight at the walking speed of 1.5 m/s (van den Bogert
et al. 1999). It is therefore considered the most crucial joint in terms of balance and
stability, particularly when in an upright, dynamic, weight-bearing position. During
weight bearing, the hip joint transfers the full upper body load to the lower extrem-
ities and as a consequence is surrounded by major muscle groups providing the
required support. The hip joint is comprised of the head of the femur and the
acetabulum of the pelvis. It is a ball- and socket-type joint, which, providing the
additional benefit of mobility in several planes, increases adaptive stability. In
knowing this, it can be assumed that a diseased and/or painful hip joint could
significantly restrict locomotion, thereby affecting day-to-day activities of the indi-
vidual with arthritis. The most common and successful surgical treatment for
end-stage hip joint disease is THA. In fact, based on the recent statistics, THA is
performed annually on approximately one million patients worldwide (OECD
2015), with the USA itself accounting for about 330,000 cases per year (Centers
For Disease Control and Prevention 2010). In THA, the diseased degenerated joint is
completely replaced by artificial implants, which are said to perform similarly in
terms of biomechanical function to a healthy joint. However, this claim is backed by
empirical data with little information related to patient gait mechanics post-THA.
One of the main reasons for this lack of objective data is that current clinical practice
utilizes time and cost-effective, subjective assessment methods, which rely solely on
a patient’s perception of their pain and functional independence, over more objective
gait assessments. The main reason is that gait assessment systems are typically large
and expensive and require technical training and expertise to be used properly.
Effects of Total Hip Arthroplasty on Gait 3
The State of the Art
Modern gait assessment methods are not only capable of providing accurate infor-
mation regarding the biomechanical aspects of gait but are also becoming both more
affordable and more accessible. Current objective gait assessment methods range
from laboratory-based methods, which include high-speed cameras, force platforms,
and video fluoroscopy, to portable ambulatory systems, which utilize 3-D inertial
sensors and pressure insoles. Furthermore, physical activity can now be assessed
both quantitatively and qualitatively, using portable and cost-effective triaxial accel-
erometers. The biomechanical function of the healthy hip joint is well studied.
However, alterations in gait mechanics following the onset of hip joint pathologies
are under continual debate, with several studies publishing contradictory results.
Most notable, mid- to long-term postoperative gait studies have shown no significant
effect of the type of surgical approach in relation to gait outcome, whereas early
postoperative studies have reported otherwise and hence such strong conclusions
cannot automatically be drawn.
Gait Assessment Methods
There are a variety of human motion analysis systems available from in vitro to
in vivo. Among the existing methods, lab-based methods consisting of both high
frame rate video cameras and high-precision force sensors embedded in a platform
are considered to be the “gold standard.” Retroreflective markers are placed on
anatomical locations, and joint movement is tracked by infrared cameras, providing
a full 3-D map of joint kinematics. Furthermore, force platforms, depending on size,
calculate the involved ground forces for one up to ten consecutive steps.
Ambulatory gait assessment (AGA) methods consist of inertial sensors and
pressure insoles. Inertial sensors are miniature devices consisting of 3-D accelerom-
eters, gyroscopes, and magnetometers (Tao et al. 2012), while pressure insoles have
flexible sensors attached to them. In terms of acquired data, AGA provides similar
information to lab-based equivalents but is far more economical in terms of cost,
space, and time. It also provides the freedom to assess individuals in their free-living
environment, testing not just the quality but also the quantity of activity. However, in
the above-described methods, the markers/sensors are placed on the body surface
and therefore subject to external environmental factors which could alter the place-
ment of the markers/body worn sensors leading to errors in the collected data (Della
Croce et al. 1999). Furthermore, as the hip is a deeply placed joint in the body, soft
tissue artifacts may appear in the data. However, mathematical algorithms are
available to minimize these artifacts. Another purely functional drawback of the
marker-/body-worn sensor systems is that it has been shown to have error in terms of
accurately locating the hip joint center, a crucial parameter, which can result in
kinematic assessment inaccuracies (Fiorentino et al. 2016).
Video fluoroscopy is another kinematic assessment method. It utilizes a series of
x-ray images, which are recorded as video frames. This method provides accurate
in vivo information of the joint motion in real time without any interference from
artifacts produced by skin or soft tissues (Bejjani et al. 1992). It has been found to be
a method for accurate measurement of the hip joint center, compared to the marker
system (Fiorentino et al. 2016). Note, however, that video fluoroscopy, even in
providing a more accurate measure of kinematics, is limited to research use only.
This is due to the continuous exposure to radiation during repeated recordings
exceeding limitations for the clinical purposes.
Another lab-based assessment method is electromyography (EMG). EMG cap-
tures the electrical signals generated by the neuromuscular activation of the skeletal
muscle. For hip joint assessment, EMG plays an important role as the hip joint is
surrounded by major muscle groups which play an important role in the stability,
balance, and smooth mobility of the joint. Muscle imbalance around a joint can
notably lead to various compensatory gait alterations, further backing the importance
of EMG assessment.
Hip Joint Gait Mechanics
Biomechanically, a healthy hip joint is known to provide more stability than

mobility. During different phases of gait, the hip joint adapts and rotates to maintain
balance. The majority of motion occurs in the sagittal plane, approximately 30 , with
a maximum hip flexion of about 25–30 and a maximum extension of about 8–10 .
Relative motion in the frontal and transverse planes is approximately 10–12 and
6–8 , respectively (Fig. 1). Upward and downward pelvic tilt has also been shown to
add 3–4 of both flexion and extension to the hip joint, respectively (Roberts 2010).
Furthermore, the forces applied by the surrounding muscles (moment/torque) adap-
tively change during walking in both sagittal and frontal planes to maintain balance.
In the sagittal plane, a peak in the external extension moment can be seen on initial
ground contact with a gradual transition into a flexion moment continuing until the
terminal stance (Fig. 1). In contrast, in the frontal plane, the external abduction
moment dominates throughout the stance phase. Lastly, the external moment in the
transverse plane showed a relatively small external rotation peak at the initial
contact, followed by a small internal rotation peak during the mid-stance which is
then observed to gradually transition to either an external rotation moment or to a
neutral position during terminal stance. Power at a joint is the product of the muscle
moment and its angular velocity. Power assessment at hip joint showed four peaks at
around 5, 45, 60, and 95% of the gait cycle (GC), the peaks during the stance phase
shows power absorption, and the peaks during the swing phase shows power
generation (Fig. 3). The ground reaction force (GRF) vector has been shown to
move progressively from an anterior to posterior position as the gait proceeds from
initial contact to terminal stance.
Fig. 1 Range of motion and internal moments at the hip joint during level walking in ten healthy
adults (age 30 4 years, BMI 23.9 2.4 kg/m2). Results are from the Motion Analysis Laboratory,
at Mayo Clinic, Rochester, MN
Pathologic Versus Instrumented Hip Gait Mechanics
An arthritic hip joint can be diagnosed via clinical and radiographic hallmarks.
Clinical hallmarks of hip osteoarthritis include pain, tenderness, stiffness, and loss
of flexibility. Radiographic hallmarks include nonuniform joint space narrowing,
osteophyte formation at the joint line, and development of subchondral sclerosis and
cysts. Notably, there are no such hallmarks for gait assessment. However, as it is
known that gait patterns vary between individuals, one cannot expect a completely
predefined gait pattern for a given body type and never mind a specific pathology.
Nonetheless, one can benchmark alterations in gait as would be seen in the majority
of patients, both with and without individualistic variances. Furthermore, it is always
beneficial to test other weight-bearing joints, on both the ipsilateral and contralateral
sides, to check for adaptations following the hip deformity (Watelain et al. 2001;
Yoshimoto et al. 2005; Foucher and Wimmer 2012). This is because the weight-
bearing joints work in a closed kinetic chain during weight-bearing activities and are
therefore directly or indirectly affected by any alterations. The following sections
compare the gait mechanics of pathological and instrumented hip joints. Arthritis is
used as an example pathology as most hip pathologies have a tendency to result in
arthritis, at which point total hip replacement becomes the treatment of choice.
Daily Physical Activity

Patients with hip arthritis are expected to see a noticeable reduction in their daily
physical activity level, due to pain and discomfort. Furthermore, an overall reduced
physical activity leads to muscle wasting around the affected hip joint. Quantita-
tively, based on accelerometer activity measured in counts per minute (cpm), healthy
elderly adults aged above 60 years show an average activity level of 200–400 cpm
(Hagstromer et al. 2007), (Evenson et al. 2012). For patients with hip arthritis, the
physical activity level is seen to decline to 100–200 cpm (Harding et al. 2014).
Following THA, short-term follow-up studies report no improvement in physical
activity level (Harding et al. 2014); however, long-term follow-up studies show
significant improvements (Lubbeke et al. 2014). It should be noted, however, that the
energy cost of moderate to vigorous physical activity has shown a significant
increase following THA (Lin et al. 2013).
Spatiotemporal
Patients with hip arthritis generally experience a reduction in walking speed,
1.01–1.08 m/s compared with 1.2–1.3 m/s for healthy adults, as a consequences of
a general reduction in stride length (1.13–1.17 m compared to 1.2–1.3 m), in single
support time (35% of cycle compared with 37%), and in cadence (103–110 steps/min
compared to 115–120 steps/min). Furthermore, step width is seen to be abnormally
wide (0.15–0.19 m compared to 0.08–0.12 m) due to the inherent lateral positioning
of the feet (Shrader et al. 2009). An increased step width also helps to provide a
larger base of support, probably to accommodate for the hip’s diminished postural
strategy. The reduced walking speed is directly related to pain and discomfort. It
follows that a reduction in stride length occurs as a result of a restricted freedom of
motion and that both a lower cadence and reduced support time are as a result of
patients consciously minimizing the frequency and duration of loading.
Following THA, patient walking speeds (1.2–1.4 m/s), stride lengths (1.3–1.4 m),
single support times (37.2 (% of cycle)), and cadence (112–119 steps/min) increase
significantly. Furthermore, the step width decreases to 0.1–0.08 m (Shrader et al.
2009), representing not only an improved gait pattern but also an improved hip
function, i.e., an enhanced freedom of motion. Do note, however, that despite the
reported improvements compared to preoperative values, the walking speed in
patients with THA is still generally slower in comparison to their healthy counter-
parts. Assuming no biomechanical issues remain, this could be due to apprehension,
fear of falling, and/or a longer than expected period of acclimatization.
Kinematics
The active and passive range of motion of an arthritic hip is shown to be reduced in
all planes. An arthritic hip is observed to stay in a relatively flexed position over most
of the gait cycle, reporting peak hip flexion between 25 and 30 and a peak extension
as low as 0 2 . Following THA, this altered motion in the sagittal plane is noted to
have improved; however, it is still seen to be significantly lower than a healthy hip
joint (Fig. 2).
In the frontal plane, overall motion at an arthritis hip is seen to be reduced. Unlike
a healthy hip, an arthritic hip begins the initial contact in a neutral or abducted
position and either stays in the abducted position or moves toward adduction
between 0 and 6 at approximately 20–25% of the gait cycle. The peak abduction
angle during toe-off stays between 0 and 10 . Following THA, an increased
adduction motion is observed (approximately 8–10 ) returning a near normal pattern
of motion in the frontal plane (Fig. 2).
Fig. 2: Hip kinematics and kinetics at sagittal, frontal, and transverse planes from ten patients with
severe arthritis, before (black) and 1 year after total hip arthroplasty (red). Average age and BMI of
the patient group were 51.03 4.4 years and 28.78 5.4 kg/m2, respectively. Results are from the
Motion Analysis Laboratory, at Mayo Clinic, Rochester, MN
Lastly, in transverse plane, an arthritic hip stays predominantly in the internally

rotated position, with an external rotation (2–5 ) only being reported at the initial
contact and terminal swing phases of the gait cycle (Fig. 2). Following THA, an
improvement is expected in the transverse plane motion as it is mostly a secondary
motion inherent to the hip muscles operating in the sagittal and frontal planes.
However, the observed improvement in range of motion is still not comparable to
a healthy hip. Overall, a large variability is seen in motion in all three planes of
patients with hip arthritis which continues to exist following THA. This would
suggest each patient could show a unique adaptation strategy.
Kinetics
The forces generated by muscles are not only necessary to support the body in an
upright position but are also necessary for forward propulsion of the body by
continuous redistribution of muscle power, between concentric and eccentric con-
tractions, around the trunk and leg muscles. In hip arthritis, a significant reduction in
muscle properties is seen on the affected side, compared to the contralateral asymp-
tomatic side (Barrett et al. 2013). In particular, arthritic hip muscles, along with the
corresponding knee extensors, have shown a reduction in overall muscle strength for
both concentric and eccentric contractions (Loureiro et al. 2013). Muscle size and
density have also been observed to have been adversely affected, as has the muscle
activation of hip muscles on the affected side (Loureiro et al. 2013). All of the above
structural and functional muscle deficiencies lead to an overall reduction in cumu-
lative force produced by the muscles, which would further lead to patient gait
asymmetry and the development of compensatory gait strategies. For example, the
Table 1: Range of available hip joint moments (nm/kg) for healthy, arthritic, and instrumented hip
joints. Results are from the Motion Analysis Laboratory, at Mayo Clinic, Rochester, MN
External joint moment Healthy hip Hip arthritis Instrumented hip
Peak flexor 0.7 to 1.3 0.25 to 0.7 0.42 to 0.8
Peak extensor 0.8 to 1.2 0.27 to 0.6 0.3 to 0.79
Peak adductor 0.94 to 1.12 0.5 to 0.8 0.63 to 0.97
Peak abductor 0.24 to 0.36 0.07 to 0.2 0.1 to 0.17
Peak internal rotator 0.09 to 0.23 0.1 to 0.26 0.15 to 0.38
Peak external rotator 0.07 to 0.16 0.03 to 0.09 0.05 to 0.1
Fig. 3: Joint power curves at the hip, knee, and ankle joint at the sagittal plane for healthy adults
(gray), patients with hip arthritis (black), and patients following 1-year total hip arthroplasty (red).
Results are from the Motion Analysis Laboratory, at Mayo Clinic, Rochester, MN
Trendelenburg gait pattern seen in hip pathology patients is mainly due to the
weakness of the hip abductors, specifically, the gluteus medius and minimus.
There are quantitative differences in the joint moments of healthy, arthritic, and
instrumented hip (Table 1). In contrast with healthy individuals, patients with hip
arthritis are shown to walk with reduced muscle moments in all three planes, which
suggest an overall reduced muscle function in an arthritic hip. Furthermore, the
altered external rotation moment peak is specifically of interest, as the small peak is
seen to occur at the early stance and continues to stay until the early swing, unlike the
healthy hip. This pattern is unfortunately noted to continue 1 year following THA
(Fig. 2). Also, the inability to extend and adduct the hip may, in long term, lead to
flexion and abduction contractures, as is evident in patients with severe hip arthritis.
The reduced joint power peaks for patients with arthritis confirm the reduced power
generation and absorption at the diseased hip and the adjacent joints on the affected
side (Fig. 3). The reduced power absorption during the initial and terminal stance, as
well as a reduced power generation during push-off, results in an overall reduced hip
moment in the sagittal plane.
Following THA, studies have shown a significant increase in both flexion and
abduction moments and corrective external rotation moment pattern in the early
stance phase. Regained muscle power, as a result of intensive and focused
Fig. 4 Pelvis motion for healthy adults (gray), patients with hip arthritis (black), and patients
following 1-year total hip arthroplasty (red). Results are from the Motion Analysis Laboratory, at
Mayo Clinic, Rochester, MN
rehabilitation protocols, is one of the reasons why patients with THA are known for
fast recovery.
Neighboring Joint Strategies

The lower extremities and the trunk work as a closed kinetic chain during weight-
bearing activities and consequently impact one another if their mechanics are altered.
As a result, a disruption in hip joint mechanics can have an adverse effect on the
spine, pelvis, knee, and ankle joints, both on the ipsilateral side (directly) and even
on the contralateral side (indirectly), as has been observed in patients with severe hip
arthritis.
The pelvis, being the closest neighboring segment, shows the most alteration in
hip pathologies. In particular, a restricted hip motion has been shown to increase the
pelvic tilt and pelvic obliquity as a compensatory mechanism (Fig. 4). Pelvic tilt is
the anterior and posterior motion of the pelvis in the sagittal plane, while pelvic
obliquity is the lateral upward and downward motion of the pelvis in the frontal
plane. During walking normal pelvic tilt is between 2 and 8 , pelvic obliquity
between 2 and 4 , and pelvic rotation between 8 and 10 (Perry and Burnfield
2010).
The increased pelvic tilt and obliquity shortens the moment arm between the hip
and the upper body center of mass, which results in extension of the hip joint,
initiating push-off at the end of the stance phase (Yoshimoto et al. 2005). This pelvic
strategy further leads to an increased trunk inclination, toward the sound side, which
is seen to have a direct impact on the lumbar spine resulting in lumbar dysfunction.
Furthermore, hip flexion contracture and an increased lumbar lordosis are seen as a
compensatory mechanism to initiate hip extension. Pelvic incidence is another
parameter which is seen to be significantly higher in patients with hip arthritis and
is itself strongly correlated to pelvic tilt, lumbar lordosis, and sacral slope
(Yoshimoto et al. 2005). Pelvic incidence is calculated by adding the pelvic tilt
and the sacral slope.
For the knee joint, two primary muscle groups control the motion. These primary
muscle groups are two jointed and cross both the hip and knee joints. Consequently,
Fig. 5: Knee motion and moment for healthy adults (gray), patients with hip arthritis (black), and
patients following 1-year total hip arthroplasty (red). Results are from the Motion Analysis
Laboratory, at Mayo Clinic, Rochester, MN
muscle weakness surrounding the hip could have direct effect on muscle control at
the knee joint as has been witnessed in patients with hip arthritis (Fig. 3). Involved
muscles include semitendinosus, semimembranosus, and long head of biceps
femoris of the hamstring group, posteriorly, and rectus femoris part of the quadri-
ceps, anteriorly.
In patients with hip arthritis, the knee motion in sagittal plane is reduced, while in
frontal and transverse plane, it is noticeably altered (Fig. 5). The altered varus and
inverted position of the knee during the early to late swing phase are seen to persist
1 year postoperatively and continue to stay significantly different from the healthy
knee motion. A noticeable reduction in joint moment is also seen in hip arthritis,
which is seen to improve 1 year following THA. However, compared to the healthy
knee, this reported improvement is still not comparable.
In terms of joint power in the sagittal plane, patients with hip arthritis show
50–60% less energy generation and absorption compared to the healthy knee
(Fig. 3). The reduced power at the knee is due to the altered functions of the muscles
connecting between the diseased hip and knee. For example, during mid-stance,
when the knee straightens to take the load, a reduced power generation is seen as a
result of the reduced hip extension. Furthermore, during push-off, patients with
arthritis demonstrate a 40–45% reduction in peak energy absorption to assist the
push-off with restricted hip extension. Lastly, at the late swing phase, a 30–35%
reduction in energy absorption can be seen in the fourth peak, representing a reduced
peak swing speed. The frontal and transverse planes also are shown to absorb less
power during the push-off (Watelain et al. 2001). This is a compensatory strategy to
limit hip abduction in order to maintain the lower limb joint alignment.
Fig. 6: Ankle motion and moment for healthy adults (gray), patients with hip arthritis (black), and
patients following 1-year total hip arthroplasty (red). Results are from the Motion Analysis
Laboratory, at Mayo Clinic, Rochester, MN
Finally, at the ankle joint, patients with hip arthritis show a significantly reduced
motion seen in all three planes. However, muscle moment showed no difference in
the sagittal plane, while significantly increased and decreased moments are found at
the frontal and transverse planes, respectively (Fig. 6). The power generation during
push-off is seen to be low compared to the healthy ankle; this could be a compen-
sation strategy to facilitate push-off with weak hip extension. The power generation
peak is seen to improve following THA but stays relatively low compared to the
healthy knee (Fig. 3).
In relation to THA, studies have demonstrated notable improvements over initial
gait alterations; however, gait mechanics typically do not return to normal
(Horstmann et al. 2013). THA is found to be effective in reducing the lower back
pain symptoms; however, gait mechanics at pelvis and spine have not been seen to
return to normal (Ben-Galim et al. 2007; Foucher and Wimmer 2012). The alter-
ations at knee and ankle joint mechanics are seen to persist up to 6–12 months
postoperatively (Reardon et al. 2001; Foucher and Wimmer 2012) (Beaulieu et al.
2010). Consequently, one can predict that the preoperative gait mechanics in the
neighboring joints, including those bilaterally, may lead to arthritic changes due to
consistent abnormal loading. Notably, there are studies reporting a lateral patellar tilt
on the ipsilateral knee, due to the internally rotated position of the hip, with an
increased adduction moment and the development of medial tibiofemoral arthritis on
the affected side as a result of the altered axial alignment (Akiyama et al. 2016;
Umeda et al. 2009).
The discussed outcome of THA is seen to be consistent with the various surgical
approaches used, including anterior, posterior, direct lateral, anterolateral, and pos-
terolateral (Queen et al. 2014; Wesseling et al. 2016). However, inconsistencies can
Table 2 Range of motion during stair ascent and descent at self-selected speed in healthy, arthritic,
and instrumented hip joint
Peak hip Healthy hip Hip arthritis Instrumented hip
motion Ascent Decent Ascent Decent Ascent Decent
Flexion 40 to 60 13 to 27 50 to 75 25 to 43 52 to 65 18 to 33
Extension 6.8 to 7.9 3.2 to 12.62 2 to 24 10.6 to 26 1.2 to 12.4 4 to 18.4
Adduction 4.9 to 12.5 3.2 to 10.2 2.5 to 13.7 0.6 to 10.2 6 to 15 2.3 to 10.7
Abduction 0 to 9.5 0.13 to 7.7 0.16 to 8.9 2 to 6.6 0.48 to 7.2 0.9 to 6.3
(Queen et al. 2013, 2015)
be found in studies comparing early postoperative outcomes (Queen et al. 2011;

Barrett et al. 2013; Madsen et al. 2004). From an anatomical perspective, the anterior
approach should help with a faster recovery of gait as the hip muscles are preserved
during the surgery. In contrast, lateral or posterior approaches result in the muscles
being dissected to access the joint, and the recovery of muscle strength following
these procedures is notably longer. It can also be seen that the minimally invasive
THA is shown to have better gait recovery than the conventional method (Weber
et al. 2014).
Stair Ascending and Descending Biomechanics

Joint mechanics during stair activity should be considered while assessing the
outcome of THA. It can be said that walking up and down stairs requires a higher
degree of freedom of motion of the hip joint (Table 2). Furthermore, good muscle
control especially of the major two jointed muscles around the hip and knee joint is
important for good power and balance (Andriacchi et al. 1980). The power gener-
ation is seen to be significantly higher during stair ascending than compared to
descending (Fig. 7). This is confirmed by the noticeably higher muscle moments in
the sagittal plane, reaching up to 1 nm/kg while ascending, in comparison to less
than 0.5 nm/kg during descent, in healthy adults (Fig. 8). However, in the frontal
plane, stair descent uses a muscle force >1 nm/kg.
In hip arthritis, difficulty with stair climbing has been consistently reported, due
to the weakness of the major hip muscles leading to reduced muscle moments
(Fig. 7) and an altered power generation (Fig. 8). Power generation is seen to reduce
significantly in hip arthritis which would lead to compensation strategies to achieve
the range of motion required to initiate stair climbing. The inability of muscles to
produce enough power for the increased range of motion required during stair
climbing is compensated by a reduced stair climbing speed in these patients. All
of the above results in patients adopting altered angular and loading strategies when
both ascending and descending stairs, which may lead to bilateral asymmetries.
Following THA, joint muscle moment and power are seen to improve significantly,
returning the power generation to near normal requirements for stair climbing.
However, the improvement is not comparable to a healthy hip. Note that the
transverse pain moments, though significantly lower compared to the other two
planes, could have some effect due to the altered pattern during both ascending
and descending stairs.
Fig. 7 Hip joint external moments curve during ascend (top) and descend (bottom) of stairs for
healthy adults (gray), patients with hip arthritis (black), and patients following 1-year total hip
arthroplasty (red). Results are from the Motion Analysis Laboratory, at Mayo Clinic,
Rochester, MN
Fig. 8: Hip joint power during ascend and descend of stairs for healthy adults (gray), patients with
hip arthritis (black), and patients following 1-year total hip arthroplasty (red). Results are from the
Motion Analysis Laboratory, at Mayo Clinic, Rochester, MN
Future Work
Research suggests that investigating the extent of remaining preoperative gait

patterns at each joint would help improve rehabilitation, by reducing preoperative
gait abnormalities as early as possible. Large-scale studies are therefore needed,
assessing all lower extremity joints both pre- and postoperatively for both short- and
long-term THA outcomes. There is also a need for a simplified objective gait score,
assessing the severity of gait alterations in patients with hip arthritis which would
assist clinicians in planning treatments, accordingly.
Cross-References
▶ 3D Dynamic Probablistic Pose Estimation Using Cameras and Reflective Markers

▶ Arthrokinematics and Joint Morphology
▶ Gait Parameters Estimated Using Magneto and Inertial Measurement Units
Analysis
▶ Three-dimensional Human Kinematics Estimation Using Magneto and Inertial
Measurement Units
References
Akiyama K, Nakata K, Kitada M, Yamamura M, Ohori T, Owaki H, Fuji T (2016) Changes in axial
alignment of the ipsilateral hip and knee after total hip arthroplasty. Bone Joint J 98-B:349–358
mechanics during stair-climbing. J Bone Joint Surg Am 62:749–757
Barrett WP, Turner SE, Leopold JP (2013) Prospective randomized study of direct anterior vs
postero-lateral approach for total hip arthroplasty. J Arthroplast 28:1634–1638
Beaulieu ML, Lamontagne M, Beaulé PE (2010) Lower limb biomechanics during gait do not
return to normal following total hip arthroplasty. Gait Posture 32:269–273
Bejjani FJ, Lockett R, Pavlidis L (1992) Videofluoroscopy system for in vivo motion analysis.
Google Patents
Ben-Galim P, Ben-Galim T, Rand N, Haim A, Hipp J, Dekel S, Floman Y (2007) Hip-spine
syndrome: the effect of total hip replacement surgery on low back pain in severe osteoarthritis
of the hip. Spine (Phila Pa 1976) 32:2099–2102
Centers For Disease Control and Prevention (2010) National Hospital Discharge Survey: 2010
table, procedures by selected patient characteristics – number by procedure category and age
Della Croce U, Cappuzzo A, Kerrigan DC (1999) Pelvis and lower limb anatomical landmark
calibration precision and its propagation to bone geometry and joint angles. Med Biol Eng
Comput 37:155–161
Evenson KR, Buchner DM, Morland KB (2012) Objective measurement of physical activity and
sedentary behavior among US adults aged 60 years or older. Prev Chronic Dis 9:E26
Fiorentino NM, Kutschke MJ, Atkins PR, Foreman KB, Kapron AL, Anderson AE (2016)
Accuracy of functional and predictive methods to calculate the hip joint center in young
non-pathologic asymptomatic adults with dual fluoroscopy as a reference standard. Ann Biomed
Eng 44:2168–2180
Foucher KC, Wimmer MA (2012) Contralateral hip and knee gait biomechanics are unchanged by
total hip replacement for unilateral hip osteoarthritis. Gait Posture 35:61–65
Hagstromer M, Oja P, Sjostrom M (2007) Physical activity and inactivity in an adult population
assessed by accelerometry. Med Sci Sports Exerc 39:1502–1508
Harding P, Holland AE, Delany C, Hinman RS (2014) Do activity levels increase after total hip and
knee arthroplasty? Clin Orthop Relat Res 472:1502–1511
Horstmann T, Listringhaus R, Haase GB, Grau S, Mundermann A (2013) Changes in gait patterns
and muscle activity following total hip arthroplasty: a six-month follow-up. Clin Biomech
(Bristol, Avon) 28:762–769
Lin BA, Thomas P, Spiezia F, Loppini M, Maffulli N (2013) Changes in daily physical activity
before and after total hip arthroplasty. A pilot study using accelerometry. Surgeon 11:87–91
Louriuro A, Mills PM, Barrett RS (2013) Muscle weakness in hip osteoarthritis: a systematic
review. Arthritis Care Res 65:340–352
Lubbeke A, Zimmermann-Sloutskis D, Stern R, Roussos C, Bonvin A, Perneger T, Peter R,
Hoffmeyer P (2014) Physical activity before and after primary total hip arthroplasty: a
registry-based study. Arthritis Care Res 66:277–284
Madsen MS, Ritter MA, Morris HH, Meding JB, Berend ME, Faris PM, Vardaxis VG (2004) The
effect of total hip arthroplasty surgical approach on gait. J Orthop Res 22:44–50
OECD (2015) Health at a glance 2015: OECD indicators (summary). OECD Publishing, Paris
Perry J, Burnfield J (2010) Gait analysis: normal and pathological function. J Sports Sci Med 9:353
Queen RM, Butler RJ, Watters TS, Kelley SS, Attarian DE, Bolognesi MP (2011) The effect of total
hip arthroplasty surgical approach on postoperative gait mechanics. J Arthroplast 26:66–71
Queen RM, Newman ET, Abbey AN, Vail TP, Bolognesi MP (2013) Stair ascending and
descending in hip resurfacing and large head total hip arthroplasty patients. J Arthroplast
28:684–689
Queen RM, Appleton JS, Butler RJ, Newman ET, Kelley SS, Attarian DE, Bolognesi MP (2014)
Total hip arthroplasty surgical approach does not alter postoperative gait mechanics one year
after surgery. PM R 6:221–226; quiz 226
Queen RM, Attarian DE, Bolognesi MP, Butler RJ (2015) Bilateral symmetry in lower extremity
mechanics during stair ascent and descent following a total hip arthroplasty a one-year longi-
tudinal study. Clin Biomech (Bristol, Avon) 30:53–58
Reardon K, Galea M, Dennett X, Choong P, Byrne E (2001) Quadriceps muscle wasting persists
5 months after total hip arthroplasty for osteoarthritis of the hip: a pilot study. Intern Med J
31:7–14
Roberts A (2010) Gait analysis: normal and pathological function (2nd edition). Bone Joint J
92-B(8):1184
Shrader W, Bhowmik-Stoker M, Jacofsky MC, Jacofsky DJ (2009) Gait and stair function in total
and resurfacing hip arthroplasty: a pilot study. Clin Orthop Relat Res 467:1476–1484
Tao W, Liu T, Zheng R, Feng H (2012) Gait analysis using wearable sensors. Sensors (Basel,
Switzerland) 12:2255–2283
Umeda N, Miki H, Nishii T, Yoshikawa H, Sugano N (2009) Progression of osteoarthritis of the
knee after unilateral total hip arthroplasty: minimum 10-year follow-up study. Arch Orthop
Trauma Surg 129:149–154
Van Den Bogert AJ, Read L, Nigg BM (1999) An analysis of hip joint loading during walking,
running, and skiing. Med Sci Sports Exerc 31:131–142
Watelain E, Dujardin F, Babier F, Dubois D, Allard P (2001) Pelvic and lower limb compensatory
actions of subjects in an early stage of hip osteoarthritis. Arch Phys Med Rehabil 82:1705–1711
Weber T, Al-Munajjed AA, Verkerke GJ, Dendorfer S, Renkawitz T (2014) Influence of minimally
invasive total hip replacement on hip reaction forces and their orientations. J Orthop Res
32:1680–1687
Wesseling M, Meyer C, Corten K, Simon JP, Desloovere K, Jonkers I (2016) Does surgical
approach or prosthesis type affect hip joint loading one year after surgery? Gait Posture
44:74–82
Yoshimoto H, Sato S, Masuda T, Kanno T, Shundo M, Hyakumachi T, Yanagibashi Y (2005)
Spinopelvic alignment in patients with osteoarthrosis of the hip: a radiographic comparison to
patients with low back pain. Spine (Phila Pa 1976) 30:1650–1657
The Effects of Ankle Joint Replacement
on Gait
Justin Michael Kane, Scott Coleman, and James White Brodsky
Abstract
Patients with ankle arthritis experience significant disability. In patient-centered
surveys, the mental and physical disability is equivalent to that associated with
end-stage hip arthritis (Glazebrook et al. Bone Joint Surg Am 90(3):499–505,
2008). End-stage ankle arthritis has a number of features differentiating it from
hip and knee arthritis. Unlike the primary osteoarthritis of the hip and knee, 70%
of ankle arthritis is post-traumatic in nature (Saltzman et al. Iowa Orthop J
25:44–46, 2005). Given the high degree of success seen with total hip
arthroplasty (THA) and total knee arthroplasty (TKA), the first generation of
total ankle arthroplasty (TAA) was introduced in the 1970s. Unfortunately, early
TAA did not have the same success as other arthroplasty procedures, and the
procedure was largely abandoned in favor of return to ankle arthrodesis (Bolton-
J.M. Kane (*)

Baylor University Medical Center, McKinney, TX, USA
Faculty, Foot and Ankle Fellowship Program, Baylor University Medical Center, Dallas, TX, USA
Orthopedic Associates of Dallas, Dallas, TX, USA
e-mail: jkane.md@gmail.com
S. Coleman
Department of Orthopaedics, Baylor University Medical Center, Dallas, TX, USA
Department of Orthopedics, Baylor Scott & White, Dallas, TX, USA
e-mail: scottcoleman12@hotmail.com
J.W. Brodsky
Faculty, Foot and Ankle Fellowship Program, Baylor University Medical Center, Dallas, TX, USA
University of Texas Southwestern Medical School, Dallas, TX, USA
Texas A&M HSC College of Medicine, Bryan, TX, USA
e-mail: jwbrodsky@gmail.com

DOI 10.1007/978-3-319-30808-1_83-1
2 J.M. Kane et al.
Maggs et al. J Bone Joint Surg Br 67:785–790, 1985, Dini and Bassett Clin
Orthop 146:228–230, 1980, Kitaoka and Patzer J Bone Joint Surg Am
76:974–979, 1994, Stauffer and Seagal Clin Orthop 160:217–221, 1981, Kofoed
and Sorensen J Bone Joint Surg 80-B:328–332, 1998, Demottaz et al. J Bone
Joint Surg Am 61 (7):976–988, 1979). While arthrodesis was more reliable with
fewer complications than arthroplasty, arthrodesis has the limitation of loss of
ankle motion, especially in cases in which there is severe arthritis but moderate
residual joint motion. Complications of arthrodesis include nonunion and
malunion, and there are reports of arthritis at adjacent joints and residual abnor-
malities of gait (Beischer et al. Foot Ankle Int 20:545–553, 1999, Buchner and
Sabo Clin Orthop Relat Res 406:155–164, 2003, Buck et al. J Bone Joint Surg
69-A:1052–1062, 1987, Coester et al. J Bone Joint Surg 83-A:219–228, 2001,
Fuchs et al. J Bone Joint Surg 85-B:994–998, 2003).
The complications associated with ankle arthrodesis, coupled with the desire
to more normally replicate ankle biomechanics, have led to renewed interest in
TAA in recent years, with encouraging reports of early- and mid-term results and
improved survivorship compared to first-generation series. Even though TAA
does not yet have the longevity of THA and TKA, there are many early and
intermediate-term reports of high levels of patient satisfaction, pain relief and
patient function, and variable survivorship at 80–95%, depending on the length of
follow-up (Haddad et al. J Bone Joint Surg Am 89:1899–1905, 2007a,
Gougoulias et al. Clin Orthop 468:199–208, 2010).
One of the postulated benefits of TAA is preservation of tibiotalar motion. A
number of studies have demonstrated improvements in gait following TAA.
While TAA does not restore normal gait, patients have improvements in nearly
all parameters of gait with many approaching normal controls (Valderrabano et al.
Clin Biomech 22:894–904, 2007, Doets et al. Foot Ankle Int 28(3):313–322,
2007, Singer et al. J Bone Joint Surg Am 95(e191):1–10, 2013, Flavin et al. Foot
Ankle Int 34(1):1340–1348, 2013). More importantly, when gait studies have
compared TAA with preoperative function, TAA offers a significant improvement
(Singer et al. J Bone Joint Surg Am 95(e191):1–10, 2013, Flavin et al. Foot Ankle
Int 34(1):1340–1348, 2013, Queen et al. J Bone Joint Surg Am 96:987–993,
2014a, Piriou et al. Foot Ankle Int 29(1):3–9, Brodsky et al. J Bone Joint Surg
Am 93:1890–1896, 2011, Queen et al. Foot Ankle Int 33(7):535–542, 2012,
Queen et al. Clin Biomech 29:418–422, 2014b).
Keywords
Total ankle arthroplasty • Implant design • Clinical outcomes • Patient-reported
outcomes • Gait analysis
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Biomechanics and Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
The Effects of Ankle Joint Replacement on Gait 3
Clinical Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Temporal-Spatial Parameters of Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Kinematic Parameters of Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Kinetic Parameters of Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Case Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
State of the Art
First-generation total ankle replacement implant designs were broadly categorized as

constrained or unconstrained. Both designs produced poor results. While
unconstrained designs allowed more mobility and decreased strain at the bone-
implant interface, they had a high level of instability and high rate of wear on the
polyethylene component resulting in an unacceptable rate of early failure.
Constrained designs reduced instability and allowed a more even force distribution
on the polyethylene, but motion was limited to the sagittal plane, and the restricted
motion resulted in an increased strain at the bone-implant interface leading to early
loosening (Bolton-Maggs et al. 1985; Dini and Bassett 1980; Kitaoka and Patzer
1994; Stauffer and Seagal 1981; Kofoed and Sorensen 1998; Lewis 1994; Kakkar
and Siddique 2011).
A better understanding of biomechanics at the ankle has led to the development of
the current generation of implants used today (Reggiani et al. 2006; Espinoza et al.
2010; Leardini et al. 2014). Of the many TAA implants on the world market today,
all are either two component or three component. In two-component designs, the
polyethylene is fixed to the tibial component, and motion occurs between it and the
talar component. In three-component designs, the polyethylene is a mobile bearing
allowing motion at the interfaces between the polyethylene and both the tibial and
talar components. The rationale for the three-component design is to achieve a more
normal restoration of multiplane motion. A lack of conclusive evidence exists to
support superiority of either implant design.
Biomechanics and Anatomy
The tibiotalar joint is a highly constrained articulation comprised of the tibia, fibula,
and dorsal surface of the talus. Poor understanding of the highly congruous joint
mechanics, the complexities of ankle motion, and the stresses on the implant-bone
interface contributed to early implant failures. A thorough understanding of the
anatomy and biomechanics is integral when considering designs of total ankle
prostheses (Deland et al. 2000; Gill 2002).
The three-dimensional osteology of the joint is such that with dorsiflexion there is
a component of external rotation and abduction and with plantarflexion there is a
4 J.M. Kane et al.
component of internal rotation and adduction. The dorsal aspect of the talus has two
domes with differing radii. The medial dome has a smaller radius of curvature than
the lateral which explains the secondary motion seen at the tibiotalar joint (Barnett
and Napier 1952). The articular surface of the talus is wedge shaped with a wider
anterior and tapers posteriorly in an asymmetric manner (Barnett and Napier 1952;
Inman 1991).
As the tibiotalar joint plantar flexes, the axis of rotation shifts from anterior to
posterior. While range of motion in the sagittal plane is 70 (Tooms 1987), studies
have suggested the working range of motion is limited to 25 (15 of plantarflexion
and 10 of dorsiflexion) during the stance phase of gait (Stauffer 1979).
During gait, a vertical load of greater than five times the body weight is trans-
mitted at the tibiotalar joint (Stauffer et al. 1977). The same vertical load is
experienced after TAA, and shear forces can exceed two to three times the body
weight during the gait cycle (Stauffer et al. 1977). Given the significant forces
exerted at the tibiotalar joint and the small weight-bearing surface of only 7 cm2
(Hintermann 2005), design considerations are paramount to implant survivorship.
Clinical Outcomes
In recent studies of TAA, investigators have reported good to excellent outcome in

almost 80% of cases at greater than 2-year follow-up (Haddad et al. 2007b). In a
meta-analysis conducted by Haddad et al., the American Orthopedic Foot and Ankle
Society (AOFAS) hind-foot score was 78.2. They found a revision rate of only 7%
and a 10-year survival rate of 77% (Haddad et al. 2007b).
In a study by Queen et al. comparing two-component and three-component
implants, the mean improvement in AOFAS score was from 32.9 to 65 at 1 year
with additional improvement to 81.4 at 2 years. Improvement in the Foot and Ankle
Disability Index (FADI) was from 60.0 to 17.4 at 1 year and 13.3 at 2 years. Both
Short Musculoskeletal Function Assessment (SMFA) function and bother scores
improved at 1 year with the function score improving from 27.3 to 11.8 and the
bother score improving from 32 to 12.9. The two-component cohort experienced
greater improvement in SF36 total score compared to the three-component cohort,
while the three-component group exhibited a greater reduction in visual analog scale
(VAS) pain scores. Unfortunately, the difference between the two cohorts was unable
to conclusively determine which was more efficacious in the treatment of end-stage
ankle arthritis (Queen et al. 2014a).
Gait Analysis
End-stage ankle arthritis is not only subjectively debilitating. Numerous studies have
demonstrated abnormality of each parameter of gait. In order to fully appreciate the
disability associated with end-stage ankle arthritis and the improvements gained with
total ankle arthroplasty, it is essential to understand the basic measurements of gait

analysis. The parameters of gait are divided into three categories.
Temporal-spatial parameters (TSPs) of gait are the most clinically appreciable
measurements and are the most intuitive and easily appreciated (chapter
“▶ Interpreting Spatiotemporal Parameters, Symmetry and Variability in Clinical
Gait Analysis”). TSPs are considered the “vital signs of gait” (Kirtley 2006),
measuring the speed of walking and its component parts, including support time
for each limb. Walking speed is the product of step length and cadence (steps/
minute). It is important to keep in mind that when evaluating TSPs, a number of
factors (i.e., pain, stiffness) influence function irrespective of ankle pathology
(Valderrabano et al. 2007; Kawamura et al. 1991; McIntosh et al. 2006; Ilgin et al.
2011; Inam et al. 2010; Ledoux et al. 2006; Potter et al. 1995; Queen and Nunley
2010).
Kinematic parameters of gait measure movement, separate from the forces
driving locomotion. Kinematics entails the displacement of an anatomic segment
(linear or angular), velocity, or acceleration (chapter “▶ 3D Kinematics of Human
Motion”) (Kirtley 2006). Most modern gait analyses rely on motion-tracking devices
to measure specific angles and trajectories as they pertain to the anatomic areas of
interest. While traditional gait analysis measured motion using a model which treats
the foot as a single segment, recent work has focused on multi-segment kinematic
models to distinguish motion between and among the different parts of the foot in the
coronal, axial, and sagittal planes (chapter “▶ Kinematic Foot Models for
Instrumented Gait Analysis”) (Mayich et al. 2014; Novak et al. 2014).
Kinetic parameters measure the forces generated with gait and are expressed both
as power and moment (force plus direction). Ground reaction forces, joint moments,
and joint mechanical power are all measured within the context of the kinetic
parameters of gait analysis (chapters “▶ Interpreting Ground Reaction Forces in
Gait” and “▶ Interpreting Joint Moments and Powers in Gait”) (Queen and Nunley
2010).
Temporal-Spatial Parameters of Gait
In patients with end-stage ankle arthritis, stride length, cadence, walking speed, and
support time for the affected limb are all abnormal (Valderrabano et al. 2007;
Stauffer et al. 1977; Khazzam et al. 2006). While some of these changes may be
attributable to the pathology itself, it is important to consider that alterations in
temporal-spatial parameters of gait have been noted as a protective strategy to reduce
the load and therefore pain on the affected joint (Mundermann et al. 2004).
A number of studies compared gait before and after total ankle arthroplasty and
demonstrated improvements in the temporal-spatial parameters of gait after surgery.
Brodsky et al. demonstrated statistically and clinically significant improvements in
gait after ankle replacement with the STAR prosthesis. Outcomes were recorded for
the affected limb, and stride length (+17 cm), cadence (+12.9 steps/min), and
walking speed (25.6 cm/s) were all measured to be improved ( p < 0.001) (Brodsky
6 J.M. Kane et al.
et al. 2011). Valderrabano et al. reported similar results with the HINTEGRA
prosthesis. Improvements in stride length (+5 cm), cadence (+5.6 steps/min), walk-
ing speed (+12 cm/s), and support time ( 0.04 s) for the affected limb were detected.
For the unaffected limb, improvements in stride length (+7 cm), cadence (+4.8 steps/
min), walking speed (+12 cm/s), and support time ( 0.07 s) were detected. While
these values did not reach those of healthy controls, the temporal-spatial parameters
measured did improve both statistically and clinically significantly and more closely
replicated the parameters of healthy controls (Valderrabano et al. 2007). Queen et al.
reported similar improvements across temporal-spatial parameters of gait. Stride
length (+13.9 cm), walking speed (+31 cm/s), and double-limb support time
( 6%) had all statistically and clinically significant improvements (Queen et al.
2012). These results parallel those of earlier authors who also found statistically and
clinically significant improvements in stride length, cadence, and walking speed
even though improvements did not reach the level of healthy controls (Doets et al.
2007; Dyrby et al. 2004; Houdijk et al. 2008; Ingrosso et al. 2009).
In studies comparing total ankle arthroplasty with tibiotalar arthrodesis, a number
of important differences were noted. Flavin et al. studied the preoperative and
postoperative gait of patients who had total ankle arthroplasty and ankle arthrodesis,
evaluating the improvements in each group and comparing them as well. Both
groups had significant improvements in multiple parameters of gait following
surgery, even though neither group approached the function of normal controls.
Certain parameters of gait were superior in each group compared to the other (Flavin
et al. 2013). This parallels the results of other authors who report similar postoper-
ative temporal-spatial parameters of gait that are improved over the gait of patients
with end-stage ankle arthritis but are not equivalent to the gait of healthy controls
(Singer et al. 2013; Hahn et al. 2012).
Kinematic Parameters of Gait
Numerous studies have reduction of movement in all three planes, in patients with
end-stage ankle arthritis (Valderrabano et al. 2007; Stauffer et al. 1977; Khazzam
et al. 2006). Predictably, the greatest loss of motion is in the sagittal plane
(dorsiflexion/plantarflexion).
Enthusiasm surrounding total ankle arthroplasty is centered on its theoretical
ability to preserve and/or recreate normal joint kinematics. While studies have failed
to demonstrate a restoration of ankle motion to that of healthy controls, improve-
ments in patients with end-stage ankle arthritis have demonstrable improvement in
motion. Brodsky et al. reported an improvement in total sagittal range of motion in
the ankle (+3.7 ), knee (+6.6 ), and hip (+4.9 ), all of which were statistically
significant. In this study the increase in tibiotalar motion was predominantly in
plantarflexion (Brodsky et al. 2011). In a study of the two-component Salto Talaris
TAA, there was also increase in ROM (+4.8 ) with the increase occurring predom-
inantly in dorsiflexion (Choi et al. 2013). Valderrabano et al. reported improvements
in ankle plantarflexion (+4.3 ), inversion movement (+1.1 ), and adduction move-

ment (+1.6 ). While these values failed to reach range of motion in healthy controls,
a statistically significant improvement was noted (Valderrabano et al. 2007). Brod-
sky et al. also compared the affected limb to the contralateral extremity using multi-
segment three-dimensional gait analysis. Range of motion in the sagittal, coronal,
and axial plane was statistically and clinically less than the unaffected limb (Brodsky
et al. 2013). In comparisons between total ankle arthroplasty and tibiotalar arthrod-
esis, kinematic parameters of gait reveal significant improvements for both cohorts
postoperatively, with neither group reaching the values seen in healthy controls for
any of the parameters. Flavin et al. reported superior sagittal plane dorsiflexion
compared to patients undergoing TAA compared to arthrodesis, while those under-
going arthrodesis had superior coronal plane eversion. However, neither procedure
resulted in normalization of kinematic parameters of gait (Flavin et al. 2013). Singer
et al. compared postoperative results of total ankle arthroplasty and tibiotalar
arthrodesis with healthy controls. Total sagittal range of motion was greater in the
arthroplasty group than the arthrodesis group (+4.4 ) but was still not normal.
Coronal plane of motion improved to a greater extent in the arthroplasty group
(+2.5 ) but was still less than that in healthy controls. Tibial rotation was greater in
the arthroplasty group (+1.4 ) and was similar to the control group (Singer et al.
2013). Hahn et al. reported a net increase in total sagittal range of motion of +3 in
patients undergoing total ankle arthroplasty with the majority of the improvement
gained in plantarflexion, while patients undergoing tibiotalar arthrodesis exhibited
decreased sagittal range of motion of 3 (Hahn et al. 2012).
Kinetic Parameters of Gait
The kinetic abnormalities associated with end-stage ankle arthritis present distinct
patterns compared to arthritis of the hips and knees. Ankle power undoubtedly is
reduced as a direct result of the abnormal temporal-spatial and kinematic parameters
of gait. Moment is also abnormal. The largest reduction in moment is seen in the
axial plane, whereas in knee osteoarthritis the largest reduction is seen in coronal
plane moment (Mundermann et al. 2004; Hurwitz et al. 2002). The differences seen
with moment may be a result of the etiology of the arthritis in the ankle compared to
the hip and knee. The vast majority of patients with end-stage ankle arthritis have a
traumatic etiology, whereas patients with hip and knee arthritis usually have primary
osteoarthritis. This post-traumatic origin may be associated with muscle and soft-
tissue compromise resulting in reduced function or may be a protective mechanism
to reduce the complex ankle joint in an effort to minimize pain. Kinetic analysis has
traditionally required that the foot be treated as a single segment in the gait model.
Brodsky et al. reported improvements in both ankle power (+0.31 W/kg) and
ankle plantarflexion moment (+0.21 Nm/kg) after total ankle replacement with the
Scandinavian total ankle replacement (Brodsky et al. 2011). Valderrabano et al.
reported similar results – ankle power improved (+0.75 W/kg), ankle plantarflexion
8 J.M. Kane et al.
moment improved (+0.04 Nm/kg), and ankle inversion moment improved

(+0.03 Nm/kg). While these results were statistically and clinically significant,
they did not reach the values of the unaffected limb (Valderrabano et al. 2007).
Queen et al. also reported improvements in kinetic parameters of gait after total ankle
arthroplasty. Peak anterior ground reaction force and posterior ground reaction force
exhibited statistically and clinically significant improvements as did ankle
plantarflexion moment (Queen et al. 2012).
In a comparison of total ankle arthroplasty and tibiotalar arthrodesis, Flavin et al.
reported that even though results did not reach statistical significance, vertical
ground-reactive forces after arthroplasty produced a more symmetric pattern closer
to that of healthy controls, compared to arthrodesis (Flavin et al. 2013). Singer et al.
reported similar results with total ankle arthroplasty patients more closely replicating
healthy controls compared to tibiotalar arthrodesis patients in ankle power (watts),
ankle extension moment (Nm/kg), and ankle moment at heel rise (Nm/kg) (Singer
et al. 2013).
Case Report
A 77-year-old male presented to the office after sustaining a right ankle fracture
approximately 20 years prior. He had progressive ankle pain and disability with
ambulation and activities. His preoperative imaging studies are listed in Fig. 1a,
b. After failing nonoperative treatment modalities including bracing, injections,
physical therapy, and activity modifications, the patient decided to undergo total
ankle arthroplasty with a mobile-bearing implant. Figure 2a, b demonstrates the
postoperative appearance of the ankle. Marked improvements were seen in patient-
reported outcomes at 2-year status post ankle replacement (Table 1). Improvements
in numerous parameters of gait were achieved. Temporal-spatial data is listed in
Table 2. Sagittal range of motion is plotted in Fig. 1. While the patient did not have
normalization of his gait, there was significant improvement in all parameters of gait
and marked improvement in his subjective outcome measures.
Conclusion/Summary
Multiple studies have demonstrated objective functional improvement after total

ankle arthroplasty, as measured by multiple parameters of gait analysis. This dem-
onstration of improvement in patients compared to their preoperative function did
not achieve improvement to a level comparable with normal controls. Interestingly,
multiple studies have shown that the similarities of improvement following tibiotalar
arthrodesis were far greater than the differences following total ankle arthroplasty,
and the loss of sagittal plane motion after arthrodesis can be little or none, at least in
patients with the most severe end-stage arthritis.
Fig. 1 Isolated post-traumatic end-stage ankle arthritis
Fig. 2 Postoperative radiographs 24 months after undergoing total ankle arthroplasty
Table 1 Patient reported Outcome Preoperative Postoperative

outcome measures
SF36 mental 30.2 63.2
comparing preoperative and
24 month postoperative SF36 physical 36.9 52.3
scores for the patient in the AOFAS hindfoot score 41 83
case report Visual analogue pain score 8 0.5
A fundamental limitation of biomechanical evaluation of gait is the inability of

any study or gait model to distinguish the relative contributions of pain relief versus
biomechanical change to the improvements in gait, which may account for the
similarities in improvements between arthrodesis and arthroplasty patients.
Much remains to be illuminated in terms of understanding the biomechanical
effects of TAA and especially of separating the effects of surgery from those of
arthritis, stiffness, and deformity in the adjacent joints of the hindfoot and midfoot.
10 J.M. Kane et al.
Table 2 Temporal spatial data from three-dimensional gait analysis of the patient in the case report
Pre-Op temporal spatial data Post-Op temporal spatial data
Cadence steps/ 74.75925066 Cadence steps/ 105.8928237
min min
Right step Right cm 29.46351107 Right step Right cm 60.80176326
length length
Left step Left cm 35.09333508 Left step Left cm 61.22609393
length length
Walking m/s 0.39657854 Walking m/s 1.073929219
speed speed
References
Barnett CH, Napier JR (1952) The axis of rotation at the ankle joint in man. Its influence upon the
form of the talus and the mobility of the fibula. J Anat 86(Pt 1):1–9
Beischer AD, Brodsky JW, Pollo FE et al (1999) Functional outcome and gait analysis after triple or
double arthrodesis. Foot Ankle Int 20:545–553
Bolton-Maggs BG, Sudlow RA, Freeman MA (1985) Total ankle arthroplasty. A long-term review
of the London Hospital experience. J Bone Joint Surg (Br) 67:785–790
Brodsky JW, Polo FE, Coleman SC et al (2011) Changes in gait following the Scandinavian total
ankle replacement. J Bone Joint Surg Am 93:1890–1896
Brodsky JW, Coleman SC, Smith S et al (2013) Hindfoot motion following STAR total ankle
arthroplasty: a multisegment foot model gait study. Foot Ankle Int 34(11):1479–1485
Buchner M, Sabo D (2003) Ankle fusion attributable to posttraumatic arthrosis: a long-term
followup of 48 patients. Clin Orthop Relat Res 406:155–164
Buck P, Morrey BF, Chao EY (1987) The optimum position of arthrodesis of the ankle. A gait study
of the knee and ankle. J Bone Joint Surg 69-A:1052–1062
Choi JH, Coleman SC, Tenenbaum S et al (2013) Prospective study of the effect on gait of a
two-component total ankle replacement. Foot Ankle Int 34(11):1472–1478
Coester LM, Saltzman CL, Leupold J et al (2001) Long-term results following ankle arthrodesis for
post-traumatic arthritis. J Bone Joint Surg 83-A:219–228
Deland JT, Morris GD, Sung IH (2000) Biomechanics of the ankle joint. A perspective on total
ankle replacement. Foot Ankle Clin 5:747–759
Demottaz JD, Mazur JM, Thomas WH et al (1979) Clinical study of total ankle replacement with
gait analysis. A preliminary report. J Bone Joint Surg Am 61(7):976–988
Dini AA, Bassett FH (1980) Evaluation of early results of Smith total ankle replacement. Clin
Orthop 146:228–230
Doets HC, van Middelkoop M, Houdijk H et al (2007) Gait analysis after successful mobile bearing
total ankle replacement. Foot Ankle Int 28(3):313–322
Dyrby C, Chou LB, Andriacchi TP et al (2004) Functional evaluation of the Scandinavian total
ankle replacement. Foot Ankle Int 25(6):377–381
Espinoza N, Walti M, Favre P et al (2010) Misalignment of total ankle components can induce high
joint contact pressures. J Bone Joint Surg Am 92(5):1179–1187
Flavin R, Coleman SC, Tenenbaum S et al (2013) Comparison of gait after total ankle arthroplasty
and ankle arthrodesis. Foot Ankle Int 34(1):1340–1348
Fuchs S, Sandmann C, Skwara A, Chylarecki C (2003) Quality of life 20 years after arthrodesis of
the ankle. A study of adjacent joints. J Bone Joint Surg 85-B:994–998
Gill LH (2002) Principles of joint arthroplasty as applied to the ankle. In: American Academy of
Orthopaedic Surgeons (ed) Rosemont (AAOS) Instructional course lectures. American Acad-
emy of Orthopaedic Surgeons, Rosemont, pp 117–128
Glazebrook M, Daniels T, Younger A et al (2008) Comparison of health-related quality of life

between patients with end-stage ankle and hip arthrosis. J Bone Joint Surg Am 90(3):499–505
Gougoulias N, Khanna A, Maffulli N (2010) How successful are current ankle replacements?: a
systematic review of the literature. Clin Orthop 468:199–208
Haddad SL, Coetzee JC, Estok R et al (2007a) Intermediate and long-term outcomes of total ankle
arthroplasty and arthrodesis: a systematic review of the literature. J Bone Joint Surg Am
89:1899–1905
Haddad SL, Coetzee JC, Estok R et al (2007b) Intermediate and long-term outcomes of total ankle
arthroplasty and ankle arthrodesis. J Bone Joint Surg Am 89:1899–1905
Hahn ME, Wright ES, Segal AD et al (2012) Comparative gait analysis of ankle arthrodesis and
arthroplasty: initial findings of a prospective study. Foot Ankle Int 33(4):282–289
Hintermann B (2005) Total ankle arthroplasty: historical overview, current concepts, and future
perspectives. SpringerWien, New York, pp 35–49
Houdijk H, Doets HC, van Middelkoop M et al (2008) Joint stiffness of the ankle during walking
after successful mobile-bearing total ankle replacement. Gait Posture 27:115–119
Hurwitz DE, Ryals AB, Case JP et al (2002) The knee adduction moment during gait in subjects
with knee osteoarthritis is more closely correlated with static alignment than radiographic
disease severity, toe out angle and pain. J Orthop Res 20:101–107
Ilgin D, Ozalevli S, Kilinc O et al (2011) Gait speed as a functional capacity indicator in patients
with chronic obstructive pulmonary disease. Ann Thorac Med 6:141–146
Inam S, Vucic S, Brodaty NE et al (2010) The 1-metre gait speed as a functional biomarker in
amyotrophic lateral sclerosis. Amyotroph Lateral Scler 11:558–561
Ingrosso S, Benedetti MG, Leardini A et al (2009) Gait analysis in patients operated with a novel
total ankle prosthesis. Gait Posture 30:132–137
Inman VT (1991) The joints of the ankle. In: Stiehl JB (ed) Biomechanics of the ankle joint, 2nd
edn. Williams & Wilkins, Baltimore, pp 31–74
Kakkar R, Siddique MS (2011) Stresses in the ankle joint and total ankle replacement design. Foot
Ankle Surg 17(2):58–63
Kawamura K, Tokuhiro A, Takechi H (1991) Gait analysis of slope walking: a study on step length,
stride width, time factors and deviation in the center of pressure. Acta Med Okayama
45:179–184
Khazzam M, Long JT, Marks RM et al (2006) Preoperative gait characterization of patients with
ankle arthrosis. Gait Posture 24:85–93
Kirtley C (2006) Clinical gait analysis: theory and practice. Elsevier, New York
Kitaoka HB, Patzer GL (1994) Clinical results of the Mayo total ankle arthroplasty. J Bone Joint
Surg Am 76:974–979
Kofoed H, Sorensen TS (1998) Ankle arthroplasty for rheumatoid arthritis and osteoarthritis:
prospective long-term study of cemented replacements. J Bone Joint Surg 80-B:328–332
Leardini A, OConnor JJ, Giannini S (2014) Biomechanics of the natural, arthritic, and replaced
human ankle joint. J Foot Ankle Res 7:8
Ledoux WR, Rohr ES, Ching RP et al (2006) Effects of foot shape on the three-dimensional
position of foot bones. J Orthop Res 24:2176–2186
Lewis G (1994) The ankle joint prosthetic replacement: clinical performance and research chal-
lenges. Foot Ankle Int 15(9):471–476
Mayich DJ, Novak A, Vena D et al (2014) Gait analysis in foot and ankle surgery – topical review,
Part 1: principles and uses of gait analysis. Foot Ankle Int 35(1):80–90
McIntosh AS, Beattyy KT, Dwan LN et al (2006) Gait dynamics on an inclined walkway. J
Biomech 39:2491–2502
Mundermann A, Dyrby CO, Hurwitz DE et al (2004) Potential strategies to reduce medial
compartment loading in patients with knee osteoarthritis of varying severity: reduced walking
speed. Arthritis Rheum 50:1172–1178
Novak AC, Mayich DJ, Perry SD et al (2014) Gait analysis for foot and ankle surgeons – topical
review, part 2: approaches to multisegment modeling of the foot. Foot Ankle Int 35(2):178–191
12 J.M. Kane et al.
Piriou P, Culpan P, Mullins M, et al (2008) Ankle replacement versus arthrodesis: a comparative gait
analysis study. Foot Ankle Int 29(1): 3–9
Potter JM, Evans AL, Duncan G (1995) Gait speed and activities of daily living function in geriatric
patients. Arch Phys Med Rehabil 76:997–999
Queen RM, Nunley JA (2010) The effect of footwear on preoperative gait mechanics in a group of
total ankle replacement patients. J Surg Orthop Adv 19:170–173
Queen RM, DeBiasio JC, Butler RJ et al (2012) Changes in pain, function, and gait mechanics two
years following total ankle arthroplasty performed with two modern fixed-bearing prostheses.
Foot Ankle Int 33(7):535–542
Queen RM, Sparling TL, Butler RJ et al (2014a) Patient-reported outcomes, function, and gait
mechanics after fixed and mobile-bearing total ankle replacement. J Bone Joint Surg Am
96:987–993
Queen RM, Butler RJ, Adams SB Jr et al (2014b) Bilateral differences in gait mechanics following
total ankle replacement: a two year longitudinal study. Clin Biomech 29:418–422
Reggiani B, Leardini A, Corazza F et al (2006) Finite element analysis of a total ankle replacement
during the stance phase of gait. J Biomech 39(8):1435–1443
Saltzman CL, Salamon ML, Blanchard GM et al (2005) Epidemiology of ankle arthritis. Iowa
Orthop J 25:44–46
Singer S, Klejman S, Pinsker E et al (2013) Ankle arthroplasty and ankle arthrodesis: gait analysis
compared with normal controls. J Bone Joint Surg Am 95(e191):1–10
Stauffer RN (1979) Total joint arthroplasty. The ankle. Mayo Clin Proc 54:570–575
Stauffer RN, Seagal NM (1981) Total ankle arthroplasty. Four years experience. Clin Orthop.
160:217–221
Stauffer RN, Chao EYS, Brewster RC (1977) Force and motion analysis of the normal, diseased,
and prosthetic ankle joint. Clin Orthop 127:189–196
Tooms RE (1987) Arthroplasty of ankle and knee. In: Crenshaw AH (ed) Campbell’s operative
orthopedics. C.V. Mosby Company, St. Louis, pp 1145–1150
Valderrabano V, Nigg BM, von Tscharner V et al (2007) Gait analysis in ankle osteoarthritis and
total ankle replacement. Clin Biomech 22:894–904
Shoulder Joint Replacement and Upper
Extremity Activities of Daily Living
Hendrik Bruttel, David M. Spranz, Jan M. Eckerle, and

Michael W. Maier
Abstract
3D motion analysis is mainly used for lower extremity, especially gait analysis.
Upper extremity is a comparably new application and less standardized. The large
range of motion and complex anatomy of the shoulder are challenging. However
3D motion analysis can provide deeper insight in the coordinated motion of the
upper extremity as multiple joints can be monitored over the whole movement.
Anatomical and reverse total shoulder arthroplasty are effective surgical treat-
ment options for patients with osteoarthritis, rheumatic arthritis, cuff tear arthrop-
athy, and traumatic shoulder injuries. 3D motion analysis helps to understand how
prostheses influence movement patterns and improve shoulder function. Thus it
may help to improve design of future prostheses and could potentially become an
objective tool for examination of patients pre- and postoperatively. Although
there are no standardized protocols yet, range of motion tasks and activities of
daily living have been used in many protocols and proved to be effective tasks for
analysis of shoulder function.
Keywords
Shoulder kinematics • Shoulder arthroplasty • Shoulder joint replacement •
Motion analysis • Upper extremity • Activities of daily living • Range of motion •
Proprioception • Upper extremity model
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Applications of 3D Motion Analysis in Patients with Shoulder Joint Replacements . . . . . . . . . . . 9
H. Bruttel • D.M. Spranz • J.M. Eckerle • M.W. Maier (*)

Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany
e-mail: hendrik.bruttel@gmail.com; David.Spranz@med.uni-heidelberg.de; JanMarc.
Eckerle@med.uni-heidelberg.de; michael.maier@med.uni-heidelberg.de

DOI 10.1007/978-3-319-30808-1_84-2
2 H. Bruttel et al.
Range of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Activities of Daily Living . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Angle Reproduction Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Introduction
The shoulder consists of the humerus and the shoulder girdle, which is made up of
the scapula and clavicle. Movement of the shoulder is the combined result of four
different joints. The clavicle connects the shoulder to the thorax via the
sternoclavicular joint and stabilizes scapular movements through the
acromioclavicular joint. The scapula is connected to the thorax by muscles only.
This allows rotational and translational movement and is called scapulothoracic
joint. The glenohumeral joint is a ball-and-socket joint between the humeral head
and the glenoid of the scapula. The tendons of the rotator cuff formed by the
subscapularis, supraspinatus, infraspinatus, and teres minor stabilize the joint. The
glenoid surface is small compared to the humeral head and thus allows a large range
of motion (ROM), which is further extended by the shoulder girdle. Fig. 1 gives an
overview of the shoulder anatomy.
History of shoulder joint replacement goes back to 1893 when French surgeon
E. J. Péan implanted the first shoulder prosthesis (Lugli 1978). In 1955 C. S. Neer
presented the first modern design for a stemmed humeral head prosthesis used in
trauma patients (Neer 1955). Today different designs are available for treatment of
degenerative and traumatic injuries of the glenohumeral joint.
Design of anatomical hemi- and total shoulder arthroplasty (TSA) aims to imitate
the original anatomical situation. A ball-shaped prosthesis replaces the humeral head
and was traditionally fixated in the diaphysis. Newer stemless models are fixated in
the metaphysis and lead to smaller surgical trauma, shorter procedure time, and
reduced loss of bone material, which facilitates revision surgery (Huguet et al. 2010;
Berth and Pap 2013; Razmjou et al. 2013; Bell and Coghlan 2014). In TSA the
glenoid is replaced by a socket-shaped prosthesis in contrast to hemiarthroplasty
where the glenoid is not replaced. Patients with primary or secondary osteoarthritis
and rheumatic arthritis can profit from this type of shoulder replacement. Good result
for reduction of pain and improvement of range of motion are reported (Deshmukh
et al. 2005; Radnay et al. 2007; Kasten et al. 2010; Raiss et al. 2012; Sandow et al.
2013). Special implant designs for traumatic shoulder injuries are also available
(Aaron et al. 2013).
Reverse shoulder arthroplasty (RSA) uses a different biomechanical design in
which the humeral head is replaced by a socket-shaped component and a ball-shaped
component is fixated to the glenoid (Grammont and Baulot 1993). This type of
prosthesis is traditionally used in patients with rotator cuff tear arthropathy, charac-
terized by rotator cuff insufficiency based on a massive irreparable rotator cuff tear,
diminished acromiohumeral distance, and secondary developed arthritic changes of
Shoulder Joint Replacement and Upper Extremity Activities of Daily Living 3
Fig. 1 Anatomy of the shoulder
Fig. 2 Different types of shoulder joint replacements: conventional stemmed TSA (a), stemless
TSA (b), and RSA (c)
the glenohumeral joint (Ecklund et al. 2007). Because of the good functional
outcome, today the RSA is also used in revision cases and in patients with proximal
humerus fractures. RSA caudalizes and medializes the glenohumeral center of
rotation (Jobin et al. 2012; Rettig et al. 2013), which improves the lever arm of the
deltoid muscle and increases muscle fiber recruitment of the anterior and posterior
deltoid to compensate for a deficient rotator cuff. Fig. 2 shows different types of
shoulder joint replacements.
4 H. Bruttel et al.
3D motion analysis is mainly used for lower extremity, especially gait analysis.
Upper extremity is a comparably new application and less standardized. The large
ROM and complex anatomy of the shoulder are challenging. However 3D motion
analysis can provide deeper insight in the coordinated motion of the upper extremity
as multiple joints can be monitored over the whole movement. In patients with
shoulder joint replacement, it can help to understand how the prosthesis implantation
influences the movement patterns of the upper extremity and thus may help to
improve design of future prostheses. It could potentially become an objective tool
for examination of pre- and postoperative patients.
State of the Art
Different systems for motion analysis are available and can be used in patients with
shoulder joint replacements. Optoelectronic systems use a set of cameras to track
skin-mounted markers. These markers can either actively emit light via LEDs or
passively reflect infrared light emitted by strobes on the cameras. Three-dimensional
positions of markers in sight of more than one camera can be calculated via
triangulation. These systems are often used in gait analysis as they can be scaled
for tracking in large volumes. Additionally passive systems allow the patient to be
unwired. However in upper extremity motion analysis, the subject is usually stand-
ing or seated, and large volumes for tracking are not necessary. Another concept is
the use of electromagnetic tracking devices that generate a magnetic field in which
sensors can track their position. These systems can be used for upper extremity
motion analysis (Meskers et al. 1998b, 1999; Hannah and Scibek 2015), and unlike
optoelectronic systems, they do not require visual sight of the sensors.
Biomechanical models divide the body into segments that are connected by
joints. Three non-collinear points per segment are necessary to track all six degrees
of freedom. Anatomical landmarks are used to define segments in a biomechanically
meaningful and comparable way (see chapter “▶ Upper Extremity Models for
Clinical Movement Analysis” for more details on upper extremity modeling).
Their position can be tracked either directly via mounting markers on the skin
over the respective landmark or indirectly via clusters of markers or electromagnetic
sensors. Direct tracking reduces calibration recordings and therefore unburdens the
motion analysis protocol. Greater distances between tracked points reduce influence
of artifacts. For indirect tracking a cluster or sensor is used to track segment
kinematics. A technical coordinate system can be created for each cluster or sensor
in which positions of the anatomical landmarks are digitized by recording reference
frames using a pointing device (see Fig. 3) attached with markers or sensors (van
Andel et al. 2008). Advantages of indirect tracking are reduced number of sensors
and markers and therefore faster patient preparation. In case an optoelectronic
system is used, they can also provide a workaround for problems with visual sight.
In most studies skin-mounted markers or sensors are used. They provide a
non-invasive way to record kinematics but are prone to soft tissue artifacts (STA).
STA arise due to skin and muscular movement during motion, which may
Fig. 3 Pointing device (a) and a cluster of four reflective markers (b)
temporarily move the marker or sensor away from its designated position. Choosing
anatomical landmarks with thin soft tissue coverage reduces STA. Some anatomical
landmarks, e.g., on the scapula, cannot be tracked directly to sufficient degree and
therefore require indirect techniques (Meskers et al. 2007; van Andel et al. 2009;
Brochard et al. 2011; Lempereur et al. 2014; Warner et al. 2015). Clusters or
electromagnetic sensors can also be attached to bone pins in order to eliminate
STA. But due to their invasive nature, the application of this method is limited to
small study groups.
Calculating Joint Kinematics Joint kinematics can be calculated from kinematic

data of two adjacent segments. Different methods to calculate angle time series to
describe rotation can be used. Translational movements are usually not
accounted for.
Euler/Cardan angles are widely used in motion analysis, as they are mathemat-
ically well defined and can be calculated to yield clinically meaningful angles
(Grood and Suntay 1983). Three angles are used to describe three subsequent
rotations around either two (Euler angles) or three axes (Cardan angles) of a
coordinate system in order to align them. The resulting angles are dependent on
the order of rotations as subsequent rotations are influenced by the rotations before.
Euler and Cardan angles therefore provide six different solutions each to achieve the
same end position. Since three-dimensional motion does not occur sequentially but
at once, it is not always easy to decide, which sequence is most useful. It is therefore
necessary to agree on a rotation sequence in order to gain comparable data (Phadke
et al. 2011). Another problem is that two points remain undefined (comparable to the
poles of the earth). Near these singularities angles change faster as the spatial
distance between degrees gets smaller (like longitudes near the poles of the earth),
and hence measurement errors increase. Most body joints have a restricted ROM,
6 H. Bruttel et al.
e.g., the hip joint, or are essentially constraint to one plane, e.g., the knee joint. In
these cases the singularities can be located outside the physiological ROM and
therefore be omitted. However for the shoulder joint, the singularities are within
the physiological ROM. Commonly Euler angles are used, and the singularities are
thus located at 0 and 180 elevation (Doorenbosch et al. 2003; Wu et al. 2005). For
many movements this can be sufficient, but it must be kept in mind for trials
involving the neutral pose and maximum elevation.
Another way to express joint kinematics is in terms of projection angles. Their
advantage is that they are independent of a rotation sequence. However if a vector
is orthogonal to one of the projection planes, the corresponding projection angle is
undefined, comparable to the singularities of Euler/Cardan angles. Another prob-
lem is that projection angles always yield two angles: clockwise and counterclock-
wise rotated. It is practicable to use the angles between 180 and 180 to reflect
the motion according to the neutral zero method. However especially when motion
occurs outside the projection planes, values above 180 can be reached, e.g., 200
flexion along with some initial abduction. This would lead to a 360 jump from
+180 to 180 in the angle time series and thus contradict intuitive expectation,
although mathematically correct. Furthermore independency of the angles leads to
the problem that movement reflects in more than one angle time series, e.g., a
flexion movement that is slightly out of sagittal plane will reflect in both the
flexion and abduction angle. Both Euler/Cardan and projection angles have in
common that a three-dimensional movement is described by a set of
two-dimensional angles. While easier to understand, they come with the
abovementioned problems. Euler/Cardan angles are useful to describe complex
movements in an easily understandable way. A projection angle can be especially
useful when movement within a plane is measured, e.g., maximum extension to
maximum flexion, as they are independent of the other rotations and are very
similar to the SFTR system used in clinical exams (American Academy of
Orthopaedic Surgeons 1965).
Another approach is the use of helical angles. Helical angles use an axis to
describe rotation around and translation along this axis. The unit vector of the axis
can be multiplied with the rotation angle resulting in three helical angles or an
attitude vector (Woltring 1994). Helical angles do not have undefined points, which
is useful for upper extremity motion analysis in order to omit the problems arising
from the large ROM as mentioned above. However their disadvantage is that they
are not intuitively interpretable. Their use for clinical motion analysis is therefore
limited, as most clinicians cannot relate to the data.
International Society of Biomechanics’ Recommendations In 2005 the Interna-

tional Society of Biomechanics (ISB) published recommendations for motion anal-
ysis of the upper extremity (Wu et al. 2005). Anatomical landmarks to define
coordinate systems for six segments (thorax, clavicle, scapula, humerus, forearm,
and hand) are proposed. Rotation sequences to calculate comparable Euler/Cardan
angles for the respective joints are proposed. The ISB intended to standardize motion
Table 1 Comparison of anatomical landmarks used in ISB recommendations and HUX model.
Short names are taken from the original publication. Hand segment definitions are not included.
(Wu et al. 2005; Rettig et al. 2009)
Segment ISB HUX
Thorax Incisura jugularis (IJ) Incisura jugularis (CLAV)
Proc. xiphoideus (PX) Proc. xiphoideus (STRN)
Proc. spinosus 7th cervical vertebra Proc. spinosus 7th cervical vertebra (C7)
(C7) Proc. spinosus 10th thoracic vertebra
Proc. spinosus 8th thoracic vertebra (T10)
(T8)
Clavicle Sternoclavicular joint (SC) Acromion (SHO)
Acromioclavicular joint (AC)
Scapula Angulus acromialis (AA)
Trigonum spinae scapulae (TS)
Angulus inferior (AI)
Processus coracoideus (PC)a
Humerus Glenohumeral center of rotation (GH) Tuberositas deltoidea (HUM)
Lateral epicondyle (EL) Ulna distal of olecranon (ELB-ELBW)
Medial epicondyle (EM)
Forearm Radial styloid (RS) Radial styloid (RAD)
Ulnar styloid (US) Ulnar styloid (ULN)
a
PC is not needed for segment definition but for estimating GH according to Meskers et al. (1998a)
analysis of the upper extremity and their recommendations are applied in the
majority of upper extremity studies. The proposed anatomical landmarks have
relatively little soft tissue coverage and are thus easy to palpate (see Table 1).
The location of the glenohumeral center of rotation (GH), which is needed to
define the humerus segment, is the only landmark not directly palpable. Its position
has to be calculated using a calibration trial. The position is then usually linked to the
technical coordinate system of a marker cluster or electromagnetic sensor on the
humerus. The ISB proposes two different methods to locate GH. GH can either be
derived from adjacent anatomical landmarks using linear regression formulas
(Meskers et al. 1998a) or via a functional method. A dynamic calibration trial is
used to calculate the pivot point of instantaneous helical axes (Woltring 1990;
Stokdijk et al. 2000). The latter method should be preferred (Stokdijk et al. 2000;
Wu et al. 2005).
The anatomical landmarks used for definition of the scapula cannot be tracked
directly as the scapula moves beneath the skin, which leads to unacceptable errors
(see Fig. 4). A cluster or electromagnetic sensor on the acromion is often used for
scapula tracking (Meskers et al. 2007; van Andel et al. 2009; Brochard et al. 2011;
Lempereur et al. 2014; Warner et al. 2015).
Heidelberg Upper Extremity Model The Heidelberg Upper Extremity Model

(HUX) proposed by Rettig et al. (2009) is another biomechanical model that has
been used for shoulder joint replacement studies. It uses a smaller number of
8 H. Bruttel et al.
Fig. 4 Difference between

marker position and palpated
location (marked with X) of
the angulus inferior of the
scapula at 90 abduction
anatomical landmarks that are all (besides GH) directly traceable. This eliminates
the need to record reference frames for clusters. It uses a least squares method to
determine GH and the elbow joint center (Gamage and Lasenby 2002). Angles
are calculated as projection angles and are thus comparable to clinical measure-
ments using a goniometer. It proved to be reliable for large ROM and for
internal/external rotation in the glenohumeral joint. The latter is described as
problematic in humerus cluster-based approaches (Cutti et al. 2005). Scapula
motion is not regarded in the model. Only humerothoracical angles are calculated
for the shoulder. Marker positions for the HUX model are summarized in
Table 1.
Methods for Joint Center Localization Both models use GH for defining the
humerus segment. Calculation of clinically meaningful humerothoracical angles is
not possible without this landmark. Different methods have been proposed, since it
is not palpable. Regression methods rely on empirical data gathered in cadaver
studies. This empirical data is scaled to the subject using regression formulas
(Meskers et al. 1998a). Shoulder joint replacements alter shoulder anatomy and
move GH. Use of these methods in this field is questionable since there is no
empirical data available for shoulder replacement patients. Another way to calculate
position of GH is using functional methods, which have been found to be superior
to empirical methods (Stokdijk et al. 2000). Several methods to estimate center of
rotation have been proposed (Woltring 1990; Halvorsen et al. 1999; Leardini et al.
1999; Halvorsen 2003; Schwartz and Rozumalski 2005; Ehrig et al. 2006) and can
be used. While ISB recommends use of the instantaneous helical axes method
(Woltring 1990; Stokdijk et al. 2000; Wu et al. 2005), the least squares method
by Gamage and Lasenby (2002) used in the HUX model might be preferable
(Lempereur et al. 2010).
Applications of 3D Motion Analysis in Patients with Shoulder

Joint Replacements
Until now 3D motion analysis is not routinely used in patients with shoulder joint
replacement pre- and postoperatively. Although development of standardized pro-
tocols is under discussion (Kontaxis et al. 2009) and several proposals have been
made (van Andel et al. 2008; Vanezis et al. 2015), there is no commonly used
protocol yet. Many study protocols use either ROM tasks or a set of activities of
daily living (ADL) or both to analyze shoulder function. Furthermore proprioception
has been studied using active angle reproduction tests (AAR). The following section
intends to give an overview over typical applications for 3D motion analysis in
patients with shoulder joint replacements.
Range of Motion
Maximum shoulder ROM in frontal, sagittal, and transversal plane is used as a

quantitative measurement for shoulder functionality. In clinical examination ROM is
measured using a goniometer. Two types of ROM measurements can be differenti-
ated: active and passive ROM. To test active ROM, the patient is asked to move their
arm from a neutral position (usually standing with arm hanging loosely besides the
body) to the maximum achievable position. This is done for flexion/extension,
abduction/adduction, and internal/external rotation. For internal/external rotation,
the elbow is usually flexed by 90 in order to have better visual representation of the
movement. The other movements are assessed with the elbow at 0 flexion. Passive
ROM is tested likewise but the examiner moves the patient’s arm. Difference or
equality of active and passive ROM is typical for certain shoulder diseases and can
therefore be important for diagnosis.
Many study protocols include ROM tasks to assess shoulder function. Usually
maximum elevation in frontal (flexion) and sagittal (abduction) plane is tested.
Further tasks differ and include maximum internal/external rotation, elevation in
scapular plane (40–45 anterior of frontal plane), extension, and adduction. The
patient is asked to move the arm to the maximum achievable position starting from a
neutral position. These tasks allow assessment of maximum active ROM in the
tested planes and therefore give a reference value for evaluation of further tasks.
For both TSA and RSA, compromised ROM has to be expected (Veeger et al.
2006; Bergmann et al. 2008; Alta et al. 2014). A larger maximum ROM can be
expected in patients treated with TSA compared to patients with RSA (Alta et al.
2014). But Alta et al. (2011) could show that ROM in RSA patients also depends on
whether it is used as revision or primary prosthesis. Primary RSA yields greater
postoperative ROM than revision RSA.
However maximum ROM can be evaluated sufficiently with standard clinical
means, i.e., with a goniometer, and 3D motion analysis does not necessarily provide
greater accuracy (Rettig et al. 2009). The advantage of motion analysis lies in its
ability to measure participating joint’s kinematics throughout the movement. This
10 H. Bruttel et al.
Scpula Protraction Scapula Rotation

60 40
50 30
40 20
Med [deg] Lat

Ret [deg] Pro
30 10
20 0
10 -10
0 -20
40 60 80 40 60 80
Dep [deg] Ele Dep [deg] Ele
Scapula Tilting
10
-10
Post [deg] Ant
-20
-30
-40
-50
40 60 80
Dep [deg] Ele
Fig. 5 Scapula protraction, rotation, and tilting plotted against shoulder elevation during an
abduction task
can help to gain an understanding of the coordination of the participating joints. The
scapulohumeral rhythm (SRH) is the relation of scapula and glenohumeral motion
(Inman et al. 1944). It is often used as a quantitative description for shoulder
coordination and can be used to assess compensation for loss of glenohumeral
ROM. Accurate recording of scapula kinematics is difficult and subject to research.
Generally contribution of the glenohumeral joint, i.e., the prosthesis, to elevation
seems to be reduced in both TSA (de Toledo et al. 2012; Alta et al. 2014) and RSA
patients (de Toledo et al. 2012; Kwon et al. 2012; Alta et al. 2014; Walker et al.
2015) compared to healthy subjects. In contrast some authors report close to normal
values (Bergmann et al. 2008; Braman et al. 2010).
Quantitative measurements gained by ROM tasks are useful for clinical evalua-
tion. Maximum ROM values build a bridge between 3D motion analysis results and
regular clinical examination. SRH provides a numeric quantification of shoulder
coordination that can be used to monitor degree of success in restoring normal
shoulder coordination. Pre- and postoperative data can be helpful to evaluate differ-
ent surgical treatments and improve future prosthetic designs. Advantage of ROM
tasks is that they can be standardized rather easily. The shoulder girdle kinematics
can be further described with angle/angle plots relating the scapula and clavicle joint
kinematics to shoulder elevation (see Fig. 5).
Activities of Daily Living
3D motion analysis is mainly used for lower extremity especially gait analysis.
Walking and standing are the main activities. Defining meaningful standardized
tasks for upper extremity is more complicated, because there is a great number of
tasks that can be performed and analyzed. Ability to handle everyday life has priority
for patients with shoulder joint replacements. ADL are basic tasks of everyday life,
e.g., drinking, eating, body hygiene, etc. Clinical scores for evaluation of shoulder
function such as Constant score (Constant and Murley 1987), DASH (Hudak et al.
1996) and ASES (Michener et al. 2002) use ADL for assessment. However patient
self-reported ability to perform certain ADL can differ greatly from examination
results (Sager et al. 1992). Therefore an objective evaluation method is necessary.
While ROM tasks can give a first impression of overall shoulder function and
insight into coordination of the shoulder, their meaning for everyday life is limited.
Many ADL do not necessarily need full-shoulder ROM, and different strategies to
accomplish the same task can be used. For this reason ADL are often used in motion
analysis protocols to assess functionality of upper extremity in a more function-
oriented way. ADL differ greatly in terms of complexity and ROM used. Selected
ADL should reflect different types of movement to cover most daily needs. They
often concentrate on the patient’s body since position of external objects may be
altered to meet patient’s abilities and are not standardized. Internal rotation is needed
to reach lower body parts, especially the lower back in order to perform perianal
care. External rotation is involved in many tasks involving upper body parts, such as
hair combing. Adduction is needed for reaching the contralateral body side, e.g., for
washing the armpit. The mentioned ADL examples are used in many study pro-
tocols. Additionally the ability to reach high objects, such as collecting a book from a
shelf, is often tested, which mainly needs a large degree of forward flexion. Other
ADL used include eating, drinking, and many more. A useful protocol should select
few ADL in order to keep examination and post-processing time low. Many ADL
consist of comparable shoulder motion and may not give further insight, e.g.,
reaching contralateral shoulder and eating use comparable ROM (Magermans
et al. 2005).
Simplified tasks that just involve touching the respective area are sometimes used
(van Andel et al. 2008). While this can facilitate instructions and analysis, their
meaningfulness for ability to perform ADL is questionable as they are often more
easy than the actual task. The patient can be given objects, e.g., a washcloth, to make
movements more realistic (Maier et al. 2014a).
Another controversial point is how patients should be instructed. Execution of
ADL differs from person to person, e.g., hair combing depends on hairstyle and
length. But also many other tasks depend on preference and routine. A high degree
of standardization is preferable to better differentiate pathological and physiological
movement abilities. To gather comparable data, standardized instructions are useful
(van Andel et al. 2008). Visual demonstrations by the examiner or by video help
patients to perform ADL in the desired way. However depending on the intention of
the study, freedom to choose own way of performance may be given (Veeger et al.
2006). While the latter method will most certainly result in more natural movement
patterns, especially, analysis of kinematic time series might be difficult.
Another factor to be considered is handedness. Some tasks, e.g., washing the
armpit, are performed with both sides in everyday life. But the dominant hand
usually does other tasks exclusively, e.g., perianal care, although this can be different
in patients with shoulder diseases. Differences between dominant and non-dominant
side in scapula motion could even be demonstrated for ROM tasks in healthy
subjects (Matsuki et al. 2011). Standardized instructions might reduce influence of
handedness to some degree, but it should still be considered.
Different quantitative measurements can be gathered from ADL tasks. The
ROM used for the ADL (AROM), i.e., the minimum and maximum of the
respective angle over the whole movement, can be calculated. AROM can be
compared to healthy individuals (Veeger et al. 2006; Kasten et al. 2010) pre- and
postoperatively (Kasten et al. 2010; Maier et al. 2014a, b) and to maximum ROM
values gathered from ROM tasks (Kasten et al. 2010; Maier et al. 2014b). AROM
compared to healthy individuals may be a better estimate of patient’s restrictions in
everyday life than maximum ROM. Amount of reduction of AROM compared to
healthy individuals depends on the ADL and angle analyzed (Maier et al. 2014b).
The effect of the implanted prosthesis on the capacity to handle everyday life can
be quantified by comparing pre- and postoperative AROM. Both RSA and TSA
have been shown to improve AROM (Kasten et al. 2010; Maier et al. 2014a).
Rehabilitation is time-consuming, and even 6 months after surgery, there is still
improvement to be expected (Maier et al. 2014b). The comparison of AROM to
ROM tasks can be helpful to evaluate the ability to make use of postoperative gain
in ROM. For elevation, patients seem to be unable to make use of their full ROM
(Kasten et al. 2010; Maier et al. 2014a, b). This might indicate room for further
improvement by training.
A deeper understanding of movement patterns can be gained by describing
kinematics with angle time series as done in gait analysis. While AROM may be
normal, abnormal movement patterns can still exist. Data is expressed time normal-
ized as 0–100% of movement in order to compare data of different individuals and to
calculate mean curves for repeated trials (see Fig. 6). However to our knowledge,
only one study by Veeger et al. (2006) provided kinematic time series for ADL in
patients with shoulder joint replacements.
Fig. 6 Averaged shoulder Shoulder Elevation

elevation during a repeated
140
hair-combing task (solid black
line) with standard deviation 120
(dashed line) and reference
data one standard deviation 100
(gray band)
80
[°]
60
40
20
0
0 20 40 60 80 100
% Movement
ADL provide a useful tool to assess patient’s ability to handle everyday life as
they can provide measurements that are closer to patient’s experience. 3D motion
analysis is an objective method for qualitative and quantitative evaluation of ADL
and therefore superior to subjective assessment in clinical scores. However devel-
opment of standardized ADL protocols is necessary in order to gain comparable
results.
Angle Reproduction Tests
Proprioception plays a key role in body motion. The motor cortex needs the joint
positions to plan and coordinate motion. It is provided by the skin, muscle spindles,
Golgi tendon organs, and joint capsules.
Joint position sense can be evaluated using either passive or active angle repro-
duction tests. The subjects arm is guided by the examiner to a certain position, e.g.,
30 abduction, and then returned to the neutral position. The blindfolded subject is
then asked to reproduce the position. This can either be done actively by the subject
him-/herself or passively when the examiner guides the arm, and the subject is asked
to tell the examiner when the reference position is reached. The difference between
the reproduced and the reference position is an indicator for the joint position sense.
In patients with shoulder joint replacements, natural anatomy is altered signifi-
cantly. The impact of shoulder replacement surgery on the body’s proprioception is
therefore a field of interest. When a goniometer is used, angle measurements are
usually either projected to sagittal or frontal plane (Cuomo et al. 2005). 3D motion
analysis can be used to measure the accuracy of reproduction in all dimensions.
Therefore it provides a useful instrument for quantifying proprioception. Depending
on pose, patients with shoulder joint replacements showed equal or progressing
deteriorating proprioception after surgery with TSA performing better than
hemiarthroplasty (Kasten et al. 2009; Maier et al. 2012). No differences could be
found between stemless and conventional stemmed TSA (Maier et al. 2015).
Changes in proprioception and their clinical relevance are not yet completely
understood, and further research is necessary.
Future Directions
Upper extremity motion analysis is mainly used as a tool for research until this day.
3D motion analysis is cost and time intensive, and therefore routine clinical use is
difficult to implement. Optoelectronic and electromagnetic motion analysis is
restricted to specialized laboratories as they require extensive equipment. However
magnetic resonance imaging, computer tomography, and many other technologies
show that complex and expensive technology does not necessarily prohibit clinical
routine use. Motion data objectively gathered by means of 3D motion analysis can
provide deeper understanding of the effects of treatments for orthopedic and trauma
patients. Clinical gait analysis is already used routinely in specialized centers for
certain orthopedic conditions such as cerebral palsy. As mentioned above upper
extremity poses challenges with respect to anatomy and standardized evaluation. For
routine use comparable results and standardized reports, which can be understood
and interpreted by clinicians, are necessary. The need for standardization regarding
modeling (Wu et al. 2005) and protocols (Kontaxis et al. 2009) is commonly
recognized, but still many differences in application exist. Partly this may be due
to ongoing development of new methods, e.g., for joint center calculation
(Lempereur et al. 2010), but clinical gait analysis shows that common ground for
routine use can be found despite arising new technologies. Generally acknowledged
proposals like the ISB recommendations by Wu et al. (2005) are therefore desirable.
New technologies like inertial sensors could lower the barrier for application of
motion analysis as they reduce equipment costs and can be used outside specialized
laboratories. Parel et al. (2012) proposed a protocol fit for routine application based
on this technique.
Another topic to be addressed especially in respect to shoulder joint replacement
research could be enhancement of upper extremity motion analysis with electromy-
ography (EMG). Surface EMG combined with motion analysis could deepen our
understanding of shoulder joint coordination and possible altered muscle function
after joint replacement even further. For example, RSA intentionally alters muscular
coordination of the shoulder enabling the deltoid to replace defective rotator cuffs,
but to our knowledge until this day, only Walker et al. (2014) have used combined
EMG and motion analysis in RSA patients.
References
Aaron D, Parsons BO, Sirveaux F, Flatow EL (2013) Proximal humeral fractures: prosthetic
replacement. Instr Course Lect 62:155–162
Alta TD, Bergmann JH, Veeger DJ, Janssen TW, Burger BJ, Scholtes VA, Willems WJ (2011)
Kinematic and clinical evaluation of shoulder function after primary and revision reverse
shoulder prostheses. J Shoulder Elbow Surg 20(4):564–570. doi:10.1016/j.jse.2010.08.022
Alta TD, de Toledo JM, Veeger HE, Janssen TW, Willems WJ (2014) The active and passive
kinematic difference between primary reverse and total shoulder prostheses. J Shoulder Elbow
Surg 23(9):1395–1402. doi:10.1016/j.jse.2014.01.040
American Academy of Orthopaedic Surgeons (1965) Joint motion: method of measuring and
recording. American Academy of Orthopedic Surgeons, Chicago
Bell SN, Coghlan JA (2014) Short stem shoulder replacement. Int J Shoulder Surg 8(3):72–75.
doi:10.4103/0973-6042.140113
Bergmann JH, de Leeuw M, Janssen TW, Veeger DH, Willems WJ (2008) Contribution of the
reverse endoprosthesis to glenohumeral kinematics. Clin Orthop Relat Res 466(3):594–598.
doi:10.1007/s11999-007-0091-5
Berth A, Pap G (2013) Stemless shoulder prosthesis versus conventional anatomic shoulder
prosthesis in patients with osteoarthritis: a comparison of the functional outcome after a
minimum of two years follow-up. J Orthop Traumatol 14(1):31–37. doi:10.1007/s10195-012-
0216-9
Braman JP, Thomas BM, Laprade RF, Phadke V, Ludewig PM (2010) Three-dimensional in vivo
kinematics of an osteoarthritic shoulder before and after total shoulder arthroplasty. Knee Surg
Sports Traumatol Arthros 18(12):1774–1778. doi:10.1007/s00167-010-1167-4
Brochard S, Lempereur M, Remy-Neris O (2011) Accuracy and reliability of three methods of
recording scapular motion using reflective skin markers. Proc Inst Mech Eng H 225(1):100–105
Constant CR, Murley AH (1987) A clinical method of functional assessment of the shoulder. Clin
Cuomo F, Birdzell MG, Zuckerman JD (2005) The effect of degenerative arthritis and prosthetic
arthroplasty on shoulder proprioception. J Shoulder Elbow Surg 14(4):345–348. doi:10.1016/j.
jse.2004.07.009
Cutti AG, Paolini G, Troncossi M, Cappello A, Davalli A (2005) Soft tissue artefact assessment in
humeral axial rotation. Gait Posture 21(3):341–349. doi:10.1016/j.gaitpost.2004.04.001
de Toledo JM, Loss JF, Janssen TW, van der Scheer JW, Alta TD, Willems WJ, Veeger DH (2012)
Kinematic evaluation of patients with total and reverse shoulder arthroplasty during rehabilita-
tion exercises with different loads. Clin Biomech 27(8):793–800. doi:10.1016/j.
Deshmukh AV, Koris M, Zurakowski D, Thornhill TS (2005) Total shoulder arthroplasty: long-term
survivorship, functional outcome, and quality of life. J Shoulder Elbow Surg 14(5):471–479.
doi:10.1016/j.jse.2005.02.009
Doorenbosch CA, Harlaar J, Veeger DH (2003) The globe system: an unambiguous description of
shoulder positions in daily life movements. J Rehabil Res Dev 40(2):147–155
Ecklund KJ, Lee TQ, Tibone J, Gupta R (2007) Rotator cuff tear arthropathy. J Am Acad Orthop
Surg 15(6):340–349
the centre of rotation of ball joints. J Biomech 39(15):2798–2809. doi:10.1016/j.
Gamage SS, Lasenby J (2002) New least squares solutions for estimating the average centre of
Grammont PM, Baulot E (1993) Delta shoulder prosthesis for rotator cuff rupture. Orthopedics 16
(1):65–68
dimensional motions: application to the knee. J Biomech Eng 105(2):136–144
Halvorsen K (2003) Bias compensated least squares estimate of the center of rotation. J Biomech 36
(7):999–1008
the center of rotation. J Biomech 32(11):1221–1227
Hannah DC, Scibek JS (2015) Collecting shoulder kinematics with electromagnetic tracking
systems and digital inclinometers: a review. World J Orthop 6(10):783–794. doi:10.5312/wjo.
v6.i10.783
Hudak PL, Amadio PC, Bombardier C (1996) Development of an upper extremity outcome
measure: the DASH (disabilities of the arm, shoulder and hand) [corrected]. The Upper
Extremity Collaborative Group (UECG). Am J Ind Med 29(6):602–608. doi:10.1002/(SICI)
1097-0274(199606)29:6<602::AID-AJIM4>3.0.CO;2-L
Huguet D, DeClercq G, Rio B, Teissier J, Zipoli B, Group T (2010) Results of a new stemless
shoulder prosthesis: radiologic proof of maintained fixation and stability after a minimum of
three years’ follow-up. J Shoulder Elbow Surg 19(6):847–852. doi:10.1016/j.jse.2009.12.009
Inman VT, JBdM S, Abbott LC (1944) Observations on the function of the shoulder joint. J Bone
Joint Surg 26(1):1–30
Jobin CM, Brown GD, Bahu MJ, Gardner TR, Bigliani LU, Levine WN, Ahmad CS (2012) Reverse
total shoulder arthroplasty for cuff tear arthropathy: the clinical effect of deltoid lengthening and
center of rotation medialization. J Shoulder Elbow Surg 21(10):1269–1277. doi:10.1016/j.
jse.2011.08.049
Kasten P, Maier M, Rettig O, Raiss P, Wolf S, Loew M (2009) Proprioception in total, hemi- and
reverse shoulder arthroplasty in 3D motion analyses: a prospective study. Int Orthop 33
(6):1641–1647. doi:10.1007/s00264-008-0666-0
Kasten P, Maier M, Wendy P, Rettig O, Raiss P, Wolf S, Loew M (2010) Can shoulder arthroplasty
restore the range of motion in activities of daily living? A prospective 3D video motion analysis
study. J Shoulder Elbow Surg 19(2 Suppl):59–65. doi:10.1016/j.jse.2009.10.012
Kontaxis A, Cutti AG, Johnson GR, Veeger HE (2009) A framework for the definition of
standardized protocols for measuring upper-extremity kinematics. Clin Biomech 24
(3):246–253. doi:10.1016/j.clinbiomech.2008.12.009
Kwon YW, Pinto VJ, Yoon J, Frankle MA, Dunning PE, Sheikhzadeh A (2012) Kinematic analysis
of dynamic shoulder motion in patients with reverse total shoulder arthroplasty. J Shoulder
Elbow Surg 21(9):1184–1190. doi:10.1016/j.jse.2011.07.031
(1999) Validation of a functional method for the estimation of hip joint Centre location. J
Biomech 32(1):99–103
Lempereur M, Leboeuf F, Brochard S, Rousset J, Burdin V, Remy-Neris O (2010) In vivo
estimation of the glenohumeral joint Centre by functional methods: accuracy and repeatability
assessment. J Biomech 43(2):370–374. doi:10.1016/j.jbiomech.2009.09.029
Lempereur M, Brochard S, Leboeuf F, Remy-Neris O (2014) Validity and reliability of 3D marker
based scapular motion analysis: a systematic review. J Biomech 47(10):2219–2230.
doi:10.1016/j.jbiomech.2014.04.028
Lugli T (1978) Artificial shoulder joint by Pean (1893): the facts of an exceptional intervention and
the prosthetic method. Clin Orthop Relat Res 133:215–218
Magermans DJ, Chadwick EK, Veeger HE, van der Helm FC (2005) Requirements for upper
extremity motions during activities of daily living. Clin Biomech 20(6):591–599. doi:10.1016/j.
Maier MW, Niklasch M, Dreher T, Wolf SI, Zeifang F, Loew M, Kasten P (2012) Proprioception
3 years after shoulder arthroplasty in 3D motion analysis: a prospective study. Arch Orthop
Trauma Surg 132(7):1003–1010. doi:10.1007/s00402-012-1495-6
Maier MW, Caspers M, Zeifang F, Dreher T, Klotz MC, Wolf SI, Kasten P (2014a) How does
reverse shoulder replacement change the range of motion in activities of daily living in patients
with cuff tear arthropathy? A prospective optical 3D motion analysis study. Arch Orthop
Trauma Surg 134(8):1065–1071. doi:10.1007/s00402-014-2015-7
Maier MW, Niklasch M, Dreher T, Zeifang F, Rettig O, Klotz MC, Wolf SI, Kasten P (2014b)
Motion patterns in activities of daily living: 3- year longitudinal follow-up after total shoulder
arthroplasty using an optical 3D motion analysis system. BMC Musculoskelet Disord 15:244.
doi:10.1186/1471-2474-15-244
Maier MW, Lauer S, Klotz MC, Bulhoff M, Spranz D, Zeifang F (2015) Are there differences
between stemless and conventional stemmed shoulder prostheses in the treatment of
glenohumeral osteoarthritis? BMC Musculoskelet Disord 16:275. doi:10.1186/s12891-015-
0723-y
Matsuki K, Matsuki KO, Mu S, Yamaguchi S, Ochiai N, Sasho T, Sugaya H, Toyone T, Wada Y,
Takahashi K, Banks SA (2011) In vivo 3-dimensional analysis of scapular kinematics: compar-
ison of dominant and nondominant shoulders. J Shoulder Elbow Surg 20(4):659–665.
doi:10.1016/j.jse.2010.09.012
Meskers CG, van der Helm FC, Rozendaal LA, Rozing PM (1998a) In vivo estimation of the
glenohumeral joint rotation center from scapular bony landmarks by linear regression. J
Biomech 31(1):93–96
Meskers CG, Vermeulen HM, de Groot JH, van Der Helm FC, Rozing PM (1998b) 3D shoulder
position measurements using a six-degree-of-freedom electromagnetic tracking device. Clin
Biomech 13(4–5):280–292
Meskers CG, Fraterman H, van der Helm FC, Vermeulen HM, Rozing PM (1999) Calibration of the
“flock of birds” electromagnetic tracking device and its application in shoulder motion studies. J
Biomech 32(6):629–633
Meskers CG, van de Sande MA, de Groot JH (2007) Comparison between tripod and skin-fixed
recording of scapular motion. J Biomech 40(4):941–946. doi:10.1016/j.jbiomech.2006.02.011
Michener LA, McClure PW, Sennett BJ (2002) American shoulder and elbow surgeons standard-
ized shoulder assessment form, patient self-report section: reliability, validity, and responsive-
ness. J Shoulder Elbow Surg 11(6):587–594. doi:10.1067/mse.2002.127096
Neer CS 2nd (1955) Articular replacement for the humeral head. J Bone Joint Surg Am 37-A
(2):215–228
Parel I, Cutti AG, Fiumana G, Porcellini G, Verni G, Accardo AP (2012) Ambulatory measurement
of the scapulohumeral rhythm: intra- and inter-operator agreement of a protocol based on inertial
and magnetic sensors. Gait Posture 35(4):636–640. doi:10.1016/j.gaitpost.2011.12.015
Phadke V, Braman JP, LaPrade RF, Ludewig PM (2011) Comparison of glenohumeral motion using
different rotation sequences. J Biomech 44(4):700–705. doi:10.1016/j.jbiomech.2010.10.042
Radnay CS, Setter KJ, Chambers L, Levine WN, Bigliani LU, Ahmad CS (2007) Total shoulder
replacement compared with humeral head replacement for the treatment of primary
glenohumeral osteoarthritis: a systematic review. J Shoulder Elbow Surg 16(4):396–402.
doi:10.1016/j.jse.2006.10.017
Raiss P, Schmitt M, Bruckner T, Kasten P, Pape G, Loew M, Zeifang F (2012) Results of cemented
total shoulder replacement with a minimum follow-up of ten years. J Bone Joint Surg Am 94
(23):e1711–e1710. doi:10.2106/JBJS.K.00580
Razmjou H, Holtby R, Christakis M, Axelrod T, Richards R (2013) Impact of prosthetic design on
clinical and radiologic outcomes of total shoulder arthroplasty: a prospective study. J Shoulder
Elbow Surg 22(2):206–214. doi:10.1016/j.jse.2012.04.016
Rettig O, Fradet L, Kasten P, Raiss P, Wolf SI (2009) A new kinematic model of the upper extremity
based on functional joint parameter determination for shoulder and elbow. Gait Posture 30
(4):469–476. doi:10.1016/j.gaitpost.2009.07.111
Rettig O, Maier MW, Gantz S, Raiss P, Zeifang F, Wolf SI (2013) Does the reverse shoulder
prosthesis medialize the center of rotation in the glenohumeral joint? Gait Posture 37(1):29–31.
doi:10.1016/j.gaitpost.2012.04.019
Sager MA, Dunham NC, Schwantes A, Mecum L, Halverson K, Harlowe D (1992) Measurement of
activities of daily living in hospitalized elderly: a comparison of self-report and performance-
based methods. J Am Geriatr Soc 40(5):457–462
Sandow MJ, David H, Bentall SJ (2013) Hemiarthroplasty vs total shoulder replacement for rotator
cuff intact osteoarthritis: how do they fare after a decade? J Shoulder Elbow Surg 22
(7):877–885. doi:10.1016/j.jse.2012.10.023
Stokdijk M, Nagels J, Rozing PM (2000) The glenohumeral joint rotation Centre in vivo. J Biomech
33(12):1629–1636
van Andel CJ, Wolterbeek N, Doorenbosch CA, Veeger DH, Harlaar J (2008) Complete 3D
kinematics of upper extremity functional tasks. Gait Posture 27(1):120–127. doi:10.1016/j.
van Andel C, van Hutten K, Eversdijk M, Veeger D, Harlaar J (2009) Recording scapular motion
using an acromion marker cluster. Gait Posture 29(1):123–128. doi:10.1016/j.
Vanezis A, Robinson MA, Darras N (2015) The reliability of the ELEPAP clinical protocol for the
3D kinematic evaluation of upper limb function. Gait Posture 41(2):431–439. doi:10.1016/j.
Veeger HE, Magermans DJ, Nagels J, Chadwick EK, van der Helm FC (2006) A kinematical
analysis of the shoulder after arthroplasty during a hair combing task. Clin Biomech 21(Suppl
1):S39–S44. doi:10.1016/j.clinbiomech.2005.09.012
Walker D, Wright TW, Banks SA, Struk AM (2014) Electromyographic analysis of reverse total
shoulder arthroplasties. J Shoulder Elbow Surg 23(2):166–172. doi:10.1016/j.jse.2013.05.005
Walker D, Matsuki K, Struk AM, Wright TW, Banks SA (2015) Scapulohumeral rhythm in
shoulders with reverse shoulder arthroplasty. J Shoulder Elbow Surg 24(7):1129–1134.
doi:10.1016/j.jse.2014.11.043
Warner MB, Chappell PH, Stokes MJ (2015) Measurement of dynamic scapular kinematics using
an acromion marker cluster to minimize skin movement artifact. J Vis Exp 96:e51717.
doi:10.3791/51717
Woltring HJ (1990) Estimation of the trajectory of the instantaneous centre of rotation in planar
biokinematics. J Biomech 23(12):1273–1274
Woltring HJ (1994) 3-D attitude representation of human joints: a standardization proposal. J
Biomech 27(12):1399–1414
Wu G, van der Helm FC, Veeger HE, Makhsous M, Van Roy P, Anglin C, Nagels J, Karduna AR,
McQuade K, Wang X, Werner FW, Buchholz B, International Society of B (2005) ISB
recommendation on definitions of joint coordinate systems of various joints for the reporting
of human joint motion – part II: shoulder, elbow, wrist and hand. J Biomech 38(5):981–992
Expert Opinion and Legal Considerations
Henry M. Silvester
Abstract
This chapter provides practical assistance to experts who may be required to give
evidence in court. It explores the development of law and rules that apply in
several jurisdictions. In particular, the legal principles are examined in the context
of an Australian jurisdiction which is at the forefront of the relatively new practice
in this area – concurrent expert evidence. Important elements are the reliability of
expert opinion, the source of knowledge, training and experience, as well as
procedural aspects. It is prepared as at October 2016.
Keywords
Opinion evidence • Court rules • Litigation • Concurrent evidence
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Evidentiary Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
The Assumptions of Fact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Court and Tribunal Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Concurrent Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Some Practical Guidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
H.M. Silvester (*)

Barry Nilsson Lawers, Sydney, NSW, Australia
e-mail: henry.silvester@bnlaw.com.au

DOI 10.1007/978-3-319-30808-1_85-1
2 H.M. Silvester
Introduction
Expert evidence in courts and tribunals over the last 20 years has undergone a
significant transformation. Gone are the days when a report is commissioned by
one party from an expert, served on the other party with little or no requirements as to
professional conduct or procedural requirements that allow the admissibility of the
opinion evidence.
The experience in one of the legal jurisdictions in Australia, namely New South
Wales (“NSW”), is used in this chapter as an example of the use of expert evidence in
litigation. The reason for this example is that NSW was the origin for one of the most
significant recent changes in how experts give evidence in courts and tribunals
(Jackson 2016). That change was the process of experts giving evidence concur-
rently in courts or tribunals having reached agreement and identified areas of
disagreement.
The purpose of this chapter is to provide an overview. More detailed consider-
ation of applicable rules of evidence and procedures can be found for the Australian
approach in Freckelton and Selby (2009) and Odgers (2014). For the United States,
see Conley and Moriarty (2011).
Justice Peter Garling of the Supreme Court of NSW helpfully identified some
relevant judicial comment on the role of experts in litigation (Garling 2015):
In The Queen v Turner [1975] QB 834 at 841, Lawton LJ expressed the basis
upon which expert evidence is received in these words:
An expert’s opinion is admissible to furnish the court with scientific information which is
likely to be outside the experience and knowledge of a judge or jury.
A similar description was given by Gaudron and Gummow JJ in Osland v The

Queen [1998] 197 CLR 316 when Their Honours said:
Expert evidence is admissible with respect to a relevant matter about which ordinary persons
are [not] able to form a sound judgment . . . without the assistance of [those] possessing
special knowledge or experience in the area.
In the Canadian decision of the Supreme Court in R v Mohan [1994] 2 SCR 9;

(1994) 89 CCC (3d) 402 the admission of expert evidence depended on the appli-
cation of the following (Freckelton and Selby 2009 p.73):
(1) Relevance;
(2) Necessity in assisting the trier of fact;
(3) The absence of any exclusionary rules; and
(4) A properly qualified expert.
The first two questions are often not difficult to address. The rules change
between jurisdictions; however, there are similarities. But what is a “properly
qualified expert” in the context of litigation? This depends upon the particular
Expert Opinion and Legal Considerations 3
knowledge, training, or experience of the expert against the backdrop of their field of
study and compliance with court procedures.
In 1995 the Evidence Act was introduced in NSW and also in relation to matters
attracting the jurisdiction of the Commonwealth of Australia in an attempt to provide
uniform evidence law, including as applied to expert evidence. Combined with the
evidentiary requirements are the procedural obligations on experts and the NSW
example is Part 31, Division 2 of the Uniform Civil Procedure Rules 2005.
In the Australian experience, in order for an expert’s opinion to be considered as a
reliable source of evidence for a judge or tribunal member to make a decision about
an issue or issues in a dispute, a number of criteria must be met (see Freckelton and
Selby 2009):
• The evidentiary requirement is that an expert must have specialized knowledge

based on training, study, or experience and that the opinion given is founded upon
these elements.
• Regardless of who has commissioned the expert, the expert’s overriding duty is to
the court or tribunal;
• They must exercise their independent professional judgment and if possible,
endeavor to reach an agreement with any other expert witness qualified on the
matter in issue.
• There must be a transparency of information relied upon by the expert, not only as
to the facts and assumptions of facts upon which the opinion is based, but other
material, examinations, tests, or other investigations upon which the expert has
relied or drawn upon to reach their opinion.
• The duty to the court or tribunal extends to identifying the extent to which the
expert may consider that their report is incomplete or inaccurate unless some
further information is provided or in the event that insufficient research or data is
identified. Also if, after providing an opinion, the expert changes their views, then
this new opinion must be given immediately.
The basis for these requirements is to ensure that in order for an expert opinion to
be relied upon by a tribunal or court, it is in no way misleading and is reliably based
upon knowledge, training, research, experience, or study and, where differing
opinions are reached between experts, that their best endeavors are made to reach
some agreement and to identify areas of disagreement and reasons for such dis-
agreement. If agreement is not possible, then the court or tribunal can clearly have
available the differing assumptions or facts or indeed the different interpretation of
scientific principle which lead to a disagreement such that findings may be made
about those matters and in turn accept or reject the expert opinion.
4 H.M. Silvester
State of the Art
Originally developed in NSW in Australia, the innovative procedure of expert

witnesses giving “concurrent evidence” in courts and tribunals has gained support
in other jurisdictions such as the United Kingdom. This procedure applies before and
during the court hearing. Before the hearing, it involves experts meeting in a
“conclave” and identifying in a joint written report areas of agreement and disagree-
ment. Then at the hearing the experts give evidence concurrently rather than the
more common procedure of separately. Later in this chapter is a discussion of this
recent development of “concurrent evidence.”
Evidentiary Requirements
For opinion evidence to be admissible in an Australian court, it is necessary for that

evidence to be relevant (s.55 of the Evidence Act 1995 (NSW and Commonwealth of
Australia), have sufficient probative value (s.135 of the Evidence Act 1995), and
comply with ss.76 and 79 of the Evidence Act 1995 which relevantly are:
Section 76 The opinion rule

(1) Evidence of an opinion is not admissible to prove the existence of a fact about the
existence of which the opinion was expressed.
(2) . . .
The exception to this is s.79 of the Evidence Act 1995.
Section 79 Exception: opinions based on specialised knowledge

(1) If a person has specialised knowledge based on the person’s training, study or experi-
ence, the opinion rule does not apply to evidence of an opinion of that person that is
wholly or substantially based on that knowledge.
(2) . . ..
If a person has specialised knowledge based on the person’s training, study or experience,
the opinion rule does not apply to evidence of an opinion of that person that is wholly or
substantially based on that knowledge.
Section 79 of the Evidence Act 1995 creates a requirement that opinion evidence must meet
the following:
• the person giving the opinion evidence must have “specialised knowledge”;
• the specialised knowledge must be “based on training, study or experience”;
• the opinion must be “wholly or substantially based on that [specialised] knowledge”.
These elements must be established on the balance of probabilities before a court

can accept that the opinion expressed in a report is admissible. The secondary
consideration, once admitted into evidence, is the persuasive weight that the court
may attach to expert opinion. For example, the opinion may be admissible (and for
example fulfil the requirements of s.79 of the Evidence Act 1995); however, an
assumption upon which the opinion is based may be established by other evidence
which is of itself found to be doubtful. This in turn would cause a court to have doubt
as to the value or weight of the expert opinion to the extent that it is based upon that
questionable assumption. Significantly, in the event that an assumption or factual
basis is not made out in the evidence, then it is doubtful that any weight may be
attached to any expert opinion that is based upon an erroneous assumption of fact
and indeed the expert opinion may well be rendered inadmissible before the court.
Sometimes a court may seek the assistance of an expert to understand something
technical without strict compliance with this section (Branson 2006).
In considering what amounted to “specialized knowledge,” Justice Gaudron of
the High Court of Australia in Velevski v R [2002] 187 ALR 233 at para. 82 observed
that “specialized knowledge” was knowledge of matters which are outside the
knowledge or experience of ordinary persons and which is sufficiently organized
or recognized to be accepted as a reliable body of knowledge.
The phrase, “specialized knowledge” is not defined in Australian legislation. In
Adler v Australian Securities and Investments Commission [2003] NSWCA 131 at
[629]; however, Justice Giles of the NSW Court of Appeal noted that the phrase “is
not restrictive; its scope is informed by the available bases of training, study and
experience”. It has considerable scope and s.80 of the Evidence Act 1995 also
permits that opinion evidence can be admissible even if it is a matter of common
knowledge.
Example: Assume that you are asked to provide an opinion on whether a
passenger in a motor vehicle was wearing a seat-belt when the vehicle was involved
in a frontal collision with another vehicle. This example will be used again through-
out this chapter to illustrate the practical application of the principles discussed. The
first question is do you have the specialized knowledge based on training, study or
experience? A bio-mechanical engineer would as they are trained in both engineer-
ing and relevant aspects of medicine. An engineer without medical knowledge may
be sufficiently expert if, for example, they had significant experience in motor
accident reconstruction over a lengthy period of time. A chemical engineer would
not pass the specialized knowledge test as their expertise is not relevant to the issue
the subject of the opinion, namely whether a seat-belt was worn.
Reliability
In determining what constitutes specialized knowledge, courts have often considered

whether the knowledge is from a “field of expertise.” An important question to be
considered is the extent of the reliability of the evidence. An example of this would
be the reliability of survey results upon which opinion or specialized knowledge
is based: see for example, Interlego AG v Croner Trading Pty Limited (1991)
102 ALR 379.
1
The 1975 Federal Rules of Evidence (US) did not refer to the Frye test.
6 H.M. Silvester
The United States helped establish the concept of the general acceptance of a field
of study as being a source of consideration of the expertise of an expert. An early
decision in the United States of Frye v United States 293 F 1013 (1923) relevantly
held that when considering the admissibility of expert evidence, “the thing from
which the deduction is made must be sufficiently established to have gained general
acceptance in the particular field in which it belongs” (also called the Frye test). With
some modification and subject to criticism, this remained largely the prevailing test
until the majority of the United States Supreme Court considered Rule 702 of the
Federal Rules of Evidence (similar to the Australian s.79 of the Evidence Act 1995)
in Daubert v Merrell Dow Pharmaceuticals 509 US 579 (1993)1. The court held:
[T]he word “knowledge” connotes more than subjective belief or unsupported speculation.
The term “applies to any body of known facts or to any body of ideas inferred from such
facts or accepted as truths on good grounds”. . .proposed testimony must be supported by
appropriate validation – i.e., “good grounds,” based on what is known.
The United States Supreme Court in Daubert went to considerable lengths to

identify whether or not expert evidence was reliable based upon valid scientific
reasoning or methodology including testing (Freckelton and Selby 2009 pp.63–64):
(1) Whether it can be or has been tested . . .;

(2) whether the theory or technique has been subject to peer review and publication . . .;
(3) the known or potential rate of error and the existence and maintenance of standards
controlling the technique’s operation;
(4) whether a technique has gained general acceptance within the scientific community.
The concept of reliability became a cornerstone and in 2000, Rule 702 was
amended in the United States to require reliability. With effect from 1 December
2000, Rule 702 provided:
If scientific, technical or other specialised knowledge will assist the trier of fact to understand
the evidence or to determine a fact in issue, a witness qualified as an expert by knowledge,
skill, experience, training or education, may testify thereto in the form of an opinion or
otherwise, if (1) the testimony is based upon sufficient facts or data, (2) testimony is the
product of reliable principles and methods, and (3) the witness has applied the principles and
methods reliably to the facts of the case.
It has been the approach in the United States that witnesses will not be permitted
to give evidence as experts if their qualifications are not relevant and sufficient
(Freckelton and Selby 2009; see also Conley and Moriarty 2011).
The Canadian experience may also assist experts in understanding how best to
provide opinion evidence. In the DNA profiling case of R v Johnston (1992) 69 CCC
395 at 413, Justice Langdon followed an earlier dissenting judgment of Justice
Wilson in R v Beland [1987] 36 CCC (3d) 481 and considered that the judgment
of Her Honour Justice Wilson was:
persuasive authority for the proposition that the [Frye 293 F 1013 (1923)] test should not be
adopted in Canada. It recommends adoption of a more expansive admissibility standard,
relevancy and helpfulness. It suggests by reference to cross-examination and opposing
experts, that if the evidence meets initial standards of relevancy and helpfulness, further
objections are relevant to weight and not to admissibility.
In relation to novel scientific evidence, Justice Langdon was of the view the
following should be considered (at 415):
(1) The potential rate of error;

(2) The existence and maintenance of standards;
(3) The care with which the scientific technique has been employed and whether it is
susceptible to abuse;
(4) Whether there are analogous relationships with other types of scientific techniques that
are routinely admitted into evidence;
(5) The presence of failsafe characteristics;
(6) The expert’s qualifications and stature;
(7) The existence of specialised literature;
(8) The novelty of the technique and its relationship to more established areas of scientific
analysis;
(9) Whether the technique has been generally accepted by experts in the field . . .;
(10) The nature and breadth of the inference adduced;
(11) The clarity with which the technique may be explained;
(12) The extent to which the basic data may be verified by the court and the jury;
(13) The availability of other experts to evaluate the technique; and
(14) The probative significance of the evidence.
In the United Kingdom, the question of admissibility of evidence was importantly

considered in the Court of Appeal decision in R v Luttrell [2004] 2 Cr App R 31;
[2004] EWCA Crim 1344. The court held that for expert evidence to be admissible,
the study or experience of the expert witness provides authority to the opinion and
the witness must be qualified to express the opinion. The court adopted the inter-
pretation of Chief Justice King of the Supreme Court of South Australia in R v
Bonython (1984) 38 SASR 45 at 46, in considering the authority of an expert:
(a) whether the subject matter of the opinion is such that a person without instruction or
experience in the area of knowledge or human experience would be able to form a sound
judgment on the matter without the assistance of witnesses possessing special knowl-
edge or experience in the area, and
(b) whether the subject matter of an opinion forms part of a body of knowledge or
experience which is sufficiently organised to be accepted as a reliable body of knowl-
edge or experience, a special acquaintance with which by the witness would render his
opinion of assistance to the court.
If the conditions are met, the evidence is admissible and may be considered by the
court whereupon the question becomes one of the persuasive “weight” to be attached
to the evidence as opposed to the admissibility of the evidence (Freckelton and Selby
2009). At p.34 in R v Luttrell (supra) the court considered the issue of reliability
regarding the question of admissibility:
8 H.M. Silvester
In established fields of science, the court may take the view that expert evidence would fall
beyond the recognised limits of the field or that methods are too unconventional to be
regarded as subject to the scientific discipline. But a skill or expertise can be recognised and
respected, and thus satisfy the conditions for admissible expert evidence, although the
discipline is not susceptible to this sort of scientific discipline.
In some cases, reliability of evidence might be relevant to admissibility. For

example, if the evidence is “based on a developing new brand of science or
medicine: until it is accepted by the scientific community as being able to provide
accurate and reliable opinion”: R v Gilfoyle [2001] 2 Cr App R 57 at para. 25. Even if
the field is recognized, an issue may be whether the particular techniques were
sufficiently recognized within the expert’s profession. See, for example, R v Robb
(1991) 93 Cr App R 161 regarding the admissibility of voice identification evidence.
In Australia the courts have noted that the focus must be on “specialized knowl-
edge,” not to be distracted by questions of reliability that are not specifically referred
to in the Evidence Act 1995. A leading case of the Australian High Court, in
considering opinion evidence, was HG v The Queen (1999) 197 CLR 414, in
which her Honour Justice Gaudron (in dissent) did raise the question of reliability:
So far as this case is concerned, the first question that arises with respect to the exception in
s.79 of the Evidence Act is whether psychology or some relevant field of psychological study
amounts to “specialised knowledge.” The position at common law is that, if relevant, expert
or opinion evidence is admissible with respect to matters about which ordinary persons are
unable “to form a sound judgment. . .without the assistance of [those] possessing special
knowledge or experience. . .which is sufficiently organised or recognised to be accepted as a
reliable body of knowledge or experience.” There is no reason to think that the expression
“specialised knowledge” gives rise to a test which is in any respect narrower or more
restrictive than the position at common law.
As to admissibility of expert evidence, it may well be that there is a trend away

from the test of “general acceptance within the professional community” towards
reliability, namely evidence given to the “validity and reliability of the technique or
theory” (Freckelton and Selby 2009 p.75) although this will vary from jurisdiction to
jurisdiction.
Example: Continuing the example of an opinion as to whether a passenger in a
motor vehicle accident was wearing a seat-belt there may be an issue as to causation.
If the passenger was not wearing the seat-belt is this irrelevant as the injuries would
have been caused regardless? In a frontal impact, there is well-known scientific
evidence to say that wearing a seatbelt reduces injury. In this example, however, let
us say that there may be a type of injury discovered relating to low impact frontal
collisions, similar to the zygapophyseal joint injury which is difficult to detect.
Going back to our example, assume a biomechanical engineer gave an opinion
that no seat-belt was worn and had it been worn then the injuries alleged to have
been caused would not have been caused (Genn 2012). The level of expertise
depending upon the medical training may not be reliable or have the necessary
specialized knowledge to comment on the type of injury and its causation if the
expert did not have the requisite training or experience with that type of injury. In
this instance, the engineer may have to work with a sufficiently experienced medical
professional.
The Assumptions of Fact
The principles that apply to the admissibility of expert evidence in Australia have
been succinctly set out in one of the leading decisions in the area by Justice Heydon
(as he then was, before becoming a High Court judge). His Honour paid particular
attention to the assumptions of fact that underpinned an expert’s opinion. In the New
South Wales Court of Appeal in Makita (Australia) Pty Limited v Sprowles [2001]
52 NSWLR 705 at para. 85 Justice Heydon observed:
In short, if evidence tendered as expert opinion evidence is to be admissible,

– it must be agreed or demonstrated that there is a field of “specialised knowledge”;
– there must be an identified aspect of that field in which the witness demonstrates that by
reason of specialised training, study or experience, the witness has become an expert;
– the opinion proffered must be “wholly or substantially based on the witness’ expert
knowledge”;
– so far as the opinion is based on facts “observed” by the expert, they must be identified
and admissibly proved by the expert;
– and so far as the opinion is based on “assumed” or “accepted” facts, they must be
identified and proved in some other way;
– it must be established that the facts on which the opinion is based form a proper
foundation for it;
– and the opinion of an expert requires demonstration or examination of the scientific or
other intellectual basis of the conclusions reach: that is, the expert's evidence must
explain how the field of “specialised knowledge” in which the witness is expert by
reason of “training, study or experience,” and on which the opinion is ‘wholly or
substantially based’ applies to the facts assumed or observed so as to produce the opinion
propounded.
If all these matters are not made explicit, it is not possible to be sure whether the opinion
is based wholly or substantially on the expert’s specialized knowledge. If the court cannot be
sure of that, the evidence is strictly speaking not admissible, and, so far as it is admissible, of
diminished weight.
Subsequent to this decision, some of the above elements have been qualified. In
Sydney-Wide Distributors Pty Limited v Red Bull Australia Pty Limited [2002]
FCAFC 157 at paras. 16 and 87, the Full Court of the Federal Court of Australia
held that many of the elements referred to by Justice Heydon in Makita “involve
questions of degree, requiring the exercise of judgment” and that where a judge is
conducting a trial alone without a jury the matters may go to weight rather than
admissibility.
In the Red Bull case, Her Honour Justice Branson noted:
The approach of Heydon JA as set out [in para. [85] of the judgment] is, as it
seems to me to be understood as counsel of perfection. As a reading of His Honour’s
reasons for judgment as a whole reveals, His Honour recognised that in the context
10 H.M. Silvester
of an actual trial, the issue of admissibility of evidence tendered as expert opinion

evidence may not be always to be addressed in the way outlined in the above
paragraph.
There were three reasons given for limiting the “counsel of perfection.” The first
was where opinion evidence was admitted without objection by the parties and in
this context, a court rarely would interfere. Secondly, a ruling on admissibility of
evidence is usually required during the course of the trial rather than at the end by
which time other evidence may impact on the weight of the opinion evidence. As
Her Honour Justice Branson noted further in the Red Bull case:
. . . It may prove to be the case that evidence ruled admissible as expert opinion will later be
found by the trial judge to be without weight for reasons that, strictly speaking, might be
thought to go to the issue of admissibility (e.g. that the witness’s opinion is expressed with
respect to a matter outside his or her area of expertise or is not wholly or substantially based
on that expertise.
The third reason was identified in the earlier case of Quick v Stoland Pty Ltd
[1998] 87 FCR 371 at 373-74:
The common law rule that the admissibility of expert opinion evidence depends on proper
disclosure of the factual basis of the opinion is not reflected as such in the Evidence Act 1995
(Cth.). The Australian Law Reform Commission recommended against such precondition to
the admissibility of expert opinion, expressing the view that the general discretion to refuse
to admit evidence would be sufficient to deal with problems that might arise in respect of an
expert opinion, the basis of which is not disclosed: ALRC Report No. 26, vol. 1 para. 750.
In a further decision in Australia, Cadbury Schweppes Pty Ltd v Darrell Lea

Chocolate Shops Pty Ltd [2006] 228 ALR 719, (Federal Court of Australia) Justice
Heerey had to determine the admissibility of opinion evidence which had been based
on “market research reports” and the like which had not been proved in evidence and
were not likely to be “proved”: [2006] 228 ALR 719 at 722 para. 6. At para.
7, Heerey J said:
However, I accept the submission of Senior Counsel for Cadbury that this aspect of Makita
has not been followed in the Federal Court. The lack of proof of a substantial part of the
factual basis of Dr Gibbs’ opinions does not of itself render his evidence inadmissible under
s.79. Such lack of proof merely goes to the weight which may be given to the opinion [and
the court referred to the decision in Red Bull and other decisions].
There are other qualifications regarding assumptions of fact.

Opinion evidence is not admissible if it is found that the opinion was based upon
different facts from those disclosed, for example, if the expert was influenced by
undisclosed facts. The fact of how the expert came to hold an opinion is relevant to
weight: Australian Securities & Investments Commission v Rich [2005] NSWCA
152 per Chief Justice Spigelman.
As long as the opinion identifies the reasoning process, it is still able to be
admitted into evidence even if there were other tests that could have been done to
underpin the opinion. An opinion may not be excluded because the reasoning
process is not fully disclosed if the area of expertise involves some subjectivity.
An opinion may be admitted into evidence where the expert has made an
observation even when all the nonopinion elements of that observation are not
set out.
The trend is that it is not necessary to always prove matters which are customarily
relied upon by experts in a particular field.
The need to prove a fact upon which a particular opinion is based was clarified by
Justice Heydon in Rhoden v Wingate [2002] NSWCA 165 at para. 86, namely, that it
was not whether the fact is proved to a particular standard but whether “there is
evidence which, if accepted, is capable of establishing” the existence of the fact.
If a written opinion sets out facts upon which the opinion has proceeded, and the
report is admitted into evidence then those facts are also admitted into evidence,
unless the court upholds an objection: s.60 Evidence Act 1995. Advocates may make
submissions to the court to exclude the admission of the facts upon which the
opinion is based until those facts are independently proven (e.g., in NSW and
Australian Federal Courts s.136 Evidence Act 1995).
Rule 703 of the United States Federal Rules of Evidence provides the following:
The facts or data in the particular case upon which an expert bases an opinion or inference
may be those perceived by or made known to him [or her] at or before the hearing. If of a
type reasonably relied upon by experts in a particular field in forming opinions or inferences
upon the subject, the facts or data need not be admissible in evidence.
Example: The expert is asked to assume that the passenger who did not wear the
seat belt was involved in a frontal collision. If in fact the collision was not purely
frontal but involved a lateral element, then the opinion may be found to be irrelevant
as it did not address the effect on causation, for example, of failure to wear a seat-belt
in a lateral collision. It is important when being asked to give an opinion based on
assumptions that when objectively considered with known facts whether the
assumptions make sense and to question the assumptions if they do not make sense.
Court and Tribunal Rules
Various jurisdictions have procedural requirements as to the admissibility of opinion

evidence.
In New South Wales, the procedural requirements on expert opinion are set out in
part in Rule 31.23 of the Uniform Civil Procedure Rules 2005 (NSW). It requires that
an expert witness must comply with the “Code of Conduct” which is set out in
Schedule 7 to the rules and any report is inadmissible unless it contains an acknowl-
edgement that the expert has read the Code of Conduct and has agreed to be bound
by it. The same applies to any oral evidence to be received from an expert. The court
has a discretion to exempt compliance but the application of the Code of Conduct is
treated very seriously and experts are rarely exempted.
12 H.M. Silvester
In essence, the Code of Conduct emphasizes the following:
(1) The duty of the expert is to the court rather than any party who has qualified or
paid the expert to appear.
(2) That the expert is not an advocate for a party but rather an independent, impartial
source of evidence and opinion to assist the court.
(3) The expert must follow any direction by the court.
(4) There is a duty to work co-operatively with other expert witnesses by exercising
their judgment and endeavoring to reach agreement.
(5) Requirements as to experts’ reports (see discussion above as to evidentiary
matters).
(6) Experts’ conference – based upon a direction by the court for experts from the
parties to proceedings to confer with each other to identify areas of agreement
and/or disagreement and if appropriate prepare a joint report.
The last aspect is discussed below under “concurrent evidence.”

The requirements under the rules for experts’ reports work together with the
evidentiary principles and arguably are more specific than the evidentiary require-
ments for admissibility of the opinion. They extend to an expert identifying if a
particular issue falls outside their area of expertise and if they believe that a report is
incomplete or inaccurate without some qualification, then the qualification must be
stated in the report. Further, if the expert considered that their opinion is not a
concluded opinion due to insufficient research or data or any other reason, then this
must be stated when the opinion is expressed. If the expert changes their opinion,
then the expert must provide a supplementary report detailing the changed opinion.
In Australia under the Legal Profession Uniform Law Australia Solicitors Con-
duct Rules 2015, Rule 24.2.3, a lawyer can draw to the expert’s attention inconsis-
tencies or other difficulties with the evidence but not encourage the expert to give
evidence different from the opinion they had formed – put simply the opinion in the
report must remain that of the expert. Communications between the expert and the
instructing lawyer should not only actually achieve this result, they also must not
appear to be otherwise.
In most jurisdictions, ethical obligations imposed upon legal practitioners will
regulate the extent to which they may engage with the expert in the formulation of an
opinion to be relied upon by a court and, while it is usually permissible for legal
practitioners to test or challenge opinions in conference prior to the completion of a
report, it is not permissible for them to suggest or require a particular opinion to be
expressed.
The orders of the court to be complied with can sometimes be onerous. Personal
explanations may be required by the court from experts as to why orders of the court
have not been complied with. Experts often have professional responsibilities
outside of their duty to the court including the pressure of business, but this often
is not a satisfactory explanation for failure to comply with the court’s directions.
Another potential area of tension that may arise are competing interests between
obligations to the court and obligations to the expert’s own governing bodies or
associations, including ethical and disclosure obligations. A clear and typical exam-
ple can be found in medical practitioners and the competing obligation of confiden-
tiality to the patient in contrast to the obligation of full candor to the court. Extreme
care must be exercised when balancing the competing obligations.
Concurrent Evidence
Concurrent evidence is a relatively new procedural innovation different from the

more traditional adversarial approach to testing expert evidence. One of the diffi-
culties with the traditional approach was that there was a separation in the giving of
evidence by experts so that evidence was not heard in the context of either other
expert evidence or witness evidence received later in the trial. Sometimes experts
would have to be recalled to give evidence if new factual evidence had emerged
during the course of the trial which changed the assumptions upon which the experts
previously gave their opinions. Also during the course of the hearing, one expert
would often have to sit in the body of the tribunal and suggest questions to an
advocate who was cross-examining another expert. This could be cumbersome. So,
not only for reasons of efficiency but also to produce more useful and relevant
evidence to a tribunal, in the jurisdiction of New South Wales, Australia, the concept
of concurrent evidence began.
In Australia, the practice of concurrent evidence was developed and has since
been copied in other jurisdictions, such as the United Kingdom2. The stages
involved in concurrent evidence are as follows (Garling 2011):
• Preparation
• The expert conclave – the meeting of experts before the hearing
• The joint written report
• Oral evidence
In the usual course, the experts would be instructed by the parties to the litigation
to prepare their separate written opinions and this was no different to the previous
approach. Then the new procedure is applied to focus attention on the issues in
dispute. Once the written opinions are exchanged, and before any trial, the experts
have a meeting (the expert conclave) to identify areas of agreement or disagreement.
A joint report is then prepared by the experts setting out those areas of agreement and
those areas where differences of opinion remain. Where there is a difference of
2
A pilot program was established in Manchester Specialist courts between 2010 and 2013 with His
Honour Judge David Waksman QC, the Manchester Mercantile Judge overseeing the pilot, as he
then was. Professor Dame Hazel Genn monitored the pilot and published findings in the Civil
Justice Quarterly, “Getting to the Truth: Experts and Judges in the ‘Hot Tub’” (2013) 32 CJQ
275–299 (Genn 2012). From 1 April 2013, in the United Kingdom Practice Direction 35 was
adopted into the Civil Procedure Rules.
14 H.M. Silvester
opinion, the report should also identify the reasons and the basis for the difference of
opinion.
Often, prior to the conclave occurring, it is incumbent upon the legal representa-
tives for parties to prepare a joint list of assumptions and questions to be put to the
experts for consideration in conclave. The preparation of such a document is often of
some significance as it tends to set the agenda for the conclave and focus the
attention of the experts upon matters considered by the parties to be relevant
considerations.
It is also important to remember that the usual course is that the experts meet and
prepare their report in the absence of the legal representatives or parties. This reflects
the basic premise that the experts overriding obligation is to the court and not to the
parties that have retained them.
Once the hearing commences, the experts of the same discipline, for example,
biomechanical engineers, give their evidence concurrently and the judge or presiding
officer of the tribunal chairs a discussion between the experts. In essence, the joint
report that the experts have created from their prehearing meeting forms the agenda
for the discussion. Advocates appearing for the parties also participate in the
“discussion” and can put questions or challenge the experts (either individually or
jointly). The process is all controlled by the judge or presiding tribunal member.
Advantages of the procedure of concurrent evidence from experts are that it not
only saves time and costs, but allows the experts to properly assist the tribunal in
reaching a decision about the particular dispute (Jackson 2016). The procedure is
also consistent with the well-known process of peer review within academic and
professional communities. The peers give their review of each other’s opinions in the
context of the hearing before the judge or tribunal.
While the process does suffer from some criticism, this is to be expected from any
innovation. The key to the process of working effectively is that the experts
understand the rules underpinning the process in each jurisdiction.
For example, the Supreme Court of NSW, Common Law Division – General Case
Management List Practice Note No. SC CL5, Parts 36 – 40 (see also SC CL7
(Professional Negligence List):
Concurrent expert evidence

36. This part of the Practice Note applies to all proceedings in which a claim is made for
damages for personal injury or disability.
37. All expert evidence will be given concurrently unless there is a single expert appointed
or the Court grants leave for expert evidence to be given in an alternate manner.
38. At the first Directions Hearing the parties are to produce a schedule of the issues in
respect of which expert evidence may be adduced and identify whether those issues
potentially should be dealt with by a single expert witness appointed by the parties or by
expert witnesses retained by each party who will give evidence concurrently.
39. In the case of concurrent experts, within 14 days of all expert witness statements/reports
being filed and served, the parties are to agree on questions to be asked of the expert
witnesses. If the parties cannot reach agreement within 14 days, they are to arrange for
the proceedings to be re-listed before the Court for directions as to the questions to be
answered by the expert witnesses.
40. In the case of concurrent experts the experts in each area of expertise are to confer and
produce a report on matters agreed and matters not agreed within 35 days of the first
Directions Hearing or such other time as the Court may order.
Regarding the joint report, rule 31.26 of the Uniform Civil Procedure Rules
(NSW) relevantly provides the following:
(1) . . .
(2) The joint report must specify matters agreed and matters not agreed and the reasons for
any disagreement.
(3) The joint report may be tendered at the trial as evidence of any matters agreed.
(4) In relation to any matters not agreed, the joint report may be used or tendered at the trial
only in accordance with the Rules of Evidence and the Practices of the court.
(5) Except by leave of the court, a party affected may not adduce evidence from any other
expert witness on the issues dealt with in the joint report.
In addition to the above, individual judges may place emphasis on certain aspects
of the procedure, for example the material with which the experts ought to be
provided (Garling 2011 p.5):
(a) an index of the documents, together with a paginated folder of the documents which is to
be put before each expert participating at a joint conference and the giving of concurrent
evidence;
(b) a complete list of the factual assumptions which are agreed, or else for which each party
contends, as the appropriate basis for the joint expert opinion; and
(c) the questions which each party contends are appropriate for the experts to be asked to
answer.
In the prehearing stage of identifying what is agreed or disagreed, the experts

when meeting may be assisted by an independent chairperson. The chairperson can
then ensure that the pre-hearing conference takes place in a manner allowing each
expert’s opinion to be heard and also be responsible for the joint report arising from
the pre-hearing meeting.
Justice Garling suggested some guidance to experts giving concurrent oral
evidence as follows (Garling 2015 pp.16–17):
It is my practice to briefly outline what is intended to happen, describe the roles and
functions of those in the court room, and to describe the ground rules for the session. In
particular, I find it necessary to remind the participants that only one person can speak at any
one time because often when a discussion takes place, the witnesses can forget that they are
in a court room. In giving this outline, I commence with identifying the role of the trial judge
and remind them that it is the judge’s task to control the proceedings by, in effect, being the
chairman of their professional meeting, so as to ensure that the agenda items are all covered
in an orderly fashion, to ensure that each of the witnesses has an opportunity to state their
opinions and the basis for them and, ultimately, to ensure that the process is conducted fairly
to the parties, and each expert and with civility.
16 H.M. Silvester
His Honour went on to note, consistent with the Code of Conduct referred to
above, that the role of the experts is to give their evidence truthfully, not as an
advocate for either party and impartially. His Honour also helpfully has identified
some of the practicalities that can apply which experts would face if they were in
court or a tribunal giving concurrent evidence (Garling 2011 p.18):
Accordingly, at the conclusion of the first issue (or item on the agenda) and after the judge
has finished raising any matters, counsel for each of the parties then, in turn, can question the
witnesses ensuring as they do that each expert has the opportunity to answer the question
asked. In other words, the examination of the experts by counsel bears little similarity to the
typical cross-examination. The purpose of counsel’s questions is to ensure that an expert’s
opinion is fully articulated and tested against a contrary opinion, even an opinion elicited by
the judge. As Giles JA said at [107] in Turjman v Stonewall Hotel Pty Ltd [2011] NSWCA
392:
“When concurrent evidence is being taken with the degree of direction by the judge,
counsel are not passengers. They can and should seek to raise material issues and put
material questions to the witnesses . . ., if necessary submitting that the judge’s view of
how the evidence should be brought out should be modified.”
This process is then repeated for each item until all of the issues have been dealt with.
Some Practical Guidance
In the context of these obligations, experts need to exercise care when they are
preparing either written reports or giving oral evidence in proceedings. An expert
must ensure that their opinion is their own, not something which has evolved
through draft opinions edited by others to something tending towards advocacy.
Rather the opinion must be an independent expert opinion.
An expert in a particular field of study may sometimes express opinions outside of
that field. The expert must however be cautious when straying into this territory and
is likely to be questioned about this in court. Such evidence may not satisfy the test
of whether the expert has “specialized knowledge.” This may not only undermine
that part of the expert’s opinion but may call into question the persuasive weight as to
the balance of the opinion.
An example of this is that a mechanical engineer would be able to comment about
the forces involved in the collision of vehicles and the transfer of energy to the
occupants of the vehicle. That expert, however, arguably would not have the
expertise sufficient for an opinion to be admissible regarding assessment of medical
evidence such as extrapolating from injuries identified as to the physical forces that
may have caused them. A mechanical engineer with biomechanical knowledge,
training, or experience, however, more likely would be able to express that as an
admissible opinion.
Chief Justice Gleeson characterized the evidence in HG v The Queen (supra), as
“a combination of speculation, inference, personal and second hand views as to the
credibility of the complainant, and the process of reasoning which went well beyond
the field of expertise”. It is important for the expert not to stray.
The expert giving the opinion must be careful not to simply engage in factual
analysis where their specialized knowledge is unnecessary. In this instance, a judge
receives no benefit from the opinion based upon the specialized knowledge of the
expert as the judge can perform the factual analysis. Indeed, such opinion is unlikely
to be admissible given that it is not based upon the specialized training, experience,
or education.
An example of this might be the opinion as to whether or not a victim in a vehicle
collision was wearing a seatbelt. A judge may have the benefit of the observations of
witnesses such as any other passengers in the vehicle and other sources of informa-
tion such as attending ambulance personnel. The expert should combine their
understanding of these observations based upon their knowledge, training, or expe-
rience, for example as a biomechanical engineer. The ambulance officers’ notes and
hospital records may show abdominal and shoulder bruising suggestive that a
seatbelt was worn. Here the expert combines the available facts derived from the
medical records and by using their expertise interprets those records to give an
opinion as to the likely forces from a seatbelt to the body that would be sufficient to
cause the bruising.
Chief Justice Gleeson, commented further in HG v The Queen (supra):
An expert whose opinion is sought to be tendered should differentiate between the assumed
facts upon which the opinion is based, and the opinion in question . . . By directing attention
to whether an opinion is wholly or substantially based on specialised knowledge based on
training, study or experience, the section [s.79] requires that the opinion is presented in a
form which makes it possible to answer that question . . . Experts who venture "opinions"
(sometimes merely their own inference of fact), outside their field of specialised knowledge
may invest those opinions with a spurious appearance of authority, and legitimate processes
of fact finding may be subverted.
The word “substantially” in the context of “substantially based on specialized

knowledge” allows the court discretion. This allows an opinion to be admissible if
based upon specialized knowledge even if it takes into account matters of “common
knowledge”: Velevski v The Queen [2002] 76 ALJR 402 (High Court of Australia)
per Justices Gaudron, Gummow, and Callinan.
Reliability of an opinion has been an important part of considering whether or not
expert evidence can be admissible. For example, the issue of body mapping, facial
mapping, and fingerprinting. Fingerprint evidence has a scientific basis and database
for purposes of comparison to allow experts to formulate an opinion as to whether
one fingerprint is similar to another. Facial or body mapping may lack a similar
database or mathematical formula as exists in the case of fingerprint comparison.
The issue perhaps is not that the opinion lacked reliability but that without the
available specialized knowledge the opinion becomes subjective: Daubert v Merrell
Dowell Pharmaceuticals 509 US 579 [1993].
An expert can and often uses materials provided by others in formulating their
opinion. A radiologist who produces a Magnetic Resonant Imaging scan which is
then used by a surgeon for forming their opinion about diagnosis does not mean that
the ultimate diagnosis is not formed “substantially based” on the specialized
18 H.M. Silvester
knowledge, of the surgeon. Care must be exercised when the expert makes their own
investigations. For example, the expert should never to contact directly a party to the
litigation or witness. A good guide is to first discuss any such investigations with
whoever has requested the written opinion.
A difficulty emerges where there are joint opinions and as can be seen from the
procedural approach to expert evidence this is becoming increasingly important. If
there is only oral evidence from 1 of the 2 experts who prepared a report, and the
report jointly authored does not identify whose opinion has been formed by which of
the two experts, then the requirement that the opinion be wholly or substantially
based on the specialized knowledge of the witness may not be fulfilled.
If an expert’s opinion has been based upon assumed facts and those facts
ultimately have not been separately proven then the expert’s opinion is of little, if
any, weight. Also important is the reasoning process, which leads to the formation of
opinion.
It is important for the experts to have assumptions clearly set out rather than, for
example, to be relying on written statements from witnesses that the experts may
have been provided. The reason is that then in the process of concurrent evidence,
one expert may be interpreting a statement and taking assumptions from it in one
way, while other experts may be interpreting the statement differently. By having
agreed assumptions or at least clearly formulated contended assumptions, then the
basis for their opinion is clear.
Conclusion
One of the judges responsible for encouraging the development of concurrent

evidence in NSW, Justice McClellan, made the following comment (McLellan
2007 p.17):
As far as the decision-maker is concerned, my experience is that because of the opportunity

to observe the experts in conversation with each other about the matter, together with the
ability to ask and answer each other’s questions, the capacity of the judge to decide which
expert to accept is greatly enhanced. Rather than have a person’s expertise translated or
coloured by the skill of the advocate, and as we know the impact of the advocate is
sometimes significant, you have the expert’s views expressed in his or her own words.
An expert’s own words are all the more persuasive if they are given impartially,
based upon identified assumptions, with the authority of knowledge, training, or
experience and if they are able to respond to peer review.
Glossary
Concurrent evidence A procedure where experts in the same field, each separately
qualified by the parties to litigation, give evidence jointly as a panel in court. They
identify areas of agreement and disagreement in a joint report before trial. At trial
each expert comments on the opinion of the other experts and advocates for the
parties also question them. The judge acts as a chairperson and ultimately decides
the issues.
Court rules Rules of procedure applied to court proceedings.
Litigation The process of dispute resolution by the members of courts or tribunals
based upon admissible evidence presented by advocates for the parties to the
dispute.
Opinion evidence Evidence based upon a person’s trainingstudy or experience. It is
given initially in writing by an expert and subject to oral or written review by
peers, court advocates, and judges based upon applicable court rules
References
Branson JC (2006) Expert evidence: a judge’s perspective. Bar News, NSW Bar Association,
Sydney, summer 2006/2007 pp. 32–38. Available online http://archive.nswbar.asn.au/docs/
resources/publications/bn/bn_summer0607.pdf
Conley I, Moriarty I (2011) Scientific and expert evidence, 2nd edn. Wolters Kluwer, New York
Freckelton I, Selby H (2009) Expert evidence – law, practice, procedure and advocacy, 4th edn.
Thomson Reuters, Sydney
Garling JP (2011) Concurrent expert evidence – reflections and development. Paper presented at the
Australian Insurance Law Association, Twilight Seminar Series, Sydney, 17 August 2011.
Available online http://www.supremecourt.justice.nsw.gov.au/Documents/garling170811.pdf
Garling JP (2015) Concurrent expert evidence – the new South Wales experience. Paper presented
at the University of Oxford Faculty of Law, Oxford, 1 December 2015. Available online http://
www.supremecourt.justice.nsw.gov.au/Documents/Speeches/2015%20Speeches/Garling_
20151201.pdf
Genn DH (2012) Manchester concurrent evidence pilot – Interim Report, UCL Judicial Institute,
London, January 2012. Available online https://www.judiciary.gov.uk/wp-content/uploads/
JCO/Documents/Reports/concurrent-evidence-interim-report.pdf
Jackson LJ (2016) Concurrent evidence – a gift from Australia. Paper presented at the conference of
the Commercial Bar Association of Victoria, London, 29 June 2016. Available online https://
www.judiciary.gov.uk/wp-content/uploads/2016/06/lj-jackson-concurrent-expert-evidence.pdf
McLellan JP (2007) Concurrent evidence. Paper presented at the medicine and law conference of
the Law Institute of Victoria, Melbourne, 29 November 2007. Available online http://www.
supremecourt.justice.nsw.gov.au/Documents/mcclellan_2007.11.29.pdf
Odgers S (2014) Uniform evidence law, 11th edn. Thomson Reuters, Sydney
Injury Mechanisms in Traffic Accidents
Brian D Goodwin, Sajal Chirvi, and Frank A Pintar
Abstract
This chapter aims to describe and explain current understandings of injury
mechanisms in motor vehicle crashes. The following sections are organized
according to anatomical regions and associated injuries. This chapter was limited
to discussing mechanisms of severe and/or fatal injuries. Real-world injury
biomechanics and injury forensics studies play an essential role that impels new
developments in vehicle design and safety enhancements. The science of impact
mechanics aims to explain injury mechanisms, characterize biomechanical sys-
tems, estimate injury risk, and analyze approaches to injury prevention. The intent
of this chapter is to review current understandings of injury mechanisms common
in traffic crashes.
Keywords
Vehicle crash • Vehicle collision • Injury scale • Injury probability • Injury
criteria • Motor vehicle
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Epidemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Current Approaches and Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Head and Neck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Head and Brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Cervical Injury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
B.D. Goodwin (*) • S. Chirvi • F.A. Pintar

Neuroscience Research Labs – Research 151, Medical College of Wisconsin, Zablocki VA Medical
Center, Milwaukee, WI, USA
e-mail: bgoodwin@mcw.edu; schirvi@mcw.edu; fpintar@mcw.edu

DOI 10.1007/978-3-319-30808-1_93-1
2 B.D. Goodwin et al.
Thorax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Traumatic Rupture of the Aorta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Thoracic Spine Injuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Intra-abdominal Injuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Pelvis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Fracture Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Lower Extremities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Introduction
An adequate understanding of injury mechanisms is necessary to implement new

engineering approaches for preventing injury. Severe injuries can cause various
degrees of life change or death. The innate desire to avoid injury along with the
value placed on a single human life has fueled safety regulations in vehicle and road
design. Generally, the efficacy of solutions in safety engineering has relied on the
explanatory power of injury mechanisms for a given injury event.
In general, injury results from the transmission of a critical amount of energy into
the body. An injury mechanism can be defined as a description of either the cause of
energy transmission resulting in injury or the circumstantial effects that lead to
unsustainable stresses or strains (mechanical failure). Beyond a mechanistic under-
standing of injury, injury criteria describe the relationship between a measureable
physical response and the risk of injury. Injury criteria are often formulated from an
understanding of the injury mechanism, or the injury mechanism is discovered
through finding the “best fit” between the injury risk and a collection of physical
responses. Injury tolerance can be defined as the point at which an injury criterion
meets a threshold that is specified based on a statistical probability representing risk
(cf. Section ▶ Head and Neck) (Hardy et al. 2015).
Validating injury mechanisms in the lab can often be difficult due to biomechan-
ical dissimilarities between a healthy human that has an active musculoskeletal
response and a postmortem human surrogate (PMHS) that passively responds.
Contributing factors such as muscle contractions tend to confound estimation of
injury mechanisms. The musculoskeletal response needs particular attention espe-
cially when the injured party can anticipate a given traffic impact. Consequently,
injury mechanism hypotheses are often deemed as confirmed (or disconfirmed)
through inferential approaches using forensic evidence from real-world crashes in
conjunction with PMHS experimentation.
Injury Mechanisms in Traffic Accidents 3
Epidemiology
In 2002, traffic crashes were the 11th leading cause of death worldwide (2.1% of all
deaths and 25% of injury-related deaths) contributing 20–50 million injuries and 1.2
million deaths per year (Kauvar and Wade 2005). By 2013, the estimated nonfatal traffic
injuries had risen to 102 million (>100% increase) and fatal to 1.5 million ( 25%
increase) (Haagsma et al. 2016). The Global Burden of Disease (GBD) has developed a
metric called the DALY (disability-adjusted life years), which quantifies the years of life
lost and disability years due to injury, disease, violence, etc. (where an increase in the
DALY corresponds to an increase in death and injury rates). From 1990 to 2013, DALY
values have decreased in rich nations (the most substantial being – 67% in Asia Pacific
and – 61% in Western Europe), while overall changes in DALY values were not
significant in developing nations indicating essentially no change. Four-wheeled injuries
significantly increased in South Asia (+22%) and sub-Saharan Africa (+20%) (Haagsma
et al. 2016). These epidemiological data indicate enhanced safety design across major
automobile manufacturers as well as improved road conditions and traffic management.
However, governments in developing nations have virtually ignored road and traffic
conditions.
Injuries incurred from traffic crashes are a significant cause of morbidity and
mortality. The continued burden of traffic crash injuries involves rigorous experi-
mental and forensic biomechanics to better understand injury mechanisms. Traffic
crash biomechanics have the end goal of adapting road and vehicle design to ever-
changing traffic conditions.
Current Approaches and Data Analysis
Injury Severity Quantification

A variety of methods have been developed for quantifying injury severity over the
years, but it remains a difficulty to standardize an objective method for injury
severity. Currently, injury severity metrics are based on consensus as well as
historical data. There remains a need for precision in injury severity quantification.
In research, defining injury risk relies heavily on the Abbreviated Injury Scale (AIS)
for both quantifying historical injury severities as well as model development (e.g.,
probability of death score; PODS). Injury severity scores are an important measure
for biomechanical research used for retrospective analyses that enable engineering
advancements with the aim to minimize injury risk from traffic-related incidents.
AIS
The Abbreviated Injury Scale (AIS) of the Association for the Advancement of
Automotive Medicine (AAAM) is a numerical rating system for quantifying injury
severity in motor vehicle crashes (MVCs). AIS is a proprietary classification system
and its use requires trained personnel to properly code a victim. As a result, not all
traffic-related incident victims are coded according to the AIS. Currently, AIS is in
its sixth revision, and it is the most widely used metric for coding injury severity. The
AIS code consists of a 6-digit value followed by a single-digit value (i.e., 123456.1).
The first “pre-dot” number identifies the body region, type of anatomical structure,
its specific structure, and level (Gabler et al. 2015). The second component of the
score has a range between 0 and 6 corresponding to no injury up to the most severe
injury, respectively. High AIS ratings indicate injuries with high mortality rates, as it
is essentially a threat-to-life scale. The AIS is implemented for all body regions, each
with different methods for quantifying injury severity. For example, AIS level for a
vertebral body compression fracture depends on the height of the vertebra after
injury, and the AIS level for a skeletal chest injury is a function of the number of
sustained rib fractures (as well as the presence of a flail chest for AIS 3). AIS
levels also apply to soft tissue damage, which depend heavily on the organ that was
damaged.
ISS, NISS, and PODS

The Injury Severity Scale (ISS) is a measure of the probability of survival. The ISS
value is simply calculated by finding the maximum AIS values from each body
region (eight body regions total: head, neck, face, chest, abdomen, spine, extremity,
and external) then summing the squares of the top three maximum AIS values. This
calculation gives the ISS metric a range of 1 to 75. The goal of the ISS level is to
quantify the overall bodily injury severity.
The New Injury Severity Score (NISS) incorporates two modifications of the ISS:
(1) the sum of the squares of the top three ISS scores no matter the body region and
(2) if any AIS score is 6, then NISS is automatically set to 75. ISS is more widely
used, but NISS has been shown to better predict hospitalization durations and organ
failure (Balogh et al. 2000, 2003; Gabler et al. 2015).
The Probability of Death Score (PODS) is an estimate of probability of survival.
PODSa is the same estimate while accounting for victim age. The two probabilities
are quantified as follows.
ex
PODS ¼ PODSa ¼ (1)
1 þ ex
for PODS
x ¼ 2:2ðAIS1 Þ þ 0:9ðAIS2 Þ 11:25 þ C (2)
for PODSa
x ¼ 2:7ðAIS1 Þ þ 1:0ðAIS2 Þ þ 0:06ðAGEÞ 15:4 þ C (3)
Where C = 0.764 for cars, and AIS1 and AIS2 are the first and second highest
AIS values. The equations above have been derived from an empirical data fit
(Somers 1983a, b). The advantage of PODS over ISS is that it describes real
historical data and its value has an explainable meaning by virtue of its basis in
probability.
Fig. 1 Example of a human injury probability curve where impact force is the independent variable
for estimating probability/risk of injury
Human Injury Probability Curve

A human injury probability curve (HIPC) is an estimate of the injury risk with
respect to a defined variable called the injury criterion. Real-world forensic data
provides knowledge of possible injuries. Safety engineering aims to minimize risk of
these known injuries. Quantifying injury risk is left to biomechanics research for
identifying injury mechanisms, establishing injury criteria, designing PMHS exper-
iments, and quantifying human tolerance through statistical analyses. Injury criteria
are the independent variables for which injury risk is estimated, and these depend on
both the method for processing experimental PMHS data (Yoganandan et al. 2014b)
and the statistical methodology for estimating risk (Parr et al. 2013; Yoganandan
et al. 2014a, 2015a). Local criteria such as stress and strain or global criteria such as
force and moment are often selected based on principles of failure mechanics
(Fig. 1). In this sense, an injury risk of 50% means that one-half of the considered
population has a tolerance limit lower than the corresponding value of the indepen-
dent variable (or injury criterion). Injury risk estimates or HIPCs do not translate
well across differing loading conditions or body regions. As a result, many injury
risk curves are necessary to assess whole body risk under well-defined conditions.
For example, risk of femur fracture under 3-point bending does not translate to
fracture risk during axial loading of the femur.
Human tolerance or risk of injury (c.f. Fig. 1) is computed using statistical
techniques such as survival analysis. A survival analysis approach lends itself to
injury risk estimation because of the nature of datasets to which the HIPC is fit
(McKay and Bir 2009; Parr et al. 2013; Yoganandan et al. 2013b). Datasets that
contain human injury and noninjury experiments contain various types of censored
(uncertain) data, which represent injurious and noninjurious experiments. A survival
analysis has been commonly carried out for clinical studies where the time of
survival needs to be estimated following diagnosis of a fatal disease. The time of
survival is considered censored when the date of death is unknown, and the survival
analysis incorporates statistical techniques to account for datasets with censored data
(Pintar et al. 1998; Funk et al. 2002; Yoganandan et al. 2014a; Petitjean et al. 2015;
Yoganandan et al. 2015a). For HIPC generation, an injury criteria variable is used for
estimating risk instead of time of survival.
Explaining Mechanisms Through Forensic Data

Retrospective studies rely heavily on traffic crash, diagnoses, and patient outcomes.
Databases of field data are managed by government departments and fall into one of
three categories (Gabler et al. 2015): (1) fatal crash data, (2) details of crash
investigations, and (3) higher-level data of all fatal and nonfatal crashes. This chapter
will give a brief overview of popular databases that are managed in the United States
(Gabler et al. 2015), and the reader is encouraged to seek other sources for more
in-depth descriptions of internationally available tools.
CIREN
Although international databases have been developed, the Crash Injury Research
Engineering Network (CIREN; managed by National Highway Traffic Safety
Administration) and National Automotive Sampling System (NASS) contain vast
amounts of forensic data. Multiple CIREN centers in the United States collected
crash and medical data from as early as 1996. CIREN has enabled detailed retro-
spective analyses from medical and biomechanical perspectives (Augenstein et al.
2000; Augenstein and Diggs 2003; Kirk and Morris 2003). For example,
Yoganandan et al. (2009) used CIREN data to survey a database of over 1,800
traffic crash victims and found that diffuse axonal injury occurred in crashes where
victims experienced contact-induced blunt force head trauma (Yoganandan et al.
2009).
NASS
The National Automotive Sampling System (NASS) is a database containing
detailed characteristics of more than 5,000 crashes each year. This database contains
data from crashes with a range of severities from no injury to minor to fatal, which
involved cars, light trucks, vans, and/or sport utility vehicles. The NASS database is
populated by NHTSA crash investigators who document data and evidence from
crash sites through forensics, photography, interviews, and hospital injury severity
codes.
FARS
The Fatality Analysis Reporting System (FARS) is a comprehensive database of all
traffic related fatalities in the United States. This database includes traffic crashes of
all vehicle types including bicyclist and pedestrian fatalities. FARS contains less
specific data than the NASS, which includes over 400 data elements per crash
compared to about 175 elements in FARS. For a crash to be included in the FARS
database, at least one fatality must occur within 30 days of the incident. Injuries are
coded using the coarse KABCO scale (killed, incapacitated, moderate injury,
complaint of pain, or property damage only). FARS has been maintained by NHTSA
since 1975 and contains data from 30,000 to 40,000 fatal accidents.
State of the Art
Elucidating injury mechanisms from MVCs requires robust experimentation and

data analysis in applied impact mechanics. Generally, exploratory research within
the field of biomechanics is rare since the field has evolved to focus heavily on
applied biomechanics for injury prevention/minimization. As a result, analytical
techniques for estimating injury risk have improved significantly (Pintar et al.
1998; Petitjean and Trosseille 2011; Takhounts et al. 2013).
New PMHS experimental methodologies are consistently published to estimate
biomechanical responses under simulated impact conditions. Multiple PMHS exper-
iments are performed with the aim to generate a mean response or response corridor
with high biofidelity. Injury and noninjury responses are then used for deriving
injury criteria, which enables injury risk estimation under defined loads. Since
PMHS specimens are unidentical, data censoring approaches are necessary to
estimate risk for the average human. The survival analysis can handle censored
data types. It is a fairly new approach to estimate injury risk along with a confidence
level of that risk, which has been called the human injury probability curve (HIPC)
(cf. Sec. Human Injury Probability Curve).
The field of applied biomechanics has advanced significantly where experimental
setups can closely simulate MVCs including impacts from rollover (Lessley et al.
2014), oblique (Yoganandan et al. 2015b), and cyclist/pedestrian (Matsui and
Oikawa 2015) collisions. High-speed videography when integrated with digital
image correlation (DIC) (Anuta 1970; Chu et al. 1985) provides the means to capture
widespread strain (and strain rate) in large-scale applications (e.g., vehicle deforma-
tion) or small-scale experiments (e.g., bone or ligament strain) under highly dynamic
or quasistatic loading conditions (McCormick and Lord 2010; Begonia et al. 2015).
Head and Neck
Head and Brain
Brain Injury
It is generally accepted that traumatic brain injury (TBI) results from sudden
movement of the head (usually caused by blunt impact), resulting in both linear
and angular acceleration. Studies show that these two modes of acceleration contain
fundamental differences in terms of their injury mechanisms.
Diffuse axonal injury (DAI) is the most common TBI where sharp, dynamic head
movements directly cause neuron damage leading to neurologic sequelae. A study
from 2009 found that a survey of 1,823 cases contained brain traumas to
67 (41 adults) vehicle occupants that were coded as DAI (3.6%; no crash resulted
in more than one DAI) (Yoganandan et al. 2009). Within the adult sample size,
33 were lateral crashes (80%), and all DAI occurrences involved head contact loads.
Hardy et al. (2001) were able to show that pure angular acceleration (20–25 ms
duration) exposes white matter to tensile strains resulting in damaged axon fibers
(Hardy et al. 2001), while Anderson et al. (2003) observed that DAI severity was
correlated with peak linear and angular accelerations in the sheep model (Anderson
et al. 2003). Neural damage occurred ipsilateral and contralateral to the impact,
which suggests that brain matter undergoes shear and/or tensile stresses due to its
abrupt displacement relative to the skull. Axonal stretching under angular acceler-
ation still requires further validation as an injury mechanism, but it is the most
reasonable hypothesis for DAI. Linear acceleration subjects the brain to a pressure
wave beginning with compression where the cells nearest to the impact experience
the highest compressive forces. An effective fluid percussion device has been used in
dogs to show that a pressure wave causes concussion (Gurdjian et al. 1955). These
waves varied in duration and pressure from 1 to 46 ms and 34.5 to 345 kPa,
respectively. VandeVord et al. (2012) were able to show that externally applied
hyperbaric blasts cause significant neurotrauma in rats, which lead to cognitive
deficits from diffuse glial cell damage (Vandevord et al. 2012). Short duration
(5–10 ms) angular accelerations from blunt impact can also rupture veins or arteries
causing acute subdural hematoma (ASDH) resulting in cell damage (Davceva et al.
2012).
Acute Subdural Hematoma

ASDH is one of the most deadly injuries having a mortality rate of greater than 50%
across studies. The ischemia resulting from vascular damage causes neuronal dam-
age, and head trauma patients die or become seriously disabled (Wilberger et al.
1991). ASDH is found in victims of blunt force trauma and is the most grave of
injuries due to its high incidence (30%), high mortality (60%), and injury severity
(common Glasgow Coma scores range 3–5) (King 2015). Data seems to point to a
50% risk of ASDH subsequent to experiencing a severe TBI (>9 Glasgow Coma
score) based on a study from 17 Austrian centers, which documented the injuries and
outcomes of 360 patients (Leitgeb et al. 2012). Among those patients with an ASDH,
47% died in the hospital, 19% survived with “unfavorable” outcome, and 32%
survived with a “favorable” outcome.
Bridging vein rupture has been viewed as the injury mechanism for ASDH.
However, as King points out, this mechanistic explanation seems unsound since
fluid mechanics principles are violated by virtue of the adhesive resistance at the
interface of the dura and arachnoid (King 2015). Blood would have to enlarge the
space between dura and arachnoid before it is visible through medical imaging.
Blood would flow into the superior sagittal sinus if it were to flow in the path of least
resistance (and it will), which implies that there is insufficient pressure for an ASDH
to form at the dura-arachnoid interface. While the evidence is strong for ASDH
formation following bridging vein rupture (Gennarelli and Thibault 1982; Depreitere
et al. 2006), this mechanism is difficult to defend in terms of fluid mechanics. Four
hypotheses have been proposed by King (2015): (1) almost all ASDHs from impacts
result from cortical arterial rupture at the dura-arachnoid interface where arachnoid
border cells need to separate to tear cortical arteries, (2) radial separation of the
border cell layers occur when under dynamic skull deformation from direct impacts,
(3) angular acceleration subjects the border cell layer to shearing deformation
causing cortical artery rupture, and (4) the bridging vein rupture is only correlated
to the events surrounding the formation of ASDH but does not cause it (King 2015).
The first hypothesis would provide a sensible explanation for ASDH formation, the
second provides a reason as to why ASDH forms at sites remote from the impact
since skull deformation does not occur only locally, the third is more unlikely due to
the “slow” motion of the brain surface when exposed to angular accelerations under
10,000 rad/s2 at the head center of gravity, and the fourth is the logical consequence
if the bridging vein rupture hypothesis is false.
Cervical Injury
Epidemiology
Traumatic spinal cord injuries (TSCIs) from MVCs represent the majority of spinal
cord trauma cases per capita on an international level (Haagsma et al. 2016). With
the exception of poor regions and regions that contain with highly dense populations
(Tropical Latin America, South Asia, Oceania, and Eastern Europe; regions per
World Health Organization), traffic-related incidents involving four-wheeled vehi-
cles, motorcycles, bicycles, or pedestrians account for approximately 50% or more
of documented TSCI cases on a per region basis (Sekhon and Fehlings 2001;
Middleton et al. 2012; Lee et al. 2014). TSCI cases from traffic-related incidents
in developed regions remains either stable or reduced, while underdeveloped regions
fail to constrain the rise in TSCI cases due to poor traffic conditions and diminished
vehicle safety standards (Haagsma et al. 2016).
Neck Pain and Whiplash

Neck pain is the most commonly reported symptom following a rear end vehicular
collision. Evidence suggests that 50% of victims of whiplash report pain 1 year after
the injury, where greater initial pain, number of symptoms, and degree of debilitation
predicted recovery rates (Carroll et al. 2009).
Nociceptors are nerve endings that act as pain receptors, and they have a high
stimulus threshold to action potential initiation compared to other nerve endings.
The signals from nociceptors are sent to the spinal cord and brain by which pain is
perceived. Intervertebral discs often become enflamed from whiplash injury, which
has the effect of lowering the stimulus threshold of surrounding nociceptors causing
an increased sensitivity to pain. The consequent pain sensations will then be
amplified under reduced loading conditions giving the perception of chronic pain.
Additionally, nociceptors become more concentrated and numerous around the
degenerated region in the soft tissue. Within the intervertebral space, nociceptors
reside in facet capsules, spinal ligaments, tendons, and muscles (Deng et al. 2000).
Barnsley et al. (1995) and Bogduk and Marsland (1988) provide clinical evidence of
cervical facet pain in patients with neck pain following whiplash (Bogduk and
Marsland 1988; Barnsley et al. 1995).
A number of hypotheses have been proposed for the precise mechanism of
whiplash that leads to chronic pain after initial injury. Anatomical complexities of
the neck make it difficult to converge on a hypothesis without conflicting premises
(Siegmund et al. 2001, 2008). Yang and Begeman (1996) proposed a shear force
hypothesis where forces are transferred up the cervical spine to the occipital con-
dyles (Yang and Begeman 1996). The torso is thrust forward and its momentum pulls
the head forward subjecting the intervertebral space to shear especially at lower
levels where the facet angle is farther from the vertical. In this case, the pain is
attributed to straining facet capsules. Deng et al. (2000) as well as Lu et al. (2005)
confirm the hypothesis that painful signals are received from nociceptors surround-
ing the facet capsule and ligament resulting from shear forces in the neck (Deng et al.
2000; Lu et al. 2005). Their PMHS and animal experiments show that the neck is
subjected to compression, tension, shear forces, flexion, and extension throughout
the duration of the impact. The shear force hypothesis has been further corroborated
through computational models (Stemper et al. 2004; Panzer et al. 2011; Fice and
Cronin 2012).
Literature seems to indicate that the mechanism of injury discussed here is quite
definitive. Close attention can be given to the engineering of methods to best prevent
whiplash injury unless compelling evidence is put forth that disconfirms the shear
force hypothesis.
Cervical Spine Injury

Since the most severe injuries of the cervical spine involve vertebral body fractures
or dislocations, this section is limited to an overview of vertebral body anatomy of
the cervical spine. Additional information regarding anatomical detail is left to other
sources (Nightingale et al. 2015). The cervical spine is comprised of seven vertebrae
where the occipital condyle (OC) sits on top of C1 (OC-C1 joint) and C7 rests on top
of T1 (C7-T1 joint) (Fig. 2). The cervical spine has a slight curve, called lordosis,
during nominal posture. The upper cervical spine (UCS; OC to C2) is anatomically
distinct from the rest of the cervical spine. Technically, C1 does not have a vertebral
body but is a ring with distinguishably larger facets (Fig. 2). The UCS is capable of a
greater range of motion than the rest of the spine. Damage to the UCS is almost
unsurvivable due to the vulnerability of the spinal cord in the UCS. As a result, data
on UCS fractures is sparse since victims are not rushed to the hospital (Yoganandan
et al. 1989). The C1 vertebra (also known as the “atlas”) is susceptible to a multipart
fracture, and fatalities are common with a four-part fracture (Jefferson 1919).
Vertebral bodies in the lower cervical spine (LCS; C3 to T1) are uniformly shaped
and have lesser range of motion than those of the UCS. Burst fractures, dislocations,
and fracture dislocations are common to the lower cervical spine vertebrae, which
include fractures of the cortical bone, endplate fractures, and loss of disk height.
Injuries to the spinal cord are less common with fractures in the LCS compared to the
UCS (Fig. 3).
Fig. 2 Cervical vertebral bodies
Much consideration has been devoted to clinical classifications of neck injuries

(Allen et al. 1982; Torg 1985; Yoganandan et al. 1989; Myers and Winkelstein
1995), but the complexity of the cervical spine leads physicians and researchers to
necessarily base classifications on inference and anecdotal evidence (Nightingale
et al. 2015).
Relatively speaking, the neck is strong during compression, but headfirst impacts
from small drop heights (0.5–2 m) have produced a variety of neck injuries in PMHS
experiments (Nusholtz et al. 1981; Yoganandan et al. 1986). Neck compression
injuries arise from conditions where the victim lands headfirst such as vehicle
occupant ejection, motorcyclist ejection, or vehicle rollover. The fracture level and
type are a function of the buckling mode during injury, which is subject to the initial
orientation of the spine (Nusholtz et al. 1981, 1983), or the degree of neck lordosis
(Culver et al. 1978). Compression injury mechanisms are further complicated by the
decoupling response of the head and neck (Yoganandan et al. 1991). Yoganandan
et al. (1991) were among the first to use whole cervical spines under compressive
loads and suggested that the neck behavior during buckling influences injury.
Vertical impact tests on well-constrained head and neck specimens without lordosis
(straightened necks where the occipital condyle was approximately concentric with
T1) were performed by Pintar et al. (1995), which produced a wide range of injuries,
and the complex neck behavior revealed no single metric as an injury predictor
(Pintar et al. 1995). Additionally, Nightingale et al. (1996a, b, 1997) performed a
variety of experiments with differing postures and impact angles and quantified
compressive failure loads (Nightingale et al. 1996a, b, 1997). A wide range of
POSTERIOR ANTERIOR
Facets
Intertransverse
ligament
Capsular ligament
Anterior longitudinal
ligament
Anterior one-half
annulus fibrosus
Interspinous and
Posterior one-half
supraspinous ligaments
annulus fibrosus
Ligamentum flavum
Posterior longitudinal ligament
Fig. 3 Vertebral body with ligaments
injuries were produced including midlevel burst fractures, odontoid fractures, and
Hangman’s fractures. Midlevel fractures are best explained in terms of the complex
kinematics of first and second order buckling of the neck.
Neck tension-extension injuries in the cervical spine are much less common than
compression injuries but are more likely to have higher injury severity and have a
fatal outcome. Etiologically, extension injuries have been found in victims who were
not wearing a seat belt or were too close to the airbag when it deployed. Tension-
extension injuries are very difficult to reproduce since neck musculature has a
substantial influence on the presence and/or nature of the injury itself (Chancey
et al. 2003). The maximum tensile tolerance of the neck essentially is increased by
19% when the occupant anticipates the injurious event by activating neck muscles.
Regardless, the ligamentous cervical spine will fail at lower loads, but the neck
tensile tolerance is increased during muscle activation. Some PMHS experiments fail
to reproduce injuries observed in real-life traffic accidents due to the relaxed
musculature of the cadaver.
It remains to be seen how each parameter that defines the initial conditions prior
to injury affect the characteristics of the resulting injury. However, the body of work
on this topic suggests that injury traits are influenced by the orientation of the neck,
head, and subject-specific anatomical characteristics (e.g., degree of lordosis) imme-
diately before injury.
Thorax
Introduction
Chest injury ranks just behind head injury in overall number of fatalities in traffic
accidents (Services UDoHaH 2007). During a vehicle collision, the thorax is
exposed to vehicle interior components including restraint systems, each of which
poses varying risks. Thorax injuries are common in frontal and side collisions as well
as their oblique counterparts. A retrospective analysis by Nirula and Pintar (2008)
shows that the incidence of severe chest injury ( AIS 3) was 5.5% and 33% in
NASS and CIREN, respectively (Nirula and Pintar 2008). The steering wheel, door
panel, armrest, and seat were all identified as contact points with substantial risk of
severe injury to the thorax.
The reader is encouraged to look to other sources for an overview of thorax
anatomy in the context of injury mechanisms (Cavanaugh and Yoganandan 2015).
The following survey of injury mechanisms will refer to various regions of the
thorax with very little review of anatomy basics.
Traumatic Rupture of the Aorta
Background
Though traumatic rupture of the aorta (TRA) does not happen frequently, it has the
highest fatality rate in traffic crashes of all injuries to the thorax. Crash data from
1998 to 2006 shows that TRAs occurred in approximately 1% of all vehicle
occupants in traffic-related incidents, but was the cause of 21% (8%) of all
fatalities (Shkrum et al. 1999). In the occurrence of a TRA, there is only a brief
time-window where the injury can be treated before it becomes fatal. Bertrand et al.
(2008) found that TRAs were twice as common in occupants involved in side impact
(2.4%) compared to frontal impact (1.1%) in the United Kingdom (Bertrand et al.
2008). The difference in TRA incidence from side impact happened after the advent
of air bag and seat belt restraints to prevent frontal crash injuries.
Fig. 4 Heart anatomy of

interest
Anatomy
The aorta extends from the base of the left ventricle of the heart at the aortic root. The
ascending arc of the aorta is relatively flexible while the descending aorta is secured
to the thoracic spine via the pleural reflection. The peri-isthmic region is between the
anastomosis of the left subclavian artery and the descending aorta. The peri-isthmus
is the most common place for TRAs to originate (Fig. 4).
Mechanism
Tears in the aorta have occurred in the peri-isthmic region for an estimated 94% of all
TRAs (Katyal et al. 1997). Past PMHS studies struggled to reproduce TRAs until
Hardy et al. (2006) was able to demonstrate that TRAs can be induced in PMHS
specimens through longitudinal quasistatic loading (Hardy et al. 2006). This study
was also the first to orient the specimen to place the diaphragm, heart, and aorta in an
anatomically consistent way with a healthy human. Interestingly, the aorta was
subjected to tension without having to induce whole body acceleration. Circumfer-
ential tears were found to be almost ubiquitous among the observed TRAs. Both the
intima (innermost layer) and media (middle, thicker layer) of the aorta were found to
be sensitive to tearing (Cammack et al. 1959).
The accepted mechanism of injury for TRA is tension in the aorta, which
apparently develops immediately after (not during) direct chest impact. Hardy
et al. (2008) used radio-opaque markers to track the response of the peri-isthmus
region while acquiring high-frequency x-ray, and they point out specific catalysts of
TRA: (1) deformation of the thorax, (2) elongation (longitudinal stretch) of the aorta,
and (3) the tethering of the aorta to the spine, which promotes stretching (Hardy et al.
2008). Here, PHMS specimens were impacted in different modes: shoveling, side
impact, submarining, and combined. All impact modes caused TRAs by transverse
tears and two oblique tears.
Thoracic Spine Injuries
Injuries to the thoracic spine have predominantly originated from traffic-related

incidents (Robertson et al. 2002b; Leucht et al. 2009), which could be a surprising
statistic since it is difficult to imagine how the thoracic spine could be exposed to
such forces in a vehicle passenger in a seated position. In the United States, 8% of
spine fractures from MVCs are AIS 3, and 9% of all thoracic spine injuries are
AIS 3 (Wang et al. 2009). In a retrospective data analysis from the United
Kingdom, almost 23,000 patients were surveyed for motorcycle and MVC victims
(Robertson et al. 2002a). Spinal injuries were present in 126 (11.2%) motorcyclists
and 383 (14.1%) car occupants. The thoracic region was the most common spine
injury in motorcyclists (54.8%; n = 126), while thoracic spine injuries in car
occupants were present in 26.6% (n = 383) of spinal injury cases. Similarly, Pintar
et al. (2012) analyzed CIREN and NASS data from MVCs and found the dorsal
spine to be particularly vulnerable despite public awareness and seatbelt use in the
United States (Pintar et al. 2012) (Fig. 5).
Compression Fracture
Perhaps nonintuitively, injuries related to spine compression are often observed in
frontal and rear-end crashes or in the presence of abrupt decelerations/accelerations.
Begeman et al. (1973) simulated frontal crash responses in PHMS using a sled
(Begeman et al. 1973). Three-point seat belts (with shoulder strap) were found to
promote spine compression compared to a single lap restraint. Begemen et al. found
that a deceleration of about 15 G’s results in over 600 lbs of shoulder belt tension and
a spine load of about 900 lbs. Additionally, it was found that the axial loads are
augmented when the body is held erect by the shoulder restraint. Yang and King
(1995) hypothesized that the shoulder strap restrains the upper body before the
forward momentum of the body can be sufficiently diminished (King and Yang
1995). The combined force from the forward momentum and the upper restraint acts
on the thoracic spine forcing it into a straightened or lordotic posture, which pro-
duces dangerous loads on more caudal segments. Furthermore, the asymmetric
3-point belt restraint will also load the spine asymmetrically exposing vertebral
bodies to concentrated loading conditions.
Forensic clinical studies provide a real-world risk of spine compression injury
from frontal crashes (Ball et al. 2000). Burst fractures between L1 and T12 levels
were found in 80% of patients where 3-point seat belts were fastened while in only
25% of patients wearing a lap belt. Nonetheless, a recent retrospective study (Pintar
et al. 2012) points out that despite the frequent occurrence of thoracolumbar frac-
tures in frontal impacts, the precise injury mechanism remains elusive.
Fig. 5 Spine load in PMHS

during horizontal deceleration
from a frontal MVC
Flexion Distraction Injury

Flexion-distraction injuries (or Chance fracture (Chance 1948)) have been found in
vehicle occupants involved in MVCs, and they are not limited to the thoracic spine.
For example, Chance fractures have been reproduced through airbag deployment in
PMHS experiments (Cheng et al. 1982). The thoracolumbar region seems to be
especially exposed to the possibility of Chance fractures during a MVC. The
mechanism of injury involves fracture initiation in the posterior aspect of the neural
arch and continues anteriorly (Fig. 6). The spine experiences hyperflexion followed
by a distraction (Stemper et al. 2015). This injury has been especially common in
occupants wearing only a lap belt or an improperly secured 3-point restraint.
A Canadian study analyzed medical data of eight children involved in an MVC
wearing lap or 3-point restraints (Santschi et al. 2005). Of the five children wearing a
Fig. 6 Flexion-distraction
injury (Chance fracture)
lap belt, four experienced flexion-distraction fractures to the lumbar spine and three
were permanently paralyzed. Flexion-distraction fractures were accompanied with
intra-abdominal injuries, and this coincidence seems to be a pattern (LeGay et al.
1990). The high incidence of spinal injury and intra-abdominal injuries that result
from wearing a lap belt restraint was apparently enough motivation for the United
States to prohibit the lap belt design in all cars sold in the United States since
September of 2007 (NHTSA 2005).
Intra-abdominal Injuries
Anatomy
The abdomen is conventionally described through nine regions: (R/L) hypo-
chontriac, (R/L) lumbar, (R/L) iliac, epigastric, umbilical, and hypoastric (Fig. 7).
This section will focus on anatomical features pertinent to traffic accident injury, but
for a detailed description of abdomen anatomy, the reader is encouraged to look to
other sources (Hardy et al. 2015).
The lower ribs offer protection from blunt trauma to the upper abdomen, and the
anterolateral abdominal wall (skin, subcutaneous tissue, muscle, fascia, and parietal
peritoneum) provides additional protection to the abdominal viscera.
Soft tissues and organs (Fig. 8) are tethered through the mesentery, which is made
up of two layers of peritoneum either between organs or tethering viscera to the
abdominal wall. A certain amount of movable freedom is granted to abdominal
Fig. 7 The abdominal region
organs based on the length of the tethering mesentery, which protects from robust
vibrations and abrupt accelerations.
The stomach, small intestine, large intestine, and gallbladder are classified as
hollow or membranous organs, and these organs are especially prone to serosal tears
and perforations in MVCs (Shinkawa et al. 2004). Abdominal organs that are
classified as solid include the liver, spleen, kidneys, and pancreas. Solid organs
tend toward increased injury severity in traffic crashes, and the liver and spleen are
the most frequently injured abdominal organs in MVCs (Klinich et al. 2010; Hardy
et al. 2015).
The liver is highly vascular and fluid-filled, and it is the largest internal organ in
the body located in the upper right region of the abdomen. The liver is vulnerable to
blunt trauma and bone fragments from a rib fracture can puncture, and depending on
the magnitude of the impact, rupture or burst.
The spleen is located in the upper left region of the abdomen, and it receives
protection from the lower rib cage from blunt impact. Spleen rupture is the most
common injury in traffic crashes.
Background
The abdomen is a highly complex region of the body where the effects of different
loads or impacts on individual organs are difficult to quantify (Yoganandan et al.
2000). A study that analyzed NASS data from 1993 to 1998 found that approxi-
mately half of 129,269 abdominal injuries resulted from front-end collisions
Fig. 8 Soft tissues and organs of the abdomen
(Yoganandan et al. 2000). AIS-6 abdominal injuries were present in 94 vehicle

occupants and occurred from right (22%) and frontal (78%) impacts. Of the abdom-
inal trauma cases, 31% were spleen injury, 30% were liver injury, and ~33% of all
injuries were coded AIS 3. Frontal collision MVCs yielded injuries most common
in the liver (39%), spleen (29%), and digestive organs (11%). Other common injuries
(>AIS-2) were of the kidney, diaphragm, arteries, and urogenital systems. The
abdomen is especially vulnerable to injury during side and oblique impacts, and
the state of automobile safety has progressed to curtail abdominal injuries through
side air bags and side curtain air bags (Baur et al. 2000; Yoganandan et al. 2007).
Due to anatomical location, the spleen or the liver has been the most seriously
injured from far-side impacts (Augenstein et al. 2000).
Mechanisms
Ball et al. (2000) point out that there is a high probability of intra-abdominal injury
requiring laparotomy in patients injured while wearing only a lap belt compared to a
3-point restraint ( 60% vs. 25%) (Ball et al. 2000). Incidence of life-threatening
intra-abdominal trauma from wearing a lap belt during vehicle collision has been
reported in approximately 50% of flexion-distraction injury patients (Gertzbein and
Court-Brown 1988; LeGay et al. 1990; Green et al. 1991). Anderson et al. (1991)
reported a high rate of abdominal trauma associated with crashes where passengers
were wearing only lap belts (Anderson et al. 1991). Thirteen of 20 patients (65%)
required laparotomy, and eight of nine younger patients (>16y.o.) had life-
threatening intra-abdominal injuries. Anderson et al. note that children appear
more likely to experience abdominal injury from lap belts in the event of a collision.
Other literature is in agreement with the study by Ball et al. (2000), which found that
the most common injuries were small-bowel perforations and serosal tears.
A PMHS study from 2015 found that specimens experience lateral abdominal
deflections >50 mm in a vehicle with side and curtain airbags exposed to an oblique
side-impact collision at approximately 25 km/h (Yoganandan et al. 2015b). These
abrupt deformations of the thorax resulted in fractures to the four lowest ribs, which
increases the risk of liver or spleen puncture and dramatically increases the internal
pressure of abdominal organs facilitating rupture. Arm placement modestly influ-
ences the nature of injury during side impact to vehicle occupants, but it seems to
mainly affect the number of rib fractures and their locations (Kemper et al. 2008).
Injury criteria for the abdomen have been extensively studied where almost every
measurable mechanical response has been investigated for use as a predictor of
injury for the abdomen (Yoganandan et al. 2001). First and perhaps foremost,
correlations have been identified between the presence of injury (and severity) and
the amount of abdominal compression (Melvin et al. 1973; Stalnaker et al. 1973;
Viano et al. 1989; Lamielle et al. 2008; Hardy et al. 2015). The most common organs
(or regions) of the abdomen that have been injured during abdominal compression
are upper and lower abdomen, liver, spleen, jejunum-ileum, and pancreas. Second,
both the nature of the injury and its severity have been found to be considerably
sensitive to the change in velocity during impact or impactor speed (Melvin et al.
1973; Kroell et al. 1981; Yoganandan et al. 2000). This velocity criteria was then
taken a step further and combined with compression (or V*C), which was called the
abdominal injury criteria (AIC) (Rouhana et al. 1985). It was found that the severity
of abdominal injury was correlated well with AIC. Other variations of this criteria
have been proposed, such as Vmax Cmax (Stalnaker and Ulman 1985) and VC(t)max
(called the viscous tolerance criterion for thoracic impact) (Viano and Lau 1983),
each having good correlation with injury severity under various impact conditions.
Kroell et al. (1986) investigated the relationships between velocity, compression,
and heart rupture trauma from blunt impact to porcine subjects, and they found a
better correlation between Vmax Cmax over VC(t)max for probability of injury and
AIS 4 (Kroell et al. 1986). Lastly, force or the manner in which force is applied
influences the nature and severity of injury (Stalnaker et al. 1973; Trollope et al.
Trollope et al. 1973; Haffner et al. 1996). Talantikite et al. (1993) performed
25 pendulum impact experiments on excised human livers at various speeds, and
they found that peak forces (as measured from the pendulum) greater than 500 N
yielded injuries AIS = 3 or greater (Talantikite et al. 1993). In this same study, whole
body experiments appeared to have a tolerance threshold of 4.4 kN, which caused
deflection of half the abdomen. Studies have also been carried out using variants of
the force injury criteria such as impact energy, Fmax Cmax, and pressure. Organ
pressure correlations to injury seem to be limited to the liver (Foster et al. 2006),
kidney (Rouhana et al. 1985), and lower abdomen (Miller 1989). Sparks et al. (2007)
were able to show that internal pressure is a reasonable predictor of abdominal injury
but found that the product of peak internal pressure and the peak time-derivative of
pressure was the best predictor of liver injury (Sparks et al. 2007). These data
suggest that there is a 50% risk of injury with a vascular pressure of 64 kPa
(9.28 psi).
The abdomen seems to have a wide variety of injury mechanisms, which vary
with organ tissue. Consequently, mechanisms of abdominal injury have been sim-
plified through individual injury criteria.
Pelvis
Pelvic fractures are considered severe injuries, and pelvic injury is often accompa-
nied by damage to other organs. A high incidence of pelvis injuries exists among
traffic crash victims (Ooi et al. 2010). The mortality rate of occupants in MVCs with
pelvic fractures seems to indicate that pelvis fractures are accompanying injuries;
i.e., pelvis injuries seem to be only correlated to incidence of death (Gokcen et al.
1994; Petrisor and Bhandari 2005). However, evidence suggests that fracture related
hemorrhaging is associated with a high risk of death following pelvic disruption
(Gabbe et al. 2011).
Anatomy
The pelvis is a ring structure that supports the flexible spinal column and transmits
the stress of weight bearing to the lower extremities. It is surrounded by a complex
arrangement of muscles providing a thick compliant covering over the pelvis. The
pelvis is stronger in the vertical and longitudinal directions to bear loads during
walking and running. Conversely, it is weaker along the lateral direction due to small
pubis bones at the front of the pelvic girdle.
Four bones form the pelvis. Two innominate bones form the anterior and lateral
walls of the ring, and the sacrum and coccyx make up the posterior wall. Each
innominate bone consists of three fused segments, the ilium, ischium, and pubis. The
fusion occurs around a cup-shaped articular cavity called the acetabulum, or the
socket part of the hip joint. The ilium is divided into the large wing-like ala and the
body of ilium forming the superior part of the acetabulum. The anterior superior iliac
spine (ASIS) and the posterior superior iliac spine (PSIS) are bony prominences at
the anterior and posterior extremities of the iliac crest. Sacroiliac joints attach the
sacrum to the ilia through weight bearing synovial joints. The articular surfaces of
the sacroiliac joint contain irregular depressions that interlock the two bones allo-
wing limited joint motion.
The ischium forms the lower lateral part of the innominate bone, and it constitutes
the posterior third of the acetabulum cup. The lowest portion of the body is the
ischial tuberosity, which supports the upper torso in a seated posture.
The pubic bone is irregularly shaped and contains a body and two rami: the
superior and inferior pubic rami. The superior ramus extends from the body to the
mid-sagittal plane, where it articulates with the corresponding ramus on the opposite
side. The joint formed by the two superior rami is called pubic symphysis, which is a
slightly movable joint containing a cartilaginous disc between the two bones. The
inferior pubic rami join each other through the pubic symphysis.
There are two major ligament groups in the pelvic region: (1) the ligaments
surrounding the vertical load bearing sacroiliac joints, and (2) the ligaments passing
between the sacrum and ischium. The former consists of anterior sacroiliac liga-
ments, posterior sacroiliac ligaments, and interosseous ligaments. The latter consists
of sacrospinous ligaments and sacrotuberous ligaments.
Fracture Types
Fracture types are classified according to the energy input to the pelvis as well as
the direction of the force impact (Linnau et al. 2007). Pennal and Tile (1980)
classified pelvic fractures based on the presumed force vector direction (Pennal
et al. 1980), and Young and Burgess (1990) refined these classifications (Burgess
et al. 1990). Pelvis fractures have been generally characterized according to the
presumed loading vector as either anteroposterior compression (APC), lateral
compression (LC), vertical shear (VS), or combined mechanical injury
(CM) (Pennal et al. 1980). Each of these injury types has a commonly used
associative severity scale from I to III, where III is considered unstable or most
severe (e.g., APC-III or LC-III).
APC fractures in the anterior aspect typically demonstrate pubic symphysis
diastasis or a vertical fracture pattern of rami and posterior injuries defined by
subsets. APC-I shows slight widening of pubic symphysis and one SI joint but
have intact anterior and posterior SI joint ligaments. APC-II results in injury
widening SI joint anteriorly affecting anterior SI ligaments and ipsilateral
sacrotuberous and sacrospinous ligaments with intact posterior SI ligaments. Stabil-
ity of APC-II depends on various ligaments involved. APC-III represents complete
separation of hemipelvis from pelvic ring with rupture of anterior and posterior SI
ligaments. APC-III is again treated as severely unstable and involves disruption of
all sacroiliac joint ligaments (Young and Resnik 1990).
LC pelvic injuries are caused through impact to the proximal femur that transfers
a load to the iliac crest. Lateral compression fractures occur as the pelvis rotates
toward the midline along the impact vector. The ilium rotates medially while its
posterior is attached to the sacrum through posterior sacroiliac (SI) ligaments (Bur-
gess et al. 1990). A secondary crush is common within the hemipelvis on the impact
side, which causes an internal rotation; i.e., the contralateral hemi pelvis rotates
externally causing an “open-book fracture.” This injury is considered LC-III or
severely unstable since the compressive force also affects the contralateral side by
way of an external rotation of the anterior pelvis (Young and Resnik 1990).
VS injury results in a diastasis of the symphysis or a vertical fracture pattern of
rami anteriorly, and it vertically displaces the hemipelvis (Burgess et al. 1990;
Manson et al. 2010). Most of these fractures result from a fall from height and are
uncommon in MVCs.
Mechanisms
The majority of experimental PMHS work has been done for pelvis injury mecha-
nisms through direct lateral or frontal impacts to the pelvis (Yoganandan et al.
2013a). Therefore, this discussion of injury mechanisms focuses primarily on
those from side, oblique, and frontal impacts in MVCs.
Lateral Impact
Lateral impact injury mechanisms are pertinent to side or oblique impacts in MVCs.
During side impact MVCs, the impact is often delivered in a direct manner from the
intruding door. Historically, PMHS studies have exposed whole cadavers to impacts
along the lateral aspect of the greater trochanter of the femur. Pubic rami were found
to experience substantial strain during lateral impact (Molz et al. 1997). According to
Bonquet et al. (Bouquet et al. 1998), a 50% injury risk of AIS 2 is incurred when
the pelvis deflection (or change in maximum lateral width) reaches 46 mm (Bouquet
et al. 1998). While the risk of injury can be developed based on test measurements,
the nature of the injury depends on the properties of the impactor. Using
20 unembalmed male cadavers, Nusholtz and Kaiker (1986) found that the nature
of the injury depends on the rigidity of the impactor (Nusholtz and Kaiker 1986). In
absence of padding, pelvic damage occurred at or around the acetabulum whereas
the pubic area was prone to fractures from impact from a rigid device.
Schiff et al. (2008) studied 728 lateral impact crashes that resulted in pelvic
fractures and compared them to 5710 control cases involved in lateral impact
without suffering any pelvic injuries (Schiff et al. 2008). The authors found statis-
tically significant factors including occupant (1) age, (2) sex, (3) weight, and
(4) vehicle type: (1) occupants 65 years and above had 70% increased risk, (2) non-
pregnant females had 60% increased risk, (3) underweight occupants had 80%
increased risk, and (4) occupants of vans had 80% decreased risk. Magnitude of
lateral intrusion had the strongest relation to pelvic fracture.
Cavanaugh et al. (1990) laterally impacted 12 whole unembalmed cadavers in a
way where the energy was directed through the greater trochanter (Cavanaugh et al.
1990). The impact velocities ranged from 6.7 to 10.5 m/s. Injuries occurred to
inferior and superior left pubic rami and the left sacroiliac joint. Pelvic injury
tolerance was found between 8 kN and 10.6 kN. A peak force of 7.98 kN yielded
25% probability of injury. A peak compression of 32.6% corresponded to a 25% risk
of fracture. Similar to intra-abdominal injury studies (cf. Sec. 0), the best injury
criterion was related to Vmax Cmax (peak velocity times peak compression).
Likewise, Viano et al. (1989) performed a series of pendulum (23.4 kg) impact tests
on unembalmed cadavers with impact velocities of 4.5, 6.7, and 9.4 m/s where the
force vector was directed through the torso (Viano et al. 1989). They found that peak
pelvic acceleration and pelvic deformation were not reliable measures, but the ratio
of pelvic deformation to pelvic width correlated well with pubic rami fracture, which
was the only type of injury observed. The tolerance level for 25% probability of
serious injury to the pelvis was found to be 27% of pelvic compression, based on the
entire width of the pelvis.
Impact energy that is delivered through the greater trochanter may result in
substantial kinetic energy, which moves the contralateral hemipelvis against the
center console of the vehicle. Tencer et al. (2007) analyzed pelvis accelerations
and reviewed CIREN for pelvic injuries of occupants in vehicles with and without a
center console. They found a higher incidence of pelvis fracture in MVCs with
center consoles (Tencer et al. 2007). The abrupt deceleration from the center console
contributes to injuries in the contralateral hemipelvis. The maximum and minimum
accelerations for (1) fixed seat without console, (2) fixed seat with console, and
(3) movable seat without console were (1) 28.5 g and 3.3 g, (2) 24.8 g and 10.5 g,
and (3) 15.3 g and 3.8 g, respectively. A 50% reduction in primary pelvic
acceleration was observed for vehicles without a center console.
Frontal/Rear Impact
Occupants involved in head-on collisions in MVCs are exposed to a high risk of a
frontal load on the pelvis. Compared to lateral impacts, frontal impacts indirectly
load the pelvis where energy is transmitted from the knee, which is impacted by the
dashboard (King 2001). Generally, injury mechanisms differ between front and rear
vehicle occupants. A patella fracture is often accompanied by posterior hip disloca-
tion in front-seat passengers, which have a tendency to sit with their knees together
and hips flexed at 90 when the knee impacts the dashboard (Markham 1972). Rear
side passengers thrown forward on impact may suffer anterior dislocation of the hip
joint where higher energy loads can give rise to sacroiliac joint separation. This
injury pattern for rear seat passengers is due to their tendency to sit with knees and
hips flexed at 90 degrees and with hips in abduction with a slight external rotation
(Markham 1972).
Salzar et al. (2006) conducted a frontal impact study using an isolated hemi-pelvis
model to study the posture-dependency of resulting fractures of the acetabulum and
proximal femur (Salzar et al. 2006). Peak forces varied with the acetabular support
area, and the highest sustained forces occurred under abduction and extension of the
specimen (i.e., having the largest contact area) while adduction and flexion failed
under lesser forces (i.e., having the least contact area).
Interestingly, Masson et al. (2005) found the critical quasistatic load to strongly
depend on sex by testing seven male and five female isolated pelves from embalmed
cadavers (Masson et al. 2005). Quasistatic loads were applied in the anterior-
posterior direction to the symphysis through a rectangular rigid plate. The sustain-
able force sustained before collapse ranged from 556 to 3981 N with average peak
loads for female equal to 1053 N and for males 2501 N. The force-displacement
corridors indicated that the female pelvic ring was more fragile due to the larger
retro-pubic angle when controlling for bone density.
Though a majority of pedestrians are struck laterally, frontal loading to the pelvis
can occur when a pedestrian is hit from the front or back. After analyzing 1014 cases,
20.7% of the cases had injuries from anteroposterior compression (Eastridge and
Burgess 1997). Pedestrian injury patterns are primarily influenced by front-end
vehicle design. Simms et al. (2006) reviewed accident data and simulated
pedestrian-vehicle interactions between sport utility vehicles (SUVs) and cars
(Simms and Wood 2006). The probability of pelvis injury was 2.5 times as likely in
pedestrians when struck by a vehicle with an elevated bumper, as is common in SUVs.
Lower Extremities
Anatomy
The lower extremities include the femur head, acetabulum joint, thigh, femur, knee,
patella, tibia, fibula, ankle, and foot. It consists of 29 distinct bone types, 72 articu-
lating surfaces, 30 synovial joints, greater than 100 ligaments, and 30 muscle
attachments (Crandall et al. 1996).
The knee is made up of the patella, femoral condyles, and knee ligaments. The
thigh comprises the supracondylar, shaft, and subtrochanteric region of the femur.
The hip consists of the femoral head, neck, and acetabulum (the hip socket). Femur
anatomy is included in the knee, thigh, and hip. The patella bone (or knee cap)
hovers over the anterior knee, which acts as a mechanism to aid in leg extension and
knee protection.
Major knee ligaments include the posterior cruciate ligament (PCL), anterior
cruciate ligament (ACL), medial collateral ligament (MCL), and lateral collateral
ligament (LCL). The reader is encouraged to seek out other literature for in-depth
explanations of ligament anatomy and functionality if so desired (Rupp 2015).
The femur is a strong bone that spans the length of the thigh, and it consists of
medial and lateral condyles at its distal head that articulate along the superior end of
the tibia. The proximal end of the femur forms the trochanteric region from which
the femoral neck sprouts allowing rotation about the hip. From a CIREN database
study, from 1997 to 2003, the femoral shaft (31.5%) and the acetabulum (21.6%)
were the most commonly reported injuries within the knee-thigh-hip arrangement
(Rupp 2006).
The tibia and fibula are attached by articulating surfaces at its proximal and distal
ends. The ankle joint contains the talocrural joint, talocalcaneal joint, talocalcaneo-
navicular joint, and transverse tarsal joint.
The hindfoot rotates in three planes of motion through internal and external
rotation (transverse plane), dorsiflexion and plantarflexion (sagittal plane), and
inversion and eversion (coronal plane) (Salzar et al. 2015). A common mis-
conception is that the hindfoot is attached to the leg through hinge joints while the
actual motion of the hindfoot is not contained within any single plane. This
particularly applies to flexion of the hindfoot joints, which is not an isolated

movement but is linked to inversion and eversion.
Mechanisms
The knee-thigh-hip complex injuries appear to be the most commonly injured region
of the lower extremities in MVC, and they are also the most severe lower extremity
injury due to the high risk of arterial rupture (Kuppa et al. 2003). The lower
extremities are the most frequently injured in MVCs where 25% of these injuries
are toward the knee due to the presence of knee bolsters. Knee injuries often have are
associated with high DALY values due to long recovery times. Knee bolster stiffness
varies across vehicles and many are sufficiently padded transmitting the critical loads
to the femur and hip (Salzar et al. 2015).
Knee Injury
Knee injuries make up about 16% of all injuries to the lower extremities (Chang et al.
2008). Database studies indicate that few knee injuries during lateral impacts have been
found to be AIS > 2, and as a result, little attention has been given to lateral loading
mechanisms throughout literature (Kuppa and Wang 2001; Rupp et al. 2002). Patellar
fractures are the most common knee injury and frequently occur in frontal impacts
when the knee interacts with the dashboard or, in the case of rear seat passengers, the
seats in front of them. Fracture occurs after critical compression force between the
external impactor and the femoral condyles. Evidence suggests that risk of injury
increases especially in drivers who have a contracted knee extensor muscle for braking
(or bracing) immediately prior to the MVC (Atkinson et al. 1998). The patella is also
susceptible to a tension load when it impacts a surface obliquely, which subjects the
patella to a sliding motion and can cause a patellar ligament avulsion.
Femur Injury
About 39% of all knee-thigh-hip injuries are thigh fractures (Chang et al. 2008).
Injury to the knee is often accompanied by fractures to the femoral condyles because
the patella is pushed into the intracondylar notch, which disunites the condyles
(Powell et al. 1975). Femoral shaft fractures are the most frequent, and they are
commonly the product of axial compression. The curvature of the femoral shaft has a
weakening effect on its axial strength. Axial loads tend to induce a posterior-to-
anterior bending moment, especially from knee loading applied medially to the
femoral condyle (Viano and Khalil 1976; Viano and Stalnaker 1980; Rupp et al.
2003; Ivarsson et al. 2009).
The mechanisms of injury of the trochanteric femur due to axial loading remain
somewhat elusive, but it has been hypothesized that the mechanism is a combination
of tension within the greater trochanter, the anchoring of the femoral head to the
acetabulum, and the moment arm of the femoral neck (Rupp 2006). In fact, evidence
suggests that fractures within the femoral neck occur through similar mechanisms
(Rupp et al. 2003).
Occupants often brace for the impending collision causing muscle contraction.
PMHS experiments designed to reproduce loads caused by MVCs have produced
injuries that are somewhat inconsistent with data collected by CIREN. The relaxed
nature of PMHS specimen yields knee, distal femur, and hip injuries from frontal
MVCs, while the CIREN database indicates that the midshaft of the femur is the
most common injury (Rupp 2006). Chang et al. (2008) proposed that muscle
activation preloads the femur prior to impact increasing the bending moment applied
to the femur (Chang et al. 2008).
Hip Injury
Fourty-five percent of knee-thigh-hip injuries are hip fractures and dislocations from
frontal impacts where fracture of the acetabulum is most frequent (Rupp et al. 2003;
Chang et al. 2008). There are many parameters that influence hip response and
susceptibility to injury including hip posture during knee loading as well as interac-
tions between the trochanter and components of the vehicle interior. Differences in
the geometry of the acetabulum between men and women also have been shown to
affect the extent and nature of fracture (Wang et al. 2004; Holcombe et al. 2011).
Similarly, body mass and the center of gravity of the upper body prior to impact have
a considerable affect on the hip response since it is coupled to the abdomen and
lower extremity masses (Rupp et al. 2008).
Approximately 70% of AIS > 2 injuries to the lower extremities from lateral
impacts in nearside occupants during MVCs are hip or pelvis injuries compared to
the 9% of injuries that are to either the thigh or knee (Banglmaier et al. 2003). Far
side occupants are exposed to different loading conditions and only 38% of AIS > 2
injuries are to the hip or pelvis.
Lower Leg Injury

The vast majority of lower extremity injuries are to the knee-thigh-hip, and injuries
to the lower leg complex rarely result in life-threatening circumstances. As vehicular
safety standards have improved and the number of airbag equipped vehicles
increased, severe injuries to the lower leg have increased since critical body regions
are more protected than the lower leg. Additionally, prior to the advent of airbags,
lower leg injuries were less relevant and generally undocumented in the event of a
life threatening or fatal injury (Salzar et al. 2015).
The effects of bending moments on the lower leg have been extensively studied,
and its tolerance depends on the loading rate, direction, and magnitude (Schreiber
et al. 1998; Kerrigan et al. 2004; Ivarsson et al. 2005; Yoganandan et al. 2014a).
Focal lateral loads directed toward the midshaft region of the tibia generated injuries
that indicated the general location of the fracture was relatively independent of the
force direction (e.g., lateral-medial vs. anterior-posterior) (Rabl et al. 1996).
Ankle Injury
Ankle injuries are the most severe lower leg injuries and are common in frontend
MVCs. Interactions between the lower leg and the break/accelerator pedal have been
attributed to approximately 25% of inversion or eversion (Xversion; rotation about
the X-axis; rotation about the anterior-posterior axis) related injuries (Morris et al.
1997). Lower leg injuries depend upon axial preloads or preflexion (e.g., through the
Achilles tendon), and injuries are likely more severe when lower leg muscles are
activated to hurriedly and forcefully depress the break pedal, which preloads the foot
in dorsiflexion (Funk et al. 2001).
It is commonly understood that injuries to the ankle occur when the foot is forced
into Xversion. PMHS experiments that employed axial preloads followed by forced
Xversion produced (medial and lateral) ligament tears, malleolar fractures, and tibial
osteochondral fractures (Funk et al. 2002). Petit et al. (1997) were able to simulate
tension in the Achilles tendon while quasistatically loading the foot into dorsiflexion
and consistently observed medial malleolus fractures and calcaneofibular ligament
tears (Petit et al. 1997). Rudd et al. (2004) conducted PMHS experiments that
simulated the force applied to the foot under dorsiflexion when the driver was
depressing the break pedal (Rudd et al. 2004). This setup provided an appropriate
simulation of the leg in the event of a front-end MVC. Bony fractures at the ankle
joint were present in 11 of 20 specimens, ligament ruptures in 4 specimens, and
osteochondral damage in almost all specimens. Rudd et al. (2004) used surface
acoustic sensors to detect fracture timing, and they point out that the fractures to the
ankle do not necessarily align with the peak force of the dynamic load. Instead, the
ankle seems to continue to bear higher load after initial fractures until the fracture is
catastrophic.
Future Directions
Despite the fact that biomechanics has largely evolved to applied biomechanics,
considerable debate surrounds many accepted mechanisms. The current debate calls
for further validation and repeated experiments. However, objections to accepted
injury mechanisms are difficult to validate since funding opportunities for experi-
mental replication are scarce. Albert King comments on this problem, “. . . [those]
who review research proposals . . . are disinclined to reverse a previous opinion and
tend to disapprove any proposal that will reverse this opinion” (King 2015). How-
ever, advancements in instrumentation will continue to allow researchers to peer into
the biomechanical response with more detail. As long as funding opportunities in
applied biomechanics remain available.
As transportation technology evolves, first-world economies will continue to
demand more sophisticated tests from the biomechanics community. However, the
complexity of tests often precedes scientific understanding. As a result, injury
explanations can contain much speculation, but PMHS tests will indeed indicate
which injuries are likely to occur under defined loading conditions.
Finally, the anthropomorphic test devices (ATDs) have provided insight into the
biomechanical response with acceptable biofidelity (Baudrit and Trosseille 2015).
The Hybrid III ATD was developed and has been used since the early 1990s (Mertz
1993). As vehicle and traffic technology advances, the body will be exposed to new
and variable loads in the event of an MVC. Recently, Danelson et al. evaluated the
biofidelity of the Hybrid III during vertical loading, and they concluded that there is a
need for further developments to improve biofidelity (Danelson et al. 2015). The
Hybrid III was designed specifically to mimic a seated automobile occupant, but the
divergence from the PMHS response was sizable. Similarly, Foreman et al. point out
the need for a biofidelic ATD response from pedestrian hits (Foreman et al. 2015).
Injuries with complex mechanisms (especially soft tissue injuries) are difficult to
predict from the ATD response. Future work is necessary to better estimate the
human response through ATDs, and a vast amount of work has been carried out by
way of biofidelity studies, response corridor generation techniques, and PMHS
experimentation to pave the way for future improvements.
Cross-References
▶ Data Analytics for Biomechanics

▶ Machine Learning Techniques for Data Analytics of Human Motion
▶ Pedestrians in Traffic Accidents
▶ Special Pediatric Consideration in Traffic Accidents
▶ Traffic Accidents with Two-Wheel Vehicles
▶ Vehicle Occupants in Traffic Accidents
References
Allen B Jr, Ferguson R, Lehmann TR, O’brien RP (1982) A mechanistic classification of closed,
indirect fractures and dislocations of the lower cervical spine. Spine (Phila Pa 1976) 7:1–27
Anderson PA, Henley MB, Rivara FP, Maier R V (1991) Flexion distraction and chance injuries to
the thoracolumbar spine. J Orthop Trauma 5:153–160
Anderson RWG, Brown CJ, Blumbergs PC et al (2003) Impact mechanics and axonal injury in a
sheep model. J Neurotrauma 20:961–974. doi:10.1089/089771503770195812
Anuta PE (1970) Spatial registration of multispectral and multitemporal digital imagery using fast
Fourier transform techniques. IEEE Trans Geosci Electron 8:353–368
Atkinson P, Atkinson T, Haut R, Eusebi C, Maripudi V, Hill T, Sambatur K (1998) Development of
injury criteria for human surrogates to address current trends in knee-to-instrument panel
injuries (No. 983146). SAE Technical Paper
Augenstein J, Diggs K (2003) Performance of advanced air bags based on data William Lehman
Injury Research Center and new NASS PSUs. Annu Proc Assoc Adv Automot Med 47:99–101
Augenstein J, Perdeck E, Martin P et al (2000) Injuries to restrained occupants in far-side crashes.
Ann Adv Automot Med 44:57–66
Ball ST, Vaccaro AR, Albert TJ (2000) Injuries of the thoracolumbar spine associated with restraint
use in head-on motor vehicle accidents. Spinal Disord 13:297–304
Balogh ZJ, Offner PJ, Moore EE, Biffl WL (2000) NISS predicts postinjury multiple organ failure
better than the ISS. J Trauma Acute Care Surg 48:624–628
Balogh ZJ, Varga E, Tomka J et al (2003) The new injury severity score is a better predictor of
extended hospitalization and intensive care unit admission than the injury severity score in
patients with multiple orthopaedic injuries. J Orthop Trauma 17:508–512. doi:10.1097/
00005131-200308000-00006
Banglmaier RF, Rouhana SW, Beillas P, Yang KH (2003) Lower extremity injuries in lateral impact:
a retrospective study. Ann Proc Assoc Adv Automot Med 47:425–444
Barnsley L, Lord S, Wallis B, Bogduk N (1995) The prevalence of chronic cervical zygapophysial
joint pain after whiplash. Spine (Phila Pa 1976) 20:20–26
Baudrit P, Trosseille X (2015) Proposed method for development of small female and midsize male
thorax dynamic response corridors in side and forward oblique impact tests. Stapp Car Crash J
59:177–202
Baur P, Lange W, Messner G et al (2000) Comparison of real world side impact/rollover collisions
with and without thorax airbag/head protection system: a first field experience study. Ann Proc
Assoc Adv Automot Med 44:187–201
Begeman PC, King AI, Prasad P (1973) Spinal loads resulting from -Gx acceleration. In: Pro-
ceedings of the 17th stapp car crash conference, Coronado, CA, pp 343–360
Begonia MT, Dallas M, Vizcarra B et al (2015) Non-contact strain measurement in the mouse
forearm loading model using digital image correlation (DIC). Bone 81:593–601. doi:10.1016/j.
bone.2015.09.007
Bergen GS, Chen LH, Warner M (2008) Injury in the United States; 2007 chartbook
Bertrand S, Cuny S, Petit P et al (2008) Traumatic rupture of thoracic aorta in real-world motor
vehicle crashes. Traffic Inj Prev 9:153–161. doi:10.1080/15389580701775777
Bogduk N, Marsland A (1988) The cervical zygapophysial joints as a source of neck pain. Spine
(Phila Pa 1976) 13:610–617
Bouquet R, Ramet M, Bermond F et al (1998) Pelvis human response to lateral impact. In:
Proceedings of the 16th international technical conference on the enhanced safety of vehicles,
Windsor, pp 1665–1686
Burgess AR, Eastridge BJ, Young JWR, Ellison TS, Ellison Jr PS, Poka A, et al (1990) Pelvic ring
disruptions: effective classification system and treatment protocols. J Trauma Acute Care Surg
30:848–856
Cammack K, Rapport RL, Paul J, Baird WC (1959) Deceleration injuries of the thoracic aorta.
AMA Arch Surg 79:244–251. doi:10.1001/archsurg.1959.04320080080010
Carroll LJ, Holm LW, Hogg-Johnson S et al (2009) Course and prognostic factors for neck pain in
Whiplash-Associated Disorders (WAD). Results of the bone and joint decade 2000-2010 Task
force on neck pain and its associated disorders. J Manip Physiol Ther 32:S97–S107.
doi:10.1016/j.jmpt.2008.11.014
Cavanaugh JM, Yoganandan NA (2015) Thorax injury biomechanics. In: Yoganandan N, Nahum
AM, Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn. Springer,
New York, pp 332–334
Cavanaugh JM, Walilko TJ, Malhotra A, Zhu Y, King AI (1990) Biomechanical response and injury
tolerance of the pelvis in twelve sled side impacts (No. 902305). SAE Technical Paper
Chance GQ (1948) Note on a type of flexion fracture of the spine. Br J Radiol 21:452
Chancey VC, Nightingale RW, C a VE et al (2003) Improved estimation of human neck tensile
tolerance: reducing the range of reported tolerance using anthropometrically correct muscles and
optimized physiologic initial conditions. Stapp Car Crash J 47:135–153
Chang C-Y, Rupp JD, Kikuchi N, Schneider LW (2008) Development of a finite element model to
study the effects of muscle forces on knee-thigh-hip injuries in frontal crashes. Stapp Car Crash J
52:475
Cheng R, Yang KH, Levine RS, King AI, Morgan R (1982) Injuries to the cervical spine caused by
distributed frontal load to the chest. SAE Paper #821155
Chu TC, Ranson WF, Sutton MA (1985) Applications of digital-image-correlation techniques to
experimental mechanics. Exp Mech 25:232–244
Crandall J, Martin P, Bass C et al. (1996) Foot and ankle injury: the roles of driver anthropometry,
footwear, and pedal controls. Paper presented at: 40th Annual Proceedings of the Association
for the Advancement of AutomotiveMedicine
Culver RH, Bender M, Melvin JW (1978) Mechanisms, tolerances, and responses obtained under
dynamic superior-inferior head impact. Ann Arbor 7:103
Danelson KA, Kemper AR, Mason MJ et al (2015) Comparison of ATD to PMHS Response in the
under-body blast environment. Stapp Car Crash J 59:445–520
Davceva N, Janevska V, Ilievski B et al (2012) The occurrence of acute subdural haematoma and
diffuse axonal injury as two typical acceleration injuries. J Forensic Legal Med 19:480–484.
doi:10.1016/j.jflm.2012.04.022
Deng B, Begeman PC, Yang KH et al (2000) Kinematics of human cadaver cervical spine during
low speed rear-end impacts. Stapp Car Crash J 44:171–188
Depreitere B, Van Lierde C, Vander SJ et al (2006) Mechanics of acute subdural hematomas
resulting from bridging vein rupture. J Neurosurg 104(6):950. doi:10.3171/jns.2006.104.6.950
Eastridge BJ, Burgess AR (1997) Pedestrian pelvic fractures: 5-year experience of a major urban
trauma center. J Trauma Acute Care Surg 42:695–700
Fice JB, Cronin DS (2012) Investigation of whiplash injuries in the upper cervical spine using a
detailed neck model. J Biomech 45:1098–1102. doi:10.1016/j.jbiomech.2012.01.016
Foreman JL, Joodaki H, Forghani A et al (2015) Whole-body response for pedestrian impact with a
generic sedan buck. Stapp Car Crash J 59:401–444
Foster CD, Hardy WN, Yang KH et al (2006) High-speed seatbelt pretensioner loading of the
abdomen. Stapp Car Crash J 50:27–51
Funk JR, Crandall JR, Tourret LJ, MacMahon CB, Bass CR, Khaewpong N, Eppinger RH (2001)
The effect of active muscle tension on the axial injury tolerance of the human foot/ankle
complex (No. 2001-06-0074). SAE Technical Paper
Funk JR, Srinivasan SCM, Crandall JR et al (2002) The effects of axial preload and dorsiflexion on
the tolerance of the ankle/subtalar joint to dynamic inversion and eversion. Stapp Car Crash J
46:245–265. doi:2002-22-0013 [pii]
Gabbe BJ, De Steiger R, Esser M et al (2011) Predictors of mortality following severe pelvic ring
fracture: results of a population-based study. Injury 42:985–991
Gabler HC, Weaver AA, Stitzel JD (2015) Automotive field data in injury. In: Yoganandan N,
Nahum AM, Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn.
Springer, New York, pp 33–47
Gennarelli TA, Thibault LE (1982) Biomechanics of acute subdural hematoma. J Trauma
22:680–686
Gertzbein SD, Court-Brown CM (1988) Flexion-distraction injuries of the lumbar spine. Clin
Gokcen EC, Burgess AR, Siegel JH et al (1994) Pelvic fracture mechanism of injury in vehicular
trauma patients. J Trauma Acute Care Surg 36:789–796
Green DA, Green NE, Spengler DM, Devito DP (1991) Flexion-distraction injuries to the lumbar
spine associated with abdominal injuries. J Spinal Disord Tech 4:312–318
Gurdjian ES, Webster JE, LH R (1955) Observations on the mechanism of brain concussion,
contusion, and laceration. Surg Gynecol Obstet 101:680–690
Haagsma JA, Graetz N, Bolliger I et al (2016) The global burden of injury: incidence, mortality,
disability-adjusted life years and time trends from the Global Burden of Disease study 2013. Inj
Prev 22:3–18. doi:10.1136/injuryprev-2015-041616
Haffner MP, Sances S, Kumaresan S et al (1996) Response of human lower thorax to impact. Ann
Proc Assoc Adv Automot Med 40:33–43
Hardy WN, Foster CD, Mason MJ, Yang KH, King AI, Tashman S (2001) Investigation of head
injury mechanisms using neutral density technology and high-speed biplanar x-ray. Stapp Car
Crash J 45:337–368
Hardy WN, Shah CS, Kopacz JM et al (2006) Study of potential mechanisms of traumatic rupture of
the aorta using insitu experiments. Stapp Car Crash J 50:247–266
Hardy WN, Shah CS, Mason MJ et al (2008) Mechanisms of traumatic rupture of the aorta and
associated peri-isthmic motion and deformation. Stapp Car Crash J 52:233–265. doi:2008-22-
0010 [pii]
Hardy WN, Howes MK, Kemper AR, Rouhana SW (2015) Impact and injury response of the
abdomen. In: Yoganandan N, Nahum AM, Melvin JW (eds) Accidental injury: biomechanics
and prevention, 3rd edn. Springer, New York, pp 373–434
Holcombe S, Kohoyda-Inglis C, Wang L et al (2011) Patterns of acetabular femoral head coverage.
Stapp Car Crash J 55:479–490
Ivarsson BJ, Kerrigan JR, Lessley DJ, Drinkwater DC, Kam CY, Murphy DB, . . . Kent RW (2005)
Dynamic response corridors of the human thigh and leg in non-midpoint three-point bending
(No. 2005-01-0305). SAE Technical Paper
Ivarsson BJ, Genovese D, Crandall JR et al (2009) The tolerance of the femoral shaft in combined
axial compression and bending loading. Stapp Car Crash J 53:251
Jefferson G (1919) Fracture of the atlas vertebra. Report of four cases, and a review of those
previously recorded. Br J Surg 7:407–422
Katyal D, McLellan BA, Brenneman FD et al (1997) Lateral impact motor vehicle collisions:
significant cause of blunt traumatic rupture of the thoracic aorta. J Trauma 42:769–772
Kauvar DS, Wade CE (2005) The epidemiology and modern management of traumatic hemorrhage:
US and international perspectives. Crit Care 9(Suppl 5):S1–S9. doi:10.1186/cc3779
Kemper AR, McNally C, E a K et al (2008) The influence of arm position on thoracic response in
side impacts. Stapp Car Crash J 52:379–420. doi:10.4271/811007
Kerrigan JR, Drinkwater DC, Kam CY et al (2004) Tolerance of the human leg and thigh in dynamic
latero-medial bending. Int J Crashworthiness 9:607–623
King AI (2001) Fundamentals of impact biomechanics: part 2-biomechanics of the abdomen,
pelvis, and lower extremities. Annu Rev Biomed Eng 3:27–55
King AI (2015) Introduction to and applications of injury biomechanics. In: Narayan Y, Nahum
AM, Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn. Springer,
New York, pp 7–14
King AI, Yang KH (1995) Research in biomechanics of occupant protection. J Trauma 38:570–576
Kirk A, Morris A (2003) Side airbag deployments in the UK – initial case reviews. In: Proceedings
of the 18th international technical conference on enhanced safety of vehicles. Nagoya, pp 1–8
Klinich KD, Flannagan CAC, Nicholson K et al (2010) Factors associated with abdominal injury in
frontal, farside, and nearside crashes. Stapp Car Crash J 54:73–91
Kroell CK, Pope ME, Viano DC et al (1981) Interrelationship of velocity and chest compression in
blunt thoracic impact. In: 25th Stapp car crash conference proceedings. SAE Technical Paper
811016, Warrendale, PA, pp 547–580
Kroell C, Allen SD, Warner CY, Perl T (1986) Interrelatinoship of velocity and chest compression
in blunt thoracic impact to Swine II. In: 30th Stapp car crash conference proceedings SAE
Technical Paper 861881, Warrendale, PA, pp 99–121
Kuppa S, Wang J, Haffner M, Eppinger R (2001) Lower extremity injuries and associated injury
criteria. In 17th ESV Conference, Paper (No. 457)
Kuppa S, Fessahaie O et al (2003) An overview of knee-thigh-hip injuries in frontal crashes in the
United States. Natl Highw Traffic Saf Adm ISSI. doi:10.1017/CBO9781107415324.004
Lamielle S, Vezin P, Verriest JP et al (2008) 3D deformation and dynamics of the human cadaver
abdomen under seatbelt loading. Stapp Car Crash J 52:267
Lee BB, Cripps RA, Fitzharris M, Wing PC (2014) The global map for traumatic spinal cord injury
epidemiology: update 2011, global incidence rate. Spinal Cord 52:110–116. doi:10.1038/
sc.2012.158
LeGay DA, Petrie DP, Alexander DI (1990) Flexion-distraction injuries of the lumbar spine and
associated abdominal trauma. J Trauma 30:436–444
Leitgeb J, Mauritz W, Brazinova A et al (2012) Outcome after severe brain trauma due to acute
subdural hematoma. J Neurosurg 117:324–333. doi:10.3171/2012.4.JNS111448
Lessley DJ, Riley P, Zhang Q et al (2014) Occupant kinematics in laboratory rollover tests: PMHS
response. Stapp Car Crash J 58:251
Leucht P, Fischer K, Muhr G, Mueller EJ (2009) Epidemiology of traumatic spine fractures. Injury
40:166–172. doi:10.1016/j.injury.2008.06.040
Linnau KF, Blackmore CC, Kaufman R et al (2007) Do initial radiographs agree with crash site
mechanism of injury in pelvic ring disruptions? A pilot study. J Orthop Trauma 21:375–380
Lu Y, Chen C, Kallakuri S et al (2005) Neurophysiological and biomechanical characterization of
goat cervical facet joint capsules. J Orthop Res 23:779–787. doi:10.1016/j.orthres.2005.01.002
Manson T, O’Toole RV, Whitney A et al (2010) Young-Burgess classification of pelvic ring
fractures: does it predict mortality, transfusion requirements, and non-orthopaedic injuries?
J Orthop Trauma 24:603–609
Markham DE (1972) Anterior dislocation of the hip and diastasis of the contralateral sacro-iliac
joint – the rear-seat passenger’s injury? Br J Surg 59:296–298
Masson C, Baque P, Brunet C (2005) Quasi-static compression of the human pelvis: an experimen-
tal study. Comput Methods Biomech Biomed Engin 8:191–192
Matsui Y, Oikawa S (2015) Risks of serious injuries and fatalities of cyclists associated with impact
velocities of cars in car-cyclist accidents in Japan. Stapp Car Crash J 59:385–400
McCormick N, Lord J (2010) Digital image correlation. Mater Today 13:52–54. doi:10.1016/
S1369-7021(10)70235-2
McKay BJ, Bir CA (2009) Lower extremity injury criteria for evaluating military vehicle occupant
injury in underbelly blast events. Stapp Car Crash J 53:229–249
Melvin JW, Stalnaker RL, Roberts VL, Trollope ML (1973) Impact injury mechanisms in
abdominal organs. In: 17th Stapp car crash conference. Society of Automotive Engi-
neers, Oklahoma City
Mertz HJ (1993) Anthropomorphic test devices. In: Accidental injury. Springer, New York, pp
66–84
Middleton JW, Dayton A, Walsh J et al (2012) Life expectancy after spinal cord injury: a 50-year
study. Spinal Cord 50:803–811
Miller M (1989) The biomechanical response of the lower abdomen to belt restraint loading.
J Trauma Acute Care Surg 29:1571–1584
Molz FJI V, George PD, Go LS et al (1997) Simulated automotive side impact on the isolated
human pelvis: Phase I: development of a containment device. Phase II: analysis of pubic
symphysis motion and overall pelvic compression. In: Stapp car crash conference proceedings,
Warrendale, PA, pp 75–89
Morris A, Thomas P, Taylor AM, Wallace WA (1997) Mechanisms of fracture in ankle and hind-
foot injuries to front seat car occupants- an in-depth accident data analysis. Stapp Car Crash
Conf 41:181–192. doi:10.4271/973328
Myers BS, Winkelstein BA (1995) Epidemiology, classification, mechanism, and tolerance of
human cervical spine injuries. Crit Rev Biomed Eng 23(5–6):307–409
NHTSA (2005) Federal motor vehicle safety standards; Occupant crash protection. Docket
No. NHTSA-04-18726
Nightingale RW, McElhaney JH, Richardson WJ et al (1996a) Experimental impact injury to the
cervical spine: relating motion of the head and the mechanism of injury. J Bone Jt Surg Am
78:412–421
Nightingale RW, McElhaney JH, Richardson WJ, Myers BS (1996b) Dynamic responses of the
head and cervical spine to axial impact loading. J Biomech 29:307–318
Nightingale RW, McElhaney JH, Camacho DL, Kleinberger M, Winkelstein BA, Myers BS (1997)
The dynamic responses of the cervical spine: buckling, end conditions, and tolerance in
compressive impacts (No. 973344). SAE Technical Paper
Nightingale RW, Myers BS, Yoganandan NA (2015) Neck injury biomechanics. In: Yoganandan
NA, Nahum AM, Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn.
Nirula R, Pintar FA (2008) Identification of vehicle components associated with severe thoracic
injury in motor vehicle crashes: a CIREN and NASS analysis. Accid Anal Prev 40:137–141.
doi:10.1016/j.aap.2007.04.013
Nusholtz GS, Kaiker PS (1986) Pelvic stress. J Biomech 19:1003–1014
Nusholtz GS, Melvin JW, Huelke DF, Alem NM, Blank JG (1981) Response of the cervical spine to
superiorinferior head impact (No. 811005). SAE Technical Paper
Nusholtz GS, Huelke DE, Lux P, Alem NM, Montalvo F (1983) Cervical spine injury mechanisms
(No. 831616). SAE Technical Paper
Ooi CK, Goh HK, Tay SY, Phua DH (2010) Patients with pelvic fracture: what factors are
associated with mortality? Int J Emerg Med 3:299–304
Panzer MB, Fice JB, Cronin DS (2011) Cervical spine response in frontal crash. Med Eng Phys
33:1147–1159. doi:10.1016/j.medengphy.2011.05.004
Parr JC, Miller ME, Pellettiere JA, Erich RA (2013) Neck injury criteria formulation and injury risk
curves for the ejection environment: a pilot study. Aviat Sp Environ Med 84:1240–1248.
doi:10.3357/ASEM.3722.2013
Pennal GF, Tile M, Waddell JP, Garside H (1980) Pelvic disruption: assessment and classification.
Clin Orthop Relat Res 151:12–21
Petit P, Portier L, Foret-Bruno J-Y et al (1997) Quasistatic characterization of the human foot-ankle
joints in a simulated tensed state and updated accidentological data. In: Proceedings of the
international research council on the biomechanics of injury conference, Hannover, Germany,
pp 363–376
Petitjean A, Trosseille X (2011) Statistical simulations to evaluate the methods of the construction
of injury risk curves. Stapp Car Crash J 55:411
Petitjean A, Trosseille X, Yoganandan N, Pintar FA (2015) Normalization and scaling for human
response corridors and development of injury risk curves. In: Yoganandan N, Nahum AM,
Melvin J (eds) Accidental injury: biomechanics and prevention, 3rd edn. Springer, New York,
pp 769–792
Petrisor BA, Bhandari M (2005) (i) Injuries to the pelvic ring: Incidence, classification, associated
injuries and mortality rates. Curr Orthop 19:327–333
Pintar FA, Yoganandan N, Voo L, Cusick JF, Maiman DJ, Sances A (1995) Dynamic characteristics
of the human cervical spine (No. 952722). SAE Technical Paper
Pintar FA, Yoganandan NA, Voo L (1998) Effect of age and loading rate on human cervical spine
injury threshold. Spine (Phila Pa 1976) 23:1957–1962
Pintar FA, Yoganandan NA, Maiman DJ et al (2012) Thoracolumbar spine fractures in frontal
impact crashes. Ann Adv Automot Med 56:277–283
Powell WR, Ojala SJ, Advani SH, Martin RB (1975) Cadaver femur responses to longitudinal
impacts (No. 751160). SAE Technical Paper
Rabl W, Haid C, Krismer M (1996) Biomechanical properties of the human tibia: fracture behavior
and morphology. Forensic Sci Int 83:39–49
Robertson A, Branfoot T, Barlow IF, Giannoudis PV (2002a) Spinal injury patterns resulting from
car and motorcycle accidents. Spine (Phila Pa 1976) 27:2825–2830. doi:10.1097/01.
BRS.0000035686.45726.0E
Robertson A, Giannoudis P V, Branfoot T et al (2002b) Spinal injuries in motorcycle crashes:
patterns and outcomes. J Trauma 53:5–8. doi:10.1097/00005373-200207000-00002
Rouhana SW, Lau IV, Ridella SA (1985) Influence of velocity and forced compression on the
severity of abdominal injury in blunt, nonpenetrating lateral impact. J Trauma 25:490–500
Rudd R, Crandall J, Millington S et al (2004) Injury tolerance and response of the ankle joint in
dynamic dorsiflexion. Stapp Car Crash J 48:1–26. doi:2004-22-0001 [pii]
Rupp JD (2006) Biomechanics of hip fractures in frontal motor vehicle crashes. Phdthesis, PhD
dissertation, The University of Michigan, Ann Arbor
Rupp JD (2015) Knee, thigh, and hip injury biomechanics. In: Yoganandan NA, Nahum AM,
Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn. Springer, New York,
pp 471–497
Rupp JD, Reed MP, Van Ee CA et al (2002) The tolerance of the human hip to dynamic knee
loading. In: 46th Stapp car crash conference, Ponte Verdra Beach, FL
Rupp JD, Reed MP, Jeffreys TA, Schneider LW (2003) Effects of hip posture on the frontal impact
tolerance of the human hip joint. Stapp Car Crash J 47:21
Rupp JD, Miller CS, Reed MP et al (2008) Characterization of knee-thigh-hip response in frontal
impacts using biomechanical testing and computational simulations. Stapp Car Crash J 52:421
Salzar RS, “Dale” Bass CR, Kent R et al (2006) Development of injury criteria for pelvic fracture in
frontal crashes. Traffic Inj Prev 7:299–305
Salzar RS, Lievers BW, Bailey AM, Crandall JR (2015) Leg, foot, and ankle injury biomechanics.
In: Yoganandan NA, Nahum AM, Melvin JW (eds) Accidental injury: biomechanics and
prevention, 3rd edn. Springer, New York, pp 499–547
Santschi M, Echavé V, Laflamme S et al (2005) Seat-belt injuries in children involved in motor
vehicle crashes. Can J Surg 48:373–376
Schiff MA, Tencer AF, Mack CD (2008) Risk factors for pelvic fractures in lateral impact motor
vehicle crashes. Accid Anal Prev 40:387–391
Schreiber P, Crandall J, Hurwitz S, Nusholtz GS (1998) Static and dynamic bending strength of the
leg. Int J Crashworthiness 3:295–308
Sekhon LH, Fehlings MG (2001) Epidemiology, demographics, and pathophysiology of acute
spinal cord injury. Spine (Phila Pa 1976) 26:S2–12. doi:10.1097/00007632-200112151-00002
Services UDoHaH (2007) Injury in the United States: 2007
Shinkawa H, Yasuhara H, Naka S et al (2004) Characteristic features of abdominal organ injuries
associated with gastric rupture in blunt abdominal trauma. Am J Surg 187:394–397
Shkrum M, McClafferty K, Green R et al (1999) Mechanisms of aortic injury in fatalities occurring
in motor vehicle collisions. J Forensic Sci 44:44–56
Siegmund GP, Myers BS, Davis MB et al (2001) Mechanical evidence of cervical facet capsule
injury during whiplash: a cadaveric study using combined shear, compression, and extension
loading. Spine (Phila Pa 1976) 26:2095–2101
Siegmund GP, Davis MB, Quinn KP et al (2008) Head-turned postures increase the risk of cervical
facet capsule injury during whiplash. Spine (Phila Pa 1976) 33:1643–1649. doi:10.1097/
BRS.0b013e31817b5bcf
Simms CK, Wood DP (2006) Pedestrian risk from cars and sport utility vehicles-a comparative
analytical study. Proc Inst Mech Eng Part D J Automob Eng 220:1085–1100
Somers RL (1983a) The probability of death score: a measure of injury severity for use in planning
and evaluating accident prevention. Accid Anal Prev 15:259–266. doi:10.1016/0001-4575(83)
90050-7
Somers RL (1983b) The probability of death score: an improvement of the injury severity score.
Accid Anal Prev 15:247–257. doi:10.1016/0001-4575(83)90049-0
Sparks JL, Bolte JH, Dupaix RB et al (2007) Using pressure to predict liver injury risk from blunt
impact. Stapp Car Crash J 51:401–432
Stalnaker RL, Ulman MS (1985) Abdominal trauma – review, response, and criteria. In: 29th Stapp
car crash conference proceedings, pp 1–16
Stalnaker RL, McElhaney JH, Roberts VL, Trollope ML (1973) Human torso response to blunt
trauma. In: King WF, Mertz HJ (eds) Human impact response: measurement and simulation.
Springer US, Boston, pp 181–199
Stemper BD, Yoganandan NA, Pintar FA (2004) Validation of a head-neck computer model for
whiplash simulation. Med Biol Eng Comput 42:333–338
Stemper BD, Pintar FA, Baisden JL (2015) Lumbar spine injury biomechanics. In: Yoganandan N,
Nahum AM, Melvin JW (eds) Accidental injury: biomechanics and prevention, 3rd edn.
Takhounts EG, Craig MJ, Moorhouse K et al (2013) Development of brain injury criteria (BrIC).
Stapp Car Crash J 57:243
Talantikite Y, Brun-Cassan F, Le coz J, Tarriere C (1993) Abdominal protection in side impact.
Injury mechanisms and protection criteria. Proc Int Res Counc Biomech Inj Conf 21:131–144
Tencer AF, Kaufman R, Huber P et al (2007) Reducing primary and secondary impact loads on the
pelvis during side impact. Traffic Inj Prev 8:101–106
Tile M, Pennal GF (1980) Pelvic disruption: principles of management. Clin Orthop Relat Res
151:56–64
Torg JS (1985) Epidemiology, pathomechanics, and prevention of athletic injuries to the cervical
spine. Med Sci Sports Exerc 17:295–303
Trollope ML, Stalnaker RL, JH ME, Frey CF (1973) The mechanism of injury in blunt abdominal
trauma. J Trauma 13:962–970
Vandevord PJ, Bolander R, Sajja VSSS et al (2012) Mild neurotrauma indicates a range-specific
pressure response to low level shock wave exposure. Ann Biomed Eng 40:227–236.
doi:10.1007/s10439-011-0420-4
Viano DC, Khalil TB (1976) Investigation of impact response and fracture of the human femur by
finite element modeling (No. 760773). SAE Technical Paper
Viano DC, Lau VK (1983) Role of impact velocity and chest compression in thoracic injury. Aviat
Sp Environ Med 54:16–21
Viano DC, Stalnaker RL (1980) Mechanisms of femoral fracture. J Biomech. doi:10.1016/0021-
9290(80)90356-5
Viano DC, Lau IV, Asbury C et al (1989) Biomechanics of the human chest, abdomen, and pelvis in
lateral impact. Accid Anal Prev 21:553–574. doi:10.1016/0001-4575(89)90070-5
Wang SC, Brede C, Lange D et al (2004) Gender differences in hip anatomy: possible implications
for injury tolerance in frontal collisions. Annu Proc Assoc Adv Automot Med 48:287–301
Wang MC, Pintar F, Yoganandan N, Maiman DJ (2009) The continued burden of spine fractures
after motor vehicle crashes. J Neurosurg Spine 10:86–92. doi:10.3171/SPI.2008.10.08279
Wilberger JE, Harris M, Diamond DL (1991) Acute subdural hematoma: morbidity, mortality, and
operative timing. J Neurosurg 74:212–218. doi:10.3171/jns.1991.74.2.0212
Yang KH, Begeman PC (1996) A proposed role for facet joints in neck pain in low to moderate
speed rear end impacts part I: biomechanics. In: 6th injury prevention through biomechanics
symposium, Wayne State University, Detroit, MI, pp 59–63
Yoganandan NA, Sances Jr A, Maiman DJ et al (1986) Experimental spinal injuries with vertical
impact. Spine (Phila Pa 1976) 11:855–860
Yoganandan N, Haffner M, Maiman DJ, Nichols H, Pintar FA, Jentzen J, . . . Sances A (1989)
Epidemiology and injury biomechanics of motor vehicle related trauma to the human spine
(No. 892438). SAE Technical Paper
Yoganandan NA, Pintar FA, Sances Jr A et al (1991) Strength and kinematic response of dynamic
cervical spine injuries. Spine (Phila Pa 1976) 16:S511–S517
Yoganandan NA, Pintar FA, Gennarelli TA, Maltese MR (2000) Patterns of abdominal injuries in
frontal and side impacts. Ann Proc Assoc Adv Automot Med 44:17–36
Yoganandan N, Pintar FA, Maltese MR (2001) Biomechanics of abdominal injuries. Crit Rev™
Biomed Eng 29(2)
Yoganandan NA, Pintar FA, Zhang J, Gennarelli TA (2007) Lateral impact injuries with side airbag
deployments-A descriptive study. Accid Anal Prev 39:22–27. doi:10.1016/j.aap.2006.05.014
Yoganandan NA, Gennarelli TA, Zhang J et al (2009) Association of contact loading in diffuse
axonal injuries from motor vehicle crashes. J Trauma 66:309–315. doi:10.1097/
TA.0b013e3181692104
Yoganandan NA, Humm JR, Pintar FA (2013a) Force corridors of post mortem human surrogates in
oblique side impacts from sled tests. Ann Biomed Eng 41:2391–2398. doi:10.1007/s10439-013-
0847-x
Yoganandan N, Stemper BD, Pintar FA et al (2013b) Cervical spine injury biomechanics: applica-
tions for under body blast loadings in military environments. Clin Biomech 28:602–609.
Yoganandan NA, Arun MWJ, Pintar FA, Szabo A (2014a) Optimized lower leg injury probability
curves from postmortem human subject tests under axial impacts. Traffic Inj Prev 15(Suppl 1):
S151–S156. doi:10.1080/15389588.2014.935357
Yoganandan N, Arun MWJ, Pintar FA (2014b) Normalizing and scaling of data to derive human
response corridors from impact tests. J Biomech 47:1749–1756. doi:10.1016/j.
Yoganandan NA, Arun MWJ, Pintar FA, Banerjee A (2015a) Lower leg injury reference values and
risk curves from survival analysis for male and female dummies: meta-analysis of postmortem
human subject tests. Traffic Inj Prev 16(Suppl 1):S100–S107. doi:10.1080/
15389588.2015.1015118
Yoganandan NA, Humm JR, Pintar FA et al (2015b) Oblique loading in post mortem human
surrogates from vehicle tests using chestbands. Stapp Car Crash J 59:1–22
Young JW, Resnik CS (1990) Fracture of the pelvis: current concepts of classification. AJR Am J
Roentgenol 155:1169–1175
Zhu F, Dong L, Jin X et al (2015) Testing and modeling the responses of Hybrid III crash-dummy
lower extremity under high-speed vertical loading. Stapp Car Crash J 59:521–536
Vehicle Occupants in Traffic Accidents
Garrett A Mattos
Abstract
Occupant motion and injury response in motor vehicle crashes is dictated by the
forces applied to the human body in combination with the relative motion
between occupant and vehicle. These responses are complex and dependent on
many factors including those relating to crash, vehicle, and occupant character-
istics. The study of occupant response is an important step in improving vehicle
design and crashworthiness. This chapter provides an overview of the important
aspects of occupant response for the four main crash modes: frontal, side,
rollover, and rear. The general characteristics of occupant kinematic response
are discussed with respect to each crash mode. Typical injury patterns and
mechanisms are identified and their relationship to occupant motion and interac-
tion with the crash environment is highlighted.
Keywords
Crash mode • Injury patterns • Occupant kinematics • Serious injury
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Frontal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Injury Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Injury Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Rollover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Injury Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Rear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
G.A. Mattos (*)

Transport and Road Safety (TARS) Research Centre, University of New South Wales, Sydney,
NSW, Australia
e-mail: g.mattos@unsw.edu.au

DOI 10.1007/978-3-319-30808-1_94-1
2 G.A. Mattos
Injury Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Countermeasures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Introduction
Traffic accidents generally involve one or more discrete impact events that result in
global acceleration and local deformation of the vehicle structure. The specific
characteristics of the vehicle response ultimately affect the motion of its occupants.
This chapter will explore the relationships between the characteristics of vehicle
crashes and occupant kinematics and injury patterns.
Significant work has been accomplished, historically, to study occupant motion
and predict injury response in motor vehicle crashes. Early testing programs utilized
live animals or human volunteers to investigate occupant motion under subinjurious
loading conditions. The use of post mortem human subjects (PMHS) to explore the
response of the human body goes back to the nineteenth century and continues to this
day. In the mid-1900s, anthropomorphic test devices (ATDs) began to be developed
which could replicate human kinematic response under specific loading conditions.
Recently, finite element (FE) modeling has been used to investigate the effects of a
wide range of select crash and vehicle factors on occupant response outcomes by
performing large amounts of simulations that could not practically be performed in a
physical setting. Over the past 10 years, significant efforts have been made to create
highly accurate FE models of the human body. These models allow for even greater
control and variability of occupant anthropometry and morphology.
Crash investigations and epidemiological studies provide the road safety com-
munity with the evidence base necessary to proceed with safety recommendations,
design changes, or research projects to investigate the causes and mechanisms of
pertinent injuries. These studies utilize large nationally representative databases or
small subsets of very detailed crash reconstructions to identify the characteristics,
and distribution of injuries occurring on the road and their findings translate into
work focused on remedying deficiencies. This process is continuous and ongoing as
the patterns of injury shift over time due to the changes in vehicle and road design
and the implementation of improved safety features.
The following will provide an overview of the current state of knowledge of the
general characteristics of occupant kinematics, injury patterns, and injury sources for
the four main crash configurations. Current capabilities in mitigating injury severity
through the use of countermeasures and design choice will also be discussed. Much
of which is provided below is available because of the aforementioned types of
research that has been conducted and is available in the literature. For brevity and
clarity, this chapter will focus on adult occupants of light passenger vehicles that are
restrained with three-point seatbelts and involved in single-event crashes, unless
otherwise noted. Injuries are described in general terms and their severity is defined
Vehicle Occupants in Traffic Accidents 3
using the Abbreviated Injury Scale (AIS) which rates injury from 1 to 6 based on the
risk of mortality (AAAM 2008).
State of the Art
Improvements in experimental methods, human surrogates, and data acquisition

technologies continue to provide opportunities to study occupant response with
more detail than before. For example, with regard to experimental methods, crash
testing facilities are continuing to push the envelope in crash severity. Frontal
crashes are being conducted at reduced amounts of overlap and considerations are
being made to evaluate the effect of automated emergency braking on occupant
response prior to and during the crash. Recently, for the first time, full-scale rollover
tests have been performed using PMHS to study occupant kinematics and injury
response. Improvements in FE model design, such as the definition of active
musculature, and in computing power have allowed for more complex and realistic
replications of the occupant behavior. Physical testing has similarly improved with
the development of advanced acquisition systems that allow for investigators to
visualize and accurately measure occupant movement. These include high-speed
x-ray systems that can record the motion of the skeletal system during impacts and
multicamera motion tracking systems that provide three-dimensional occupant
motion measurements that are accurate to a fraction of a millimeter.
Frontal
Frontal crashes account for over half of all crashes involving passenger vehicles.
They occur when the front plane of a forward-traveling vehicle strikes another
vehicle or fixed object. These impacts can involve the entire face or just a fraction
of overlap of the struck vehicle’s front plane. The vehicle and occupant kinematics
are dependent on the amount of overlap, defined by the amount of the frontal plane
of the struck vehicle that is impacted by the striking vehicle or object. In full frontal
crashes, the vehicle decelerates almost purely along its longitudinal axis as the crush-
zone structures of the vehicle frame deform and absorb the crash energy, while
generally preventing intrusion into the occupant compartment. This impact scenario
results in the occupants moving directly forward and into their seatbelts and forward
airbags. The combination of seatbelt, airbags, and crush tubes aims to allow the
occupant to ride down the crash event under a survivable deceleration.
As the amount of overlap decreases, greater lateral and rotational motions begin
to come into play, especially when the impact is applied at an oblique angle. Small-
overlap crashes, in which less than one-quarter of the front structure is impacted,
often do not engage the crush-zone structure of the vehicle frame. In this scenario,
the vehicle sustains major damage that can extend from the forward wheel to the
passenger footwell, and even to the B-pillar in severe crashes. While the vehicle
4 G.A. Mattos
Fig. 1 Occupant and vehicle motion in small-overlap frontal crash
decelerates along its longitudinal axis, it also moves laterally and rotates, as shown in
Fig. 1. The multidirectional vehicle response in small-overlap and oblique frontal
crashes causes the occupants to move toward the impact on the front corner of the
vehicle. Moving in this angled direction, forward and lateral relative to the vehicle,
reduces the effectiveness of the airbags or prevents engagement altogether.
A common phenomenon that occurs in frontal crashes, especially to rear-seated
occupants, is known as submarining. Defined as insufficient restraint of the pelvis by
the lap belt, submarining occurs when the lap belt fails to effectively engage the
pelvis due to a combination of poor belt geometry and improper placement. For ideal
performance, the lap belt is designed to be placed across the pelvis which is able to
withstand the forces applied by the lap belt in a frontal crash. During submarining,
the occupant’s pelvis moves forward while the lap belt rides up onto the abdomen.
This can result in severe loading of the abdomen and lumbar spine, increased
excursion of the lower extremities, and decreased displacement of the head.
Injury Patterns
Injuries sustained in frontal crashes are primarily due to direct impact with the
vehicle interior or loading on the body by the seatbelt or seatpan. The pattern of
injuries is dependent on the crash scenario and occupant seat position, both relative
to the location of impact and differentiated by front or rear location as demonstrated
in Fig. 2. Small-overlap crashes tend to be more severe in terms of injury outcome,
vehicle structural deformation, and occupant response. Children seated in front seats
experienced significantly increased injury risk due to the mismatch between their
size and injury tolerance and the performance of front seat restraints and airbags
which are optimized for the 50th percentile male. The use of seatbelts greatly reduces
Fig. 2. Distribution of seriously (AIS3+) injured occupants in large-overlap (left) and small-
overlap (right) frontal impact crashes by body region sustaining serious injury (Hallman et al. 2011)
the risk of injury to all body regions, especially the head and thorax, in low- and
high-speed frontal crashes (Viano and Parenteau 2010).
Head injury patterns differ between large- and small-overlap frontal crashes.
With a decrease in the amount of overlap comes an increase in the likelihood of the
head impacting components not protected by airbags, such as the A-pillar or center
instrument panel. This results in a disparity of head injuries which is highlighted in
field data. Occupants in large-overlap frontal crashes generally experience impacts
between their head and the steering wheel or dash panel airbag directly in front of
their seat. Impacts with the frontal airbag generally have a lower incidence of minor
head injuries, such as bruising, and a greater incidence of moderate (AI2+) injuries
involving a loss of consciousness (Hallman et al. 2011). The interaction between an
occupant’s head and rigid interior components is more common in small-overlap
frontal crashes and results in a higher rate of skull fractures and brain tissue damage.
Moderate and severe head injuries are relatively rare for rear seat occupants. This is
due to the combined effect of the vehicle’s structural performance and the rear seat
occupant’s kinematics. Structural intrusion rarely reaches the rear seat occupants
and, thus, limits their exposure to impacts with intruding objects. Secondly, rear seat
occupants are more likely to experience submarining resulting in reduced upper
torso and head excursion and limiting their exposure to direct impact.
Spinal injuries, specifically those involving the posterior aspects of the cervical,
i.e., upper, vertebra, occur at twice the rate in small-overlap versus large-overlap
6 G.A. Mattos
frontal crashes due to the oblique nature of the event (Hallman et al. 2011).
The mechanism of these types of fractures is head impact with the A-pillar resulting
in compression-extension of the cervical spine. Injuries to the lumbar, i.e., lower,
spine are likely caused by seatpan loading of the pelvis that compresses the
spine (Pintar et al. 2014).
Chest injuries on the lower end of the injury severity spectrum frequently
sustained in frontal crashes include skin contusions and abrasions and sternum
fractures directly related to seatbelt loading. The most common serious (AIS3+)
thoracic injury, also most often attributed to belt loading, is a unilateral lung
contusion in higher severity crashes. Small-overlap impacts have a higher rate of
bilateral lung contusions than large-overlap crashes, while both crash configurations
have similar rates of injuries involving multiple uni- and bilateral rib fractures
(Hallman et al. 2011). Thoracic injury risk, specifically to the lungs and heart, is
increased in severe impact events that result in the occupant bottoming-out the airbag
and directly impacting the steering column (Chen and Gabler 2014). For rear-seated
occupants in frontal crashes, the thorax is the most commonly injured body region
and occurs almost exclusively due to seatbelt loading (Beck et al. 2016).
The balance between practicality and performance in seatbelt design has resulted
in almost universal use of three-point seatbelt systems that create asymmetric chest
loading in frontal impacts that is suboptimal for thoracic injury mitigation. However,
the overall benefits of seatbelt use far outweigh any minor deficiencies in their
design. Unbelted drivers are significantly more likely than belted drivers to impact
the steering wheel or front dash in a frontal crash. Such impacts greatly increase the
risk of serious thoracic injuries such as aortic, heart, and liver lacerations (Chen and
Gabler 2014). A factor that can increase the severity of front belt-restrained driver
and passenger injuries in a frontal impact is the presence of an unrestrained rear seat
occupant. The rear seat occupant can load the rear of the front seat increasing the
deformation of the driver or passenger’s thorax.
Abdomen injuries are relatively rare in frontal crashes for front seatbelt-
restrained occupants and account for approximately 5% of all AIS3+ injuries. For
frontal occupants, they are primarily caused by interaction between the occupant and
the steering wheel or lap belt (Reichert et al. 2013). Rear seat occupants, especially
adolescents, are at a much higher risk of sustaining abdomen injuries due to their
increased risk of submarining. Occupants that experience submarining are subjected
to loading of the abdomen by the lap belt which can result in upper abdominal
injuries as well as concomitant fractures of the lower rib cage.
Upper extremity injuries in frontal crashes commonly consist of fractures of the
hand, radius, and ulna. The incidence of injury to the outboard extremity is increased
in small-overlap crashes due to its interaction with intruding components and
exposure to crash forces. Airbag deployment increases the risk of upper extremity
injury, especially to the hand and forearm for drivers that have their hands on the
steering wheel (Jernigan et al. 2005).
The lower extremities constitute the most commonly injured body region in
frontal crashes, and for large-overlap crashes, they are the most frequently injured
body region at the moderate (AIS2+) and serious (AIS3+) levels. Moderate (AIS2+)
lower extremity injuries sustained in frontal crashes most often involve the pelvis,
femur, knee, and tibia in the form of fractures (Hallman et al. 2011). The oblique
occupant kinematics resulting from small-overlap frontal crashes increases pelvis
loading while decreasing loading of the feet, as compared to large-overlap crashes.
These injuries are sustained due to footpan loading caused by intrusion or from
interaction between the knee and instrument panel as the occupant moves forward
relative to the vehicle.
Side
Side impact crashes are generally defined as planar crashes between two vehicles, or
between one vehicle and a fixed object, in which the primary direction of force is
within 45 of the lateral axis of the vehicle. While multivehicle crashes, in which the
front of one vehicle impacts the side of another, are more common than those
involving a single vehicle, i.e., fixed object collisions, the resulting occupant kine-
matics is similar.
This crash configuration is dominated by a lateral acceleration of the vehicle, but
generally also involves a longitudinal acceleration component. Typical impact
scenarios include intersection crashes, in which both the struck and striking vehicles
are moving, and single-vehicle crashes, in which loss of control events lead to a side
impact with a fixed object. Occupants involved in side impact crashes can be
classified by their seat position relative to the impacted side of the vehicle. Those
seated on the struck-side are identified as near-side occupants, while those seated on
the nonstruck side are identified as far-side occupants. The kinematic and injury
response differs for near- and far-side occupants.
The kinematic response of the near-side occupant is dominated by the intrusion of
the adjacent door and B-pillar. The term “intrusion” is used here to describe relative
motion between the vehicle’s nominal shape and its deformed shape. In the case of a
side-impact against a stationary fixed object such as a tree, the intruding structure
does not necessarily move relative to the global reference frame, e.g., surrounding
environment, since it is also stationary against the fixed object. In a vehicle-to-
vehicle side impact crash, the intruding structure moves relative to both the
global and the local vehicle reference frame. At impact, the struck side of the vehicle
begins to intrude inward while the vehicle is simultaneously accelerated laterally.
This results in a compounded relative lateral motion between the occupant and
the vehicle interior, resulting in impact between the occupant and the intruding
structure at essentially the preimpact speed. This impact typically begins with the
pelvis and progresses with time upward through to the upper thorax and head
(Yoganandan et al. 2015). Near-side (also referred to as struck side) occupant
responses may be modulated by the deployment of torso side airbags and side air
curtains and can be affected by the position of the upper extremity. As the event
proceeds, the occupant will unload the vehicle’s struck side interior and begin to
move toward the center of the vehicle. At this point, the occupant’s inboard pelvis
will decelerate against the lap belt and center console, if one exists, and the upper
8 G.A. Mattos
torso may displace toward the far-side of the vehicle, depending on the performance
of the shoulder belt.
Unlike the response of the near-side occupant, the initial response of the far-side
occupant is driven by the vehicle’s lateral acceleration, rather than its deformation,
and is heavily influenced by the geometry and performance of the restraint system.
As the vehicle accelerates laterally, the occupant is displaced toward the struck side.
The displacement of the pelvis is more effectively mitigated than the displacement of
the torso. The geometry of the lap belt and the additional support provided by the
center console, if one exists, limits pelvis displacement to 100–300 mm. The
shoulder belt is the sole source of torso restraint, and its effectiveness depends on
its ability to remain in position over the shoulder. Increased belt slip, defined as the
amount the shoulder belt moves off of its nominal position, reduces the effectiveness
of the restraint. Depending on the amount of belt slip, the head can displace laterally
up to 732 mm (Forman et al. 2013). As the head and torso move across the center of
the vehicle, they interact with the near-side occupant, if there is one, and the
deformed vehicle structure. By the time the far-side occupant reaches the struck
side, the deformation event is generally over. The upper extremities, which are
essentially unrestrained, often flail toward the struck side.
Injury Patterns
Injury characteristics, risk, and patterns differ between near- and far-side occupants
as indicated in Fig. 3. The proximity of near-side occupants to the impact zone
increases their risk of injury for a given crash due to the direct loading that occurs
between their body and the door.
Head injuries have been found to almost exclusively result from direct contact
and most often with the roof rail, B-pillar, or striking vehicle/fixed object
(Yoganandan et al. 2010). Serious head injuries more frequently involve the brain
than the skull, which is likely the result of an airbag’s ability to mitigate skull
fractures more effectively than brain injuries (Yoganandan et al. 2010). Depending
on the severity of the crash and the performance of the side air curtain, if present, the
near-side occupant’s head may come into direct contact with the impacting face of
the striking vehicle or fixed object. The head of the far-side occupant is prone to
impacting a wide range of interior components, often on the struck-side of the
vehicle, and the adjacent occupant, if there is one, due to its relatively unrestrained
nature (Gabler et al. 2005).
Spine injuries are somewhat rare in side impact crashes, though they comprise a
significant portion of severe injuries and are often cited as a main cause of death.
They can result from head impacts which load the neck axially and laterally
(McIntosh et al. 2007). Soft tissue and joint injuries are also possible under inertial
loading in side impacts.
Thoracic injuries sustained by far-side occupants typically result from impact
with the seatback, seatbelt, and vehicle interior (Gabler et al. 2005). Far-side
occupants are likely to impact near-side seats that have deformed into the far-side
Fig. 3. Distribution of seriously (AIS3+) injured near-side (left) and far-side (right) occupants in
side impact crashes by body region sustaining moderate serious (AIS3+) injury (Rupp et al. 2013)
occupant’s trajectory. For near-side occupants, structural intrusion into the

occupant space is believed to be a causal factor in producing chest injuries (Pintar
et al. 2007). As the impact direction becomes more oblique, chest injury severity
tends to increase due to the reduced injury tolerance of the ribcage to oblique
loading. Rib fractures and lung contusions dominate the injury profile, with the
vast majority of injuries located on the struck side of the body. While aortic injury
is rare in side impact crashes, sustained by less than <4% of occupants, side-
impact crashes provide the greatest risk to aortic injury of all crash configura-
tions (Fitzharris et al. 2004).
Abdominal injuries are sensitive to occupant position. For far-side occupants,
abdominal injuries are generally isolated to the organs located on the inboard aspect
of the vehicle (Fildes et al. 2007). For example, a far-side occupant seated on the left
side of the vehicle would be more likely to suffer an abdomen injury to the liver,
which is located on the right side of the abdomen, than the spleen, which is located
on the left side of the abdomen. The positions of interior door components are
important to consider relative to an occupant’s anthropometry.
Lower extremity injuries mostly involve the pelvis and, for near-side occupants,
are caused by direct loading from the intruding door or B-pillar. Interaction with
the center console and belt latch has also been proposed as a mechanism of
pelvic injury for both near- and far-side occupants as they move toward the center
of the vehicle during the initial impact, i.e., far-side occupants, or as they
10 G.A. Mattos
rebound toward the center of the vehicle after the initial impact, i.e., near-side
occupants (Tencer et al. 2007).
Rollover
In a rollover crash, a vehicle overturns one or more times as a result of severe driver
input, e.g., cornering too fast, or tripping mechanism, e.g., tire impact with a curb.
While the circumstances of rollover crashes vary widely, more so than other crash
modes, there are general trends and similarities between them that have been found
to correlate to occupant response and injury outcome. A rollover crash can be
separated into three main phases: vehicle loss of control, initiation of roll, and
rollover.
The first two phases are relatively benign with regard to injury risk. However,
they significantly alter occupant motion and positioning that has carry-on effects for
the remainder of the event. During these phases, occupants experience primarily
horizontal inertial accelerations due to yawing, longitudinal braking, and lateral
skidding of the vehicle. This causes near-side occupants, i.e., those seated adjacent
to the direction of roll, to move forward and toward their door while far-side
occupants move forward and toward the center of the vehicle.
During the rollover phase, occupants undergo a series of discrete inertial and
direct loading scenarios due to a combination of free-flight rotation and vehicle-to-
ground contact events. Similar to the first two phases of a rollover crash, the free-
flight segments of the rollover phase present a relatively low injury risk. The primary
loading experienced by occupants during the free-flight segment is inertial and is due
to centrifugal accelerations that cause the occupants to move upward and outward
relative to their nominal seat position. This motion is primarily in the lateral and
vertical directions and is poorly mitigated by conventional restraints which are
designed to manage longitudinal motion of occupants. As a consequence, occupants
head move into close proximity to the upper interior vehicle roof structure and the
torso moves toward the door, irrespective of near- or far-side seat position.
Unlike other crash configurations, which last on the order of 100 ms, rollover
crashes can easily last for 5 s or more. This allows for occupants to react and
activate their musculature in an attempt to counteract the forces they are subjected
to during the event. Test occupants subjected to preroll maneuvers and pure
rollover events without impact have been observed to engage their neck and torso
musculature to maintain a nominal seated position and resist the inertial forces at
play (Lai et al. 2005; Yamaguchi et al. 2005). While the effects of such instances of
muscle activation are unknown for real-world rollovers, it is clear that occupant
motion in rollover crashes is highly complex, chaotic, and dependent on many
variables.
The highest risk of injury for belted occupants occurs during the instances of
direct impact between the vehicle and the ground as these instances result in the
greatest levels of deceleration during the event. The worst outcomes overall typically
occur as a result of complete occupant ejection, which is most often associated with
Fig. 4 Distribution of moderately (AIS2+) injured (left) or seriously (AIS3+) injured (right)
occupants in rollover crashes by body region sustaining moderate (AIS2+) or serious (AIS3+)
injury (Mattos et al. 2013; McMurry et al. 2016)
failure to wear a properly adjusted and/or fastened seatbelt. As a result of the vehicle
response that precedes the vehicle-to-ground impact events, i.e., loss of control,
initiation of roll, and free-flight rotation, occupants are typically not effectively
restrained and are in direct contact or close proximity to the adjacent door and
roof. The combination of multi-directional occupant response and vehicle loading
produces a crash event that has a higher rate of fatality and injury than any other
(National Center for Statistics and Analysis 2007).
Injury Patterns
The distribution of injuries by body region is summarized in Fig. 4. The majority of

occupants that sustained an AIS2+ or AIS3+ injury only received injuries to a single
body region.
Head injuries are one of the most common injuries sustained in rollover crashes.
Nearly one-quarter of all seriously injured (AIS3+) occupants in a rollover crash
suffer a serious head injury and 42% of moderately (AIS2+) injured occupants
sustain an AIS2+ head injury (Mattos et al. 2013; McMurry et al. 2016). For 30%
of moderately (AIS2+) injured occupants, the head is the only body region to sustain
an AIS2+ injury. The high rate of head injuries is due to the head’s relatively exposed
12 G.A. Mattos
position and proximity to the upper interior roof and side window: the most common
sources of head injury. The head is generally in contact with these structures when
they impact the ground during the rollover. These impacts directly transmit forces to
the head, which can be injuries depending on the impact speed and protection
provided by the upper interior padding. The head may also be subjected to impact
events at the end of a rollover event if the vehicle’s rotation is stopped abruptly as it
lands on its wheels and the occupant is flung across the cabin.
Spine injuries affect 19% of all moderately (AIS2+) injured and 36% of all
seriously (AIS3+) injured occupants. These injuries most often involve the cervical
spine in the form of fractures (Bambach et al. 2013b). While the majority of head
injuries occur without concomitant spine injuries, injuries to the cervical spine are
accompanied by head injuries approximately 75% of the time (Hu et al. 2005).
Injuries to the upper and middle regions, i.e., cervical and thoracic, of the spine are
generally thought to occur due to compressive loading of the spine. In a rollover
crash, this can result from either one or a combination of two mechanisms: intrusion
of a deforming roof onto the head of a seated occupant or an occupant diving into the
roof which is stationary against the ground. As such, the majority of cervical and
thoracic spine injuries are believed to result from interaction between the head and
interior vehicle roof. While less frequent, injuries to the lumbar spine do occur and
are most often attributed to interaction with the seatback or due to bottom-up loading
when the vehicle chassis impacts the ground.
Thorax injuries are more severe, on average based on the AIS score, than
injuries to other body regions. The most common AIS3+ injury scenario in a rollover
crash is the thorax being the only body region that suffers an AIS3+ injury, which
accounts for 13% of all occupants (McMurry et al. 2016). Thorax injuries often
involve a combination of single or multiple unilateral rib fractures and unilateral
lung contusions (Bambach et al. 2013a). As severity increases, bilateral injuries are
more common. These injures have been attributed to impact with the door, seatbelt
engagement, or seatback interaction. Intrusion of the door into the occupant space is
rare and not believed to be required to produce thoracic injuries. The mechanism of
injury is believed to be due to instances in which the occupant laterally impacts the
interior door or center console as the vehicle’s wheels decelerate laterally during
discrete impacts with the ground.
Abdominal injuries are relatively rare in rollover crashes as the abdomen is not
often subjected to intruding components or large impact forces. Relatively benign
interaction with the lap belt as well as limited fore-aft motion helps to limit
abdominal injury risk.
The upper extremities are more frequently injured than the lower extremities in
rollover crashes. The duration of the rollover provides adequate time for the upper
extremities to flail and become partially ejected through portals created when side
windows and windscreens disintegrate early in the event. Often, the upper extremity
is the only body region to suffer an AIS2+ injury.
Lower extremity injuries occur in a similar manner to upper extremity injuries.
The duration of the rollover event provides ample opportunity for the extremities to
flail and contact multiple interior components.
Rear
Rear impacts account for approximately one-quarter of all crashes. While the risk of
injury or fatality in rear impacts is relatively low, moderate and serious injuries that
do occur are often long-lasting. Similar to frontal crashes, rear crashes can occur as
large-overlap or small-overlap events. As in frontal crashes, large-overlap impacts
result in a primarily longitudinal crash pulse while those with small-overlap involve
lateral and rotational movement of the vehicle. Additionally, rear impact crashes may
also involve the striking vehicle overriding the struck vehicle without engaging with
the crush zone of the struck vehicle, as in the case of a large vehicle with a high ride-
height impacting the rear of a small car. This scenario results in extensive damage to
the struck vehicle since the impacted structures are not capable of managing the
energy applied during the impact.
A typical kinematic response in a low-severity rear impact is as follows
(McConnell et al. 1995):
Over the first 100 ms, the pelvis begins to load the forward moving seatback. This
loading is followed by the lower back and upper thoracic spine. During this time, the
seatback will begin to rotate rearward. In the next 100 ms, the neck begins to move
forward causing the head to rotate rearward until it contacts the head restraint. At this
point, the seatback has reached its maximum rearward rotation. The next phase is
defined by the torso, followed by the head, reaching and slightly exceeding the
forward velocity of the vehicle. This is followed by the occupant fully regaining
head control and returning to their preimpact position.
Occupant motion in high-severity rear crashes is dependent on seat performance.
Unyielding seats have the combined effect of generating greater loads on the occupant
as well as causing them to be launched forward at the end of the crash pulse which can
result in a greater opportunity for injury (Viano et al. 2008). Seats that deform during
the event provide more gradual loading and generate lower peak forces. The large
contact area between the occupant and seatback provides excellent restraint which is
able to very efficiently manage the loading of the occupant as they ride-down the crash
pulse. In crashes which result in severe intrusion or include passengers seated behind
front-seat occupants, too much seatback deformation can result in the occupant or
seatback impacting the intruding structure or rear-seated passenger. This highlights the
need for a balance of seatback compliance to provide optimal performance.
The risk of injury is increased for children in rear-facing seats, since in this crash
configuration they essentially experience a frontal crash. The relatively large head
size compared to neck and body size for children and infants results in a higher
probability of neck, spine, and basal skull injuries.
Injury Patterns
Serious injuries in rear-end crashes are relatively rare in low-severity, i.e., delta-V
less than 30 km/h, events. Only 0.1% of belted occupants exposed to a low-severity
impact sustain a serious injury. Occupants in high-severity, i.e., delta-V greater than
14 G.A. Mattos
Fig. 5. Distribution of seriously (AIS3+) injured occupants in low-severity (left) and high-severity
(right) rear crashes by body region sustaining serious (AIS3+) injury (Imler et al. 2014)
30 km/h, rear impacts are approximately ten times more likely to sustain a serious
injury (Imler et al. 2014). Injuries to the head and neck are generally the most severe
out of all injuries sustained. A summary of the distribution of injured body regions is
given in Fig. 5.
Head injuries most frequently involve the brain without concomitant skull
fractures. During a rear impact, the head typically only strikes the cushioned
headrest or seatback which provides good protection against skull fractures. In
rear crashes that involve oblique trajectories, occupants are more likely to strike
the side door or B-pillar which increases the risk of skull fracture. In high-
severity impacts, the front seatback can be deformed rearward such that it
impacts rear-seated occupants. This is especially severe for forward facing
children as their size and the dimensions of the seatback typically result in direct
head contact. Rebound events are also known to cause head injuries as the
occupant is flung forward to impact components that are not protected by airbags
(Galli and Digges 2008).
Spine injuries in the form of strains and sprains, identified commonly as whip-
lash injuries or whiplash-associated disorders, which result from motor-vehicle
crashes, most frequently occur to occupants in rear-end collisions. Although

whiplash-associated disorders are the most common motor vehicle injuries treated
in hospitals in the US, they are one of the least understood injury mechanisms in
automotive safety. These injuries are generally thought to be produced early in the
event when the head fails to effectively engage the headrest and is allowed to rotate
backward causing the spine to undergo a combination of extension and tension.
This scenario can occur in very minor crashes that often result in little or no damage
to the vehicle (Quinlan et al. 2004). If the impact and occupant kinematic response is
severe enough, it can also cause ligament damage, vertebral fractures, and spinal
cord trauma. While the exact time of injury generation is still under debate, the
basic mechanism of injury is thought to be excessive strain of tissues related to facet
joints, ligaments, vertebral arteries, nerve roots, and muscles of the neck (Siegmund
et al. 2009).
The thoracic spine can be injured due to asymmetric loading by an unyielding
seat in oblique or override impacts. In this scenario, the occupant’s head and
shoulders move rearward relative to the seat while the pelvis remains restrained by
the lap belt subjecting the thoracic spine to a combination of bending and shear
loading (Viano 2011). Three-quarters of all serious spine injuries involve the cervical
spine with the remainder sustained by the thoracic spine. Lumbar spine injuries are
rare in rear impacts, especially for belted occupants, due to the support afforded to
the lower back by the seatback and restraint system.
Thoracic injuries are generally limited to lung contusions and rib fractures that
occur due to the direct loading between the occupant and seatback.
Abdomen injuries rarely occur in any type of rear impact, and are nearly
nonexistent in low-severity crashes. The most commonly injured organs are the
liver and spleen and are likely caused by direct loading between the occupant and
seatback.
Upper and lower extremity injuries are similarly rare and generally involve
fractures of the long bones and pelvis. The upper extremities are often injured
due to flailing or interaction with a deploying airbag. The risk of injury to the
extremities is quite low for a low-severity single-event rear-end crash as the
extremities are not likely to contact any interior component of the vehicle with
appreciable velocity (Rundell et al. 2013). However, high-severity crashes, in
which the seatback deformation is limited, can result in the occupants being
launched into the front components of the vehicle at the end of the crash pulse.
Often, these secondary impacts occur without the injury mitigating effects of
frontal airbags, which do not usually deploy in rear crashes. If the rebound event
is severe enough, the occupant, who is likely to be out of position and not
effectively restrained, will be at risk of suffering extremity injuries due to contacts
with the front interior components. This effect can be exacerbated in events that
involve secondary frontal impacts. One notable proposed injury mechanism
involves loading of the back of the knee by the seatpan causing injury to the
ligaments in the knee (Rundell et al. 2013).
16 G.A. Mattos
Countermeasures
A wide range of countermeasures and design features exist that have been shown to
reduce the risk and severity of injury in each crash configuration. This section
provides a summary review of notable features that promise to significantly reduce
injury risk in future vehicles.
The combined effectiveness of airbags and seatbelts on reducing serious injuries
in all crashes is estimated to be as high as 64% (Kent et al. 2005), and the seatbelt
alone is credited with saving the lives of over 15,000 passenger vehicle occupants
per year in the US (Kahane 2015). While the benefits of restraints are most notable in
frontal crashes, their ability to keep the occupant within the vehicle in all crash
configurations significantly reduces the risk of injury. Advances in restraint design
have the potential to even further increase the safety benefit for all crash modes. For
example, active restraints, such as motorized retractors, can be used to remove slack
and ensure the occupant is optimally restrained prior to the crash.
Airbags continue to be improved and implemented to different aspect of the
vehicle. Some examples include the front center airbag, designed to limit occupant-
to-occupant contact during side and rollover crashes, and inflatable belts which
promise to further reduce injury risk in frontal crashes, especially for smaller and
older occupants. Rollover activated side curtain airbags with long-duration inflators
are designed to prevent partial ejections in rollover crashes by covering potential
ejection portals created when side windows disintegrate.
The primary safety feature for reducing the risk of injury in rollover crashes is an
adequately strong roof. A roof that is capable of maintaining the integrity of the
occupant space will reduce the occupant’s risk of coming into contact with the
intruding structure. An important countermeasure that is required to work with a
strong vehicle roof is the upper interior padding, which is installed on the interior
roof structures above the vehicle beltline. This passive countermeasure has been
shown to reduce severe head injuries due to impacts with the interior roof in rollover
crashes by simply softening impacts to the occupant’s head (NHTSA 2011).
Summary
The kinematic and injury response of motor-vehicle occupants is complex and depen-
dent on a wide range of crash, vehicle, and occupant factors. The knowledge of
occupant response and injury risk in relation to crash configuration is vital to many
aspects of road safety from vehicle design to post-crash care, where the ability to
diagnose traumatic injuries is greatly benefitted by knowing where to look. As
demonstrated above, occupants in crashes are at the mercy of the vehicle’s kinematics
and structural performance. The human body is limited in its ability to tolerate impact,
and the efforts to improve vehicle safety are focused on providing an occupant
compartment that prevents these thresholds from being exceeded. As the understand-
ing of both injury tolerance and occupant response improves, so does the opportunity
to increase the crashworthiness of vehicles and better protect all occupants.
References
AAAM (2008) Abbreviated injury scale, 2005 revision – update 2008. Association for the
Advancement of Automotive Medicine, Des Plaines
Bambach MR, Grzebieta RH, McIntosh AS (2013a) Thoracic injuries to contained and restrained
occupants in single-vehicle pure rollover crashes. Accid Anal Prev 50:115–121. doi:10.1016/j.
aap.2012.03.033
Bambach MR, Grzebieta RH, Mcintosh AS, Mattos GA (2013b) Cervical and thoracic spine injury
from interactions with vehicle roofs in pure rollover crashes. Accid Anal Prev 50:34–43.
doi:10.1016/j.aap.2012.10.006
Beck B, Bilston LE, Brown J (2016) Injury patterns of rear seat occupants in frontal impact: an
in-depth crash investigation study. Inj Prev 22(3):165–170. doi:10.1136/injuryprev-2015-
041715
Chen R, Gabler H (2014) Risk of thoracic injury from direct steering wheel impact in frontal
crashes. J Trauma Acute Care Surg 76(6):1441–1446
Fildes BN, Fitzharris M, Gabler H, Digges K, Smith S. Chest and abdominal injuries to occupants in
far side crashes. Paper presented at the 20th international technical conference on the enhanced
safety of vehicles; 2007. Lyon, France.
Fitzharris M, Franklyn M, Frampton RJ, Yang K, Morris A, Fildes BN (2004) Thoracic aortic injury
in motor vehicle crashes: the effect of impact direction, side of body struck, and seat belt use. J
Trauma Injury Infect Crit Care 57(3):582–590
Forman JL, Lopez-Valdes F, Lessley DJ, Riley P, Sochor M, Heltzel S, Ash J, Perz R, Kent RW,
Seacrist T, Arbogast KB, Tanji H, Higuchi K (2013) Occupant kinematics and Shoulder Belt
retention in far-side lateral and oblique collisions: a parametric study. Stapp Car Crash J
57:343–385. doi:10.4172/2155-6210.S1-002
Gabler HC, Digges K, Fildes BN, Sparke L (2005) Side impact injury risk for belted far side
passenger vehicle occupants. SAE Trans J Passenger Car Mech Syst 114(6):114–116
Galli R, Digges K (2008) Severe head and neck injuries in NASS rear impacts. SAE Technical
Paper 2008-01-0190, doi:10.4271/2008-01-0190
Hallman JJ, Yoganandan N, Pintar FA, Maiman DJ (2011) Injury differences between small and
large overlap frontal crashes. Ann Adv Automot Med 55:147–157
Hu J, Lee JB, Yang KH, King AI (2005) Injury patterns and sources of non-ejected occupants in
trip-over crashes: a survey of NASS-CDS database from 1997 to 2002. Annu Proc Assoc Adv
Automot Med 49:119–132
Imler S, Heller M, Raasch C, Watson H, et al. (2014) The effect of rear impact collision delta-V and
restraint status on injury outcome. SAE Technical Paper 2014-01-0524, doi:10.4271/2014-01-
0524
Jernigan MV, Rath AL, Duma SM (2005) Severe upper extremity injuries in frontal automobile
crashes: the effects of depowered airbags. Am J Emerg Med 23(2):99–105. doi:10.1016/j.
ajem.2004.02.045
Kahane C. Lives saved by vehicle safety technologies and associated federal motor vehicle safety
standards, 1960 to 2012 – passenger cars and LTVs. Paper presented at the 24th technical
conference on the enhanced safety of vehicles; 2015. Gothenburg, Sweden.
Kent R, Viano DC, Crandall J (2005) The field performance of frontal air bags: a review of the
literature. Traffic Inj Prev 6(1):1–23. doi:10.1080/15389580590903131
Lai W, Ewers B, Richards D, Carhart M, Newberry W, Corrigan CF (2005) Evaluation of human
surrogate models for rollover. Society of Automotive Engineers, Warrendale
Mattos GA, Grzebieta RH, Bambach MR, McIntosh AS (2013) Head injuries to restrained
occupants in single-vehicle pure rollover crashes. Traffic Inj Prev 14(4):360–368.
doi:10.1080/15389588.2012.722735
McConnell WE, Howard RP, Poppel JV, Krause R, Guzman HM, Bomar JB, Raddin J, Benedict JV,
Hatsell CP (1995) Human head and neck kinematics after low velocity rear-end impacts –
understanding “whiplash”. Society of Automotive Engineers, Warrendale
18 G.A. Mattos
McIntosh AS, Kallieris D, Frechede B (2007) Neck injury tolerance under inertial loads in side
impacts. Accid Anal Prev 39(2):326–333. doi:10.1016/j.aap.2006.09.001
McMurry TL, Bose D, Ridella SA, Eigen AM, Crandall JR, Kerrigan JR (2016) Epidemiology of
moderate-to-severe injury patterns observed in rollover crashes. Accid Anal Prev 90:36–40.
doi:10.1016/j.aap.2016.01.018
National Center for Statistics and Analysis (2007) An analysis of motor vehicle rollover crashes and
injury outcomes. National Center for Statistics and Analysis, Washington, DC
NHTSA (2011) Evaluation of the 1999–2003 head impact upgrade of FMVSS no. 201 upper-
interior components (DOT HS 811 538). NHTSA, Washington, DC
Pintar F, Maiman DJ, Yoganandan N (2007) Injury patterns in side pole crashes. Ann Adv Automot
Med 51:419–433
Pintar F, Curry W, Halloway DE, Malik W, Yoganandan N (2014) Thoracolumbar spine fractures in
frontal crashes: an update of field data. Traffic Inj Prev 15(sup1):S254–S256. doi:10.1080/
15389588.2014.956646
Quinlan KP, Annest JL, Myers B, Ryan G, Hill H (2004) Neck strains and sprains among
motor vehicle occupants – United States, 2000. Accid Anal Prev 36(1):21–27. doi:10.1016/
S0001-4575(02)00110-0
Reichert R, Morgan R, Park C, Digges K, Kan C-D. Thoracic and abdominal injuries to drivers
in between-rail frontal crashes. Paper presented at the International Research Council on the
Biomechanics of Injury (IRCOBI); 2013 Sep 11–13. Gothenburg, Sweden.
Rundell S, Guiang A, Weaver B, Meyer E (2013) Characterization of occupant lower extremity
behavior during moderate-to-high speed rear impacts. SAE Technical Paper 2013-01-0222,
doi:10.4271/2013-01-0222
Rupp JD, Flannagan CAC, Leslie AJ, Hoff CN, Reed MP, Cunningham RM (2013) Effects of BMI
on the risk and frequency of AIS 3+ injuries in motor-vehicle crashes. Obesity 21(1):E88–E97.
doi:10.1002/oby.20079
Siegmund GP, Winkelstein BA, Ivancic PC, Svensson MY, Vasavada A (2009) The anatomy
and biomechanics of acute and chronic whiplash injury. Traffic Inj Prev 10(2):101–112.
doi:10.1080/15389580802593269
Tencer AF, Kaufman R, Huber P, Mock C, Rout ML (2007) Reducing primary and secondary
impact loads on the pelvis during side impact. Traffic Inj Prev 8(1):101–106. doi:10.1080/
15389580600944250
Viano D (2011) Fracture-dislocation of the thoracic spine in extension by upright seats in severe rear
crashes. SAE Technical Paper 2011-01-0274, doi:10.4271/2011-01-0274
Viano D, Parenteau C, Prasad P, Burnett R (2008) Occupant responses in high-speed rear crashes:
analysis of government-sponsored tests SAE Technical Paper 2008-01-0188, doi:10.4271/2008-
01-0188
Viano DC, Parenteau CS (2010) Injury risks in frontal crashes by Delta V and body region with
focus on head injuries in low-speed collisions. Traffic Inj Prev 11(4):382–390. doi:10.1080/
15389581003751623
Yamaguchi GT, Carhart MR, Larson R, Richards D, Pierce J, Raasch CC, Scher I, Corrigan CF
(2005) Electromyographic activity and posturing of the human neck during rollover tests.
Society of Automotive Engineers, Warrendale
Yoganandan N, Baisden JL, Maiman DJ, Gennarelli TA, Guan Y, Pintar FA, Laud P, Ridella SA
(2010) Severe-to-fatal head injuries in motor vehicle impacts. Accid Anal Prev
42(4):1370–1378. doi:10.1016/j.aap.2010.02.017
Yoganandan N, Pintar F, Humm J, Rudd R (2015) Injuries in full-scale vehicle side impact moving
deformable barrier and pole tests using Postmortem human subjects. Traffic Inj Prev 16(sup2):
S224–S230. doi:10.1080/15389588.2015.1062887
Body Movements in Music Performances:
The Example of Clarinet Players
Manfred Nusseck, Marcelo M. Wanderley, and Claudia Spahn
Abstract
Musicians move in many ways when performing music. Such movements, or
gestures, may have a variety of roles in a performance. They can directly and
indirectly control the sound of the instrument and build up communication
between performers or between performers and the audience. Although omni-
present in performances, some of these gestures, however, do not seem to have a
clear relationship to the process of sound production. In this chapter, general
aspects of movements of musicians in music performances, and the distinction of
these movements, are introduced and illustrated with an analysis of ancillary
movements of clarinet performers.
Keywords
Music performance • Musical gestures • Ancillary movements • Clarinet • Motion
capture
M. Nusseck (*) • C. Spahn

Freiburg Institute for Musicians’ Medicine, University of Music Freiburg, Medical Center –
University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany
e-mail: Manfred.nusseck@uniklinik-freiburg.de; Claudia.spahn@uniklinik-freiburg.de
M.M. Wanderley
Input Devices and Music Interaction Laboratory (IDMIL), CIRMMT, McGill University, Montreal,
QC, Canada
e-mail: marcelo.wanderley@mcgill.ca

DOI 10.1007/978-3-319-30808-1_107-1
2 M. Nusseck et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Measuring Gestures: The Example of Clarinet Players . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Influences of Gestures on the Musical Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Similarities and Differences Between Music and Sport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Introduction
When watching a professional clarinetist on stage, one can observe a variety of

movements, such as the motions of the fingers on the instrument, the clarinet moving
around lifted, lowered, or in circles, and the player leaning forward and backward,
possibly with associated facial expressions following the emotional melodic flow.
Furthermore, one may perceive oneself as an observer also moving, following the
music or the movements of the performer.
Musicians make music with body movements. The goal of the performance is to
convey musical structures and intended expressions with an instrument1 to an
audience. For instance, in classical western music, the musician interprets the
musical content of the piece and addresses it during the performance in terms of
phrasing, rhythm, and melody, by using timing, articulation, dynamic, and timbre
(Kendall and Carterette 1990; Palmer 2013) as well as expressive nonverbal com-
munication signals (Davidson 2012). The performance might depend on the score,
the performance situation, and the audience. Therefore, this process can be described
as an overlapping interaction between creation, perception, and performance (Palmer
2015).
Everyone can make music to some extent. Clapping, humming, and swaying to
music require no musical training. They are built upon certain kinds of body motions
that stand in direct relation to the music and to musical activities and represent the
basics of music performance. Leman and Godøy (2010) speak of a profound and
inseparable engagement between both sound and movements, and note that the
experience of music is rooted in the sensations of movements.
The theory of embodied cognition proposes that cognitive aspects are strongly
linked with properties and experiences of the body. Embodied music cognition
supposes that the musical mind is based on specific body experiences (see Leman
and Godøy 2010). This works for both the production and the perception of music.
On the production side, the body is always included in the procedure of making
music; therefore music needs the body. For the perception of music, the body
individually reacts to the musical features with accompanying movements and/or
expressive and emotional manners; hence music induces movements.
1
In this text, the term “instrument” always also includes vocals.
Body Movements in Music Performances: The Example of Clarinet Players 3
Body movements in music performances have been an increasingly common

topic in music research disciplines such as musicology and music perception, and
also in human movement science, cognitive psychology, and computer sciences. A
specific interest has been devoted to the role of gestures in music (Gritten and King
2006; Godøy and Leman 2010). Aside from the music itself, musical activities also
involve certain aspects of communicative movements. The so-called musical ges-
tures are biologically and culturally grounded and shape affectively the expressive
and emotional elements of the musical performance (Hatten 2004). Research in
musical gestures has focused on several aspects, such as characteristic movements
of musicians, sensorimotor feedback, and the connection between perception and
action (Leman and Godøy 2010).
Music, however, can easily be appreciated without seeing the musician (Dahl
et al. 2010), as visible gestures of the musician are not directly necessary to perceive
music. It is the product of the performance, i.e., the music itself, which is being
enjoyed in the first place. Possibly, a listener would be able to infer intended gestures
only with access to the sounds produced by a performer. Those gestures, however,
may not be the gestures actually performed by the musician, but individually
imagined by the listener. Musical gestures may therefore be included in any form
of musical performance, although the music itself is not reducible to them (Hatten
2004). Nevertheless, seeing a musician during a musical performance can influence
the perception of the performance (Vines et al. 2006b; Schutz and Lipscomb 2007;
Platz and Kopiez 2012).
The goal of this article is to provide an overview of motion aspects in musical
performances. The state-of-the art theories on gestures in music and the distinction
between sound-producing and ancillary gestures are introduced in the first section.
An overview of researches on ancillary gestures of clarinet players is given in the
second section. Further research on motion aspects for other instruments can be
found in other chapters of this handbook (for drumming see “▶ Movements, Timing
and Precision of Drummers”; for violin see “▶ Investigating Aspects of Movement
in Violin Performance”; for piano see “▶ Movement and Touch in Piano Perfor-
mance”; for guitar see “▶ Observing and Learning Complex Actions: On the
Example of Guitar Playing”). That musical experience influenced by ancillary
gestures seen in clarinet performances is discussed in the third section. As musicians
have often been compared with high-performance athletes, similarities and differ-
ences between music and sports will be discussed in the fourth section.
State of the Art
Movements in music performance are commonly called gestures, as this term

describes intentional bodily actions conveying meaning, and are often associated
with particular hand movements, body postures, and facial expressions. In this sense,
music made with acoustic instruments cannot exist without gestures (Cadoz 1988).
Several functional classifications of gestures have explored the movements made
by musicians. In general, body movements in music performances can be
4 M. Nusseck et al.
distinguished into movements which are directly responsible for the production of
sound, and into movements which are not involved in sound production (Delalande
1988; Cadoz and Wanderley 2000). Sound-producing movements include, for
instance, the actions of hitting a key, pulling the strings, bowing, etc. However,
there are other, ancillary, movements that do not necessarily produce sounds, but
which might have a supportive or communicative goal, and carry expressive inten-
tions and meanings. Additionally, there are also movements that accompany the
music and might be made even when the instrument is not played to maintain the
musical and expressive flow in the performance. Above and beyond these move-
ments, there are also gestures used to directly interact with others, i.e., to commu-
nicate and signal to other players or the audience. These non-sound-producing
movements are called ancillary (Wanderley 2002) or expressive (Davidson 1993)
gestures.
In this chapter, we will focus on ancillary or expressive gestures in clarinet
performance and examine two main questions:
– Are they necessary at all? In other words, can musicians perform without
ancillary gestures?
– Are they repeatable? That is, do musicians make similar ancillary gestures when
performing the same piece multiple times?
First, we will introduce the theoretical background of gestures in music perfor-

mances. Jensenius et al. (2010) broadened the classification of music-related move-
ments with a four-tier typology of gestures according to their specific function:
• Sound-producing gestures, which effectively produce the sound on the instrument

• Sound-facilitating gestures, which act as support for sound-producing gestures
and are not directly involved in sound production
• Sound-accompanying gestures, which are movements that follow the music,
following or mimicking sound-producing and sound-facilitating gestures without
actually producing sounds
• Communicative gestures, which intend to directly communicate with others, such
as another performer or the audience. These can be gestures such as eye contact or
a head nod to provide cues for controlling and timing of musical content and
expression between co-performers (Davidson 2011).
Sound-producing gestures can be further divided into two subtypes: excitation

gestures, which directly create the sound, such as the key strokes on the piano, and
modification gestures, which do not produce sound by themselves, but modify the
quality of the sound, such as the pedal of the piano (Cadoz and Wanderley 2000;
Jensenius et al. 2010).
The sound-facilitating gestures are used as specific nuances made at certain points
in the music such as structural or stylistic features, these are, for instance, a slowing
down or stopping at certain musical passages. These gestures convey movements of
intended expressiveness and specific personal components of the player and carry a
great deal of emotional aspect. However, they occur more in solo performances and
with an audience present than in ensemble performances and without an audience
(Davidson 2012; Glowinski et al. 2013).
Davidson (1993) showed that detecting expressive intentions, such as playing
deadpan, projected, and exaggerated, can be done just by seeing these gestures. In a
task to identify emotional intentions (happy, sad, angry, and afraid) of, in this case a
marimba performer, observers could easily detect the intended expression when only
the head, the arms, or the trunk movements, i.e., not the sound-producing body area,
were seen (Dahl and Friberg 2007). These examples show that ancillary gestures
convey especially expressive information.
Theoretically, expressive ancillary gestures are not strictly necessary for the
music production, and they can even become distracting and obstructive (Davidson
1993). In pop or rock events, on the other hand, performers rely heavily on these
gestures and use them as a main source of communication with the audience
(Lehmann and Kopiez 2013). In general, the necessity and interpretation of these
gestures are very subjective and depend on the performer and the performing
situation. It is also important to consider cultural biases that strongly influence a
person’s appreciation of such movements.
Sound-accompanying gestures can be seen as filling elements between sound-
producing parts, such as rhythmical postural swaying or head nodding. They also
function as show elements for the audience. Between performers they can help to
keep the pace (Goebl and Palmer 2009) or the emotional and expressive flow alive.
These gestures can therefore provide some sort of continuity, even when no musical
sound has to be produced (Hatten 2004).
Generally, movements in music performances serve multiple purposes (Davidson
2011). In fact, the various gestures can overlap each other, and, as in the case of most
functional classifications, their functions may be difficult to distinguish and separate
(Dahl et al. 2010). For instance, the postural sway of a performer is not a simple
product of the beat plus the expression. It contains interactive elements between
these gestures (Demos et al. 2014). Therefore, music performances are complex
systems which differ each time the player performs (Davidson 2011). Furthermore,
the performance differs for the same player between playing solo or in an ensemble
(Palmer 2013). However, specific movement patterns reoccur in repeated perfor-
mances of the same player, as they belong to the execution of the particular piece.
This will be demonstrated in the next section.
Measuring Gestures: The Example of Clarinet Players
To investigate performance gestures, a musician has to be recorded and analyzed

during an actual musical performance, thus the measuring equipment should not
distract from the musical act or the instrumental play. Every instrument has its own
particular regions of motion interest, and some instruments provide a clear view on
these movement areas, while in other instruments, the direct view may be blocked
out by the instrument.
6 M. Nusseck et al.
Fig. 1 Screenshot of a clarinetist during a motion capture recording and the resulting digital stick
figure (Source: Nusseck et al. 2007)
Recent video analysis techniques and the development of motion capture by

tracking small wireless markers make it possible to directly measure motions in a
real performance. These motion capture systems allow the recording of certain body
areas while playing, without drastically impeding the musician’s movements. After
the recording, the tracked markers can be transformed into a digital 3D dataset to
allow the analysis of each marker position and to build specific illustrations of the
player such as a digital skeletal representation of the player by plotting the markers,
the so-called stick figures (see Fig. 1). By using motion capture techniques, diverse
sets of features can be extracted to reveal behaviors of body parts and their gestures
in music performances.
Analyses of particular motion areas of interest can help to understand general
movement behaviors of performers and investigate the information flow of specific
gestures, for instance, expressive motivations.
For clarinet players, finger movements, lip pressure (embouchure), and blowing
are the sound-producing gestures, whereas motions like weight shifting, knee bend-
ing, clarinet lifting, and body curvature are ancillary gestures. Creating sound on a
clarinet is therefore a less visible process with small and inconspicuous gestures. The
most visually salient movements are ancillary gestures.
Though not directly involved in sound production, clarinetists’ ancillary gestures
are not randomly produced or just a visual effect. It has been shown that they are not
essential, i.e., they can be eliminated (or at least attenuated), that they are repeatable
(same musician performing a piece multiple times), that they are consistent across
performance manner (standard and expressive performances, performer standing up
or seated), and that they are influenced by several factors (physiological, instrumen-
tal, and musical) (Wanderley 2001, 2002; Wanderley and Vines 2006). For instance,
certain movements of the clarinet bell are in conjunction with specific passages in the
musical notation (Teixeira et al. 2014).
Performer MJ, 1st Session

200
150
Functional Values for Bell Height
100
50
–50
–100 Immobile
Standard
Exaggerated
–150
8 10 12 14 16 18 20
Time (seconds)
Fig. 2 Vertical movement of the clarinet bell for the first two staff lines of Stravinsky’s second of
the Three Pieces for Solo Clarinet. Three performance manners (immobile, standard, and exagger-
ated), first session performance
To further investigate ancillary gesture differences between performances of the

same musician, Vines et al. (2006a) analyzed multiple performances recorded at
different times. Figures 2 and 3 show the vertical movement of the clarinet bell over
time, for two performance sessions by the same performer, of the first few bars of
Igor Stravinsky’s second of the Three Pieces for Clarinet Solo. The figures show
multiple time-warped performances of this excerpt recorded on two different occa-
sions and with three performance manners: immobile (while restraining body move-
ment as much as possible), standard (as in a concert), and exaggerated (with
intentional overexaggeration of expressiveness, nothing was mentioned about move-
ments). The data shown correspond to one immobile performance (green), the
average of three standard performances (in session 1, and four standard perfor-
mances in session 2) (red), and the average of two exaggerated performances (blue).
Looking at these figures, one can see that, when the same clarinet player performs
a piece multiple times, similar patterns can be detected at the same points in the score
across the different performances, showing that, for this performer, ancillary gestures
of the bell are not random, but clearly repeatable, even over large periods of time
(Dalca et al. 2013). One can also note that the movements seem to have some form of
periodicity, in this case, with respect to the phrasing of the piece (Wanderley et al.
2005).
8 M. Nusseck et al.
Performer MJ, 2nd Session

Functional Values for Bell Height (mm) 200
150
100
50
–50
–100 Immobile
Standard
Exaggerated
–150
8 10 12 14 16 18 20
Time (seconds)
Fig. 3 Vertical movement of the clarinet bell for the first two staff lines of Stravinsky’s second of
the Three Pieces for Solo Clarinet. Three performance manners (immobile, standard, and exagger-
ated), second session performance, several months afterwards
Furthermore, Figs. 2 and 3 show that at certain times, for instance, between
12 and 14.5 s, the performer either augmented the amplitude of the movements
(in the exaggerated performance, in blue) or dramatically reduced it (in the immobile
performance, in green), while at other moments, smaller movement differences
occurred, for example, between the standard and the exaggerated averages.
This suggests that the change in magnitude of expressive gestures depends both
on how the piece is played (i.e., performance manner), as well as on what is being
played (i.e., the score). In other words, although the amplitude of ancillary gestures
can be manipulated with respect to performance manner, this does not necessarily
happen all the time. For instance, in overly technical passages, there is not much
room for either movement attenuation or exaggeration, while in other parts of the
piece, where more freedom is possible, the amplitude of ancillary gestures can be
more freely modulated according to performance manner (Vines et al. n.d.).
In short, quoting Dalca et al. (2013, pp. 853), “expressive musical gestures are,
thus, a durable indicator of the performer’s stylistic intentions (reflecting his or her
cognitive and affective interpretation of the score) and are an integral element of the
ongoing music, rather than a parallel, complementary behaviour reflecting additional
cues of momentary disposition, or embellishments to performance demands.”
Influences of Gestures on the Musical Experience
Several studies have investigated the influence of movements on the perception of

the performance, focusing on why musicians use more movements than necessary.
Generally, it was found that seeing movements changes the perceived musical
experience (Vines et al. 2006b; Schutz and Lipscomb 2007; Platz and Kopiez
2012). Although there are several other research questions on the topic of effects
of musicians’ motions, this chapter will only focus on examples related to the
questions of (a) the influence on the musical experience and (b) the influence on
the judgment of the performer’s skill.
Research on the effects of gestures in shaping the musical experience focuses
mainly on the detection of – and/or the differences between – expressive intentions
in musical gestures. For that, Davidson (1993) asked musicians (i.e., a violinist and a
pianist) to play with different performance manners (deadpan, projected, and exag-
gerated) and presented the performances only visually, only acoustically or with both
sound and vision. Results showed that expressiveness was easily distinguished in all
three presentation modes, where vision only revealed the best differentiation rates.
The author summarized that visual-only information carries a great amount of
information about expressiveness in a performance. As mentioned above, this has
also been shown for the recognition of emotional intentions when only different
parts of the marimba player, i.e., the head, arms, and torso, were visible and no sound
was presented (Dahl and Friberg 2007).
In these studies, the musicians were asked to perform in specific ways. In a
different research design, Nusseck and Wanderley (2009) used one original perfor-
mance and afterwards manipulated it to produce different performances. By gener-
ating stick figures using data obtained with motion capture techniques, the
amplitudes of clarinet players’ movements were altered (reduced or exaggerated)
or even blocked out in the way that only arm movements or the swaying of the body
were shown, with other parts frozen.
By letting observers rate the musical intensity of the performance, it was found
that the perceived intensity clearly correlated with changes in motion amplitudes
(Fig. 4). The larger the motions were, the higher was the experienced musical
intensity. Results showed that manipulating the original presentation led to different
experiences of the performance. However, when presenting the stick figures with
different body parts of the player frozen, the ratings were less different, suggesting
that the main musical intentions were stored rather equally, i.e., holistically, in the
whole body (Nusseck and Wanderley 2009).
Judgments of the skill of the performer can also be changed by seeing the
musician’s movements. In the same study, the observers were also asked to rate
the professionalism of the player. By reducing the motion amplitudes down to half of
their range, the impression of professionalism decreased significantly (Fig. 4).
However, increasing the motions above the range of the original did not raise the
perception of professionalism of the performance.
As in the study of Nusseck and Wanderley (2009), in which the same acoustical
source was played in all conditions, Rodger and colleagues (Rodger et al. 2012)
10 M. Nusseck et al.
Fig. 4 Mean ratings of the Intensity

musical intensity (top) and the 7
professionalism of the player
(bottom) according to the 6
amplitude manipulations in
motion capture stick figures 5
(From Nusseck and
Wanderley 2009) 4
1
150% Original 50% 20%
Professionalism
7
1
150% Original 50% 20%
performed experiments by switching video and audio sources between a novice and
a professional player and asked observers to rate the expertise of the player. The
results showed that the visual information from experts on top of the sound of a
novice increased the rating of expertise, compared to when displaying both video
and sound of the novice. In contrast, the sound from experts with the movements
from the novice player did not change the expertise ratings compared to the
presentation of both video and sound of the expert. The authors conclude that acting
(visually) like an expert can increase the professionalism impression of a novice
player, whereas a visually presented beginner did not alter an acoustically expert
presentation.
In a study by Tsay (2013), observers were asked to identify the winner among the
three best players in an international music competition when only the video, only
the audio, or both video and audio were presented. The best results were found for
the video-only presentation. The author concludes that the motions of a professional
solo performance were used as the main information source.
Similarities and Differences Between Music and Sport
The analogy between expert musicians and high-performance athletes is often made
as there are obvious similarities. Generally, music and sports performances require
distinct skill acquisition, intensive practice (mental and physical), good motor
coordination, and great demands on cognitive capacity and memory (Kenny
2011). In both performances, complex motor activities are involved that require a
certain amount of perfection in timing and movement accuracy (Wulf 2013). During
the learning process, different stages of motor learning are achieved, from a more
cognitive stage of focusing on skill learning toward an autonomous stage with less
awareness and minimum cognitive effort on the motor action and more capacity for
additional aspects (Mesagno et al. 2016). However, learning strategies cannot be
simply transferred to improve performance. For instance, letting novice musicians
learn an easy disk-throwing task with the same strategy as that used by experts did
not increase their performance outcome (Hodges et al. 2011).
As both musical and sport performances normally take place in front of an
audience, they share the opportunity to enjoy the experience of excellence (Yoshie
et al. 2009) and also develop the ability to perform under pressure and consequen-
tially to deal with the experience of performance anxiety (Mesagno et al. 2016;
Spahn 2015; Kenny 2011; Yoshie et al. 2009). Therefore, research insights from
music and sports can clearly benefit each other, and there is an overlapping research
on the topics of training/practicing methods and especially on recovery activities
(Wagner et al. 2016).
Despite the similarities between musicians and athletes, there are also extensive
differences. The most prominent ones are (a) that music conveys artistic elements
such as musical, emotional, and expressive components and (b) that the criteria used
to judge the quality of a performance are very different between musical and athletic
performances (Yoshie et al. 2009).
While a music performance is a communicative and expressive act and stands in
an interactive process with an audience, sport performances mainly follow the goal
of increasing an objective measurable score. For instance, a musician’s primary
ambition is not necessarily to perform as accurately as possible, whereas accurate-
ness is essential in sports. Accuracy is obviously important in music performance
too, but it is only one aspect helping to perform as intended and to realize specific
expressive intentions. Music cannot therefore be reduced to accuracy only.
Similarly, while athletes attempt to avoid as much as possible any accompanying
movements that are not necessary for reaching the goal of achieving shorter timing,
longer distance, better throwing, etc., musical performances involve body move-
ments that evoke expression and create communicational elements which are not
directly linked to sound production (see section “▶ State of the Art”). Although
musicians also avoid extra extreme motions which affect their instrumental perfor-
mance, other additional movements may help them express personal and musical
intentions.
Summary
The goal of this chapter was to introduce the topic of movements in music perfor-
mances. The specific distinctions between musician’s movements during perfor-
mances were presented. Also, similarities and differences between movements in
sports and in music performance were discussed. Though sharing many similarities
(e.g., high accuracy and precision of timing control), they mainly differ in their
goals: music performance body movements are not only responsible for the sound
production but also contribute communicative and expressive elements of the
performance.
In this chapter we focused on sound-producing and ancillary gestures, specifically
the production and perception of ancillary gestures by clarinet performers. However,
to clearly quantify individual gestures and their meaning for the musician and the
audience is still an ongoing challenge.
Nonetheless, several important related topics were not discussed here, such as
teachers’ communication of musical knowledge through gestures, which represents a
valuable pedagogical aspect of musical gestures (Simones et al. 2015), or the
movements in conducting (Wöllner 2008). Conductor’s movements can be thought
of as a special class of communicative gestures that have an influence on the sound
result (Nakra 2000; Gambetta 2005).
In summary, the research of motion in music performances sheds insights into
how musicians interact with the instrument, the audience, and with other performers
and help not only to understand the motion behavior but also to improve playing.
Acknowledgments The second author would like to thank Bradley Vines and Ioana Dalca for their
long-term collaboration, Mauricio Alves Loureiro and Euler Teixeira for several comments and
suggestions to this manuscript, as well as the Natural Sciences and Engineering Research Council of
Canada (NSERC) for partially funding his research.
References
Cadoz C (1988) Instrumental gesture and musical composition. In: Proceedings of the 1998
International Computer Music Conference, Den Haag, Netherlands, 60–73
Cadoz C, Wanderley MM (2000) Gesture – music. In: Wanderley MM, Battier M (eds) Trends in
gestural control of music. IRCAM – Centre Pompidou, Paris, pp 71–94
Dahl S, Friberg A (2007) Visual perception of expressiveness in musicians’ body movements.
Music Percept 24(5):433–454
Dahl S, Bevilacqua F, Bresin R, Clayton M, Leante L, Poggi I, Rasamimanana N (2010) Gestures in
performance. In: Godøy RI, Leman M (eds) Musical gestures: sound, movement, and meaning.
Routledge, New York
Dalca IM, Vines BW, Pearce MT, Wanderley MM (2013) Expressivity as time-dependent move-
ment for music performance: a statistical exploration. Proc. of the 10th International Symposium
on Computer Music Multidisciplinary Research (CMMR2013), 845–854
Davidson J (1993) Visual perception and performance manner in the movements of solo musicians.
Psychol Music 21:103–113
Davidson J (2011) Movement and collaboration in musical performance. In: Hallam S, Cross I,
Thaut M (eds) Oxford handbook of music psychology. OUP, Oxford
Davidson JW (2012) Bodily movement and facial actions in expressive musical performance by
solo and duo instrumentalists: two distinctive case studies. Psychol Music 40:595–633
Delalande F (1988) La Gestique de Gould. Élements pour une sémiology du geste musical. In:
Guertin G (ed) Glenn Gould Pluriel. Louise Courteau Éditrice
Demos AP, Chaffin R, Kant V (2014) Toward a dynamical theory of body movement in musical
performance. Front Psychol 5:477
Gambetta, C (2005) Conducting outside the box: creating a fresh approach to conducting gesture
through the principles of laban movement analysis. DMA Dissertation, University of North
Carolina at Greensboro
Glowinski D, Mancini M, Cowie R, Camurri A, Chiorri C, Doherty C (2013) The movements made
by performers in a skilled quartet: a distinctive pattern, and the function that it serves. Front
Psychol 4:841
Godøy RI, Leman M (eds) (2010) Musical gestures: sound, movement, and meaning. Routledge,
New York
Goebl W, Palmer C (2009) Synchronization of timing and motion among performing musicians.
Music Percept 26:427–438
Gritten A, King E (eds) (2006) Music and gesture: new perspectives on theory and contemporary
practice. Ashgate Publishing, Aldershot
Hatten R (2004) Interpreting musical gestures, topics, and tropes: Mozart, Beethoven, Schubert.
Indiana University Press, Bloomington
Hodges NJ, Edwards C, Luttin S, Bowcock A (2011) Learning from the experts: gaining insights
into best practice during the acquisition of three novel motor skills. Res Q Exerc Sport
82:178–187
Jensenius AR, Wanderley MM, Godøy RI, Leman M (2010) Musical gesture: concepts and methods
in research. In: Godøy RI, Leman M (eds) Musical gestures: sound, movement and meaning.
Routledge, New York, pp 12–35
Kendall RA, Carterette EC (1990) The communication of musical expression. Music Percept 8
(2):129–163
Kenny DT (2011) Negative emotions in music making: performance anxiety. In: Juslin P, Sloboda J
(eds) Handbook of music and emotion: theory, research, applications. Oxford University Press,
Oxford
Lehmann M, Kopiez R (2013) The influence of on-stage behavior on the subjective evaluation of
rock guitar performances. Music Sci 17(4):472–494
Leman M, Godøy RI (2010) Why study musical gestures? In: Godøy RI, Leman M (eds) Musical
gestures: sound, movement, and meaning. Routledge, New York
Mesagno C, Mornell A, Quinn AL (2016) Choking under pressure in sport and music: exploring the
benefits of theory transfer across domains. In: Mornell A (ed) Art in motion III – performing
under pressure. Peter Lang, Frankfurt, pp 23–58
Nakra TM (2000) Searching for meaning in gestural data. interpretive feature extraction and signal
processing for affective and expressive content. In: Wanderley MM, Battier M (eds) Trends in
gestural control of music. IRCAM – Centre Pompidou, Paris, pp 269–299
Nusseck M, Wanderley MM (2009) Music and motion: how music-related ancillary body move-
ments contribute to the experience of music. Music Percept 26(4):335–353
Nusseck, M, Wanderley, MM, Schoonderwaldt, E (2007) Validating kinematic displays for the
perception of musical performance. Proceedings of the ENACTIVE’07 conference, Grenoble,
381–384
Palmer C (2013) Music performance: movement and coordination. In: Deutsch D (ed) The psy-
chology of music, 3rd edn. Elsevier Press, Amsterdam, pp 405–422
Palmer C (2015) Listening, imagining, performing: melody as a life cycle of musical thought.
Music Percept 33:3–11
Platz F, Kopiez R (2012) When the eye listens: a meta-analysis of how audio-visual presentation
enhances the appreciation of music performance. Music Percept 30(1):71–83
Rodger M, Craig C, O’Modhrain S (2012) Expertise is perceived from both sound and body
movement in musical performance. Hum Mov Sci 31:1137–1150
Schutz M, Lipscomb S (2007) Hearing gestures, seeing music: vision influences perceived tone
duration. Perception 36:888–897
Simones L, Rodger M, Schroeder F (2015) Seeing how it sounds: observation, imitation and
improved learning in piano playing. In: 3rd European symposium on multimodal communica-
tion, Dublin
Spahn C (2015) Treatment and prevention of music performance anxiety. In: Altenmüller E,
Finger S, Boller F (eds) Progress in brain research, vol 217. Elsevier, Amsterdam, pp 129–140
Teixeira ECF, Loureiro MA, Wanderley MM, Yehia CH (2014) Motion analysis of clarinet
performers. J New Music Res 44(2):97–111
Tsay CJ (2013) Sight over sound in the judgment of music performance. Proc Natl Acad Sci 110
(36):14580–14585
Vines BW, Dalca I, Wanderley MM (2006a) Variation in expressive physical gestures of clarinetists.
In: Proceedings of ICMPC2006, Bologna, pp 524–525
Vines BW, Krumhansl CL, Wanderley MM, Levitin DJ (2006b) Cross-modal interactions in the
perception of musical performance. Cognition 101(1):80–113
Vines B, Dalca I, Wanderley MM (n.d.) Variation in expressive physical gestures of clarinetists.
Unpublished report, McGill University. http://idmil.org/publications#unpublished_reports. Last
Accessed on March 1, 2017 (2006)
Wagner V, Altfeld S, Kallus KW, Kellmann M (2016) The impact of Stress and Recovery on
performance. In: Mornell A (ed) Art in motion III – performing under pressure. Peter Lang,
Frankfurt, pp 59–72
Wanderley MM (2001) Performer-instrument interaction: applications to gestural control of sound
synthesis. PhD thesis Ircam
Wanderley MM (2002) Quantitative analysis of non-obvious performer gestures. In: Wachsmuth I,
Sowa T (eds) Gesture and sign language in human-computer interaction. Springer Verlag,
Berlin, pp 241–253
Wanderley MM, Vines BW (2006) Origins and functions of clarinettists’ ancillary gestures. In:
Gritten A, King E (eds) Music and gesture: new perspectives on theory and contemporary
practice. Ashgate Publishing, Aldershot, pp 165–191
Wanderley MM, Vines B, Middleton N, McKay C, Hatch W (2005) The musical significance of
clarinetists’ ancillary gestures: an exploration of the field. J New Music Res 34(1):97–113
Wöllner C (2008) Which part of the conductor’s body conveys most expressive information? A
spatial occlusion approach. Music Sci 12(2):249–272
Wulf G (2013) Attentional focus and motor learning: a review of 15 years. Int Rev Sport Exerc
Psychol 6(1):77–104
Yoshie M, Shigemasu K, Kudo K, Ohtsuki T (2009) Effects os state anxiety on music performance:
relationship between the revised competitive state anxiety inventory-2 subscales and piano
performance. Music Sci 13(1):55–84
Investigating Aspects of Movement
in Violin Performance
Gongbing Shan, Peter Visentin, Manfred Nusseck, and

Claudia Spahn
Abstract
The violin plays a central role in Western classical music traditions. Performing
on the violin requires a complex interaction between motor control and auditory
perception. The development of such skills requires extensive training over a
period of years. With modern movement analysis techniques, characteristic motor
behaviors associated with violin performance can be revealed. This chapter
overviews research findings regarding the kinematics and kinetics of posture,
bowing, fingering, and limb movements of violin players. These provide insights
into the mechanics of playing the violin. More importantly, they can help to
improve skill acquisition, control efficiency, and, concomitantly, musical effec-
tiveness through their discussions of motor control economy, compensatory
strategies, anthropometrics, and ergonomics.
Keywords
Violin performance • Gross motor control • Fine motor control • Biomechanics •
Movement coordination • Compensatory movements
G. Shan (*)
Department of Kinesiology, Faculty of Arts & Science, University of Lethbridge, Lethbridge,
Alberta, Canada
e-mail: g.shan@uleth.ca
P. Visentin
Department of Music, Faculty of Fine Arts, University of Lethbridge, Lethbridge, Canada
e-mail: visentin@uleth.ca
M. Nusseck •
C. Spahn
Freiburg Institute for Musicians’ Medicine, University of Music and University Clinic Freiburg,
Freiburg, Germany
e-mail: Manfred.nusseck@uniklinik-freiburg.de; Claudia.spahn@uniklinik-freiburg.de

DOI 10.1007/978-3-319-30808-1_108-1
2 G. Shan et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Posture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Bowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Left-Hand Fingering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Movement Coordination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Introduction
To an audience, an artful violin performance can appear to be simple, but it is not.

High degrees of motor control and organization are needed to create even the most
basic auditory and visual stimuli that communicate this illusion. Since the violin is
an acoustic string instrument, the design of which has barely changed over the last
300 years; kinematic control is directly related to the physics of its sound production.
Thus, the auditory product during performance is governed by the interaction of the
performer with the tools of performance, and variation of motor control among
performers should be kinematically narrow. It is clearly the case that stable/repeat-
able motor control is a necessary prerequisite for high-level violin playing, but
researchers must understand that it is not by itself sufficient for an artistic perfor-
mance. Performers manipulate motor control to evoke both auditory and visual
discourse during performance. Small variations of control are used in performance
to produce auditory effects; these amount to colors available on the artistic palette of
a performer. Visual cueing is also used to communicate to an audience. By building
upon kinematic stability learned through long-term practice, conscious manipulation
of gross and fine motor control can thus become hallmarks of an individual artist’s
“style.”
In terms of human movement research, gross and fine motor controls are typically
examined severally. This practice finds legitimacy in the fact that, for most physical
activities, motor control patterns exhibit characteristics wherein either gross or fine
motor control appears to dominate the behavior. For example, many daily-living
activities (e.g., walking, running, and lifting) mainly rely upon large muscles (gross
motor control), where smaller muscles function in significant stabilizing roles. Other
activities, like writing and typing, or using devices such as smartphones, rely mainly
on smaller muscle group coordination (fine motor control). In this case, gross motor
control functions to provide a stabilizing foundation – often, it is nearly arrested to
do so.
In the case of violin performance, there is visibly codependent interaction of gross
and fine motor control – skill development that requires extensive training. Consid-
ering the importance of motor learning for violin playing, it is surprising how few
violin studies examine performance-related gross or fine motor control. Even fewer
Investigating Aspects of Movement in Violin Performance 3
studies examine interaction between gross and fine motor coordination. Perhaps one
explanation for this is that repeatability – a cornerstone of scientific methodology – is
not desirable for its own sake in music. Violin performance is a temporal art which
requires a performer to simultaneously “respond to the moment” while forward
planning for upcoming musical events. Since it is undesirable for performers to
merely act as mechanical executors of instructions on a printed page (Von
Hasselbach 2012), violin performance should be thought of as a process, not as a
kinematic reconstruction of instructions on a printed page of music. This reality
creates significant, but not insurmountable, challenges for researchers who apply
scientific research methodologies in the study of violin performance.
A particular confounding factor for movement analysis methods is that the
physical dimensions of a performer’s body (anthropometry) influence motor control.
In great part, the personal “style” of a performer may derive from his or her
development of a variety of motor behaviors that are compensatory in nature.
These accommodate body size and shape as well as produce outcomes that a
performer would perceive as falling within acceptable ranges of musico-cultural/
aesthetic “expectations.”
This chapter outlines the current state of knowledge related to motor control
during violin performance. The discussion is organized according to the following
topics: posture, bow control, left-hand (LH) fingering, and left and right limb
movement coordination. These will be considered in light of gross and fine motor
control, combinations of these control types, and compensatory control mechanisms,
providing food for thought for individuals interested in undertaking quantitative
research investigating violin performance.
State of the Art
Posture
Postural concerns pertaining to violin playing are evident in pedagogical and

theoretical treatises as early as the seventeenth century. The first decades of the
nineteenth century marks the beginnings of music pedagogy as a science. Most
modern styles of playing have a lineage that descends from pedagogical materials
created during the formation of the French Conservatoire in this era (Baillot 1834).
Violin method books of many subsequent authors emphasize the development of
technique as a means to achieve the highest degrees of artistry, with a focus on
technical training. Regarding kinematics, kinetics, and physiology, the vast majority
of this instructional documentation may be summarized as rudimentary (Fig. 1);
most are experience/practitioner based with superficial descriptions of postures or
mechanistic movements. However, the high frequency with which postural control is
addressed among the nineteenth- and twentieth-century sources can be taken as an
indicator that it is import to violin performance.
From a kinematic perspective, postural control is a process of corporeal stabili-
zation related, in greatest part, to gross motor control mechanisms. Such control is
4 G. Shan et al.
Fig. 1 The position of the right arm and wrist is held closer to the body when the violinist plays
sitting down. The neck of the violin is then held slightly lower. Position of the hand, wrist, and
fourth finger in extensions (top right). Forced and improper position of the hand, wrist, and fourth
finger in extensions (bottom right) (Adapted from Baillot (1834), Shan and Visentin (2010))
needed during violin performance because the precision of fine motor control
governing bowing and fingering is contingent on predictability of arm and body
movement. On a fundamental level, changes in posture affect body kinematics and
kinetics. For violinists, it is common to sometimes play while standing and, at other
times, while sitting. Solo playing normally occurs in a standing position and sitting is
standard in ensemble and orchestral playing. In an orchestral environment, two
violinists will share one music stand. In a recent article (Spahn et al. 2014),
balance/weight distribution and kinematics were studied in relation to standing and
sitting postures during violin performance. For the sitting position, kinematics and
balance were quantified for two different orientations – slightly to the left and
slightly to the right of a music stand (emulating an orchestral environment). In this
study, results revealed mean weight distribution in the standing position to be
symmetrical, i.e., 50.0% on each foot. In the sitting positions, weight distribution,
measured under the sitting bones, is both asymmetrical and significantly different in
the right and left sitting orientations. Essentially, when seated to the right of a stand, a
player’s weight is distributed more heavily on the left and vice versa.
In the same article, details of posture were elaborated using multiple joint angles:
head and neck, upper limb, trunk, and hip. Comparing standing to sitting, significant
differences were found in the following: (1) the angle between right shoulder, neck,
and left shoulder; (2) mean angle of the thoracic kyphosis; (3) the lumbar lordosis;
and (4) elbow angle and range of motion for the right (bowing) arm. Comparing left
and right orientations in the seated position, significant differences were found for
the head angle, neck lordosis angle, thoracic kyphosis, and elbow angle. Although
the purpose of Spahn’s study was to discuss its findings in terms of postural effects
that might influence the development of musculoskeletal disorders/disease among
violinists, its significance goes far beyond this purpose. The article supplies a
detailed examination of postural variables in violin performance, and it supplies
foundational concepts for future studies.
Bowing
An underlying theory for a mechanical basis to study optimization of violin bowing

may be traced to the work of Nelson (Nelson 1983). In his article, movements
needed during bowing were discussed in terms of optimization with respect to
various bowing objectives. Using a purely mechanical approach, motor control
variables were described in terms of their “costs” with respect to achieving these
objectives. The concept of performance trade-offs between competing objectives
was used in order to identify and quantify physical parameters and constraints related
to economy of bowing movement. The article presumed a performative objective of
“cost” minimization. In this way, Nelson imposes a value judgment, perhaps with
considerable legitimacy, on what constitutes optimization in violin bowing. The
objective of the article was to investigate various performance constraints which
might influence trade-offs among competing objectives. The article is particularly
relevant because (1) it provides a theory for the mechanical basis of variables
associated with bowing, (2) it opens the door to discussions of “trade-offs” that
might occur as a result of optimizing competing variables, and (3) it discusses them
as providing physical limitations with regard to more complex objectives.
Although this study was theoretical with no actual biomechanical data to validate
its results, it laid a strong foundation for future biomechanical studies related to
bowing. A missing piece in Nelson’s work is consideration of auditory quality.
Indeed, he acknowledges tone production as an important variable governing the
performative act, but the influence of tone can only be tested in a performance
situation, not in a theoretical simulation. From a violinist’s standpoint, and indeed for
anyone who has attended the violin performance of a 6-year-old, tone quality is an
overriding constraint.
In the research literature, there are relatively few examples of quantitative
research related to bowing in a performance situation. Quantitative studies
employing advance movement analysis methods are a relatively recent development
in music research (Shan and Visentin 2003; Visentin and Shan 2003; Shan et al.
2007; Visentin et al. 2008). In the identified studies, researchers set out to determine
whether or not it was possible to accurately characterize biomechanics related to
violin bowing – both kinematics and kinetics – for smooth and bounced bowing
techniques. The test protocols involved bowing at various bow stroke frequencies
6 G. Shan et al.
G-String E-String G-String

Joint angles (Degrees)
120
100 RE (Flex/Ext)
80
60
40
20
0
–20 0 5 10 15 20 25 30
120 RE (Flex/Ext)
100
80
60
40
20
0
–20 0 4 8 12 16 20 24 28 32
Time [s]
Fig. 2 Right elbow and shoulder flexion/extension. Ranges of motions (ROMs) are measured by
peak to trough differentials (Visentin et al. 2008)
(tempi) and on all four strings of the violin. Professional-level subject groups were
used. Collectively, these studies found that it was possible to make some general-
izations about physical parameters related to bowing. Kinematic analyses revealed
stable motor control patterns (i.e., ranges of motion of single joints and dynamic
interactions among joints), and kinetic calculations quantified physical effort. For
both the identified bowing techniques, shoulder and elbow movements, those joints
that dominated gross motor control, showed generalizable patterns when posture,
positions, and time-based dynamic interactions among joints were considered. Fine
motor control, such as that found in the wrists, was highly variable.
For the smooth bowing condition, absolute changes in amplitudes of the bowing
arm joints were dependent on the string played (Fig. 2). Control pattern changes of
the shoulder and elbow joints appeared to be complimentary – the range of motion of
one gets bigger as the other gets smaller. From G-string to E-string, the range of
motion in shoulder flexion increased, while overall shoulder flexion/extension angle
decreased. The elbow acted in exactly the opposite manner. Complementary motor
control was also apparent in data for the bounced bowing condition (Fig. 3).
Through biomechanical modeling, the studies of Shan and Visentin also identi-
fied tempo-dependent phases of motor control (Shan et al. 2007; Visentin et al. 2008)
by examining the relative movement of the bow to the strings of violin. In this way,
motor control is linked to the acoustic necessities of producing sound on the
instrument. Three discrete control pattern phases were found for both smooth and
bounced bowing. Forward dynamic analysis revealed each phase to have distinct
combination of kinematic factors (Figs. 4 and 5). The first phase was marked by
increased bow acceleration and speed (indicators of exertion), while travel distance
Fig. 3 Joint angles through the cycle of the scale for the right shoulder (RS) and right elbow
(RE) during a typical trial set (Shan et al. 2007)
of the bow decreased. The second phase was characterized by constant bow accel-
eration, revealing workload to remain steady, while distance and speed decreased
(showing efficiency gains). And the third, revealing an approach to physical limits of
the player, showed bow acceleration, speed, and distance to all gradually approach
their respective plateaus. These results amount to “trade-offs,” showing a
8 G. Shan et al.
Fig. 4 Kinematic factors showing three discrete phases of motor control defined by two critical
points (vertical dotted lines). From left to right: increasing physical effort, optimization, and
approaching physiological limits (Visentin and Shan 2003)
Fig. 5 Bowing kinematics as a function of frequency: (a) distance, (b) velocity, and (c) acceler-
ation. “I-beams” indicate standard deviations at the respective bow stroke frequencies tested in this
study (Shan et al. 2007)
10 G. Shan et al.
compensatory strategy (whether consciously invoked or a subliminal result from

long-term training) that mitigates effort “costs” with respect to increases in tempo.
Whereas Nelson’s study discussed performance trade-offs between competing
objectives in order to identify and quantify physical parameters and constraints, the
research of Shan and Visentin does the opposite – it quantified physical parameters
using movement analysis as a means to determine performance trade-offs between
competing objectives. Notwithstanding their polar approaches, results of Shan and
Visentin appear to validate Nelson’s purely theoretical model. Moreover, their
performance-based testing conditions incorporate consideration of the one important
parameter that Nelson identified but could not measure – tone.
Left-Hand Fingering
Certainly, violin playing must be counted among the most complex of tasks for
motor control. To date, motion capture studies quantifying fine motor movements
needed to play the violin have concentrated on the left hand. The reason for this is
straightforward. Visible movement of the fingers on the bow (right hand) is small;
they remain in a relatively stable position, and fine control is in greatest part directed
toward the manipulation of forces that the fingers apply to the bow. The LH is
another matter – movement is visible in all digits. The four fingers of the LH are used
to press the strings onto a fingerboard, changing string lengths to give pitches of
varied frequencies, while the thumb is used as an opposable digit. Accuracy of pitch
(tuning) is a vital element for successful violin performance. It depends on submil-
limeter positional accuracies of the fingers on the fingerboard. Adding to complexity
is the fact that the physics governing pitch generation on a string requires logarithmic
positional control patterns, i.e., control must be both nonlinear, and when the same
pitches are played on different strings, a different nonlinear finger positioning is
needed.
Two quantitative studies of LH fingering will be discussed below: one employing
2D motion capture with the hand remaining in a single position relative to the
fingerboard (Baader et al. 2005) and the other using 3D capture for a more complex
performance task (Visentin et al. 2015). While both provide valuable perspective on
finger control during violin performance, these studies employ very different bio-
mechanical models to underpin their analyses. It is important to acknowledge that all
biomechanical modeling represents a simplification of the reality it models.
Depending on the research question being asked, differing degrees of modeling
complexity can provide valuable, contextual perspective.
In the study of Baader et al. (2005), LH finger movement during violin playing
was measured using a relatively simple model with five points on the hand, one on
the distal part of each finger and one on the thumb. This permitted determination of
finger position and movement relative to the fingerboard of the violin. When a finger
depresses a string against the fingerboard to change pitch, the starting and stopping
of finger movements provide information about timing. Spatial data provides
distances, velocity, and accelerations of the fingers. The researchers used these data
to reveal several generalizable motor control patterns in the LH digits:
• Intra-subject finger displacement profiles were consistent. This might be expected

among high-level players. However, the subject group in this study included both
professionals and amateurs with varying levels of proficiency and currency in
terms of playing the violin. From this, one might infer that consistency of finger
movement is vital at all stages of the development of playing proficiency.
• Speed of finger movement as it approached the finger board (to change notes/
pitches) was generally fast, but there were consistent instances when finger
movements toward the string were more gradual and/or slower. These occurred
during specific note sequences and were present both intra- and intersubject.
• Serial and parallel finger actions were observed. Serial (note-to-note) finger action
represents baseline motor control mechanisms of this high-skill activity, while
parallel control may indicate anticipatory actions needed to effectively accom-
plish the assigned task.
In another study (Visentin et al. 2015), a more complex LH model was employed
to study “shifting” – a violinistic technique where proximal/distal LH positioning
with respect to the fingerboard is deliberately changed. In violin performance,
shifting is needed to generate pitches for all but the simplest of musical contexts.
In addition to having practical utility of extending the range of pitches available to
the performer, shifting has artistic merits. Because each of the four strings of the
violin has a different tone-color profile, it can be employed to artfully manage the
musical intentions. Clearly, the complexity of shifting as a task involves both gross
and fine motor control. To investigate fine motor activity, the researchers designed a
hand model that allowed examination of fingertips, finger joints, and wrist move-
ments. The model employed 25 markers to divide the hand into 16 segments
connected by 15 joints (Fig. 6). In this model, finger joints were considered to be
simple hinge joints, and the thumb was modeled for three degrees of freedom
(flexion/extension, ab/adduction, and rotation). Finally, in this study, the hand
model was attached to a full-body model in order to link fine control of the fingers
with gross control of the wrist, elbow, and shoulder.
The study measured 540 shifting samples from six professional-level subjects and
at three different tempi (playing speeds). Results showed some elements of the skill
to be individualized in surprising ways, while others could be explainable by
anthropometry, ergonomics, or musical entrainment. Intra- and intersubject timing
of the assigned task was highly accurate. Timing errors ranged from 0.027 to 0.068 s.
Intra-subject duration of the shifting process was also highly consistent even at
different tempi: variation across the tempi tested was less than 0.025 s for all but one
subject. The researchers hypothesized that the duration of shifts would be both
influenced by tempo and vary due to the performance context. Regarding shifts at
different tempi, it seems intuitively reasonable that, to go faster, one has to go faster,
i.e., duration of shifts should become shorter at faster tempi. As for complexity of
shifting, every movement element has the potential to influence timing, e.g., some of
12 G. Shan et al.
Fig. 6 Marker set for the

hand in an integrated full-
body model (Shan et al. 2012)
the shifts in the current study required synchronous changes of strings and a
concomitant height adjustment in the position of the right arm. Greater variability
in shifting times based on complexity of context was expected. With the exception of
one subject, neither of these hypotheses proved to be true. This seems to suggest that
timing elements of motor control were stabilized for skill execution. Thus, the
shifting process was resistant to tempo- and complexity-based adaptation. This is
an unexpected finding that could have real-world applications in violin pedagogy.
Movement Coordination
Asymmetric motor control is a notable feature of playing the violin. Complex and
precise management between the left and right limbs as well as between gross and
fine motor control in each limb is required. All of this activity has a central
motivation, i.e., controlling the interaction of the bow with the violin to produce
desired auditory effects. Each of the studies discussed below evaluates different
motor complexities/coordination in an effort to deepen understanding of violin
performance (Baader et al. 2005; Schoonderwaldt 2009; Shan et al. 2012; Visentin
et al. 2015). All of them find complex relationships, codependencies, and compen-
satory behaviors to exist. Notably, all of the identified studies in some way consider
motor control in terms of auditory phenomena. This is an important research ethic for
violin studies because auditory feedback ultimately drives motor coordination
choices for high-level players during performance.
Fundamental examinations of coordination complexity are found in Shan et al.
(2012) and Baader et al. (2005). Shan et al. (2012) showed joint angles and
orientations to change according to finger placement on the fingerboard on the
violin. In a consideration of just two pitches (both on the same string, but one octave
apart), they found significant differences in average joint angles in the left wrist and
digits (fingers and thumb); ranges of motion showed significant differences for all
Fig. 7 A visual rendering of

complex finger orientations
(3D data) for four different
pitches on the violin
(Modified from Shan et al.
2012)
LH fingers. As well, comparison of the same protocol on different strings revealed

complex left-limb joint control relationships (Fig. 7), while right limb (bowing)
control was stable. The authors postulated that this control stability was developed
through years of training for the purpose of eliminating sympathetic “cross talk”
between left and right limb functions.
In Baader et al. (2005), synchronization between the right bowing arm and LH
fingering was observed by comparing timing of LH fingering (finger impact on the
string and finger lifts) in relation to bow direction reversals at two different tempi. On
the violin, fingering control achieves pitch changes either by using a finger to depress
the string onto the fingerboard (shortening the string) or by lifting a finger that was
already depressing the string (lengthening the string). For the condition of depress-
ing a finger on the string, synchronization intervals ranged from ~0.000 to 0.075 s.
For the lifting condition, mean values varied between 0.030 and 0.050 s. Mean
values were rather variable among subjects, possibly due to subtle individualization
of fingering strategies.
Coordination among motor control elements, whether related to postural control
or to the tools of performance (violin and bow), has been linked to compensatory
behaviors or physical trade-offs in several articles (Schoonderwaldt 2009; Shan et al.
2012; Visentin et al. 2015). All of the identified articles used experienced/expert
performers and uncovered motor control optimization related to the musical condi-
tions dictated by each study protocol. This suggests that resultant compensatory/
trade-off control can be linked to long-term training.
In Schoonderwaldt (2009), parameters related to bow movement were used to
deepen insight into the interaction between the player and the instrument. The study
examined bow position (the linear distance between the point of bow-hair contact
14 G. Shan et al.
with the violin string and the proximal end of the bow), bow speed, bow force (which
measured tension changes in the bow hair as it contacted the string), and the distance
between the violin bridge and the bow contact point on the string (bow bridge). All
of these are relevant to sound production on a violin. The following was found:
1. Production of dynamic (volume) differences involved a trade-off between bow

velocity and bow-bridge distance.
2. Differing note durations lead to higher contrasts in the observed control strate-
gies, as well as larger differences between the extreme values of the bowing
parameters.
3. For the longest notes, volume level was almost entirely determined by
bow-bridge distance, whereas for notes of shortest duration, bow velocity was
the dominating parameter.
4. Bow-bridge distance showed much less variation between volume conditions,
mainly because small values of bow-bridge distance were avoided at high bow
velocities.
5. Players adapted bowing parameters to the physical properties of the strings and
the instrument, showing the players were sensitive both to the differences in
bow-force limits across strings and to limits related to playing speed.
In summary, bowing parameters were highly interrelated. Observed

compensations/trade-offs indicated specific adaptations that were needed in order
to optimize sound quality and motor control in relation to speed and volume
changes.
Compensatory motor control observed in Shan et al. (2012) showed that, because
of performative constraints, whether anthropometric or equipment related, some
joints had to move more than others. In Figs. 8 and 9, gross and fine motor controls
are compared for a standard playing protocol beginning on two different pitches.
Significant differences were observed in the elbow and wrist (gross motor control) as
well as in multiple finger joints (fine motor control). The study linked these interre-
lated coordination patterns to ergonomic necessity.
In addition to ergonomics, anthropometry plays an important role in motor
control optimization during violin performance. Since a violin is fixed in size,
playing it requires a performer to find playing strategies adapted to his/her body
shape. In the shifting study of Visentin et al. (2015), two expert performers of very
different statures were compared. The 40 cm height difference substantially defined
motor behavior possibilities of each. At slower playing speeds, the shorter subject
initiated left arm shifting with the thumb and followed in sequence by the wrist and
elbow joints. The taller subject could execute the required shifting almost exclu-
sively using wrist movement. When faster tempi were used, the shorter subject
changed motor control patterns, initiating with the wrist and following with simul-
taneous movement of the elbow and thumb (a motor control grouping strategy to
compensate for having to play faster), while no change was observed for the taller
subject (motor control efficiency seems to have already been maximized). This same
study also examined the length of time that it took for each subject to accomplish the
Fig. 8 Gross motor as evidenced by joint control of both arms. Left and right columns show data
for trials beginning on A-flat below middle C and A-flat above middle C, respectively (Revised from
Shan et al. (2012))
required shifting tasks. Remarkably, shifting times were stable intra-subject, not-
withstanding changes in tempo. However, intersubject shifting time was signifi-
cantly different due to the influence of anthropometry; the shorter subject required
nearly 50% more time to do each shift. Underlying principles in motor theory seem
to be borne out in this finding. Because accuracy decreases as joints controlling
larger segments become more active, moving larger segments takes more effort and
influences stability of fine motor control (Magill and Anderson 2007) – requiring
more time for execution.
A novel and important element in this study was the use of several professional
adjudicators to blindly evaluate audio recordings from the trials. This permitted
researchers to link observed motor control variations to the quality of the perfor-
mances. Results revealed no quality differences between the two players, suggesting
that more than one control option may be used to achieve outcomes that are musico-
culturally appropriate.
16 G. Shan et al.
Fig. 9 Select components of fine motor control in the LH. The thumb (top row) shows postural
compensation mainly evident in the first and second joints. Both index and fourth fingers show
compensatory behaviors of all of the joints (Shan et al. 2012)
Summary
During training, violinists spend countless hours exploring alternative motor control
sequences in order to synchronize and stabilize elements of performance. For the
violinist, entrainment explores strategies that coordinate both gross and fine motor
control, rely on nuanced audial and neurosensory discrimination, and require adap-
tation in a temporal endeavor. The studies above demonstrate some of the challenges
facing researchers interested in quantifying motor control associated with violin
performance. A motor control research methodology may be able to quantify some
technical elements of the activity, but, in and of itself, it will never be “music
research.” Artistry is the driving force in high-level performance, and consideration
of it must be integrated in fundamental research design before results of research can
achieve acceptance by those who are being studied – performers.
In a violin performance, musical success or failure is measurable in both milli-
meters and milliseconds, while the conceptual integrity of a performance may unfold
over hours. Each of the studies discussed above addresses violin performance with
respect to one or more characteristics of left and right limb management. Collec-
tively, they find complex relationships, codependencies, and compensatory/trade-off
behaviors. Such scientific work can play an important role in challenging the inertia
of tradition that exists in the performing arts. For researchers, it may be helpful to
remember that “technique” in performance is not the acquisition of repeatable,
idealized skills. Rather it is the application of intention to a musical process –
something that may be successfully accomplished using a variety of motor behav-
iors. An understanding of biomechanics, ergonomics, and the strategic use of motor
control can help explain the interaction of the artist with the tools of performance in
the context of a desired musical outcome. Ultimately, by providing objective ground
upon which to discuss elements of performance, it may help demystify musical
artistry during violin performance.
Cross-References
• ▶ Observing and revealing the hidden structure of the human form in motion
throughout the centuries
• ▶ 3D Kinematics of Human Motion
• ▶ Optimal Control Strategies for Human Movement
• ▶ Movements, Timing and Precision of Drummers
• ▶ Movement and Touch in Piano Performance
• ▶ Body Movements in Music Performances - on the Example of Clarinet Players
• ▶ Brain and Behavioural Bases of Guitar Learning in Novices
18 G. Shan et al.
References
Baader AP, Kazennikov O, Wiesendanger M (2005) Coordination of bowing and fingering in violin
playing. Cogn Brain Res 23:436–443
Baillot P 1834 L’art du violon. Nouvelle méthode dédiée à ses élèves. par P. Baillot, membre de la
légion d’honneur, de la musique particulière du Roi et professeur au conservatoire de musique.
Paris, Dépôt Central de la musique
Magill RA, Anderson D (2007) Motor learning and control: concepts and applications. McGraw-
Hill, New York
Nelson WL (1983) Physical principles for economies of skilled movements. Biol Cybern
46:135–147
Schoonderwaldt E (2009) The player and the bowed string: Coordination of bowing parameters in
violin and viola performance. J Acoust Soc Am 126:2709–2720
Shan G, Visentin P (2003) A quantitative three-dimensional analysis of arm kinematics in violin
performance. Med Probl Perform Art 18:3–10
Shan G, Visentin P (2010) Arts biomechanics: an infant science: its challenges and future. Nova
Science Publishers, New York
Shan G, Visentin P, Tardif G, Robinson A, Evans J, Crowe S (2012) Quantifying compensation
strategies between gross and fine motor control in violin performance. Arts BioMech 1:143
Shan G, Visentin P, Wooldridge L, Wang C, Connolly D (2007) A frequency-based characterization
of spiccato bowing in violin performance. Percept Mot Skills 105:1027–1051
Spahn C, Wasmer C, Eickhoff F, Nusseck M (2014) Comparing violinists’ body movements while
standing, sitting, and in sitting orientations to the right or left of a music stand. Med Probl
Perform Art 29:86
Visentin P, Li S, Tardif G, Shan G (2015) Unraveling mysteries of personal performance style;
biomechanics of left-hand position changes (shifting) in violin performance. Peer J 3:e1299
Visentin P, Shan G (2003) The kinetic characteristics of the bow arm during violin performance: an
examination of internal loads as a function of tempo. Med Probl Perform Art 18:91–98
Visentin P, Shan G, Wasiak EB (2008) Informing music teaching and learning using movement
analysis technology. Int J Music Educ 26:73–87
Von Hasselbach J 2012 Für “lebendige’und” beziehungsfähige’Töne im Streichinstrumentspiel.
Eine Reaktion auf Peter Röbkes Einleitung zu Das Musizieren und die Gefühle. Musik und
Gefühl. Interdisziplinäre Annäherungen in musikpädagogischer Perspektive. Georg Olms
Verlag, Hildesheim (Olms)/Germany
Movement and Touch in Piano Performance
Werner Goebl
Abstract
Pianists achieve extreme levels of virtuosity on their instrument, requiring a
combination of talent and decade-long continuous and deliberate practice, train-
ing, and experience. As with all musical behaviors, body movements in piano
performance are goal directed, aiming at producing intended sounds with utmost
precision and accuracy in expressive parameters such as timing, dynamics,
timbre, and articulation. Body movements in piano performance may also serve
communicative purposes such as to express emotional states or to coordinate with
co-performers. Pianists control the timing and velocities of the individual piano
hammers by varying the forces applied to the piano key surfaces, as well as to the
three pedals through their feet. The key forces are accomplished by coordinating
the kinematic chain from their shoulders to the fingertips aligned with feet
movements to manipulate the pedals. As kinematic properties such as finger
velocity covary with performance parameters (tempo, dynamics, etc.), pianists
have to stabilize several parameters of movement kinematics and musical expres-
sion simultaneously. The intrinsic way the fingers arrive at the piano key surface,
referred to as piano touch (i.e., pressing versus striking a piano key), yields
different tactile and other sensory percepts to the pianists themselves and the
audiences alike, making this parameter an important one in accomplished piano
performance.
Keywords
Piano performance • Motion capture • Piano technique • Piano touch • Movement
efficiency
W. Goebl (*)
Department of Music Acoustics – Wiener Klangstil (IWK), University of Music and Performing
Arts Vienna, Vienna, Austria
e-mail: goebl@mdw.ac.at

DOI 10.1007/978-3-319-30808-1_109-1
2 W. Goebl
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Purpose of Movement in Piano Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Pianistic and Scientific Views on Piano Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
The Scope of Piano Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Measuring Movement Effects on the Piano Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Measuring the Pianist’s Body Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Biomechanical and Pianistic Attributes of Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Piano Touch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Effects of Intended Tempo and Dynamics on Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Role of Expertise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Movement Disorders in Piano Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Cross-Referencs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Introduction
Human movement in skilled piano performance is a remarkable example for the

extreme capabilities that our bodies are able to achieve: professional pianists are able
to produce highly regular sequences of up to about 16 tones per second such as in a
Chopin Etude (Goebl and Palmer 2013), they are able to precisely control hammer
velocities of individual keystrokes between about 0.2 and 8 m/s through variations of
force on the key surface (Goebl et al. 2005), they easily achieve about 30 strictly
increasing dynamics on one repeated tone (Tro 2000), their fingertips experience
impact peaks of more than 10 G when arriving at the piano key surface (Goebl and
Palmer 2008) – more than a fighter jet pilot would survive – and during a typical
piano recital, they perform easily over 50,000 tones (Flossmann et al. 2010), each
minutely timed and dynamically shaped to bring the musical image of the artist to
perfection.
Human movement and music are closely intertwined, not only because music
necessarily requires human body motion to be generated but also because music
induces the perception of movement in many different ways, such as with the
impression of rising or falling of (the pitch of) a melody or a fast beat suggesting a
fast musical movement able to drive humans to move and dance along with it
(Phillips-Silver 2009). This chapter focusses on the first, the highly skilled human
movement that is required and precisely executed for and during playing the piano.
Purpose of Movement in Piano Performance
Human movement in piano performance is primarily directed to produce sound

imagined by the performing musician. Rather than being the goal of the artist (such
as body movements in ballet dancing [“▶ Functional Movement Analysis in
Dance”]), the movements serve another primary purpose – that is – the creation of
Movement and Touch in Piano Performance 3
sounds to be perceived by the audience (also referred to as “sound-producing

gestures,” Dahl et al. 2010, “▶ Body Movements in Music Performances – On the
Example of Clarinet Players”). In this light, by studying human movement during
music performance, we measure an important behavioral dimension that is located in
the middle of the artistic chain from the musical idea, the musical work to its
realization in actual sound (Bishop and Goebl 2016). With increasing skill, pianists
more successfully connect the auditory image of the sound to be played with the
internal motor command that is required to produce it, so that they can just think of
the sound and the required movement arises automatically (audio-motor integration,
see e.g., Stewart et al. 2013). When practicing specific novel movements, pianist
direct their attention to the particular movements (“internal focus”) but should shift
their attentional focus to the sound afterwards to ensure seamless performance later
on stage (“external focus,” cf. Wulf and Mornell 2008).
Beyond the goal of producing sound, human motion during piano performance
may also serve other (in part unconscious) functions, such as communicating
emotional aspects to the audience (“▶ Postural Movements of Violin Players”),
coordinating with other musicians or facilitating the own performance such as
tapping with the feet to help keeping the beat (Dahl et al. 2010, “▶ Movements,
Timing and Precision of Drummers”). Viewing and listening to a piano performance,
rather than just listening to it, has been shown to alter its perception considerably
(Behne and Wöllner 2011), to the degree that visual information alone is more
predictive of identifying the winner of a piano competition than auditory information
alone or auditory and visual information combined (Tsay 2013).
Pianistic and Scientific Views on Piano Technique
Considering human movement in piano performance is central for pianists, piano

educators, and piano students themselves. They are primarily interested in how these
movements can be trained and optimized through practice to establish an unlimited
virtuoso technique on the instrument while simultaneously minimizing the risk of
injury. Pianists require a minimum of 10,000 h or about 10 years of deliberate
practice to achieve medium levels of expert performance skill; professional pianists
exhibit even more (Krampe and Ericsson 1996). Renowned pianists and piano
educators have provided a wealth of writings describing different pedagogical and
pianistic approaches to piano technique and how the piano ought to be played,
ranging from pure verbal descriptions (Neuhaus 1973) to using illustrative pictures
or sequences of pictures to illustrate a particular movement (as in Gat 1965, for an
overview see Gerig 1974).
Pianistic accounts of how to play the piano and to develop a proper keyboard
technique started to appear in the eighteenth century (Couperin 1716; Bach 1753)
with reference to a range of keyboard instruments of that period. In the nineteenth
century, when the piano developed to its modern form in tone compass and sound,
pianism developed, as did theoretic writings about it. While one method of piano
playing in the nineteenth century, the finger school, emphasized having every action
4 W. Goebl
executed by the fingers alone, other parts of the kinematic chain, hands, wrists, and
arms were fixated and unmoving (e.g., advocated by pianists and teachers
M. Clementi, J. B. Cramer, or N. Hummel, see Gerig 1974; MacRitchie 2015).
Some proponents of the finger school even proposed mechanical practicing devices
that helped to secure the wrist or the hand during piano playing (such as Johann
Bernhard Logier’s Chiroplast, cf. Rainbow 1990). A radically opposing view
became very popular in the late nineteenth century, when terms as “arm weight”
and “relaxation” were the guiding vision of piano playing and “fixation” of any kind
was to be avoided (e.g., Matthay 1903; Breithaupt 1905). Many pianists use such a
terminology as metaphor for learning pianists to imagine the “correct” movements to
be trained and executed on the piano. This terminology is often in conflict with what
scientists actually find when analyzing piano technique experimentally. There is
certainly no kind of piano technique that eliminates the muscular fixation of joints,
be it only for short durations during a keystroke (Ortmann 1929), despite the
statements of some piano schools (e.g., Matthay 1903).
On the other hand, scientists have been studying piano technique systematically
already for almost a century. First systematic approaches to quantitatively monitor
arm and finger movements during piano performance were provided by the piano
professor Otto Ortmann (1929) in Baltimore and the scientists Nikolai Bernstein and
Tatiana Popova in Moscow (see Bernstein and Popova 1929 for the German
publication or the edited English translation by Kay et al. 2003 of the same
experiment published in Russian), both challenged by several extreme statements
of the arm weight school such as by Matthay (1903) or Breithaupt (1905).
Bernstein and Popova (1929) measured and analyzed repeated octave strikes of
professional pianists with an optical 2D motion capture system able to record up to
600 fps (called the kymocyclographic method) and manipulated tempo and dynam-
ics as independent variables. They concluded that dynamics only changed the
quantity of the arm movement while tempo changed its entire construction. As
they started out from an artistic perspective, they implied that demonstrating fast
movement at a slow tempo as well as practicing fast movements at a slow pace
would therefore not make sense (Kay et al. 2003). They also conclude that arm
weight is physiologically only relevant at slow tempi.
Scientific investigation of piano technique has continued ever since, and record-
ing and analysis methods developed considerably, providing precise details and
insights into its basic levels of control (see overviews by Furuya and Altenmüller
2013; Metcalf et al. 2014; MacRitchie 2015). However, both approaches to
comprehending piano technique, the artistic and the scientific still have their funda-
mental difficulties in understanding each other’s scope of insight (MacRitchie 2015).
This chapter attempts to reconcile some aspects of artistic and scientific approaches.
The Scope of Piano Technique
The pianists use the piano keyboard and the three pedals as interfaces to generate
their intended sounds on a piano (Goebl et al. 2005). By controlling the variations of
force applied by their fingertips to the piano key surfaces, they vary the timing and
the speed with which the hammers strike the strings. (There are certainly many other
ways of producing sound on the piano keyboard, such as playing glissandi with the
nail-side of the finger or playing note clusters with the fist or the entire lower arm, not
to speak from the many possibilities to play directly on the strings or other parts
inside the piano.) All possible timbral and dynamic variations on the piano are
controlled by the exact timing and the speed of the (usually) 88 piano hammers in
coordination with the dampers that regulate the duration of the sound by attenuating
the string vibrations when released. A damper is either held up by a pressed piano
key, the sostenuto (middle) pedal that extends the lifting of selected dampers, or –
most commonly – by the right pedal that lifts all of them at once, creating a rich
vibrant sound. This control space is only amended by the una corda (left) pedal that
shifts the piano action sideways to make the hammers strike only a subset of the one-
to three-stringed courses, causing a variation in timbre and a prolonged acoustical
decay (Fletcher and Rossing 1998).
The piano action interrupts the mechanical connection of the key and the hammer
just before the hammer hits the strings (escapement) to allow the hammer to hit the
strings and rebound from them freely (Goebl et al. 2014a). The pianist loses control
over the hammer at this point of escapement, which is adjusted by piano technicians
as close to the piano strings as possible. The final hammer velocity is thus the only
physical parameter controlling the intensity and the sound of an isolated piano tone,
independent of the intrinsic acceleration pattern of the key. The “single variable
hypothesis” states that variations of what pianists call the “piano touch” are irrele-
vant for the sounding outcome; only variations in hammer speed are relevant for
expressive piano performance (Bryan 1913; White 1930). This created a debate
between pianists and scientists that spanned over a century (see discussion below),
finally resolved toward the pianists’ side (Goebl et al. 2014a; MacRitchie 2015).
The effects of pianists’ movements may be reduced to variations of force applied
to the piano key surfaces in order to create variations in hammer velocities (White
1930). Pianists make use of the entire kinematic chain of their arms, hands, and
fingers that offer an unlimited number of possible movement combinations for a
given task, usually referred to as the degree-of-freedom (DOF) problem (Bernstein
1967). The scope of measurement techniques for assessing human movement in
piano performance have to include kinematic (posture, motion), kinetic (dynamic
and static forces and torques), and muscular (contraction and co-contraction
of antagonist muscles) aspects, discussed in the following section.
State of the Art
Measuring the movement effects inside the piano action has been easier and there-
fore more readily available for research than studying the human movements directly
(for an overview on measurement techniques, see Goebl et al. 2008, 2014b; Metcalf
et al. 2014). Data derived from parts of the piano actions such as the hammer or key
motion (as provided by MIDI-based instruments) or extracted from audio recordings
6 W. Goebl
have been used to study aspects of musical expression in piano performance by

analyzing and modeling the performed realization itself (see e.g., Palmer 1997;
Gabrielsson 2003; Widmer and Goebl 2004, for reviews). In the following, we
report approaches that make use of data of the movement effects or movement
measurements, respectively.
Measuring Movement Effects on the Piano Action
Quantitative measurements of the key movements were attempted in the late nine-
teenth century by Binet and Courtier (1895) who used a graphical cylinder to record
the air compression variations inside a rubber tube positioned under the piano keys.
More systematic investigations were reported by Ortmann (1925) who mounted a
tuning fork onto a piano key, the vibrating end of which wrote oscillating traces into
a piece of smoked glass during a keystroke, capturing the specific velocity profiles of
the key. With this setup, Ortmann (1925) was able to analyze different playing
techniques through the velocity profile of the piano key (for more details and other
approaches, see Goebl et al. 2008).
The first large-scale performance collections were made with recording systems
that aimed to reproduce the pianist’s performances as convincingly as possible
through pneumatic reproducing systems such as systems by Welte or Ampico
(Hagmann 1984). Piano rolls have been used as source for systematic research
into the musical expression of piano playing (Hagmann 1984) as have specially
developed devices such as the Iowa piano camera in the 1930s or Shaffer’s
photocell Bechstein in the 1980s (see Goebl et al. 2008). MIDI-based instruments
such as digital pianos, hybrid pianos, or reproducing pianos usually provide the
exact information on the onset and offset timing and the dynamics in terms of a key
or hammer velocity measure for each performed note (Goebl and Bresin 2003).
These data contain information on properties of the movements that generated
those performance patterns. For example, Jabusch et al. (2004) use such MIDI
information of performances of a simple scale task as dependent variable to
distinguish healthy pianists from those who suffer from focal dystonia (see also
below).
Current reproducing pianos may also provide additional information on the key
position history, an important component as it yields information on the type of
touch used for a keystroke (i.e., the CEUS system by Bösendorfer, see Bernays
and Traube 2014; Goebl et al. 2014a). Similar data is provided by the custom-
made gesture-sensing keyboard that measures key position optically (McPherson
and Kim 2011). Other measurement devices track the location of finger-key
contact on the key surface with a capacitive foil glued on it (McPherson 2012)
or provide information on finger forces applied to the key (force transducers
underneath the keyboard, Parlitz et al. 1998 or onto the key surface, Kinoshita
et al. 2007).
Measuring the Pianist’s Body Motion
The detailed measurement of the pianist’s body during piano performance in a

natural and minimally intrusive setting (that is a pianist performing on an acoustic
grand piano with equipment that does not alter her or his performance considerably
but still offering kinematic and kinetic data on the entire movement chain from
shoulder to fingertip) is still an open challenge (for overviews, see Dahl et al. 2010;
Furuya and Altenmüller 2013; Goebl et al. 2014b; Metcalf et al. 2014; MacRitchie
2015). There is still a trade-off between data detail, equipment complexity, and
measurement intrusion. For example, a Microsoft Kinect depth camera allows
unobtrusive recording of unequipped pianists on an acoustic piano but offers only
coarse data on upper body and arm motion (e.g., Hadjakos 2012). In contrast,
marker-based optical motion capture at the level of finger movements requires
minimizing occlusion problems, by employing digital pianos that allow free camera
sight to the fingertips from the front of the pianist (that an acoustic piano prohibits
due to its construction); however, using a digital piano reduces ecological validity
(e.g., Tits et al. 2015).
The earliest motion capture approaches employed tracking devices specially
developed for the purpose of piano movement analysis. Mechanical levers attached
to the pianists’ fingers or the arm were used to monitor the movement trajectories
during a keystroke, to investigate specific kinematic properties of keystrokes exe-
cuted with different types of touch (percussive versus non-percussive Ortmann
1929). Ortmann also explored the use of lightbulbs, attached to the wrist and fingers,
which left a trace on a photo plate, to visualize a wide range of arm, hand, and finger
movements (Ortmann 1929). Using a similar basic principle, Bernstein and Popova
(1929) also used active light-emitting markers mounted on the head, shoulder,
elbow, and wrist to monitor the movement chain of a pianist performing repeated
octave keystrokes at different dynamics and tempi. The recording apparatus, termed
the “kymocyclographic method,” was able to capture up to 600 fps, which is
impressive even for today’s standards (Kay et al. 2003).
The optical methods developed and employed by Bernstein and Ortmann were
active marker-based optical systems similar to those used in today’s piano movement
research (e.g., Goebl and Palmer 2009b). These systems have the advantage of
delivering clean data and correct marker mappings but are restricted in the number
of markers and sampling rate and exhibit limitations in their size and cabling. The
idea of mechanical motion tracking developed after Ortmann (1929), and more
recent methods include, for example, electrogoniometers (devices mounted on
joints, see Chung et al. 1992) and data gloves (a glove with open fingertips equipped
with a number of joint angle sensors providing posture information from joint
angles, Winges and Furuya 2014).
Probably the most detailed motion data with a tolerable level of intrusion is
obtained with passive optical marker-based motion capture system that uses a
variable number of infrared cameras to track the position of small reflective markers
8 W. Goebl
that may be mounted on the pianists’ body parts (for an overview, see Goebl et al.
2014b). The number of markers is not limited, and sampling rates go up to about
500 fps. There have been studies measuring pianists’ body motion at finger level
(e.g., Ferrario et al. 2007; Tits et al. 2015; Dalla Bella and Palmer 2011; Goebl and
Palmer 2013). But particularly at tracking pianists’ finger movements, optical
motion capture has to deal with marker occlusion at thumb-under maneuvers or
when curled fingers render fingertips invisible. To overcome the latter, cameras are
placed such that they also monitor the movements from the front-side of the
keyboard, which is only possible for digital pianos. The use of acoustic pianos is
still an open challenge, as the key lid makes the front-side view impossible and
fingertip occlusion is omnipresent.
In passive optical systems, marker tracking and identification may present a
postprocessing problem. A recent study formulated a proportional relationship
between minimal required frame rate, the smallest marker distance, and the largest
marker speed that a motion capture algorithm is able to track (Song and Godøy
2016). For tracking pianist’s fingers, typical marker distances of down to 20 mm and
finger speeds of 2 m/s require a motion sampling rate of larger than 244 fps.
A solution to fingertip marker occlusion at optical systems are magnetic tracking
systems (such as used by Rahman et al. 2011, who used 32 magnetic 6DOF sensors
on both hands of a pianist to reconstruct hand and finger movements). This setup also
works on an acoustic piano, but pianists are limited in their performance by the many
cabled receivers and transmitters attached to their hands and fingers.
Inertial sensors (normally containing accelerometers and gyroscopes) have also
been used to track aspects of pianists’ body motion. Hadjakos et al. (2008) mounted
cabled 6DOF inertial sensors on the pianists’ wrists to visualize the motion of the
lower arms together with a piano roll display. Current commercial systems link real-
time data of multiple inertial sensors to reconstruct full-body motion; however,
finger data has yet to be added.
Minimally intrusive solutions with maximal ecological validity are video based
with either some graphical markers painted on the hand (MacRitchie and Bailey
2013) or even completely without markers, e.g., by using depth image data from a
Microsoft Kinect (Hadjakos 2012). MacRitchie and Bailey (2013) used a single
camera image placed above the piano keyboard to track the motion of the ten fingers
of a pianist, even estimating the height dimension from marker distances. Combined
with finger-keyboard interaction data provided from additional devices, the data
quality was enhanced (MacRitchie and McPherson 2015). Hadjakos (2012) tracked
head, shoulder, arm, and hand movements with a Microsoft Kinect placed above the
piano keyboard. The tracking accuracy and precision was in the range of 3–5 cm, so
possible applications may be in visualization of large-scale body movements but not
any more detailed analysis. There are attempts of multi-camera markerless motion
capture to study the biomechanics of full bodies, such as by Corazza et al. (2006), but
they have not been adapted for precise finger tracking, required for the study of piano
performance.
Biomechanical and Pianistic Attributes of Movement
The artistic control space of a piano is spanned by the wide range of possible force
variations applied to the surfaces of the piano keys and the three pedals (also referred
to as “endpoint redundancy,” Furuya and Altenmüller 2013). Pianists generate the
force variations at the piano keys predominantly by controlling the joint movement
and torques of the kinematic chain of the upper limbs, from the shoulder, elbow, and
wrist to the finger joints. The infinite possibilities of organizing the degrees of
freedom of this kinematic chain are described by the “kinematic redundancy”
(joint movement), “kinetic redundancy” (joint torques, developing from muscular,
gravitational, intersegmental and reaction forces), and the “muscular redundancy”
(e.g., the balance of agonist and antagonist muscles, Furuya and Altenmüller 2013).
An important aspect in piano technique is the movement independence of the
fingers from each other. While non-musicians usually display larger finger enslaving
between the middle and the ring finger than for the index and the little finger, which
is in part due to biomechanical coupling (Häger-Ross and Schieber 2000; Zatsiorsky
et al. 2000), pianists exhibit similar levels of finger independence across all fingers
(Furuya et al. 2011a), suggesting that finger independence can be achieved through
long-term training. Furthermore, a timed finger tapping study involving all possible
finger combinations showed that finger motion was modulated by biomechanical
constraints, but this did not influence the timing accuracy, suggesting that cognitive
control surmounts biomechanical restrictions in skilled performance (Loehr and
Palmer 2007).
Piano playing is more than the successive execution of individual movements;
often individual movements are overlapping or require anticipatory movements to
secure seamless transition of successive events. This effect is termed “coarticulation”
in speech research (Fowler and Saltzman 1993) but has been shown in finger spelling
(Jerde et al. 2003) and piano playing (Engel et al. 1997; Goebl and Palmer 2006).
Pianists change their movement trajectories up to 500 m/s before a leap or a thumb-
under movement to master the horizontal relocation in comparison to a baseline
condition (Engel et al. 1997). Practicing anticipatory movements is explicitly
recommended by renowned piano educators to develop fast and difficult passages
with horizontal relocation requirements (in the case of the left hand in Chopin’s
Prelude Op. 28 No. 3, see Neuhaus 1973).
Piano Touch
“There may be players for whom touch and assault are synonymous. (In German, we find the
deplorable word Anschlag, to strike). The pianist can indeed assault the piano and, for good
measure, the composer and the public. To languages that propose a more loving vocabulary,
like touch and touché, we owe a debt of gratitude.” (Brendel 2013, p. 100, emphasis in
original)
10 W. Goebl
The way pianists manipulate the key surfaces of the piano keys with their fingers
is referred to as piano touch and has been a vibrant topic of debate among pianists,
piano educators, piano students, and piano lovers since the first pianos were intro-
duced (Goebl et al. 2014a; MacRitchie 2015).
In the early twentieth century, a physicist’s view on piano tone and dynamics
began to spread that supported the so-called “single variable hypothesis” (Bryan
1913). This hypothesis refers to the fundamental mechanics of the piano action in
which the hammer travels freely to the strings after the jack has been thrown aside
by contact with the escapement dolly, causing the pianist to lose contact with and
control over the hammer (Fletcher and Rossing 1998, p. 354). The single variable
hypothesis assumes that only the speed with which the piano hammer hits the
strings determines the loudness and the timbre of an isolated piano tone. Therefore,
it does not matter whether it is the well-trained finger of an exquisite piano virtuoso
or a volume of Beethoven sonatas slipping down from the note stand that is
producing a given tone. Only when the final hammer velocities are the same will
the two tones be indistinguishable even for the most refined ear. This view has been
the technical basis for modern reproducing pianos that measure and reproduce the
final hammer velocities of each single piano hammer irrespective of the specific
touch and succeed to create convincing reproduction results (Goebl and Bresin
2003).
However, pianists have a more complex view of the way they interact with the
piano action in an artistic manner (for an overview, see MacRitchie 2015). To
describe different ways of touching a piano key, pianists and scholars use antago-
nistic terms such as percussive versus non-percussive (Ortmann 1929), staccato
versus legato (Askenfelt and Jansson 1990), or struck versus pressed (Goebl et al.
2005; Kinoshita et al. 2007). The former denotes a keystroke in which the fingertip
arrives at the key surface with a certain speed, accelerating the key in a sudden,
percussive manner, while the latter describes a fingertip already resting at the key
surface and pressing it down by accelerating it gradually.
These opposite ways of accelerating a piano key may generate identical final
hammer velocities, but they still entail typical kinematic, mechanical, acoustical, and
perceptual differences: struck keystroke movements exhibit larger shoulder and
finger flexion velocities in repeated arm strikes than pressed ones (Furuya et al.
2010), struck keystrokes require less time to generate a sound than pressed ones for
piano tones that are equally loud (Goebl et al. 2005), and struck keystrokes have a
characteristic finger-key “noise” component in the radiated sound that is absent in
pressed tones (Askenfelt and Jansson 1990). This finger-key sound enables listeners
to identify the particular touch movement that generated an individual piano sound
(Goebl et al. 2014a).
Today, piano touch is regarded as a multimodal phenomenon including auditory,
visual, tactile, and proprioceptive components and therefore concerning all aspects
of piano technique. In this light, the century-old single variable hypothesis of
hammer velocity is outdated in current approaches of studying piano performance
(MacRitchie 2015).
Effects of Intended Tempo and Dynamics on Movement
Expressive piano performance requires pianists not only to accurately time their
many keystrokes but to shape their dynamics at the same time by controlling the
acceleration patterns of the keys (see above). These parameters interact with one
another, so a louder tone requires a faster striking movement, which needs to be
initiated later than does the striking movement for a softer tone to create a sounding
tone at a given time instant. The human motor system must control these interacting
parameters simultaneously.
To create louder octave strikes, pianists reorganize their movements in a partic-
ular way: experts increase their shoulder deceleration to increase elbow rotation,
while beginners use more muscular load at the elbow to come to the same goal
(Furuya and Kinoshita 2008; Furuya and Altenmüller 2013). Experts additionally
reduce the elbow flexor activity working against gravity, thus utilizing “arm weight”
to produce a keystroke, while beginners use agonist elbow muscles for the same goal
(Furuya et al. 2009).
While fundamental movement kinematics such as joint contributions to the final
fingertip movement (Goebl and Palmer 2013) or movement covariation between
striking and non-striking fingers (Furuya and Soechting 2012) have been shown to
remain stable across different performance tempi, several other parameters do change.
With a faster performance tempo, pianists raise their finger higher above the keyboard
(Dalla Bella and Palmer 2011), contradicting common recommendations from peda-
gogues to keep the fingers as close as possible to the key surface when playing very
fast passages. Similarly, the instants of finger movement initiation (maximum finger
height) and of movement goal (tone onset) move closer together in time with faster
tempi (Goebl and Palmer 2009a) but not as much as the inter-onset interval, the time
interval between successive tone onsets, decreases. Keystroke finger movements that
are successive at slow tempi (one is accomplished before the next is initiated) do
overlap largely at very fast tempi, so pianists need to modify the temporal organization
of their finger movements at different tempi (Goebl and Palmer 2009a).
The occurrence of the type of touch is also modulated by the performance
parameters tempo (faster tempi have larger proportions of struck keystrokes,
Goebl and Palmer 2008) and dynamics (soft tones tend to be produced by pressed
touches, while loud by struck touches, Goebl et al. 2005). At slow tempi, pianists
have control over the choice of touch, while at fast tempi, only struck touches are
feasible (Goebl and Palmer 2008). Different ways of touching the keys induce
different tactile sensations in the performer that might be used to stabilize timing
control in performing isochronous sequences (Goebl and Palmer 2008).
Role of Expertise
A practical and effective experimental paradigm to understand movement organiza-

tion in skilled piano performance is to compare performances of highly trained
12 W. Goebl
expert pianists to those of novice pianists (Furuya and Altenmüller 2013). Move-
ment organization in highly skilled pianists differs fundamentally from those of
beginners. Experts strive to achieve physiological efficiency by reducing muscular
load and making more use of proximal joints and gravitational, intersegmental, and
reaction forces than beginners (Furuya and Altenmüller 2013).
In a forced repetition task that required pianists and novices to keep up playing
octaves loudly ( forte) at a rate of 4 Hz over the period of half an hour (thus
producing over 7,000 keystrokes), only experts were successfully accomplishing
this task, suggesting that the specialized movement organization of experts enables
them to avoid muscular fatigue (Furuya and Kinoshita 2008). Similar results have
been confirmed for a tremolo task (quick and repeated alternation of two tones
played with thumb and little finger), in which experts used more proximal muscles
and less finger flexion than amateurs (Furuya et al. 2011b). Experts applied less force
for a shorter duration to the piano keys than beginners to keep a piano key pressed
down after a keystroke (Parlitz et al. 1998).
As physiological movement efficiency becomes more important for the perfor-
mance of very fast passages, pianists with more efficient keystroke finger move-
ments achieve faster maximum tempi than others. A kinematic measure of keystroke
efficiency confirmed this experimentally with a dozen pianists performing isochro-
nous finger exercises from medium fast to very fast tempi. The pianist achieving the
fastest possible tempo condition also showed highest keystroke efficiency values,
while those with lower keystroke efficiency only reached medium fast maximum
tempi (Goebl and Palmer 2013). To illustrate, Fig. 1 shows stick figures of example
keystrokes of two contrasting pianists. The upper graph represents a finger keystroke
with a distal interphalangeal joint (DIP) clearly breaking in and a metacarpal
phalangeal joint (MCP) flexing excessively to compensate this breaking in of the
DIP joint, resulting in a less efficient keystroke movement. The lower graph shows a
keystroke from one of the fastest pianists with high kinematic efficiency. The
majority of endpoint motion is generated in the MCP joint, while the other finger
joints barely move. This latter keystroke exhibits significantly larger efficiency
values than the former (Goebl and Palmer 2013).
Movement Disorders in Piano Performance
The high demands of the concert podium and the resulting competition among
pianists require them to undertake enormous practice efforts, which may often result
in overuse syndromes or long-lasting injuries, also referred to as playing-related
musculoskeletal disorders (Metcalf et al. 2014, “▶ Motion Analysis as Preventive
and Rehabilitative Measures in Dance Medicine”). Among the most frequent pathol-
ogies in professional pianists are tendinitis (tenosynovitis, Sakai 1992), the tennis
elbow (lateral epicondylitis, Metcalf et al. 2014) and, most dramatically, focal
dystonia (Konczak and Abbruzzese 2013).
Focal dystonia is a neurological motor disorder that affects specific regions of the
body (such as the right hand of the pianist) and is typically reflected in involuntary
Max Height Key Bottom

100
S17
Z Position (mm) 80 MCP time

PIP
Wrist
60 DIP
40
20 Tip
0
−50 −100 −150 −200 −250
100 S24
MCP
PIP
Z Position (mm)
80 Wrist
60
DIP
40
20
Tip
0
−50 −100 −150 −200 −250
Y Position (mm)
Fig. 1 Examples of finger keystrokes on the piano with contrasting movement efficiency. Stick
figures of the joints of the middle finger and the wrist of two pianists, seen from a side perspective.
The time from the maximum finger height above the keyboard as the start of a finger striking
movement through to the key-bottom impact as its end are color-coded from blue to red. Pianist S24
showed higher efficiency values than S17 (Figure taken from Goebl and Palmer (2013) p.7)
movement impairments in task-specific actions that are highly trained (for over-
views, see Altenmüller et al. 2012; Konczak and Abbruzzese 2013). The prevalence
of focal dystonia in professional musicians is roughly 1% and affects considerably
more males than females (Altenmüller et al. 2012). Focal dystonia is associated with
a genetic predisposition (Altenmüller et al. 2012), depression, sleep problems, a
decreased precision of tactile and proprioceptive perception (Konczak and
Abbruzzese 2013), as well as behavioral triggering factors such as practicing
repetitive passages over prolonged time spans (Furuya et al. 2015). Cortical
maladaptations in the somatosensory cortex, the basal ganglia, and the cerebellum
have been documented for focal dystonia patients (Furuya et al. 2015).
In pianists, focal dystonia causes involuntary contraction of antagonist muscles
during the execution of well-trained movement sequences, such as playing scales or
chords. Pianists suffering focal dystonia show smaller finger flexion maxima, lower
between-finger independence, and larger timing variability than healthy controls,
according to a recent sensor-glove study (Furuya et al. 2015), documenting clearly
the deterioration in fine motor control in focal dystonia patients.
14 W. Goebl
Future Directions
Despite a long history of scientific investigation of human movement in piano

performance, research in this direction is still to develop. Current restrictions in
capturing the detailed body motion during piano performance such as fingertip
occlusion or measurement intrusion need to be overcome with new technologies
and analysis approaches. Systems are required that are more reliable and less
intrusive (like optical marker-less motion capture), which use only a set of high-
speed cameras, combined with refined and complex computer vision algorithms able
to reconstruct the skeletal movements at the level of individual finger, hand, and arm
phalanges.
The opposition of pianists to integrate scientific findings into the daily musical
experience is still pronounced, not only because of technical hurdles that restrict the
use of motion capture technology to the professional laboratory rather than the stage
of the practice studio. Initiatives in the future should try to fill the gap between
artistic and scientific investigation of the same phenomenon by finding a mutual
language and accepting the fundamental differences in approach and goal
(MacRitchie 2015). It would be desirable to convince pianists to trust and apply
quantitative methods in optimizing their playing technique (Wulf and Mornell 2008)
just as expert athletes in many sport disciplines do on a regular basis.
Cross-Referencs
▶ Body Movements in Music Performances – On the Example of Clarinet Players

▶ Brain and Behavioural Bases of Guitar Learning in Novices
▶ Investigating Aspects of Movement in Violin Performance
▶ Movements, Timing and Precision of Drummers
Acknowledgements I want to thank Laura Bishop and Manfred Nusseck for invaluable comments
on an earlier version of this chapter. This work has been supported by the Austrian Science Fund
(FWF, projects P 24546, P 23248, and J 2526).
References
Altenmüller E, Baur V, Hofmann A, Lim VK, Jabusch HC (2012) Musician’s cramp as manifes-
tation of maladaptive brain plasticity: arguments from instrumental differences. Ann N Y Acad
Sci 1252:259–265. doi:10.1111/j.1749-6632.2012.06456.x
Askenfelt A, Jansson EV (1990) From touch to string vibrations. In: Askenfelt A (ed) Five lectures
on the acoustics of the piano, vol 64. Publications issued by the Royal Swedish Academy of
Music, Stockholm, pp 39–57
Bach CPE (1753) Versuch über die wahre Art, das Clavier zu spielen [An essay on the true art of
playing keyboard instruments]. Henning, Berlin
Behne KE, Wöllner C (2011) Seeing or hearing the pianists? A synopsis of an early audiovisual
perception experiment and a replication. Music Sci 15(3):324–342. doi:10.1177/
1029864911410955
Bernays M, Traube C (2014) Investigating pianists’ individuality in the performance of five timbral
nuances through patterns of articulation, touch, dynamics, and pedaling. Front Psychol Cogn Sci
5(157):1–19. doi:10.3389/fpsyg.2014.00157
Bernstein NA (1967) The coordination and regulation of movements. Pergamon Press, Oxford
Bernstein NA, Popova TS (1929) Untersuchung über die Biodynamik des Klavieranschlags [Study
of the biodynamics of piano touch]. Arbeitsphysiologie 1(5):396–432. doi:10.1007/
BF02012845
Binet A, Courtier J (1895) Recherches graphiques sur la musique [Graphical research into music]
L’Anne’e Psychologique 2:201–222, available also in a German translation by Schmitz H-W
(1994) Das Mechanische Musikinstrument 61: 16–24
Bishop L, Goebl W (2016) Music and movement: musical instruments and performers. In:
Ashley R, Timmers R (eds) Routledge companion to music cognition. Taylor and Francis,
London, to appear
Breithaupt RM (1905) Die natürliche Klaviertechnik [The natural piano technique], 2nd edn. C. F.
Kahnt, Leipzig
Brendel A (2013) A pianist’s A–Z. Faber and Faber, London
Bryan GH (1913) Pianoforte touch. Nature 91(2271):246–248. doi:10.1038/091246a0
Chung IS, Ryu J, Ohnishi N, Rowen B, Headrich J (1992) Wrist motion analysis in pianists. Med
Probl Performing Artists 7(1):1–5
Corazza S, Mündermann L, Chaudhari AM, Demattio T, Cobelli C, Andriacchi TP (2006) A
markerless motion capture system to study musculoskeletal biomechanics: visual hull and
simulated annealing approach. Ann Biomed Eng 34(6):1019–1029. doi:10.1007/s10439-006-
9122-8
Couperin F (1716) L’art de toucher le clevcin [the art of playing the harpsichord]. Che’s l’Auteur, le
Sieur Foucaut, Paris
Dahl S, Bevilacqua F, Bresin R, Clayton M, Leante L, Poggi I, Rasamimanana N (2010) Gestures in
performance. In: Godøy RI, Leman M (eds) Musical gestures: sound, movement, and meaning.
Dalla Bella S, Palmer C (2011) Rate effects on timing, key velocity, and finger kinematics in piano
performance. PLoS One 6(6):e20,518. doi:10.1371/journal.pone.0020518
Engel KC, Flanders M, Soechting JF (1997) Anticipatory and sequential motor control in piano
playing. Exp Brain Res 113(2):189–199. doi:10.1007/BF02450317
Ferrario VM, Macrì C, Biffi E, Pollice P, Sforza C (2007) Three-dimensional analysis of hand and
finger movements during piano playing. Med Probl Performing Artists 22(1):18–23
Fletcher NH, Rossing TD (1998) The physics of musical instruments, 2nd edn. Springer, New York
Flossmann S, Goebl W, Grachten M, Niedermayer B, Widmer G (2010) The magaloff project: an
interim report. J New Music Res 39(4):363–377. doi:10.1080/09298215.2010.523469
Fowler CA, Saltzman E (1993) Coordination and coarticulation in speech production. Lang Speech
36(Pt. 2–3):171–195
Furuya S, Altenmüller E (2013) Flexibility of movement organization in piano performance. Front
Hum Neurosci 7(173):1–10. doi:10.3389/fnhum.2013.00173
Furuya S, Kinoshita H (2008) Expertise-dependent modulation of muscular and non-muscular
torques in multi-joint arm movements during piano keystroke. Neuroscience 156(2):390–402
Furuya S, Soechting JF (2012) Speed invariance of independent control of finger movements in
pianists. J Neurophysiol 108(7):2060–2068
Furuya S, Osu R, Kinoshita H (2009) Effective utilization of gravity during arm downswing in
keystrokes by expert pianists. Neuroscience 164(2):822–831. doi:10.1016/j.
Furuya S, Altenmüller E, Katayose H, Kinoshita H (2010) Control of multi-joint arm movements
for the manipulation of touch in keystroke by expert pianists. BMC Neurosci 11:82–96.
doi:10.1186/1471-2202-11-82
Furuya S, Flanders M, Soechting JF (2011a) Hand kinematics of piano playing. J Neurophysiol
106:2849–2864
16 W. Goebl
Furuya S, Goda T, Katayose H, Miwa H, Nagata N (2011b) Distinct inter-joint coordination during
fast alternate keystrokes in pianists with superior skill. Front Hum Neurosci 5(50):1–13.
doi:10.3389/fnhum.2011.00050
Furuya S, Tominaga K, Miyazaki F, Altenmüller E (2015) Losing dexterity: patterns of impaired
coordination of finger movements in musician’s dystonia. Sci Rep 5(13360). doi:10.1038/
srep13360
Gabrielsson A (2003) Music performance research at the millennium. Psychol Music 31
(3):221–272. doi:10.1177/03057356030313002
Gat J (1965) The technique of piano playing, 3rd edn. Corvina, Budapest
Gerig RR (1974) Famous pianists and their technique. Robert B, Luce
Goebl W, Bresin R (2003) Measurement and reproduction accuracy of computer-controlled grand
pianos. J Acoust Soc Am 114(4):2273–2283. doi:10.1121/1.1605387
Goebl W, Palmer C (2006) Anticipatory motion in piano performance. J Acoust Soc Am 120
(5):3002
Goebl W, Palmer C (2008) Tactile feedback and timing accuracy in piano performance. Exp Brain
Res 186(3):471–479. doi:10.1007/s00221-007-1252-1
Goebl W, Palmer C (2009a) Finger motion in piano performance: touch and tempo. In:
Williamon A, Pretty S, Buck R (eds) Proceedings of the International Symposium on Perfor-
mance Science 2009 (15–18 December 2009). European Association of Conservatoires, Auck-
land, pp 65–70
Goebl W, Palmer C (2009b) Synchronization of timing and motion among performing musicians.
Music Percept 26(5):427–438. doi:10.1525/mp.2009.26.5.427
Goebl W, Palmer C (2013) Temporal control and hand movement efficiency in skilled music
performance. PLoS One 8(1):e50,901. doi:10.1371/journal.pone.0050901
Goebl W, Bresin R, Galembo A (2005) Touch and temporal behavior of grand piano actions.
J Acoust Soc Am 118(2):1154–1165. doi:10.1121/1.1944648
Goebl W, Dixon S, De Poli G, Friberg A, Bresin R, Widmer G (2008) ‘Sense’ in expressive music
performance: data acquisition, computational studies, and models. In: Polotti P, Rocchesso D
(eds) Sound to sense – sense to sound: a state of the art in sound and music computing. Logos,
Berlin, pp 195–242
Goebl W, Bresin R, Fujinaga I (2014a) Perception of touch quality in piano tones. J Acoust Soc Am
136(5):2839–2850. doi:10.1121/1.4896461
Goebl W, Dixon S, Schubert E (2014b) Quantitative methods: motion analysis, audio analysis, and
continuous response techniques. In: Fabian D, Timmers R, Schubert E (eds) Expressiveness in
music performance – empirical approaches across styles and cultures. Oxford University Press,
Oxford, UK, pp 221–239
Hadjakos A (2012) Pianist motion capture with the kinect depth camera. In: Proceedings of the 9th
sound and music computing conference. Aalborg University – Copenhagen, Copenhagen, pp
303–310
Hadjakos A, Aitenbichler E, Mühlhäuser M (2008) Syssomo: a pedagogical tool for analyzing
movement variants between different pianists. In: Proceedings of 5th International Conference
on Enactive Interfaces, Scuola Universitaria Superiore Sant’Anna, Pisa
Häger-Ross C, Schieber MH (2000) Quantifying the independence of human finger movements:
comparisons of digits, hands, and movement frequencies. J Neurosci 20(22):8542–8550
Hagmann P (1984) Das Welte-Mignon-Klavier, die Welte-Philharmonie-Orgel und die Anfänge der
Reproduktion von Musik. Europäische Hochschulschriften: Reihe 35, Musikwissenschaft,
Bd. 10, Peter Lang, Bern, available at http://www.freidok.uni-freiburg.de/volltexte/608/
Jabusch HC, Vauth H, Altenmüller E (2004) Quantification of focal dystonia in pianists using scale
analysis. Mov Disord 19(2):171–180
Jerde TE, Soechting JF, Flanders M (2003) Coarticulation in fluent fingerspelling. J Neurosci 23
(6):2383–2393
Kay BA, Turvey MT, Meijer OG (2003) An early oscillator model: studies on the biodynamics of
the piano strike (Bernstein & Popova, 1930). Mot Control 7(1):1–45
Kinoshita H, Furuya S, Aoki T, Altenmüller E (2007) Loudness control in pianists as exemplified in

keystroke force measurements on different touches. J Acoust Soc Am 121(5):2959–2969.
doi:10.1121/1.2717493
Konczak J, Abbruzzese G (2013) Focal dystonia in musicians: linking motor symptoms to somato-
sensory dysfunction. Front Hum Neurosci 7(297):1–10. doi:10.3389/fnhum.2013.00297
Krampe RT, Ericsson KA (1996) Maintaining excellence: deliberate practice and elite performance
in young and older pianists. J Exp Psychol Gen 125(4):331–359
Loehr JD, Palmer C (2007) Cognitive and biomechanical influences in pianists’ finger tapping. Exp
Brain Res 178(4):518–528
MacRitchie J (2015) The art and science behind piano touch: a review connecting multi-disciplinary
literature. Music Sci 19(2):171–190. doi:10.1177/1029864915572813
MacRitchie J, Bailey NJ (2013) Efficient tracking of pianists’ finger movements. J New Music Res
42(1):79–95. doi:10.1080/09298215.2012.762529
MacRitchie J, McPherson AP (2015) Integrating optical finger motion tracking with surface touch
events. Front Psychol Perform Sci 6(702):1–14. doi:10.3389/fpsyg.2015.00702
Matthay T (1903) The act of touch in all its diversity. An analysis and synthesis of pianoforte tone
production. Bosworth & Co. Ltd., London. http://www.archive.org/details/
actoftouchinalli009163mbp
McPherson AP (2012) Touchkeys: capacitive multi-touch sensing on a physical keyboard. In:
Essl G, Gillespie B, Gurevich M, O’Modhrain S (eds) Proceedings of the International Confer-
ence on New Interfaces for Musical Expression (NIME). University of Michigan, Ann Arbor
McPherson AP, Kim YE (2011) Multidimensional gesture sensing at the piano keyboard. In:
Proceedings of the SIGCHI conference on human factors in computing systems. ACM, Van-
couver, pp 2789–2798. doi:10.1145/1978942.1979355
Metcalf CD, Irvine TA, Sims JL, Wang YL, Su AWY, Norris DO (2014) Complex hand dexterity: a
review of biomechanical methods for measuring musical performance. Front Psychol Cogn Sci
5(414):1–12. doi:10.3389/fpsyg.2014.00414
Neuhaus H (1973) The art of piano playing. Barrie & Jenkins, London
Ortmann O (1925) The physical basis of piano touch and tone. Kegan Paul, Trench, Trubner/J.
Curwen/E. P. Dutton, London/New York
Ortmann O (1929) The physiological mechanics of piano technique. Kegan Paul, Trench, Trubner,
E. P. Dutton, London/New York. paperback reprint: E. P. Dutton, New York, 1962
Palmer C (1997) Music performance. Annu Rev Psychol 48:115–138. doi:10.1146/annurev.
psych.48.1.115
Parlitz D, Peschel T, Altenmüller E (1998) Assessment of dynamic finger forces in pianists: effects
of training and expertise. J Biomech 31(11):1063–1067
Phillips-Silver J (2009) On the meaning of movement in music, development and the brain.
Contemp Music Rev 28(3):293–314
Rahman MM, Mitobe K, Suzuki M, Takano C, Yoshimura N (2011) Analysis of dexterous finger
movement for piano education using motion capture system. Int J Sci Technol Educ Res 2
(2):22–31. http://www.academicjournals.org/IJSTER
Rainbow B (1990) Johann Bernhard Logier and the chiroplast controversy. Music Times 131
(1766):193–196. http://www.jstor.org/stable/966259
Sakai N (1992) Hand pain related to keyboard techniques in pianists. Med Probl Performing Artists
7(2):63–65
Song MH, Godøy RI (2016) How fast is your body motion? Determining a sufficient frame rate for
an optical motion tracking system using passive markers. PLoS One 11(3):e0150,993.
Stewart L, Verdonschot RG, Nasralla P, Lanipekun J (2013) Action–perception coupling in pianists:
learned mappings or spatial musical association of response codes (smarc) effect? Q J Exp
Psychol 66(1):37–50, doi:10.1080/17470218.2012.687385
Tits M, Tilmanne J, d’Allesandro N, Wanderley MM (2015) Feature extraction and expertise
analysis of pianists’ motion-capture finger gestures. In: Proceedings of the 2015 International
18 W. Goebl
Computer Music Conference, International Computer Music Association, pp 102–105, http://

hdl.handle.net/2027/spo.bbp2372.2015.019
Tro J (2000) Aspects of control and perception. In: Rocchesso D, Signoretto M (eds) Proceedings of
the COST–G6 Conference on Digital Audio Effects (DAFX–00), December 7–9, 2000,
Universita` degli Studi di Verona. Dipartimento Scientifico e Tecnologico, Verona, pp 171–176
Tsay CJ (2013) Sight over sound in the judgment of music performance. Proc Natl Acad Sci 110
(36):14,580–14,585. doi:10.1073/pnas.1221454110
White WB (1930) The human element in piano tone production. J Acoust Soc Am 1:357–367.
doi:10.1121/1.1915190
Widmer G, Goebl W (2004) Computational models of expressive music performance: the state of
the art. J New Music Res 33(3):203–216. doi:10.1080/0929821042000317804
Winges SA, Furuya S (2014) Distinct digit kinematics by professional and amateur pianists.
Neuroscience 284:643–652
Wulf G, Mornell A (2008) Insights about practice from the perspective of motor learning: a review.
Music Perform Res 2:1–25
Zatsiorsky VM, Li ZM, Latash ML (2000) Enslaving effects in multi-finger force production. Exp
Brain Res 131(2):187–195
Movements, Timing, and Precision
of Drummers
Sofia Dahl
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
A World of Percussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Measuring Drumming Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Measuring Concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Percussive Movements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Player-Instrument Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Preparing a Stroke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Interaction Between Drumstick and Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Contextual Influence on System Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Timing Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
A Case of Drum Set Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Conclusion and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Abstract
Percussion instruments vary a great deal in material properties and characteristics
of the interaction. Most drums, however, produce sounds with well-defined
onsets, and contact time between player and instrument can be very brief. Players
of percussion need to have precise control of timing and striking force for
sometimes very rapidly repeating, fast movement patterns. This chapter starts
by outlining what is specific for drumming movements and how they can be
measured. Some general movement strategies used in drumming movements are
described and also how tempo, dynamic level, and the surface struck affect the
player-instrument interaction and timing control. Lastly, full-body motion capture
S. Dahl (*)
Department of Architecture, Design and Media Technology, Aalborg University Copenhagen,
Copenhagen, Denmark
e-mail: sof@create.aau.dk

DOI 10.1007/978-3-319-30808-1_110-1
2 S. Dahl
data from a drum set performance is used to illustrate some of the points
addressed in the chapter.
Keywords
Drumming • Percussion • Music • Player-instrument interaction • Rhythm •
Timing • Striking force • Striking • Downstroke • Beater • Preparatory height •
Rates • Rebound • Beat interval • Drum set
Introduction
As a starting point for this chapter, it should be useful to clarify some ways in which
the playing of percussion instruments in general, and drumming in particular, differ
from other human activities. There are a number of other hitting actions in everyday
and sports movements, for example, hammering a nail, knocking on a door, or using
a bat or club to hit a ball (chapters “▶ Movement Analysis of the Golf Swing” and
“▶ Hitting and Pitching”). These are all repeated, iterative, actions that require
(to different extent) anticipation and planning of the interaction if they are to be
performed well without hitting ourselves or missing the ball. But playing musical
instruments in general, and percussion in particular, is also subjective to a musical
context where the timing of actions typically need to fit a larger “grid” of timed
events, a constraint of metrical hierarchy. Moreover, the goal of musical activities is
communication (chapter “▶ Body Movements in Music Performances – on the
Example of Clarinet Players”) and expression, and this separates musical expertise
with respect to, e.g., sports. Although timing is extremely important to master as a
musician, a strictly isochronous performance where all intervals are equally long is
not the desired goal. Neither is the fastest, hardest striking player who generates the
loudest sounds the best musician. This might seem obvious, but in view of the body
of work that is done about maintaining temporal regularity (and particular isochro-
nous sequences), let’s just remember that it is how a musician uses these abilities in a
musical way that makes the performer we want to hear and see.
A World of Percussion
One thing that makes percussion stand out from most other instrument groups is the
vast variety. Throughout the world, percussion instruments appear in so many forms,
shapes, and sizes that the sheer diversity can fill volumes. Drums are commonly
membranophones, where the sound is produced by vibrations in some kind of
membrane that can be made from animal skin or plastic and “tuned” to different
Movements, Timing, and Precision of Drummers 3
tension.1 The membrane is struck by the hand or some kind of beater that also can
vary with respect to shape, weight, and material properties. For instance, a snare
drum is typically played with wooden drumsticks (sometimes with nylon tips) or
bundles of thin sticks or bristles (brushes), whereas orchestral timpani typically are
played using mallets with felt tips where the hardness can differ. Again, the different
materials and properties of the beater will affect the interaction with the instrument
played and hence influence the sound. A professional percussionist performing in a
variety of musical genres would have learned to master a range of these different
combinations,2 but also an amateur drum set player who prefers one set of sticks will
interact with at least five different instruments involving four limbs: snare drum,
cymbals, and tom-toms played with drumsticks and bass drum and hi-hat played
with feet on pedals.
Given the wide range of instruments, beaters, possible interactions, and music
traditions, it is perhaps not surprising that only the tiniest fraction of all possible
drumming movements have been studied empirically. And also with the investiga-
tions that have been made, there is an emphasis on Western instrument traditions.
Depending on the scope of the investigation, researchers typically need to control for
factors influencing the player-instrument interaction, and thus investigations on
drumming tend to focus on single strokes on (electronic) drums or force plates,
whereas timing production and synchronization studies predominantly have used
finger tapping. In fact, the number of studies that have investigated movement and/or
timing performance on acoustical drums is very low, and movement analysis on
hand drumming is practically absent.3 Hence, the studies that form the basis of this
chapter only reflect a very small portion of the drumming variety over the world.
Reasons for this limited scope are probably sociocultural, but another factor could be
the general need of selecting and controlling the independent variables in an
experiment (which may be difficult enough when studying music performance). In
addition, it is typically difficult to capture good, reliable data in a (reasonable)
ecological setting, a challenge that we will turn to in the next section.
Measuring Drumming Performance
As might be concluded from the introduction, the data we can measure and approx-
imately what we can expect in terms of values will vary with the type of instrument
played. Large, full-body movements with large amplitudes for a standing taiko
drummer hitting the big drum with thick wooden sticks versus a player sitting
1
Drums can also be idiophones, such as slit or steel drums, where a more solid structure is set into
vibration, and in some drums also, an air column is set to vibrate (e.g., udo clay drums or goblet
drums where part of the drum is a Helmholtz resonator).
2
World-renown solo percussionist Evelyn Glennie allegedly travels with 1–2 t of equipment.
3
Although hand drumming interactions have occasionally been measured in other fields (e.g., Jones
et al. 2007; Tindale et al. 2005)
4 S. Dahl
down and using fingers to produce differently pitched strokes on an Indian tabla
drum. Thus, there are different constraints on the equipment that can be used, and the
methods used for measurement and analysis will differ widely. Typically, however,
drumming movements are fast, which makes it difficult for both performer and
observer to study the motion. Recordings of drumming movement can be of great
help for students of percussion to understand the mechanics of a stroke (see Moeller
1956, for an early example of motion display for didactic purposes using still images
from filmed playing).
Measuring Concerns
Generally, one main constraint when measuring drumming performance is the

timing resolution of the measurement system (chapter “▶ Observing and Revealing
the Hidden Structure of the Human Form in Motion”). If the main interest is in
players’ overall movement patterns, 30–60 Hz can be sufficient which enables
affordable and easily accessible options such as video or Microsoft Kinect. For
more detailed information on fast movements of, e.g., drumsticks, however, it may
be necessary to use high-speed cameras or other types of sensors that react fast
enough. Today, the high-end optical motion capture systems are capable of the high
sampling rates typically needed to capture stick motion also at the time of contact.
Exactly the sampling rate needed depends on the type of interaction (where a hard
surface and high impact velocities would require more) and also on the position of
the marker to be tracked, but 400 Hz or even higher can be necessary in order to
ensure that a marker close to the tip of a drumstick is sampled at impact. In reality,
however, the placement of markers or sensors on a drumstick or mallet is a
compromise between getting an accurate estimate of the stick motion and the risk
of the marker obstructing the playing or falling off during playing (see Fig. 1). In
order to fully capture the rotation of a drumstick, a combination of at least three
markers in a triangular configuration is needed, something that may also be difficult
to do on a regular drumstick or thin beater, without attaching the cluster as an extra
structure (with the risk of affecting the balance during playing).
Fig. 1 Hands and drumsticks

during motion capture
recording. The markers’
placement is a compromise
between getting an accurate
estimate of stick motion and
rotation and the risk of falling
off or obstructing playing
Fig. 2 Close-up of modified drumstick used for recording interaction between drumstick and
drumhead. Only one of the pair of strain gauges is visible. The light-emitting diode is placed as
close as possible to the very tip without obstructing playing. The copper foil at the tip closes an
electrical circuit when the drumstick is in contact with a conducting layer on the drum or surface
struck, giving precise reading on the time and duration of contact
Movement and other interaction data can also be recorded by attaching sensors to
the limbs and sticks of players. Accelerometers, piezo crystals, and strain gauges can
be used to record relative motion and interaction forces, e.g., between drumstick and
drumhead (Dahl et al. 2011a; Wagner 2006). Figure 2 shows the modified drumstick
used in Dahl et al. (2011a). Here a pair of strain gauges (placed on each side of the
stick, only one visible in the picture) registers the bending deformation of the
drumstick (which is dependent on the force during impact), and an electrical circuit
is closed when a thin copper foil at the very tip of the drumstick is in contact with a
graphite layer sprayed on the drumhead, giving information on the contact time
between drumstick and drumhead. Infrared light from a light-emitting diode is
registered by the motion capture system. An obvious disadvantage with the attach-
ment of sensors to a drumstick or mallet is the attachment of cables that need to be
carefully adjusted so that they do not obstruct the playing. The added sensors and
cables also alter the weight distribution and balance of the stick which may affect
playing and also result in a rotation of the sensor and less accurate measurements.
Playing with a modified stick means paying attention to that sensors face the right
way, and this can be difficult during fast playing. By combining force measurements
with microphone recordings (carefully calibrated at a defined distance), there is a
possibility of detecting and removing unreliable data points. Similarly, a missing
contact time can mean that the stick was rotated (or hit off the mark).
An easier access to response data from hits without attaching cables or sensors to
the beater is to collect the data from the surface struck. Participants can play on a
force plate or commercially available electronic drum pads (which typically give
output using MIDI4 coded as values between 0 and 127). Although not being a real
acoustical instrument, a force plate or drum pad can still be assumed to correspond to
4
Musical Instrument Digital Interface (MIDI) is a protocol and standard to communicate between
different devices and software programs.
6 S. Dahl
Fig. 3 Recording setup of movements during drum set playing. The drummer is playing on an
electronic drum set with MIDI interface. In order to reduce interference from reflections, the shiny
stands have been covered in black fabric. The electronic drum set cymbals have plastic covering that
reduces reflections which makes them more favorable in a motion capture setting compared to
acoustic cymbals. The markers on the player’s feet are (barely) visible, and here camera placement
is important so that the markers are not occluded by the legs or drums
the normal practicing situation for many drummers, who for practical considerations
often need to choose more quiet alternatives to acoustical drums.
An example of a real recording situation is seen in Fig. 3, which shows recording
of full-body movement during drum set playing. The drummer is wearing a motion
capture suit with reflective markers attached and is playing an electronic drum set
(Roland V-drums) where the MIDI output is synchronized with motion data. Passive
motion capture systems rely on that cameras record the reflections of emitted
infrared light, and here undesired reflections from metal stands or cymbals can be
a real challenge. In this particular example, the cymbal pads reduced this risk thanks
to the plastic covering and stands were covered up. Another challenge is occlusion
from instrument parts or the player’s body which may result in data gaps. A large set
of cameras at different positions are needed in order to have all markers in view
during the performance. For this reason, the quality of the motion data needs to be
thoroughly checked, and often cleanup and repair of data gaps are necessary.
Surface electromyography (EMG) (chapter “▶ Surface EMG”) can also be used
to monitor muscle recruitment and movement patterns (Fujii et al. 2009a, b; Fuji-
sawa and Miura 2010). As with other sensors, cables may obstruct playing if not
attached carefully. Cable movements can also generate artifacts through playing due
to the rapid changes in direction at hits, especially if attached on the hand which can
be moved by stick vibration at impact.
Percussive Movements
A professional player not only needs to be able to control striking force and timing
but do this at, sometimes a very fast, rate. A movement strategy that can help doing
that and enable a player to keep on playing for long periods of time would be a good
investment for the individual player (chapters “▶ Optimal Control Strategies for
Human Movement” and “▶ Movement Efficiency in Piano Players”).
Player-Instrument Interaction
A general characteristic of percussion instruments is that they produce impulse-like

sounds with onsets that are well defined. In terms of the interaction, the timbre and
sound level are related to the history of the force pulse from the contact between
player/beater and instrument. A brief force pulse with high amplitude will produce a
rich sound spectrum with many partials. Playing with a beater makes it possible to
excite the instrument more vigorously than what is possible using the bare hands. A
mallet or drumstick can be accelerated to achieve high striking velocities at a low
physiological “cost.” However, often this comes at the expense of quickly being able
to change characteristics of the beater used. Where a conga player has the option of
changing hand shapes to change the timbre between subsequent strokes (a bass
stroke compared to slap), this could require different mallets for another
percussionist.5
A player’s movement defines the velocity and effective mass at impact, as well as
the contract duration which can be very brief. Measured contact durations between
drumstick and drumhead are 5–8 ms for a mezzo forte stroke at the center of a
tom-tom (Dahl 1997), rototom (Dahl et al. 2011a), or snare drum (Wagner 2006). In
such a short time, there is no time for voluntary adjustment from the player, but
whatever striking force and dampening effect need to be part of the striking
movement from the start. Any intention of dampening out a stroke during impact
has to be part of the striking movement itself.
Preparing a Stroke
A way to achieve the striking force required to set the instrument in vibration is to
create some kind of “runway” during which the hand or beater can accelerate by
lifting it to a height some distance away. Several studies on drumming show how
players initiate a stroke by lifting the drumstick in a preparatory movement to a
height from which the actual downstroke is then initiated. The same principle can be
seen in preparations for strokes in racquet sports or for batters, where the stroke is
initiated from a position with the body and torso turned (chapter “▶ Hitting and
5
Another option is to alternate; samba surdo players use one hand and a mallet in the other hand.
8 S. Dahl
10 cm
Fig. 4 Arm, hand, and stick trajectories during a mezzo forte stroke on a rototom. The image shows
movement trajectories of markers attached to a player’s shoulder (upper left), elbow (bottom left),
wrist, hand, and stick (rightmost loop). The “stick figure” is made by straight lines connecting the
markers every 25 ms. The preparation for the stroke starts with the hand moving upward and the
stick following (see upward arrow). After reaching the preparatory height, the actual downstroke
starts (downward arrow), and the stick quickly gains velocity. After the impact, the stick is thrown
back up into another, smaller trajectory by the rebound from the drumhead
Pitching”). An example of a preparatory movement can be seen in Fig. 4, which

shows the positions of markers attached to a player’s shoulder, elbow, wrist, knuckle,
and at the end of a drumstick during a mezzo forte stroke at a slow tempo. The
rightmost trajectory in the figure shows the larger path of the drumstick marker, but
angle of the line shows how the upward movement is initiated from the wrist. The
wrist is leading the upward movement while the hand and stick are following and the
tip of the stick is pointing downward. As the stick approaches the maximum vertical
displacement, its preparatory height, the hand initiates the downstroke, flicking the
stick down toward the drumhead. This kind of “whiplash” movement is described in
books on playing technique (e.g., Famularo 1999; Moeller 1956; Shivas 1957); and
its smooth, wavelike characteristics have been observed in studies on movements of
drummers using motion capture (Dahl 2000, 2004; Trappe et al. 1998; Waadeland
2006b).
There is a clear, almost linear, relationship between preparatory height and
striking velocity at impact (Dahl 2004; Dahl et al. 2011a). However, coefficients
and slopes of the relationship vary due to the considerable individual differences in
playing strategies between players and also because players interpret dynamic levels
as having different loudness. For example, the maximum vertical velocity for the
stick marker during fortissimo strokes for one player in Dahl (2004) was about 9 m/s,
whereas the maximum stick velocity for another player using the same stick was
closer to 20 m/s. (The real velocity of the tip of the drumstick would have been
higher, Dahl estimated the difference to 30 %.) In addition to a high preparatory

height, the player that produced the highest striking velocity in Dahl’s study used
specific technique called Moeller stroke (Moeller 1956), where a larger elbow
movement is used to lift the stick high and also accelerates it to a very high peak
velocity at impact.
Interaction Between Drumstick and Surface
When a player has the time to prepare, strokes at higher dynamic levels are initiated
from a greater height compared to strokes played at softer dynamic levels. For
instance, four players in a study by Dahl (2004) increased preparatory heights for
accented strokes compared to unaccented strokes, resulting in (on average) a dou-
bling in striking velocity for the three dynamic levels used (pp, mf, ff). Dahl (2004)
also found that players adjust the striking velocity to match which surface (soft,
normal, and hard) that was attached to the force plate used in the study. In general,
players reduced striking velocity when playing on the hard surface as compared to
normal. By comparison, the soft surface resulted in an increase in striking velocity.
An expected result, as professional players of percussion are required to play the
same rhythms on many different surfaces with different tactile feedback.
The rebound from the struck surface or drumhead, which throws the drumstick
back up directly after impact, is also dependent on the surface and striking velocity.
Figure 4 shows how the stick abruptly changes movement direction after the hit and
completes a smaller loop compared to the preparatory movement. In effect, the
rebound means that a player can play several impacts with one striking movement,
as has been reported for swing patterns (Waadeland 2006a), and this is a prerequisite
for playing very fast stroke patterns like ruffs or closed rolls (with clusters of up to
six hits per stroke reported (Miura 2005)). Keeping a grip that is relaxed enough to
let the stick “do the work” while maintaining control over the interaction between
instrument and drumstick is something that players spend considerable amount of
time learning to do.
There are several different types of grips used for a beater, but typically a player
strives to hold a drumstick or mallet in such a way that it is free to rotate around a
fulcrum, most commonly between the thumb and index finger.6 In this way, the stick
is free to rotate in the vertical plane, but the player can use the fingers to stabilize or
“lock” it if needed. In some cases, there is a need to tighten the grip to stabilize the
stick to actively control the effect of the rebound, for instance, when players have to
play strokes at different dynamic levels, as in the case of accents (see Dahl 2004). In
a “controlled” stroke, a player restrains the rebound in preparation for a following
stroke at a softer dynamic level (see, e.g., Famularo 1999). Restraining the rebound
6
The type of grip used naturally varies with beater and instrument. For instance, a common way to
hold the double-headed beater often used in bodhrán playing more resembles holding a pencil, and a
rotational movement allows both ends to strike the vertically held drumhead.
10 S. Dahl
after a strong hit also means an altered interaction in comparison to the “normal”
case where the player let the drumstick rebound freely from the surface. Since the
contact times are so short and players need to make these adjustments beforehand,
the “controlled” stroke also results in a different sound (Dahl and Altenmüller 2008).
Percussion instruments typically have a large dynamic range and can generate
both very soft and very loud sounds. In terms of contact force, Dahl and colleagues
(2011a) reported peak forces ranging from 1.8 to 106.8 N from four professional
percussionists playing strokes at different dynamic levels on a drum, using the
modified drumstick showed in Fig. 2. Fujii and Oda (2009a) had 17 drummers tap
rapidly on a force plate and reported mean peak forces across participants to be
roughly 20 N when using a drumstick and about 7.7 N when tapping with their
finger. By means of comparison, Kinoshita et al. (2007) reported maximum finger
contact forces in piano reaching up to 80 N.
Contextual Influence on System Control
The specific characteristics of the instrument played (e.g., hardness of surface or

tension of the drumhead), in combination with the properties of the beater used,
partly define the limits of the temporal rates and sound levels played. Utilizing the
rebound from the playing surface and its upward lift of the stick appears to be part of
an energy efficient movement strategy. At medium tempi, the rebound can simply be
incorporated in the preparatory motion for the following stroke. If the time between
strokes is too long, the player has to initiate each stroke from the start, lifting the stick
anew. At faster tempi, on the other hand, where intervals between strokes are 200 ms
or less, there is typically no time for larger preparatory movements, and a decrease in
movement amplitude (and loudness) is necessary.
Together with the instrument, the player’s arm, hand, and any beater used make
up a dynamic system. A system has its resonance frequencies at which a large output
can be obtained with only small energy supplied. This is similar to the bouncing of a
ball, which can require comparable little input energy at a certain natural frequency.
Bouncing at a high rate with large amplitude, however, is quite demanding, but can
be made much easier if one moves closer to the ground to reduce the amplitude. The
optimal bouncing rate depends on the size and “bounciness” of the ball. In the same
way, the combination of a beater’s weight and stiffness together with the elasticity of
a drumhead makes certain combinations of tempo and dynamic level more playable
than others. In general, a proficient player will adjust the fulcrum and tightness of the
grip to take advantage of the system while maintaining control and minimizing
undesired variability in timing and sound level.
The fastest drumming rates reported are just over 1200 single strokes in 1 min
using both hands,7 corresponding to around 100 ms between strokes for one hand
(10 Hz). At these rates, larger hand or arm movements are not possible, but finger
7
See http://www.extremesportdrumming.com/wfdchampions.htm.
action can provide energy input to the system, which does not necessarily “stiffen
up” despite the fast rates. Fujii et al. (2009a) measured muscle activity of the 2005
world record holder who produced an average beat interval of 98 ms with his left
hand during 10 s. Surface electromyography showed low level of co-contraction,
that is, little overlap between activation of wrist extensor and flexor muscles. This
means that the muscles were not “locking” the wrist, making it stiff during playing,
but that the high drumming rate were produced with a compliant joint. In addition,
the drummer showed low variability in timing intervals with standard deviations of
9.5 and 6.6 ms for the left and right hand, respectively. Muscle activity for other
drummers in the study showed similar pattern, with low degree of overlap between
antagonist wrist muscle groups, compared to non-drummers (Fujii et al. 2009a).
Similar results showing differences in average muscle recruitment between
drummers and non-drummers have been reported by Fujisawa and Miura, recording
single strokes on a drum for 3 min at different tempi (Fujisawa and Miura 2010).
Comparing average EMG for the first and the last minute of playing, three
non-drummers showed significantly higher mean EMG during the last minute of
playing more often than did the three trained drummers. Fujisawa and Miura also
compared playing with stick with and without hitting a drumhead and found
decreased average EMG for the condition with the drumhead, indicating that expe-
rienced drummers were able to utilize the rebound from the drum more efficiently
than non-drummers.
Balance between limbs is also important in the control of temporal patterns,
especially when the same rhythm is to be performed with both hands. As can be
expected, trained drummers have been shown to have good inter-hand coordination,
also for rapid movements (Fujii et al. 2010; Fujii and Oda 2009b), and in terms of rates
produced, drummers display less difference between hands than non-drummers (Fujii
et al. 2009a). It is typically the non-preferred hand (i.e., left for right-handed players)
that shows the higher variability in force and timing of strokes and also is unable to
match the rates produced by the preferred (right) hand (Fujii et al. 2009a, b).
Timing Control
For instruments with clearly defined onsets, such as percussion, precise control of
timing is essential if a performance is to be musically acceptable. Comprehensive
reviews of the literature on sensory motor synchronization has been published by
Repp (2005, 2010; Repp and Su 2013) so this section will be limited to briefly go
through timing variability in drumming actions with some comparisons with finger
tapping. Madison et al. (Madison et al. 2013) compared ways of tapping (with finger,
a drumstick, and with or without tactile feedback) for a group of university students
and found lower variability for tapping with stick compared to finger tapping. Not
surprisingly, the amount of training participants had also helped to reduce the
variability.
Table 1 gives an overview of coefficient of variation (CV, calculated as standard
deviation relative to mean) for studies using different tasks and effectors for
12 S. Dahl
Table 1 Overview of timing control in drumming, values are coefficient of variation (CV,
calculated as standard deviations divided by mean interval) and are estimated from data reported
in the studies cited in the rightmost column. Where CV has not been reported, values are estimated
from reported variability and tempo. Ranges are shown for studies reporting values for different
conditions. With the exception of Dahl et al. (2011b), Danielsen et al. (2015), and Madison (1992),
values are based on averages across participants, which would give lower values. The ranges shown
are therefore underestimating the real CV in most cases
Beat
intervals
Task Synch Effector (ms) CV(%) Participants
Isochrony Click Unimanual 333–1000 3.1–8.7 6 unspecified (Sternad et al.
finger 2000)
Isochrony Free Unimanual 500–574 ca 5 8 university studentsa
finger (Madison et al. 2013)
Isochrony Free Unimanual 500–574 2.7 8 university studentsa
stick (Madison et al. 2013)
Beat Click Foot 300–1000 1.2–1.6 15 drummers (Fujii et al.
rhythm 2011)
Beat Click Bimanual 150–1000 0–4 15 drummers (Fujii et al.
rhythm stick 2011)
Beat Clickb Bimanual 412–938 1–3 10 drummers (Danielsen et al.
rhythm stick 2015)
Beat Click Bimanual 500–600 0.9–1.4 3 drummers (Madison 1992)
rhythm stick
Isochrony Free Unimanual 200 1.5–6.5 4 drummers (Dahl et al.
c
stick 2011b)
Speed Free Unimanual 147 4.2 17 drummers (Fujii et al.
stick 2009b)
Speed Free Unimanual 143–161 18–24 23 non-drummers (Fujii et al.
stick 2009a)
Speed Free Unimanual 151–156 5–7 44 drummers (Fujii et al.
stick 2009a)
a
After six 30 min training sessions, no professional musicians
b
Only case “On the beat” considered here
c
Values with extra bounces removed for two drummers with focal dystonia
non-drummers and drummers. The values reported in studies vary between individ-
ual and group means, and so the CV values in the table are ballpark estimates meant
to give a rough estimate of overall performance. As can be seen by the values, CV
lies within three percent for drummers playing with stick or foot at medium tempi
and a bit higher for non-drummers playing with stick (after the training in Madison
et al. (2013)). At higher playing rates, for beat interval of 200 ms (corresponding to
300 beats per minute) or higher, variability increases, substantially so for the speed
tapping task.
Many studies have concentrated on interval production where the task is to
produce (close to) isochronous intervals between strokes at different tempi. The
task is straight forward and makes it easy to quantify performance, also between
players. For actual drum performance using regular rhythms, there are a more
modest number of studies, predominantly concentrated on timing performance with

click track (Danielsen et al. 2015; Fujii et al. 2011; Hellmer and Madison 2015;
Madison 1992), but also relative timing in the playing of swing patterns (Friberg and
Sundström 2002; Honing and De Haas 2008) and Malian jembe drumming have
been investigated (Polak et al. 2016). The focus in studies of swing drumming has
been on how the relative length of the shorter interval in the typical cymbal pattern
used in much jazz music (“ding-chicka-ding”) is altered with overall playing tempo.
Measurements of commercial recordings of famous drummers (Friberg and
Sundström 2002) as well as experimentally controlled recordings in the lab (Honing
and De Haas 2008) show that the drummers have minute control over the timing
performance and adjust the relative timing of the pattern to the base tempo. The three
drummers in the study by Honing and Bas De Has displayed an impressive precision
over the timing pattern showing standard deviations ranging between 5 and 10 ms
for the nine tempi tested (Honing and De Haas 2008). Similarly, the jembe drum-
ming performances measured by Polak et al. showed that the drummers are very
precise in timing the strokes within each measure and adjust according to different
musical pieces played. The pieces typically increase in tempo across a performance,
but mean asynchronies were in the range of 6–12 ms for beat intervals between
300 and 600 ms (Polak et al. 2016).
A Case of Drum Set Performance
So far, much of the drumming work done has looked at performance using drum-
sticks and frequently measured drum set players, but movement measurements
during drum set performance have previously not been reported. Specifically for
drum set players compared to other percussionists and other instrumentalists (with
the exception of organists) is that players regularly play using all four limbs. A drum
set (or kit) can be set up in different ways and with varying numbers of instruments,
but basically it consists of a bass drum, snare drum, one or more tom-toms, hi-hat,
and a set of cymbals. Together these instruments form the percussion basis for much
of the (Western) popular music, spanning across widely different genres and played
in a multitude of ways. We will here take a look some data of a simple rhythm played
on drum set as an example.
Figures 5 and 6 show motion capture data from approximately two bars of slow
beat pattern played on hi-hat, snare drum, and bass drum in the same setup showed in
Fig. 3. Figure 5 shows the player as a stick figure viewed both from the front and the
side, with traces showing the movement trajectories of the markers on the player’s
head, toes, hand knuckles (the index MPC joint), and drumsticks. The loops formed
by the hand and stick trajectories during the preparatory movements are clearly seen
from both views. Note that the head marker is also moving systematically despite the
player being seated.
A projection of the hand and stick trajectories in the vertical direction versus time
is shown in Fig. 6. The pattern played is a standard beat pattern (“boom-chack,
boom-boom-chack”) at roughly 73 beats per minute, where the right hand is playing
14 S. Dahl
Fig. 5 Stick figures based on full-body motion capture during approximately one measure of a
slow standard beat pattern viewed from the front (left) and player’s right side (right). The colored
traces show the movement trajectories from the player’s head (yellow), right hand (magenta) and
stick (red) playing the hi-hat, left hand (green), stick (blue) playing snare drum, left toe (cyan) on
hi-hat pedal, and right toe (magenta) on bass drum pedal. The drummer’s hands are crossed while
playing the hi-hat, a typical setup most right-handed drummers use. Figure created using the MoCap
toolbox (Burger and Toiviainen 2013)
eight notes on the hi-hat (red line) with an interval of 410 ms, while the bass drum
(magenta) and snare drum (dark blue) are marking first and third beat each measure/
bar. The second bar has two bass drum strokes along with the first two eight notes on
the hi-hat, and the player’s foot on the hi-hat pedal is starting to lift the hi-hat for the
final stroke in the pattern (the stroke itself is just outside the plot here).
The preparatory movements for the drumstick playing the hi-hat vary somewhat
depending on which beat in the pattern that is played. The first hit seen is the first
beat in the pattern and is emphasized by being played on both hi-hat and bass drum.
The following stroke is a less emphasized eighth note and is played slightly softer, as
is every other hi-hat stroke.
We can observe several things from the trajectories. Firstly, the markers on the
hand knuckle are slightly early in phase in the preparatory movements, meaning that
the hand is moving up before the drumstick for strokes both on hi-hat and on snare
drum, similar to the trajectory in Fig. 4. At the downstroke, however, the hand
reaches the minimum slightly after the stick. Secondly, the influence of the rebound
from the surface is seen as some oscillatory behavior of the stick markers after the
stronger impacts (every other stroke on the hi-hat). Because of the higher striking
1100
1000
900
Vertical displacement (mm)
800
700
600
500
400
300
200
100
0 0.5 1 1.5 2 2.5 3
Time (s)
Fig. 6 Vertical projection of stick and hand movement during the beat pattern played in Fig. 5. The
lines show the displacement of markers on the players’ right stick (red line) and knuckle (dash-
dotted line), while playing the hi-hat, and on the right stick (blue line) and knuckle (dashed line) for
a total of two strokes on the snare drum. At the bottom, the displacement of the left toe (cyan) on
hi-hat pedal and right toe (magenta) on bass drum pedal can be seen. The stick markers are
positioned approximately 10 cm from the tip of the drumstick. The lead of the hand can be seen
in the phase differences of the maximum vertical displacement, whereas the minima of the knuckle
occurs slightly after that for the stick. Oscillations of the stick after stronger hits on both instruments
can be seen
velocity on the emphasized strokes, the stick would bounce up to a greater height if
not held by the hand. The hand cannot change direction as fast as the stick, and
therefore we can observe the fluctuations in stick height before or during the
following preparatory movement. For the softer, unaccented strokes, the stick
rebound is simply incorporated in the preparatory movement for the following
stroke.
The rebound also gives a welcome assistance when the player needs to move
swiftly between different parts of the drum set. Fast transitions between a drum and
cymbal at one end of the set to another at the other end mean that a player needs to
cover distances of more than a meter. Initiating the transition movement as fast as
possible is necessary in order to arrive on time. Indeed, players can start the
preparatory movement for a following stroke already during the ongoing one
(Dahl 2004).
Quite specific to drum set playing is that a player is also playing with both feet,
using pedals with springs that move the hi-hat cymbals up and down, where another
pedal is moving a beater (of felt or wood) in contact with the drumhead of the bass
drum. A spring in the pedal positions the beater some distance (adjustable, but
roughly 5–10 cm away) from the drumhead in its resting position. Depending on
16 S. Dahl
where the pedal is pressed, the leverage with which the beater is thrown toward the
drum will differ. The drummer’s foot is lifted in a preparatory position before
strokes, either with the heel still touching the base of the pedal (“heel down”) or
the whole leg slightly lifted with the toe part lightly on the pedal (“heel up”), and
some techniques use combinations of heel-toe motions for sequences of strokes.
Quick succession of strokes can also be done by moving the toe up the pedal during a
stroke.
Playing heel up can generate considerable power in the stroke as the whole leg
add to the force in the stroke, but it also requires the use of larger muscles in the thigh
and hip. When two bass pedals are played at fast rates at some length, as is common
in several genres of metal music, both legs of the player are moving rapidly and here
both good sense of balance and stamina are needed. Fastest rate reported for double
pedal playing is about 1030 beats in 1 min, or roughly over 117 ms between strokes
per foot. Also for more standard drum patterns, keeping four limbs drumming for an
extended period of time is a vigorous activity and has been estimated to have the
same metabolic demand as running 8 km/h and other sports such as ice hockey and
volleyball (De La Rue et al. 2013). Sadly, overstraining and lower limb dystonia
related to pedal playing has also been reported (chapter “▶ Functional Dystonias”)
(Lee and Altenmüller 2014; Rosset-Llobet et al. 2012).
Conclusion and Future Directions
In many ways, the playing of percussion is to produce discrete events, sounds that
have brief attacks and often cannot be modified once initiated. Brief as the sounds
may be, they are linked together by continuous motion. Players may initiate prepa-
ratory motions as early as during an ongoing stroke in order to deliver the following
stroke on time. The feedback and rebound from the instrument also affect the
preparation for the next coming stroke, something that skilled players can take
advantage of.
Drumming movements are fast; in particular certain instruments and beaters
produce high accelerations and contact forces. This poses a challenge when studying
movements and interaction between player and instrument, both in terms of required
sampling rates for optical tracking but also in the attachment of markers and sensors
that should stay fixed without obstructing playing or falling off during strokes. Hand
drumming movements, without beaters, may not produce the same striking veloci-
ties, but placing markers and sensors can be even more tricky because of the different
interaction with the instrument. Another challenge for future studies to tackle is that
the human hand is a complex structure that cannot easily be modeled as a rigid
segment during movement (chapter “▶ Rigid Body Models of the Musuloskeletal
System”). This complexity is naturally true for both hand drumming and with a
beater, and to obtain detailed measurements of grip forces during playing is yet
another challenge. Indeed, knowing more on the interaction and forces on the hand
could prove helpful to avoid injuries and movement-related disorders for players.
Something that has been out of the scope of this particular chapter are the
movements that are not directly related to producing strokes but still may be
important for timekeeping, regulation, or communication with co-performers. Sev-
eral studies have shown that percussionists’ striking movements can convey visual
information on expression and timing to observers. (The interested reader may read
more in other sources, e.g., Dahl et al. 2010; Schutz 2016). In general, the link
between movement control and sound production needs to be further investigated,
and with respect to the movement studies of rhythms on real musical instruments,
there is a world to explore and much more to be done. With the faster, more portable
(and more affordable) technology now available, it should be possible to measure
players during more realistic tasks and settings. In particular much percussion
playing has a strong element of improvisation, where rhythms may enfold and be
elaborated in relation (or contrast) to the base pattern. Spontaneous drum breaks
require movement strategies that allow fast actions executed after minimal time for
motor planning. As noted in the “Introduction,” keeping perfect isochrony or
striking fast is not equal to musical excellence. However, a skilled musician has
acquired a technique that enables the temporal control needed for the patterns that
she or he wants to play. We may be in awe of the impressive technical playing skills a
performer have, but it is how the rhythm makes us move or how the melody makes
us smile that really is the key to their success.
Acknowledgments The author would like to thank Marcus Lewin for his patience and profes-
sional playing and allowing me to use the pictures and data for several of the figures in this chapter
and also to Minho Song and Diana Kayser at fourMs, University of Oslo, for assisting with the
recordings.
Cross-References
▶ Body Movements in Music Performances – on the Example of Clarinet Players

▶ Hitting and Pitching
▶ Movement Analysis of the Golf Swing
▶ Movement Efficiency in Piano Players
▶ Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
▶ Rigid Body Models of the Musuloskeletal System
▶ Surface EMG
References
Burger B, Toiviainen P (2013) Mocap toolbox – a matlab toolbox for computational analysis of
movement data. In: Proceedings of the sound and music computing conference, Stockholm.
KTH Royal Institute of Technology
18 S. Dahl
Dahl S (1997) Spectral changes in the tom-tom related to striking force. In: Speech music and
hearing quarterly progress and status report 1. KTH, Department of Speech, Music and Hearing,
Royal Institute of Technology, Stockholm
Dahl S (2000) The playing of an accent – preliminary observations from temporal and kinematic
analysis of percussionists. J New Music Res 29(3):225–233
Dahl S (2004) Playing the accent – comparing striking velocity and timing in an ostinato rhythm
performed by four drummers. Acta Acust United Acust 90(4):762–776
Dahl S, Altenmüller E (2008) Motor control in drumming: influence of movement pattern on
contact force and sound characteristics. In: Proceedings of acoustics’08 international meeting
Acoustical Society of America, ASA, the European Acoustics Association, EAA, and the Socité
Francaise d’Acoustique, SFA, pp 1489–1494
Dahl S, Bevilacqua F, Bresin R, Clayton M, Leante L, Poggi I, Rasamimanana N (2010) Move-
ments and analysis of drumming. In: Leman M, Godøy RI (eds) Gestures in performance.
Dahl S, Großbach M, Altenmüller E (2011a) Effect of dynamic level in drumming: measurement of
striking velocity, force, and sound level. In: International conference of the forum acusticum.
Danish Acoustical Society, Aalborg, Denmark, pp 621–624
Dahl S, Großbach M, Altenmüller E (2011b) Good playing practice when drumming: influence of
tempo on timing and preparatory movements for healthy and dystonic players. In: International
symposium on performance science 2011. pp 237–242
Danielsen A, Waadeland CH, Sundt HG, Witek MA (2015) Effects of instructed timing and tempo
on snare drum sound in drum kit performance. J Acoust Soc Am 138(4):2301–2316
De La Rue SE, Draper SB, Potter CR, Smith MS (2013) Energy expenditure in rock/pop drumming.
Int J Sports Med 34(10):868–872
Famularo D (1999) It’s your move: motions and emotions. Alfred Music Publishing, Coram
Friberg A, Sundström A (2002) Swing ratios and ensemble timing in jazz performance: evidence for
a common rhythmic pattern. Music Percept Interdiscip J 19(3):333349
Fujii S, Oda S (2009a) Effect of stick use on rapid unimanual tapping in drummers. Percept Mot
Skills 108(3):962–970
Fujii S, Oda S (2009b) Effects of stick use on bimanual coordination performance during rapid
alternate tapping in drummers. Mot Control 13(3):331–341
Fujii S, Kudo K, Ohtsuki T, Oda S (2009a) Tapping performance and underlying wrist muscle
activity of non-drummers, drummers, and the world’s fastest drummer. Neurosci Lett 459
(2):69–73
Fujii S, Kudo K, Shinya M, Ohtsuki T, Oda S (2009b) Wrist muscle activity during rapid unimanual
tapping with a drumstick in drummers and nondrummers. Mot Control 13(3):237–250
Fujii S, Kudo K, Ohtsuki T, Oda S (2010) Intrinsic constraint of asymmetry acting as a control
parameter on rapid, rhythmic bimanual coordination: a study of professional drummers and
nondrummers. J Neurophysiol 104(4):2178–2186
Fujii S, Hirashima M, Kudo K, Ohtsuki T, Nakamura Y, Oda S (2011) Synchronization error of
drum kit playing with a metronome at different tempi by professional drummers. Music Percep
Interdiscip J 28(5):491–503
Fujisawa T, Miura M (2010) Investigating a playing strategy for drumming using surface electro-
myograms. Acoust Sci Technol 31(4):300–303
Hellmer K, Madison G (2015) Quantifying microtiming patterning and variability in drum kit
recordings. Music Percept Interdiscip J 33(2):147–162
Honing H, De Haas WB (2008) Swing once more: relating timing and tempo in expert jazz
drumming. Music Percept Interdiscip J 25(5):471–476
Jones R, Lagrange M, Schloss A (2007) A hand drumming dataset for physical modeling. In:
Proceedings of the international computer music conference (ICMC), vol 2. International
Computer Music Association (ICMA), Copenhagen, pp 379–381
Kinoshita H, Furuya S, Aoki T, Altenmüller E (2007) Loudness control in pianists as exemplified in
keystroke force measurements on different touches. J Acoust Soc Am 121(5):2959–2969
Lee A, Altenmüller E (2014) Heavy metal curse: a task-specific dystonia in the proximal lower limb
of a professional percussionist. Med Probl Perform Art 29(3):174–176
Madison G (1992) Drumming performance with and without clicktrack-the validity of the internal
clock in expert synchronisation. In: Proceedings of fourth workshop on rhythm perception &
production. L’Université René Descartes, Paris & Le Centre National de la Recherche
Scientifique, Bourges, pp 117–122
Madison G, Karampela O, Ullen F, Holm L (2013) Effects of practice on variability in an
isochronous serial interval production task: asymptotical levels of tapping variability after
training are similar to those of musicians. Acta Psychol (Amst) 143(1):119–128
Miura M (2005) Inter-player variability of a roll performance on a snare-drum performance. In:
International conference of the forum acusticum. Budapest, pp 563–568
Moeller SA (1956) The Moeller book. Ludwig Music Publishing Co, Cleveland OH
Polak R, London J, Jacoby N (2016) Both isochronous and non-isochronous metrical subdivision
afford precise and stable ensemble entrainment: a corpus study of Mailan jembe drumming.
Front Neurosci 10(285). doi:10.3389/fnins.2016.00285
Repp BH (2005) Sensorimotor synchronization: a review of the tapping literature. Psychon Bull
Rev 12(6):969–992
Repp BH (2010) Sensorimotor synchronization and perception of timing: effects of music training
and task experience. Hum Mov Sci 29(2):200–213
Repp BH, Su YH (2013) Sensorimotor synchronization: a review of recent research (2006–2012).
Psychon Bull Rev 20(3):403–452
Rosset-Llobet J, Fàbregas-Molas S, Pascual-Leone A (2012) Drummer’s lower limb dystonia. J
Neurol 259(6):1236–1237
Schutz M (2016) Lessons from the laboratory: the musical translation of scientific research on
movement. In: Hartenberger R (ed) The Cambridge companion to percussion. Cambridge
University Press, Cambridge, pp 267–280
Shivas AA (1957) The art of the tympanist and drummer. Dobson Books, London
Sternad D, Dean WJ, Newell KM (2000) Force and timing variability in rhythmic unimanual
tapping. J Mot Behav 32(3):249–267
Tindale AR, Kapur A, Tzanetakis G, Driessen P, Schloss A (2005) A comparison of sensor
strategies for capturing percussive gestures. In: Proceedings of the 2005 conference on New
interfaces for musical expression. National University of Singapore, Vancouver, BC, Canada,
pp 200–203
Trappe W, Parlitz D, Katzenberger U, Altenmüller E (1998) 3-D measurement of cyclic motion
patterns in drummers with different skill. In: Proceedings of the fifth international symposium
on the music perception and cognition (ICMPC9). Chattanooga, pp 97–99
Waadeland C (2006a) The influence of tempo on movement and timing in rhythm performance. In:
Baroni M, Addessi AR, Caterina R, Costa M (eds) Proceedings of 9th international conference
on music perception and cognition (ICMPC9). Bologna, p 29
Waadeland CH (2006b) Strategies in empirical studies of swing groove. Stud Musicol Nor
32:169–191
Wagner A (2006) Analysis of drumbeats: interaction between drummer, drumstick and instrument.
KTH Royal Institute of Technology, Stockholm
Functional Movement Analysis in Dance
Andrea Schärli
Abstract
Functional movement analysis is often adopted to better understand – and hence
optimize – complex movement skills of athletes. This chapter aims to uncover the
hitherto untouched potential of functional movement analyses for dancers. Both
sports and dance involve complex movement skills acquired and refined over
years of practice and training. The difference between dance and sports lies in the
ultimate goal of this skill development: the dancer aims to “maximize” his or her
aesthetic effect, whereas the athlete regularly aims to maximize some objective
performance criterion. Functional movement analysis was originally applied to
sports, like gymnastics and athletics, and focused mainly on the mechanical
principles underlying top-level sports movements. The deepened mechanical
understanding has helped athletes improve many a gymnastic and athletic skill.
The scope of functional movement analysis has broadened in recent years to also
include the athlete performing a movement and the context in which this takes
place. Accordingly, the analysis no longer exclusively focuses on the mechanics
of a movement but further considers anatomical, physiological, and psycholog-
ical factors. The present chapter aims to show that dance can benefit greatly from
functional movement analysis, both in its original narrow conception and partic-
ularly in its current broader one. This claim rests on the central argument that
maximizing aesthetic effects and maximizing motor performance are often very
closely related in dance, so that many fruits of functional movement analysis
currently only reaped by athletes are there for the taking for dancers as well.
A. Schärli (*)
Institute of Sport Science, University of Bern, Bern, Switzerland
e-mail: andrea.schaerli@ispw.unibe.ch

DOI 10.1007/978-3-319-30808-1_112-1
2 A. Schärli
Keywords
Dance teaching • Functional movement analysis • Dance performance • Chore-
ography • Aesthetics • Biomechanics • Ballet • Breakdance • Sub-actions •
Modalities • Functional assignment
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Functional Movement Analysis: More Than Just Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
FMA Examples from Dance: Grand Jeté and Single Pirouette en Dehors from Ballet Technique
and Multiple Head Spins from Breakdance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
FMA in Dance Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
FMA in Dance Teaching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
Dance is the art of human movement. In this respect, dance is unique among the arts:
no other art form can pride itself with an instrument of such magnificent complexity,
and in no other art form does its maintenance and perfection require so much
knowledge and skill. Like athletes, dancers perform highly complex motor skills,
acquired over years of training and practice. Yet quite unlike professional athletes,
dancers spend relatively little time analyzing their movements from a scientific
perspective. Whereas athletes and their coaches are becoming more and more
scientific in approaching crucial motor tasks from psychological, physiological,
anatomical, and – importantly – biomechanical perspectives, dancers, dance
teachers, and choreographers owe little to the current state of the art in human
movement science.
On the one hand, this is understandable: the ultimate goals in sports can reason-
ably be summarized by the famous Olympic motto Citius, Altius, Fortius (Faster,
Higher, Stronger), and modern-day human movement science has a lot to say about
how to reach these objective performance goals. They are, however, not the ultimate
goals of the dancer as a performing artist. Although artists and philosophers alike
will probably debate the ultimate goals of dance as an art form for many centuries to
come, it is safe to say that they are in the realm of aesthetics and expression rather
than mere objective performance. Whereas optimizing running speed, jumping
height or bodily strength is – or in recent decades has become – a science, optimizing
aesthetics is much more elusive, indeed more an art than a science.
On the other hand, it cannot be denied that there is a clear relationship between a
dancer’s objective performance and its aesthetic effect (e.g., Angioi et al. 2009).
Take the grand jeté, one of the most magnificent leaps in ballet. The dancer jumps off
one foot, performs a full leg split in midair, and lands on the other foot – to a great
Functional Movement Analysis in Dance 3
aesthetic effect. It is clear that this effect depends, at least in part, on parameters such
as jumping height and split amplitude – objective performance measures that
modern-day sports science is all about. It is precisely due to the relationship between
aesthetic effect and objective performance that an occasional switch from an artistic
to a scientific perspective would greatly benefit the professional and even the
recreational dancer. The present chapter will take just such a scientific perspective
and explore the benefits of functional movement analysis (FMA) for dancers.
State of the Art
What is FMA? Originally conceived in the 1970s by German sport scientist Ulrich
Göhner (Göhner 1979), FMA has been developed further in recent years into an
“action-oriented FMA” approach (Göhner 1992; Hossner et al. 2015). The differ-
ences between a mere movement and an action are that the latter, unlike the former,
is (1) goal-directed, (2) requires an actor (i.e., a performer), and (3) is performed in a
particular real-world context. Accordingly, action-oriented FMA focuses on a move-
ment task, performed by a performer, with a particular goal, in a particular context –
rather than a movement task per se. Hence, it is (1) the movement task itself, (2) the
performer performing it, (3) the task’s goal, and (4) the task’s real-world context that
are explicitly considered in action-oriented FMA.
Looking first at the movement task itself, FMA seeks to analyze a complex action
– say the Fosbury flop high jump in athletics – into its constituent parts. Those parts
are themselves actions, but simpler, and usually of shorter duration. Hossner et al.
(2015) label these parts sub-actions. This analysis into sub-actions may be summa-
rized by three questions:
1. What sub-actions must be performed?

2. How must they be performed?
3. Why must they be performed in this way?
Regarding question 1, an obvious sub-action of the Fosbury flop would be the

arching of the back while the body clears the bar. Another is the takeoff, which
precedes the aforementioned back arching. Another would be the typical curved
approach, which precedes the takeoff. What all of these sub-actions have in common
is that they contribute to the overall goal of jumping as high as possible: they are
functional parts. Moreover, they contribute to this goal in their own unique way,
which brings us to question 2.
Question 2 on the how of the performance deals with so-called modalities of each
sub-action. The modality of a sub-action refers to the unique manner in which that
sub-action contributes to the overall goal. For example, in a well-executed Fosbury
flop, the approach should be fast and curved, and the take-off foot should be placed
well in front of the body’s center of gravity (CG). Of course, a modality only makes
sense if we know why a sub-action must be performed in that way. Thus, an adequate
FMA of the Fosbury flop should result in an understanding of why all its sub-actions
4 A. Schärli
– performed perfectly – contribute to the overall goal of jumping maximally high.

This brings us to the third and final question in FMA.
Why does the approach in the Fosbury flop need to be fast? It should be fast in
order to generate the kinetic energy needed to jump high. The take-off foot should be
placed ahead of the body’s CG in order to direct the kinetic energy of the approach
upward. After takeoff, the fate of the body’s CG is sealed: the jumper can no longer
push off anything to make his CG go higher. What he can do in flight, however, is to
reach each body part as high as possible relative to the CG exactly when it crosses
the bar’s vertical plane. He does this in order to clear the highest possible bar, given
the CG trajectory achieved by his approach and takeoff. This is why he arches his
back when his hips cross the bar: to push his buttocks up relative to his CG – and
hence relative to the bar.
So far, we have performed the first phase of a FMA of the Fosbury flop or at least
a rough sketch of it. We analyzed an action as a solution for a specific movement task
and split this action into its functional constituents, its sub-actions. We thus deep-
ened our understanding of the movement task in light of its goal. Consequently, we
uncovered different aspects of the task that may be optimized through training and
practice: we could work on running speed in order to make the approach faster; we
could work on explosive leg strength to make the takeoff more buoyant; or we could
work on flexibility of the lumbar spine to bring the buttocks up higher. According to
our analysis, all of these should result in a higher jump.
Yet FMA does not stop here. As pointed out above, FMA not only addresses the
sub-actions, modalities, and functional assignments of a particular movement task; it
also explicitly considers the performer and the real-world context in which the action
takes place. With respect to the performer, optimal high jump technique might look
different for a young child as compared to a professional athlete given that their body
proportions are so radically different. With respect to the action’s context, the
approach of the Fosbury flop would most certainly look different if it were
performed on asphalt rather than tartan. Rules form another crucial piece of action
context: in high jump, one important rule is that the athlete must take off from one
leg. If he were allowed to take off from two legs, the optimal high jump and its
sub-actions would look radically different. Objects and devices contribute additional
components to define the action context: the device that serves as the performance
criterion in the high jump (i.e., the vertical posts with the small platforms and the bar
lying on top of them) clearly affects what a jumper can afford to do in order to meet
said criterion. If the bar were lighter and shorter and lay on smaller platforms, the
ideal jump would look different. If the jumper landed in a sand pit – as were past
circumstances– rather than on a soft matt, in order to prevent injuries, the ideal jump
would also look different. Hence, many more aspects of the performer and the action
context can and should be considered in a comprehensive FMA of the high jump. All
of these aspects can be thought of as constraints that affect the optimal movement
solution of a given task (see Fig. 1).
Fig. 1 Constraints affecting

the optimal movement
solution for a given task
(Hossner et al. 2015, p. 5; with
permission)
Functional Movement Analysis: More Than Just Mechanics
Our analysis of the Fosbury flop so far exclusively concerns the mechanics of the
task. Our answers to the questions of what, how, and why sub-actions were
performed were of a mechanical nature, focusing on speed, curvature, or kinetic
energy. In its original conception, FMA focused primarily on movement tasks whose
optimization was first and foremost a mechanical problem. This is often the case in
sports: maximizing jump height (in the high jump) is mainly a mechanical puzzle, as
is maximizing throwing distance (in javelin throwing) or running speed (in the
100 m dash). In order to achieve maximum performance in these three disciplines,
it is very important to optimize the mechanics of the movements involved.
However, even in sports, there is more than just mechanics. The consideration of
the performer and the action’s real-world context over and above the actual move-
ment itself naturally led Hossner et al. (2015) to consider nonmechanical consider-
ations in task optimization. Such nonmechanical considerations could, for example,
be of a tactical nature, as in football where the penalty shooter refrains from a
biomechanically optimal approach in order to hide his intended direction of ball
strike. Another example of nonmechanical optimization would be the perceptual
optimization of a pistol shooter who might adopt a posture that is not mechanically
optimized for holding the pistol as steady as possible, but rather for keeping his head
as quiet as possible to ensure optimal visual location of the target. How one should
6 A. Schärli
go about optimizing such nonmechanical performance criteria in sports is an ongo-

ing challenge in FMA today.
In dance, the aesthetical criterion should be treated as just such a nonmechanical
criterion. At first it might seem that the aesthetic criterion renders FMA all but
useless in dance. Whereas it is possible to assess how and why the curved approach
in the high jump contributes to the ultimate goal of jumping high, it seems much
more difficult to assess how and why the approach in the grand jeté contributes to the
ultimate goal of “being maximally beautiful.” Yet because, as noted, a higher grand
jeté will generally be considered more beautiful than a lower one, the mechanical,
objective criterion of height may serve – at least in part – as a useful substitution for
the nonmechanical, subjective criterion of aesthetics. This useful substitution again
illustrates the pivotal argument of the present chapter that maximizing aesthetic
effect and maximizing objective performance are often very closely related in dance.
Hence, despite the subjective nature of aesthetics, the scientific perspective of FMA
can still be very useful for dancers, their teachers, and choreographers.
In order to investigate whether the scientific lens of FMA indeed deepens our
understanding of common movements in dance and, consequently, may even stim-
ulate new ideas for optimizing their execution, we now shift to provide the reader
with some examples of FMAs from the world of dance.
FMA Examples from Dance: Grand Jeté and Single Pirouette en

Dehors from Ballet Technique and Multiple Head Spins from
Breakdance
To consider the grand jeté (Table 1) as an exemplary dance movement task for an
in-detail FMA, we must first think about possible constraints affecting the choice of
sub-actions and modalities for an optimal task solution. In other words, we have to
define the goal of our movement since the optimal task solution will differ if the goal
of the dancer is to execute the grand jeté either maximally high or maximally far. In
the following, we will consider a standard grand jeté in which the jump travels
approximately one step ahead and should be high enough to allow for a 180 leg split
in the air. The execution of the grand jeté depends on the dancer’s individual
anatomy and abilities (i.e., length of extremities, strength in the legs, flexibility of
the hamstrings and hips, etc.). Hence, the optimal movement execution strategy will
differ for dancers with different body proportions, such as a young child, a male
professional, or a female beginner dancer. For our analysis, we consider a male
professional ballet dancer. He is to perform the grand jeté in traditional classical
ballet technique, which means that he needs to perform the jump with externally
rotated legs. No other dancers or objects are involved and the dancer performs the
jump on a wooden stage. The dancer wears tight clothes and ballet shoes, which do
not restrict his movements in any way. The dancer wants to improve his floating
illusion in the air and thus performs a FMA.
Secondly, after reflecting on relevant task constraints, we must list the crucial
sub-actions of the movement task at hand. In a grand jeté, as previously specified,
Table 1 FMA of a grand jeté (standard variant performed by a professional ballet dancer)
Sub-actions Modalities Functional assignment
... in order to reach an

Running... ...with 2-5 steps... optimal horizontal
speed for take-off.
... in order to transfer

horizontal to vertical
...diagnoally upwards
energy and in order to
from on leg in an
Pushing-off.. optimal (not maximal) increase the way in
turn-out postion... which force can be
produced during
pushing-off.
...with legs in split

...in order to produce
Flying... and arms up and to the
a floating illusion.
side...
...on one leg with toes

...in order to
touching the ground
decelarate the dancer
Landing... first, then foot rolling
vertically and
off, and knee
horizontally.
bending...
these sub-actions are running, pushing off, flying, and landing. Third, we must define
the modalities of each sub-action. How are the sub-actions performed? Running with
2–5 steps, pushing off diagonally upward from one leg with turned-out legs, flying
with legs split at 180 and the arms out to the side, and landing on the other leg with
toes grounding first and then consecutively bending ankle, knee, and hip joints.
8 A. Schärli
Table 2 FMA of a single pirouette en dehors (performed by a beginning ballet dancer)
Sub-actions Modalities Functional assignemnt
…in order to start the

turning motion as
…with the leg à la
vertical as possible
Preparing… seconde and body
and thus reducing the
upright… topple angle to a
minimum.
…in order to lengthen

Bending knees
… slowly… the path for torque
(demi plié)…
production.
…forcefully from …in order to produce

fourth position by an optimal rotational
quickly putting the impulse for one
Pushing-off
gesture leg into passé revolution and keep
position and bringing rotational inertia
the arms up… small.
…with body in
compact position and
…in order to keep
gesture leg in passé
turning speed constant
position, arms in front
Turning… and to stop the turning
of the body, head
to the front after one
turning around
revolution.
quickly to catch front
as soon as possible…
……in order to
increase rotational
…by opening arms
inertia and thus slow
and putting gesture
down the rotational
Landing… leg foot to the ground
speed and in order to
into required
land in balance and
position…
continue with the next
movement.
Fourth and finally, we must assign a function to each sub-action, which would then
round off our FMA of a grand jeté jump. Details of the FMA can be found in Table 1.
The so-called floating illusion (Laws 2002) of the grand jeté mainly depends on
well-coordinated in-flight leg movements: during the flight phase, split amplitude
increases during the upward phase and decreases during the downward phase to
prepare for landing. Maximal split amplitude is reached at the zenith of the CG’s
flight path. In this way, the flight path of the dancer’s trunk and head (presumably the
spectator’s main focus) is “flattened” relative to that of the body’s CG. Accordingly,
the trunk spends a relatively large portion of the total flight time close to its highest
point. For a little while, the trunk seems to move near horizontally through the air: it
seems to float. Of course, a longer flight time would further strengthen the desired
floating illusion. Functionally, optimizing such sub-actions as running and pushing
off should result in maximum flight time, much like in our high jump example.
Further details on the floating illusion phenomenon will follow in the next section.
Our next analysis is about a single pirouette (i.e., a rotation around the longitu-
dinal axes) from ballet technique (Table 2). Let us consider a beginner female dancer
performing the pirouette in the framework of a ballet class on a dance floor with
many other dance students around her and a mirror in front of them. The dancer has
the ability to balance on one leg and to perform simple turning movements on one
and two legs. We want to unravel why the ballet pupil falls out of the pirouette before
finishing one revolution and therefore conduct a FMA. Details can be found in
Table 2.
Low floor friction could be a cause for falling out of a pirouette (an error in the
sub-action turning). Accordingly, the dancer may choose to turn barefoot instead of
turning in socks or ballet shoes. Often, however, errors in a specific sub-action (here
turning) can be traced back to previous sub-actions. It could be that the dancer
prepares with the body’s long axis too far from vertical. Physically, starting a
pirouette with a topple angle of more than 3 from vertical does not allow for the
completion of even one revolution (Lott and Laws 2012). Moreover, it could be that
the push-off is too strong and/or not directed properly, resulting in a rotational
impulse that is too large and/or directed around an axis other than the body’s long
axis, making it difficult to control the turning motion properly.
Our last example of a functional movement analysis is a typical dance movement
from breakdance: the head spin (Table 3). Let us consider a professional breakdancer
performing multiple head spins in a dance studio on a sports floor wearing a hat.
Circled round him his team mates and the opposing crew of dancers who watch his
performance. Details of the respective FMA can be found in Table 3.
In order to produce multiple head spins, the breakdancer must maintain the body’s
long axis close to vertical and minimize rotational inertia by bringing arms and legs
toward the midline. To initiate the turn, he must (1) place his arms far apart to
facilitate a large rotational impulse, (2) twist his trunk and legs against the desired
turning direction to prepare for the turn, (3) forcefully “unwind” legs and trunk in the
10 A. Schärli
Table 3 FMA of a head spin (multiple-spin variant performed by a professional breakdancer)
Sub-actions Modalities Functional assignment
... in order to transfer

Shifting weight from
...with control... the weight to a
feet to head...
balanced head stand.
...in order to produce

... brusquely with legs the largest possible
Pushing off... twisted and hands far rotational impulse
apart... for the spinning
motion.
... in order to
...by pulling legs decrease rotational
Spinning around...
towards the midline... inertial and thus allow
for several turns.
...by lowering legs ...in order to enhance

into a freeze position rotational inertia and
Landing
(or a following decelerate the
movement)... spinning motion.
desired turning direction to generate a large rotational impulse, and, finally, (4) lift up
both hands. The reasons for a breakdancer falling out of the spin after only a few
rotations are very similar to those discussed for a dancer falling out of a pirouette:
low floor friction, non-vertical body axis, or poorly dosed and directed initial
impulse.
FMA in Dance Performance
Now that we have performed actual FMAs of a grand jeté and pirouette from ballet
technique and a head spin from breakdance technique, let us discuss the benefits of
such an investigation for dance performance. In dance, the movement requirements
have changed considerably over the centuries. Comparing a dancer from a French
court ballet in the seventeenth century to a ballet dancer of a professional company
today, one is impressed by the development toward “higher,” “faster,” and “more” –
similar to the development of sports performance. When a present-day professional
ballet dancer turns five pirouettes, it is crucial for him and his teacher or choreog-
rapher to understand the mechanics of turning. What position of the feet allow for the
production of the largest torque? How does the dancer position his arms and legs to
reach maximal angular velocity? It is often overlooked in dance that the rotational
impulse required for a pirouette can only be produced during the initial push-off
from the floor, a moment of crucial importance to the pirouette(s) that follow.
Turning speed, however, may be manipulated by changing the body’s mass distri-
bution around the rotational axis and, hence, its rotational inertia: placing more mass
closer to the axis will speed up the pirouette, whereas placing more mass further
away from the axis will slow it down. Thus, changing arm positions during a turn
will produce different aesthetical effects of which choreographers should be aware.
In addition, if a dancer wants to perform a turn with an increased number of
rotations, he benefits from placing his feet farther apart during preparation in order
to generate the greatest possible impulse during push-off. This holds true for the
breakdancer as well, who wants to spin on his head many times. He too must initially
generate a large rotational impulse, which is facilitated by wide, stable hand place-
ment and twisted leg position before push-off. Subsequently, in order to execute as
many turns as possible (an objective criterion) and hence have the greatest aesthetic
effect (the related subjective criterion), the breakdancer and the ballet dancer alike
must minimize the horizontal shear forces acting on their head and foot, respectively,
and adopt a posture that minimizes rotational inertia. Many lifts, turns, and jumps in
dance can be likewise improved aesthetically by maximizing performance in respect
to objective, mechanical criteria. Hence, means of performance enhancement should
benefit from the deep mechanical understanding contributed by FMA.
Interestingly, FMA may occasionally uncover an actual conflict between the
aesthetic requirements of a dance movement on the one hand and its mechanically
optimal execution on the other. In the grand jeté, for example, the traditional
aesthetic requires external rotation of the hips. However, the biomechanics of the
grand jeté dictate that performing it in a parallel position (i.e., without external hip
12 A. Schärli
rotation) would result in a higher jump (Laws 2002). This is because the parallel
position affords a larger flexion range in the ankle joint, allowing for a larger plantar
flexion range during push-off and hence increasing the upward impulse imparted on
the body.
Given that both a higher jump and greater hip turnout are aesthetically desirable,
the dancer is presented with a dilemma. He may comply with one or the other
criterion, but not with both at the same time. Of course, a dancer may adopt a
technique that constitutes a compromise between these conflicting goals. It is,
however, crucial for dancers, choreographers, and teachers to be aware of the
restrictive role of turned-out leg positions during jumps. The analytic mind-set
given by FMA creates just this kind of awareness and allows dance professionals
to break with tradition, which can be refreshing, exciting, and, of course, beautiful.
Another interesting aesthetical requirement of the grand jeté is the so-called
floating illusion discussed above. It should seem as if the dancer briefly defies
gravity and “floats,” leaving the typical parabolic path that he would usually follow
after jumping into the air. From a mechanical perspective, this illusion is produced
by specific changes of different body parts relative to the body’s CG, which, of
course, mandatorily follows a parabolic trajectory after takeoff. In this sense,
creating the floating illusion should involve similar well-aimed movements in midair
to those of the high jumper discussed earlier. In the same way that the high jumper’s
mid-air movements make him clear the bar despite his CG failing to do so, the ballet
dancer’s mid-air movements (i.e., splitting the legs to a maximum at the peak of the
CG’s flight path) make him seem to float despite his CG failing to do so. Here again,
a deeper mechanical understanding of the grand jeté leads to very specific guidelines
for its aesthetic optimization.
A notable added benefit of FMA in dance is that it may open up a realm of
hitherto unexplored choreographic possibilities: optimizing the approach of the
grand jeté with respect to jump height may bring about novel movement patterns.
For example, the dancer may consider the typical curved take-off approach of the
high jumper in order to jump high without jumping very far. Hence, in addition to
objectively optimizing aspects of existing dance forms (such as grand jeté height),
functional movement analysis may bring about radically novel movement patterns
(such as the curved approach for the grand jeté) that – precisely due to their
functional nature – may be aesthetically valuable in themselves.
Another area in which functional movement analysis should be beneficial is in
dance improvisation – an often adopted tool in many dance styles to discover new
movements. In improvisation, a task forms the starting point of the process and
dancers try to find solutions for this task. At the end of the process, the choreogra-
pher chooses the movement solutions he likes best. In this process of improvisation,
FMA might be a source of inspiration for the dancers and, when applied routinely,
may facilitate a “functional mindset” that dancers and choreographers adopt (con-
sciously or even unconsciously) when creatively searching for new dance
movements.
FMA in Dance Teaching
The requirements of many dance styles are markedly different. A great ballet dancer
is not necessarily a great breakdancer as the movement techniques and the physical
requirements are very different for these two styles. However, many basic movement
categories, such as turning, jumping, traveling through space, holding positions, and
shifting weight, are present across all dance styles, and their underlying functional
mechanisms are mainly the same. In addressing the functional similarities of these
overlapping movement categories, FMA represents a useful tool for creating transfer
of skills between different dance styles. Take as an example turning around the
vertical axis. As discussed in the context of the pirouette and the head spin above, the
basic movement pattern invariably starts with the production of a rotational impulse
by pushing off from the ground (or from a partner). During the turn, rotational
velocity varies with rotational inertia, which can be manipulated by altering leg or
arm positions. This is true whether we are talking about a ballet pirouette, a head spin
in breakdance, or a tango turn in ballroom dancing. FMA allows for an overarching
description of general dance skills and, therefore, opens the door to the transfer of
movement skills from one dance style to another.
Furthermore, as we now know, a functional approach not only considers the
movement itself but also the dancer, the movement goal, and the environment,
including other dancers, rules, and objects (see Fig. 1). The explicit consideration
of the dancer may lead to the observation that male and female dancers have different
body proportions, generally rendering the male dancer top-heavier compared to the
female dancer. Lott and Laws (2012) claim that this makes a male dancer less likely
to topple out of a pirouette simply because his physical build cause him to do so
more slowly. This difference is exacerbated in children who are even more
top-heavy. Such anatomical differences imply that the optimal movement solution
for a pirouette may be different for men, women, and children. There is, therefore, no
such thing as “the” perfect pirouette, and this fact is explicitly acknowledged
in FMA.
Beyond the discussion of FMA basics, Hossner et al. (2015) relate the FMA
approach to the concept of a task solution space: a space of appropriate task solutions
to replace the concept of a single optimal task solution. The space is multi-
dimensional because it is shaped by all the constraints (i.e., dimensions) that are
presented in Fig. 1. Where in this space the actual optimal task solution lies thus
depends on all these constraints. It is our contention that the concept of a task
solution space would help dance teachers detect and accommodate the different
constraints acting on their students (are they beginners or experts, male or female,
etc.) and hence properly define the optimal task solution for each individual in each
individual context. It would allow the teacher to better differentiate between move-
ments that lie outside the task solution space (i.e., movement errors), on the one
hand, and functional variations within the task solution space on the other. If a
movement does not resemble that of the professional dancer, this does not imply that
something is wrong. Only by grasping the underlying functional assignments of each
14 A. Schärli
sub-action can a teacher pinpoint errors clearly and in this way help the dancers most
efficiently.
Conclusion
Originally adopted to movements that can be measured objectively and analyzed

(mainly) mechanically, modern FMA expands its explanations to other disciplines
such as anatomy, physiology, or psychology. In dance, FMA must come to terms
with the decidedly nonmechanical and subjective aesthetic and expressive criteria.
This chapter argued that this is not as daunting as it may seem at first sight. The
argument revolved around the observation that there is a clear relationship between
objective mechanical parameters of complex dance movements on the one hand and
their subjective aesthetic and expressive effect on the other. Acknowledging this
relationship opens the door for dancers, choreographers, and dance teachers for a
deeper understanding of the complex movements they are performing, conceiving,
and teaching, respectively. This deeper understanding will help the dancer improve
performance in a more focused manner, one sub-action at a time. It will help the
choreographer develop novel movement ideas, through the occasional focused
search for optimal task solutions for sub-actions rather than the usual intractable
search for “beautiful movements.” It will help the teacher provide appropriate
feedback, fully acknowledging the individual student, the context in which the
movement is performed, and the accompanying task solution space. FMA thus
promises to offer a refreshingly new perspective for the dancer, choreographer,
and dance teacher alike.
Cross-References
▶ Jumping for Maximum Distance and Height

▶ Long Swing Manoewers in Gymnastics
▶ Motion Analysis as a Pedagogic Tool in Dance
▶ Motion Analysis as Preventive and Rehabilitative Measures in Dance Medicine
▶ Movement Analysis of Scull and Oar Rowing
▶ Somatic Practices: How Motion Analysis and Mind Images Work Hand in Hand
in Dance
▶ The Motor Solutions of Throws in Sports
▶ The Movement of Kicking a Ball
▶ The Segmental Movements in Front Crawl Swimming
References
Angioi M, Metsios GS, Twitchett E, Koutedakis Y, Wyon M (2009) Association between selected
physical fitness parameters and esthetic competence in contemporary dancers. J Dance Med Sci
Off Publ Int Assoc Dance Med Sci 13(4)115–123
Göhner U (1979) Bewegungsanalyse im sport (movement analysis in sports). Hofmann, Schorndorf
Göhner U (1992) Einführung in die Bewegungslehre des Sports: Teil 1: Die sportlichen
Bewegungen. Hofmann Verlag, Schorndorf
Hossner E, Schiebl F, Gohner U (2015) A functional approach to movement analysis and error
identification in sports and physical education. Front Psychol 6:1339. doi:10.3389/
fpsyg.2015.01339
Laws KL (2002) Physics and the art of dance: understanding movement. Oxford University Press,
New York
Lott MB, Laws KL (2012) The physics of toppling and regaining balance during a pirouette. J
Dance Med Sci Off Publ Int Assoc Dance Med Sci 16(4)167–174
Somatic Practices: How Motion Analysis
and Mind Images Work Hand in Hand
in Dance
Sarah Whatley
Abstract
Somatic Practices are body-based movement practices that foreground self-
awareness and a first person experience of moving. Increasingly, somatic prac-
tices are informing how dance is taught, created, and performed with many
dancers turning towards somatics to ensure a healthy and holistic approach to
dance. Several somatic practices draw on imagery as a source for moving, for
stimulating a more sensorial engagement with movement and to encourage a
sense of moving “naturally” and with respect for the “natural environment.”
When somatic practices and the imagery that is important for many of these
practices are coupled with motion analysis tools, the necessary processing of
movement often requires an intervention that can disrupt the “natural” sense of
moving somatically. This processing can thus appear to be at odds with a somatic
approach to dance. However, there are many examples where motion analysis and
mind images do work hand in hand in dance and can generate exciting new
insights to the production, teaching, and making of dance. It is this intersection
between somatic principles, imagery, and motion analysis tools that is the focus
for this essay, which discusses projects that have explored and exploited the
intersection between motion analysis, imagery, and somatic practices.
Keywords
Dance • Somatic practices • Imagery • Motion capture • WhoLoDance • Ruth
Gibson • Hologram
S. Whatley (*)
Centre for Dance Research, Coventry University, Coventry, UK
e-mail: s.whatley@coventry.ac.uk

DOI 10.1007/978-3-319-30808-1_113-1
2 S. Whatley
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Somatic Practices in Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Imagery and Dance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Mind Images and Motion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Motion Analysis in Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Introduction
Somatic practices are body-based movement practices that foreground self-

awareness and a first person experience of moving. Increasingly, somatic practices
are informing how dance is taught, created, and performed with many dancers
turning towards somatics to ensure a healthy and holistic approach to dance. Several
somatic practices draw on imagery as a source for moving, for stimulating a more
sensorial engagement with movement and to encourage a sense of moving “natu-
rally” and with respect for the “natural environment.” When somatic practices and
the imagery that is important for many of these practices are coupled with motion
analysis tools, an interesting friction arises that might seem to intervene in the
premise on which somatic practices rely. This is because the introduction of tech-
nology that requires the processing of movement often requires an intervention that
can disrupt the “natural” sense of moving somatically, whether for analyzing the
biomechanical properties of movement or to generate visualizations for aesthetic
purposes. This processing can thus appear to be at odds with a somatic approach to
dance. However, there are many examples where motion analysis and mind images
do work hand in hand in dance and can generate exciting new insights to the
production, teaching, and making of dance. It is this intersection between somatic
principles, imagery, and motion analysis tools that is the focus for this essay.
Beginning with an overview of somatic practices and their influence on dance
more generally, the discussion will then examine the role of imagery in dance and
how imagery has played a role in motion analysis, with specific reference to motion
capture. Reference will be made to two projects that have explored and exploited the
intersection between motion analysis, imagery, and somatic practices, and which
have contributed to revealing more about the subtleties and the “hidden” complex-
ities of dance.
State of the Art
For many years, dance and somatic-informed dance practices have been a source for
experts who are interested in the complexity of human motion and expression. But
dance is also a broad category and takes on myriad roles: as art form, a traditional
Somatic Practices: How Motion Analysis and Mind Images Work Hand in Hand in. . . 3
cultural practice, a route to enhanced health and well-being, or as a recreational

activity. There have been many projects that have brought together experts to explore
the use of new technological tools for multimodal analysis of dance movement for
different purposes including the teaching of dance, for biomechanical purposes as
well as for performance creation. Different areas of dance have been covered by a
number of projects and many have focused in particular on Motion Capture, Dance
Notation, and Animation. For example, Antonio Camurri has developed a four-layer
conceptual model of automated analysis of dance (2003), as well as modules for real
time analysis of movement and mapping, and an interactive sonification component
to transform a movement quality in the auditory domain (2005). Starting from
low-level physical measures of the video-tracked whole-body silhouette, Camurri
et al. identified motion features, such as the overall amount of motion computed with
silhouette motion images, the degree of contraction and expansion of the body, or the
motion fluency computed on the basis of the changes of the overall amount of
motion over time. On the basis of these motion features, they were able to distinguish
between four emotional performances of a dance sequence (anger, fear, grief, and
joy) by four dancers. This framework of vision- based bodily expression analysis
was used for a number of multimodal interactive applications, in particular in
performing arts, and mainly includes support vector machine techniques for real-
time affective classification (2005, 2007). Kapur et al. (2005) used full-body skeletal
movement data (obtained with the VICON motion capture system on five partici-
pants) to distinguish automatically between the same four basic emotional states, and
Camurri et al. developed a qualitative approach to human full-body movement for
affect recognition (2003).
Motek Entertainment1 and its sister company Motek Medical have developed
several in-house tools for real-time blending of motion capture data. Motek have also
been active in the last decade in progressing the building of systems with multimodal
input and output capabilities. Similarly, from a technological standpoint, holographic
displays are becoming faster and less expensive to produce, and research on elec-
tronic holography has yielded systems which provide full color, moving holographic
images. These developments have the potential to support the teaching and learning
of dance, as will be discussed later. Other projects have worked on motion capture
systems to support the teaching of dance. For example, Chan et al. (2011) have
described a new dance training system that they have developed based on motion
capture and virtual reality technologies that can evaluate the difference between the
learner and the virtual teacher. Their system claims to be able to carry out “a more
comprehensive analysis of the user’s whole body motion in order to let him/her
know, which body parts are moved incorrectly” (2011, 187) and can provide
feedback by comparing two motions, producing, they assert, a fun and motivating
learning experience.
The motion analysis of dance for biomechanical purposes has also spawned
interest in motion capture for other physical activities. For example, James Shippen’s
1
See http://www.motekentertainment.com/
4 S. Whatley
work in using a VICON motion capture lab for creating a software based on
engineering principles to analyze the movement of the human body, and analyze
how it reacts to different loads and postures, was initially developed to examine the
effect of dance on the musculoskeletal system for injury prevention. With a focus on
Irish dancers, using an optical tracking system and floor mounted force plates,
Shippen researched the kinematics, kinetics, and internal muscle loadings in order
to improve performance and reduce the risk of injuries (Shippen and May 2010).
Working with the dancers and their teachers, Shippen noted the importance of
imagery in the dancers’ rehearsal process and created a color coded graphics for
visualization of individual muscle activity and tabulated muscle loading data for
easy interpretation of the results. More recently he transferred this knowledge to
examining the physical aspects of the most common horticultural activities, using
the Xsens motion capture system (Shippen et al. 2015).
A very different project is Motion Bank,2 which has developed a range of
computer-aided visualizations of dance, and the structures that underpin dance
works, for arts education and interdisciplinary research. The project brought together
researchers, leading dance choreographers, designers, educators, and computer
scientists. The aim was to “explore how digital technology can be uniquely applied
to the challenge of documenting, analysing, notating/annotating and presenting
dance” (2011, 12) by archiving a number of choreographers’ conceptual approaches
along with video recordings and three-dimensional data documenting the perfor-
mances and the depictions created by the designers. Incorporating different motion
analysis tools including Kinect and Motionbuilder to visualize different aspects of
the choreographer’s work, a number of digital scores have been created as a result of
this interdisciplinary design process.
A leading expert in motion analysis and motion capture technology Kim Vincs,
based at Motion.lab Deakin, Australia, has conducted a number of projects that have
contributed to furthering knowledge about the affordances of motion analysis for
dance research and practice. In a recent project (2014), Vincs and Barbour employ
principal component analysis (PCA) techniques in conjunction with motion capture
to uncover more about variability and consistency within and across different dance
practices. Their findings are very interesting for a number of reasons. First, they
advise careful consideration in using PCA because as “dance can potentially com-
municate across both high- and low-level features, analysis that focuses solely on
high-level features to minimise computational complexity and optimise recognition
risks eliminating key artistically and culturally valent aspects of the movement”
(2014, 65). They then argue that their data suggests that the organization of dance
into different practices “operate more cohesively on a conceptual level than it does at
the microlevel of movement patterning made visible by PCA analysis” (2014, 75)
challenging “our culturo-aesthetic assumptions about how certain kinds of dance are
produced” (2014, 75). Their research has thus contributed to thinking about the place
of different statistical analyses in the analysis of dance, suggesting that it is “not only
2
See Motion Bank website: http://motionbank.org/
larger, locomotor movement that elicits more variable movement patterning” (2014,
74) and by extension, variability might therefore be an aesthetic choice (made by the
dancer) as well as a biomechanical one (2014, 74).
Somatic Practices in Context
Somatic practices have a long lineage. Beginning as far back as the late nineteenth
and twentieth centuries in Europe and America, somatics can be seen to have its
roots in the work of Francois Delsarte, an actor and opera singer (1811–71), who
created a system in the mid-nineteenth century based on his observations of human
interaction that aimed to connect the inner emotional experience with a systematized
set of gestures, speech, and movements. His work had a significant influence on
modern dance as it was emerging, including the early days of somatics. Many of the
underlying theories that have informed somatic systems originated in Germany
around that time – movement teachers Elsa Gindler, Heinrich Gimmler, Bess
Mensendieck, Ilse Middendorf, and Marion Rosen were all important in constructing
the foundation for somatic work.
Almost a century later and in 1976, Thomas Hanna, a US practitioner and
philosopher, named the field Somatics (1980), recognizing a number of embodied
disciplines that collectively share a focus on sensory awareness. Contemporary
practices that are generally regarded as promoting somatic principles include Alex-
ander Technique, the Feldenkrais Method ®, and particularly Feldenkrais’ Awareness
Through Movement Method, Body Mind Centering ®, Skinner Releasing Tech-
nique™, and Authentic Movement, amongst others, all with their own innovators
and legacy of teachers. As examples of reflection-in-action, these techniques facil-
itate self-awareness and engage with what Michel Foucault termed “Technologies of
the Self” (1988). Many are intended to be used “by the self on the self” and maintain
the goal of educating attentional skills. Some of these contemporary somatic prac-
tices have been influenced by eastern philosophy and body practices including yoga,
martial arts, Shiatsu, Chi Kung, and Buddhist meditation. Along with Kinesiology
education, they support a principle of developing mind through body, so that the
training of one creates knowledge within the other.3
Somatics are thus variously described as centered in first-person sensory and
body-based techniques, experiences, and methodologies, and self-reflexive tech-
niques structured to transform one’s experience of the self in the world. They
promote the unity of mind and body, and the body as contextual and fluid. They
give rise to theories of embodiment4 and encourage verbalization, questions, and
3
For a valuable and extended account of the history of somatic practices as they have developed and
in particular formed a relationship with HCI in thought and practice, see Thecla Schiphorst’s PhD
thesis (2009): http://www.sfu.ca/~tschipho/PhD/PhD_thesis.html
4
There are many and varied theories of embodiment; in broad terms, embodiment refers to how our
bodies and active experiences shape how we think, feel, and perceive.
6 S. Whatley
comments. Participants are encouraged to speak from their movement explorations,

to develop a consciousness of their bodily practices, and an awareness of their
movement patterns (Harbonnier-Topin and Simard 2015). The appeal of the body-
based practices of somatics to dancers has grown through a long and interconnected
history, resulting in a close relationship, particularly with contemporary dance
techniques.
Many dancers and somatic movement practitioners will consider themselves
philosophers as much as movers. Recent writings (Fraleigh 2015; Williamson
2016) have demonstrated not only how established philosophy (particularly phe-
nomenology) has inspired and informed somatic movement practices, but the mov-
ing, thinking, and writing contributes to a somatic philosophy that forms its own
particular way of being and moving in the world. Hence it may be useful to think of
somatic work as a philosophy rather than as a philosophy of somatic practices. Many
dancer scholars draw substantially on philosophical frameworks and particularly
phenomenology, and the philosophers most closely aligned with its origins, Edmund
Husserl (1859–1938) and Merleau-Ponty (1908–1961). Phenomenology promotes
the belief that meaning comes from our experiences of phenomena, fosters the idea
that consciousness is based in the lived body, and emphasizes the importance of
bodily knowledge. And as philosopher Manen argues, as a methodology, phenom-
enology looks at how a thick description of lived experience helps us explicate
meanings as we live them (1990). By arguing for the body as a primary source of
knowing, for intersubjectivity (an intimate connection between subjectivity and the
subject) and shared experience, it has been widely drawn upon by somatic dance
practitioners/philosophers, notably Maxine Sheets-Johnstone (1966), Sondra
Fraleigh (1996, 2015), Williamson (2010), and Kozel (2007).
Other theorists that have informed those working in somatics include the Amer-
ican philosopher and educationalist John Dewey (1859–1952) who studied with
Matthias Alexander for more than 20 years in the early part of the twentieth century
and was influenced by his teachings, particularly those that informed his educational
theory relating to self-agency and ethics. Dewey’s reference to “body-mind” as an
essential unity, based on his argument that mind emerges from body’s more basic
physical and psycho-physical functions, rejected the idea that mind and cognition are
superimposed on the soma by transcendent powers of reason emanating from a
spiritual world beyond nature (1958). A contemporary philosopher of aesthetics,
Richard Schusterman has also influenced somatics with his argument that the body is
our basic medium of perception and action by drawing on insights from body-mind
awareness and the aesthetic dimensions of everyday life (2008). He is perhaps best
known for his development of the field of somaesthetics (1999) through an integra-
tion of the disciplines of philosophy of mind, ethics, politics, social theory, gender
studies, and aesthetics. These and other theories have fuelled a broader somatic turn
within dance, movement, and therapeutic practices, and a range of somatic modal-
ities are now practiced widely within dance teaching and training.
A focus on psycho-social-somatic learning gives rise to what could thus be
termed a corporeal philosophy; a philosophy that attends to the lived body, promot-
ing both an inward and outward focus that relates to context, condition, experience,
cultural connections, ethics, and ecological factors. Dance scientist Glenna Batson
has argued for the importance of attention in dance practice, claiming that attending
to bodily sensations is essential to coordinating mind and body and is fundamental to
the capacity and capability of building relationships to self and others (Batson and
Wilson 2014, 106). Attending to sensory data coming from interior states, move-
ment, and environment provides a bridge between inner states and the outer world –
expressing being and becoming in the world, of coupling thinking with world-
making (Batson and Wilson 2014, 108). Attention is thus a basic somatic function
and tuning attention is a core part of a dancer’s practice.
Imagery and Dance
Related to attention is imagery, which is often part of the dancer’s toolkit, to prompt
new movement creation, to notice movement “habits,” and to observe changes in
perceptual focus. Batson notes how image can operate for dancers:
First, an image can arise intuitively simply from exploring movement; second, images can
shape and alter the dynamics of an ongoing movement; and third, images can allow the
dancer to project outside of self, offering ways to observe the self in interacting with space,
time and others.(Batson and Wilson 2014, 61)
Imagery is used regularly in the teaching of dance and has a long history. Several
texts have focused on the way image supports the biomechanical principles of dance
movement and the role of imagery in a dancer’s understanding of anatomical
structure and function (Todd 1968; Sweigard 1974; Franklin 1996). Several dancers
have since developed their own methods and movement systems, which are now
very widely practiced. These include Joan Skinner, who was a pupil of dance teacher
Cora Belle Hunter, who was herself a student of Todd. Skinner is one of several who
have evolved her own method, in her case Skinner Releasing Technique™ (SRT)
using guided poetic imagery, which she describes as “image clusters” that are core to
her work (cited in Neuhaus 2010).
Imagery may be visual, auditory (sonic), or kinesthetic. How imagery enters the
dance studio and rehearsal space varies according to the dance practice, technique, or
choreographic process. For example, in a contemporary dance class the teacher may
focus on one part of the body, providing anatomical information, advice about spatial
factors, what body parts to move, as well as appropriate images to encourage
different qualities of action and to engage the “internal” physiological processes in
moving from stillness to action. As the dancer moves in response, further spatial and
dynamic instruction may be offered as well as imagery, which might be kinesthetic,
metaphorical, or biomechanical. As the dancer continues to practice, the action may
form itself into repeatable sequences. The dancer responds through conscious
embodiment, combining awareness of imagery, anatomical and mechanical struc-
tures and constraints, and physiological processes, learning to manipulate abstract
forms of kinesthetic imagery.
8 S. Whatley
Studies on imagery in dance have also been conducted by cognitive and behav-
ioral scientists (McCarthy et al. 2006; May et al. 2011) who have uncovered the
ways in which dancers are particularly expert in drawing on and sourcing imagery
and move with ease between different sense modalities. For example, the studies
conducted by a team of psychologists led by Jon May and working in collaboration
with Wayne McGregor|Random Dance in the UK set out to “record dancers’
awareness of their use of forms of imagery during movement creation, and to relate
these measures to evidence of patterns of brain activity from neuroimaging studies”
(May et al. 2011, 405; italics in original). Their work showed how novel movement
creation can start from both mental and physical processes. In dance practices that
emphasize somatic principles, dancers often attune to imagery for the transformative
potential of the image in their dancing. So for example, an image of a feather floating
to the floor might initiate movement that is light in quality and free flowing in space.
The image is internalized and encourages personal autonomy in the dancer while
promoting sensory awareness and the tuning of the body to the environment and
other dancers. The image thus enables the dancer “to experience a personalized
version of them so they are relevant to each individual’s process and needs as they
are situated in that precise moment in time” (Emslie 2009, 173). In these cases, the
image is a mental image, offered to the dancer to motivate movement.
Mind Images and Motion Analysis
The role of mind images in dance when motion analysis is involved opens up
questions about how the somatic properties of the dance might be made more visible
in the data, or might stimulate particular kinds of data visualizations that might then
feedback to the dancer more information about the dance. The feedback loop
invokes again Merleau-Ponty’s views of the phenomenal body and particularly
how he divided into the body schema (the body as organism, caused by and in
turn causing movement) and the body image (the visualization of the body as an
external, represented, object) (Merleau-Ponty 2002). So, for example, “when a
performer is offered visual stimuli from a video loop, the visualized body image
drives the body schema” (Polydorou et al. 2015, 194). But the intelligence of the
corporeal dancing body is not always supported by digital technologies and can
enable “a blinded technologization of human movement” (Salazar Sutil 2015, 50).
When the tools for motion analysis serve the needs of the dance and not the reverse,
we avoid a “dependence on total machine automation” (Salazar Sutil 2015, 50) and
ensure that movement extracted from the body supports an increased attentiveness
and consciousness of the dancer’s actions, behaviors, and relationships with others.
However, at the same time, motion analysis tools have led to dance artists being
increasingly fascinated with what technologies can offer to their practice and many
performance makers are exploring how the physical dancing body can even be
absent from the live performance event. These projects have perhaps contributed
to a further move away from an ocular-centric choreographic approach, towards a
more conceptual, process-driven methodology. Alongside the practice is therefore
also a growing discourse that seeks to articulate the particular nature of the human/
technology interface in performance and might contribute thought to how the
material body comes into being through bodily expressions in myriad technological
environments.
Movement data for analytic purposes can be captured through different kinds of
motion capture technologies and new systems are evolving all the time, often in a
move towards making what can be quite sophisticated and complex tools more
accessible, usable, and affordable. (See also chapters: Observing and revealing the
hidden structure of the human form in motion throughout the centuries; Rigid Body
Models of the Musculoskeletal System; Estimation of the Body Segment Inertial
Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis;
Arthrokinematics and Joint Morphology; 3D Dynamic Pose Estimation Using
Reflective Markers or Electromagnetic Sensors; 3D Dynamic Pose Estimation
Using Wearable Robotics; Measurement of 3D Dynamic Joint Motion Using
Biplane Videoradiography; 3D Musculoskeletal Kinematics Using Dynamic MRI;
Cross-platform Comparison of Imaging Technologies for Measuring Musculoskel-
etal Motion; Ultrasound Technology for Examining the Mechanics of Muscle,
Tendon, and Ligament; 3D Dynamic Probablistic Pose Estimation from Data Col-
lected Using Cameras and Reflective Markers; 3D Dynamic Pose Estimation Using
Cameras and No Markers; Three-dimensional Human Kinematics Estimation Using
Magneto and Inertial Measurement Units; Gait Parameters Estimated Using Mag-
neto and Inertial Measurement Units; Physics-based Models for Human Gait Anal-
ysis; 3D Kinematics of Human Motion) Technologies might include gyroscopes,
accelerometers as well as simpler tools such as Kinect. Sophisticated motion capture
labs require experts to set up multiple range and depth-sensing cameras. The
dancer’s movement is captured by means of having sensors (reflective or magnetic
markers) attached to various body parts. A common characteristic of motion capture
is the extracting movement from a body, and the subsequent abstraction of the body
from the physical site of the dance (although performances may involve the dancer
moving in reaction to, or in collaboration with, real time motion capture). The
disconnection between the live dancer and her data, which can be used in various
ways and at different times, can enable the dancer to examine her movement from
outside the experience of dancing, and others to also analyze the data sets for myriad
purposes. The extraction of data, and data that once processed and turned into a
digital avatar that usually appears to carry a clear signature of the dancer and her
gestures in the dots, lines, and trajectories, can be unsettling or induce an uncanny
experience for the dancer. The animation or “digital portrait” (Dils 2002, 94) that
emerges is not a mirror image, nor is it necessarily a representation of the dancer so
the dancer’s sense of self can be disrupted, particularly if glitches enter the animation
process.
However, motion capture can produce measurable data, by producing quantita-
tive information about the load on a muscle, joint, etc., and qualitative information
by representing experience (Schiphorst and Calvert 2015, 243), as well as providing
an interesting defamiliarizing experience for the dancer while simultaneously mak-
ing a bridge between “self and other.” The role of imagery in this process is
10 S. Whatley
interesting. If the dancer is repeating a set phrase of movement, image can influence
the quality, dynamics, and temporal properties, which may reveal more properties of
the dance than could otherwise be observed or felt from the “inside.” Questions arise
such as: does the image carry through into the data, or transform through the data?
And how might the practicalities of the motion capture process impact on the
dancer’s habitation of the image? For example, dancers are often required to repeat
movement patterns or explorations for the captures to succeed in computational
terms. If the dancer is improvising then the process of repetition might disturb the
dancer’s ability to be “present,” “in process,” and authentic in the improvisation.
Even a repeatable or “set” movement phrase is inherently challenging for dancers
who are required to perform a repeated movement for the purposes of capture. Vincs
and Barbour recognized this in their project, noting that “for our dancers, performing
the same phrases over and over presented the danger of repetitive use injuries, and
the very real challenge of finding ways to keep the movement ‘alive’ in a performa-
tive sense in such a repetitive situation. This is, perhaps, one reason that very few
large datasets comprising complex contemporary dance movement exist, and this, in
and of itself, has been a barrier to the development of motion capture analysis
techniques for dance” (Vincs and Barbour 2014, 66).
Motion Analysis in Practice
To examine further the potential partnership between motion analysis and mind
images in dance, and how that partnership is informed by or reveals more about
somatic practices, I turn now to two examples. The first is the ongoing practice of
dance and visual artist Ruth Gibson who working with her partner Bruno Martelli
explores how motion capture can reveal more about the imagery that underpins the
somatic practice and pedagogy of SRT. Gibson creates moving images and installa-
tions, and virtual worlds as locations for enquiry. In some of her work the dancing
body, although a source for what is generated, is apparently absent whereas in other
works such as Summerbranch5 (2005) the dancing avatar is more clearly present. A
recent enquiry focused on whether and how the materiality of the dancing body
could be “captured” with motion capture technology using Skinner’s metaphoric
imagery as a source for movement generation.6 As with many somatic practices,
SRT prioritizes aspects of movement action that are less visible to cameras or
challenge the capture process that uses sensors on the external body. These more
elusive aspects include stillness, breath, floor-based movement, and partnering work.
Capturing what might appear to be “stillness” reveals the continual activity,
5
The title Summerbranch is a pun as with many of Gibson/Martelli works. It is a play on the title of
two choreographic works by Merce Cunningham, Summerspace (1958) and Winterbranch (1964).
6
The project, “Capturing Stillness” was the first study to interrogate SRT in the field of motion
capture, computer game worlds and interactive virtual environments. It was funded by the Arts and
Humanities Research Council and took place at Coventry University (2010–2013).
hesitations, and fluctuations of the body in an apparent state of motionlessness.

Gibson then developed these captures into kinetic landscape visualizations to
attempt to reveal the metaphoric aspects of the image actions in SRT, such as
“mist” and “moss.” By attending to the subtleties of human motion, the captures
unearthed more about the pedagogies and philosophies of SRT and somatic princi-
ples at play (Whatley 2015). This work then led to the creation of an application,
MocApp, for mobile and tablet screens, using the same captures of experienced SRT
practitioners to enable the user to explore human action. The immediacy of the
dancing “stick figures” creates a chance to play with an accessible version of motion
capture for experimentation and to create new captures. A related project by Gibson
and Martelli, MAN A (2014), focused on kinetic, stylized humanoid forms derived
from 3D motion-captured dancers to play with the viewer’s proprioception and to
elicit kinesthetic responses akin to the mirror-neuron response creating the potential
for embodied empathy with the virtual dancers.
Gibson’s projects are not concerned with statistical analyses or measuring/com-
paring data sets but they demonstrate the synergistic potential of imagery and
technology for revealing some of the more intangible aspects of dance practice,
and particularly practice that privilege imagination, empathetic communication, and
movement generation that holistically tune the mind and body. The corporeal
dancing body is on one hand rendered into data and then reconstituted in multiple
ways, into artworks and applications, but each shows how dance is a valuable site for
seeing how imagery can source movement that provides access to more insights to
the human body in action. As with all projects where the capture is of a human
subject, issues of agency and ethics are at stake. In somatic practices, the emphasis is
on individual agency, resisting the conditioning models of some more traditional
dance techniques, such as classical ballet and some contemporary dance techniques.
Dancers working with Gibson are expert SRT practitioners and so have a sharp
awareness of their own subjectivity but reported finding the motion capture inter-
esting, one acknowledging the transformative experience of being captured, noting:
You need to regularly pause in a particular position (arms outstretched) so that the computer
can recalibrate, and actually finding yourself in this place from time to time became quite
interesting - like revisiting a familiar place but finding a new resonance each time. I had
feared that the outcome of ‘capture’ would encourage me to become over focused on form
rather than the process of the image unfolding but this was not the case, and . . ... I think
sticking with the process of the image unfolding allowed for a certain clarity of data –that is the
forms captured are clearer as a result of not predetermining them.(cited in Whatley 2015, 202)
The second example is the WhoLoDance project which has motion analysis at the
core of the project7 and by generating large datasets of dance movement may well
7
WhoLoDance is a Research and Innovation Action funded under the European Union’s Horizon
2020 Programme. The project’s aim is to develop and apply breakthrough technologies for dance
learning; for practitoners, researchers, professionals, dance students and the general public. See
http://www.wholodance.eu/
12 S. Whatley
address the problem raised by Vincs and Barber noted earlier. Using machine
intelligence tools and methodologies, the aim of WhoLoDance is to apply sequence
similarity and clustering methods for analysis of motion captured dance data in order
to allow for multiple novel applications in the area of dance analysis and education.
Four dance genres are the principal focus for the project because each is to at least
some extent codified, so is built around a lexicon of movement actions and
sequences that form a basis for the genre’s pedagogy: Flamenco, classical ballet,
Greek folk dance, and contemporary dance. As the project continues and the dance
community is more informed of the project and begins user testing the tools as they
develop, the aim is to extend the exploration towards the capture of more somatically
informed improvisational dance practices, to test out the premise on which the
project is based that the projection of the dancer in relation to the live dancer
produces a novel sense of embodiment and a different kind of relationality.
The data that is being generated in WhoLoDance is being analyzed in a number of
ways, to identify the movement principles and connections between different dance
practices and to support the learning principles that have evolved through the many
years of dance teaching within each of the genres. The aim is not to replace the
teacher or to provide a virtual proxy for the teacher but rather to enable the dancer
and teacher to discover the more hidden properties of the dance genre. The motion
capture production is taken through various stages and two pipelines of develop-
ment. The first, covering both high-end and low-end capture devices, will create a
blendable motion capture repository and perform data collection and curation as well
as 3D position reconstruction for the modeling of the avatar. The second pipeline is
concerned with creating the interactive visualization of the virtual bodies that will be
used in the installation (polygonal 3D avatars, or real-time visualizations of force
fields, vectors of movement and particle point-clouds) that identifies when a physical
body is intersecting with a virtual body and feeds back sensory signals to the user.
The ambition is for the WhoLoDance immersive environment to innovate dance
teaching and to encourage a greater sense of three-dimensionality by developing a
life-size volumetric display that will enable a dancer to literally “step inside” the
dance teacher’s body. While other motion training projects have used motion capture
to create a virtual dance teaching tool (Chan et al. 2011) and have gone some way to
collect enough data for evaluating the difference between the learner and teacher,
none have yet combined motion capture with virtual reality and hologram technol-
ogies to support the teaching of dance with a focus on the qualities of movement and,
in particular, the imagery that generates metaphors of motion for the dance learner
and teacher. In WhoLoDance, building a volumetric moving hologram capable of
displaying real-time motion capture content will test how dancers respond to virtual
environments that feed back information to the dancer about movement “accuracy”
and “feeling states” as well as the imagery that is the source for their dancing and
may open up new questions about embodiment. The concept is that, by inhabiting a
virtual avatar/projection space /holographic projection, the tools will elicit for the
dancer a particular experience of being “in” the body, the dancer’s own body and the
body of another that will be informative, and provide new ways to learn movement
and perceive movement from the outside and inside simultaneously.
The main goal from the education perspective is therefore to create methods for
learning dance that incorporate novel digital and virtual tools, and which are
appropriate for a range of learners providing access to a range of different dance
genres through immersive representations of virtual bodies. By giving the student
the chance to experiment with different modalities of feedback (audio, visual, audio-
visual, verbal, etc.) and also with different virtual environments or different avatars,
the project will discover whether it provides interesting opportunities to design
adaptive and personalized paths to learning.
Conclusion
This essay has sketched out the territory of somatic practices and how imagery
participates in the dancer’s experience of making, performing, and teaching dance.
Discussion has also focused on motion analysis and outlined some ways in which
projects and artists are using motion capture technologies to unearth more of the
“hidden” aspects of dance, whether performed by expert practitioners or enjoyed by
those dancing for pleasure and general well-being. Reference has also been made to
how motion analysis technologies can support the teaching and learning of dance. In
this context, it is important to recognize that dance is a heterogeneous art form
incorporating many styles and techniques. As Vincs and Barbour observe, the
“semiotic variability of dance” means that there is “no single ‘grammar’ of the
body [that] can be relied upon to carry the communicative valence of any particular
dance movement or practice” (2014, 65).
Somatic practices promote a more person-centered approach to movement devel-
opment than some dance techniques that are more didactic or conditioning. Somatics
also encourage embodied thinking and help to tune and train attention, bringing
mind and body together. Most dance practices including those with a somatic
emphasis recognize and make use of imagery in the training of dancers. Imagery
is frequently core to a dancer’s perceptual awareness and for enhancing the com-
munication of kinesthetic data. Imagery can also help dancers make new movement,
to notice and break out of movement habits, and gain insights to movement qualities.
There are multiple ways in which an image can be transformative, “so real that one
can become the image” (Skinner, cited in Neuhaus 2010). When brought into the
domain of motion capture, image can participate in both the input and output stages,
virtualizing the sensory, corporeal world of the dancer. Motion analysis also intro-
duces a new perspective, and a new lens on dance, revealing more and sometimes
surprising aspects of dance because of the precise detail that the captures can record,
as Vincs and Barbour discovered in their integration of PCA analysis of dance
(2014). In WhoLoDance, the interrelationship between imagery and technology is
core to creating a new sensorial environment and experience for the dancer. Simi-
larly, in Gibson’s work, the interrelationship between somatic-informed dance,
motion capture, and poetic imagery promotes an enhanced embodied perception.
The projects discussed here are part of a growing collection of initiatives that are
incorporating a variety of motion analysis tools and technologies to support the
14 S. Whatley
making and teaching of dance; each exploring the hand in hand relationship between
technology, imagery, and dance.
References
Batson G, Wilson MA (2014) Body and mind in motion: dance and neuroscience in conversation.
Intellect, Bristol
Camurri A, Lagerlof I, Volpe G (2003) Recognizing emotion from dance movement: comparison of
spectator recognition and automated techniques. Int J Hum-Comput Stud 59:213–225
Camurri A, Volpe G, De Poli G, Leman M (2005) Communicating expressiveness and affect in
multimodal interactive systems. IEEE Multimedia 12(1):43–53
Castellano G, Villalba S D, Camurri A (2007) Recognising human emotions from body movement
and gesture dynamics. In: Paiva A, Prada R, Picard R W (eds) 2nd international conference on
affective computing and intelligent interaction, Lisbon, September 2007. Lecture notes in
computer science. Springer, Heidelberg, pp 71–82
Chan JC, Leung H, Tang JK, Komura T (2011) A virtual reality dance training system using motion
capture technology. IEEE Trans Learn Technol 4(2):187–195
Dewey J (1958) Experience and nature, vol 1. Courier Corporation, North Chelmsford,
Massachusetts
Dils A (2002) The ghost in the machine: Merce Cunningham and Bill T. Jones. PAJ: J Perform Art
24(1):92–104
Emslie M (2009) Skinner releasing technique: dancing from within. J Dance Somatic Pract 2
(1):169–175
Forsythe W, deLahunta, S (2011) Motion Bank Brochure and Logo Generator. Available via http://
motionbank.org/sites/motionbank.org/files/mb_brochure.pdf. Accessed 1 Oct 2016
Foucault M, Martin LH, Gutman H, Hutton P (1988) Technologies of the self: a seminar with
Michel Foucault. University of Massachusetts Press, Amherst
Fraleigh S (1996) Dance and the lived body: a descriptive aesthetics. University of Pittsburgh Press,
Pittsburgh
Fraleigh S (2015) Moving consciously: somatic transformations through dance, yoga, and touch.
University of Illinois Press, Urbana
Franklin E (1996) Dance imagery for technique and performance. Human Kinetics Publishers,
Champaign
Hanna T (1980) The body of life: creating new pathways for sensory awareness and fluid
movement. Healing Arts Press, Rochester
Harbonnier-Topin N, Simard H (2015) Somatic education and introspective verbalisation. In:
Whatley S, Garrett Brown N, Alexander K (eds) Attending to movement: somatic perspectives
on living in this world. Triarchy, Axminster, pp 239–250
Kapur A, Virji-Babul N, Tzanetakis G, Driessen P F (2005) Gesture-based affective computing on
motion capture data. In: Tao J, Tan T, Picard RW (eds) 1st international conference on affective
computing and intelligent interaction, Beijing, October 2005. Lecture notes in computer science.
Springer, Heidelberg, pp 1–7
Kozel S (2007) Closer: performance, technologies, phenomenology. MIT Press, Cambridge, MA
Manen MV (1990) Researching lived experience: human science for an action sensitive pedagogy.
State University of New York Press, New York
May J, Calvo-Merino B, deLahunta S, McGregor W, Cusack R, Owen A, Veldsman M, Ramponi C,
Barnard P (2011) Points in mental space: an interdisciplinary study of imagery in movement
creation. Dance Res 29(2):402–430
McCarthy R, Blackwell A, Wing A, Hollands K, Barnard P, Nimmo-Smith I, Marcel A (2006)
Bodies meet minds: choreography and cognition. Leonardo 39(5):475–478
Merleau-Ponty M (2002) [1962] Phenomenology of perception (trans: Smith C). Routledge,

London
Neuhaus B (2010) The Kinaesthetic Imagination; an interview with Joan Skinner. In Contact
Quarterly Online Journal, July 2010. Available via https://community.contactquarterly.com/
journal/view/skinner. Accessed 9 Oct 2013
Polydorou D, Michailidis T, Bullock J (2015) Communication through haptic interaction in digital
performance. In: Salazar Sutil N, Popat S (eds) Digital movement: essays in motion technology
and performance. Palgrave MacMillan, Basingstoke, pp 189–205
Salazar Sutil N (2015) Intelligence behind movement: laboratories of biomechanics and the making
of movement utopia. In: Salazar Sutil N, Popat S (eds) Digital movement: essays in motion
technology and performance. Palgrave MacMillan, Basingstoke, pp 35–54
Schiphorst T (2009) The varieties of user experience: bridging embodied methodologies form
somatics and performance to human computer interaction. PhD dissertation, University of
Plymouth
Schiphorst T, Calvert T (2015) Practising choreography: digital movement as emergent Praxis. In:
Salazar Sutil N, Popat S (eds) Digital movement: essays in motion technology and performance.
Palgrave MacMillan, Basingstoke, pp 239–259
Sheets-Johnstone M (1966) The phenomenology of dance. Dance Books, London
Shippen J, May B (2010) Calculation of muscle loading and joint contact forces during the rock step
in Irish dance. J Dance Med Sci 14(1):11–18
Shippen J, May B, Alexander P, Kolk E (2015) Growing healthier. In: Christer K (ed) Proceedings
of the 3rd European conference on Design4Health, Sheffield, 2015
Shusterman R (1999) Somaesthetics: a disciplinary proposal. J Aesthet Art Critic 57(3):299–313
Shusterman R (2008) Body consciousness: a philosophy of mindfulness and somaesthetics. Cam-
bridge University Press, Cambridge
Sweigard LE (1974) Human movement potential: its ideokinetic facilitation. University Press of
America, Lanham
Todd M E (1968) [1937] The thinking body: a study of the balancing forces of dynamic man.
Princeton Book Company, Princeton
Vincs K, Barbour K (2014) Snapshots of complexity: using motion capture and principal compo-
nent analysis to reconceptualise dance. Digit Creat 25(1):62–78
Whatley S (2015) Motion capture and the dancer: visuality, temporality and the dancing image. In:
Whatley S, Garrett Brown N, Alexander K (eds) Attending to movement: somatic perspectives
on living in this world. Triarchy, Axminster, pp 193–204
Williamson A (2010) Reflections and theoretical approaches to the study of spiritualities within the
field of somatic movement dance education. J Dance Somatic Pract 2(1):35–61
Williamson A (2016) Reflections on existential phenomenology, spirituality, dance and movement-
based somatics. J Dance Somatic Pract 8(2):273–299
Running Shoes: Injury Protection
and Performance Enhancement
Steffen Willwacher
Abstract
In addition to being possibly the most frequently performed sport in the world,
running is associated with a high incidence of overuse injuries. Running shoes
can modify the load imposed on the runner’s body by changing the interface
between the biological system and the running surface. The key shoe design
features that can affect biomechanical loading are the compression and bending
stiffness of the midsole and the medial support elements. These features can also
affect the energy consumption of the runner, which is key to running perfor-
mance. In general, no footwear design feature is able to reduce the load in the
entire leg. Instead, what can be achieved is a shift of the loading from distal to
more proximal regions of the limb, or vice versa, often by shifting the point of
force application of the ground reaction force. Distance-running performance is
strongly diminished by increases in the mass of footwear; however, this can be
negated to some extent when the additional mass is due to the presence of
cushioning materials, which can reduce impact costs in heel–toe runners. Future
research in this field needs to identify more clearly the biomechanical risk factors
associated with running-related injuries. Based on this, recommendations with
respect to individually optimal footwear can be made with greater confidence.
This will be essential for successful implementation of customized footwear
within the running-shoe market.
Keywords
Running mechanics • Footwear biomechanics • Sports technology • Running-
related injuries • Overuse injuries • Barefoot running • Impact cushioning •
Midsole bending stiffness • Sports performance • Running economy
S. Willwacher (*)
Institute of Biomechanics and Orthopedics, German Sport University, Cologne, Germany
Institute of Functional Diagnostics, Cologne, Germany
e-mail: s.willwacher@dshs-koeln.de

DOI 10.1007/978-3-319-30808-1_121-1
2 S. Willwacher
Contents
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
A Brief History of the Running Shoe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Running Shoes and Running-Related Injuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Running Shoes and Distance Running Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
State of the Art
Today, most running shoes are designed to either reduce the risk of sustaining an
overuse injury or improve performance. In general, this can lead to different and
sometimes contradictory design solutions. Theoretically, multiple extrinsic and
intrinsic risk factors contribute to the development of overuse injuries, including
biomechanical factors, which can be altered by means of footwear technology.
However, in practice, quality prospective studies aimed at the identification of
relevant biomechanical risk factors are lacking. In addition, individual responses
to footwear modifications can be highly varied. The reasons for this variability are
quite poorly understood at present, making it difficult to predict the ideal footwear
for a given individual. Nonetheless, the effects of the core footwear-design features
on lower extremity loading and distance-running performance have been studied
extensively, so the average responses of the running population to such design
modifications can be identified. The most important will be summarized in this
chapter.
A Brief History of the Running Shoe
Humans regularly use two different modes of locomotion: walking and running.
While walking is by far the dominant mode for people today, running has played an
important role over the course of human evolution (Bramble and Lieberman 2004).
In fact, the human body possesses some unique anatomical features that make
humans remarkably good endurance runners compared to other primates and curso-
rial mammals (Lieberman and Bramble 2007). In this endurance running hypothesis,
it is suggested that the capacity of early humans to endure long periods of continuous
running enabled their implementation of a particularly effective hunting strategy,
namely to chase animals over long distances until, exhausted and overheated, they
became easy prey (Carrier et al. 1984). This might have allowed our ancestors to
access energy- and protein-rich sources of food, which in combination with the
invention of cooking techniques, fueled the development of our remarkably power-
ful brains (Carmody and Wrangham 2009; Herculano-Houzel 2016). Evolutionary
adaptations for endurance running in early humans took place over thousands, if not
millions of years, likely without footwear or at least with only minimal varieties.
Running Shoes: Injury Protection and Performance Enhancement 3
Footwear and other cloth-based artifacts rarely survive to form part of the fossil
record, making the roots of their development difficult to determine. There is no
archeological evidence regarding the use footwear prior to the middle Upper Paleo-
lithic (Trinkaus 2005). Early footwear was mainly used for thermal and mechanical
protection, using thin layers of animal skin in simple configurations, sometimes
combined with dried grass to improve thermal insulation. It is reasonable to assume
that such footwear was used for all kinds of locomotion for thousands of years
without any further specialization. The common use of specific types of footwear for
specific purposes is a relatively new development, which probably started at the end
of the eighteenth century, as the Industrial Revolution was beginning.
The “invention” of running-specific shoes occurred at this time. The revitalization
of running in the late eighteenth century, this time as a sport, resulted in the
development of relatively lightweight shoes aimed at providing the extra traction
needed to counter the more forceful and dynamic motions involved in running.
Initial attempts were made in England, where leather running shoes were produced.
Major improvements in the quality of sports shoes were made when rubber was first
introduced and used as the material for the sole. With improvements in manufactur-
ing techniques, such as vulcanization, the quality and durability of these shoes
further improved.
During this period, the sport of running was mostly performed by athletes at
competitions, and not by the general public as a recreational activity. Therefore, most
running was performed on running tracks, or on natural surfaces like grass and soil in
the case of cross-country running. In order to improve traction on these grounds,
spiked shoes were invented.
After the Second World War, distance running slowly became more and more
popular as a recreational sport. The resulting bump in demand for dedicated footwear
led to the foundation, and in turn prosperity, of many sports footwear companies.
Recreational running in its current form started in the 1970s, when work in Western
Europe and the USA generally became less time consuming and less physically
demanding as a result of socioeconomic changes. It quickly became a pastime for
millions of people, particularly because of its positive effects on general well-being
and cardiovascular health. Nonetheless, running-shoe designs of that time were still
quite simple: the main goal was to provide running shoes that were lightweight while
providing some protection against cold temperatures, humidity, and surface
roughness.
This changed in the late 1970s when the first running shoes with impact cush-
ioning systems were invented, starting the era of technical running shoes as we know
them today. In the following years, additional technological innovations were
introduced like motion-control features, improved lacing systems, and even
microcomputer-controlled systems that adjust the cushioning properties in response
to sensor data obtained while running.
A countermovement to the technical footwear trend emerged in 2009 when the
book Born to Run was released by Christopher McDougal, in which he refers to the
members of the reclusive Tarahumara tribe in Mexico (McDougall 2009). The
members of this tribe are all remarkably good long-distance runners, but they run
4 S. Willwacher
barefoot, or at most they use very minimal footwear. McDougal found that these
people did not suffer from running-related injuries and concluded from this that the
use of modern technical running shoes might be the root cause of most injuries in
recreational runners. He further cites the work of Harvard professor Daniel
E. Lieberman, who also stated that modern technical running shoes are not well
suited to the particular form of running that has evolved in humans over thousands of
years (Lieberman et al. 2010). They both propose that modern running shoes
facilitate an “unnatural” running style, leading to overuse injuries in runners. The
footwear industry quickly responded to these popular publications by introducing
minimalistic footwear or even “barefoot” running shoes, which only offer minimal
impact cushioning and support elements.
Today, distance running is performed in one of three categories of running shoes:
Barefoot, minimal footwear, or technical footwear. The dividing lines between these
can be somewhat blurred, and the range of running shoes on the market might even
be considered a continuum, extending from barefoot running without any technical
support, through minimal shoes with basic elements to provide thermal and surface-
roughness protection, and on to classical technical shoes that include diverse tech-
nical components to perform various functions such as impact cushioning and
motion control (Fig. 1).
Ever since the running boom started in the late 1970s, footwear companies have
been interested in evaluating their products and acquiring new design ideas from
biomechanics research (Nigg 2010). Therefore, it is not surprising that the barefoot
running boom also resulted in extensive scientific debate among biomechanists,
arguing about the biomechanical consequences of using different kinds of footwear
and the effect of associated running styles on lower-extremity loading.
The purpose of this chapter is to summarize current knowledge on the relation-
ships between running-shoe technology, lower-extremity loading, injury develop-
ment, and the energetic cost of running. Lower-extremity loading is potentially the
most important factor in the design of running shoes owing to the high prevalence of
running-related overuse injuries in recreational and competitive runners. An inci-
dence of up to 79% in nonelite runners was reported by van Gent et al. (2007), with
different injury proportions found in different populations of runners (Kluitenberg
et al. 2015); this highlights the importance of effective prevention strategies.
Running Shoes and Running-Related Injuries
Despite an extensive amount of research performed in the field of running-related

loading and injury development, there is relatively little consensus regarding potential
risk factors. The major reason for this is the multifactorial nature of running-related
overuse injuries. Overuse results from repetitively stressing biological tissues of the
body above a certain threshold level, resulting in a degenerative response (Fig. 2).
Biomechanical loading during running can induce both positive and negative
adaptations in biological tissues, depending on the kind of load imposed and the
characteristics of the tissues involved. Tissue adaptation by itself is a function of
Fig. 1 Schematic overview of running-footwear categories
Fig. 2 Schematic overview

of factors involved in the
adaptive response of
biological tissues to
biomechanical loading during
running
tissue stress, which can be quantified by its amplitude, rate, duration, frequency, and
other measures. The stress response of the tissue is governed by biochemical factors,
which, among others, are related to the hormonal and nutritional status of the runner.
At the same time, the stress is dependent upon the biomechanical loading on the
tissue, which is characterized by its geometry, its internal architecture, and
6 S. Willwacher
extracellular-matrix composition (Nigg and Herzog 2007). A positive tissue adap-

tation might therefore reduce the stress experienced for the same amount of mechan-
ical loading, thereby allowing the biological structure to operate with a greater safety
factor. Biomechanical loading during running is determined by the individual’s
running biomechanics, which are strongly influenced by the training regime and
the running environment in which the sport is performed. While the training regime
can be described by the training volume, intensity (speed), resting periods, and other
variables, the running environment is characterized by environmental factors,
running-surface characteristics (including stiffness, inclination, cross-slope, and
traction), and footwear characteristics. Psychological and sociological factors deter-
mine the motivation of the runner to perform the sport and therefore can strongly
influence the training goals and the environment in which the sport is performed.
From the above remarks, it becomes clear that running-related overuse injuries
are influenced by several factors within a complex scenario. Moreover, interactions
between factors can make determining the contribution of each individual factor
over the course of injury development a very complex and difficult task. The gold-
standard approach in this field is the prospective study. Here, runners are analyzed in
noninjured state and are followed for a certain period of time, during which statistical
approaches are used to find significant (and clinically meaningful) differences
between those that incur an overuse injury and those that don’t. As there are multiple
running-related overuse injuries and multiple risk factors under consideration, this
study design requires large amounts of participants to be investigated in order to
reach an adequate level of statistical power. Other problems are related to identifying
the specific role of different risk factors, particularly biomechanical factors, in the
development of overuse injuries. Biomechanical assessment is mainly carried out
prior to data collection in a highly controlled laboratory setting, including level
running over short distances or on treadmills, most often in an unfatigued state. The
question arises whether these settings and testing modes are sufficiently representa-
tive to be applicable to the biomechanics of individuals running under real-world
conditions. Fatigue could alter the individual’s running biomechanics in a nonlinear
manner, challenging the validity of laboratory-based motion analyses of subjects in
an unfatigued state. The role of running-related fatigue is generally under-
researched, even though the reduced ability of the muscular system to control the
motion of the musculoskeletal system may strongly affect the load distribution
within the body.
Despite these methodological issues, some risk factors for the onset of running-
related injuries have been identified by researchers, though for the reasons men-
tioned above, the supporting evidence is generally in the range of limited to minimal:
Previous injury within the last 12 months is reported to be a risk factor for sustaining
another running-related injury in the future. This result has been found in most high-
quality studies, indicating that runners with a history of injury should be even more
careful when performing distance running (Saragiotto et al. 2014). Even though the
statistical basis for this risk factor is probably better in comparison to other risk
factors, it does not allow us to understand the underlying mechanisms of injury
development, and moreover, it is not a variable that can be altered by the runner or by
technological aids. Other risk factors being discussed are excessive mileage, age,
gender, running experience, previous sports activity, marathon participation, use of
orthotics and inserts, and abnormal running biomechanics (Saragiotto et al. 2014).
Some of these variables, mostly those related to running biomechanics can be
modified by means of technological interventions, like running shoes. Consequently,
the footwear industry has tried to develop shoes addressing these biomechanical risk
factors. The two most influential factors emerging in the last 40 years have been
rearfoot eversion and impact loading. In order to control the amplitude of these
variables, motion control and impact cushioning technologies have been developed
and marketed. Their development has been largely based on theoretical consider-
ations relating to these variables and rational design principles, but there remains a
lack of strong empirical support from prospective studies. Therefore, the role of
footwear in the development of running injuries is still not very well understood. In
the following section, I will try to summarize how running shoes can modify the risk
of overuse injury in a more general way.
Today, the most common running injuries are patella femoral pain syndrome,
Achilles tendinitis, plantar fasciitis, shin splints (medial tibial stress syndrome), and
iliotibial band syndrome (Nielsen et al. 2014; Kluitenberg et al. 2015). The origin of
these injuries can be found in abnormal stress that is applied to the tissues involved,
which is in part the result of abnormal biomechanics. In all of the aforementioned
running injuries, this can take the form of an increased load imposed on the
structures and/or altered joint kinematics, which can change the loading situation
of the tissues involved. For example, when considering patella femoral pain syn-
drome, it can be expected that chronically increased load transmitted through the
quadriceps and patella tendon will lead to increased contact pressure between the
patella and femur, which over time will result in an overuse-related response,
including tissue irritation and/or inflammation. In addition, the stress distribution
between the patella and femur may be affected by the knee joint kinematics. If, for
example the tibia is rotating too much internally during the stance phase, the patella
may not be gliding along its natural path, causing pressure peaks at its edges and
correspondingly increased stress. Therefore, both joint kinematics and the loading of
the structures surrounding the joint need to be addressed in the risk assessment for
overuse injuries. The loading of structures surrounding a joint is usually described
using joint moments determined by inverse dynamics; more sophisticated modeling
techniques allow an analysis of the load distribution on the different structures
(muscles, ligaments, capsules) involved (Robertson et al. 2013). In the next section,
the role of the core footwear design features in modifying ground-reaction-force
characteristics, joint loading, and kinematics will be discussed; in particular, midsole
hardness, midsole geometry, and support elements.
One of the most often studied design features in running shoes are the material
properties and geometry of the midsole in the rearfoot area. This is due to the
importance that was initially given to the first peak of the vertical ground-reaction-
force (GRF) component. It was believed that the amplitude and loading rate of this
force peak was related to overuse injury development. The study of impact forces
has (re)gained popularity during the minimal/barefoot running boom observed in
8 S. Willwacher
recent years, even though the evidence of a relationship between impact-force

characteristics and overuse injuries is inconclusive (Nigg et al. 2015). Despite
what was initially hypothesized from the results of simple mechanical testing, it
was frequently observed that impact-force amplitudes did not change in response to
altered midsole hardness. Reasons for this phenomenon are twofold: On the one
hand, runners have been found to change certain parameters like footfall pattern and
joint stiffness in response to altered interface stiffness (Ferris et al. 1998; Nigg and
Liu 1999; Hardin et al. 2004; Willwacher et al. 2014b), which can counteract
changes in impact force expected from a purely passive mechanical viewpoint. On
the other hand, it was shown that the first peak of the vertical ground-reaction force
(GRF) is the result of a superimposition of high frequency components transmitted
through the heel of the runner and a low frequency component related to the
acceleration of the center of mass (CoM) of the runner. Shorten and Mientjes
(2011) clearly showed that changing the hardness of the midsole in the heel area
reduces the high frequency component of the vertical GRF as predicted by impact-
attenuation theory. Nevertheless, due to the complex interplay with the low fre-
quency component and the biomechanical adaptations in response to altered inter-
face hardness, no systematic change in the first impact of the vertical GRF was
observed, challenging the validity of the first peak of the vertical GRF as an indicator
of impact severity.
Altered footfall pattern is the most important biomechanical change that occurs in
response to altered midsole stiffness. Runners who are used to running in well-
cushioned footwear usually adapt their footfall pattern towards a more midfoot or
even forefoot-oriented pattern when running in harder shoes (De Wit et al. 2000;
Willwacher. 2014b). This response is observed in the majority of habitually shod
runners, even though some runners retain a rearfoot landing. In addition, the
response to altered midsole stiffness is moderated by the stiffness of the running
surface, so that the change in the footfall pattern is less pronounced when running on
a softer running surface (Willwacher et al. 2014b). Changes in the strike pattern of
runners need to be considered as a function of the combined stiffness of the running
shoe and the running surface. While it is generally assumed that rearfoot strikers
switch to a flatter footfall pattern when running in harder shoes or on harder surfaces,
to avoid pressure peaks and associated pain in their heels (De Wit et al. 2000), it is
not understood why some keep their rearfoot running style in a harder running
environment (shoe and surface). Nonetheless, changing the footfall pattern necessi-
tates changes in leg biomechanics, leading to a different load distribution and a
spatiotemporally altered running motion.
Running in minimalistic or barefoot shoes, that is with almost no impact cush-
ioning, generally leads to a lower horizontal distance from the CoM to the center of
pressure at heel strike, a reduced ground contact time, a higher step frequency, and a
shorter stride length. Thus, over a given distance, runners who make these adjust-
ments need more ground contacts than those maintaining a typical rearfoot strike
pattern. As a consequence of the altered leg configuration at heel strike, the lever
arms of the GRF with respect to the lower extremity joints are systematically
changed as well. Lever arms are increased at the ankle joint, both in the sagittal
plane of motion and in the frontal plane of motion, leading to increased loading of
active and passive structures resisting the external moment created by the GRF and
the increased lever arms at the ankle. Consequently, there is a greater load on the
triceps surae muscle–tendon unit, including the Achilles tendon. Additionally, foot
muscles, providing an internal plantar flexion and inversion moment, like M. tibialis
posterior or m. flexor halluces longus, are more highly loaded than during heel–toe
running. It can be further assumed that higher bending load is acting on bones within
the mid-portion of the foot. If there is no adequately adaptive response of the more
highly loaded biomechanical structures, they will experience chronically increased
stress, which may eventually lead to an overuse injury. In accordance with this,
overuse injuries of the calf, the Achilles tendon, plantar fasciae, and the midfoot are
reported in the literature, often when insufficient time for adaptation was provided in
the training protocol. On the other hand, running with a midfoot or forefoot strike
pattern can result in shorter GRF lever arms at the knee, leading to reduced external
flexion moments and reduced load on the quadriceps muscle group, patella tendon,
and other structures involved in creating an internal knee extension moment. There-
fore, changing the hardness of the midsole of a running shoe can be used to shift the
load imposed on the lower extremities away from the knee towards the foot and
structures surrounding the ankle joint. For runners suffering from pain in the
aforementioned structures of the knee it might be helpful to consider an increase
in midsole hardness. Nonetheless, the biological structures of the foot need to be
capable of handling the load imposed on them. Therefore, prior to changing foot-
wear, an adequate adaptation period or an additional strengthening program for foot
muscles is recommended.
Another very common running-shoe design feature is addition of stabilization
elements, usually in the medial aspect of the midsole or within the upper of the shoe.
These elements originate in the idea that some runners are not capable of handling
the high loads that act on the feet during the ground contact phase; therefore, they
need additional support from their shoes to avoid excessive motion of the rear- and
mid-foot, and the associated tissue stresses. The invention of stabilization elements
originated from observations of runners showing high amplitudes of rearfoot ever-
sion during mid-stance. It was considered that this motion is harmful to the runner
and that foot motion needs to be guided to stay within a “normal” joint excursion. It
is interesting to note that the external moments causing the eversion increased with
the arrival of impact cushioning systems, which almost always increase the heel
height of running shoes. This greater elevation increases GRF lever arms and
corresponding external joint moments at the talocrural joint (Stacoff et al. 1988).
It was initially believed that high eversion amplitudes and eversion velocities are
harmful to the runner because they could lead to an inhomogeneous stress distribu-
tion within the Achilles tendon and plantar fascia. Furthermore, it was shown that
rearfoot eversion motion is coupled to internal rotation motion of the tibia
(Hintermann et al. 1994). It was therefore believed that excessive rearfoot eversion
might be responsible for rotation-related injuries at the knee, like iliotibial band
syndrome or patellofemoral stress syndrome. Different approaches were taken to
reduce the amount of rearfoot eversion by means of running-shoe technology.
10 S. Willwacher
Popular designs include an elevated midsole either underneath the medial longitu-
dinal arch of the foot or underneath the sustentaculum tali. Additionally, harder
materials are used within this region of the midsole to provide increased resistance
against eversion motion of the rear- and mid-foot. Other, less common approaches
include the use of stabilizing elements, like elastic bands within the upper or hard
plastic guiderails built into the intersection between the midsole and the upper.
Depending on the details of the footwear stabilization design and the subject samples
investigated, motion-control shoes have been shown to reduce the amount of
rearfoot eversion, although sometimes only to a minimal extent.
On the other hand, recent prospective studies challenge the idea that amplified
rearfoot eversion is really a risk factor for running-related overuse injuries, showing
that only runners exhibiting very high levels of rearfoot eversion are at a higher risk
of injuries. Based on these results, it seems that the role of rearfoot eversion has been
overestimated by running shoe companies over the past 30 years and that many
runners have been using these kinds of footwear without actually benefitting from
them (Nigg et al. 2015). In fact, there is evidence suggesting that rearfoot eversion
acts as a natural mechanism for absorbing impact energy (Perry and Lafortune 1995)
and that energy stored within the elastic structures of the medial aspect of the
longitudinal arch and within the plantar fascia can be effectively reused during the
push-off, thereby reducing the metabolic energy demands of the runner (Stearne
et al. 2016). It has also been shown that coupling of rearfoot eversion to tibial
rotation can vary greatly across individuals, indicating that threshold levels between
“normal” and “harmful” tibial rotation are subject-specific and cannot be the same
for the entire running population (Hintermann et al. 1994).
During the push-off phase of the ground contact, impact cushioning and motion
control elements of running shoes only minimally affect the load distribution of the
lower extremities. This is because the heel is lifted from the ground and GRFs are
transmitted through the forefoot, which is not affected by footwear elements located
at the rear- and mid-foot. Nevertheless, footwear technology can affect the load
distribution within the lower extremities during this phase. The three most important
design features in this respect are the longitudinal bending stiffness, the forefoot
offset, and the forefoot rocker of the shoe. The longitudinal bending stiffness has a
direct effect on the range of motion of the metatarsophalangeal (MTP) joints
(Stefanyshyn and Wannop 2016). Due to the so-called windlass-mechanism, the
range of MTP dorsiflexion directly affects the tension of the plantar fascia. Runners
who are sensitive to increased stress of the plantar fascia might therefore choose a
running shoe with increased longitudinal bending stiffness in order to put less stress
on the tissue. Apart from affecting the stress of the plantar fascia, the longitudinal
bending stiffness affects the pressure distribution underneath the foot (Willwacher
et al. 2013). With higher longitudinal bending stiffness, the center of pressure is
shifted to the anterior portion of the foot. This shift has a direct effect on joint
moments for the MTP joints, ankle, and to a lesser extent on the knee joint. This is
because the center of pressure strongly correlates with the point of force application
for the GRF. Shifting the point of force application for the GRF vector affects the
GRF lever arms and therefore has a direct effect on the joint moment necessary at the
MTP, ankle, and knee joints (Willwacher. 2014a). Increasing longitudinal bending
stiffness (LBS) leads to higher joint moments at the MTP and ankle, whereas they
are reduced at the knee.
Some running shoes include rocker elements that can also affect MTP biome-
chanics. Rocker elements are usually implemented in shoes by increasing the
thickness of the forefoot midsole. By increasing the midsole height, a pronounced
curvature of the midsole can be created under the forefoot allowing easier rollover of
the foot during push-off, including a reduced range of MTP motion. This is partic-
ularly beneficial for people suffering from motion-related pathologies like hallux
rigidus or joint arthritis. Another way of facilitating foot rollover is to increase the
forefoot offset. This is realized by curving the anterior of the shoe upwards, which
puts the toes into dorsiflexion and reduces the resistance of the midsole to bending,
because midsole bending is required later, and to a lesser extent, compared to
midsoles with a straight geometry. However, the effect of rocker and forefoot offset
on the load distribution at more distal joints has not been adequately assessed in the
literature to date.
The results outlined above show that the interplay between the technical features
of running shoes and the load distribution within the body can be complex and
subject specific. The gain in popularity of running as a recreational sport has meant
that the anatomical characteristics of runners have diversified greatly in recent
decades. Whereas initially, only well-trained competitive athletes were doing dis-
tance running, today people with very diverse fitness levels and backgrounds partake
in the sport. Therefore, the demands put on running shoes differ enormously between
individuals. However, researchers are only now beginning to understand how
individual runners with individual characteristics benefit from specific footwear
design features. Further improvement of this understanding will be needed to
develop footwear that is optimally customized to the needs of the individual runner,
allowing them to use a running technique that minimizes the load on their muscu-
loskeletal system (Nigg et al. 2015).
Running Shoes and Distance Running Performance
In most distance running races the main target is to finish within the shortest possible
time, or in other words, to run a given distance with the highest average speed.
Running at a given speed incurs a metabolic cost. Running at submaximal speeds
involves mainly aerobic respiration and can be estimated by measuring the oxygen
consumption of the runner (Cavanagh and Kram 1985). Better runners are capable of
running at the same speed while consuming less oxygen, or running at a faster speed
with the same oxygen consumption. The amount of oxygen consumed per kg of
body mass for a given running speed is used to define the running economy; elite
runners usually exhibit superior running economy when compared to less skilled
runners (Saunders et al. 2004).
The metabolic cost of running stems largely from the conversion of metabolic
energy into mechanical energy. This is necessary to propel the body forward and to
12 S. Willwacher
support the runner’s body weight. The mechanical work performed by the body is
usually divided into external and internal work (Cavagna et al. 1964; Cavagna and
Kaneko 1977). External work describes the work done to lift and accelerate the
CoM, while internal work refers to the work performed by moving the limbs relative
to the CoM. Wearing running shoes can affect the metabolic cost of running in both
positive and negative directions via different pathways, which will be described in
the following section.
Modern running shoes exhibit great variability with respect to their weight. A pair
of minimal shoes can weigh less than 200 g, while a pair of heavily cushioned shoes
that include sophisticated motion control elements can easily weigh more than 1 kg.
Adding weight to the feet affects both the internal and external work that has to be
performed during running. Firstly, the whole body mass of the runner is increased
thereby increasing the external work necessary to lift the body vertically and to
accelerate it forward.
Adding mass to the distal end of the legs increases the mass and moment of inertia
of the legs. Consequently, increased internal work is required to accelerate the legs
with respect to the center of mass and to lift the legs upward during the swing phase.
More concretely, it has been found that increasing shoe mass by 100 g per shoe leads
to a 1% increase in the energy consumption of a runner (Frederick 1984; Hoogkamer
et al. 2016).
It is frequently observed that runners change their running mechanics when
switching from traditional well-cushioned shoes to barefoot or minimalistic varie-
ties. Some components of these changes are a direct result of the lower footwear
mass, while others seem to be related to loss of the footwear’s cushioning and
stabilizing features. Changes in running mechanics include altered step length,
step frequency, ground contact times, and footfall pattern, along with associated
changes in joint loading and joint work patterns. Of particular interest in this case is
the change in footfall pattern when switching to barefoot shoes, where no impact
cushioning is provided by the shoe midsoles or by the running surface. In this
situation, most rearfoot runners modify their running pattern by striking the ground
with a flatter foot orientation and generally position the foot closer to their center of
mass. This adaptation necessitates shorter ground contact time and a higher step
frequency compared to the unmodified rearfoot running style.
From the perspective of energetics, it is interesting to note that these two running
styles differ in terms of how the kinetic energy of the ground impact is absorbed.
When running barefoot on a hard running surface, this energy has to be absorbed by
the musculoskeletal system. The energy can either be absorbed by the negative work
of muscle fibers and thereby be transformed into mechanically unusable forms of
energy (mostly heat), or it can be passively stored in the passive structures (tendons
and ligaments). The latter strategy has the advantage that some of the stored energy
can be used during the push-off phase, making the running motion more efficient.
However, it must be noted that energy storage and return within passive elements of
the musculoskeletal system also requires force generation by muscle fascicles in
order to allow tendon stretch. This muscle activation also incurs a metabolic cost.
When running in shoes with heel cushioning or when running on a soft running
surface, habitual rearfoot strikers make use of the energy absorbing properties of
midsole or surface materials. It has been shown that energy expenditure is decreased
when running barefoot on a treadmill that was covered with a 1 cm thick ethylene-
vinyl acetate (EVA) material, which is used as the midsole material in most running
shoes (Tung et al. 2014). Thus, an impact cost must be paid when running, and this is
reduced by cushioned footwear because some of the impact energy is passively
absorbed by the footwear materials.
Another important design feature for running performance is the bending stiffness
of the midsole. During the second half of the stance phase, the heel is lifting off from
the ground, resulting in a dorsiflexion of the metatarsal phalangeal joints and
consequently a bending of the midsole at the forefoot. In this process, energy is
absorbed and partly stored as strain energy within the midsole. It is generally
believed that most of the stored energy is returned after the foot has lost contact
with the ground, making it irrelevant for acceleration of the runner (Stefanyshyn and
Wannop 2016). Therefore, assuming that this energy was being wasted on work that
does not improve running performance, efforts were made to reduce such energy
absorption by stiffening the midsole via carbon fiber inserts. Using this approach,
decreases in energy absorption at the forefoot and 1% reductions in metabolic energy
consumption were observed in distance runners by Roy and Stefanyshyn (2006).
However, it was also shown that there is an individually optimal level of bending
stiffness, above which the energy consumption increases again.
More recent studies have shown that stiffening of the midsole does not always
lead to reductions in energy absorption at the MTP. Energy absorption is the product
of the bending moment created at the MTP and the MTP dorsiflexion range of
motion. Usually, with increased bending stiffness, the bending moment increases
while the dorsiflexion range of motion decreases. If the reduction in the range of
motion is proportionally greater than the increase in the bending moment, less
energy is absorbed at the MTP, but if there is a proportionally greater increase in
the bending moment, then more energy is absorbed at the MTP. This has been found
mainly for medium-stiffness midsoles, whereas at higher stiffness values, the reduc-
tion in the range of motion is usually proportionally greater than the increase in
bending moment. In some recent studies it has been shown that midsole bending
stiffness does not only affect the biomechanics of the MTP joint, it can also affect the
mechanics of the ankle, and to a lesser extent, the knee and hip (Willwacher. 2014a).
It has been shown that the moment arms of the GRF with respect to the joint centers
of the MTP, ankle, and knee vary in proportion to the longitudinal-bending stiffness;
therefore, the bending stiffness of footwear can be considered to have a gearing
function with respect to the lower extremities during running. In addition to the
effect on MTP and ankle joint kinematics, this gearing function determines the
working conditions of the leg muscle–tendon units during locomotion. Working
conditions include the force amplitudes, length, and contraction velocities of
muscle–tendon units. Depending on the strength characteristics, muscle architecture,
tendon stiffness, and other factors, the footwear bending stiffness can be tuned to an
optimal level for a given individual (Stefanyshyn and Wannop 2016). This tuning
must be considered in addition to the reduction of passive energy losses due to
14 S. Willwacher
midsole bending when attempting to find the optimal longitudinal bending stiffness
for an individual. Nevertheless, the complex interplay between technological, ana-
tomical, and neural features that is relevant to the task of minimizing energy
consumption is currently not well understood. Furthermore, the role of fatigue also
has to be considered in this task: Perhaps the capacity to modulate bending stiffness
during a run is needed, ideally as a function of runner fatigue.
As stated in the section on injury prevention, keeping joint motion within an
individually preferred movement path is considered an important principle of muscle
activation in runners (Nigg et al. 2015). If a footwear design feature constrains the
ability of a runner to maintain their ideal movement path, they will generally increase
the action of certain muscle groups to counteract this. Additional muscle activation
entails additional metabolic costs, which are detrimental to running economy.
However, the preferred-movement-path theory is a challenging one to prove, as it
is currently difficult to determine the individually preferred movement path of a
subject. Despite this, some indirect evidence is available from studies relating to
footwear comfort and energy consumption in runners. These studies indicate that
running in footwear that is perceived as more comfortable is associated with a lower
energetic demand. If it is considered that comfort is at least partly related to having
design features that do not conflict with the runner’s preferred movement path, then
these results provide indirect support for the hypothesis.
Future Directions
In the near future, researchers will need to work on methods that reliably identify the
preferred movement path of individual runners, and more generally, to find ways of
aiding footwear selection based on individual characteristics and needs. This will be
necessary for the development of footwear that minimizes the biomechanical risks
associated with its use. In addition, a better understanding of these risk factors is
urgently needed, necessitating more high-quality prospective studies aimed at iden-
tification of biomechanical risk factors that are influenced by footwear design. Smart
sensors implanted in shoes may help researchers with these efforts by providing
individualized data from real-world running scenarios. Such new sources of data
could result in the development of novel footwear technologies that are better
optimized to individual runners.
Cross-References

▶ Observing and Revealing the Hidden Structure of the Human Form in Motion
Throughout the Centuries

References
Bramble DM, Lieberman DE (2004) Endurance running and the evolution of homo. Nature 432
(7015):345–352
Carmody RN, Wrangham RW (2009) The energetic significance of cooking. J Hum Evol 57
(4):379–391
Carrier DR, Kapoor AK, Kimura T, Nickels MK, Satwanti SEC, So JK, Trinkaus E (1984) The
energetic paradox of human running and hominid evolution [and comments and reply]. Curr
Anthropol 25(4):483–495
Cavagna G, Kaneko M (1977) Mechanical work and efficiency in level walking and running. J
Physiol 268(2):467
Cavagna G, Saibene F, Margaria R (1964) Mechanical work in running. J Appl Physiol 19
(2):249–256
Cavanagh PR, Kram R (1985) The efficiency of human movement–a statement of the problem. Med
Sci Sports Exerc 17(3):304–308
De Wit B, De Clercq D, Aerts P (2000) Biomechanical analysis of the stance phase during barefoot
and shod running. J Biomech 33(3):269–278
Ferris DP, Louie M, Farley CT (1998) Running in the real world: adjusting leg stiffness for different
surfaces. Proc R Soc Lond B Biol Sci 265(1400):989–994
Frederick EC (1984) Physiological and ergonomics factors in running shoe design. Appl Ergon 15
(4):281–287
Hardin EC, van den Bogert AJ, Hamill J (2004) Kinematic adaptations during running: effects of
footwear, surface, and duration. Med Sci Sports Exerc 36(5):838–844
Herculano-Houzel S (2016) The human advantage: a new understanding of how our brain became
remarkable. MIT Press, Cambridge
Hintermann B, Nigg B, Sommer C, Cole G (1994) Transfer of movement between calcaneus and
tibia in vitro. Clin Biomech 9(6):349–355
Hoogkamer W, Kipp S, Spiering BA, Kram R (2016) Altered running economy directly translates to
altered distance-running performance. Med Sci Sports Exerc 48(11):2175–2180
Kluitenberg B, van Middelkoop M, Diercks R, van der Worp H (2015) What are the differences in
injury proportions between different populations of runners? A systematic review and meta-
analysis. Sports Med 45(8):1143–1161
Lieberman DE, Bramble DM (2007) The evolution of marathon running. Sports Med 37
(4–5):288–290
Lieberman DE, Venkadesan M, Werbel WA, Daoud AI, D’Andrea S, Davis IS, Mang’Eni RO,
Pitsiladis Y (2010) Foot strike patterns and collision forces in habitually barefoot versus shod
runners. Nature 463(7280):531–535
McDougall C (2009) Born to run. Knopf Doubleday Publishing Group. New York City
Nielsen RO, Rønnow L, Rasmussen S, Lind M (2014) A prospective study on time to recovery in
254 injured novice runners. PLoS One 9(6):e99877
Nigg BM (2010) Biomechanics of sport shoes. University of Calgary, Calgary
Nigg BM, Herzog W (2007) Biomechanics of the Musculo-skeletal system. Wiley, Chichester
Nigg BM, Liu W (1999) The effect of muscle stiffness and damping on simulated impact force
peaks during running. J Biomech 32(8):849–856
Nigg B, Baltich J, Hoerzer S, Enders H (2015) Running shoes and running injuries: mythbusting
and a proposal for two new paradigms:‘preferred movement path’and ‘comfort filter’. Br J
Sports Med 49(20):1290–1294
16 S. Willwacher
Perry S, Lafortune M (1995) Influences of inversion/eversion of the foot upon impact loading
during locomotion. Clin Biomech 10(5):253–257
Robertson DGE, Caldwell GE, Hamill J, Kamen G, Whittlesey SN (2013) Research methods in
biomechanics, 2nd edn. Human Kinetics, Champaign
Roy J, Stefanyshyn DJ (2006) Shoe midsole longitudinal bending stiffness and running economy,
joint energy, and EMG. Med Sci Sports Exerc 38(3):562
Saragiotto BT, Yamato TP, Junior LCH, Rainbow MJ, Davis IS, Lopes AD (2014) What are the
main risk factors for running-related injuries? Sports Med 44(8):1153–1163
Saunders PU, Pyne DB, Telford RD, Hawley JA (2004) Reliability and variability of running
economy in elite distance runners. Med Sci Sports Exerc 36(11):1972–1976
Shorten M, Mientjes MI (2011) The ‘heel impact’ force peak during running is neither ‘heel’ nor
‘impact’ and does not quantify shoe cushioning effects. Food Sci 3(1):41–58
Stacoff A, Denoth J, Kaelin X, Stuessi E (1988) Running injuries and shoe construction: some
possible relationships. Int J Sport Biomech 4(4):342–357
Stearne SM, McDonald KA, Alderson JA, North I, Oxnard CE, Rubenson J (2016) The foot’s arch
and the energetics of human locomotion. Sci Report 6:19403
Stefanyshyn DJ, Wannop JW (2016) The influence of forefoot bending stiffness of footwear on
athletic injury and performance. Food Sci 8(2):1–13
Trinkaus E (2005) Anatomical evidence for the antiquity of human footwear use. J Archaeol Sci 32
(10):1515–1526
Tung KD, Franz JR, Kram R (2014) A test of the metabolic cost of cushioning hypothesis during
unshod and shod running. Med Sci Sports Exerc 46(2):324–329
van Gent RN, Siem D, van Middelkoop M, van Os AG, Bierma-Zeinstra SM, Koes BW (2007)
Incidence and determinants of lower extremity running injuries in long distance runners: a
systematic review. Br J Sports Med 41(8):469–480. discussion 480
Willwacher S, König M, Potthast W, Brüggemann G-P (2013) Does specific footwear facilitate
energy storage and return at the metatarsophalangeal joint in running. J Appl Biomech 29
(5):583–592
Willwacher S, König M, Braunstein B, Goldmann J-P, Brüggemann G-P (2014a) The gearing
function of running shoe longitudinal bending stiffness. Gait Posture 40(3):386–390
Willwacher S, Regniet L, Mira Fischer K, Oberländer KD, Brüggemanna G-P (2014b) The effect of
shoes, surface conditions and sex on leg geometry at touchdown in habitually shod runners.
Food Sci 6(3):129–138
Airborne Movements: Somersaults
and Twists
Maurice R. Yeadon
Abstract
The angular momentum required for somersault is produced during the approach
and final contact phase prior to becoming airborne, while twist may be initiated
either during contact or in the aerial phase. The aerial phase of twisting and
non-twisting somersaults may be understood in terms of the two modes of rigid
body rotations: the twisting somersault and the wobbling somersault. Elite per-
formance of twisting somersaults is characterized as having a large aerial twisting
technique contribution arising from asymmetrical movements of the arms and
hips. Investigation of optimum technique using computer simulation indicates
that elite technique may be characterized as maximizing success within an
environment of motor system noise. Computer simulation has established limit-
ing performances to be triple straight somersaults in tumbling and high bar
dismounts, and the handspring triple salto in vaulting. Each elite performance
of a twisting somersault requires in-flight adjustments to be made on the basis of
feedback provided by the vestibular and visual systems.
Keywords
Twisting somersault • Computer simulation • Angular momentum • Gymnastics •
Trampolining • Springboard diving • Aerials • Optimum technique • Limiting
performances • Motor control • Variability • Twisting techniques • Biomechanics
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Somersault Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Airborne Rotational Movements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
M.R. Yeadon (*)

Loughborough University, Loughborough, UK
e-mail: M.R.Yeadon@lboro.ac.uk

DOI 10.1007/978-3-319-30808-1_128-1
2 M.R. Yeadon
Rigid Body Motions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Contact Twist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Aerial Twist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Tilt Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Optimum Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Limits of Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Control of Twisting Somersaults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Introduction
Almost all sporting movements have an airborne phase. In throwing movements,

the athlete is typically airborne at the release of the discus, shot, or basketball. When
running or jumping, the athlete spends longer in the airborne phase than in the
contact phase (e.g., sprinting, long jumping, high jumping). In acrobatic sports, the
aerial phase is often the main focus for judging the skill (e.g., tumbling, trampolin-
ing, diving). For the effective coaching of complex multiple somersaults with
multiple twists, it is advantageous to have a good understanding of the mechanics
of aerial movement.
During the contact phase, the athlete generates the vertical velocity at takeoff
which determines the time of flight. For movements that take off and land at the same
level (e.g., tumbling, trampolining), the time of flight t is given by the constant
acceleration equation:
t ¼ 2v=g
where v is the vertical takeoff velocity and g is the acceleration due to gravity. In
sports such as ski jumping, the vertical acceleration is not constant due to aerody-
namic drag and lift forces while airborne and the flight time will be somewhat greater
than that given by constant acceleration equations.
It is during the contact phase that the athlete also generates the required angular
momentum for the somersault rotation in the aerial phase (e.g., trampolining,
diving). In the case of dismounts from high bar, the angular momentum is built up
in a sequence of accelerated giant circles, whereas in tumbling and vaulting, angular
momentum is developed prior to the final contact phase during which it typically
decreases.
The angular momentum about the longitudinal axis required for producing twist
during a somersaulting movement may also be developed during the contact phase,
in which case the technique used is termed contact twist. On the other hand, the twist
may be produced entirely during the aerial phase where the technique is described
as aerial twist. In this case, the angular momentum for the twist is derived from the
somersault momentum generated during the contact phase.
Airborne Movements: Somersaults and Twists 3
State of the Art
The advent of digital computing facilities at universities 50 years ago allowed the
development of whole-body multi-jointed computer simulation models of sporting
movements. Miller (1970) used a model with symmetry about the sagittal plane to
simulate the airborne phase of somersaulting dives. Pike (1980) simulated a full
twisting forward dive in which the twist was initiated after takeoff using asymmet-
rical arm movement.
While numerical solutions of the equations of motion of a dynamical system are
sometimes the only viable approach, the availability of explicit analytical solutions
can provide insight into the general solution space. Yeadon (1993a) provided
analytical solutions in terms of elliptic functions for the two modes of aerial motion
of a rigid body.
Computer simulation of the contact phase has provided insight into the generation
of somersault momentum in diving (Kong 2005), trampolining (Burke 2014),
tumbling (King and Yeadon 2004), vaulting (Hiley et al. 2015), high bar (Arampatzis
and Brüggemann 1999; Hiley and Yeadon 2003a), and the aerials event in freestyle
skiing (Jones 2012). Most of these models are driven using joint torque time histories
since this makes it easier to impose realistic limitations on the strength of the athlete.
Yeadon (1993b) compared the capabilities of various techniques for initiating twist
during the contact phase by considering the relative angular twist momenta calcu-
lated using segmental inertias.
An understanding of aerial twisting techniques (Yeadon 1993c) has been gained
using an angle-driven simulation model of aerial movement (Yeadon et al. 1990a).
Asymmetrical movements of arms and hips produce tilt of the longitudinal twist axis
away from the vertical somersault plane, resulting in a component of angular
momentum about the twist axis. Contributions of various twisting techniques have
been quantified using the tilt angle (Yeadon 1993d) in trampolining, springboard
diving, tumbling, rings dismounts, high bar dismounts, and the aerials event in
freestyle skiing.
Optimization techniques such as simulated annealing (Goffe et al. 1994) and
genetic algorithm (Carroll 2001) have been used in the evaluation of torque-driven
models of the contact phase and in the determination of the generation of height and
somersault rotation in the flight phase of springboard diving, tumbling, and vaulting.
Optimizations for consistent performance in the presence of motor system noise
have been undertaken in high jumping (Wilson et al. 2007) and high bar dismounts
(Hiley and Yeadon 2016). Methods such as direct collocation (Ackermann and van
den Bogert 2010) for solving optimal control problems allow the determination of
complex dynamical multiphase movements in which only the initial and final
conditions are pre-specified.
Limiting performances in sports can be determined provided there are realistic
estimates of segmental inertias and strength of an athlete as well as limiting initial
conditions. Yeadon (2013) determined twisting limits for various techniques of produc-
ing twist in triple somersaults in the aerials event of freestyle skiing. Hiley et al. (2015)
4 M.R. Yeadon
found limiting movements in gymnastics vaulting, while Hiley and Yeadon (2005)
determined maximal dismounts from high bar.
Since there will always be some variation in technique in the performance of
a movement due to motor system noise, there will be a need to minimize the
propagated variation in the performance criteria using feedforward or feedback
techniques. Hiley and Yeadon (2016) showed that the feedforward giant circling
technique used prior to double layout somersault dismounts could be characterized
as that which maximized success in the presence of motor system noise. King and
Yeadon (2003) found the limits of variation for a layout somersault in tumbling that
could be corrected using feedback adjustments in the aerial phase. Yeadon and
Mikulcik (1996) showed how asymmetrical arm movements could prevent the
build-up of twist in unstable double layout somersaults, while Yeadon and Hiley
(2014) demonstrated the potential of feedback techniques for controlling twist in a
twisting somersault.
Somersault Production
Somersault is developed during contact with the apparatus prior to takeoff. For a
forward somersault on trampoline preceded by a plain jump there will be zero
angular momentum at the start of the contact phase so that all of the required angular
momentum will be developed during the contact phase. If there is little or no travel
during the flight phase of the somersault, there will be zero horizontal velocity of the
mass center at takeoff and so the mean force exerted by the trampoline bed on the
feet will be vertical during contact. During the descending phase of the bed, the mass
center moves forward and the horizontal force on the feet is forward. During the
ascending phase, the hips flex and the horizontal force on the feet is backward,
reducing the mass center horizontal velocity to zero at takeoff. Thus on average
during the contact phase, the mass center is a distance in front of the center of
pressure of the feet and the force F on the feet is vertical. This produces a moment Fd
about the mass center which generates the somersault angular momentum (Fig. 1).
Thus in a forward somersault without travel, the angular momentum is produced
primarily by the vertical bed reaction force during contact. During the descending
phase, little rotation occurs since the moments produced by the horizontal and
vertical forces are in opposing directions. During the ascending phase, both hori-
zontal and vertical forces contribute to the generation of somersault momentum. The
mechanics of backward somersaults are similar using hip and back hyperextension
rather than hip flexion.
In springboard diving, the production of somersault is similar to that of trampo-
lining. In a forward dive, the diver lands from a hurdle step approach with near-zero
angular momentum and with near-zero forward velocity of the mass center. At
takeoff, the mass center is travelling forward in order to take the dive a safe distance
away from the end of the board. As a consequence, the contribution of the horizontal
force to somersault generation is somewhat smaller. In an inward dive, the diver
depresses the board from a standing position and then during the ascending phase
Fig. 1 In a forward
somersault on trampoline, the
vertical force F on the feet
produces angular momentum
about the mass center G
generates the rotation and backward velocity of the mass center by flexing at the
hips. As a consequence, the contribution of the horizontal force to somersault
production is somewhat greater. In all cases, it is the vertical component of the
force on the feet that makes the major contribution to the somersault momentum in a
good performance.
In tumbling the gymnast generates horizontal velocity and angular momentum
during the approach using a roundoff and backward handspring prior to the final
backward somersaulting skill. In the final contact phase, the horizontal velocity
decreases, the vertical velocity increases, and the angular momentum about the
mass center also decreases. In this case, the angular momentum is not generated
during the contact phase: rather the loss of angular momentum is minimized while
generating the required vertical velocity for sufficient flight time. During contact, the
vertical reaction force of the floor will reduce the angular momentum, while the
horizontal force will tend to increase the angular momentum (Fig. 2). The loss of
angular momentum produced by the vertical force will typically be greater than the
increase due to the horizontal force so that there will be a net loss. This loss of
angular momentum will be smaller if the body changes from a hips-flexed position at
touchdown to a straight or arched configuration at takeoff.
Vaulting in gymnastics is similar to tumbling in that the body has high somersault
momentum at first contact with the vaulting table, and this decreases during the
contact phase while the vertical velocity increases and the horizontal velocity
decreases (Hiley et al. 2015). Since the hands are in contact rather than the feet,
the actions for forward and backward rotations are reversed with the body hyper-
extending and the shoulder angle opening for forward rotating vaults and the hips
flexing and the shoulder angle closing for backward rotating vaults.
In high bar dismounts, the angular momentum arises from rotating in an extended
position about the bar. This angular momentum is increased using a sequence of
6 M.R. Yeadon
Fig. 2 In a backward
somersault takeoff in
tumbling, the horizontal force
Fh tends to increase the
angular momentum about the
mass center G while the
vertical force Fv reduces it
Fig. 3 Angular momentum

about the high bar in a
backward giant circle is
increased by flexing after the
lowest point and extending
around the highest point
accelerated giant circles in which the body flexes after the lowest point and extends
around the highest point (Hiley and Yeadon 2003a; Fig. 3).
In the aerial event of freestyle skiing, the athlete gains velocity by skiing down an
in-run and then enters a curved takeoff ramp called a kicker which produces the
angular momentum. The angular momentum can be increased by arching during this
contact phase similar to takeoffs in trampolining and tumbling.
Airborne Rotational Movements
Rigid Body Motions
The rotational movement of an airborne rigid body may be described by successive

rotations through angles ϕ, θ, and ψ about transverse, frontal, and longitudinal axes
(Fig. 4). The somersault angle ϕ specifies the rotation of the longitudinal axis about a
fixed spatial horizontal direction; the tilt angle θ is the angle the longitudinal axis
makes with the plane perpendicular to the somersault axis; the twist angle ψ gives
the rotation about the longitudinal body axis.
There are two general motions of an airborne rigid body: the wobbling somersault
in which the twist angle oscillates and the twisting somersault in which the twist
angle continually increases. In a wobbling somersault, the body somersaults in the
same direction throughout, while it twists in one direction and then the other and tilts
in one direction and then the other (Fig. 5). Thus somersault is a monotonic function
of time, while tilt and twist are oscillating functions of time. In a twisting somersault,
both somersault and twist continually increase, while the tilt angle remains to the
same side of the vertical plane perpendicular to the somersault (Fig. 6).
These two general solutions approach each other along a singularity in which the
twist slowly builds up to a rapid half twist repeatedly. This instability constitutes a
potential problem for the control of non-twisting straight somersaults.
Fig. 4 Somersault (ϕ), tilt

(θ), and twist (ψ) rotations
8 M.R. Yeadon
Fig. 5 In a wobbling somersault, the somersault increases, while the tilt and twist oscillate
Fig. 6 In a twisting somersault, both somersault and twist increase
For twisting somersaults in the straight position, in which the lateral and frontal
moments of inertia are approximately equal, the somersault and twist rates are
related to the tilt angle by the equation:

IT
p¼ω 1 sin θ
IL
where p is the mean twist rate, ω is the mean somersault rate, IT is the transverse
moment of inertia, IL is the longitudinal moment of inertia, and θ is the mean tilt
angle. Thus for a ratio of 15:1 between moments of inertia and a tilt angle of 10 ,
there will be around 2½ twists per somersault while the body is straight.
When the lateral and frontal moments of inertia are not equal, there are analytical
solutions for twisting somersaults in terms of elliptic functions in which the somer-
sault, tilt, and twist rates oscillate with the period of a half twist (Yeadon 1993a). For
wobbling somersaults, there are analytical solutions in terms of elliptic functions
in which the somersault rate and twist angle oscillate with the tilt angle (Yeadon
1993a). Mikl and Rye (2016) derived equations describing 12 cases spanning the
range of possibilities for relative moments of inertia and initial conditions.
Bharadwaj et al. (2015) obtained an analytical solution for a special case of a nonrigid
body executing a twisting somersault.
By changing body configuration during flight, there exists the possibility of
changing a twisting somersault into a wobbling somersault and thereby effectively
stopping the twist. An appreciation of the features of the two general solutions of
rigid body motion can indicate how this may be accomplished.
Contact Twist
Twist may be generated during the contact phase by turning body segments in the
direction of twist so that a reaction torque acts on the feet building up angular
momentum about the longitudinal twist axis. If the twist is produced by moving
wide arms relative to the trunk, then adducting the arms in flight results in a whole-
body twisting angular velocity some 50% greater than the original arm angular
velocity (Yeadon 1993b). If the whole body initially rotates relative to the feet, the
twisting angular velocity after arm adduction will be three times that of the initial
angular velocity. It is possible to produce seven twists in a single somersault on
trampoline using this method although such a skill would not be performed as part of
a competition routine.
When there is angular momentum about both the somersault axis and the twist
axis at takeoff, rigid body mechanics indicates that during flight the longitudinal axis
will make an approximately constant angle with the non-horizontal angular momen-
tum vector. As a consequence, the body will be vertical at takeoff and landing and
tilted away from the vertical after a half somersault (Fig. 7). In tumbling and
trampolining with takeoff and landing on the feet, this tilt will be self-correcting,
whereas in twisting dives with 1½ somersaults, the tilt may be problematic.
In a twisting somersault, rigid body mechanics indicates that the tilt angle
between the longitudinal axis and the plane perpendicular to the angular momen-
tum vector will oscillate with each half twist since the lateral and frontal moments
of inertia are unequal (Yeadon 1993a). This nutation effect will be greatest when the
arms are held wide at takeoff and subsequently the tilt angle will increase to a
maximum after a quarter twist. If the arms are adducted at this time rather than
earlier, the tilt angle and the twist rate will be greater (Fig. 8).
If the body changes from a straight position to a piked position during a twisting
somersault, the motion can change into a wobbling somersault, and this will effectively
Fig. 7 During a contact twist, the visible tilt away from the vertical will disappear of its own accord
after one somersault
10 M.R. Yeadon
stop the twist. In the computer simulation shown in Fig. 9, the nutation effect is used
to boost the initial tilt angle and then the body pikes after 1½ twists so that the tilt
angle oscillates in the wobbling somersault mode allowing extension after 1½
somersaults when the body is vertical.
Fig. 8 Rather than adducting the arms immediately after takeoff (upper sequence), delaying arm
adduction until the ¼ twist position allows the tilt angle to increase by 3 resulting in a faster twist
(lower sequence)
Fig. 9 A backward 1½ somersault dive with 1½ twists in which the twisting somersault mode
changes into the wobbling somersault mode by piking
Aerial Twist
In order to convert a somersault movement into a twisting somersault, it is necessary

to tilt the longitudinal twist axis away from the vertical somersault plane so that
there is a component of angular momentum along the twist axis. This may be
achieved using asymmetrical movements of the arms after takeoff (Frolich 1979)
or asymmetrical movements of the hips (Yeadon 1993c). In the simulation depicted
in Fig. 10, the arms rotate from a symmetrical position producing tilt and starting the
twist. As one revolution of twist nears completion, the arm movement is reversed,
removing the tilt and stopping the twist.
In Fig. 11, the body extends from a piked position with wide arms, passing
through a side arch before straightening. This produces the tilt which results in 1½
twists when the arms are adducted. The tilt is removed using a combination of
asymmetrical hip and asymmetrical arm movements.
In Fig. 12 the body is twisting at takeoff, and the tilt angle is increased using a
symmetrical lowering of the arms parallel to the sagittal plane around the quarter
twist position. This increases the tilt angle by a factor of four and allows two twists to
be completed in the somersault.
Fig. 10 Asymmetrical arm movement produces aerial twist in a somersault
Fig. 11 Asymmetrical hip movement produces aerial twist in the second somersault of a double
somersault
12 M.R. Yeadon
Fig. 12 Symmetrical lowering of the arms parallel to the sagittal plane around the quarter
twist position increases the tilt angle by 10 allowing two twists to be completed in the backward
somersault
Tilt Contributions
Since the tilt angle is approximately proportional to the twist rate for a freely rotating
rigid body (Yeadon 1993a) it may be used to quantify the twist contributions of
contact and aerial techniques of competitive performances of twisting somersaults.
The contact contribution may be defined as the initial tilt angle at takeoff and aerial
contributions may be determined by simulating performances with asymmetries of
arms and hips removed and noting the tilt angle in mid-twist (Yeadon 1993d).
In 18 single somersaults with one twist in a competition gymnastics floor
exercise, the mean contributions to the tilt angle from contact and aerial techniques
were 30% and 70%, respectively. The aerial contributions arose primarily from a
symmetrical lowering of the arms with minor contributions from asymmetrical arm
and hip movements (Yeadon and Kerwin 1999). In eight full twisting single somer-
sault dismounts from high bar, aerial techniques accounted for more tilt than contact
techniques in every case, with the majority of aerial contributions arising from
asymmetrical movements of the arms and hips (Yeadon et al. 1990b). In eight
twisting double somersault dismounts from high bar, the twisting techniques varied
with the timing of the twist within the two somersaults (Yeadon 1997). Contact
contributions were larger when there was more twist in the first somersault. When
there was little or no twist in the first somersault, the major contribution came from
aerial techniques that comprised arm movements and asymmetrical hip movements.
In six full twisting double somersault dismounts from the rings, the majority of the
tilt was produced using aerial techniques that were predominantly asymmetrical
movements of the arms. Contributors to the subsequent removal of tilt were again
primarily asymmetrical arm movements (Yeadon 1994).
In four single and double somersaults on trampoline with up to two twists, the
contact contributions comprised less than 25% with symmetrical and asymmetrical
arm movements being the main aerial contributors (Yeadon 1993d). In eight reverse
1½ somersault dives with 2½ twists, seven produced the majority of the tilt using
aerial techniques which were predominantly asymmetrical movements of the arms
and hips, although the mean contact contribution amounted to one third of the total
tilt (Yeadon 1993e). In six triple somersaults with three or four twists in the aerials
event of freestyle skiing, two used contact techniques, two used aerial techniques,
and two used a combination of contact and aerial techniques (Yeadon 1989). In
summary elite performances of twisting somersaults are characterized primarily by
the use of aerial techniques even when the twist starts during the contact phase.
Optimum Technique
Koschorrek and Mombar (2012) used optimal control methods to generate natural-
looking platform dives by minimizing joint torques during the contact and aerial
phases of twisting and non-twisting somersault dives. While it is understandable
that such a minimization criterion will produce smooth changes in joint angles and
may therefore appear to look natural, the claim that this also provides insight into
technique is unwarranted. The normal operation of muscle enervation will result in
smooth movements, and if this aspect is modeled, there is no need to invoke torque
minimization. Indeed maximal effort is typically used in acrobatic movements rather
than minimizing joint torque (Hiley 2012). Three optimization criteria were inves-
tigated for the undersomersault on parallel bars by Hiley and Yeadon (2012). It was
found that the techniques used by two elite gymnasts closely resembled the optimi-
zation that minimized horizontal velocity at release.
Using a simple optimization criterion to determine optimum technique may result
in a solution that produces inconsistent performance due to motor system noise.
Incorporating such noise into the optimization process showed that a triple-piked
somersault dismount from high bar could be performed consistently (Hiley and
Yeadon 2008). Comparing the results of various optimization criteria with the
technique used by an elite gymnast, in a double somersault dismount from high
bar (Fig. 13), indicated that technique was characterized by maximizing success
within an environment of motor system noise (Hiley and Yeadon 2016).
Limits of Performance
In order to determine the limits of performance for a given type of sports movement
using computer simulation, it is necessary to have realistic information on approach
characteristics, athlete strength, and motor system noise. Hiley and Yeadon (2005)
showed that it was theoretically possible to perform a triple straight backward
somersault dismount from high bar, providing the release could be timed to within
28 ms. Brüggemann and Arampatzis (1993) determined that a quadruple tucked
backward somersault dismount was possible in principle. Heinen et al. (2011) found
that the double Jaeger piked (two somersaults behind the bar to recatch) was a
realistic possibility, while Čuk et al. (2009) showed that the Tkachev salto (forward
somersault to recatch from backward giant circle) may be possible.
14 M.R. Yeadon
Fig. 13 A straight double somersault dismount from high bar
Fig. 14 A handspring triple somersault tucked vault
King and Yeadon (2004) found that large linear and angular approach velocities
were key for maximizing rotation in tumbling and that a straight triple somersault
should be possible. In gymnastics vaulting, it was found that a handspring double
somersault with 1½ twists and a handspring triple somersault tucked (Fig. 14) were
realistic possibilities (Hiley et al. 2015).
In twisting somersaults, the number of twists that can be achieved is limited by
the time that the body can be extended in a straight position and so flight time and
somersault momentum will be limiting factors. In double somersaults on trampoline,
it is possible to use aerial twisting techniques to produce 3½ twists in the second
somersault when rotating forward (Fig. 15) and three twists in the second somersault
Fig. 15 A double forward somersault with 3½ twists in the second somersault produced by
asymmetrical arm and hip movements during flight
when rotating backward (Yeadon 2015). In triple somersaults in the aerials event of
freestyle skiing, six twists will be the limit according to Yeadon (2013).
Control of Twisting Somersaults
Each performance of a given acrobatic movement is unique in the sense that there is
always variability in a coordinated movement due to planning errors, execution
errors, and noise in the motor-sensory system (van Beers et al. 2004). As a conse-
quence, there is often a need to make adjustments during flight in order to meet the
required final somersault and twist rotations. Such adjustments are made using
feedback from the vestibular balance organs of the inner ear as well as visual
information obtained by making head movements to keep the landing in view during
flight (Yeadon and Knight 2012).
In a tumbling performance of a straight backward somersault, King and Yeadon
(2003) used computer simulation to show that in-flight feedback adjustments were
capable of correcting rotation errors of up to 8%. Hiley and Yeadon (2003b) found
that the scooped giant circle technique used by all finalists in an Olympic high bar
competition had a greater margin of variability than the traditional giant circle
technique for timing the release of a double somersault dismount.
In the unstable double straight somersault, the build-up of twist may be prevented
using asymmetrical arm movements providing the time delay in the feedback loop is
16 M.R. Yeadon
Fig. 16 Graphics of target full twisting backward somersault produced by asymmetrical arm
movement (upper sequence), perturbation arising from less arm abduction (middle sequence),
corrected simulation (lower sequence)
not greater than a quarter somersault (Yeadon and Mikulcik 1996). The same method
of continual control was used by Yeadon and Hiley (2014) to correct a perturbed
full twisting backward somersault (Fig. 16). In contrast a single symmetrical arm
adduction adjustment was capable of correcting perturbations to a 1½ twisting
forward somersault providing the feedback time delay was less than 200 ms (Fig. 17).
Future Directions
The use of computer simulation models may give further insight into twisting
somersaults providing such models incorporate realistic representations of segmen-
tal inertia parameters, joint torque-generating capacity, motor system noise, and
sensory feedback time delays. In tandem with such models exhibiting biofidelity,
there is a need for efficient algorithms for finding robust solutions to human movement
Fig. 17 Graphics of target 1½ twisting forward somersault produced by asymmetrical hip move-
ment (upper sequence), perturbation arising from more hip flexion (middle sequence), corrected
simulation (lower sequence)
problems in an environment of motor system noise and feedback time delay. There is
potential for using athlete-specific models to assess the viability of a proposed
movement and to determine personalized limiting movements. Additionally there
is scope to gain insight into learning mechanisms whereby near-optimal technique
can be achieved within a relatively small number (hundreds) of iterations.
18 M.R. Yeadon
Cross-References

▶ Estimation of the Body Segment Inertial Parameters for the Rigid Body Biome-
chanical Models Used in Motion Analysis
▶ Jumping for Maximum Distance and Height
▶ Jumping on Elastic Surfaces: Diving and Trampolining
▶ Long Swing Manoeuvres in Gymnastics
▶ Ski Jumping: Aerodynamics and Kinematics in Take-off and Flight
References
Ackermann M, van den Bogert AJ (2010) Optimality principles for model-based prediction of
human gait. J Biomech 43:1055–1060
Arampatzis D, Brüggemann G-P (1999) Mechanical and energetic processes during the giant
swing exercise before dismount and flight elements on the high bar and uneven parallel bars.
Bharadwaj S, Duignan N, Dullin HR et al (2015) The diver with a rotor. https://arxiv.org/pdf/1510.
02978.pdf. Accessed 14 Sep 2016
Brüggemann G-P, Arampatzis D (1993) Simulation of a quadruple backward somersault dismount
from the high bar. In: Bouisset S, Métral S, Monod S (eds) Abstracts, International society of
biomechanics XIVth congress, Paris, pp 222–223, 4–8 July 1993
Burke DJ (2014) The mechanics of the contact phase in trampolining. Doctoral dissertation,
Loughborough University. https://dspace.lboro.ac.uk/2134/16744. Accessed 14 Sep 2016
Carroll DL (2001) FORTRAN genetic algorithm driver. Downloaded from: http://cuaerospace.com/
carroll.ga.html
Čuk I, Atiković A, Tabaković M (2009) Tkachev Salto on high bar. Sci Gymnast J 1(1):5–13
Frolich (1979) Do springboard divers violate angular momentum conservation? Am J Phys
47:583–592
Goffe WL, Ferrier GD, Rogers J (1994) Global optimization of statistical functions with simulated
annealing. J Econ 60:65–99
Heinen T, Jeraj D, Vinken PM et al (2011) What it takes to do the double Jaeger on the high bar?
Sci Gymnast J 3(3):7–18
Hiley MJ (2012) Incorporating aspects of motor control in the optimisation of human performance.
Sci Motricité 75:49–57
Hiley MJ, Yeadon MR (2003a) Optimum technique for generating angular momentum in acceler-
ated backward giant circles prior to a dismount. J Appl Biomech 19:119–130
Hiley MJ, Yeadon MR (2003b) The margin for error when releasing the high bar for dismounts.
Hiley MJ, Yeadon MR (2005) Maximal dismounts from high bar. J Biomech 38:2221–2227
Hiley MJ, Yeadon MR (2008) Optimisation of high bar circling technique for consistent perfor-
mance of a triple piked somersault dismount. J Biomech 41:1730–1735
Hiley MJ, Yeadon MR (2012) The effect of cost function on optimum technique of the
undersomersault on parallel bars. J Appl Biomech 28:10–19
Hiley MJ, Yeadon MR (2016) Investigating optimal technique in the presence of motor system
noise: application to the double layout somersault dismount on high bar. J Sports Sci
34:440–449
Hiley MJ, Jackson MI, Yeadon MR (2015) Optimal technique for maximal forward rotating vaults
in men’s gymnastics. Hum Mov Sci 41:1730–1735
Jones PE (2012) The mechanics of takeoffs in the aerials event of freestyle skiing. Doctoral
dissertation, Loughborough University. https://dspace.lboro.ac.uk/2134/10784. Accessed
14 Sep 2016
King MA, Yeadon MR (2003) Coping with perturbations to a layout somersault in tumbling.
King MA, Yeadon MR (2004) Maximising somersault rotation in tumbling. J Biomech 37:471–477
Kong P-W (2005) Computer simulation of the takeoff in springboard diving. Doctoral dissertation,
Loughborough University. https://dspace.lboro.ac.uk/2134/9051. Accessed 14 Sep 2016
Koschorreck J, Mombaur K (2012) Modeling and optimal control of human platform diving with
somersaults and twists. Optim Eng 13:29–56
Mikl J, Rye DC (2016) Twist within a somersault. Hum Mov Sci 45:23–39
Miller DI (1970) A computer simulation of the airborne phase of diving. Doctoral dissertation,
Pennsylvania State University
Pike NL (1980) Computer simulation of a forward, full twisting dive in a layout position. Doctoral
dissertation, Pennsylvania State University
van Beers RJ, Haggard P, Wolpert DM (2004) The role of execution noise in movement variability.
J Neurophysiol 91:1050–1063
Wilson C, Yeadon MR, King MA (2007) Considerations that affect optimised simulation in a
running jump for height. J Biomech 40:3155–3161
Yeadon MR (1989) Twisting techniques used in freestyle aerial skiing. Int J Sport Biomech
5:275–284
Yeadon MR (1993a) The biomechanics of twisting somersaults. Part I: rigid body motions. J Sports
Sci 11:187–198
Yeadon MR (1993b) The biomechanics of twisting somersaults. Part II: contact twist. J Sports Sci
11:199–208
Yeadon MR (1993c) The biomechanics of twisting somersaults. Part III: aerial twist. J Sports Sci
11:209–218
Yeadon MR (1993d) The biomechanics of twisting somersaults. Part IV: partitioning performance
using the tilt angle. J Sports Sci 11:219–225
Yeadon MR (1993e) Twisting techniques used by competitive divers. J Sports Sci 11:337–342
Yeadon MR (1994) Twisting techniques used in dismounts from rings. J Appl Biomech 10:178–188
Yeadon MR (1997) Twisting double somersault high bar dismounts. J Appl Biomech 13:76–87
Yeadon MR (2013) The limits of aerial twisting techniques in the aerials event of freestyle skiing.
J Biomech 46:1008–1013
Yeadon MR (2015) Twist limits for late twisting double somersaults. In: Abstracts: XV
International symposium on computer simulation in biomechanics, Edinburgh, 9–11 July 2015.
http://iscsb2015.lboro.ac.uk/ISCSB_book_of_abstracts.pdf. Accessed 14 Sep 2016
Yeadon MR, Hiley MJ (2014) The control of twisting somersaults. J Biomech 47:1340–1347
Yeadon MR, Kerwin DG (1999) Contributions of twisting techniques used in backward somersaults
with one twist. J Appl Biomech 15:152–165
Yeadon MR, Knight JP (2012) A virtual environment for learning to view during aerial movements.
Comput Methods Biomech Biomed Eng 15:919–924
Yeadon MR, Mikulcik EC (1996) The control of non-twisting somersaults using configurational
changes. J Biomech 29:1341–1348
Yeadon MR, Atha J, Hales FD (1990a) The simulation of aerial movement – IV: a computer
simulation model. J Biomech 23:85–89
Yeadon MR, Lee S, Kerwin DG (1990b) Twisting techniques used in high bar dismounts. Int J Sport
Biomech 6:139–146
The Segmental Movements in Front Crawl
Swimming
Ross H. Sanders, Jordan T. Andersen, and Hideki Takagi
Abstract
This chapter explores the cooperative coordination of segmental motions that
optimizes performance in front crawl swimming. The chapter reveals how under-
standing of swimming technique and performance has advanced through the
application of technology. The mystery of how propulsion is achieved in
human swimming is being unraveled through observation of water flow associ-
ated with upper and lower limb motions using particle image velocimetry (PIV)
and computational fluid dynamics (CFD) combined with video-based three-
dimensional analysis of swimming motions. These methods are supplemented
with direct pressure measurement of forces acting on the hands and feet and in
combination provide input for computer simulations and programmed swimming
actions of robots. Improved understanding of the mechanisms of propulsion
emerging from these coordinated approaches will translate into improved
coaching methods for sprint and distance swimming with consideration of the
energy cost and physiological constraints.
Keywords
Freestyle swimming • Computational fluid dynamics • Robotics • Particle image
velocimetry • Three-dimensional motion analysis • Pressure measurement • Swim
coaching • Sports technology
R.H. Sanders (*) • J.T. Andersen

Exercise and Sport Science, Faculty of Health Sciences, The University of Sydney, Sydney, NSW,
Australia
e-mail: ross.sanders@sydney.edu.au; jordan.andersen@sydney.edu.au
H. Takagi
Faculty of Health and Sport Sciences, University of Tsukuba, Ibaraki, Japan
e-mail: takagi.hideki.ga@u.tsukuba.ac.jp

DOI 10.1007/978-3-319-30808-1_132-1
2 R.H. Sanders et al.
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Propulsion in Front Crawl Swimming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Contributions of the Hand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Influence of Hand Motion by Other Body Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Involvement of the Trunk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Motions of the Lower Limbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Introduction
It is probably fair to say that our knowledge and understanding of segmental motions
in swimming and their contribution to performance remain at a rudimentary level
due to the complexity of the dynamics of motion of irregular objects in fluids.
Nevertheless, for over 50 years coaching practice with respect to technique devel-
opment of swimmers has been based on application of rationales rooted in scientific
principles, and there has been considerable confidence in the ideas prevailing at any
time. However, while the scientific principles have been sound, their application to
swimming has often been inappropriate and based on untenable assumptions. This
has been the case particularly with regard to the mechanisms of propulsion. Unfor-
tunately, misunderstanding of the mechanisms of propulsion has led to inappropriate
coaching practices with regard to technique development. Fortunately, swimmers
have optimized technique effectively despite, rather than because of, our lack of
knowledge of the mechanisms of propulsion.
In this chapter we look at the motions of body segments and their contribution to
swimming performance in the light of evidence from video-based three-dimensional
motion analysis and recent application of technologies involving direct pressure
measurement, observation of fluid motion using particle image velocimetry (PIV),
computer simulation, robotics, and computational fluid dynamics (CFD).
State of the Art
The greatest challenge to researchers examining propulsion in swimming is that of

measurement. Propulsion is the result of water flow that is manipulated by swimmers
in a way that hydrodynamic reaction forces are directed towards the swimming
direction. Water moves with turbulent flow all around the swimmer and sophisticated
techniques are required to measure the resulting fluid forces. In addition, the
air/water interface presents a barrier for three-dimensional motion analysis of swim-
ming technique. Due to the problems of capturing body markers clearly in turbulent
water researchers have resorted to manual digitization of individual video frames to
collect kinematic data (e.g., Sanders and Psycharakis 2009; McCabe et al. 2011;
The Segmental Movements in Front Crawl Swimming 3
McCabe and Sanders 2012). This is very time consuming and limits research output
significantly.
Fortunately, technology and new techniques are emerging that allow researchers
to acquire data with increasing accuracy, reliability, and speed. Wearable inertial
magnetic sensors, or inertial magnetic measurement units (IMMUs), produce reli-
able three-dimensional kinematic data of swimming motions (Fantozzi et al. 2016).
IMMUs are protected by waterproof cases, enabling researchers to acquire kinematic
data without cameras (Fantozzi et al. 2015). Water flow across body segments can be
measured using differences in pressures on either side of a segment. Specialized
gloves fitted with pressure sensors have been developed to measure flow over the
hand in front crawl swimming (Takagi and Wilson 1999; Takagi and Sanders 2002).
This technology has been adapted to enable measurement of pressure differences
across the feet in the eggbeater (Kawai et al. 2016) and breaststroke kicks
(Tsunokawa et al. 2015). PIV allows researchers to visualize the flow of water
particles (Matsuuchi et al. 2009) providing insights into the production of force
associated with acceleration of masses of water. Computer simulations and CFD are
being used to calculate fluid forces around a human swimming model created from
anthropometric and kinematic data of real swimmers (Nakashima et al. 2007; Kudo
et al. 2013). These technologies are producing “snapshots” of various aspects of
swimming propulsion that, when put together, create a more detailed picture of
human swimming than has been possible in the past.
Propulsion in Front Crawl Swimming
Contributions of the Hand
When one considers swimming technique, emphasis is most commonly placed on

the motion of the hand as the main instrument of propulsion. The contribution of the
other body segments is to enable the hand to move through a path that maximizes
propulsion without disturbing the alignment of the body. The path of the hand has
received considerable attention. Prior to 1968, coaches generally taught swimmers to
enter the hand well in front of the head and to then pull the hand directly backwards
along the midline of the body with the palm oriented perpendicular to the line of
motion (see, for example, Colwin 1992). There were several logical elements of
rationale: (i) forces produced would be predominantly in the swimming direction,
(ii) propulsion by drag forces would be the most efficient technique, (iii) having the
hand perpendicular to the line of motion maximized the propulsive forces, and
(iv) moving the hand along the midline of the body minimized the torques tending
to rotate the body out of alignment with the swimming direction and thereby
minimized resistive drag. The rationale seemed sound and was readily accepted
and applied by coaches in their poolside demonstrations. However, Counsilman
(1968) filmed the underwater path of the hands of elite front crawl swimmers and
showed that they scribed a curved path during the pull. This led to the concept that
good swimmers are “scullers” and use a series of lateral “sweeps” with the hand
oriented with small angles of attack to generate lift forces constituting a major
contribution to propulsion.
Computer simulations by Hay et al. (1993), supported by video-based digitized
2D–experimental data from side and underneath views of the swimmers (Liu et al.
1993), stimulated reassessment of the “good swimmers are scullers” idea by show-
ing that the curvilinear path is not a consequence of deliberate sweeping actions
across the body. Indeed, the hand path is strongly influenced by body roll and
swimmers actions straighten the path considerably to remove much of the effect of
body roll.
In subsequent studies, the “quasi-static” approach (Berger et al. 1995; Sanders
1999; Schleihauf 1979) to analysis of digitized data of the three-dimensional path
and orientation of a hand, in conjunction with its corresponding lift and drag
coefficients, has revealed that the angle of attack is mostly large during the under-
water stroke. Thus, while lift may contribute to the propulsive force, drag forces
would make a greater contribution to propulsion. However, in the “quasi-static”
approach steady flow conditions are assumed. That is, the effect of accelerations,
turbulence, and production and shedding of vortices are not considered. Further, it is
assumed that the flow of water near the hand is not influenced by the angular motion
of the hand or by the adjoining limbs, i.e., the forearm and arm.
Another assumption in the “quasi static” approach is that the flow of water around
the hand is directly opposite the direction of hand motion. However, Toussaint et al.
(2002) provided evidence of flow along the long axes of the upper limbs during the
pull. Toussaint suggested that water is “pumped” along the arm and forearm,
establishing pressure differences between the palmar and dorsal surfaces of the
hand as well as the forearm and arm.
Recognition of the inadequacy of the “quasi static” approach has led to the use of
emerging technologies to take into account the complex and dynamically changing
flow patterns. One of these technologies is computational fluid dynamics (CFD).
Sato and Hino (2013) have conducted CFD analyses on actual hand path data of elite
swimmers in competition using stereo cameras. The “S”-shaped hand path of Ian
Thorpe was found to be very efficient, particularly during the upsweep phase of the
underwater stroke and more sustainable than the “I” pull.
To investigate the differences in propulsion achieved by “S”-shaped and “I”-
-shaped hand paths (Fig. 1), Takagi et al. (2014) replicated the swimming motion of
swimmers’ upper limbs using a robotic hand, forearm, and upper arm with five
degrees of freedom controlled independently by a computer. They identified two
important mechanisms.
As shown in Fig. 2, the first was the generation of a Kármán vortex street, i.e., a
flow phenomenon in which clockwise or counterclockwise vortices are alternately
shed from the side of the little finger or the thumb. A Kármán vortex street is
known to be generated when the hand moves in a linear manner with a large angle
of attack as in the “I”-shaped stroking pattern. At that time of shedding each vortex,
the pressure difference between the palm and dorsal sides increases, producing
a drag force which contributes to the thrust force. According to the results of
Fig. 1 Illustrations of
“S”-shaped and “I”-shaped
S-Stroke I-Stroke
hand paths and three phases
(Downsweep, Insweep,
Upsweep) during “S” shape Downsweep
Uh Insweep
Upsweep
Kudo et al. (2013), these irregular oscillations occur in the hydrodynamic forces
while the hand is accelerating.
The second is an unsteady lift force generated when the hand changes direction
when scribing the “S” (Fig. 3). When the leading edge of the hand changes from the
thumb side to the little finger side, the clockwise flow around the hand (bound
vortex) changes to counterclockwise. This coincides with the shedding of a vortex
from the hand on the thumb side. This circulation adds to the flow due to the linear
velocity of the hand to increase the surface velocity; the surface pressure decreases,
and a lift force is produced.
Thus, it may be concluded that skilled swimmers can benefit from both an
“S”-shaped hand path and an “I”-shaped hand path. Simulations have shown that
the “I”-shaped pull produces large forces and is therefore suited to sprint front crawl
swimming. The “S”-shaped pull produces smaller forces but with less energy loss to
the water. Thus, the “S”-shaped pull is typically used by swimmers in middle-
distance and long-distance events (Takagi et al. 2016).
Fig. 2 Diagrammatic illustration of the flow and vortex during “I”-shaped stroking pattern
Fig. 3 Diagrammatic
illustration of the flow and
vortex during “S”-shaped
stroking pattern
Insweep Upsweep
Ultralow pressure
area Main flow velocity
Lift Moving hand
velocity
Relative flow velocity
Circular vortex Circular vortex
(Clockwise) (anticlockwise)
Shedding vortex
Influence of Hand Motion by Other Body Segments
Having established that the path of the hand is important in front crawl swimming,
we must now look at how the motions of the upper limbs influence the hand path.
The forearm plays an important role in moving the hand through the desired path and
its motion, in turn, is influenced by the actions at the shoulder joint via the upper arm
with respect to the trunk, that is, an internal reference frame, and the flexion and
extension of the elbow. The orientation of the upper limb lever system with respect to
the external reference frame is also influenced strongly by the rotation of the upper
body about its long axis commonly referred to as “shoulder roll” (Psycharakis and
Sanders 2008). These actions, in combination with subtle pronation and supination
of the hand via rotation of the forearm about its long axis, change the pitch angle of
the hand (i.e., the angle between the plane of the hand and the direction of motion of
the hand), influencing hydrodynamic forces on the hand and forearm (Matsuuchi and
Muramatsu 2011; Nakashima et al. 2007).
Elbow flexion and extension make use of the long lever of the forearm to change
the hand’s trajectory (Figueiredo et al. 2011, 2013a). Nakashima et al. (2012) have
shown through simulation that the high elbow position during entry and at the instant
of catch is efficient and associated with the “S”-shaped pull and distance swimming.
This high elbow position is due to some elbow flexion and internal rotation of the
shoulder (Sanders 2002; Payton et al. 1999, 2002). The elbow is somewhat less
flexed for the “I”-shaped pull common in sprinting. The upsweep is executed
primarily by elbow extension, permitting the swimmer to achieve a large amount
of thrust at the end of the underwater stroke (Figueiredo et al. 2013a; Suito et al.
2008).
The motion of the entire upper limb segmental chain is greatly influenced by the
actions at the shoulder joint. Although shoulder extension contributes most to hand
velocity (Payton et al. 2002), other changes at the shoulder joint have a large impact
on a swimmer’s stroke. Shoulder roll during the insweep results in an “S” pull
(Payton et al. 1999) whereas slight abduction of the shoulder produces more of an
“I”-shaped hand path. Internal rotation of the shoulder is also greater in the “S”-
-shaped pull than the “I”-shaped pull (Payton et al. 2002). Shoulder roll contributes
to the outsweep and upsweep and shoulder abduction contributes to hand velocity
only towards the end of the upsweep (Suito et al. 2008).
Swimmers have less shoulder roll when sprinting than when distance swimming
(approximately 106 degrees versus 111 degrees, respectively) (McCabe and Sanders
2012; McCabe et al. 2011). Greater shoulder roll tends to generate a curvilinear path
of the hand, and partly accounts for the observation that the “S”-shaped pull is
common among distance swimmers.
The breathing action tends to produce an asymmetry of body roll for many
swimmers (Psycharakis and Sanders 2008; Psycharakis and McCabe 2011; Seifert
et al. 2005). Psycharakis and Sanders (2008) reported body roll to the breathing side
of about 57 degrees and to the nonbreathing side of about 49 degrees during 200 m
swims. Interestingly, these data were from stroke cycles in which there was no breath
taken. This means that the bias exists even when the swimmer has no intention of
taking a breath, as is the case for much of the distance comprising a 50 m sprint. The
cause of the bias is quite obvious – the swimmer turns the head enough to clear the
water to breathe. Nevertheless, the hollow that forms naturally during high-speed
swimming enables swimmers to breathe with the mouth very close to water level,
and perhaps even below the level of the water undisturbed by the swimmer.
Although the head can be rotated around its long axis independently of the upper
trunk and of the rotation of the shoulder, there is a natural tendency for the upper
body to be rotated further to the breathing side than the nonbreathing side.
The recovery of the upper limbs does not affect propulsion directly. However, the
limb motions in the recovery can significantly impact performance by influencing
body rotation. Some swimmers have a “high elbow” action characterized by a flexed
elbow that positions the hand below the level of the elbow throughout the recovery
from the instant of hand exit to the instant of hand entry, which could contribute to
body roll efficiency (Yanai 2004). In swimming textbooks, it is common to see the
high elbow recovery recommended as a means of reducing the moment of inertia of
the upper limbs about a vertical axis of the body and thereby reducing the tendency
for counter rotations that affect the alignment of the body and increase resistive drag
(Maglischo 2003). However, some excellent world champions, including the Amer-
ican middle- and long-distance swimmer Janet Evans, had a style in which the elbow
was considerably extended throughout recovery. Clearly, body alignment can be
maintained from other actions that balance the rotational torques produced by the
arm recovery.
Swimmers also differ with respect to the entry. Some enter the hand just above the
head and glide it forward while others enter the hand further in front of the head (i.e.,
“over the top”). The high elbow recovery tends to be associated with the former while
the extended elbow tends to be associated with the latter. The latter may be conducive
to reducing the time between entry and the catch due to the downward and somewhat
circular motion of the hand (as viewed from the side) as it enters the water, continuing
into the downward and backward motion of the catch. This makes the straight-arm
recovery attractive to sprinters. For example, short course 50 m and 100 m freestyle
world record holder Stefan Nystrand had a very distinct straight-arm recovery.
Because of the downward motion at entry, less time is spent in the nonpropulsive
phase between entry and catch, and the swimmer can move into the propulsive
phases sooner than with a high elbow recovery and glide. The applied literature
supports the straight-arm recovery for sprinting (SwimOutlet.com; Palmer 2009).
However, evidence of the benefits of either recovery technique remains anecdotal.
To maximize propulsive forces, the hand moves backwards very quickly follow-
ing the catch. To maximize the time of force application in sprinting, there is little
pause between the entry and catch. However, in distance events, a high stroke rate
cannot be maintained without adverse effects of fatigue on performance (Alberty
et al. 2005; Aujouannet et al. 2006) and reduction in the propulsive forces generated
(Seifert et al. 2007; Alberty et al. 2009). This means that distance swimmers have a
pause following entry during which time the recovering hand “catches up” (Chollet
et al. 2000). That is, the upper limbs on either side of the body are not exactly out of
phase. Thus, the coordination of the right and left upper limbs changes consider-
ably between sprint and distance swimming (Chollet and Seifert 2011). This “catch
up” is reflected in a change in the Index of Coordination (Chollet et al. 2000)
defined as the “time lag between the beginning of propulsion with the first right
arm stroke and the end of propulsion with the first left arm stroke, and between the
beginning of propulsion with the second left arm stroke and the end of propulsion
with the first right arm stroke” (Chollet and Seifert 2011). It is common to have an
overlap in the production of force by the right and left upper limbs in sprinting,
while there is a short period during which no propulsive force is produced by the
upper limbs in distance swimming (Schnitzler et al. 2008; Barbosa et al. 2015;
Figueiredo et al. 2013b).
Involvement of the Trunk
This brings us to the motion of the lower trunk. Psycharakis and Sanders (2008) have
shown that there is some independence of the magnitude of the hip rotation with
respect to the shoulder rotation. This is counterintuitive as one would expect the
trunk to rotate through an equivalent angle along its whole length. Sanders and
Psycharakis (2009) proposed that the much smaller rotation of the hips about the
long axis of the trunk than the shoulders was due to the reaction of the kicking
actions which dampen the transmission of the rotation from the shoulders to hips.
Several other researchers have echoed this effect of the flutter kick in front crawl
swimming. Vorontsov and Rumyantsev (2000) and McCullough et al. (2009) stated
that one role of the leg movements is to neutralize forces created by the arm stroke
and Maglischo (2003) believes that the flutter kick helps to stabilize the hips.
Sanders and Psycharakis (2009) explored the possibility that the trunk plays a
very active role in the transmission of motion to the feet in front crawl swimming.
This is the case in butterfly swimming in which the body undulates with timing such
that a wave is transmitted along the body culminating in the kick (Sanders et al.
1995). Sanders and Psycharakis (2009) proposed that in front crawl swimming a
torsional wave might serve a similar purpose. However, the difference in phase of the
hip rotation relative to the shoulder rotation was not sufficient to support that idea.
Nevertheless, there was a definite sequencing of motion of the hip, knee, and ankle
indicating progression of a torsional wave from the hips towards the ankles contrib-
uting to the kicking action. Further, the body wave in the caudal direction was at a
slower velocity among the faster swimmers. This is interesting given the knowledge
that the body wave velocity of marine animals matches the swimming velocity more
closely with increasing swimming mechanical efficiency and corresponding physi-
ological economy (Sfakiotakis et al. 1999).
Motions of the Lower Limbs
Despite the rhythmical “wave-like” actions of the lower limbs in front crawl
swimming that contribute to the kick, the evidence suggests that the kicking action
from the hips is at considerable physiological expense (Holmer 1974; Sanders 2007).
Although the actions of the thighs and shanks are out of phase, thereby opening the
possibility of energy transmission from thighs to shanks, the maximum mechanical
energy possessed by the thighs is small compared to the maximum mechanical
energy of the shanks. This means that rather than a flow of mechanical energy from
thighs to shanks in a wave-like manner, most of the energy gained by the shanks has
to be from the effort of the muscles acting about the knee joints. Indeed, Matsuda
et al. (2016) observed high levels of activation of the rectus femoris muscle to slow
down knee flexion and initiate knee extension during flutter kicking in experienced
competitive swimmers. The same pattern was also observed from the biceps femoris
muscle nearing the end of knee extension and into the initiation of knee flexion. This
is in contrast to the transmission of energy from the trunk to the thighs and to the
shanks via “wave-like mechanism” in butterfly swimming (Sanders et al. 1995). This
difference partly explains why butterfly swimming is almost as efficient as front
crawl swimming at the elite level. Note that Michael Phelps’ personal best time in
the 200 m butterfly is only approximately 8 s slower than his best 200 m freestyle
time. This is despite the need to perform work to raise the upper body to breathe in
butterfly swimming.
Like the propulsion gained from the hand, understanding the mechanisms of
propulsion gained from the kicking action of the lower limbs is emerging with
applications of technology to observe water flow. Propulsion in human swimming
is likely to be generated by accelerating masses of water backwards including masses
of swirling water in vortices (Matsuuchi et al. 2009). Rudimentary methods have
evolved into more sophisticated visualization, and modeling of water flow, to
understand these mechanisms. Since the movements of the lower limbs in the
dolphin kick are similar to the flutter kick (Arellano et al. 2002), insights into the
propulsive mechanisms in the flutter kick may arise from analyses of propulsion in
the dolphin kick. Arellano (1999) and his research group were among the first to
observe water flow changes in human swimming by injecting bubbles into the water
from a tube attached to a swimmer’s toe while the swimmer performed underwater
dolphin kicking. The bubble traces showed vortices shed into the swimmer’s wake at
the end of both the upbeat and the downbeat of the dolphin kick. A considerably
larger vortex was shed during the downbeat than upbeat, suggesting a mechanism for
the greater thrust that is achieved during the downbeat (Arellano et al. 2002).
Arellano (1999) also observed a vortex created and shed from the feet during the
downbeat of the flutter kick using a “bubble wall” through which the swimmers pass.
The formations of vortices in human swimming are similar to those that marine
animals use for propulsion (von Loebbecke et al. 2009a). These visualization
techniques helped guide future research and the development of increasingly sophis-
ticated technology to observe and measure water flow.
Visualization and simulation of water flow during a dolphin kick are also used to
explain how propulsion is generated by the lower limbs. Particle image velocimetry
(PIV) provides a representation of flow velocity by capturing the movement of water
particles illuminated with a laser. During the downbeat of the dolphin kick, a vortex
moves along the dorsal aspect of the knees towards the feet. The swimmer then kicks
into the vortex and “smashes” it with the foot with the subsequent upbeat. Propulsion
is generated from the energy of the rotating mass of water. This process is called
vortex recapturing (Hochstein and Blickhan 2011) and its propulsive thrust has been
quantified using CFD modeling (Pacholak et al. 2014). CFD techniques have also
Fig. 4 CFD visualization of vortices shed from the downbeat of the dolphin kick. At the end of the
downbeat, a vortex ring is shed from the feet (left). Water particles create a propulsive jet backwards
(middle). A 2D slice through the vortex ring shows counter-rotating vortices that are responsible for
the propulsive jet (right – solid lines represent counterclockwise rotation, dotted lines represent
clockwise rotation) (Adapted from von Loebbecke et al. (2009b))
been used to model the propulsion from a vortex ring that is shed into a swimmer’s
wake in dolphin kicking (Fig. 4, left). When viewed from the side, a 2D slice through
the vortex ring reveals two vortices rotating in opposite directions (Fig. 4, right). The
vortex ring creates a propulsive jet that thrusts the swimmer forward (Fig. 4, middle)
(Hochstein and Blickhan 2011; Ungerechts et al. 2000; von Loebbecke et al. 2009b;
Arellano et al. 2002; Cohen et al. 2009).
In addition to the contribution to propulsion of the lower limbs in front crawl
swimming, the lower limb motions have an important function in maintaining
streamlined body alignment to minimize resistive drag. In particular, the rolling of
the hips facilitates kicking in oblique planes rather than being constrained to kicking
in a vertical plane. This affords the opportunity to produce torques about the vertical
axis of the body to offset the torques produced by the upper limbs during the pull and
recovery phases of the stroke (Yanai 2001b). The ability of the lower limbs to
compensate in this way is evident when swimmers have difficulty maintaining
alignment without “wobbling” during training drills with constraints at the ankles
(e.g., rubber bands that hold the ankles together and prevent flutter kicking). This
difficulty also besets Paralympic swimmers with reduced function in the lower limbs
(Fulton et al. 2009). Further, it has been hypothesized that one of the functions of the
six-beat kick is to assist the rotation of the shoulders and facilitate body roll (Yanai
2003).
The flutter kick also produces rotation about the horizontal axis of the body to
help reduce resistive drag by maintaining horizontal body alignment. The torques
produced by the kick offset those produced by the arm actions and the buoyancy of
the upper trunk (Yanai 2001a).
Future Directions
To effectively measure and fully appreciate how propulsion is generated, it is critical

to understand water flow characteristics. Swimming research is moving towards
demystifying the mechanisms of propulsion with the technologies that are available.
CFD and PIV have provided exciting new ways to examine the mechanisms of
propulsion in swimming. The way we measure kinematics of swimmers is also
changing dramatically. Motion capture techniques have progressed significantly
over the past 30 years. The development of automatic digitizing systems that work
reliably for data capture underwater and at the interface of air and water is advancing.
Inertial-magnetic measurement units, or IMMUs, allow for measurement of kine-
matics without the issues that are inherently involved in video-based techniques.
IMMU systems are becoming more reliable for quantifying swimming movements
(Fantozzi et al. 2016).
The pioneering work of Takagi et al. (2014) using a combination of approaches
including three-dimensional cinematography, simulation, CFD, PIV, robotics, and
pressure-sensing devices ushers a new era of understanding how swimming perfor-
mance can be optimized. The challenge for researchers of swimming is to convey the
knowledge effectively so that it can be applied by coaches and swimmers to
maximize performance. For example, coaches and swimmers need to understand
the technique required to optimize performance for specific event distances and to
develop swimmers’ techniques accordingly.
Cross-References
▶ Cross-platform Comparison of Imaging Technologies for Measuring Musculo-

skeletal Motion
▶ Movement Analysis of Scull and Oar Rowing
▶ Sprinting and Running at Maximum Speed
References
Alberty M, Sidney M, Huot-Marchand F, Hespel JM, Pelayo P (2005) Intracyclic velocity varia-
tions and arm coordination during exhaustive exercise in front crawl stroke. Int J Sports Med
26(06):471–475
Alberty M, Sidney M, Pelayo P, Toussaint HM (2009) Stroking characteristics during time to
exhaustion tests. Med Sci Sports Exerc 41(3):637–644. doi:10.1249/MSS.0b013e31818acfba
Arellano R (1999) Vortices and propulsion. In: Sanders R, Linsten J (eds) Applied proceedings:
swimming. XVII International symposium on biomechanics in sport, Perth, 1999. Edith Cowan
University Press, pp 53–65
Arellano R, Pardillo S, Gavilán A (2002) Underwater undulatory swimming: kinematic character-
istics, vortex generation and application during the start, turn and swimming strokes. In: Pro-
ceedings of the XXth international symposium on biomechanics in sports, Universidad de
Granada
Aujouannet YA, Bonifazi M, Hintzy F, Vuillerme N, Rouard AH (2006) Effects of a high-intensity
swim test on kinematic parameters in high-level athletes. Appl Physiol Nutr Metab
31(2):150–158
Barbosa TM, de Jesus K, Abraldes JA, Ribeiro J, Figueiredo P, Vilas-Boas JP, Fernandes R (2015)
Effects of protocol step length on biomechanical measures in swimming. Int J Sports Physiol
Perform 10(2):211–218
Berger MA, de Groot G, Hollander AP (1995) Hydrodynamic drag and lift forces on human hand/
arm models. J Biomech 28(2):125–133
Chollet D, Chalies S, Chatard JC (2000) A new index of coordination for the crawl: description and
usefulness. Int J Sports Med 21(1):54–59
Chollet D, Seifert L (2011) Inter-limb coordination in the four competitive strokes. In: Seifert L,
Chollet D, Mujika I (eds) The world book of swimming: from science to performance. Nova
Science, New York, pp 153–172
Cohen R, Cleary PW, Mason BR (2009) Simulations of human swimming using smoothed particle
hydrodynamics. In: 7th international conference on CFD in the minerals and process industries,
Melbourne, Australia
Colwin C (1992) Swimming into the 21st century. Human Kinetics Publishers, Champaign
Counsilman JE (1968) The science of swimming. Prentice Hall, Englewood Cliffs
Fantozzi S, Giovanardi A, Borra D, Gatta G (2015) Gait kinematic analysis in water using wearable
inertial magnetic sensors. PLoS One 10(9):e0138105
Fantozzi S, Giovanardi A, Magalhães FA, Di Michele R, Cortesi M, Gatta G (2016) Assessment of
three-dimensional joint kinematics of the upper limb during simulated swimming using wear-
able inertial-magnetic measurement units. J Sports Sci 34(11):1073–1080
Figueiredo P, Nazário R, Cereja A, Vilas-Boas JP, Fernandes R (2011) The effect of fatigue on the
200 m front crawl underwaterstroke pattern. In: Vilas Boas JP, Machado L, Kim W et al. (eds)
XXIX international conference on biomechanics in sports, Porto, Portugal, vol 1, ISBS-
conference proceedings archive
Figueiredo P, Sanders R, Gorski T, Vilas-Boas JP, Fernandes R (2013a) Kinematic and electromyo-
graphic changes during 200 m front crawl at race pace. Int J Sports Med 34(01):49–55
Figueiredo P, Toussaint H, Vilas-Boas JP, Fernandes R (2013b) Relation between efficiency and
energy cost with coordination in aquatic locomotion. Eur J Appl Physiol 113(3):651–659
Fulton SK, Pyne DB, Burkett B (2009) Quantifying freestyle kick-count and kick-rate patterns in
paralympic swimming. J Sports Sci 27(13):1455–1461. doi:10.1080/02640410903062936
Hay JG, Liu Q, Andrews JG (1993) Body roll and handpath in freestyle swimming: a computer
simulation study. J Appl Biomech 9(3):227–237
Hochstein S, Blickhan R (2011) Vortex re-capturing and kinematics in human underwater undula-
tory swimming. Hum Mov Sci 30(5):998–1007. doi:10.1016/j.humov.2010.07.002
Holmer I (1974) Energy cost of arm stroke, leg kick, and the whole stroke in competitive swimming
styles. Eur J Appl Physiol 33(2):105–118
Kawai E, Tsunokawa T, Tsubakimoto S, Takagi H (2016) A study of fluid forces acting on a foot
during eggbeater kicks of water polo players. In: ISBS-conference proceedings archive, vol 1
Kudo S, Vennell R, Wilson B (2013) The effect of unsteady flow due to acceleration on hydrody-
namic forces acting on the hand in swimming. J Biomech 46(10):1697–1704
Liu Q, Hay JG, Andrews JG (1993) Body roll and handpath in freestyle swimming: an experimental
study. J Appl Biomech 9(3):238–253
Maglischo EW (2003) Swimming fastest. Human Kinetics, Champaign
Matsuda Y, Hirano M, Yamada Y, Ikuta Y, Nomura T, Tanaka H, Oda S (2016) Lower muscle
co-contraction in flutter kicking for competitive swimmers. Hum Mov Sci 45:40–52
Matsuuchi K, Miwa T, Nomura T, Sakakibara J, Shintani H, 04/888) Unsteady flow field around a
human hand and propulsive force in swimming. J Biomech 42 (1):42–47
Matsuuchi K, Muramatsu Y (2011) Investigation of the unsteady mechanism in the generation of
propulsive force while swimming using a synchronized flow visualization and motion analysis
system. In: Biomechanics in applications. InTech, Rijeka, pp 389–408
McCabe CB, Psycharakis S, Sanders RH (2011) Kinematic differences between front crawl sprint
and distance swimmers at sprint pace. J Sports Sci 29(2):115–123. doi:10.1080/
02640414.2010.523090
McCabe CB, Sanders RH (2012) Kinematic differences between front crawl sprint and distance
swimmers at a distance pace. J Sports Sci 30(6):601–608. doi:10.1080/02640414.2012.660186
McCullough AS, Kraemer WJ, Volek JS, Solomon-Hill GF Jr, Hatfield DL, Vingren JL, Ho JY,
Fragala MS, Thomas GA, Hakkinen K, Maresh CM (2009) Factors affecting flutter kicking
speed in women who are competitive and recreational swimmers. J Strength Cond Res
23(7):2130–2136. doi:10.1519/JSC.0b013e31819ab977
Nakashima M, Maeda S, Miwa T, Ichikawa H (2012) Optimizing simulation of the arm stroke in
crawl swimming considering muscle strength characteristics of athlete swimmers. J Biomech
Sci Eng 7(2):102–117
Nakashima M, Satou K, Miura Y (2007) Development of swimming human simulation model
considering rigid body dynamics and unsteady fluid force for whole body. J Fluid Sci Technol
2(1):56–67
Pacholak S, Hochstein S, Rudert A, Brücker C (2014) Unsteady flow phenomena in human
undulatory swimming: a numerical approach. Sports Biomech 13(2):176–194
Palmer B (2009) Why is michael phelps tinkering with success? Wasn't his old stroke good enough
for Olympic gold. The Slate Group. http://www.slate.com/articles/sports/sports_nut/2009/05/
why_is_michael_phelps_tinkering_with_success.html. Accessed 23 Aug 2016
Payton C, Baltzopoulos V, Bartlett R (2002) Contributions of rotations of the trunk and upper
extremity to hand velocity during front crawl swimming. J Appl Biomech 18(3):243–256
Payton C, Bartlett RM, Baltzopoulos V, Coombs R (1999) Upper extremity kinematics and body
roll during preferred-side breathing and breath-holding front crawl swimming. J Sports Sci
17(9):689–696
Psycharakis SG, McCabe C (2011) Shoulder and hip roll differences between breathing and
non-breathing conditions in front crawl swimming. J Biomech 44(9):1752–1756
Psycharakis SG, Sanders RH (2008) Shoulder and hip roll changes during 200-m front crawl
swimming. Medi Sci Sport Exerc 40(12):2129–2136. doi:10.1249/MSS.0b013e31818160bc
Sanders RH (1999) Hydrodynamic characteristics of a swimmer’s hand. J Appl Biomech 15(1):3
Sanders RH (2002) New analysis procedures for giving feedback to swimming coaches and
swimmers. In: Gianikellis KE (ed) XX international symposium on biomechanics in Sports,
Caceres, Spain, 2002, ISBS – Conference proceedings archive, pp 1–14
Sanders RH (2007) Rock and roll rhythms in swimming. In: Menzel HJ, Chagas MH (eds) XXV
international symposium on biomechanics in sports, Ouro Preto, Brazil, 2007, ISBS – Confer-
ence proceedings archive
Sanders RH, Cappaert JM, Devlin RK (1995) Wave characteristics of butterfly swimming.
J Biomech 28(1):9–16
Sanders RH, Psycharakis SG (2009) Rolling rhythms in front crawl swimming with six-beat kick.
J Biomech 42(3):273–279. doi:10.1016/j.jbiomech.2008.10.037
Sato Y, Hino T (2013) A computational fluid dynamics analysis of hydrodynamic force acting on a
swimmer's hand in a swimming competition. J Sports Sci Med 12(4):679–689
Schleihauf R (1979) A hydrodynamic analysis of swimming propulsion. Swimming III 8:70–109
Schnitzler C, Seifert L, Ernwein V, Chollet D (2008) Arm coordination adaptations assessment in
swimming. Int J Sports Med 29(6):480–486. doi:10.1055/s-2007-989235
Seifert L, Chollet D, Allard P (2005) Arm coordination symmetry and breathing effect in front
crawl. Hum Mov Sci 24(2):234–256
Seifert L, Chollet D, Chatard JC (2007) Kinematic changes during a 100-m front crawl: effects of
performance level and gender. Med Sci Sports Exerc 39(10):1784–1793. [Erratum appears in
Med Sci Sports Exerc. 2008 Mar;40(3):591]
Sfakiotakis M, Lane DM, Davies JBC (1999) Review of fish swimming modes for aquatic
locomotion. IEEE J Ocean Eng 24(2):237–252. doi:10.1109/48.757275
Suito H, Ikegami Y, Nunome H, Sano S, Shinkai H, Tsujimoto N (2008) The effect of fatigue on the
underwater arm stroke motion in the 100-m front crawl. J Appl Biomech 24(4):316–324
SwimOutlet.com How & when to swim straight-arm freestyle. https://www.swimoutlet.com/guides/
how-when-to-swim-straight-arm-freestyle. Accessed 23 Aug 2016
Takagi H, Nakashima M, Ozaki T, Matsuuchi K (2014) Unsteady hydrodynamic forces acting on a
robotic arm and its flow field: application to the crawl stroke. J Biomech 47(6):1401–1408
Takagi H, Nakashima M, Sato Y, Matsuuchi K, Sanders RH (2016) Numerical and experimental

investigations of human swimming motions. J Sports Sci 34(16):1564–1580. doi:10.1080/
02640414.2015.1123284
Takagi H, Sanders R (2002) Measurement of propulsion by the hand during competitive swimming.
In: The engineering of sport, vol 4, pp 631–637
Takagi H, Wilson B (1999) Calculating hydrodynamic force by using pressure differences in
swimming. In: Keskinen K, Komi P, Hollander AP (eds) Biomechanics and medicine in
swimming VIII. Gummerus Printing, Jyvaskyla, pp 101–106
Toussaint HM, Van den Berg C, Beek WJ (2002) “pumped-up propulsion” during front crawl
swimming. Med Sci Sports Exerc 34(2):314–319
Tsunokawa T, Nakashima M, Takagi H (2015) Use of pressure distribution analysis to estimate fluid
forces around a foot during breaststroke kicking. Sports Eng 18(3):149–156. doi:10.1007/
s12283-015-0174-6
Ungerechts BE, Persyn U, Colman V 2000 Analysis of swimming techniques using vortex traces.
In: ISBS-conference proceedings archive, vol 1
von Loebbecke A, Mittal R, Fish F, Mark R (2009a) A comparison of the kinematics of the dolphin
kick in humans and cetaceans. Hum Mov Sci 28(1):99–112
von Loebbecke A, Mittal R, Mark R, Hahn J (2009b) A computational method for analysis of
underwater dolphin kick hydrodynamics in human swimming. Sports Biomech 8(1):60–77
Vorontsov AR, Rumyantsev VA (2000) Propulsive forces in swimming. In: Zatsiorsky VM
(ed) Biomechanics in sport: performance, enhancement and injury prevention, vol XI. Wiley,
Hoboken, pp 205–231
Yanai T (2001a) Rotational effect of buoyancy in frontcrawl: does it really cause the legs to sink?
J Biomech 34(2):235–243
Yanai T (2001b) What causes the body to roll in front-crawl swimming? J Appl Biomech
17(1):28–42
Yanai T (2003) Stroke frequency in front crawl: its mechanical link to the fluid forces required in
non-propulsive directions. J Biomech 36(1):53–62. doi:10.1016/s0021-9290(02)00299-3
Yanai T (2004) Buoyancy is the primary source of generating bodyroll in front-crawl swimming.
J Biomech 37(5):605–612
Movement Analysis of Scull and Oar Rowing
Patria A. Hume
Abstract
Rowing coaches ask “What is an ideal rowing technique?” and “How can
biomechanics help me improve technique in my rower?”. This chapter discusses
biomechanical variables that have been shown to help predict boat velocity, and
how biomechanists can provide profiling of body kinematics (joint angles,
segment velocities and segmental sequencing, drive to recovery ratios) and forces
(handle, pin/oar, and feet). Rower movement has traditionally been assessed on
ergometers using power output and two-dimensional high-speed video or opto-
electronic systems to provide joint angles and segment velocities. Advances in
technology have seen the use of three-dimensional video, force plates at the foot-
stretchers, strain gauges at the pins and oars, pressure sensors at the seat and
feet, electromyography, computer simulations/modelling, inertial sensors, and
instrumented rowing boats (i.e., pitch, yaw, roll, velocity), for rowing technique
analyses. The greater use of customized telemetered sensors on the rowing skiff
can assist the coach and biomechanist with judging when performance (skiff
velocity) improves with a training intervention. Biomechanical variables can be
measured and feedback presented in real-time on-water. Proper proximal to distal
sequencing of joint rotations is important in the production of high power output
in rowing technique. The biomechanist must be able to measure rowing perfor-
mance technique and outcome variables reliably and with accuracy/sensitivity, so
feedback can be given to rowers and coaches on what needs to be the focus of
technique improvement. This chapter focuses on how biomechanics technologies
are applied to improve performance and help prevent injury in rowing.
P.A. Hume (*)

Auckland University of Technology, Auckland, New Zealand
e-mail: patria.hume@aut.ac.nz

DOI 10.1007/978-3-319-30808-1_133-1
2 P.A. Hume
Keywords
Boat velocity • Body kinematics • Drive-to-recovery ratio • Ergometer • Feet
force • Force profiles • Handle force • Inertial sensors • Injury prevention • Joint
angles • Performance • Pin/oar force • Power output • Real-time feedback •
Rowing • Segment velocities • Segmental sequencing • Strain gauges • Stroke
rate • Technique • Telemetered sensors
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Predictors of Boat Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Ergometer Versus On-Water Data Collection Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Rower Kinematic Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Joint Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Segmental Sequencing of Rower . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Drive-to-Recovery Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Force Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Oar Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Foot-Stretcher Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Blade Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Introduction
Rowing is popular worldwide, with people of all ages (under 23, juniors, university,
Masters) and genders participating in many formats of the sport (2000-m rowing
with one oar and sculling with two oars, indoor ergometer rowing, coastal/tour
rowing, and para-rowing).
Race time is the critical measure of performance and is determined from mean
skiff (boat) velocity during a race (usually 2000 m). Although a high proportion of
race training is completed on-water, rowing ergometers are commonly used for
performance testing, technique coaching, crew selection, or for training during
poor weather (Soper and Hume 2004b).
Winning an Olympic gold medal can come down to a boat-ball length (e.g.,
Men’s Rio 2016 Olympic single scull final). Therefore, small changes in technique
that can improve boat velocity with each stroke can make a large difference to the
performance outcome.
Rowing performance outcome measures from a biomechanics perspective may
be categorized as direct (boat or ergometer time over 2000 m) or indirect (power
output, boat velocity). Boat velocity is determined by stroke rate, propulsion per
stroke, stroke velocity, stroke drive distance, and stroke power. Rowers must
develop a consistent movement pattern (body angles, joint forces, and muscle
activity patterns) to secure these qualities. Proper proximal to distal sequencing of
Movement Analysis of Scull and Oar Rowing 3
joint rotations is important in the production of high power output in rowing

technique. The biomechanist must be able to measure rowing performance technique
and outcome variables reliably and with accuracy/sensitivity, so feedback can be
given to rowers and coaches on what needs to be the focus of technique
improvement.
Rower movement has traditionally been assessed on ergometers using power
output and two-dimensional high-speed video or optoelectronic systems, to provide
joint angles and segment velocities. Advances in technology have seen the use of
three-dimensional video, force plates at the foot-stretchers, strain gauges at the pins
and oars, pressure sensors at the seat and feet, electromyography, computer simula-
tions/modelling, inertial sensors, and instrumented rowing boats (i.e., pitch, yaw,
roll, velocity) for rowing technique analyses. The greater use of customized tele-
metered sensors on the rowing skiff can assist the coach and biomechanist with
judging when performance (skiff velocity) improves with a training intervention.
Biomechanical variables can be measured and feedback presented in real-time
on-water.
Coaches ask “What is an ideal rowing technique?” and “How can biomechanics
help me improve technique in my rower?” (Soper et al. 2005). This chapter discusses
biomechanical variables that have been shown to help predict boat velocity, and how
biomechanists can provide profiling of body kinematics (joint angles, segment
velocities and segmental sequencing, drive to recovery ratios) and forces (handle,
pin/oar, and feet). This chapter focuses on how biomechanics technologies are
applied to improve performance and help prevent injury in rowing.
State of the Art
State-of-the-art research evidence for use of boat instrumentation systems by sports

biomechanists to assess technique and provide real-time oar/pin and foot-stretcher
force feedback for on-water rowing performance is provided in the following
sections. Quantification of boat velocity and its determinants via ergometer and
on-water studies has enabled profiling of rower kinematics and forces acting on
the oars, blades, and foot-stretchers.
Predictors of Boat Velocity
Race time is the critical measure of a sculler’s performance and is determined by race
distance and average skiff velocity. The mechanical determinants of producing
average skiff velocity were indicated in a deterministic biomechanical model of
rowing (Soper and Hume 2004b) (see Fig. 1). There has not been a study to date that
provides an estimate of the magnitude of impact of all of these independent variables
to average and/or instantaneous boat velocity. Boat velocity should remain as
constant as possible during a stroke cycle (i.e., limited acceleration or deceleration
changes) to minimize the effects of drag on the boat from the water and to a lesser
4 P.A. Hume
Total rowing race time
t=d/v
Average boat Average boat

distance per stroke velocity per stroke
Distance Distance travelled Boat velocity Boat velocity during

travelled during during recovery during drive phase recovery phase
drive phase phase v=axt
Average time Average acceleration
Time taken for Time taken for Acceleration Acceleration

drive phase recovery phase during the drive during the recovery
phase phase
a=F/m
Average force Average mass
Propulsive Resistive Mass of the Mass of the boat

forces/torques forces rower(s) and rigging
Foot stretcher Oar Air Water

force torque resistance resistance
Time of force Angle and Time of force Angle at force Drag Viscous Wave
application with height of the application application drag drag
respect to catch time foot stretcher
Number of Time of Angle of Angle of
strokes each oar at the the oar at
completed stroke catch finish
Fig. 1 A deterministic model showing the basic mechanical factors of rowing that influence
performance
extent air mechanics. Greatest boat velocity occurs during the recovery phase when
the oars are out of the water and the rowers’ mass is moving in the opposite direction
to the boat.
Higher stroke rate is significantly correlated (r = 0.66–0.76) to increased average
skiff velocity (r = 0.66–0.76) (Martin and Bernfield 1980). Longer stroke drive
lengths (McBride 1998) will result in greater average skiff velocity. Reduced
intrastroke fluctuations in skiff velocity (Dal Monte and Komor 1989; Nolte
1991), as stroke rate increases (Martin and Bernfield 1980), may be an indicator of
performance level. It is important to note that significant changes in biomechanical
performance indicators may lead to only minimal alteration in boat velocity (Doyle
et al. 2010).
The stroke rate (strokes per minute or spm) is the number of strokes divided by
1 min of time. A 1-min period divided by the number of strokes will give the time per
stroke. Anywhere from 1000 m out from the finish, but more normally from the final
300 m, an all-out sprint starts with a high stroke rating. All crews will have their own
efficient stroke rate. Gearing of the blade in the skiff allows female rowers to rate the
same as men even if they are not as strong. Velocity fluctuations have been shown to
be highly correlated (r = 0.93) with stroke rate because the accelerations of the
rowers’ body mass and the mass of the counteracting boat increase with stroke rate.
Race time may be slightly reduced by a moderate reduction in stroke rate that is
compensated by an increased force output for each stroke (Hill and Fahrig 2009).
Average skiff velocity is the average velocity of each individual stroke. Stroke
velocity is the product of the oar excursion (distance of oar/blade travel) and time.
Stroke drive velocity (ms 1) is the blade stroke drive displacement (m) – known as
stroke drive length – divided by the blade stroke drive time (s), i.e., the time from
catch to the finish (when the blade leaves the water). The total blade stroke time
(from catch to catch) is divided into the drive time and the recovery time with the
ratio between these times known as the drive to recovery ratio. Rowers tend to place
more emphasis on the stroke time than on the oar excursion during competition, as
when stroking at very high rates, it is hard to maintain a good length.
The stroke drive distance, known as oar excursion, is determined as the total arc
the blade moves through from blade entry at the catch until the blade is fully
removed at the finish and is reported in degrees. Force application is most inefficient
at the catch and finish positions due to increased transverse force application
(Sanderson and Martindale 1986) and therefore a reduction in oar excursion of this
magnitude may be acceptable, and possibly desirable, if stroke rate and therefore
average skiff velocity increased.
Boat velocity is determined by the amount of propulsive and drag forces acting
about the rowing system (rower(s), shell, and blades) (Barrett and Manning 2004).
Blade force is the only source of propulsion, and is the key factor that will determine
performance (Sanderson and Martindale 1986). A rower’s capacity to produce large
forces at the handle to be transferred to the blade via the pin will in part depend upon
their physical capacity and the largest, strongest individuals are seen to produce
higher boat velocities (Barrett and Manning 2004). Clear guidelines for coaches,
selectors, and rowers on the ideal biomechanical rowing stroke for a given rower of
determined anthropometric characteristics are needed (Soper and Hume 2004b).
However, it is not always possible to increase physical capacity sufficiently, and
gains must then be made through the coordination of this force production (i.e., the
shape of the force profile) (Anderson et al. 2005).
Kinetic and kinematic profiles during a stroke cycle are dependent upon gender
(Elliott et al. 2002; Hartmann et al. 1993), skill level (Smith and Spinks 1995),
rowing subdiscipline (Korndle and Lippens 1988), and potentially seat allocation
when in a team skiff. The time course of force production, in association with oar
position, body segment velocities, and contribution of other crewmembers affect
performance. Additionally, a rower’s anthropometry and skiff setup also contribute
to performance (Soper and Hume 2004b). Mean propulsive power output per
kilogram of body mass, propulsive work consistency, stroke-to-stroke consistency,
and stroke smoothness have successfully distinguished rowers of three different
ability levels (Smith and Spinks 1995).
A validated biomechanical model (Pelz and Vergé 2014) for efficiency and speed
of rowing determined that the speed of a competitive rowing crew depends on the
number of crew members, their body mass, sex, and the type of rowing (sweep
rowing or sculling). The time-averaged speed is proportional to the rower’s body
6 P.A. Hume
mass to the 1/36th power, to the number of crew members to the l/9th power, and to
the physiological efficiency (accounted for by the rower’s sex) to the 1/3rd power.
Ergometer Versus On-Water Data Collection Methods
In making a decision about what measurement method to use, the biomechanist

needs to understand what the biomechanical similarities or differences are between
ergometer and on-water sculling or sweep rowing (see Fig. 2).
Ergometer design (e.g., stationary or dynamic) affects rowing stroke mechanics
(Benson et al. 2011; Greene et al. 2013). Power delivery and coordination patterns of
rowing strokes for 14 elite rowers were tested in three ergometer conditions: the
fixed stretcher Concept IIc ergometer, the Concept IIc ergometer mounted on sliding
rails, and the sliding stretcher RowPerfect ergometer. Ergometers were instrumented
to measure the external force generated at the handle and the foot-stretcher. A nine-
segment inverse dynamics model was used to calculate joint and overall power. Peak
power generation and absorption at the knee joint was significantly greater, and total
power on the ergometer delayed on the fixed stretcher ergometer compared to the
sliding stretcher ergometers. No differences were observed in the mechanical energy
at the handle of the three ergometers; however, greater joint mechanical energy
production of the lower limb reduced mechanical efficiency when rowing the
Concept IIc fixed ergometer. The fixed foot-stretcher on the Concept IIc fixed
ergometer acts to increase the inertial forces that the rower must overcome at the
catch, increasing the moment and power output at the knee, affecting the coordina-
tion pattern during the recovery phase (Greene et al. 2013).
Ergometers enable a controlled and technically simpler environment than collect-
ing data on-water. The findings from research (Elliott et al. 2002) investigating the
ability of an ergometer to simulate on-water sculling or sweep rowing technique
typically support the use of ergometers. Trunk, thigh, and shank positions were
similar at the catch and finish positions on-water or on the RowPerfect for eight
national junior rowers. However, at the catch (24 spm), 3.1% greater knee flexion
(1.6 , p < 0.01) occurred on-water than on a RowPerfect ergometer. Stroke rate did
not significantly affect joint kinematics when using a RowPerfect (Elliott et al. 2002)
or Concept II (Halliday et al. 2001) ergometer or when sculling on-water with knee
angles of 52.5 2.0 (Elliott et al. 2002). Ergometers, however, do not allow good
reproduction of upper limb body patterns compared with on-water rowing or
sculling due to the central pulley system used. There are significant differences in
segment energies between the two activity modes (Martindale and Robertson 1984).
The widespread use of the air-braked Concept II is evident by the inception of the
Concept II Indoor Rowing Olympics in 1981. The Concept II is reported to be
reliable when testing well-trained rowers (2%, 95% CL = 1.3–3.1%) (Schabort et al.
1999) and national-level competitive rowers (1.3%, 95% CL = 0.8–2.9%) (Soper
and Hume 2004a) during 2000-m races. The high reliability of rowers’ performance
on the Concept II ergometer is thought to be present due to the ergometer being
inherently stable, requiring no calibration, and the rowers being accustomed to
Fig. 2 Ergometers are used on-land and can be fixed or sliding, while rowing single scull skiff
(boat) are used on-water
8 P.A. Hume
completing 2000-m time trials on the ergometer (Schabort et al. 1999). A moderate
positive correlation (r = 0.74) between 2000 m performance time on a Concept II
and during on-water 2000 m standing start time trials has been reported (Ryan-
Tanner et al. 1999).
An issue affecting the ecological validity of rowing biomechanical research has
been the difficulty in examining biomechanics of rowing on-water due to environ-
mental conditions. There are now a number of instrumentation systems that allow
measures for boat velocity and acceleration. For example, the Accrow system
measures boat acceleration with an acceleration sensor and boat velocity with GPS
(Mattes and Schaffert 2010). The software performs three analysis routines every
stroke: the boat motion for on-water training or tests, a race analysis over 2000 m,
1000 m, or 500 m to improve the race profile, and an improvement of the rowing
start. Boat acceleration, boat velocity, stroke rate, and propulsion per stroke are
measured.
With development of new technologies such as wearable inertial sensors that
contain gyroscopes and triaxial accelerometers, sports biomechanists are now able to
more easily quantify on-water motion of rowers. Estimation of the center of mass
trajectory is possible using these wearable sensors during golf swings (Najafi et al.
2015) and could be applicable to rowing. Linking the output from these devices to
video collected on the boat via systems such as GoPro cameras could allow
quantitative and qualitative kinematics to be examined. However, these devices
have not yet been applied to rowing. Studies evaluating the use and accuracy
would be needed before they could be used in the aquatic environment for rowing.
Rower Kinematic Profiling
To determine an ideal rowing technique, it is essential to first understand the

movement patterns of the stroke cycle and the underlying forces. The overall
movement shape is what a coach views and therefore manipulates to produce
a more powerful and efficient stroke cycle. Kinematics represents the overall move-
ment pattern of the sculling motion, which is the result of internal and external forces
acting on the rower-skiff system. Many researchers have reported joint and segments
motions during ergometer rowing (Hawkins 2000; Martindale and Robertson 1984;
Nelson and Widule 1983; Torres-Moreno et al. 1999) and during on-water rowing
(Elliott et al. 2002; Soper et al. 2002).
Joint Angles
The high incidence of lower back pain among scullers and rowers, in part due to
the amount of time a rower spends in a flexed position (Ogurkowska et al. 2015),
the number of rowing stroke cycles completed, and the forces on the body during
the rowing stroke (Reid and McNair 2000) have led to the assessment of the lumbo
pelvic angle (included angle between the pelvis and lower back) at the catch. Rowers
should adopt a less flexed lumbar spine posture, particularly at the catch phase when
the oar is placed in the water (Stallard 1999). In this respect, if the pelvis could be
rotated more anteriorly, less motion would be required in the lumbar spine. As
rowing distance get longer, rowers tend to change their body angles and action
(Bingul et al. 2014).
Kinematics of the upper arm and forearm segments are significantly different in
sweep oar on-water and sweep oar ergometer rowing (Lamb 1989). However, these
differences are of minor importance because of the small contributions made by the
arms at the catch. The lower leg angular velocity contributed 76% for sweep oar
on-water rowing and 68% for sweep oar ergometer rowing to total linear oar velocity
(Lamb 1989).
An individual’s ankle range of motion, lower limb anthropometry, and possibly
discipline (sculling or sweep rowing) may affect the optimal foot-stretcher position.
Foot-stretcher position influences rowing technique (Hume et al. 2000) and perfor-
mance (Herberger and Klavora 1985; Hume et al. 2005a) and can be optimized. The
foot-stretcher can be adjusted in three directions: vertically, horizontally, and
obliquely (changing the rake). The rake of the foot-stretcher has been recommended
to be between 41 and 46 but will be dependent upon a rower’s ankle range of
motion. Horizontal forces should increase as the foot-stretcher angle increases or
gets steeper. Higher, steeper, and/or less stern-ward foot-stretcher settings should be
used to prevent excessive over-reaching at the catch when span lengths are reduced
in larger boat classes. A novel foot-stretcher with a rigid wedge shoe fixed at a pivot
around the heel has allowed contact of the whole foot surface to the foot-plate
throughout the entire stroke (Millar et al. 2016) resulting in 2% on-water race time
performance enhancement and improved comfort.
Biomechanics researchers and coaches can utilize feedback real-time devices to
help change technique (Smith and Loschner 2002). For example, visual feedback of
body position using a real-time Goggle Training System has been shown to be
effective in changing technique (Hume et al. 2001, 2003) and power output during
ergometer rowing in New Zealand rowers (Hume et al. 2005b). Video from the coach
boat of the rower was provided in real-time to the rowers’ goggles via a telemetry
system. Pelvic angle was changed to help reduce back injury.
Segmental Sequencing of Rower
Reports presenting segmental interaction of the rowing stroke for athletes (Hume
and Soper 2001; Manning et al. 2001) have provided visual and quantitative
feedback using video analysis of the rowing stroke. Sequential sequencing of the
lower limbs, trunk, and arms may lead to a more effective rowing stroke (Kleshnev
and Kleshnev 1998) and therefore greater average skiff velocity.
The lower limbs are responsible for initiating the drive phase through knee
extension followed sequentially by trunk movement. Trunk velocity contribution
maximizes at approximately 70% of the drive phase, surpassing the contribution at
this point of lower leg velocity as the rower moves toward the bow of the skiff (Lamb
10 P.A. Hume
1989). When a stroke is performed correctly, the legs, followed by the lower trunk,
mid-trunk, arms, and wrist segments will accelerate the skiff (Nelson and Widule
1983). Consecutive segment movements have been shown in 62 experienced oars-
men to produce greater power than synchronous segment movements; however,
synchronous movements produced a more mechanically efficient rowing pattern
(Kleshnev and Kleshnev 1998). The movement choice should be based upon a
rower’s individual musculature and biomechanical features of their style. For exam-
ple, following 1 month of intensive training, female rowers showed a tendency to
have a higher effective work output, which was attributed to better sequencing of the
main muscle groups (Mason et al. 1988).
When comparing segment velocities of novice and skilled rowers (Nelson and
Widule 1983), novice rowers have a greater delay between peak angular velocities of
the knee and trunk than skilled rowers. Delay in peak segment angular velocities
resulted in a smaller sum of knee and trunk velocity when the oar was perpendicular
to the shell.
The asymmetrical motion of the rowing oar action (Fohanno et al. 2015) and the
segmental sequencing of both the oar and scull actions involve substantial forces and
ranges of rotational movement. Good technique is needed if the athlete is to avoid
injuring their back, knees, elbows, and wrists (Smoljanovic et al. 2009; Wilson et al.
2008) or developing rib stress fractures (McDonnell et al. 2011). Real-time segmen-
tal sequencing information is still limited during on-water analyses.
Drive-to-Recovery Ratio
Skiff velocity should remain as constant as possible during a stroke cycle (i.e.,
limited acceleration or deceleration changes) to minimize the effects of drag on the
skiff from the water. Greatest skiff velocity occurs during the recovery phase when
the oars are out of the water and the rowers’ mass is moving in the opposite direction
to the skiff (Affeld et al. 1993). The catch and initial drive phase of the stroke are
characterized by low skiff velocity which has been attributed to a delay in the ability
of the rower to overcome water resistance (Martin and Bernfield 1980) and inertia
of the system, the relative velocity of the skiff and oars, the need to change the
movement direction of the oarsman, and the time spent in the stern of the skiff prior
to placing the blades in the water (Martin and Bernfield 1980). The ability of the
sculler to limit the reduction in skiff velocity during the catch and initial drive phase
may be an indicator of performance level.
Oar excursion can be divided into the drive oar excursion and the recovery
oar excursion. The relative time to complete the drive oar excursion and the recovery
oar excursion is known as the drive-to-recovery ratio. During on-water sweep oar
rowing and ergometer rowing, the drive-to-recovery ratios have ranged from 0.9 to
1.7, and are strongly negatively correlated to stroke rate (McBride 1998) and average
skiff velocity.
As stroke rate increases, the absolute time in the drive and recovery phases of the
stroke cycle decreases; however, greater reductions occur in the recovery phase
(Dawson et al. 1998). Increases in average skiff velocity resulted in increases in the
percent of total drive time during on-water sweep rowing (r = 0.73) (Martin and
Bernfield 1980), on-water sculling, and on an ergometer (Dawson et al. 1998).
Although percent leg drive time is significantly related to average velocity,
a constant drive time irrespective of the overall stroke time may be indicative of
a limit in the maximum rate of knee and hip extension during the stroke cycle to
maintain the required level of force (Martin and Bernfield 1980). Optimizing the
drive-to-recovery ratio by slowing down a rower’s recovery phase raises some
concerns as a performance strategy, due to the influence it may have on skiff velocity
fluctuations (McBride 1998).
Auditory feedback has been effective in improving time structure of rowing
cycles (extended time of positive acceleration) for six athletes in the German
National Para-Rowing team who were visually impaired (Schaffert and Mattes
2015). Boat acceleration–time traces were converted online into acoustic feedback
and presented via speakers during rowing. The acoustic feedback provided important
functional information about the boat motion independent of vision.
Force Profiling
Coaches and selectors need guidelines on how to use force application profiles to
reliably or accurately predict a rower’s current or future on-water sculling or sweep
rowing performance (Coker et al. 2008). Coaches can ask “What is the ideal oar
force–angle profile for scullers and rowers in bow and stroke seat positions?” and
“What is the ideal interval between oar and foot-stretcher force application for
a rower, and between different crew members?” Biomechanists need to be able to
collect quality force data and to be able to interpret it meaningfully for coaches.
Early studies used measurement of force–time relations in rowing racing skills
with modified strain gauges (Baird and Soroka 1951). Technique improvements
were suggested, such as greater force on the ergometer handle being generated
when the drive phase was initiated with extended elbows and the finish position
was reached with the elbows close to the trunk (Bompa 1980).
Force profiling has been reported using Concept II ergometers (MacFarlane et al.
1997; Torres-Moreno et al. 1999), Gjessing ergometers (Hartmann et al. 1993;
Mason et al. 1988; Peltonen and Rusko 1993), and RowPerfect ergometers (Elliott
et al. 2002). For example, on a RowPerfect ergometer handle, forces ranged from
318 to 541 N for four male and four female national junior or under-23 rowers at
24, 26, and 28 spm (Elliott et al. 2002). Similarly, experienced national-level male
rowers produced greater power output at race pace compared to females of the same
competitive level (Buck et al. 2000). Greater peak force recordings for males
compared to females were also reported following a series of 5 and 10 maximal
effort strokes and a 6-min maximal effort row on a Gjessing ergometer by 81 elite
male and 21 elite female heavyweight rowers (Hartmann et al. 1993).
Boat instrumentation allows force profiles from the oar/pin and foot-stretcher
(Elliott et al. 2002; McBride and Elliott 1999; Smith and Loschner 2002), and oar
12 P.A. Hume
position, skiff velocity, and skiff orientation to be gained (Coker et al. 2009). Ideally
blade force rather than pin force would be collected because this is the origin of
propulsion and therefore performance. Large forces and angular velocities at the pin
will indicate high rower powers despite no knowledge of whether this has been
transferred as propulsive force at the blade. At the elite level, where blade work will
be skilled, it can be assumed that the pin forces should reflect blade forces well, but
force and power data should carefully be compared with boat velocity and video
footage to ensure that pin forces and power increases do result in enhanced perfor-
mance. Inertial blade forces, created as the blade changes direction at the catch, will
add to the gate force reading (Kleshnev 2002).
Oar Forces
On-water force profiles are known to differ from those seen on the ergometer;
therefore, it is preferable that oar/pin force data are collected on-water (Korndle
and Lippens 1988; Li et al. 2007; Roth et al. 1993; Wing and Woodburn 1995).
Modern boat instrumentation systems (e.g., PowerLine™) allow analysis of
on-water force profiles. Peak oar force increases with stroke rate (McBride 1998),
is greater when the handle is pulled at the umbilicus level (Bompa 1980), is
significantly greater in males (Elliott et al. 2002; Hartmann et al. 1993), and may
be asymmetrical between oars during on-water sculling (Elliott et al. 2002). Oar/pin
forces change as stroke rating changes (Roth et al. 1993). For well-trained, compet-
itive coxless pairs, average oarlock force equaled 841.3 N at 20 spm and signifi-
cantly (p = 0.001) increased (12%) to 942.0 N at 35.7 spm (maximal rating)
(McBride 1998).
Effective rowing oar force profiles have a large area under the force–time, or
force–angle, curve, generally assisted by having a smooth shape with no dips
in force (Smith and Spinks 1998). Achieving a more rectangular- rather than
triangular-shaped force profile with a high rate of force development will maximize
drive-phase work and impulse for a fixed peak force magnitude (Kleshnev and
Kleshnev 1998; Millward 1987).
Peak oar force occurring in front of the position where the oar shaft is perpen-
dicular to the longitudinal axis of the boat will maximize the use of lift forces acting
at the blade (Nolte and Morrow 2002). Earlier peak positioning will also better
utilize the recoil of elastic energy stored in the bend of the oar shaft. Peak oar force
occurs earlier in the drive phase and closer to the bow of the skiff as stroke rate
increases (McBride 1998). The ability to maintain peak oar force application when
the oar is perpendicular to the skiff may be an indicator of performance level.
The area underneath the force–angle curve relates to the amount of work done per
stroke, while the area underneath the force–time curve equates to the impulse added
during the drive. Any increase in momentum (and therefore increase in boat velocity,
including the reversal of the unavoidable deceleration occuring each stroke) will be
determined by the size of this impulse. The area under either curve should therefore
be as large as possible. Tangents drawn from the steepest points of the curve at the
catch and finish to the baseline allow the area to be calculated from this modified
curve (Hill and Fahrig 2009). This method manages the negative areas under the
0 kgF line and also attempts to minimise the addition of intertial forces in these areas.
Increasing the average force relative to maximal force and achieving a high rate of
force development at the beginning of the drive will result in a larger area for the
same, or even lower, peak force (Kleshnev and Kleshnev 1998; Millward 1987).
This ratio has been found to be significantly higher in elite than subelite rowers
(Smith and Draper 2006). Visually, this will create a more rectangular, rather than
triangular shape.
The smoothness of the force profile is correlated with rowing level (Smith and
Spinks 1998). This may be due to reduced velocity fluctuations which will increase
efficiency (Baudouin and Hawkins 2004; Smith and Loschner 2002). Eliminating
“dips” will increase the area under the curve and increase performance accordingly.
Although small, the differences in the shape and smoothness of the force profile
between experienced rowers are real (Wing and Woodburn 1995). Individuals will
develop a highly repeatable, “signature” force profile through training, allowing
rowers to be identified by subtle characteristics in their force profile during the drive
phase (Baudouin and Hawkins 2004; Hill 2002; Wing and Woodburn 1995). These
characteristics do not appear to change without specialized training; therefore, if
alterations are needed, it is necessary to investigate what exactly is causing the
undesirable feature and make specific changes accordingly (Hill 2002).
The oar angle at which peak force should occur has been disputed. Mechanically,
it was traditionally thought that peak force should be applied when the blade was
perpendicular to the long axis of the boat. At that position, the transverse component
of the rower’s force would be minimal and force would be maximal in the propulsive
direction (Martin and Bernfield 1980; Spinks 1996). Conversely, front-loaded force
profiles are thought to be preferable because the resulting power curve is more
evenly distributed and will allow for reduced velocity fluctuations and increased
mechanical efficiency (Kleshnev 2006; Nolte and Morrow 2002). It is also known
that the elastic energy stored in the bend of the oar shaft, and later recoiled, is used
more effectively if an early peak force is applied (Kleshnev 2007b). Although
arguments for both middle- and front-loaded force–angle profiles exist, peak force
in front of the perpendicular point is preferable (Coker 2010).
Force–time and force–angle profiles are often used in presenting data to coaches
(Fig. 3). Gate force is measured at the pins at the oar gates in units of kilograms.
Evaluation of force versus time is a simple method of analysis, useful in conjunc-
tion with synchronized video and for examining coordination of timing within a
crew (Kleshnev 2007a). Force–angle curves provide a visual representation of
work done per stroke and are effective when comparing different stroke rates.
Force–angle curves can be useful when considering crew setup as it is clear
whether a rower is too far forward or back in the boat. If a rower’s foot-stretcher
is further to the bow than other crew members, this rower’s force–angle curve will
be shifted to the right in comparison to the others (when presented on the same axis
and scale) and, if the foot-stretcher is further to the stern, the curve will be shifted
to the left.
14 P.A. Hume
Force-angle profle Force-time profle

100 100
80 80
Gate force (kg)

Gate force (kg)
60 60
40 40
20 20
0 0
–20 –20
–60 –40 –20 0 20 40 0 0.5 1 1.5
Gate angle (degrees) Time (s)
Fig. 3 Force–angle profile showing the full stroke, and force–time profile of the same stroke
showing time period from minimum to maximum angle. The vertical line indicates the point where
the blade changes direction
Selecting rowers with similar force profile characteristics was thought to most
likely result in a successful rowing crew (Mason et al. 1988; Wing and Woodburn
1995). However, differences in timing and magnitudes of force application may aid
a pair crew in minimizing lateral skiff deviations and therefore improve their
success.
Blade entry should be timed so that the blade matches the water speed when it is
approximately 50% covered, the hands should be raised 0.1 s before the catch, and
upward force should be applied to the handle to achieve as fast an entry as possible
(Kleshnev and Baker 2007). In order to prevent back splash and the negative forces
associated with it, a minimum angular velocity of the blade at the moment of catch
(blade first touching the water) should be (sinθ)/L, where L is the outboard and θ is
the angle the oar shaft makes with the hull (Macrossan and Macrossan 2006). This
minimizes the negative effects of early or late entry.
Due to the inboard length of the oars used during sculling, the oar handles must
overlap when the blades are perpendicular to the skiff, resulting in upper body
asymmetry. This asymmetry may lead to discrepancy in stroke and bow side force.
Bow oar force of eight rowers sculling has been reported as 11.7%, 9.5%, and 4.0%
greater than stroke oar force at 24, 26, and 28 spm, respectively (Elliott et al. 2002).
Greater force application on one blade may result in greater yawing (movement
about the longitudinal axis of the skiff), which is reported to negatively correlate to
skiff velocity (Loschner et al. 2000).
Foot-Stretcher Forces
Foot-stretchers are the plates used to position the feet in a skiff. Foot-stretchers
provide a solid base (with respect to the skiff) against which the athlete can apply
force to produce forward propulsion of the skiff. At the catch, foot-stretcher forces
have ranged from 100 to 392 N depending on whether data were collected during
on-water sweep rowing by elite coxless pairs (Smith and Draper 2002) or single
scullers (Korndle and Lippens 1988). Foot-stretcher forces are reported to peak
between 299 and 600 N during the drive phase (Zatsiorsky and Yakunin 1991). The
large variation in peak foot-stretcher forces recorded may be due to analysis of sweep
rowers versus scullers, on-water versus simulator rowing, or the skill level of
participants.
It is important to initiate each stroke with powerful and symmetrical loading of
the foot-stretchers. The magnitude and symmetry of bilateral foot forces (relative
resultant, vertical, and horizontal forces) was reported (Buckeridge et al. 2014) for
five heavyweight female scullers, six heavyweight female sweep rowers, and six
lightweight male rowers who performed an incremental step test on the Concept II
ergometer. Asymmetries ranged from 5.3% for average resultant force to 28.9% for
timing of peak vertical force, but were not sensitive to stroke rate.
The magnitude of applied foot-stretcher forces during on-water sculling by elite
female scullers have differed between stroke and bow side feet by 55 N (18.4%)
(Loschner and Smith 2000). Due to the asymmetrical nature of on-water sweep
rowing, discrepancies in the magnitude of force application (~33% or 200 N greater
force application on the oar-side leg compared to the nonoar-side leg to each foot-
stretcher) are more significant (Korndle and Lippens 1988).
The time of peak foot-stretcher forces during a stroke cycle is most commonly
described with respect to the time or angle of peak oar/pin force and is reported to
reach peak magnitude prior to (Korndle and Lippens 1988; MacFarlane et al. 1997),
at the same time (Korndle and Lippens 1988), or following (Loschner and Smith
2000) peak oar forces.
The relationship between kinematics and foot kinetics have been reported.
Twenty elite female rowers performed an incremental rowing test on an instru-
mented rowing ergometer, which measured force at the handle and foot-stretchers,
while three-dimensional kinematic recordings of the ankle, knee, hip, and lumbar-
pelvic joints were made. Hip kinematics were a key predictor of foot force output
(R2 = 0.48), whereas knee and lumbar-pelvic kinematics were the main determi-
nants in optimizing the horizontal foot force component (R2 = 0.41). Bilateral
asymmetries of the foot-stretchers significantly influenced lumbar-pelvic kinematics
(R2 = 0.43) and pelvic twisting (R2 = 0.32) during the rowing stroke (Buckeridge
et al. 2015).
Blade Forces
Blade force is difficult to measure as opposed to handle or pin force. Blade force
during on-water sculling has been measured by positioning a linear proximity
transducer on the outboard of each oar, closer to the blades (Elliott et al. 2002).
Force readings were substantially smaller (107.3–187.0 N) than previously reported
handle or pin force values. Changes in right blade maximum force increased
8.9% and 9.7% when stroke rate increased from 24 to 28 spm ( p = 0.02) and
26 to 28 spm, respectively (Elliott et al. 2002). Equivalent changes in left oar forces
were 1.0% and 2.7%.
16 P.A. Hume
Given elite rowers’ perceptions of on-water sculling stroke success have been
associated with catch efficiency variables measured with a boat instrumentation
system (Millar et al. 2015), it may be that instrumentation systems should be used
with developing rowers until they are able to perceive the differences between good
and poor strokes.
Summary
Laboratory-based biomechanical studies of rowing using predominately videogra-

phy and force methods have provided key information regarding rowing technique.
With advances in boat instrumentation, sports biomechanists are now able to assess
and provide relatively real-time oar/pin and foot-stretcher force feedback for
on-water rowing performance. Real-time segmental sequencing information is still
limited during on-water analyses. However, feedback may improve with advances in
technology such as isoinertial units and real-time feedback via haptic or auditory
systems. More information is required to better match rower characteristics to
specific boat types for effective performance and to understand how precise alter-
ations to technique and setup can alter rowing force profiles. Estimates of the
magnitude of impact of all the independent variables contributing to average
and/or instantaneous boat velocity are needed. Modelling concepts in rowing are
needed.
Acknowledgements Thanks are given to Dr. Clara Soper for providing Fig. 1 from her Sports
Medicine paper that I adapted for this text book. Thanks are given to Dr. Sarah Kate Millar for
getting photos taken of herself on ergometers and on-water for Fig. 2. Thanks are given to Dr. Jennie
Coker for providing Fig. 3 on pin force profiles contributed from her PhD thesis (Coker 2010)
toward this chapter.
Cross-References

▶ 3D Dynamic Pose Estimation Using Inertial Measurement Units
Sensors
▶ 3D Dynamic Pose Estimation Using Surface Parameterization (MoSh)
▶ 3D Dynamic Pose Estimation Using Wearable Robotics
▶ 3D Dynamic Probabilistic Pose Estimation Using Cameras and Reflective
Markers.
▶ 3D Muscle, Ligament and Tendon Motion Using Ultrasound
▶ A Physics-Based Statistical Model for Human Gait Analysis
▶ Conventional Gait Model - Success and Limitations

▶ Frequency Analysis/Wavelet Analysis
▶ Induced Acceleration and Induced Power Analyses
▶ Simulation of Soft Tissue Loading That Generates Observed Movement
Dynamics
▶ Surface EMG
▶ The Movement of Scull and Oar Rowing
▶ Time Series Analysis/Time Series Processing of Signals
References
Affeld K, Schichl K, Ziemann A (1993) Assessment of rowing efficiency. Int J Sports Med
14(Suppl 1):S39–S41
Anderson R, Harrison AJ, Lyons GM (2005) Accelerometry-based feedback – can it improve
movement consistency and performance in rowing? Sports Biomech 4:179
Baird D, Soroka WW (1951) Measurement of force-time relations in racing skills. Am Soc Mech
Eng 58:77–85
Barrett RS, Manning JM (2004) Relationships between rigging set-up, anthropometry, physical
capacity, rowing kinematics and rowing performance. Sports Biomech 3:221–235
Baudouin A, Hawkins D (2004) Investigation of biomechanical factors affecting rowing perfor-
mance. J Biomech 37(7):969–976
Benson A, Abendroth J, King D, Swensen T (2011) Comparison of rowing on a Concept 2 station-
ary and dynamic ergometer. J Sports Sci Med 10(2):267–273
Bingul BM, Bulgan C, Aydin M, Buyukdemirtas T, Ozbek A (2014) Two-dimensional kinematic
analysis of catch and finish positions during a 2000 m rowing ergometer time trial. South Afr
J Res Sport Phys Educ Recreation 36(3):1–10
18 P.A. Hume
Bompa TO (1980) Technique and muscle force. Can J Appl Sport Sci 5(4):245–249
Buck DP, Smith R, Sinclair PJ (2000) Dynamic characterisation of Concept II and Rowperfect
ergometer rowing. Symposium conducted at the meeting of the Australiasian biomechanics
conference 3, Brisbane, Australia
Buckeridge EM, Bull AMJ, McGregor AH (2014) Foot force production and asymmetries in elite
rowers. Sports Biomech 13(1):47–61
Buckeridge EM, Bull AMJ, McGregor AH (2015) Biomechanical determinants of elite rowing
technique and performance. Scand J Med Sci Sports 25(2):e176–e183
Coker J (2010) Using a boat instrumentation system to measure and improve elite rowing technique
and set-up for on-water rowing performance. PhD Dissertation. AUT University, Auckland
Coker J, Hume PA, Nolte V (2008) Evaluating rowing force profiles: implications from literature.
Symposium conducted at the meeting of the New Zealand sports medicine and science confer-
ence, Dunedin
Coker J, HumePA, Nolte V (2009) Validity of the powerline boat instrumentation system.
In: Harrison D, Anderson R, Kenny I (Chair) Symposium conducted at the meeting of the
27th international conference of biomechanics in sports, Limerick, Ireland
Dal Monte A, Komor A (1989) Rowing and sculling mechanics. In: Vaughan CL
(ed) Biomechanics of Sport. CRC Press, Boca Raton, pp 53–119
Dawson RG, Lockwook RJ, Wilson JD, Freeman G (1998) The rowing cycle: sources of variance
and invariance in ergometer and on-the-water performance. J Mot Behav 33(1):33–43
Doyle MM, Lyttle AD, Elliott BC (2010) Comparison of force-related performance indicators
between heavyweight and lightweight rowers. Sports Biomech 9(3):178–192
Elliott BC, Lyttle AD, Birkett O (2002) The RowPerfect ergometer: a training aid for on-water
single scull rowing. J Sports Biomech 1(2):123–134
Fohanno V, Nordez A, Smith RM, Colloud F (2015) Asymmetry in elite rowers: effect of ergometer
design and stroke rate. Sports Biomech 14(3):310–322
Greene AJ, Sinclair PJ, Dickson MH, Colloud F, Smith RM (2013) The effect of ergometer design
on rowing stroke mechanics. Scand J Med Sci Sports 23(4):468–477
Halliday SE, Zavatsky AB, Andrews BJ, Hase K (2001) Kinematics of the upper and lower
extremities in three-dimensions during ergometer rowing. In: Muller R, Gerber H, Stacoff A
(Chair) Symposium conducted at the meeting of the international society of biomechanics,
Interrepro AG, Zurich, Switzerland, 8–13 July
Hartmann U, Mader A, Wasser K, Klauer I (1993) Peak force, velocity, and power during five and
ten maximal rowing ergometer strokes by world class female and male rowers. Int J Sports Med
14(Suppl 1):S42–S45
Hawkins D (2000) A new instrumentation system for training rowers. J Biomech 33:241–245
Herberger E, Klavora P (1985) Rigging: part two. Foot stretcher adjustment. Rowing 28(307):
ISSN: 0035–8584
Hill H (2002) Dynamics of coordination within elite rowing crews: evidence from force pattern
analysis. J Sports Sci 20(2):101–117
Hill H, Fahrig S (2009) The impact of fluctuations in boat velocity during the rowing cycle on race
time. Scand J Med Sci Sports 19(4):585–594
Hume PA, Soper CS (2001) Segmental sequencing reports (Athlete report). Auckland University of
Technology, Auckland
Hume PA, Soper CS, Joe G, Williams T, Aitchison D, Gunn S (2000) Effects of foot-stretcher angle
on the drive phase in ergometer rowing. Symposium conducted at the meeting of the sports
medicine Australia, Book of abstracts: 2000 Pre-Olympic congress: international congress on
sport science, sports medicine and physical education, Brisbane, Australia
Hume PA, Reid DA, Benschop A (2001) Back injury in rowing – the effectiveness of the Glasstron
goggles as a coaching tool in changing pelvis angle: a pilot study. Sport Science New Zealand.
Symposium conducted at the meeting of the developing sport – the next steps: conference 2001,
Wellington, New Zealand
Hume PA, Soper CS, Zeinstra J (2003) The effectiveness of visual feedback on changing rowing
technique. NZ Coach 11(4):22–24
Hume PA, Soper CS, Reid D, Tong R (2005a) Foot-stretcher angle does affect rowing ergometer
performance. Successful Coach UK 21(April):4
Hume PA, Soper CS, Zeinstra J (2005b) Visual feedback to change rowing technique. Successful
Coach UK 19(February):7–9
Kleshnev V (2002) Oar inertia force. Different catch and release angles in stroke and bow seats of
a pair. Rowing Biomech Newsl 2(4)
Kleshnev V (2006) Why is a long catch not a waste of energy; Why is a front loaded drive more
efficient? Rowing Biomech Newsl 6(63)
Kleshnev V (2007a) Catch and release slips expressed in video frames; Plotting force curve relative
time or oar angle. Rowing Biomech Newsl 7(75)
Kleshnev V (2007b) Using the elastic energy of the oar shaft. Rowing Biomech Newsl 7(80)
Kleshnev V, Baker T (2007) Understanding rowing technique. The timing of the catch. Rowing and
Regatta
Kleshnev V, Kleshnev I (1998) Dependence of rowing performance and efficiency on motor
coordination of the main body segments. J Sports Sci 16(5):418–419
Korndle H, Lippens V (1988) Do rowers have a particular “footwriting”? In Biomechanics in Sport
(266502, pp 7–11). Institution of Mechanical Engineers, London
Lamb DH (1989) A kinematic comparison of ergometer and on-water rowing. Am J Sports Med
17(3):367–373
Li C, Ho W, Lin H (2007) Strength curve characteristics of rowing performance from the water and
the land. J Biomech 40(Suppl 2):S770
Loschner C, Smith R (2000) The relationship between pin forces and individual feet forces applied
during sculling. Symposium conducted at the meeting of the 3rd Australasian biomechanics
conference, Griffith University, Gold Coast
Loschner C, Smith R, Galloway M (2000) Intra-stroke boat orientation during single sculling.
In: Hong Y, Johns D (Chair). Symposium conducted at the meeting of the XVIIIth international
symposium on biomechanics in sports, The Chinese University of Hong Kong, Hong Kong,
25–30 June
MacFarlane DJ, Edmond IM, Walmsley A (1997) Instrumentation of an ergometer to monitor the
reliability of rowing performance. J Sports Sci 15(2):167–173
Macrossan MN, Macrossan NW (2006) Back-splash in rowing-shell propulsion. University of
Queensland, Brisbane
Manning JM, Osbourne M, Hume PA (2001). Video analysis of the rowing stroke: segmental
interaction biomechanics. NZ Rowing Mag, February
Martin TP, Bernfield YS (1980) Effect of stroke rate on velocity of a rowing shell. Med Sci Sports
Exerc 12(4):250–256
Martindale WO, Robertson DGE (1984) Mechanical energy in sculling and in rowing an ergometer.
Can J Appl Sport Sci 9:153–163
Mason BR, Shakespear PB, Doherty P (1988) The use of biomechanical analysis in rowing to
monitor the effect of training. Excel 4(4):7–11
Mattes K, Schaffert N (2010) New measuring and on water coaching device for rowing. J Hum
Sport Exerc 5(2):226–239
McBride ME (1998) The role of individual and crew technique in the optimisation of boat velocity
in rowing. PhD Dissertation. University of Western Australia, Perth
McBride ME, Elliott BC (1999) Use of real-time telemetry to monitor instantaneous seat and boat
velocity in pair oared rowing. In: Sanders RH, Gibson BJ (Chair) Symposium conducted at the
meeting of the XVIIth international symposium on biomechanics in sports, School of Biomed-
ical and Sports Science, Edith Cowan University, Perth, Western Australia
McDonnell LK, Hume PA, Nolte V (2011) Rib stress fractures among rowers: definition, epidemi-
ology, mechanisms, risk factors and effectiveness of injury prevention strategies. Sports Med
41(11):883–901
20 P.A. Hume
Millar SK, Oldham RH, Hume PA, Renshaw I (2015) Using rowers’ perceptions of on-water stroke
success to evaluate sculling catch efficiency variables via a boat instrumentation system. Sports
3(4):335–345. doi:10.3390/sports3040335
Millar SK, Hume PA, McDonnell L (2016) Does a modified foot-stretcher improve 500-m on-water
and ergometer rowing performance time and comfort? Symposium conducted at the meeting of
the 34th international society of biomechanics in sports, Tsukuba, Japan
Millward A (1987) A study of the forces exerted by an oarsman and the effect on boat speed.
J Sports Sci 5(2):93–103
Najafi B, Lee-Eng J, Wrobel JS, Goebel R (2015) Estimation of center of mass trajectory using
wearable sensors during golf swing. J Sports Sci Med 14(2):354–363
Nelson WN, Widule CJ (1983) Kinematic analysis and efficiency estimate of intercollegiate female
rowers. Med Sci Sports Exerc 15(6):535–541
Nolte V (1991) Introduction to the biomechanics of rowing. FISA Coach 2(1):1–6
Nolte V, Morrow A (2002, Winter) Coach boat view. Biomechanics of rowing. Rowing Canada
Aviron
Ogurkowska M, Kawalek K, Zygmanska M (2015) Biomechanical characteristics of rowing.
Trends Sport Sci 22(2):61–69
Peltonen J, Rusko H (1993) Interrelations between power, force production and energy metabolism
in maximal leg work using a modified rowing ergometer. J Sports Sci 11(3):233–240
Pelz PF, Vergé A (2014) Validated biomechanical model for efficiency and speed of rowing.
J Biomech 47(13):3415–3422
Reid D, McNair P (2000) Factors contributing to low back pain in rowers. Br J Sports Med
34(5):321–325
Roth W, Schwanitz P, Pas P, Bauer P (1993) Force-time characteristics of the rowing stroke and
corresponding physiological muscle adaptations. Int J Sports Med 14(Suppl 1):S32–S34
Ryan-Tanner R, Hahn A, Lawton E, Bellenger S, Kearns A (1999) A comparison of “performance”
during on-water racing and ergometer rowing. Symposium conducted at the meeting of the 5th
IOC world congress on sport sciences, Sports Medicine Australia, Sydney, Australia
Sanderson B, Martindale W (1986) Towards optimizing rowing technique. Med Sci Sports Exerc
18(4):454–468
Schabort EJ, Hawley JA, Hopkins W, Blum H (1999) High reliability of performance of well-
trained rowers on a rowing ergometer. J Sports Sci 17(8):627–632
Schaffert N, Mattes K (2015) Effects of acoustic feedback training in elite-standard Para-Rowing.
J Sports Sci 33(4):411–418
Smith R, Draper C (2002) Quantitative characteristic of coxless pair-oar rowing. In: Gianikellis KE
(Chair) Symposium conducted at the meeting of the XXth international symposium on biome-
chanics in sports, Universidad de Extremadura. Servicio de Publicaciones, Caceres, Spain
Smith R, Draper C (2006) Skill variables discriminate between the elite and sub-elite in coxless pair
oared rowing. In: International Society of Biomechanics in Sports, Proceedings of XXIV
international symposium on biomechanics in sports 2006 (pp. 343–346), University of Salz-
burg, Salzburg, Austria
Smith R, Loschner C (2002) Biomechanics feedback for rowing. J Sports Sci 20(10):783–791
Smith R, Spinks WL (1995) Discriminant analysis of biomechanical differences between novice,
good and elite rowers. J Sports Sci 13(5):377–385
Smith R, Spinks WL (1998) A system for the biomechanical assessment of rowing performance
(Rowsys). J Hum Mov Stud 34:141–157
Smoljanovic T, Bojanic I, Hannafin JA, Hren D, Delimar D, Pecina M (2009) Traumatic and
overuse injuries among international elite junior rowers. Am J Sports Med 37(6):1193–1199
Soper CS, Hume PA (2004a) Reliability of power output during rowing changes with ergometer
type and race distance. Sport Biomech 3(2):237–247
Soper CS, Hume PA (2004b) Towards an ideal rowing technique for performance: the contributions
from biomechanics. Sports Med 34(9):825–848
Soper CS, Hume PA, Reid D, Tonks R (2002) The effectiveness of the Goggle Training System as
a coaching tool in changing pelvis angle at the catch during on-water rowing. In: Gianikellis KE
(Chair) Symposium conducted at the meeting of the XXth international symposium on biome-
chanics in sports, Universidad de Extremadura, Servicio de Publicaciones, Caceres, Spain.
Soper CS, Hume PA, Tonks D (2005) What is an ideal rowing technique? How can biomechanics
help? NZ Rowing Magaz
Spinks WL (1996) Force-angle profile analysis in rowing. J Hum Mov Stud 31(5):211–233
Stallard M (1999) The challenge of rowers backache. Sport Medicine Today 1(2):53–55
Torres-Moreno R, Tanaka C, Penney KL (1999) Joint excursion, handle velocity, and applied force:
a biomechanical analysis of ergonometric rowing. Int J Sports Med 21(1):41–44
Wilson F, Gissane C, Simms C, Gormley J (2008) A 12 month prospective cohort study of injury in
international rowers. Br Med J 44:207–214
Wing AM, Woodburn C (1995) The coordination and consistency of rowers in a racing eight.
J Sports Sci 13(3):187–197
Zatsiorsky VM, Yakunin N (1991) Mechanics and biomechanics of rowing: a review. Int J Sport
Biomech 7:229–281
Movement Analysis of the Golf Swing
P. A. Hume and J. Keogh
Abstract
Golf is popular worldwide. The complex and asymmetrical motion of golf
swings, involving substantial forces and ranges of rotational movement, demands
good technique if the player is to perform well while avoiding injury. Biome-
chanical analyzes can help in maximizing distance and accuracy of golf shots, and
reducing injury risk, by providing qualitative and quantitative evidence of body
angles, joint forces, and muscle activity patterns. We highlight how lab-based
biomechanical studies of the golf swing, using 3D videography, force, electro-
myography, modelling, and robots, have provided key information regarding the
golf swing. With recent advances in technology such as isoinertial devices and
wearable sensors, much of this knowledge can now be applied on the golf course.
Sports biomechanists are now able to assess and provide relatively real-time
feedback on golf swing performance. Golf performance outcome measures
from a biomechanics perspective may be categorized as direct (ball displacement,
shot accuracy) or indirect (clubhead velocity, clubface angle). The magnitude and
direction of the linear clubhead velocity at impact is determined by the angular
velocity of the clubhead during the swing and the length of the arm–club system.
Golfers must develop a consistent fundamental swing pattern to secure these
qualities. Proper proximal to distal sequencing of body rotations is important in
the production of high-clubhead velocities in the golf swing. The biomechanist
must be able to measure golf performance technique and outcome variables
reliably and with accuracy/sensitivity so feedback can be given to golfers and
coaches on technique improvement.
P.A. Hume (*)

Auckland University of Technology, Auckland, New Zealand
e-mail: patria.hume@aut.ac.nz
J. Keogh
Bond University Australia, Robina, QLD, Australia
e-mail: jkeogh@bond.edu.au

DOI 10.1007/978-3-319-30808-1_137-1
2 P.A. Hume and J. Keogh
Keywords
Golf • Clubhead velocity • Clubface angle • Shot accuracy • Kinematics •
Kinetics • Feedback • Technique • Body kinematics • Joint angles • Segment
velocities • Segmental sequencing • Inertial sensors • Telemetered sensors • Real-
time feedback • Performance • Injury prevention
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Predictors of Clubhead Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Biomechanics of Golf Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Video Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Electromagnetic Motion Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Wearable Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Force Plate Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Electromyography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Modelling/Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Robotics and Instrumented Golf Clubs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
Golf is popular worldwide, with people of all ages and genders participating in the
sport. The aim of golf is to hit the golf ball into a small hole in as few shots as
possible (Hume et al. 2005). The techniques used for golf swings during driving,
chipping, and putting are key to performance and reduction of injury risk (Alexander
and Kern 2005; Hellström 2009). The complex and asymmetrical motion of the golf
swing, involving substantial forces and ranges of rotational movement, demands
good technique if the player is to avoid injuring their back, knees, elbows, or wrists.
For example, the golf swing generates forces and torques which tend to cause
internal or external rotation of the tibia on the femur, which are resisted by the
knee ligaments and menisci (Marshall and McNair 2013). Golfers may also develop
medial epicondylosis, commonly known as “golfer’s elbow,” because of the repet-
itive stress placed on the medial elbow soft tissues during pronation and wrist flexion
(Hume et al. 2006). Back pain and injuries are most frequently reported by golfers,
and may be due to golf swing techniques that place an emphasis on restricting pelvic
turn while increasing thorax rotation during the backswing to generate higher
clubhead speeds at impact (Cole and Grimshaw 2016).
Biomechanical analyzes can help in maximizing distance and accuracy of
golf shots, and reducing injury risk, by providing qualitative and quantitative
evidence of body angles, joint forces, and muscle activity patterns (Hume et al.
2005; Keogh and Hume 2012). Golfer movement has traditionally been assessed
Movement Analysis of the Golf Swing 3
using two-dimensional (2D) or three-dimensional (3D) high-speed video or opto-

electronic systems, to provide joint angles and segment velocities. Changes in
technology have made it possible to measure and analyze golf movements in
many ways and with new levels of precision. Advances in technology have seen
the use of 3D force plates, pressure sensors, electromyography, computer simula-
tions/modelling, inertial sensors, and instrumented golf clubs, for golf technique
analyzes. Ball velocity has been measured using a variety of ball launch systems
(Bertram and Guadagnoli 2008; Wallace et al. 2007), with the TrackMan system
being one of the newer devices (Oliver et al. 2016; Robertson et al. 2012). TrackMan
(www.trackman.dk) enables 3D golf ball trajectory and other animated data to be
superimposed and compared relative to a target direction indicated in an image
(by combining capture from a camera and radar ball direction).
Biomechanics methods have progressed over the years to enable measurement of
golf techniques. Golf performance outcome measures from a biomechanics perspec-
tive may be categorized as direct (ball displacement, shot accuracy) or indirect
(clubhead velocity, clubface angle) (Keogh and Hume 2012). The magnitude and
direction of the linear clubhead velocity at impact is determined by the angular
velocity of the clubhead during the swing and the length of the arm–club system.
Golfers must develop a consistent fundamental swing pattern to secure these qual-
ities (Kwon 2010). Proper proximal to distal sequencing of body rotations is
important in the production of high-clubhead velocities in the golf swing. The
biomechanist must be able to measure the golf performance technique and outcome
variables reliably and with accuracy/sensitivity so feedback can be given to golfers
and coaches on what needs to be the focus of technique improvement.
State of the Art
This chapter will focus on summarizing the literature in relation to the movement
analysis of the golf swing. Initially, this chapter will seek to provide a state-of-the-art
review of the primary predictors of clubhead velocity. State-of-the-art knowledge of
biomechanical methods used to assess golf technique including biomechanical
modeling of the swing is then presented. The applications and potential limitations
of these biomechanical methods for improving golf performance and preventing
overuse injuries are also described.
Predictors of Clubhead Velocity
Golf swing clubhead velocity is associated with driving distance (Keogh et al. 2007).
Low-handicap performance is also associated with high-clubhead velocities. For
example, clubhead velocity has been correlated (r = 0.950) with golf handicap
(clubhead velocity = 4.065–0.0214 hcp) for 45 male golfers (18–80 years,

2–27 hcp) (Fradkin et al. 2004).
Biomechanical studies have improved our understanding of the variables that
characterize a successful golf swing. Clubhead velocity has also traditionally been
measured using laboratory-based, 3D kinematic analysis systems (Egret et al. 2003;
Horan et al. 2010). More recently, clubhead velocity has been measured in the field
using a Stalker Professional Sports radar gun (Applied Concepts, Plano, TX, USA)
operating at a frequency of 34.7 GHz (Bradshaw et al. 2009; Keogh et al. 2009) or
launch monitor systems such as the TrackMan (Oliver et al. 2016; Robertson et al.
2012). Clubhead velocity may be improved by increasing downswing amplitude
(Delay et al. 1997), the sequential acceleration of body parts, (Fletcher and Hartwell
2004) and shaft flexibility (Nesbit 2005a). Individual golfer analysis of work, power,
force, and torque during golf swings may help improve clubhead velocity (Nesbit
2005b). Differences in reported clubhead velocities in studies are probably a result of
the experience and skill of the golfers (novice-, high-, and low-handicap players, and
professional golfers), the clubs used (drivers, irons, putters), as well as analysis
methodologies. Researchers have reported correlations with clubhead velocity for
golfer swing characteristics such as total work (r = 0.431) (Nesbit 2005b), centre of
pressure (r = 0.46–0.58) (Ball and Best 2007b), and ball impact locations higher on
the club faces of drivers and 5-irons (r = 0.65) (Williams and Sih 2002).
Biomechanical studies have suggested that clubhead velocity can be improved
by technique alterations such as reducing the radius path of the hands during the
downswing, increasing wrist torque and/or range of motion, delaying wrist motion to
late in the downswing, increasing downswing amplitude, improving sequential
acceleration of body parts, improving weight transfer, and utilizing X-factor stretch
(Keogh and Hume 2012). There is also the potential that a number of anthropometric
(Kawashima et al. 2003), flexibility, muscular strength, and endurance characteris-
tics may predict clubhead velocity in low- and high-handicap golfers (Keogh et al.
2009) and young elite players (Torres-Ronda et al. 2014).
When using biomechanics methods, researchers have attempted to identify the
predictors of performance using low-handicap golfers. For example, a study of the
relationship between swing technique characteristics and performance in golf driv-
ing for low-handicap female golfers (Brown et al. 2011) reported there was no
common driver swing technique for optimal performance. The recommendation
for individual swing characteristic determination was based on analyzes for the
drive swings of 16 golfers using a 12-camera 3D motion capture system and a
stereoscopic launch monitor. Three covariates (increased pelvis–thorax differential
at the top of the backswing, increased pelvis translation during the backswing, and
a decrease in absolute backswing time) were determinants of the variance in club-
head velocity (adjusted r2 = 0.965, P < 0.05). There were significant correlations
between clubhead velocity and left-hand grip strength (r = 0.54, P < 0.05), and
sitting flexibility (i.e., trunk rotation in the transverse plane while sitting) (clock-
wise: r = 0.522, P < 0.05; counterclockwise: r = 0.711, P < 0.01).
Biomechanics of Golf Technique
Video Kinematics
Golf coaches can use simple 2D video analysis to perform a range of qualitative and
quantitative analyses of the golf swing in the field. Qualitative and quantitative
assessments of relevant biomechanical variables such as the radius path of the
hands during the downswing, range of motion and timing of wrist motion in the
downswing, downswing amplitude, and sequential acceleration of body parts may
be obtained from 2D analyzes and help provide feedback to golfers that will improve
clubhead velocity and golf ball distance (Keogh and Hume 2012). However, the
disadvantages are that the accuracy of the quantitative measurements is less than
laboratory-based 3D analyses, and usually data extraction is time intensive due to the
manual nature of analysis, rather than semiautomatic analysis provided during
laboratory based assessments.
Due to the highly complex multijoint motions in golf, kinematic studies have
focused on both the clubhead and various body segments (Egret et al. 2003; Zheng
et al. 2008). As a total body sequential movement, the golf swing requires certain
generic coordinative patterns to be effectively performed. Many studies have exam-
ined the nature of the coordination between the pelvis, torso, and upper limb during
the golf swing (Evans et al. 2012; Horan et al. 2011; Horan et al. 2010). Strong
coupling between the thorax and pelvis (average R2 = 0.92 across all directions) is a
method for simplifying the motor control strategy used during the downswing, and
a way of ensuring consistent motor patterns (Horan and Kavanagh 2012). Three-
dimensional data collected for 14 male professional golfers (age 27 8 years, golf-
playing experience 13.3 8 years) using an optical motion analysis system showed
the thorax segment had the highest peak velocities for the upper body during the
downswing. The head and thorax showed a more variable coupling relationship
(average R2 = 0.76 across all directions).
Thorax and pelvis kinematics were shown to change during golf swings for male
and female skilled golfers following the performance of 40 min of putting practice
that challenged the trunk muscle endurance of the golfers (Horan et al. 2014). The
golfers were encouraged not to move from the spot, therefore maximizing the time in
which their trunk was inclined forward. Angular displacement at address, top of
backswing and ball contact for the pelvis, thorax, and pelvis–thorax interaction were
calculated, in addition to the magnitude of peak angular velocity and repeatability of
continuous segment angular velocities. Female golfers had less pelvis and thorax
anterior–posterior tilt at address, less thorax and thorax–pelvis axial rotation at top of
backswing, and less pelvis and thorax axial rotation and pelvis lateral tilt at ball
contact pre- to post-putting. Females had significantly lower thorax–pelvis lateral tilt
peak angular velocity pre- to post-putting. Therefore, the endurance-based putting
intervention affected females’ thorax and pelvis orientation angles and velocities to a
greater extent than males (Horan et al. 2014).
Trunk and X-Factor Variables

According to the kinetic link principle, rapid sequential stretching of the hip, trunk,
and upper limb muscles during the backswing would allow greater muscular forces
and torques to be produced, with the ultimate result being an increase in clubhead
velocity. The difference in hip and trunk angles at the top of the backswing was
initially referred to as the X-factor (McLean 1992) with the increase in this angle
early in the downswing called the X-factor stretch. Evidence for the benefit of the
X-factor and X-factor stretch in increasing clubhead velocity (Brown et al. 2013;
Cole and Grimshaw 2009; Evans et al. 2012; Kwon et al. 2013) remains equivocal.
The equivalence may reflect a range of methodological differences between studies.
For example, X-factor and trunk kinematic (flexion/extension, lateral bend, axial
rotation) variables may be quite different when calculated by 2D compared to 3D
methods (Smith et al. 2016). Differences between 2D and 3D X-factor (~16 ) result
can be explained by projection errors introduced to 2D analysis through trunk
flexion ( 6.5 3.6 ) at the top of the backswing and trunk right-side lateral bend
(8.7 2.9 ) of the trunk and pelvis segments at impact. Therefore, a 3D method for
kinematic data calculation would appear necessary to accurately analyze most
aspects of the golf swing.
In a study examining the performance of 20 drives, eight high-handicap golfers
(11–20 hcp) had greater hip rotation at the top of the backswing and reduced
maximum X-factor compared with seven low-handicap golfers (0–10 hcp) (Cole
and Grimshaw 2009). In a study of ten professional and five amateur male golfers
(Meister et al. 2011), clubhead velocity at impact was correlated to X-factor at
impact (r = 0.94), peak X-factor (r = 0.90), peak upper-torso rotation (r = 0.90),
and peak free moment per kilogram (r = 0.94). Downswing was initiated by reversal
of pelvic rotation, followed by reversal of upper-torso rotation. Peak X-factor
preceded peak free moment in all swings for all golfers and occurred during initial
downswing.
However, methodological considerations for the 3D measurement of the X-factor
and lower trunk movement in golf continue to be debated (Brown et al. 2013; Cole
and Grimshaw 2009; Joyce et al. 2010; Kwon et al. 2013; Wheat et al. 2007). There
is currently no universal method of measuring rotational angles during the golf
swing. Different marker placement methods and analysis models can result in
different outcomes for X-factor.
A lateral bending/flexion-extension/axial rotation (ZYX) order of rotation was
deemed to be the most suitable Cardanic sequence to use in the assessment of the
X-factor and lower trunk movement in the golf swing in one study (Joyce et al.
2010). However, in an evaluation (Brown et al. 2013) of three common methods
used to analyze the X-Factor there were significant differences for events during
the swing. A twelve-camera 3D motion capture system captured driver swings
of 19 golfers. Subject-specific 3D biomechanical models were created with the
position and orientation of each model estimated using a global optimization algo-
rithm. Data for each kinematic measure were derived as a times series for all three
methods. Regression analysis of these data showed that whilst one method could be
successfully mapped to another, the mappings between methods were subject-

dependent. A consistent methodology considering the X-Factor from a joint angle
approach was recommended for describing intersegment kinematics during golf
swings.
The validity of X-factor computation methods and whether direct relationships
exist between X-factor parameters and clubhead velocity have been examined in a
group of skilled male golfers (n = 18, 0.6 2.1 hcp) (Kwon et al. 2013). Five
driver trials were captured from each golfer using an optical motion capture system
(250 Hz). Two plane-based methods (conventional vs. functional swing plane-based)
and one Cardan rotation-based method (relative orientation) were used to compute
select X-factor (end of pelvis rotation, top of backswing, ball impact, and maxi-
mum), X-factor stretch (stretch and maximum stretch), and X-factor velocity (ball
impact and maximum) variables. The three computational methods generated sig-
nificantly different X-factor parameter values. The conventional method provided
substantially larger X-factor values than the other methods in the untwisting phase,
and the correlation analysis revealed that X-factor variables were not directly related
to maximum clubhead velocity. The meaningfulness of select X-factor parameters
generated by the conventional method was deemed questionable (Kwon et al. 2013).
Hip Rotation Motion Improvements

Appropriate pelvic rotations are important in producing an efficient strike on the ball
so that the maximum amount of energy is transferred from the club to the ball. The
smash factor has been defined as the ratio of club velocity before impact to ball
velocity after impact (Lynn et al. 2013). The smash factor is one of the variables
included in the TrackMan launch monitor outputs (http://trackmangolf.com/). There
are differences in pelvic kinematics and ball flight determinants between golfers of
different skill levels. Skilled golfers have greater pelvic accelerations and a larger
time between peak pelvic velocity and ball contact (peak to impact time) than less-
skilled individuals (Lynn et al. 2013).
While flexibility has long been considered important for golfers, improving
passive hip ROM may not be associated with increased hip rotation utilized during
the golf swing. In a study of seven skilled male golfers between the ages of 18 and
40 years, dynamic kinematic hip rotation data were captured with a high-speed
optoelectric 3D motion capture system during a golf swing. There were weak
correlations between passive and dynamic range of rotation in lead and trail hips
during the golf swings (Alderslade et al. 2015). Additionally, there was a lack of any
significant between-group difference and hip internal or external rotation between
lower and high handicap players; or any significant correlation between club head
speed and these hip rotational measures (Keogh et al. 2009).
Trunk Kinematics and Club Characteristics

Significant 3D trunk kinematic differences exist when using different clubs for
golf swings. For example, multiple regression analyzes with trunk and lower
trunk flexion/extension and lower trunk axial rotation variables explained variance
in clubhead velocity for a driver (33.7%) and five-iron (66.7%) in a study of
15 low-handicap male golfers (2.5 1.9 hcp) (Joyce et al. 2013). A 10-camera
motion capture system operating at 250 Hz collected player kinematics, and a real-
time launch monitor collected clubhead velocity and ball velocity.
Kinematic factors that contribute to greater hitting distance when using the five-
iron club are greater hip joint movement and X-factor angle during the downswing,
shoulder flexion and less left shoulder internal rotation in the backswing, greater
extension angular velocity in both shoulders at early downswing, greater left shoul-
der adduction angular velocity at ball contact, and greater left elbow extension early
in the downswing (Healy et al. 2011). Three-dimensional marker coordinate data
were collected at 250 Hz to calculate joint kinematics, while a swing analyzer
measured club swing and ball launch characteristics for 30 male golfers in high or
low ball launch velocity (high: 52.9 2.1 m s 1; low: 39.9 5.2 m s 1) groups.
Kinematics and clubhead velocity are different depending on the type of clubs
used (e.g., driver, five-iron, pitching-wedge). However, similar timing (movement
time and proportion for each phase of the swing) may occur when using different
clubs according to the results from a study of seven right-handed male golfers
(Egret et al. 2003). Kinematic data during swings using three different clubs were
collected with an optoelectronic system with five cameras operating at 50 Hz. Low
sampling systems (i.e., at 50 Hz) may not be sufficient for such analyzes when the
driver is being used. Clubhead speed was measured using a swing-made detector
(Bell-Tronics, Ltd., Covington, USA).
Club Characteristics and Golf Performance

Golf club shaft stiffness can affect ball launch conditions by altering clubhead speed
and/or loft angle at ball impact (Worobets and Stefanyshyn 2012). Forty golfers hit
10 balls with each of five drivers varying in shaft stiffness from “Ladies” to “Extra-
Stiff,” in a double-blind study design. The motions of three reflective markers
attached to the clubhead were captured with a high-speed motion analysis system.
At ball impact, shaft stiffness had a statistically significant influence on clubhead
velocity for 27 subjects, on loft angle for 11 subjects, and on lie angle for all
40 subjects. No significant effect of shaft stiffness was observed on face angle, in
to out path angle, or attack angle.
Shaft length affects golf-driving performance (Kenny et al. 2008). Clubhead and
ball launch conditions and drive distance and accuracy were determined for seven
golfers (0.21 2.41 hcp) who performed shots on a purpose-built practice hole
using a range of drivers (1.168 and 1.270 m). As shaft length increased initial ball
velocity increased (+1.8 m/s), and there was no decrease in accuracy. Ball launch
conditions of spin components and launch angle remained unaffected by shaft
length.
The effect of shaft stiffness on the dynamics of golf drives has been quantified
(Betzler et al. 2012). Twenty golfers performed swings with two clubs designed to
differ only in shaft bending stiffness. Wrist kinematics and clubhead presentation to
the ball were determined using optical motion capture systems in conjunction with a
radar device for capturing ball speed, launch angle, and spin. The less stiff club
increased clubhead (0.45%) and ball speeds (0.7%) as a result of a faster recovery of
the lower flex shaft from lag to lead bending just before impact and an increase of
0.4% in angular velocity of the grip of the lower flex club at impact. Decreases in
shaft stiffness led to more shaft bending at the transition from backswing to
downswing. However, taking effect sizes into account, the changes in shaft stiffness
in isolation did not have a meaningful effect on the measured parameters, for the type
of shaft investigated.
A 3D optoreflective system operating at 400 Hz has been used to capture swings
of 21 male golfers using their own drivers in an attempt to quantify 3D ball flight
characteristics and their relationship to clubhead kinematics at impact. Club-head
kinematics at impact (velocity, orientation, path, and centeredness) can explain a
significant proportion (adjusted r2 = 0.71–0.82) of early ball flight characteristics
including resultant velocity, launch angle, side angle, back spin, and side spin even
when generalized across individual clubs (Sweeney et al. 2013).
Golfer Characteristics and Golf Performance

Studies comparing golfers of varying age, gender, ability, and fitness may increase
our understanding of the determinants of golf ball displacement and accuracy
and provide golf coaches with additional information on how to coach a variety of
clients. Several studies have reported gender-related differences in golf swing
characteristics (Egret et al. 2006; Horan et al. 2010; Zheng et al. 2008). Egret et al.
(2006) reported that their five female golfers (20.4 4.2 years; 6.1 3.4 hcp)
seemed to produce a wide swing with larger hip and shoulder joint rotation angles
at the top of the backswing than their seven male golfers (23.1 2.9 years;
6.6 1.7 hcp), and the male golfers flexed their left knee more during the back-
swing. However, neither of these significant gender-related differences in the kine-
matics of the golf swing showed any significant relationships with clubhead velocity.
Horan et al. (2010) reported that 19 male skilled golfers (hcp <4; 26 7 years) had
different lateral and anterior posterior pelvis tilt than their 19 female skilled golfers
(hcp <4; 25 7 years). The results of these two studies suggest that what might be
considered optimal swing characteristics for male golfers may not necessarily apply
to female golfers. There been no studies that have directly compared the effect of age
on golf swing technique.
Golf also provides opportunities for athletes with a disability. Kinetic and
kinematic differences in the golf swing in one leg and two legs amputees has
been examined for two males and one female golfers (professional or amateur
levels) with different degrees of disability (Stastny et al. 2015). Each amputee
golfer performed five golf swings with the iron six with 3D data recorded. There
was a significant relationship between the level of amputation and a number of
kinetic and kinematic variables such as the X-factor and individual body angles.
The fluency and similarity of the golf swing did not depend on the level of
amputation. The most variable parameter was a weight-shift in all golfers. The
takeaway and horizontal force angle depended on the level of amputation rather
than individual technique.
Electromagnetic Motion Capture
One of the major issues affecting the ecological validity of golf biomechanical
research has been the difficulty in examining the biomechanics of the golf swing
in realistic field conditions. However, with the development of new technologies,
sports biomechanists are now able to quantify the 3D motion of golfers on the
golf course, thereby allowing research to be conducted in this environment. The
repeatability of 3D thorax and pelvis kinematics in the golf swing measured using a
field-based electromagnetic motion capture (Evans et al. 2012) has been reported as
suitable to assess swing kinematics for golfers in context-specific environments. Two
golfers (14 hcp) performed five swings with each of two clubs (five-iron and
driver) on consecutive days in indoor and outdoor locations.
Kinematic proximal-to-distal sequencing and speed-summation (successive
increase in peak segment angular velocity) have been demonstrated as common
characteristics of both partial and full-swing shots in golf players of different skill
levels and genders using a 3D electromagnetic tracking system (Polhemus Liberty)
(Tinmark et al. 2010). Forty-five golfers (11 male tournament professionals, 21 male
and 13 female elite amateurs) performed partial shots with a wedge to targets at three
submaximal distances, 40, 55, and 70 m, and full-swing shots with a five-iron and a
driver for maximal distance. Pelvis, upper torso, and hand movements were recorded
at 240 Hz. The magnitude of the resultant angular velocity vector of each segment
was computed.
Wearable Sensors
Another technological advance which has allowed quantification of the biome-

chanics of the golf swing on the golf course are wearable inertial sensors that
contain gyroscopes and triaxial accelerometers. Estimation of the center of mass
trajectory is possible using these wearable sensors during golf swings (Najafi et al.
2015). The accuracy of the system to estimate 3D body segment angles compared
to a 3D camera-based motion analyzer was conducted for three golfers and showed
relatively high agreement between wearable sensors and the 3D reference for
angles (r > 0.99, random error <1.2 = 1.5%) for anterior-posterior;
<0.9 = 2% for medial-lateral; and <3.6 = 2.5% for internal-external rotation
direction). The accuracy of three simplified center of mass trajectory models,
conducted for four golfers, showed the two-link model (r > 0.93) had better
agreement with the reference system compared to the one-link model (r = 0.52).
The two-link model estimated center of mass trajectory during multiple golf
swings, conducted for 18 golfers, showed relatively good accuracy (r > 0.9,
anterior-posterior random error <1 cm = 7.7%) and <2 cm = 10.4%) for
medial-lateral).
Wearable sensors may allow the quantification of 3D motion characteristics of a
variety of body segments of golfers and the clubhead in the field.
Force Plate Kinetics
Using common biomechanical concepts, kinetic determinants of the golf swing

include the kinetic-link principle, angular impulse-momentum relationship, summa-
tion of forces (including ground reaction forces), and stretch-shorten cycle (Hume
et al. 2005). When a golfer wants to maximize the distance of their drives and long
irons, relatively large ground reaction forces and joint torques would appear to be
required (Worsfold et al. 2008). During the golf swing, the reaction forces applied at
the feet control translation and rotation of the body-club system. Skilled players
(n = 12) using a six-iron regulated shot distance by scaling the magnitude of the
resultant horizontal reaction force applied to each foot with minimal modifications in
force direction (McNitt-Gray et al. 2013). During each swing, reaction forces were
measured using dual force plates (1200 Hz) and 3D kinematics were simultaneously
captured (110 Hz).
Within-swing changes in ground reaction forces, often referred to as weight
transfer, may help in optimizing golf ball displacement (Ball and Best 2007a, b;
Jenkins 2008). It has been advocated that during the backswing a greater proportion
of the ground reaction forces be on the back foot, with transfer of the ground reaction
forces onto the front foot during the downswing/acceleration phase. Elite golfers
used weight transfer toward the target throughout the swing to provide extra energy
to the ball (Gryc et al. 2015). Kinematics were collected using the 3D CODA Motion
System, and weight transfer data were collected using two force plates.
Electromyography
Electromyography (EMG) studies support the view that skilled golf performance
involves sequential muscle activation (McHardy and Pollard 2005), with different
muscles becoming more active in varying phases (e.g., downswing) (Glazebrook
et al. 1994). Although the level of muscle activation depends on the swing phase,
relatively high levels of hip and knee (Bechler et al. 1995) and trunk (Watkins et al.
1996) muscle activity are observed, with less activation of the muscles around the
shoulder girdle/joints (Kao et al. 1995).
Electromyography may also be used in conjunction of other data collection
approaches to estimate internal forces during the golf swing. Four super VHS
camcorders and two force plates were used to obtain 3D kinematics and kinetics
of golf swings performed by five male collegiate golfers. Average EMG levels for
different phases of golf swings were determined for activity of right and left rectus
abdominis, external and internal obliques, erector spinae, and latissimus dorsi
muscles. An EMG-assisted optimization model was applied to compute the contact
forces acting on the L4–L5. Mean peak compressive load at L4–L5 was over six
times body weight during the downswing. Mean peak anterior and medial shear
loads approached 1.6 and 0.6 BW during the follow-through phases. Average EMG
levels of different muscles are highest in the acceleration and follow-through phases,
suggesting a likely link between cocontractions of paraspinal muscles and lumbar

spinal loads (Lim et al. 2012).
Myoelectric activity of the lumbar erector spinae and the external obliques was
recorded via surface EMG, while 12 male golfers who experienced low back pain
while playing or practicing golf, and 18 asymptomatic golfers, performed 20 drives
(Cole and Grimshaw 2008). Low-handicap (0–12 hcp) low back pain golfers tended
to demonstrate reduced lumbar erector spinae activity at the top of the backswing
and at impact and greater external obliques activity throughout the swing than
golfers without low back pain. The reduced erector spinae activity may be associated
with a reduced capacity to protect the spine and its surrounding structures at the top
of the backswing and at impact, where the torsional loads are high.
There are differences in muscle activity patterns between golfers of various
standards such as more pronator teres muscle activity during the golf swing for
professional and low handicap (<4 hcp) than amateur (10–20 hcp) golfers (Farber
et al. 2009). More consistent shoulder and wrist muscle activity patterns and
temporal patterning of the different phases of chip shots have been reported for
expert than novice golfers performing pitching wedge, nine-iron and seven-iron
shots to targets located 20, 40 and 60 m away (Abernethy et al. 1990).
Modelling/Simulation
Simulation studies utilzing 2D three segment models with joint torque generators
(Sprigings and Mackenzie 2002) or 3D full body models (Nesbit 2005a) have
demonstrated the important role of the wrist in the golf swing. These studies have
shown that reducing the radius path of the hands during the downswing (Nesbit
2005a), increasing wrist torque (Nesbit 2005a; Sprigings and Mackenzie 2002),
increasing wrist range of motion (Nesbit 2005b), and delaying relative wrist motion
to late in the downswing (Nesbit 2005a) all increase clubhead velocity. Simulation
studies have provided an insight into the role of the kinetic link principle in the golf
swing (Nesbit 2005a, b; Sprigings and Neal 2000).
A 2D three-segment model comprising torso, left arm, and golf club was used to
model the downward phase of the golf swing. Torque generators that adhered to the
activation and force-velocity properties of muscle were inserted at the proximal end
of each segment. Separate simulations were performed, with the wrist joint generator
enabled then disabled. Delaying the uncocking of the wrists to when the lead arm
was about 30 below the horizontal took advantage of the kinetic link principle
resulting in a greater summation of angular velocities, and more linear clubhead
velocity at ball impact and ball displacement (Sprigings and Neal 2000).
Work and power (energy) analyses of golf swings can provide a more compre-
hensive description of the downswing (Nesbit 2005b). Two computer models were
used to estimate the energy production, transfers, and conversions within the body
and the golf club by employing standard methods of mechanics to calculate work of
forces and torques, kinetic energies, strain energies, and power during the golf
swing. The model of the golf club determined the energy transfers and conversions
within the club during the downswing. The full-body computer model of the golfer
determined the internal work produced at the body joints during the downswing.
Four diverse amateur subjects were analyzed and compared using these two models.
The energy approach yielded new information on swing mechanics, determined the
force and torque components that accelerated the club, illustrated which segments of
the body produced work, determined the timing of internal work generation, mea-
sured swing efficiencies, calculated shaft energy storage and release, and proved that
forces and range of motion were equally important in developing clubhead velocity.
An additional 3D kinematic and kinetic study of the golf swing (Nesbit 2005a)
performed by 84 male and one female amateur subjects of various skill levels was
reported. The analysis was performed using a variable full-body computer model
of a human coupled with a flexible model of a golf club. Data to drive the model
were obtained from subject swings recorded using a multicamera motion analysis
system. Model output included club trajectories, golfer/club interaction forces and
torques, work and power, and club deflections. The study further highlighted the
importance of the wrists in generating clubhead velocity and orienting the club face
as it approaches ball impact. The trajectory of the hands and the ability to do work
were the factors most closely related to skill level.
Hand path and club angular trajectories were manipulated to increase clubhead
velocity at impact for a further four golfers with diverse swing capabilities and styles
(Nesbit and McGinnis 2014). A 2D dynamic model provided kinetic constrained
optimization of the golf swing hub path. The manner of the modifications, the degree
of velocity improvement, the amount of kinetic reduction, and the associated kinetic
limitation quantities were subject-dependent. By artificially minimizing selected
kinetic inputs within the optimization algorithm, it was possible to identify swing
trajectory characteristics that indicated relative kinetic weaknesses of a golfer.
Robotics and Instrumented Golf Clubs
Putting proficiency is also a key determinant of golfing success. In a study of

10 single figure handicap golfers (Delphinus and Sayers 2012), proficient (>79%)
putters tended to move predominantly in the frontal plane (towards the target), while
the non-proficient (<79%) putters moved more sagittally. Golfers completed
18 putts (2 m) toward a cup in an indoor facility, with their linear movements of
the center of mass and head, and the angular movements of the pelvis and trunk
recorded using 3D analysis at 100 Hz using a six-camera motion capture system.
Robots have been used to help learn golf putting (Reinkensmeyer and Patton
2009). TopSwing is a parallel robot developed by Sofea Software GmbH in
Germany that provides trajectory training for golf swings (Kummel et al. 2014).
Haptic guidance successfully changed the golf backswing technique for a group of
healthy golf novice participants. This improvement in golf swing technique lasted
for at least 7 days after the training.
The TOMI ® putting robot (Mackenzie and Evans 2010) was developed for
measuring putting stroke kinematics (face angle, stroke path, putter speed, and
impact spot at the moment of ball contact) has been found to be reliable and valid
when assessed against a 3D high-speed camera system. All putts were executed from
a distance of 4 m on a straight and flat synthetic putting surface.
The InPutter engineered putter is equipped with an inertial measurement unit, force
sensitive resistors, and heartbeat radio-frequency receiver compatible with Polar
electrocardiogram transmitters. With a sample frequency of 100 Hz, the device can
be combined easily with other alternatives, such as cameras, to increase the range of
applications and variables for further analysis. The InPutter has been validated exper-
imentally using data collected from expert golfers (Couceiro et al. 2015).
Sampling at a higher frequency of 1000 Hz, the SmartSwing instrumented driver
golf club (model LS300) (www.smartswinggolf.com/site/tic/science.html) was used
in a project to develop an augmented video coaching system using qualitative
analysis and post-production open source software. The machine learning diagnostic
outputs were similar (89.5 2.6%) to those produced by the instrumented golf club
(Bačić and Hume 2012).
Summary
We have highlighted findings from the literature regarding the evidence for how
number of potential technical, muscular strength/power, range of motion, and
anthropometric characteristics may be predictors of clubhead velocity. A summary
of the kinematic and kinetic biomechanical analyzes of the golf swing was provided;
with the majority of the research examining the 3D motion of the golf swing in
laboratory environments. While these laboratory-based biomechanical studies of the
golf swing using 3D videography, force, electromyography, modelling, and robots
have provided key information regarding the golf swing, the ecological validity of
the studies remains somewhat questionable. With recent advances in technology
such as isoinertial devices and wearable sensors, much of this knowledge can now be
obtained on the golf course. Sports biomechanists and coaches are now in a position
to more readily access such technology, thereby allowing more valid and relatively
real-time feedback on golf swing performance.
Cross-References

▶ 3D Dynamic Pose Estimation Using Inertial Measurement Units
Sensors
▶ 3D Dynamic Pose Estimation Using Surface Parameterization (MoSh)
▶ 3D Dynamic Pose Estimation Using Wearable Robotics
▶ 3D Dynamic Probablistic Pose Estimation Using Cameras and Reflective
Markers.
▶ 3D Muscle, Ligament and Tendon Motion Using Ultrasound

▶ A Physics-Based Statistical Model for Human Gait Analysis
▶ Frequency Analysis/Wavelet Analysis
▶ Induced Acceleration and Induced Power Analysis
▶ Simulation of Soft Tissue Loading that Generates Observed Movement Dynamics
▶ Surface EMG
▶ The Movement of Scull and Oar Rowing
▶ Time Series Analysis/Time Series Processing of Signals
References
Abernethy B, Neal RJ, Moran MJ, Parker AW (1990) Expert-novice differences in muscle activity
during the golf swing. In: Cochran AJ (ed) (Chair), E & F N Spon. Symposium conducted at the
meeting of the science and golf I: proceedings of the first world scientific congress of golf,
London
Alderslade V, Crous LC, Louw QA (2015) Correlation between passive and dynamic range
of rotation in lead and trail hips during a golf swing. S Afr J Res Sport Phys Edu Recreat
37(3):15–28
Alexander DL, Kern W (2005) Drive for show and putt for dough?: an analysis of the earnings of
PGA tour golfers. J Sport Econ 6(1):46–60
Bačić B, Hume PA (2012) Augmented video coaching, qualitative analysis and post-production
using open source software. In: Bradshaw E, Burnett A, Hume PA (eds) (Chair), Australian
Catholic University. Symposium conducted at the meeting of the 30th conference of the
international society of biomechanics in sports, Melbourne
Ball KA, Best RJ (2007a) Different centre of pressure patterns within the golf stroke I: cluster
Ball KA, Best RJ (2007b) Different centre of pressure patterns within the golf stroke II: group-based
Bechler JR, Jobe FW, Pink M, Perry J, Ruwe PA (1995) Electromyographic analysis of the hip and
knee during the golf swing. Clin J Sport Med 5:162–166
Bertram C, Guadagnoli MA (2008) The effects of custom-fitted clubs versus “placebo” clubs on
golf-swing characteristics. Int J Sports Sci Coach 3:93–98
Betzler NF, Monk SA, Wallace ES, Otto SR (2012) Effects of golf shaft stiffness on strain, clubhead
presentation and wrist kinematics. Sports Biomech 11(2):223–238
Bradshaw EJ, Keogh JWL, Hume PA, Maulder P, Nortje J, Marnewick M (2009) The effect of
biological movement variability on the performance of the golf swing in high and low
handicapped players. Res Q Exerc Sport 80(2):185–196
Brown SJ, Nevill AM, Monk SA, Otto SR, Selbie WS, Wallace ES (2011) Determination of the
swing technique characteristics and performance outcome relationship in golf driving for low
handicap female golfers. J Sports Sci 29(14):1483–1491
Brown SJ, Selbie WS, Wallace ES (2013) The X-factor: an evaluation of common methods used to
analyse major inter-segment kinematics during the golf swing. J Sports Sci 31(11):1156–1163
Cole MH, Grimshaw PN (2008) Electromyography of the trunk and abdominal muscles in golfers
with and without low back pain. J Sci Med Sport 11(2):174–181
Cole MH, Grimshaw PN (2009) The X-factor and its relationship to golfing performance. J Quant
Anal Sports 5(1):9
Cole MH, Grimshaw PN (2016) The biomechanics of the modern golf swing: implications for lower
back injuries. Sports Med 46(3):339–351
Couceiro MS, Dias G, Araújo A, Pereira S (2015) Inputter: concept and evaluation of an engineered
golf putter. S Afr J Res Sport, Phys Educ Recreat 37(3):43–54
Delay D, Nougier V, Orliaguet JP, Coello Y (1997) Movement control in golf putting. Hum Mov Sci
16(5):597–619
Delphinus EM, Sayers MGL (2012) Putting proficiency: contributions of the pelvis and trunk.
Sports Biomech 11(2):212–222
Egret CI, Vincent O, Weber J, Dujardin FH, Chollet D (2003) Analysis of 3D kinematics concerning
three different clubs in golf swing. Int J Sports Med 24(6):465–469
Egret CI, Nicolle B, Dujardin FH, Weber J, Chollet D (2006) Kinematic analysis of the golf swing
in men and women experienced golfers. Int J Sports Med 27(6):463–467
Evans K, Horan SA, Neal RJ, Barrett RS, Mills PM (2012) Repeatability of three-dimensional
thorax and pelvis kinematics in the golf swing measured using a field-based motion capture
system. Sports Biomech 11(2):262–272
Farber AJ, Smith JS, Kvitne RS, Mohr KJ, Shin SS (2009) Electromyographic analysis of forearm
muscles in professional and amateur golfers. Am J Sports Med 37(2):396
Fletcher IM, Hartwell M (2004) Effect of an 8-week combined weights and plyometrics training
program on golf drive performance. J Strength Cond Res 18(1):59–62
Fradkin AJ, Sherman CA, Finch CF (2004) How well does club head speed correlate with golf
handicaps? J Sci Med Sport 7(4):465–472
Glazebrook MA, Curwin S, Islam MN, Kozey J, Stanish WD (1994) An electromyographic
analysis and an investigation of intervention strategies. Am J Sports Med 22(8):674–679
Gryc T, Zahalka F, Maly T, Mala L, Hrasky P (2015) Movement’s analysis and weight transfer
during the golf swing. J Phys Educ Sport 15(4):781–787
Healy A, Moran KA, Dickson J, Hurley C, Smeaton AF, O’Connor NE, . . . Chockalingam N
(2011) Analysis of the 5 iron golf swing when hitting for maximum distance. J Sport Sci
29(10):1079–1088
Hellström J (2009) Competitive elite golf: a review of the relationships between playing results,
technique and physique. Sports Med 39:723–741
Horan SA, Kavanagh JJ (2012) The control of upper body segment speed and velocity during the
golf swing. Sports Biomech 11(2):165–174
Horan SA, Evans K, Morris NR, Kavanagh JJ (2010) Thorax and pelvis kinematics during the
downswing of male and female skilled golfers. J Biomech 43(8):1456–1462
Horan SA, Evans K, Kavanagh JJ (2011) Movement variability in the golf swing of male and
female skilled golfers. Med Sci Sports Exerc 43(8):1474–1483
Horan SA, Evans K, Morris NR, Kavanagh JJ (2014) Swing kinematics of male and female skilled
golfers following prolonged putting practice. J Sports Sci 32(9):810–816
Hume PA, Keogh JWL, Reid D (2005) The role of biomechanics in maximising distance and
accuracy of golf shots. Sports Med 35(5):429–449
Hume PA, Reid D, Edwards T (2006) Epicondylar injury in sport: epidemiology, type, mechanisms,
assessment, management and prevention. Sports Med 36(2):151–170
Jenkins S (2008) Weight transfer, golf swing theory and coaching. Int J Sports Sci Coach
3(0):29–51
Joyce C, Burnett A, Ball K (2010) Methodological considerations for the 3D measurement of the
X-factor and lower trunk movement in golf. Sports Biomech 9(3):206–221
Joyce C, Burnett A, Cochrane J, Ball K (2013) Three-dimensional trunk kinematics in golf:
between-club differences and relationships to clubhead speed. Sports Biomech 12(2):108–120
Kao J, Pink M, Jobe FW, Perry J (1995) Electromyographic analysis of the scapular muscles during
the golf swing. Am J Sports Med 23(1):19–23
Kawashima K, Kato K, Miyazaki M (2003) Body size and somatotype characteristics of male
golfers in Japan. J Sports Med Phys Fitness 43(3):334–341
Kenny IC, Wallace ES, Otto SR (2008) Influence of shaft length on golf driving performance.
Keogh JWL, Hume PA (2012) Evidence for biomechanics and motor learning research improving
golf performance. Sports Biomech 11(2):288–309
Keogh JWL, Nortje J, Maulder P, Marnewick M, Hume PA, Bradshaw E (2007) Kinematic
predictors of golf swing velocity. In: Symposium conducted at the meeting of the 6th Biennial
Australia and New Zealand Society of Biomechanics Australasian Biomechanics conference,
University of Auckland, Auckland
Keogh JWL, Marnewick MC, Maulder PS, Nortje JP, Hume PA, Bradshaw EJ (2009) Are
anthropometric, flexibility, muscular strength, and endurance variables related to clubhead
velocity in low- and high-handicap golfers? J Strength Cond Res 23:1841–1850
Kummel J, Kramer A, Gruber G (2014) Robotic guidance induces longlasting changes in the
movement pattern of a novel sport-specific motor task. Hum Mov Sci 38:23–33
Kwon Y (2010) Kinematic analysis of sports movements: golf swing plane analysis. Presented at
the meeting of the XXVIII international symposium of biomechanics in sports, Marquette
Kwon Y, Han K, Como C, Lee S, Singhal K (2013) Validity of the X-factor computation methods
and relationship between the X-factor parameters and clubhead velocity in skilled golfers.
Lim Y-T, Chow JW, Chae W-S (2012) Lumbar spinal loads and muscle activity during a golf swing.
Lynn SK, Frazier BS, New KN, Wu WFW, Cheetham PJ, Noffal GJ (2013) Rotational kinematics of
the pelvis during the golf swing: skill level differences and relationship to club and ball impact
conditions. Int J Golf Sci 2(2):116–125
Mackenzie SJ, Evans DB (2010) Validity and reliability of a new method for measuring putting
stroke kinematics using the TOMI® system. J Sports Sci 28(8):891–899
Marshall RN, McNair PJ (2013) Biomechanical risk factors and mechanisms of knee injury in
golfers. Sports Biomech 12(3):221–230
McHardy A, Pollard H (2005) Muscle activity during the golf swing. Br J Sports Med 39:799–804
McLean J (1992) Widen the gap. Golf Mag 34(12):49–53
McNitt-Gray JL, Munaretto J, Zaferiou A, Requejo PS, Flashner H (2013) Regulation of reaction
forces during the golf swing. Sports Biomech 12(2):121–131
Meister DW, Ladd AL, Butler EE, Zhao B, Rogers AP, Ray CJ, Rose J (2011) Rotational
biomechanics of the elite golf swing: benchmarks for amateurs. J Appl Biomech 27(3):242–251
Najafi B, Lee-Eng J, Wrobel JS, Goebel R (2015) Estimation of center of mass trajectory using
wearable sensors during golf swing. J Sports Sci Med 14(2):354–363
Nesbit SM (2005a) A three dimensional kinematic and kinetic study of the golf swing. J Sports Sci
Med 4:499–519
Nesbit SM (2005b) Work and power analysis of the golf swing. J Sports Sci Med 4:520–533
Nesbit SM, McGinnis RS (2014) Kinetic constrained optimization of the golf swing hub path. J
Sports Sci Med 13(4):859–873
Oliver MH, Horan SA, Evans KA, Keogh JWL (2016) The effect of a seven-week exercise program
on golf swing performance and musculoskeletal measures. Int J Sports Sci Coach 11(4):
610–618. doi:10.1177/1747954116654784
Reinkensmeyer DJ, Patton JL (2009) Can robots help the learning of skilled actions? Exerc Sport
Sci Rev 37(1):43–51
Robertson SJ, Burnett AF, Newton RU et al (2012) Development of the nine-ball skills test to
discriminate elite and high-level amateur golfers. J Sports Sci 30:431–437
Smith AC, Roberts JR, Wallace ES, Pui K, Forrester SE (2016) Two- and three-dimensional
methods have been used for analysis of trunk kinematic variables in the golf swing. J Appl
Biomech 32(1):23–31
Sprigings EJ, Mackenzie SJ (2002) Examining the delayed release in the golf swing using computer
simulation. Sports Eng 5(1):23–32
Sprigings EJ, Neal RJ (2000) An insight into the importance of wrist torque in driving the golf ball:
a simulation study. J Appl Biomech 16(4):356–366
Stastny P, Maszczyk A, Tománková K, Kubový P, Richtrová M, Otáhal J, . . . Cięszczyk P (2015)
Kinetic and kinematic differences in a golf swing in one and both lower limb amputees. J Hum
Kinet 48(1):33–41
Sweeney M, Mills P, Alderson J, Elliott B (2013) The influence of club-head kinematics on early
ball flight characteristics in the golf drive. Sports Biomech 12(3):247–258
Tinmark F, Hellström J, Halvorsen K, Thorstensson A (2010) Elite golfers’ kinematic sequence in
full-swing and partial-swing shots. Sports Biomech 9(4):236–244
Torres-Ronda L, Delextrat A, Gonzalez-Badillo JJ (2014) The relationship between physical and
fitness requirements in elite golf performance. Int SportMed 15(2):156–164
Wallace ES, Otto SR, Nevill A (2007) Ball launch conditions for skilled golfers using drivers of
different lengths in an indoor testing facility. J Sports Sci 25(7):731–737
Watkins RG, Uppal GS, Perry J, Pink M, Dinsay JM (1996) Dynamic electromyographic analysis of
trunk musculature in professional golfers. Am J Sports Med 24(4):535–538
Wheat JS, Vernon T, Milner CE (2007) The measurement of upper body alignment during the golf
drive. J Sports Sci 25(7):749–755
Williams KR, Sih BL (2002) Changes in golf clubface orientation following impact with the ball.
Sports Eng 5(2):65–80
Worobets J, Stefanyshyn D (2012) The influence of golf club shaft stiffness on clubhead kinematics
at ball impact. Sports Biomech 11(2):239–248
Worsfold P, Smith NA, Dyson RJ (2008) Low handicap golfers generate more torque at the shoe-
natural grass interface when using a driver. J Sports Sci Med 7(3):408–414
Zheng N, Barrentine SW, Fleisig GS, Andrews JR (2008) Swing kinematics for male and female pro
golfers. Int J Sports Med 29:965–970
Three-Dimensional Reconstruction
of the Human Skeleton in Motion
Valentina Camomilla, Aurelio Cappozzo, and Giuseppe Vannozzi
Abstract
This chapter illustrates the conceptual background underlying the in silico recon-
struction of the human skeletal motion. A specific focus is given to the experi-
mental and analytical methods that allow acquiring information related to both
bone movement and morphology in vivo in the framework of rigid body mechan-
ics. This process involves the definition of global and local frames of reference.
Common anatomical and mathematical conventions that are used to describe
global bone pose and joint kinematics are illustrated. Issues concerning accuracy
and reliability of the estimated quantities when using skin markers and stereo-
photogrammetry and magneto-inertial measurement units are also dealt with.
Keywords
Rigid body mechanics • Human movement analysis • Bone pose estimation •
Anatomical calibration • Joint kinematics
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Global and Local Frames: Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Global Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Local Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Estimate of the Bone-Embedded Frame Pose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Stereophotogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Magneto-Inertial Measurement Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Anatomical Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
V. Camomilla (*) • A. Cappozzo • G. Vannozzi

Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department
of Movement, Human and Health Sciences, University of Rome Foro Italico, Rome, Italy
e-mail: valentina.camomilla@uniroma4.it; aurelio.cappozzo@gmail.com; giuseppe.
vannozzi@uniroma4.it

DOI 10.1007/978-3-319-30808-1_146-1
2 V. Camomilla et al.
Subject-Specific Morphology Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Movement-Morphology Data Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Construction of the Anatomical Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Joint Kinematics Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Translational Degrees of Freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Rotational Degrees of Freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Precision and Accuracy of Joint Kinematics Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Introduction
Protecting and enhancing human motor function is an important strategic aim of

modern society dwelling within the grand challenges concerning health and well-
being (personalized care, prevention and early diagnosis, integrative and holistic
approaches, healthy aging, sustainability of chronic diseases). Efforts must be made
to strengthen our multi-scale understanding of both structure and function of the
human locomotor system and to develop investigative and operational methods that
can be translated into professional practice. In this context, quantitative human
movement analysis is of paramount importance, and, for many applications, it
must be carried out in the three-dimensional space. This analysis is based on
measurements (motion capture, dynamometry, electromyography, calorimetry, med-
ical imaging, etc.) and computational modeling (mathematical models of the anat-
omy and physiology of the tissues, organs, and systems involved). It provides
information on the functions of the locomotor subsystems and on the overall strategy
of motor activity. These outcomes contribute to the understanding of the key factors
that affect joint motion and internal loading and, thus, injury, tissue degeneration or
regeneration, as well as motor control and its adaptation, energy consumption, and
fatigue. Quantification of subject-specific variables can be effectively used in pre-
vention, early diagnosis (e.g., monitoring of functional status in the elderly, specific
workers, or athletes), intervention (e.g., prognosis, therapeutic programming, work-
place optimization, training), and quantifying relevant outcomes. Prospectively,
advanced movement analysis technology may be used for the implementation of
real-time biofeedback (virtual and augmented reality) both in rehabilitation and
training (institutionalized or not).
State of the Art
A prerequisite for many of the above-mentioned endeavors is the accurate three-

dimensional reconstruction in silico of the portion of the locomotor system of
interest while performing a motor task. The solution of this problem is based on
the following fundamental considerations.
A body may be thought to be made of P particles. The larger the number P, the
more details of the body shape are portrayed. The motion of this body may be
Three-Dimensional Reconstruction of the Human Skeleton in Motion 3
Fig. 1 Global (gx, gy, gz) and

local (lx, ly, lz) frames. P
represents a particle of the
body of interest and gp and lp
its position vectors in the two
frames, respectively. Left
superscripts denote the frame
with respect to which a
position vector or an
orientation of a frame is
represented. Subscripts denote
the entity the vector or the
matrix describes
described by providing information about the motion of each of those particles in the
form, for instance, of three Cartesian coordinates (or a position vector), relative to a
given set of orthogonal axes (global frame), and their time derivatives, in each
sampled instant of time (1. . .N ). The resulting set of numbers describing position
is made of 3 P N numbers. This data set may be split into two parts: one
describing movement and the other describing shape. This is a profitable exercise
because, if we are not interested in shape variation, i.e., in deformation, the second
data subset may be considered to be time invariant, and this results in a remarkable
reduction of the data set dimension with the obvious experimental and computa-
tional advantage. This is achieved through the following procedure.
A set of orthogonal axes, generally named local frame (l), is defined rigidly
associated with the body while assuming its current shape (Fig. 1). The location in
space of this frame relative to the global frame (g), in a given instant of time, is
described using two separate bits of information: one deals with its position and the
other with its orientation (also named attitude). The former information is provided
by the position vector of the origin of the local frame (Fig. 1):
g
g
o¼ ox g oy g oz (1)
The orientation of the local frame may be described by providing the orientation
of each of its axes. This is done through the components of the three unit vectors
aligned with these axes, relative to the global frame axes. Through obvious geomet-
rical considerations, it can be seen that these components are equal to the cosines
(named direction cosines) of the angles formed by each unit vector with the global
frame axes:
h i
g
uxl ¼ cos θxl xg cos θxl yg cos θxl zg (2)
h i
g
uyl ¼ cos θyl xg cos θyl yg cos θyl zg (3)
h i
g
uzl ¼ cos θzl xg cos θzl yg cos θzl zg (4)
Equations 2, 3, and 4 are normally represented in a matrix named orientation

matrix (also rotation or attitude matrix):
2 3
cos θxl xg cos θyl xg cos θzl xg
g
Rl ¼ 4 cos θxl yg cos θyl yg cos θzl yg 5 (5)
cos θxl zg cos θyl zg cos θzl zg
It is useful to mention here that given the position vector and the orientation
matrix of the local frame, relative to a global frame, and the position of a point (or of
an ensemble of points, that is, a body) in the local frame (Fig. 1), it is possible to
represent this position in the global frame through the following operator (coordinate
transformation):
g
p ¼ g Rl l p þ g ol (6)
Of course the use of this operator may be generalized, since it allows representing
the position of a point given in a reference frame to any other referred frame.
It is important to note that the nine direction cosines are not independent; in fact
the sum of the squares of each triplet equals one (the amplitude of the unit vector),
and the vector product of each pair of unit vectors, being orthogonal, equals zero.
The resulting six relationships leave only three of the nine direction cosines inde-
pendent. In summary, the orientation of the local frame relative to the global frame is
fully described by three scalars, which, added to the three scalars necessary to
describe position (Eq. 1), makes a total of six scalars (the degrees of freedom of
the body). The ensemble of position and orientation of a local frame and, thus, of a
body, is named pose.
Another and more compact way of representing the orientation of a body, i.e., of a
local frame, is to exploit the Euler’s theorem and use the kinematic variables that
describe the fictitious rotation that takes the local frame from being parallel to the
global frame to the target orientation. We can achieve this by decomposing this
rotation into three finite successive rotations about body-fixed axes in their current
orientation. The corresponding three rotation angles, called Euler angles, completely
describe the given rotation. There are twelve possible sets of Euler angles. Six imply
the first and third rotation about the same axis (symmetric sets), and six imply
rotations about the three different axes, in all possible combinations (asymmetric
sets). The latter angle sets are also referred to as Cardan or Bryant angles. A second
approach explicitly identifies the axis of rotation n (unit vector) and the angle of
rotation θ that realize the target rotation and uses the following orientation
(or rotation) vector:
θ ¼ θn (7)
Both the orientation vector and the Euler angles are illustrated in standard
kinematics and dynamics texts and are not discussed here (Shuster 1993). However,
a few remarks will be made in a subsequent section with regard to the use of these
quantities to describe joint kinematics.
After the above-illustrated representation of the instantaneous pose of a body, it is
seen that, if the body can be hypothesized to keep its shape unaltered during the
analyzed movement, the number of scalars necessary to describe it is reduced from
3 P N to 6 N + 3 P. Under these very favorable circumstances, we talk of a
rigid body and use the part of classical mechanics that operates under the mentioned
hypothesis.
The question arises now whether the rigid body hypothesis is applicable to the
constituents of the locomotor apparatus in motion. In this respect we have two orders
of problems: one regards the mathematical modeling of the locomotor apparatus and
the other the acquisition of experimental data. It is commonly accepted that if the
focus of the investigation is skeletal motion, then bones may be considered to be
rigid bodies without having a significant impact on the end results. When aiming at
the estimation of internal loads or mechanical energy, then entire body segments are
involved in the modeling exercise and may be disputable to consider them rigid
bodies due to the fact that deformation and displacements occurring during move-
ment of muscular and visceral masses may significantly change some inertia param-
eters with consequent inertial effects (mass moments of inertia and location of the
center of mass; Clark and Hawkins 2010; Pain and Challis 2001; Zelik and Kuo
2010) (“▶ Estimation of the Body Segment Inertial Parameters for the Rigid Body
Biomechanical Models Used in Motion Analysis”). It is evident that this issue
becomes critical when high accelerations are involved (Challis and Pain 2008;
Gruber et al. 1998; Liu and Nigg 2000; Riddick and Kuo 2016). An experimental
issue consists on the fact that bone movement can be accurately recorded only using
invasive techniques (intracortical pins or medical imaging involving the use of
ionizing radiations). As better illustrated later, under normal circumstances, we can
only capture the movement of 3-D markers (through stereophotogrammetry or
magneto-inertial measurement units) attached to the skin surface above the bony
segment of interest. When using stereophotogrammetry, the 3-D marker is
constructed using three or more point markers (marker cluster). A 3-D marker is
unavoidably mobile with respect to the underlying bone due to the interposed soft
tissue, and, thus, its global motion defers from that of the bone. In general, this local
displacement of the marker is caused by skin sliding associated joint movement, soft
tissue volumetric deformation due to muscular contraction, gravity, and inertial
effects on relevant masses (wobbling). In this context, the relative motion between
marker and underlying bone is to be regarded as an artifact, the so-named soft tissue
artifact, which, if not properly dealt with, has very serious consequences on the
reliability of the results of the analysis (Cappozzo 1991; Garling et al. 2007;
Lamberto et al. 2016; Leardini et al. 2005; Li et al. 2012). If the objective is a
kinetic or energy analysis, then quantitative information concerning the
abovementioned soft tissue relative motion, which is concerning the instantaneous
body segment shape and mass distribution during movement and the estimate of the
consequent time-varying inertia parameters, may be required posing an experimental

problem very difficult to be dealt with using the present state of the art of measure-
ment technology (Wakeling and Nigg 2001) (“▶ Simulation of Soft Tissue Loading
From Observed Movement Dynamics”).
In the general framework illustrated above and accepting the related limitations,
in human and animal movement analysis, rigid body mechanics is normally used to
model the locomotor apparatus. The objective of this chapter is to illustrate the way
this is done.
From now on we will refer to the estimate of bone pose being confident of the fact
that, to all practical purposes, bones may be considered rigid bodies and that models
of the relevant soft tissues may be associated to the bone in its current pose. Note that
most analyses demand not only the estimate of the instantaneous pose of the bones
involved but also the full reconstruction of their outer surface in the 3-D in silico
space.
It is important to acknowledge the fact that the quantitative description of human
motion involves the use of kinematic and kinetic vector components, the values of
which depend on the Cartesian reference axes with respect to which they are defined.
Thus, for the sake of their repeatability, equally repeatable set of axes must be made
available. As will be clearer later, it is easy to accomplish this requirement for the
global frames, while, by cause of technical difficulties and morphological complex-
ity, it is more difficult for the local frames associated with the bones (Della Croce
et al. 1999, 2005). Furthermore, these sets of axes must be defined so that the
resulting kinematic and kinetic scalar quantities have a clear functional meaning
and effectively and consistently describe the anatomical and functional entities used
in the medical-biological literature.
Global and Local Frames: Terminology
As reported above, describing the skeletal-system movement involves the use of sets
of coordinate system axes that define the global and the local frames. The definition
and the methods for identifying these coordinate systems vary with the objective of
the analysis and the measuring instrument involved.
Global Frames
In a movement analysis laboratory, the following inertial, global frames can be

defined (Fig. 2; Cappozzo et al. 1995, 1997b).
Motion capture system frame: it is composed of the set of axes used by either the
stereophotogrammetric system or the magneto-inertial system to represent point-
marker position or 3-D marker orientation, respectively. This is arbitrarily defined
during the motion capture system calibration procedure.
Motor task frame: this frame is consistent with the analyzed motor task and
sometimes describes its basic features. For instance, when locomotor acts are
Fig. 2 Movement analysis laboratory equipped with a stereophotogrammetric system and two
force plates. Global frames are depicted. If a locomotor act is analyzed, the motor task frame may
coincide with the frame of one of the two force plates
investigated, one axis of the frame indicates the mean direction of progression.
According to the general recommendations from the International Society of Bio-
mechanics (Wu and Cavanagh 1995), in human locomotion analysis, right-handed
orthogonal coordinate systems should have the x-axis pointing in the direction of
progression, y pointing vertically upward, and z pointing to the right.
Dynamometer frame: this is the frame in which force and moment components
are given by the instrument and is defined during its calibration.
Plumb line: this is a single axis and represents the orientation of the gravity line,
usually assumed to point downward.
Therefore, within the same experiment, different mechanical quantities can be
measured with respect to different global frames. However, normally, their interpre-
tation, or their use as input to musculoskeletal models allowing the estimation of
further non-measurable quantities, requires that all of them be represented in the
same frame (primary global frame). The latter role is usually assumed by the motor
task frame. This procedure involves the determination of the position vector and the
orientation matrix of all secondary global frames involved relative to the primary
frame, allowing to obtain any vector quantity in the primary frame using Eq. 6. In
order to achieve this result, the dynamometers and the pathway, staircase, or any
Fig. 3 Local frames
other implement used to perform the motor task must be equipped with a 3-D
marker, the pose of which may be detected by the motion capture system being
used (Rabuffetti et al. 2003).
Local Frames
We deal here with the frames associated with a bony segment (Fig. 3).
Anatomical frame. This frame is made of orthogonal axes that exhibit the
following properties: (i) compatibility with the anatomical axes and planes defined
in the anatomy literature and (ii) intra- and inter-subject repeatability. The concept of
repeatability must, of course, be extended to the portability of the results of move-
ment analysis among different laboratories (Benedetti et al. 2013). This entails the
standardization of the definition of the anatomical frames (Cappozzo et al. 1995; Wu
and Cavanagh 1995; Wu et al. 2002).
The following local frames depend on the specific technique used for the deter-
mination of their pose. For this reason they are named technical frames.
Bone-embedded frame. This is a frame rigidly connected with the bone, but with no
repeatable relationship with its anatomy.
Motion technical frame. This is the frame the pose of which is provided by the
motion capture system and related data processing. It, normally, represents an
estimate of the bone-embedded frame.
Morphology technical frame. This is the frame used by the measurement instrument
(i.e., imaging apparatus, such as MRI or CT) that provides a digital model of the
bone and, eventually, of the relevant soft tissues.
Fig. 4 Input and output data of bone pose estimators using skin markers and stereo-
photogrammetry (a) or magneto-inertial measurement units (b)
Estimate of the Bone-Embedded Frame Pose
Bone movement may be directly tracked using medical imaging techniques. X-ray
fluoroscopy, either planar or biplanar, is mostly used to record a single joint
movement (Banks and Hodge 1996). This technique, however, exhibits the follow-
ing limitations: it uses ionizing radiation, the measurement volume and the sampling
frequency are limited, and the procedure is operator time-consuming. Recently
systems have been developed that move alongside the subject, thus overcoming
the limitation of the measurement volume, at least in one dimension (Guan et al.
2016) (“▶ Measurement of 3D Dynamic Joint Motion Using Biplane Video-
radiography”). Other techniques are under development that exploit magnetic reso-
nance and ultrasonic technologies (“▶ 3D Musculoskeletal Kinematics Using
Dynamic MRI”).
In this chapter we deal only with the motion capture techniques mostly used in
human movement analysis that track bone motion indirectly, but display the advan-
tages of being noninvasive, having large measurement volumes and high sampling
rates: stereophotogrammetry associated with skin markers, or magneto-inertial mea-
surement units (Fig. 4). The fundamental concepts illustrated regarding the use of
rigid body mechanics for the modeling of the human locomotor apparatus in motion
are, however, valid whatever technique is used to monitor this motion.
Stereophotogrammetry
Optoelectronic stereophotogrammetry is, to date, the most widely used solution for
measuring skeletal kinematics. It is made of a number of video cameras, connected
to a computer, the fields of view of which intersect defining the measurement
volume. This system provides the position (Cartesian coordinates) of point markers,
either emitting or retro-reflecting light, located on the skin surface relative to a global
frame. This is done through mathematical operators that receive the 2-D coordinates
of the point-marker images, measured in the image plane of at least two cameras at
any given instant of time, and parameters that describe the location in space and
optical features of the cameras. These parameters are obtained through the calibra-
tion of the stereophotogrammetric system.
The reconstructed positions in the global frame of three or more nonaligned skin
markers (marker cluster or 3-D marker) located above the bone of interest and a
mathematical estimator are used to construct a motion technical frame and determine
its pose relative to the global frame (Fig. 4) (“▶ 3D Dynamic Pose Estimation Using
Reflective Markers” and “▶ 3D Dynamic Probablistic Pose Estimation From Data
Collected Using Cameras and Reflective Markers”). Markers are located in such a
way to comply with technical requirements such as visibility to a sufficient number
of cameras, maximal relative distance between markers (Cappozzo et al. 1997a), and
minimal relative movement between them and the underlying bone. Virtual markers
may also be used to construct a marker cluster. A virtual marker is a point, the
position of which is calculated using the positions of the available physical markers
and a geometric or statistical rule. A typical example of virtual marker is a point
approximating a joint center, which is in common between two adjacent bones. If the
pose of one bony segment is successfully reconstructed, the global position time
history of the virtual marker, the position of which in this segment is known, can be
obtained using Eq. 6 and used to reconstruct the pose of the second body segment as
if it were a physical marker. This procedure allows reducing the number of physical
markers to be tracked. Note that, while this economical attitude could be justified
years ago, this is not the case with current optical motion capture technology that can
deal with a very high number of markers simultaneously.
A marker cluster observed, for instance, in a selected instant of time or subject’s
posture, is taken as a rigid cluster model. Using some geometric rule, a set of
orthogonal axes is associated to it (model frame). If the skin markers were rigidly
attached to the underlying bone, the rigid cluster model could be superimposed onto
the current marker cluster and the model frame taken as motion technical frame.
Unfortunately, this is not the case: during movement, the reconstructed marker local
positions undergo variations due to stereophotogrammetric errors (Chiari et al. 2005)
and, as already mentioned, the soft tissue deformation (Leardini et al. 2005). Thus,
the marker cluster changes its shape and moves relative to the bone (Fig. 5).
As a consequence, the above-mentioned superimposition exercise must be carried
out using an approximation approach (“single-body optimization”). This consists in
a least-squares method, the most popular of which minimizes the sum of the squared
distances between corresponding points of the cluster model and of the marker
Fig. 5 The artifact

displacements of the skin
markers (arrows), occurring
in a given interval of time,
result in a rigid transformation
(a translation plus a rotation)
plus a nonrigid transformation
(a change in size and shape) of
the marker cluster (Andersen
et al. 2012; Barré et al. 2013;
Benoit et al. 2015; De Rosario
et al. 2012; Dumas et al. 2014;
Grimpampi et al. 2014). The
former transformation has
been proved to be the
dominant part (Andersen et al.
2012; Barré et al. 2013;
Benoit et al. 2015; Bonci et al.
2015; Dumas et al. 2015)
cluster, the so-named Procrustes distance (Soderkvist and Wedin 1993). In this way
the instantaneous pose of the motion technical frame can be determined and taken as
an estimate of the bone-embedded frame.
Instead of dealing with one marker cluster, and thus one bone, at a time, it is
possible to carry out the least-squares superimposition procedure illustrated above
using a multi-body system made of a chain of rigid cluster models connected by
kinematic models of the joints involved (“multi-body kinematics optimization”).
The joint models proposed in the literature typically involve major simplifications
with respect to real and subject-specific joints and have less than six independent
degrees of freedom (Charlton et al. 2004; Duprey et al. 2010; Li et al. 2012; Lu and
O’Connor 1999; Reinbolt et al. 2005; Richard et al. 2016; Scheys et al. 2011).
It should be emphasized that while instrumental errors that are random in nature
have no significant effect on the end result, the illustrated procedures (single-body
and multi-body kinematics optimization) do not attenuate the propagation of the
largest portion of the soft tissue artifact (Andersen et al. 2010; Bonci et al. 2015;
Clément et al. 2015, 2017; Gasparutto et al. 2015; Li et al. 2012). The latter issue
remains the greatest obstacle to the accurate reconstruction of skeletal movement and
deserves further attention by the human movement analyst community.
Magneto-Inertial Measurement Units
Miniature magneto-inertial measurement units, embedding a microprocessor and

often endowed with wireless communication technology, are an increasingly popular
alternative to stereophotogrammetry for 3-D human movement analysis (“▶ Three-
dimensional Human Kinematics Estimation Using Magneto-Inertial Measurement
Units”). These units are attached to the body segment of interest. They comprise
three-axis linear accelerometer and angular rate sensors and a three-axis magnetom-
eter. The physical quantities provided by each sensor are measured with respect to
the axes of a unit-embedded frame generally aligned with the edges of the unit case.
Through algorithms able to fuse the redundant information available and compen-
sate for sensor noise and drift, the 3-D orientation of the unit-embedded frame
relative to a global frame is provided. The unit-embedded frame is assumed to be
the motion technical frame and, again, despite the soft tissue artifact, is assumed to
be an estimate of the bone-embedded frame (Fig. 4). As opposed to stereo-
photogrammetry, magneto-inertial measurement units do not supply reliable posi-
tional information. The advantage of this technique is that it does not impose limits
to the measurement volume.
Anatomical Calibration
So far we have seen how a bone-embedded frame, as observed during the execution
of a motor task, may be reconstructed in silico in its current pose. The next step is the
representation of a digital model of the relevant bone in this frame. In order to
comply with the repeatability issues discussed in a previous section, this digital
model must carry an anatomical frame with it. The experimental acquisition of the
parameters that allow the construction in silico of the bone model in its current pose
and of the relevant anatomical axes is referred to as “anatomical calibration.”
Subject-Specific Morphology Data
The 3-D digital model of the bone of interest may be defined at different levels of
resolution that depend on the number of points used to describe the bone external
surface. The minimum number of these points is 3, as imposed by the possibility of
constructing an anatomical frame. A larger number of points allow a more realistic
rendering of the bone in silico and a more accurate association of the soft tissue
digital models to the bone.
A subject-specific bone digital model may be measured or estimated using
medical imaging. A full reconstruction of that model may be obtained using mag-
netic resonance, but this is rarely possible for logistic and economic reasons. A way
of estimating a subject-specific bone digital model, with sufficient accuracy for most
purposes, is through two planar X-rays of the bone and a relevant statistical model
and shape recognition algorithm (Chaibi et al. 2012). This method is made applicable
by a low-dose X-ray imaging technology performed at a low dose and with an
expanded dynamic range that allows for whole-body scanning while the subject is
under normal weight-bearing conditions (Melhem et al. 2016). It is evident that with
all methods that involve medical imaging, the bone digital model is represented in a
morphology technical frame that is different from the motion technical frame. For a
number of reasons that will be illustrated later, the subject-specific bone digital model
must carry labeled anatomical landmarks (sites with recognizable anatomical features).
Given a clear definition of these landmarks, this labeling may be carried out in silico
through a virtual palpation (Van Sint Jan et al. 2003).
An alternative approach is the estimate of the subject-specific bone model using
subject-specific partial information collected in the movement analysis laboratory
and a template bone model that is made to match the above-mentioned morpholog-
ical information. This partial information may consist of the 3-D position of isolated
points or clouds of adjacent points that lie on the bone surface and of lines oriented as
anatomical axes or lie in anatomical planes. In this case, morphological information
is available in the motion technical frames. The position of anatomical landmarks
may be determined with an ad hoc stereophotogrammetric acquisition by either
temporarily locating skin markers over them (Cappozzo 1984) or using a wand
equipped with a marker cluster (Fig. 6a; Cappozzo et al. 1995), after identification by
manual palpation (Van Sint Jan et al. 2003). In some experimental protocols, skin
markers used to track motion are located over anatomical landmarks, the position of
which is therefore readily available, or in an anatomical plane (Fig. 6b; Davis et al.
1991; Frigo et al. 1998; Leardini et al. 2007) (“▶ Conventional Gait Model – Success
and Limitations” and “▶ Variations of Marker-sets and Models for Standard Gait
Analysis”). More detailed morphological information may be acquired by using the
above-mentioned wand over clouds of points of the bone covered with a layer of soft
tissue that allows for their palpation through the skin (Fig. 6c; Donati et al. 2007, 2008).
To these points, internal anatomical landmarks may be added when noninvasively
identifiable. The position of these landmarks may be determined as a function of the
position of other accessible anatomical landmarks and/or readily available anthro-
pometric measures using population statistical models. Typical examples in this
respect are the regression equations used to determine the position of the center of
the acetabulum (Bell et al. 1990; Davis et al. 1991; Hara et al. 2016; Harrington et al.
2007; Leardini et al. 1999; Seidel et al. 1995) or the position of the clavicle, scapula,
and humeral bone (Sholukha et al. 2009). In some cases, an internal anatomical
landmark may be considered to coincide with a joint center of rotation (“▶ Next-
Generation Models Using Optimized Joint Center Location”). This allows determin-
ing the position of these landmarks using a “functional approach.” This entails
performing an ad hoc experiment during which the subject, equipped with suitable
markers (skin-marker clusters or magneto-inertial measurement units), is asked to
execute a 3-D movement of the joint involved. The time history of the pose of one
bone relative to the other forming the joint is estimated and used to determine the
position of the joint center and, thus, of the relevant anatomical landmark (Crabolu
et al. 2016; Halvorsen 2003; McGinnis and Perkins 2013). An example in this
respect is the center of the acetabulum that can be assumed to coincide with the
center of the femoral head and the center of rotation of the femur relative to the
pelvic bone (Fig. 7; Camomilla et al. 2006; Cappozzo 1984; Cereatti et al. 2010;
Kainz et al. 2015; Leardini et al. 1999; Piazza et al. 2004). Similarly, the center of the
head of the humerus may be identified as the rotation center of the glenohumeral
joint (Campbell et al. 2009; Lempereur et al. 2010).
Lines that lie in anatomical planes also carry useful morphological information.
By suitably choosing a joint movement, the resulting rotation axis may be assumed
Fig. 6 Anatomical calibration in the stereophotogrammetric laboratory. (a) Skin markers are
located consistently with technical requirements; anatomical landmarks are identified by manual
palpation and their position measured using a wand. (b) Some skin markers are located on
anatomical landmarks or lying in an anatomical plane. (c) As in (a), but portions of the bone
surfaces are digitized by moving the tip of the wand over them. The center of the acetabulum is
estimated using mathematical models or identified through a functional approach
Fig. 7 Movement normally

used for the estimate of the
center of the acetabulum using
the functional approach. The
figure depicts the trajectory of
a point located on the distal
portion of the femur and
represented in the transverse
plane of the pelvis. Numbers
represent the movement
sequence (Camomilla et al.
2006)
to have this property. Such an axis may be determined through the already mentioned
functional approach (Gamage and Lasenby 2002; Halvorsen et al. 1999). Examples
in this respect are the knee (Colle et al. 2016; De Rosario et al. 2017; Ehrig et al.
2007), the elbow joint (Fraysse and Thewlis 2014), and the talocrural joint (Sheehan
2010; van den Bogert et al. 1994) moving in the sagittal plane. The estimated
Fig. 8 Anatomical calibration using magneto-inertial measurement units. The orientations of the
two dashed lines shown in the figure are detected by the unit mounted on the calibration device and
define the sagittal plane of the femur. The line joining the medial epicondyle (LE) and the greater
trochanter (GT) may be used for the orientation of the y-axis. The orientation of the other two axes
is, thus, available as well. If the center of the femoral head is identified using the functional
approach (Crabolu et al. 2016), a digital model of the femur may be matched with the available
subject-specific morphological information and the definition of the femoral anatomical frame
illustrated in Fig. 9 may be used
rotation axis is assumed to lie in the frontal plane and, together with at least an
anatomical landmark, allows the determination of this plane. It is worth emphasizing
that, as opposed to what in some cases may appear in the literature, an axis or line,
obtained using this method, is not an anatomical axis per se, although sometime it
may be considered to approximate it.
When magneto-inertial measurement units are used, the position of anatomical
landmarks cannot be reliably measured. The only morphological information that
can be collected concerns the orientation of lines that lie in anatomical planes. This
information can be obtained using either a functional approach (Bouvier et al. 2015;
Cutti et al. 2008, 2010; Favre et al. 2009; Luinge et al. 2007; Seel et al. 2012) or a
specifically designed calibration device consisting in a rod carrying a
magneto-inertial measurement unit and two mobile pointers perpendicular to it
(Fig. 8). This unit provides the orientation relative to the motion technical frame,
made available by the unit mounted on the body segment of interest, of a line joining
two palpable anatomical landmarks pointed by the calibration device (Picerno et al.
2008).
Movement-Morphology Data Registration
After the subject-specific morphological information is collected, the entire subject-

specific digital model of the bone must be associated to the motion technical frame in
each sampled instant of time during the analyzed movement and, therefore,
represented in silico in its current pose relative to the global frame.
When the entire subject-specific digital model is made available, either measured
or estimated through medical imaging, and represented in a morphology technical
frame different from the motion technical frame, a transformation of the position
vectors given in the former frame into position vectors in the latter frame must be
carried out (the procedure is known as movement-morphology data registration). To
this purpose Eq. 6 can be used, provided that the anatomical calibration procedure
includes the determination of the orientation matrix and origin position vector of one
frame with respect to the other. This may be achieved by having the position of at
least three points in both frames and applying a superimposition procedure similar to
that described in section “Estimate of the Bone-Embedded Frame Pose.” These
points may be anatomical landmarks or, when using stereophotogrammetry, the
skin markers. In the latter circumstance, the medical imaging procedure must be
carried out with the subject carrying the marker set that will be used for the motion
tracking. While doing this with the subject lying on a horizontal surface, as normally
occurs with magnetic resonance, we may incur in a problem. His/her soft tissues are
deformed in a different fashion then when assuming the static or dynamic postures
under analysis. The thus obtained pose of the motion technical frame relative to the
morphological technical frame is, therefore, different from that occurring during the
motor task of interest.
If partial subject-specific morphology information is collected in the motion
capture laboratory, then the subject-specific bone model in its current pose may be
obtained by matching a suitably chosen template bone model, provided in a mor-
phology technical frame, with the available morphological data available in the
motion technical frame. To this purpose a superimposition procedure associated
with non-isomorphic scaling and a reorientation of the template model may be used
(Chaibi et al. 2012; Donati et al. 2007, 2008; Quijano et al. 2013). In this case, the
bone model is already represented in the motion technical frame, and, therefore, no
registration procedure is required.
The bone model thus obtained normally carries a number of labeled anatomical
landmarks or lines, those that have been used for its construction and registration. If
the position of all anatomical landmarks required for further processing is not
provided by this procedure, then a virtual palpation or geometric construction is
carried out on the bone model. An example of geometrical construction is that of a
line that may be supposed to lie in the frontal plane of the femur. This is the
so-named cylinder axis, which is the axis that joins the centers of the two spheres
that fit the medial and lateral posterior condyles of the femur and that is supposed to
lie in the femur frontal plane (Yin et al. 2015).
Fig. 9 Example of anatomical frame definition (Cappozzo et al. 1995). Right-handed set of axes
with the following characteristics (only two axes are defined because the third axis is the vector
product of them). (a) Pelvis; origin: midpoint between the right anterior superior iliac spine
(RASIS) and left anterior superior iliac spine (LASIS); zp-axis: oriented as the line passing through
the RASIS and the LASIS with positive direction to the right; xp-axis: lies in the plane defined by
the RASIS and LASIS and the midpoint between RPSIS and LPSIS and with its positive direction
forward. (b) Femur; origin: midpoint between the lateral epicondyle (LE) and medial epicondyle
(ME); yf-axis: it joins the origin with the center of the femoral head (FH) and its positive direction is
proximal; zf-axis: it lies in the plane defined by FH, ME, and LE, and its positive direction is from
left to right. (c) Tibia and fibula; origin: midpoint between the lateral malleolus (LM) and medial
malleolus (MM); ys-axis: it is the line of intersection between the plane defined by the head of the
fibula (HF), LM and MM (frontal plane), and the plane orthogonal to it passing through the tibial
tuberosity (TT), LM, and MM; its positive direction is proximal; zs-axis: it lies in the frontal plane
with positive direction from left to right
Construction of the Anatomical Frame
Once the subject-specific digital model of a bone is available, the relevant anatom-
ical frame must be constructed. As already mentioned, this is done using the position
of anatomical landmarks, and/or points or lines that lie on an anatomical plane, and a
geometric rule. The anatomical frame may be constructed in the morphology
technical frame or in the motion technical frame and, then, represented in whatever
other frame it is required through the transformation represented by Eq. 6. It is
evident that, given the many anatomical landmarks and other morphological features
of the bone that can be made available, many anatomical sets of axes may be defined
and have, in fact, been defined and illustrated in the literature (Fig. 9; Cappozzo et al.
1995; Kadaba et al. 1990; Wu and Cavanagh 1995; Wu et al. 2002).
As already mentioned, while defining the term “anatomical frame,” the issue here
is the inter- and intra-subject repeatability of this set of axes and their portability.
Critical factors are the following. Anatomical landmarks are areas and not points, as
assumed to be in the present construction, and their definition may slightly change
depending on the source that describes them. When an operator palpates an external
anatomical landmark, these circumstances may lead to large inter- and intra-operator
variability as quantified in Della Croce et al. (1999). The uncertainty that affects
internal anatomical landmarks or lines, such as those assumed to coincide with a
joint center or a joint rotation axis, may also impact very negatively on repeatability
(Stagni et al. 2000). By merging the human movement and the ever-developing
medical imaging laboratories, we may be able, in the next future, to find better
solutions to this problem. In addition, these enhanced solutions should be shared
within the human movement analyst community and possibly undergo a standard-
ization process for the sake of data portability.
Joint Kinematics Estimation
The assessment of the stability and mobility of a joint is based on the observation of
the relative movement of the two adjacent bones involved, that is, of joint kinemat-
ics. For each bone of interest, the procedures described in the previous sections
provide an estimate of the pose of an anatomical frame (only orientation if
magneto-inertial sensors are used) relative to a global frame of choice in each
sampled instant of time during movement (Eqs. 1 and 5). As already mentioned in
a previous section, the relative pose between two rigid bodies is described by six
scalar quantities, the six degrees of freedom; three describing the mutual orientation
and three the mutual position. The objective of this section is to review the possible
descriptions of joint kinematics and the issues associated with them.
Given the orientation matrices and the position vectors of the anatomical frames
of the proximal and distal bones of the joint under analysis relative to a global frame
in any given instant of time (Fig. 10)
g
Rp g op and g Rd g od, (8)
we may represent the position vector of any given point in the global frame as a
function of its position vector expressed in both local frames:
g
p ¼ g Rp p p þ g op (9)
and
g
p ¼ g Rd d p þ g od: (10)
By eliminating gp in Eqs. 9 and 10 and pre-multiplying both left and right terms
of the resulting equation by the transposed matrix g RpT , p p may be calculated as
Fig. 10 Proximal and distal

bone anatomical frames, their
pose relative to the global
frame (gRp , gop ,
and gRd, god respectively)
and the pose of the distal bone
relative to the proximal bone
(pRd , pod)

p
p ¼ g RpT g Rd d p þ g RpT g od g op : (11)
From this equation, and keeping in mind Eq. 6, it results that the orientation and
position of the distal bone relative to the proximal bone, are given by

p
Rd ¼ g RpT g Rd and p od ¼ g RpT g od g op , respectively: (12)
Although pRd and pod fully describe joint kinematics, the scalar quantities that
appear in them do not necessarily comply with the requirements of consistency with
the anatomical and physiological terminology and thus of an effective description of
function. This issue is tackled as illustrated in the following.
Translational Degrees of Freedom
The relative position of two adjacent bones in a given instant of time is described
making reference to a vector (t) joining the position of a point defined in the
proximal (Kp) and that of a point defined in the distal (Kd) local frames. These two
points are normally chosen so that they coincide while the subject assumes a
reference posture (zero joint translation). The translational degrees of freedom are
described by the way the three components of vector t vary in time during the
movement. Since during function translation is coupled with rotation, vector
t depends on the location of the above-mentioned points. This location must
therefore be chosen with care and accounting for the characteristics of the joint
Fig. 11 A possible definition

of knee translation vector (t)
during flexion. Kp is made to
coincide with the midpoint of
the transepicondylar axis; Kd
and Kp are rigid with the tibia
and with the femur,
respectively, and they
coincide during a selected
reference posture
dealt with and based on the objective of the analysis. For instance, when dealing with
the hip joint, Kd and Kp may be defined as coinciding with the mean center of
rotation of that joint and associated with both pelvic bone and femur in a selected
instant of time during movement or during a selected posture (Cereatti et al. 2010). If
the analysis regards a joint that is more complex from the mechanical point of view,
such as the tibiofemoral joint, than the midpoint of the transepicondylar axis
(Fig. 11; Grood and Suntay 1983), a contact point between the two bones involved
may be of interest.
The other choice that needs to be made regards the set of axes with respect to
which vector t is represented, again in each sampled instant of time. This set of axes
may be that of the proximal or distal anatomical frame. Normally, the axes used to
describe the three rotational degrees of freedom, illustrated later, are used in their
current orientation.
It is worthwhile noticing that, due to the fact that joint linear displacements are
normally smaller or in the order of the resolution of most motion capture systems,
there is little literature dealing with them and no shared convention for their
description.
Rotational Degrees of Freedom
The quantification of the rotational degrees of freedom is typically based on the

observation, in the sampled instants of time during movement, of the orientation of
the distal bone relative to the proximal bone using a sequence of three rotations about
the distal bone anatomical axes. The rotation sequence proposed by Grood and
Suntay (1983) and recommended by Wu and Cavanagh (1995) and Wu et al.
(2002), that is, the Cardan sequence, yields three angles that best match the way
that functional anatomy uses for most joints (Fig. 12): (i) angle γ around zd
(coinciding with zp) for flexion-extension, (ii) angle α around the current orientation
of xd for abduction-adduction, and (iii) angle β around the current orientation of yd
for internal-external rotation. If, during the illustrated sequence, each of these three
Fig. 12 The Cardan angles that describe an orientation of the distal bone relative to the proximal
bone are defined according to the following sequence of operations: (i) to begin with the anatomical
frames of the proximal ( p, hip bone) and distal bone (d, femur) are parallel; (ii) the first rotation γ
occurs around the dz-axis; (iii) the second rotation α around the dx-axis in the orientation, it assumes
after the first rotation (d1x); (iv) the third rotation β around the dy-axis in the orientation, it assumes
after the second rotation (d2y). Axes in red (dz, d1x, d2y) are the joint axes and d1x is the floating axis
axes is frozen with the orientation it assumes during the related rotation, a set of
non-orthogonal axes is obtained (Fig. 12). The axis (xd) around which the second
rotation occurs is referred to as floating axis, and the three axes are named joint axes.
The equations that allow the calculation of the Cardan angles from the orientation
matrix pRd are
α ¼ sin 1p Rd ð3, 2Þ
β ¼ sin 1 ½p Rd ð3, 1Þ= cos α (13)
γ ¼ sin 1 ½p Rd ð1, 2Þ= cos α
The demonstration of these equations can be found in the standard mechanical

literature. Note that, if the second rotation equals π/2 or a multiple of it, then we incur
in a singularity condition (gimbal lock). This circumstance prevents the use of the
Cardan convention for those joints which, during movement, may undergo an
abduction-adduction greater than π/2.
Moreover, using this method, the instantaneous orientation of the distal bone is
represented starting from a reference orientation that is equal to that of the proximal
bone. Since during upright posture, the two anatomical frames are not necessarily
parallel; the corresponding joint angles are not zero as often assumed to be in the
biomedical literature.
It is also important to remark that the three angles thus obtained univocally
describe an orientation, but not true rotations. In addition, the Cardan angles, as
more in general the Euler angles, do not enjoy the additive property. This means that,
in principle, we cannot subtract two values of a given Cardan angle, observed in two
different instants of time during movement or different postures, and, indeed, the
result cannot be interpreted as an angular displacement, nor can we calculate relevant
arithmetical averages (Pierrynowski and Ball 2009). For the same reason, when
plotting a Cardan angle versus time, a continuous line should not be made to join two
sampled values because this would imply an interpolation between them. We may
conclude that, paraphrasing what Churchill is told to have said with regard to
democracy, the Cardan angles is the worst form of joint rotation description, except
for all the others.
The second approach, definitely less popular than and as abstract as the previous
one, is based on the orientation vector pθd applied to the orientation, observed in the
sampled instants of time during movement, of the distal bone relative to the proximal
bone anatomical frame (Fioretti et al. 1997; Woltring 1994). This vector may be
represented in any set of local axes of choice, the proximal bone or the distal bone
anatomical axes, or the non-orthogonal joint axes. This choice shall be tailored upon
the consistency of the three resulting angles with functional anatomy. A positive
property of this method is that it does not suffer the gimbal lock issue, but displays all
other limitations of the Cardan angles.
A third method to describe joint kinematics uses the following geometric
approach (Paul 1992). The flexion-extension rotation is measured through the
angle formed by the y-axis of the distal bone and the projection onto the xy (sagittal)
plane of this bone of the y-axis of the proximal bone. The abduction-adduction
rotation is given by the angle formed by the y-axis of the distal bone and the
projection of the y-axis of the proximal bone onto the yz (frontal) plane of the distal
bone. The internal-external rotation angle is the angle formed by the x-axis of the
distal bone and the projection of the x-axis of the proximal bone onto the xz
(transverse) plane of the distal bone. This approach is very intuitive; in fact it is
similar to the way used in functional anatomy to describe joint motion. It is as
abstract as the other methods, but it may deserve some attention from the human
movement analyst community.
Precision and Accuracy of Joint Kinematics Estimates
Two major issues affect the estimate of joint kinematics. One impacts on accuracy
and it is mainly associated with the soft tissue artifact. The other affects precision and
depends on the variability with which anatomical landmarks and, therefore, anatom-
ical frames are determined.
In a preceding section, we have illustrated the origin of the soft tissue artifact and
how and in what measure it impacts the estimate of bone pose while using nonin-
vasive motion capture techniques. Figure 13 shows an example of the propagation of
this artifact to the six degrees of freedom of the knee during the support phase of
running, when the single-body optimization method previously illustrated is used.
Inaccuracies in the order of 10 and 10 mm for rotations and linear displacements,
respectively, may be observed.
Fig. 13 The six degrees of freedom of the knee joint during the support phase of running of a lean
young adult able-bodied male subject. Black lines represent reference values obtained using clusters
of markers mounted on intracortical pins (Cereatti et al. 2017). Gray lines represent the values
obtained using clusters of skin markers and a single-body least-squares bone pose estimator
(Camomilla et al. 2015)
As far as the precision issue is concerned, in Table 1 we report data that provide
an idea of the impact that the variability in the identification of the orientation of the
mediolateral anatomical axis of the femur may have on knee kinematics during level
walking as described using for different conventions.
Future Directions
Although the potential of quantitative movement analysis in research, professional

decision-making, and intervention practice is fully recognized, its application to
large portions of the locomotor apparatus that move in sizeable volumes is currently
limited by several problems. The experimental and analytical protocols, which have
been described in this chapter, provide results regarding skeletal motion with
precision and accuracy that are insufficient to answer many of the questions posed
by scientists and professionals. At the moment, however, using optical or magneto-
inertial technology, the reconstruction of skeletal movement may be carried out with
errors that may be in the order of 10 mm and 10 or more, for position and
orientation, respectively. This collides with the fact that some relative movements
Table 1 Maximum variation of the knee joint angles (degrees) with respect to the reference values
as a result of the variation of the orientation of the mediolateral anatomical axis of the femur over the
range 15 (Fioretti et al. 1997). Method used for the joint kinematics description: (1) Cardan
angles, (2) projection of the joint orientation vector onto the femoral anatomical axes, (3) projection
of the joint orientation vector onto the knee joint axes, and (4) geometric approach (see previous
section for relevant details)
Method Flexion-extension Abduction-adduction Internal-external rotation
1 1.0 13.8 9.0
2 3.9 17.5 0.0
3 2.5 12.4 6.5
4 1.6 14.3 21.0
between adjacent joints, among those of great interest in several applications, are in
the order of a few millimeters and degrees. Thus, future research must aim at this
resolution by exploiting emerging technology (including other domains, e.g., med-
ical imaging, robotics, animation) and innovative experimental and analytical
methods. In particular it should focus on the following issues:
(a) Development of algorithms able to process the data provided by the motion
capture systems and make data describing the movement of the skeleton with the
mentioned accuracy available. A specific challenge is the compensation for the
artifact movement between the superficially tracked motion and the underlying
structures. Two approaches are being attempted. One uses “intelligent markers”
which, through an ultrasound technology, gathers trial-specific information
about this artifact movement (Masum et al. 2016). Another approach uses
bone pose estimators, either single-body or multi-body kinematics optimization,
that embed a mathematical model of the artifact (Bonnet et al. 2017; Camomilla
et al. 2015).
(b) Devising non- or minimally invasive experimental techniques, possibly usable
in the movement analysis laboratory, and statistical shape analysis able to
provide accurate and detailed subject-specific bone and soft tissue morphology
to be registered with motion data.
(c) The definition and construction of anatomical frames that allow for an adequate
repeatability of the results of the analysis and their standardization for the sake of
data portability.
(d) Implementation of advanced probabilistic approaches that accommodate for
intraindividual variability and input data uncertainty.
In addition, other open issues are the following. Computational models of the
neuromusculoskeletal system are not as integrative across space scales as demanded
by a holistic approach and encounter difficulties in incorporating the characteristics
of a specific subject. Finally, there is no consensus on the optimal metrics to use
when assessing motor function.
References
Andersen MS, Benoit DL, Damsgaard M, Ramsey DK, Rasmussen J (2010) Do kinematic models
reduce the effects of soft tissue artefacts in skin marker-based motion analysis? An in vivo study
of knee kinematics. J Biomech 43:268–273
Andersen MS, Damsgaard M, Rasmussen J, Ramsey DK, Benoit DL (2012) A linear soft tissue
artefact model for human movement analysis: proof of concept using in vivo data. Gait Posture
35:606–611
Banks SA, Hodge WA (1996) Accurate measurement of three-dimensional knee replacement
kinematics using single-plane fluoroscopy. IEEE Trans Biomed Eng 43:638–649
Barré A, Thiran JP, Jolles BM, Theumann N, Aminian K (2013) Soft tissue artifact assessment
during treadmill walking in subjects with total knee arthroplasty. IEEE Trans Biomed Eng
60:3131–3140
Bell AL, Pedersen DR, Brand RA (1990) A comparison of the accuracy of several hip joint center
location prediction methods. J Biomech 23:617–621
Benedetti MG, Merlo A, Leardini A (2013) Inter-laboratory consistency of gait analysis measure-
ments. Gait Posture 38:934–939
Benoit DL, Damsgaard M, Andersen MS (2015) Surface marker cluster translation, rotation, scaling
and deformation: their contribution to soft tissue artefact and impact on knee joint kinematics.
J Biomech 48:2124–2129
van den Bogert AJ, Smith GD, Nigg BM (1994) In vivo determination of the anatomical axes of the
ankle joint complex: an optimization approach. J Biomech 27:1477–1488
Bonci T, Camomilla V, Dumas R, Chèze L, Cappozzo A (2015) Rigid and non-rigid geometrical
transformations of a marker-cluster and their impact on bone-pose estimation. J Biomech
48:4166–4172
Bonnet V, Richard V, Camomilla V, Venture G, Cappozzo A, Dumas R (2017) Joint kinematics
estimation using a multi-body kinematics optimisation and an extended Kalman filter, and
embedding a soft tissue artefact model. J Biomech (in press)
Bouvier B, Duprey S, Claudon L, Dumas R, Savescu A (2015) Upper limb kinematics using inertial
and magnetic sensors: comparison of sensor-to-segment calibrations. Sensors 15:18813–18833
Camomilla V, Cereatti A, Vannozzi G, Cappozzo A (2006) An optimized protocol for hip joint
centre determination using the functional method. J Biomech 39:1096–1106
Camomilla V, Bonci T, Dumas R, Chèze L, Cappozzo A (2015) A model of the soft tissue artefact
rigid component. J Biomech 48:1752–1759
Campbell AC, Lloyd DG, Alderson JA, Elliott BC (2009) MRI development and validation of two
new predictive methods of glenohumeral joint center location identification and comparison
with established techniques. J Biomech 42:1527–1532
Cappozzo A (1984) Gait analysis methodology. Hum Mov Sci 3:27–50
Cappozzo A (1991) Three-dimensional analysis of human locomotor acts: experimental methods
and associated. Hum Mov Sci 10:589–602
Cappozzo A, Catani F, Della Croce U, Leardini A (1995) Position and orientation in space of bones
during movement, anatomical frame definition and determination. Clin Biomech 10:171–178
Cappozzo A, Cappello A, Della Croce U, Pensalfini F (1997a) Surface-marker cluster design
criteria for 3-D bone movement reconstruction. IEEE Trans Biomed Eng 44:1165–1174
Cappozzo A, Della Croce U, Lucchetti L (1997b) Gait data, terminology and definition. In: Allard P,
Cappozzo A, Lumberg A, Vaughan K (eds) Three-dimensional analysis of human locomotion.
Wiley, New York, pp 129–132
Cereatti A, Margheritini F, Donati M, Cappozzo A (2010) Is the human acetabulofemoral joint
spherical? J Bone Joint Surg (Br) 92:311–314
Cereatti A, Bonci T, Akbarshahi M, Aminian K, Barré A, Begon M, Benoit DL, Charbonnier C, Dal
Maso F, Fantozzi S, Lin CC, Lu TW, Pandy MG, Stagni R, van den Bogert AJ, Camomilla V
(2017) Standardization proposal of soft tissue artefact description for data sharing in human
motion measurements. J Biomech. doi:10.1016/j.jbiomech.2017.02.004
Chaibi Y, Cresson T, Aubert B, Hausselle J, Neyret P, Hauger O, de Guise JA, Skalli W (2012) Fast
3D reconstruction of the lower limb using a parametric model and statistical inferences and
clinical measurements calculation from biplanar X-rays. Comput Methods Biomech Biomed
Eng 15:457–466
Challis JH, Pain MTG (2008) Soft tissue motion influences skeletal loads during impacts. Exerc
Sport Sci Rev 36:71–75
Charlton IW, Tate P, Smyth P, Roren L (2004) Repeatability of an optimised lower body model. Gait
Posture 20:213–221
Chiari L, Cappozzo A, Della Croce U, Leardini A (2005) Human movement analysis using
stereophotogrammetry. Part 2: experimental errors. Gait Posture 21:197–211
Clark T, Hawkins D (2010) Are fixed limb inertial models valid for dynamic simulations of human
movement? J Biomech 43:2695–2701
Clément J, Dumas R, Hagemesiter N, de Guise JA (2015) Soft tissue artifact compensation in knee
kinematics by multi-body optimization: performance of subject-specific knee joint models.
J Biomech 48:3796–3802
Clément J, Dumas R, Hagemeister N, de Guise JA (2017) Can generic knee joint models improve
the measurement of osteoarthritic knee kinematics during squatting activity? Comput Methods
Biomech Biomed Eng 20:94–103
Colle F, Lopomo N, Visani A, Zaffagnini S, Marcacci M (2016) Comparison of three formal
methods used to estimate the functional axis of rotation: an extensive in-vivo analysis performed
on the knee joint. Comput Methods Biomech Biomed Eng 19:484–492
Crabolu M, Pani D, Raffo L, Cereatti A (2016) Estimation of the center of rotation using wearable
magneto-inertial sensors. J Biomech 16:3928–3933
Cutti AG, Giovanardi A, Rocchi L, Davalli A, Sacchetti R (2008) Ambulatory measurement of
shoulder and elbow kinematics through inertial and magnetic sensors. Med Biol Eng Comput
46:169–178
Cutti AG, Ferrari A, Garofalo P, Raggi M, Cappello A, Ferrari A (2010) “Outwalk”: a protocol for
clinical gait analysis based on inertial and magnetic sensors. Med Biol Eng Comput 48:17–25
Davis RB, Ounpuu S, Tyburskky D, Gage R (1991) A gait analysis data collection and reduction
technique. Hum Mov Sci 10:575–587
De Rosario H, Page A, Besa A, Mata V, Conejero E (2012) Kinematic description of soft tissue
artifacts: quantifying rigid versus deformation components and their relation with bone motion.
Med Biol Eng Comput 50:1173–1181
De Rosario H, Page A, Besa A (2017) Analytical study of the effects of soft tissue artefacts on
functional techniques to define axes of rotation. J Biomech. doi:10.1016/j.
Della Croce U, Cappozzo A, Kerrigan DC (1999) Pelvis and lower limb anatomical landmark
calibration precision and its propagation to bone geometry and joint kinematics. Med Biol Eng
Comput 37:155–161
Della Croce U, Leardini A, Chiari L, Cappozzo A (2005) Human movement analysis using
stereophotogrammetry. Part 4: assessment of anatomical landmark mislocation and its effects
on joint kinematics. Gait Posture 21:226–237
Donati M, Camomilla V, Vannozzi G, Cappozzo A (2007) Enhanced anatomical calibration in
human movement analysis. Gait Posture 26:179–185
Donati M, Camomilla V, Vannozzi G, Cappozzo A (2008) Anatomical frame identification and
reconstruction for repeatable lower limb joint kinematics estimates. J Biomech 41:2219–2226
Dumas R, Camomilla V, Bonci T, Chèze L, Cappozzo A (2014) Generalized mathematical
representation of the soft tissue artefact. J Biomech 47:476–481
Dumas R, Camomilla V, Bonci T, Chèze L, Cappozzo A (2015) What portion of the soft tissue
artefact requires compensation when estimating joint kinematics? J Biomech Eng 137:064502.
doi:10.1115/1.4030363
Duprey S, Chèze L, Dumas R (2010) Influence of joint constraints on lower limb kinematics
estimation from skin markers using global optimization. J Biomech 43:2858–2862
functional joint axes. J Biomech 40:2150–2157
Favre J, Aissaoui R, Jolles BM, de Guise JA, Aminian K (2009) Functional calibration procedure
for 3D knee joint angle description using inertial sensors. J Biomech 42:2330–2335
Fioretti S, Cappozzo A, Lucchetti L (1997) Joint kinematics. In: Allard P, Cappozzo A, Lumberg A,
Vaughan K (eds) Three-dimensional analysis of human locomotion. Wiley, New York, pp
173–189
Fraysse F, Thewlis D (2014) Comparison of anatomical, functional and regression methods for
estimating the rotation axes of the forearm. J Biomech 47:3488–3493
Frigo C, Rabuffetti M, Kerrigan DC, Deming LC, Pedotti A (1998) Functionally oriented and
clinically feasible quantitative gait analysis method. Med Biol Eng Comput 36:179–185
Gamage SSHU, Lasenby J (2002) New least squares solutions for estimating the average Centre of
rotation and the axis of rotation. J Biomech 35:87–93
Garling EH, Kapteina BL, Mertens B, Barendregt W, Veeger HEJ, Nelissen RGHH, Valstar ER
(2007) Soft-tissue artefact assessment during step-up using fluoroscopy and skin-mounted
markers. J Biomech 40:S18–S24
Gasparutto X, Sancisi N, Jacquelin E, Parenti-Castelli V, Dumas R (2015) Validation of a multi-
body optimization with knee kinematic models including ligament constraints. J Biomech
48:1141–1146
Grimpampi E, Camomilla V, Cereatti A, De Leva P, Cappozzo A (2014) Metrics for describing soft-
tissue artefact and its effect on pose, size, and shape of marker clusters. IEEE Trans Biomed Eng
61:362–367
dimensional motions: application to the knee. J Biomech Eng 105:136–144
Gruber K, Ruder H, Denoth J, Schneider K (1998) A comparative study of impact dynamics:
wobbling mass model versus rigid body models. J Biomech 31:439–444
Guan S, Gray HA, Keynejad F, Pandy MG (2016) Mobile biplane X-ray imaging system for
measuring 3D dynamic joint motion during overground gait. IEEE Trans Med Imaging
35:326–336
Halvorsen K (2003) Bias compensated least square estimate of the center of rotation. J Biomech
36:999–1008
the center of rotation. J Biomech 32:1221–1227
Hara R, McGinley J, Briggs C, Baker R, Sangeux M (2016) Predicting the location of the hip joint
centres, impact of age group and sex. Sci Rep 6:37707
imaging. J Biomech 40:595–602
during level walking. J Orthop Res 8:383–392
Kainz H, Carty CP, Modenese L, Boyd RN, Lloyd DG (2015) Estimation of the hip joint centre in
human motion analysis: a systematic review. Clin Biomech 30:319–329
Lamberto G, Martelli S, Cappozzo A, Mazzà C (2016) To what extent is joint and muscle mechanics
predicted by musculoskeletal models sensitive to soft tissue artefacts? J Biomech. doi:10.1016/j.
jbiomech.2016.07.042i
(1999) Validation of a functional method for the estimation of hip joint centre location.
Leardini A, Chiari L, Cappozzo A, Della Croce U (2005) Human movement analysis using
stereophotogrammetry. Part 3: soft tissue artifact assessment and compensation. Gait Posture
21:212–225
based protocol for gait analysis in children. Gait Posture 26:560–571

estimation of the glenohumeral joint center by functional methods: accuracy and repeatability
assessment. J Biomech 43:370–374
Li K, Zheng L, Tashman S, Zhang X (2012) The inaccuracy of surface-measured model-derived
tibiofemoral kinematics. J Biomech 45:2719–2723
Liu W, Nigg BM (2000) A mechanical model to determine the influence of masses and mass
distribution on the impact force during running. J Biomech 33:219–224
Lu TW, O’Connor JJ (1999) Bone position estimation from skin marker coordinates using global
optimization with joint constraints. J Biomech 32:129–134
Luinge HJ, Veltink PH, Baten CTM (2007) Ambulatory measurement of arm orientation.
Masum MA, Pickering MR, Lambert AJ, Scarvell JM, Smith PN (2016) Multi-slice ultrasound
image calibration of an intelligent skin-marker for soft tissue artefact compensation. J Biomech.
doi:10.1016/j.jbiomech.2016.12.030
McGinnis RS, Perkins NC (2013) Inertial sensor based method for identifying spherical joint center
of rotation. J Biomech 46:2546–2549
Melhem E, Assi A, El Rachkidi R, Ghanem I (2016) EOS ® biplanar X-ray imaging: concept,
developments, benefits, and limitations. J Child Orthop 10:1–14
Pain MTG, Challis JH (2001) High resolution determination of body segment inertial parameters
and their variation due to soft tissue motion. J Appl Biomech 17:326–334
Paul JP (1992) Terminology and units. Deliverable n. 4, C.E.C. Program AIM, Project A-2002,
CAMARC-II
Piazza SJ, Erdemir A, Okita N, Cavanagh PR (2004) Assessment of the functional method of hip
joint center location to reduced range of hip motion. J Biomech 37:349–356
Picerno P, Cereatti A, Cappozzo A (2008) Joint kinematics estimate using wearable inertial and
magnetic sensing modules. Gait Posture 28:588–595
Pierrynowski MR, Ball KA (2009) Oppugning the assumptions of spatial averaging of segment and
joint orientations. J Biomech 42:375–378
Quijano S, Serrurier A, Aubert B, Laporte S, Thoreux P, Skalli W (2013) Three-dimensional
reconstruction of the lower limb from biplanar calibrated radiographs. Med Eng Phys
35:1703–1712
Rabuffetti M, Ferrarin M, Mazzoleni P, Benvenuti F, Pedotti A (2003) Optimised procedure for the
calibration of the force platform location. Gait Posture 17:75–80
Reinbolt JA, Schutte JF, Fregly BJ, Koh B, Haftka RT, George AD, Mitchell KH (2005) Determi-
nation of patient-specific multi-joint kinematic models through two-level optimization.
Richard V, Lamberto G, Lu TW, Cappozzo A, Dumas R (2016) Knees kinematics estimation using
multi-body optimisation embedding a knee joint stiffness matrix: a feasibility study. PLoS ONE.
Riddick RC, Kuo AD (2016) Soft tissues store and return mechanical energy in human running.
Scheys L, Desloovere K, Spaepen A, Suetens P, Jonkers I (2011) Calculating gait kinematics using
MR-based kinematic models. Gait Posture 33:158–164
Seel T, Schauer T, Raisch J (2012) Joint axis and position estimation from inertial measurement data
by exploiting kinematic constraints. In: Proceedings of IEEE international conference on control
applications, Dubrovnik, Croatia, pp 45–49
Seidel GK, Marchinda DM, Dijkers M, Soutas-Little RW (1995) Hip joint center location from
palpable bony landmarks – a cadaver study. J Biomech 28:995–998
Sheehan FT (2010) The instantaneous helical axis of the subtalar and talocrural joints:
a non-invasive in vivo dynamic study. J Foot Ankle Res 3:13. doi:10.1186/1757-1146-3-13
Sholukha V, Van Sint JS, Snoeck O, Salvia P, Moiseev F, Rooze M (2009) Prediction of joint center
location by customizable multiple regressions: application to clavicle, scapula and humerus.
Shuster MD (1993) A survey of attitude representations. J Astronaut Sci 41:439–517

Soderkvist I, Wedin PA (1993) Determining the movements of the skeleton using well-configured
markers. J Biomech 26:1473–1477
Stagni R, Leardini A, Cappozzo A, Benedetti MG, Cappello A (2000) Effects of hip joint centre
mislocation on gait analysis results. J Biomech 33:1479–1487
Van Sint JS, Hilal I, Salvia P, Sholukha V, Poulet P, Kirokoya I, Rooze M (2003) Data representation
for joint kinematics simulation of the lower limb within an educational context. Med Eng Phys
25:213–220
Wakeling JM, Nigg BM (2001) Soft-tissue vibrations in the quadriceps measured with skin
mounted transducers. J Biomech 34:539–543
Woltring HJ (1994) 3-D attitude representation of human joints, a standardisation proposal.
J Biomech 27:1399–1414
Wu G, Cavanagh PR (1995) ISB recommendations for standardization in the reporting of kinematic
data. J Biomech 28:1257–1261
joint coordinate system of various joints for the reporting of human joint motion – part I, ankle,
hip, and spine. J Biomech 35:543–548
Yin L, Chen K, Guo L, Cheng L, Wang F, Yang L (2015) Identifying the functional flexion-
extension axis of the knee: an in-vivo kinematics study. PLoS ONE 10:e0128877
Zelik KE, Kuo AD (2010) Human walking isn’t all hard work: evidence of soft tissue contributions
to energy dissipation and return. J Exp Biol 213:4257–4264
Estimation of the Body Segment Inertial
Parameters for the Rigid Body
Biomechanical Models Used in Motion
Analysis
Raphaël Dumas and Janis Wojtusch
Abstract
Body segment inertial parameters (BSIPs) of the human body are key parameters
in biomechanics to study the dynamics of human motion. BSIPs can be obtained
in different ways including direct measurements on cadavers or photogrammetry
and medical imaging on living humans, but they are more generally estimated by
regression equations (based on those measurements).
This chapter overviews three widely used regression equations reported by
Winter (2009), de Leva (1996a), and Dumas et al. (2007a). These regression
equations are presented for the head with neck, thorax, abdomen, pelvis, and right
upper arm, forearm, hand, thigh, shank, and foot segments.
The segment endpoints and segment reference frames defined at the time of the
BSIPs assessment and regression computation are reviewed so that the reader can
consider how they match with the construction of the rigid body biomechanical
models they would like to use for motion analysis. The segment definitions and
regression equations that remain undefined or unavailable are indicated, and some
assumptions are proposed to amend them, where found applicable. The compu-
tation of the segment mass, position of center of mass, moments, and products of
inertia from these regression equations are fully detailed, including the modifica-
tion of the designation of the segment axes and the transformation from right to
left segments.
R. Dumas (*)
LBMC UMR_T9406, Univ Lyon, Université Claude Bernard Lyon 1, IFSTTAR, Lyon, France
e-mail: raphael.dumas@ifsttar.fr
J. Wojtusch
Department of Computer Science, Simulation, Systems Optimization and Robotics Group, TU
Darmstadt, Darmstadt, Germany
e-mail: wojtusch@sim.tu-darmstadt.de

DOI 10.1007/978-3-319-30808-1_147-1
2 R. Dumas and J. Wojtusch
Keywords
Segment mass • Center of mass • Moments of inertia • Regression equations •
Segment length • Segment endpoints • Joint center • Segment reference frame
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Computation of the BSIPs from the Regression Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Segment Definition and Regression Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Introduction
Body segment inertial parameters (BSIPs) are required in biomechanics for the
computation of intersegmental moments, angular momentum, mechanical work,
and for the study of the whole body dynamic stability (see “▶ 3D Kinetics of
Human Motion,” “▶ Induced Acceleration and Induced Power Analyses of
Human Motion,” “▶ Optimal Control Modeling of Human Movement,” and
“▶ Physics-based Models for Human Gait Analysis”). BSIPs refer to the segment
mass, the position of the segment center of mass with respect to a segment reference
frame, and the segment moments and products of inertia with respect to a segment
point (typically the segment center of mass or a segment endpoint) and with respect
to a segment reference frame (see “▶ Three-dimensional Reconstruction of the
Human Skeleton in Motion”).
BSIPs of the human body can be obtained in different ways. The history and the
description of the techniques used to assess BSIPs can be found in Pearsall and Reid
1994, Reid and Jensen 1990, and Drillis et al. 1964. BSIPs can be directly measured on
cadavers (Dempster 1955; Clauser et al. 1969; Chandler et al. 1975) and indirectly
measured on living subjects, typically through photogrammetry (Ackland et al. 1988;
Jensen 1978; McConville et al. 1980; Young et al. 1983) or medical imaging (Bauer
et al. 2007; Dumas et al. 2005; Durkin et al. 2002; Mungiole and Martin 1990; Pearsall
et al. 1996; Zatsiorsky et al. 1990; Cheng et al. 2000). Based on these measurements,
estimations of the BSIPs from regression equations are more classically used. Obvi-
ously, these regression equations are limited by the number and the nature of the
subjects on which they have been established. In other words, regression equations for
BSIPs are better established for older Caucasian non-pathological males than for
females, children, and, as a matter of fact, for pathological subjects. Nevertheless,
regression equations are widely used because of their expediency.
Furthermore, when studying the dynamics of human motion with a 3D rigid body
biomechanical model, another difficulty is to match the segment definition used for
the BSIP assessment with the model construction. Indeed, BSIPs have been obtained
with their own rationale for segment endpoints and segment reference frames,
generally constrained by experimental and technical limitations. Conversely, rigid
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 3
body biomechanical models (see “▶ Rigid Body Models of the Musculoskeletal

System” in this book) and especially “conventional gait models” widely used in
clinical motion analysis (Davis et al. 1991; Kadaba et al. 1990) (see “▶ Conventional
Gait Model – Success and Limitations” in this book), are based on a chosen marker
set and calibration protocol that aims at approximating the joint centers and axes. In
this context, several adjustment procedures of existing regression equations have
been proposed (de Leva 1996a; Dumas et al. 2007a; Hinrichs 1990) for a better
correspondence between segment definition and model construction. Still, no spe-
cific adjustment for “conventional gait models” has been proposed, and it is com-
monly assumed that the segment definition used for the BSIP assessment and the
model construction are consistent. Therefore, classically, the segment center of mass
is expected to align with the axis linking the joint centers, and this axis plus two
orthogonal axes are expected to be principal axes of inertia.
This chapter overviews three widely used sets of regression equations for BSIPs,
allowing for a 16-segment rigid body biomechanical model. The segment definitions
(i.e., segment endpoints and segment axes or planes) used for the BSIP assessment,
regression computation, and adjustment are reviewed so that the reader can consider
how they match with the construction of the “conventional gait models” or any rigid
body biomechanical models they would like to use.
The segment definitions and regression equations that remain undefined or
unavailable are specified, and some assumptions are proposed to amend them,
where found applicable. Note that the planes of segmentation used for the BSIP
assessment can be also an issue (leading to some differences between the regression
equations, especially for the trunk segments), but, conversely to the segment end-
points, axes, and planes, this does not directly interfere with the construction of the
biomechanical models used in motion analysis and this is not reviewed in this chapter.
State of the Art
Three widely used regression equations for BSIP are the regression equations of
Winter (2009) derived from the data of Dempster (1955), the regression equations of
de Leva (1996a) adjusted from the data of Zatsiorsky et al. (1990), and the regression
equations of Dumas et al. (2007a, b) adjusted from the data of McConville et al.
(1980) and Young et al. (1983).
Dempster (1955) directly measured the BSIPs of eight male cadavers (mean age
68.5 years old, mean weight 61.1 kg, mean stature 1.69 m) using equilibrium and
pendulum methods. Zatsiorsky et al. (1990) indirectly measured the BSIPs by frontal
gamma-ray scanner on 100 males (mean age 23.8 years old, mean weight 73.0 kg,
mean stature 1.74 m) and 15 females (mean age 19.0 years old, mean weight 61.9 kg,
mean stature 1.73 m). They obtained the surface density of the body from subjects
lying supine. The foot segment was scanned in a lateral view separately from the rest
of the body. The whole human body was modeled as rectangular cuboids of 2 cm
width and length and of different heights estimated from anthropometric measure-
ments. The masses and the distances from the geometrical centers of the cuboids to
reference anatomical landmarks were known from the surface density. The centers of
the mass of the cuboids were assumed at their geometrical centers and the principal
axes of inertia of the cuboids were assumed aligned with the axes of symmetry of the
cuboids. Then, the BSIPs of the segments were computed by summing up the BSIPs
of the cuboids (i.e., using weighted barycenter and parallel axis theorem).
McConville et al. (1980) and Young et al. (1983) indirectly measured the BSIPs
by photogrammetry on 31 males (mean age 27.5 years old, mean weight 77.3 kg,
mean stature 1.77 m) and 46 females (mean age 31.2 years old, mean weight 63.9 kg,
mean stature 1.61 m), respectively. The BSIPs are defined relative to skin anatomical
landmarks assuming a homogenous density of 1 g/cm3.
Based on these datasets and according to the aforementioned adjustments, linear
regression equations have been proposed (de Leva 1996a; Dumas et al. 2007a;
Winter 2009). The segment mass is computed as a percentage of the body mass.
The position of the center of mass is computed as a percentage of the segment length,
defined as the distance between the segment endpoints. The radii of gyration (i.e., the
square roots of the moments of inertia divided by the segment mass) are computed as
percentages of the segment length. In these regression equations, the number of
segments is 16: the head with neck, thorax, abdomen, pelvis, right and left upper
arms, forearms and hands, and thighs, shanks, and feet.
In the literature, other segmentations and other regression equations (i.e., nonlinear
(Zatsiorsky et al. 1990), involving anthropometric measurements such as segment
circumferences (Yeadon and Morlock 1989; Zatsiorsky et al. 1990)) exist but appear
hardly used, probably because they do not fit well with the motion analysis protocols
(i.e., they involve not only skin markers but also calipers and tape measures).
Computation of the BSIPs from the Regression Equations
As previously mentioned, the segment mass, ms, of segment s = 1,. . ., 16 is

estimated as:
ms ¼ ps M (1)
where ps is the percentage of the body mass M.

The position of the center of mass with respect to the segment reference frame is
estimated as:
0 1
cXs
rs ¼ Ls @ cY
s
A (2)
Z
cs
where cXs , cY Z
s , cs are coordinates of the center of mass expressed as percentages of
the segment length Ls.
The inertia matrix (moments and products of inertia in and out the diagonal,
respectively) with respect to the center of mass and the segment reference frame is
estimated as:
2 2 XY 2 XZ 2 3
r XX
s r r
6 sYY 2 sYZ 2 7
Is ¼ ms ðLs Þ2 4 rs r 5 (3)
sZZ 2
sym: rs
where r XX YY ZZ
s , r s , r s are radii of gyration (i.e., the square roots of the moments of
inertia divided by the segment mass) expressed as percentages of the segment length
Ls. The products of inertia are expressed in the same way (i is indicated in case of
negative products of inertia).
The designation of the segment axes can be different from one rigid body
biomechanical model to another. The regression equations in the following tables
will be labeled with respect to the anterior-posterior, superior-inferior, and medial-
lateral axes. These labels are intended for a subject in anatomic posture (standing
upright with arms at the sides, palms facing forward, and feet parallel). Note that in
the “conventional gait models,” I axis is anterior, J axis is lateral (to the left), and
K axis is superior (Davis III et al. 1991; Kadaba et al. 1990). However, according to
the standardization of the International Society of Biomechanics (ISB), X axis is
anterior, Y axis is superior, and Z axis is lateral (to the right) (Wu et al. 2002, 2005).
The sign convention used in the following tables complies with this standardization
of the ISB (X, Y, and Z axes). Nevertheless, the position of the center of mass and
the inertia matrix with respect to the I, J, and K axes can be easily computed using a
permutation matrix P:
" # 0 X1
1 0 0 cs
rs ¼ 0 0 1 L s @ cY s
A (4)
0 1 0 c Z
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} s
P
and
2 2 XY 2 XZ 2 3
r XX
s r r
6 sYY 2 sYZ 2 7 T
Is ¼ ms ðLs Þ2 ½P4 rs r 5 ½ P : (5)
sZZ 2
sym: rs
In the same way, the regression equations in the following tables are for the right
upper and lower limb segments. The position of the center of mass and the inertia
matrix for the left segments can be computed (in the X, Y, and Z axes) using a
symmetry matrix S (in place of P) in Eqs. 4 and 5:
2 3
1 0 0
S ¼ 40 1 0 5: (6)
0 0 1
The inertia matrix (moments and products of inertia) in Eq. 3 is given with respect
to the center of mass and the segment reference frame. The inertia matrix with
respect to the origin of the segment reference frame can be computed with the
parallel axis theorem:

Is ¼ Is þ m s ðrs ÞT rs E33 rs ðrs ÞT (7)
where E3 3 is the identity matrix.
Segment Definition and Regression Equations
For the regression equations of Winter (2009), the BSIPs for the segments of the trunk
are revised based on the original data of Dempster (1955) and the additional estima-
tions of Plagenhoef (1971): the center of mass of the abdomen and pelvis segments
was estimated by equilibrium methods applied to a cardboard-lead cutout modeling
the anthropometry and mass distribution of the trunk as reported by Dempster (1955).
Note that, in the original data of Dempster (1955), only an abdominopelvic segment
is reported. Its mass is 26.4% of body mass and its segment density is 1.01 g/cm3. The
position of its center of mass about the superior-inferior axis is 59.9% of segment
length (i.e., thoracic joint center to midpoint between hip joint centers). The abdomen
and pelvis segments are presented in the following tables for consistency among the
different regression equations. Nevertheless, this abdominopelvic segment may be
preferred if this segmentation matches with the rigid body biomechanical model used
for motion analysis. Moreover, the abdomen and pelvis segments were also available
in the regression equations reported by Winter (2009), but the position of the center of
mass of these segments, as well as the thorax segment, was given in percentage of the
longitudinal distance between midpoints between glenohumeral and hip joint centers
(i.e., 30.4% of total height (Plagenhoef 1971)).
As for the regression equations based on the data of McConville et al. (1980) and
Young et al. (1983), the regression equations provided in the following tables include
updates for the thorax and abdomen segments (Dumas et al. 2015) and a revision made
for this chapter (i.e., some typos and inconsistency in the regression equations for the
joint centers previously used in the adjustment procedure were corrected).
The segment endpoints and segment axes are presented in Fig. 1 with the
related skin landmarks and joint centers estimated by different regression equations.
The regression equations for the BSIPs of the different segments are given in
(Tables 1–10).
Future Directions
As previously mentioned in the introduction, regression equations (specifically

regression equations based on body mass and segment lengths) are widely used
because of their expediency but remain limited and sometimes difficult to harmonize
with the rigid body biomechanical models used for motion analysis.
10.4%
Acromion
Gleno-humeral joint centre Head vertex
Sellion
100%
4.3%
Medial humeral epicondyle
Lateral humeral epicondyle 55%
Elbow joint centre 8°
Radiale 7th cervical vertebra Cervical joint centre
Acromion Suprasternal notch
100%
33% 11° Gleno-humeral joint centre
Wrist joint centre
0.6%
Radius styloid Ulnar styloid 100%
2nd metacarpal heads 5th metacarpal heads Xiphoid
8th thoracic vertebra
3rd metacarpale 94°
Thoracic joint centre
0.7% 12th thoracic vertebra
Hip joint centre 52%
Greater trochanter
Midpoint between
posterior-superior iliac spines
100% Lumbar joint centre 33.5%
7.4% 3.2%
Medial femoral epicondyle 37.0% Left anterior-superior iliac spine
Lateral femoral epicondyle
Knee joint centre 36.1%
Fibula head
Tibiale 100%
Right anterior-superior iliac spine
100%
Hip joint centre
9.5%
Lateral maleollus 3.2% Medial maleollus

Ankle joint centre Shyrion
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . .
Heel
5th metatarsal head

1rst metatarsal head
Tip of 2nd toe
Fig. 1 Skin landmarks, joint centers estimated by regression equations, and body segments
7
8
Table 1 Head with neck

(Dempster 1955; Winter (de Leva 1996a; (Dumas et al. 2007b; McConville et al. 1980;
The head with neck 2009) Zatsiorsky et al. 1990) Young et al. 1983)
Segment endpoints Head vertex skin Skin landmark on head Skin landmark on the head vertex
landmarka vertex projected on Cervical joint center estimated by the regression
Cervical joint center superior-inferior axis equations (Dumas et al. 2007a)
obtained by dissection Skin landmark on 7th Female: cervical joint center on a direction forming
(center of C7-T1 disc) cervical vertebra an angle of 14 in the sagittal plane with the vector
projected on superior- from 7th cervical vertebra to suprasternal notch
inferior axis skin landmarks and at 53% of the thorax width (i.e.,
distance between 7th cervical vertebra and
suprasternal notch skin landmarks) from 7th
cervical vertebra skin landmark
Male: angle of 8 and percentage of 55%
Segment Superior-inferior axis Superior- Superior-inferior axis from cervical joint center to
reference frame from cervical joint inferior axis: N/A head vertex skin landmark
center to head vertex Frontal plane parallel to Sagittal plane containing cervical
skin landmark the gamma-ray scanner joint center and skin landmarks on the head vertex
Sagittal plane set as acquisition plane and sellion
plane of oscillation of Origin at head vertex Origin at cervical joint center
the pendulum method skin
Origin at head vertex landmark projected on
skin landmarka superior-inferior axis
Segment length N/A Female: 243.7 and Female: 243 and male: 278
(mm) male: 242.9
Segment density 1.11 N/A Assumed to be 1
(g/cm3)
Segment mass (% of Female: N/A and male: Female: 6.68 and male: Female: 6.7 and male: 6.7
total body mass) 8.10 6.94
R. Dumas and J. Wojtusch
(continued)
Table 1 (continued)
(Dempster 1955; Winter (de Leva 1996a; (Dumas et al. 2007b; McConville et al. 1980;
The head with neck 2009) Zatsiorsky et al. 1990) Young et al. 1983)
Position of center of Anterior-posterior axis Assumed negligible N/A Female: 0.8 and male: 2.0
mass (% of the (X)
segment length) Superior-inferior axis Female: N/A and male: Female: 48.41 and Female: 55.9 and male: 53.4
(Y) 43.3a male: 50.02
Medial-lateral axis (Z) Assumed negligible Assumed negligible Female: 0.1 and male: 0.1
Moment of inertia Anterior-posterior axis Assumed equal to Female: 29.5 and male: Female: 30 and male: 28
(radius of gyration in (X) medial-lateral axis 31.5
% of the segment Superior-inferior axis N/A Female: 26.1 and male: Female: 24 and male: 21
length) (Y) 26.1
Medial-lateral axis (Z) Female: N/A and male: Female: 27.1 and male: Female: 31 and male: 30
28.1a 30.3
Product of inertia Sagittal plane (X,Y) Assumed negligible Assumed negligible Female: 5(i) and male: 7(i)
Transverse plane (X, Assumed negligible Assumed negligible Female: 1 and male: 2(i)
Z)
Frontal plane (Y,Z) Assumed negligible Assumed negligible Female: 0 and male: 3
a
Value of Winter (2009) replaced by original value of Dempster (1955), radius of gyration worked out with a rule of 3
9
Table 2 Thorax
10
(Dempster 1955; (de Leva 1996a; Zatsiorsky (Dumas et al. 2015; McConville et al. 1980; Young
The thorax Winter 2009) et al. 1990) et al. 1983)
Segment endpoints Cervical joint center Skin landmark on the Cervical joint center (see head with neck segment)
(see head with neck seventh cervical vertebra Thoracic joint center estimated by the regression
segment) projected on superior- equations (Dumas et al. 2015)
Thoracic joint center inferior axis Female: thoracic joint center on a direction forming
obtained by dissection Skin landmark on xiphoid an angle of 92 in the sagittal plane with the vector
(center of T12-L1 projected on superior- from 12th to 8th thoracic vertebra skin landmarks
disc) inferior axis and at 50% of thorax width (see head with neck
segment) from 12th thoracic vertebra skin landmark
Male : angle of 94 and percentage of 52%
Segment reference Superior-inferior axis Superior-inferior axis: N/A Superior-inferior axis from thoracic to cervical joint
frame from thoracic to Frontal plane parallel to the center
cervical joint center gamma-ray scanner Sagittal plane containing skin landmarks on 7th
Sagittal plane set as acquisition plane cervical and 8th thoracic vertebra and suprasternal
plane of oscillation of Origin at 7th cervical notch
the pendulum method vertebra skin landmark Origin at cervical joint center
Origin at cervical joint projected on superior-
center inferior axis
Segment length (mm) N/A Female: 228.0 and male: Female: 322 and male: 334
242.1
(g/cm3)
Segment mass (% of Female: N/A and male: Female: 15.45 and male: Female: 26.3 and male: 30.4
total body mass) 21.60 15.96
Position of center of Anterior-posterior Assumed negligible N/A Female: 1.5 and male: 0.0
mass (% of the axis (X)
segment length) Superior-inferior Female: N/A and Female: 50.50 and male: Female: 54.2 and male: 55.5
axis (Y) male: 62.7a 50.66
Medial-lateral Assumed negligible Assumed negligible Female: 0.1 and male: 0.4
axis (Z)
(continued)
Table 2 (continued)
(Dempster 1955; (de Leva 1996a; Zatsiorsky (Dumas et al. 2015; McConville et al. 1980; Young
The thorax Winter 2009) et al. 1990) et al. 1983)
Moment of inertia Anterior-posterior N/A Female: 31.4 and male: Female: 38 and male: 42
(radius of gyration in axis (X) 32.0
% of the segment Superior-inferior N/A Female: 44.9 and male: Female: 32 and male: 33
length) axis (Y) 46.5
Medial-lateral N/A Female: 46.6 and male: Female: 34 and male: 36
axis (Z) 50.5
Product of inertia Sagittal plane (X, N/A N/A Female: 12(i) and male: 11(i)
Y)
Transverse plane Assumed negligible Assumed negligible Female: 3(i) and male: 1
(X,Z)
Frontal plane (Y, Assumed negligible Assumed negligible Female: 1 and male: 3
Z)
a
Value of Winter (2009) replaced by original value of Dempster (1955)
11
Table 3 Abdomen
12
(de Leva 1996a;

(Dempster 1955; Plagenhoef Zatsiorsky et al. (Dumas et al. 2015; McConville et al. 1980;
The abdomen 1971; Winter 2009) 1990) Young et al. 1983)
Segment Thoracic joint center (see thorax Skin landmark on Thoracic joint center (see thorax segment)
endpoints segment) xiphoid projected Lumbar joint center estimated by the regression
Inferior endpoint: N/A on superior-inferior equations (Dumas et al. 2007a)
axis Female: in the pelvis reference frame (see pelvis
Skin landmark on segment) with origin translated at midpoint
omphalion between anterior-superior iliac spine skin
projected on landmarks, lumbar joint center at 34.0%,
superior-inferior 4.9% and 0% of pelvis width (see pelvis
axis segment) about the anterior-posterior, superior-
inferior, and medial-lateral axes, respectively
Male: percentages of 33.5%%, 3.2%, and
0.0%, respectively
Segment Superior-inferior axis: N/A Superior-inferior Superior-inferior axis from thoracic to lumbar
reference frame Sagittal plane of the cardboard- axis: N/A joint center
lead cutouts modeling the Frontal plane Sagittal plane: N/A (no axial rotation at lumbar
anthropometry and mass parallel to the joint center assumed)
distribution of the trunk gamma-ray scanner Origin at thoracic joint center
Origin at thoracic joint centera acquisition plane
Origin at xiphoid
skin landmark
projected on
superior-inferior
axis
Segment N/A Female: 205.3 and Female: 125 and male: 151
length (mm) male: 215.5
Segment N/A N/A Assumed to be 1
density (g/cm3)
(continued)
Table 3 (continued)
(de Leva 1996a;
(Dempster 1955; Plagenhoef Zatsiorsky et al. (Dumas et al. 2015; McConville et al. 1980;
The abdomen 1971; Winter 2009) 1990) Young et al. 1983)
Segment mass Female: N/A and male: 13.90 Female: 14.65 and Female: 4.1 and male: 2.9
(% of total body male: 16.33
mass)
Position of Anterior-posterior axis (X) N/A N/A Female: 21.9 and male: 17.6
center of mass Superior-inferior axis (Y) Female: N/A and male: 34.6a Female: 45.12 Female: 41.0 and male: 36.1
(% of the and male: 45.02
segment length) Medial-lateral axis (Z) Assumed negligible Assumed negligible Female: 0.3 and male: 3.3
Moment of Anterior-posterior axis (X) N/A Female: 35.4 and Female: 65 and male: 54
inertia (radius male: 38.3
of gyration in % Superior-inferior axis (Y) N/A Female: 41.5 and Female: 78 and male: 66
of the segment male: 46.8
length) Medial-lateral axis (Z) N/A Female: 43.3 and Female: 52 and male: 40
male: 48.2
Product of Sagittal plane (X,Y) N/A N/A Female: 25 and male: 11
inertia Transverse plane (X,Z) Assumed negligible Assumed negligible Female: 3(i) and male: 6(i)
Frontal plane (Y,Z) Assumed negligible Assumed negligible Female: 5(i) and male: 5(i)
a
Adapted from Plagenhoef (1971), position of center of mass expressed as percentage of the length of the abdominopelvic segment (thoracic joint center to the
midpoint between the hip joint centers)
13
Table 4 Pelvis
14
(Dempster 1955; Plagenhoef (Dumas et al. 2007a; McConville et al.

The pelvis 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) 1980; Young et al. 1983)
Segment Superior endpoint: N/A Skin landmark on omphalion projected on Lumbar joint center (see abdomen
endpoints Midpoint between hip joint superior-inferior axis segment)
centers obtained by Midpoint between hip joint centers Midpoint between hip joint centers
dissection (acetabulum estimated by the regression equations estimated by the regression equations
center) (de Leva 1996b) (Dumas et al. 2007a)
Female: N/A Female: in the pelvis reference frame (see
Male: on superior-inferior axis, hip joint below) with origin translated at midpoint
center at 0.7% of thigh length (i.e., between anterior-superior iliac spine skin
longitudinal distance between tibial and landmarks, right/left hip joint center at
greater trochanter skin landmarks) from the 13.9%, 33.6%, and +/37.2% of pelvis
greater trochanter skin landmark width (i.e., distance between left and right
(percentages on anterior-posterior and anterior-superior iliac spine skin
media-lateral axes: N/A) landmarks) about the anterior-posterior,
superior-inferior, and medial-lateral axes,
respectively
Male: percentages of 9.5%, 37.0%, and
+/36.1%, respectively
Segment Superior-inferior axis: N/A Superior-inferior axis: N/A Medial-lateral axis from left to right
reference frame Sagittal plane of the Frontal plane parallel to the gamma-ray anterior superior iliac spine skin landmarks
cardboard-lead cutouts scanner acquisition plane Transverse plane containing skin
modeling the anthropometry Origin at omphalion skin landmark landmarks on left and right anterior-
and mass distribution of the projected on superior-inferior axis superior iliac spines and midpoint between
trunk posterior-superior iliac spines
Origin at midpoint between Origin at lumbar joint center
hip joint centers
Segment length N/A Female: 181.5 and male: 145.7 Female: 103 and male: 93
(mm)
Segment N/A N/A Assumed to be 1
density (g/cm3)
(continued)
Table 4 (continued)
The pelvis 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) 1980; Young et al. 1983)
Segment mass Female: N/A and male: 14.20 Female: 12.47 and male: 11.17 Female: 14.7 and male: 14.2
(% of total body
mass)
Position of Anterior- N/A N/A Female: 7.2 and male: 0.2
center of mass posterior
(% of the axis (X)
segment length) Superior- Female: N/A and male: Female: 49.20 and male: 61.15 Female: 22.8 and male: 28.2
inferior 15.6a
axis (Y)
Medial- Assumed negligible Assumed negligible Female: 0.2 and male: 0.6
lateral
axis (Z)
Moment of Anterior- N/A Female: 40.2 and male: 55.1 Female: 95 and male: 102
inertia (radius posterior
of gyration in % axis (X)
of the segment Superior- N/A Female: 44.4 and male: 58.7 Female: 105 and male: 106
length) inferior
axis (Y)
Medial- N/A Female: 43.3 and male: 61.5 Female: 82 and male: 96
lateral
axis (Z)
Product of Sagittal N/A N/A Female: 35(i) and male: 25(i)

inertia plane (X,Y)
Transverse Assumed negligible Assumed negligible Female: 3(i) and male: 12(i)
plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 2(i) and male: 8(i)
plane (Y,Z)
a
15
Adapted from Plagenhoef (1971), position of center of mass expressed as percentage of the length of the abdominopelvic segment (thoracic joint center to the
midpoint between the hip joint centers)
16
Table 5 Upper arm

The upper arm 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) 1980; Young et al. 1983)
Segment Glenohumeral joint center Glenohumeral and elbow joint centers Glenohumeral joint center estimated by
endpoints obtained by dissection (center of estimated by the regression equations the regression equations (Dumas et al.
curvature of humeral head) (de Leva 1996b) 2007a)
Elbow joint center obtained by Female: N/A Elbow joint center estimated at midpoint
dissection (axis of humeral Male: on superior-inferior axis, between lateral and medial humeral
trochlea at narrowest cross glenohumeral joint center at 10.4% of epicondyle skin landmarks
section of ulnar articulation) upper arm length (i.e., longitudinal Female : glenohumeral joint center on a
distance between acromion and radial direction forming an angle of 5 in the
skin landmarks) from acromion skin sagittal plane with the vector from 7th
landmark and elbow joint center at 4.3% cervical vertebra to suprasternal notch
of upper arm length from radial skin skin landmarks and at 36% of thorax
landmark (percentages on anterior- width (see head with neck and thorax
posterior and media-lateral axes: N/A) segments) from the acromion skin
landmark
Male: angle of 11 and percentage of
33%
Segment Superior-inferior axis from Superior-inferior axis from elbow to Superior-inferior axis from elbow to
reference frame elbow to glenohumeral joint glenohumeral joint center glenohumeral joint center
center Frontal plane parallel to the gamma-ray Frontal plane containing glenohumeral
Sagittal plane set as plane of scanner acquisition plane joint center and skin landmarks on lateral
oscillation of the pendulum Origin at glenohumeral joint center and medial humeral epicondyles
method Origin at glenohumeral joint center
Origin at glenohumeral joint
center
Segment length Female: N/A and male: 286a Female: 275.1 and male: 281.7 Female: 251 and male: 277
(mm)
Segment 1.07 N/A Assumed to be 1
density (g/cm3)
(continued)
Table 5 (continued)
The upper arm 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) 1980; Young et al. 1983)
(% of total body
mass)
Position of Anterior- Assumed negligible Assumed negligible Female: 5.5 and male: 1.8
(% of the axis (X)
segment length) Superior- Female: N/A and male: 43.6 Female: 57.54 and male: 57.72 Female: 50.0 and male: 48.2
inferior
axis (Y)
lateral
axis (Z)
Moment of Anterior- Assumed equal to medial-lateral Female: 26.0 and male: 26.9 Female: 30 and male: 29
inertia (radius posterior axis
length) inferior
axis (Y)
Medial- Female: N/A and male: 32.2 Female: 27.8 and male: 28.5 Female: 30 and male: 30
lateral
axis (Z)
Product of Sagittal Assumed negligible Assumed negligible Female: 3(i) and male: 5
inertia plane (X,
Y)
Transverse Assumed negligible Assumed negligible Female: 5 and male: 3
plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 3 and male: 13(i)
plane (Y,Z)
17
a
Value of Plagenhoef (1971), 16.9% of total height
18
Table 6 Forearm
(Dempster 1955;
Plagenhoef 1971; (Dumas et al. 2007a; McConville
The forearm Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1980; Young et al. 1983)
Segment Elbow joint center Elbow joint center (see upper arm segment) Elbow joint center (see upper arm
endpoints (see upper arm Wrist joint center estimated by the regression equations (de Leva segment)
segment) 1996b) Wrist joint center estimated at
Wrist joint center Female: N/A midpoint between radial and ulna
obtained by Male: on superior-inferior axis, wrist joint center at 0.6% of styloid skin landmarks
dissection (center forearm length (i.e., longitudinal distance between radial and
of curvature of radius styloid skin landmarks) from radius styloid skin landmark
proximal end of (percentages on anterior-posterior and media-lateral axes: N/A)
capitate bone)
Segment Superior-inferior Superior-inferior axis from wrist to elbow joint center Superior-inferior axis from wrist to
reference frame axis from wrist to Frontal plane parallel to the gamma-ray scanner acquisition elbow joint center
elbow joint center plane Frontal plane containing elbow joint
Sagittal plane set as Origin at elbow joint center center and skin landmarks on radial
plane of oscillation and ulna styloids
of the pendulum Origin at elbow joint center
method
Origin at elbow
joint center
Segment length Female: N/A and Female: 264.3 and male: 268.9 Female: 247 and male: 283
(mm) male: 269a
density (g/cm3)
Segment mass Female: N/A and Female: 1.38 and male: 1.62 Female: 1.4 and male: 1.7
(% of total body male: 1.60
mass)
(continued)
Table 6 (continued)
(Dempster 1955;
Plagenhoef 1971; (Dumas et al. 2007a; McConville
The forearm Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1980; Young et al. 1983)
(% of the axis (X)
segment length) Superior- Female: N/A and Female: 45.59 and male: 45.74 Female: 41.1 and male: 41.7
inferior male: 43.0
axis (Y)
lateral axis
(Z)
Moment of Anterior- Assumed equal to Female: 25.7 and male: 26.5 Female: 27 and male: 28
inertia (radius posterior medial-lateral axis
length) inferior
axis (Y)
Medial- Female: N/A and Female: 26.1 and male: 27.6 Female: 25 and male: 28
lateral male: 30.3
axis (Z)
Product of Sagittal Assumed negligible Assumed negligible Female: 10 and male: 8
inertia plane (X,Y)
Transverse Assumed negligible Assumed negligible Female: 3 and male: 1(i)

plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 13(i) and male: 2
plane (Y,Z)
a
19
20
Table 7 Hand
(Dempster 1955; Winter (de Leva 1996a; Zatsiorsky (Dumas et al. 2007a; McConville et al. 1980;
The hand 2009) et al. 1990) Young et al. 1983)
Segment endpoints Wrist joint center (see Wrist joint center (see Wrist joint center (see forearm segment)
forearm segment) forearm segment) Midpoint between 2nd and 5th metacarpal head
Interphalangeal knuckle of 3rd metacarpal skin skin landmarks
3rd finger skin landmark landmark projected on
superior-inferior axis
Segment reference frame Superior-inferior axis: N/A Superior-inferior axis: N/A Superior-inferior axis from midpoint between
Sagittal plane set as plane Frontal plane parallel to the 2nd and 5th metacarpal head skin landmarks to
of oscillation of the gamma-ray scanner wrist joint center
pendulum method acquisition plane Frontal plane containing wrist joint center and
Origin at wrist joint center Origin at wrist joint center skin landmarks on 2nd and 5th metacarpal heads
Origin at wrist joint center
86.2
Segment density (g/cm3) 1.17 N/A Assumed to be 1
Segment mass (% of total Female: N/A and male: Female: 0.56 and male: Female: 0.5 and male: 0.6
body mass) 0.60 0.61
Position of center of mass Anterior- N/A N/A Female: 7.7 and male: 8.2
(% of the segment length) posterior
axis (X)
Superior- Female: N/A and male: Female: 74.74 and male: Female: 76.8 and male: 83.9
inferior 50.6 79.00
axis (Y)
lateral
axis (Z)
(continued)
Table 7 (continued)
(Dempster 1955; Winter (de Leva 1996a; Zatsiorsky (Dumas et al. 2007a; McConville et al. 1980;
The hand 2009) et al. 1990) Young et al. 1983)
Moment of inertia (radius of Anterior- N/A Female: 45.4 and male: Female: 64 and male: 61
gyration in % of the posterior 51.3
segment length) axis (X)
Superior- N/A Female: 33.5 and male: Female: 43 and male: 38
inferior 40.1
axis (Y)
Medial- Female: N/A and male: Female: 53.1 and male: Female: 59 and male: 56
lateral 29.7 62.8
axis (Z)
Product of inertia Sagittal N/A Assumed negligible Female: 29 and male: 22
plane (X,Y)
Transverse N/A Assumed negligible Female: 23 and male: 15
plane (X,Z)
Frontal N/A Assumed negligible Female: 28(i) and male: 20(i)
plane (Y,Z)
21
Table 8 Thigh
22
(Dempster 1955; Plagenhoef 1971; (Dumas et al. 2007a; McConville

The thigh Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1980; Young et al. 1983)
Segment Hip joint center obtained by Hip joint center (see pelvis segment) Hip joint center (see pelvis
endpoints dissection (center of curvature of Knee joint center estimated by the segment)
femoral head) regression equations (de Leva 1996b) Knee joint center estimated at
Knee joint center obtained by Female: N/A midpoint between midpoint
dissection (middle of a line through Male: on superior-inferior axis, knee joint between lateral and medial
the center of curvature of the center at 7.4% of thigh length (i.e., femoral epicondyle skin
posterior aspect of femoral condyles) longitudinal distance between greater landmarks
trochanter and tibial skin landmarks) from
tibial skin landmark (percentages on
anterior-posterior and media-lateral axes:
N/A)
Segment Superior-inferior axis from knee to Superior-inferior axis from knee to hip Superior-inferior axis from knee
reference frame hip joint center joint center to hip joint center
Sagittal plane set as plane of Frontal plane parallel to the gamma-ray Frontal plane containing hip joint
oscillation of the pendulum method scanner acquisition plane center and skin landmarks on
Origin at hip joint center Origin at hip joint center lateral and medial femoral
epicondyles
(mm)
(g/cm3)
(% of total body
mass)
(continued)
Table 8 (continued)
(Dempster 1955; Plagenhoef 1971; (Dumas et al. 2007a; McConville
The thigh Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1980; Young et al. 1983)
Position of center Anterior- Assumed negligible Assumed negligible Female: 7.7 and male: 4.1
of mass (% of the posterior
segment length) axis (X)
Superior- Female: N/A and male: 43.3 Female: 36.12 and male: 40.95 Female: 37.7 and male: 42.9
inferior
axis (Y)
lateral
axis (Z)
Moment of Anterior- Assumed equal to medial-lateral axis Female: 36.4 and male: 32.9 Female: 31 and male: 29
inertia (radius of posterior
gyration in % of axis (X)
the segment Superior- N/A Female: 16.2 and male: 14.9 Female: 19 and male: 15
length) inferior
axis (Y)
lateral
axis (Z)
Product of inertia Sagittal Assumed negligible Assumed negligible Female: 7(i) and male: 7
plane (X,Y)
plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 7(i) and male: 7(i)
plane (Y,Z)
a
23
24
Table 9 Shank
(Dumas et al. 2007a;
(Dempster 1955; Plagenhoef McConville et al. 1980; Young
The shank 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1983)
Segment Knee joint center Knee joint center (see thigh segment) Knee joint center (see thigh
endpoints (see thigh segment) Ankle joint center estimated by the regression segment)
Ankle joint center obtained by equations (de Leva 1996b) Ankle joint center estimated at
dissection (center of the area Male: on superior-inferior axis, ankle joint center at midpoint between lateral and
of the cut body of the talus) 3.2% of shank length (i.e., longitudinal distance medial malleolus skin
between tibial and sphyrion skin landmarks) from landmarks
sphyrion skin landmark (percentages on anterior-
posterior and media-lateral axes: N/A)
Segment Superior-inferior axis from Superior-inferior axis from ankle to knee joint center Superior-inferior axis from
reference frame ankle to knee joint center Frontal plane parallel to the gamma-ray scanner ankle to knee joint center
Sagittal plane set as plane of acquisition plane Frontal plane containing knee
oscillation of the pendulum Origin at knee joint center and ankle joint centers and the
method fibula head skin landmark
Origin at knee joint center Origin at knee joint center
(mm)
density (g/cm3)
(% of total body
mass)
(continued)
Table 9 (continued)
(Dumas et al. 2007a;
(Dempster 1955; Plagenhoef McConville et al. 1980; Young
The shank 1971; Winter 2009) (de Leva 1996a; Zatsiorsky et al. 1990) et al. 1983)
(% of the axis (X)
segment length) Superior- Female: N/A and male: 43.3 Female: 43.52 and male: 43.95 Female: 40.4 and male: 41.0
inferior
axis (Y)
lateral
axis (Z)
Moment of Anterior- Assumed equal to medial- Female: 26.3 and male: 24.6 Female: 28 and male: 28
inertia (radius posterior lateral axis
length) inferior
axis (Y)
lateral
axis (Z)
Product of Sagittal Assumed negligible Assumed negligible Female: 2 and male: 4(i)
inertia plane (X,Y)

plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 6 and male: 4
plane (Y,Z)
a
25
Table 10 Foot
26
(Dempster 1955; Winter (de Leva 1996a; Zatsiorsky (Dumas et al. 2007a; McConville et al.
The foot 2009) et al. 1990) 1980; Young et al. 1983)
Segment endpoints Heel skin landmarksa Heel skin landmark Ankle joint center (see shank segment)
Tip of 2nd toe skin Tip of longest toe skin Midpoint between 1st and 5th metatarsal
landmarksa landmark head skin landmarks
Segment reference frame Anterior-posterior axis from Anterior-posterior axis from Anterior-posterior axis from heel to
heel to tip of 2nd toe skin heel to tip of longest toe skin midpoint between 1st and 5th metatarsal
landmarks landmarks head skin landmarks
Sagittal plane set as plane of Frontal plane parallel to the Transverse plane containing skin
oscillation of the pendulum gamma-ray scanner landmarks on heel and 1st and 5th
method acquisition plane metatarsal heads
Origin at heel skin Origin at heel skin landmark
landmarka
258.1
Segment density (g/cm3) 1.09 N/A Assumed to be 1
Segment mass (% of total Female: N/A and male: 1.45 Female: 1.29 and male: 1.37 Female: 1.0 and male: 1.2
body mass)
Position of center of mass Anterior- Female: N/A and male: 42.9a Female: 40.14 and Male: Female: 38.2 and male: 50.2
(% of the segment length) posterior 44.15
axis (X)
Superior- Assumed negligible Assumed negligible Female: 30.9 and male: 19.9
inferior
axis (Y)
lateral
axis (Z)
(continued)
Table 10 (continued)
(Dempster 1955; Winter (de Leva 1996a; Zatsiorsky (Dumas et al. 2007a; McConville et al.
The foot 2009) et al. 1990) 1980; Young et al. 1983)
Moment of inertia (radius of Anterior- N/A Female: 13.9 and male: 12.4 Female: 24 and male: 22
gyration in % of the segment posterior
length) axis (X)
Superior- Assumed equal to medial- Female: 27.9 and male: 24.5 Female: 50 and male: 49
inferior lateral axis
axis (Y)
Medial- Female: N/A and male: 40.7a Female: 29.9 and male: 25.7 Female: 50 and male: 48
lateral
axis (Z)
Product of inertia Sagittal N/A N/A Female: 15(i) and male: 17
plane (X,Y)
Transverse N/A N/A Female: 9 and male: 11(i)
plane (X,Z)
Frontal Assumed negligible Assumed negligible Female: 5(i) and male: 0
plane (Y,Z)
a
Value of Winter (2009) replaced by original value of Dempster (1955), radius of gyration worked out with a rule of 3
27
The adaptation of the previously developed adjustment procedures to elderly

subjects has been recently proposed (Ho Hoang and Mombaur 2015) and may be
extended to other specific populations.
There is an increasing interest in using subject-specific BSIPs of the lower limb
indirectly measured by medical imaging (Bauer et al. 2007; Dao et al. 2012; Ganley
and Powers 2004; Sreenivasa et al. 2016; Taddei et al. 2012; Valente et al. 2014) for
gait analysis, especially in case of pathologic subjects (see “▶ Cross-platform
Comparison of Imaging Technologies for Measuring Musculoskeletal Motion”).
Here again, the issue of matching segment definition used for the BSIP assessment
with the biomechanical model used for motion analysis exists. The segment defini-
tion and model construction are generally assumed consistent or anatomical land-
marks are virtually palpated on the MRI or CT scan reconstructions (Dao et al. 2012;
Sreenivasa et al. 2016; Valente et al. 2014). Occasionally, skin markers are placed on
the subject before CT scan and gait analysis, both performed consecutively (Taddei
et al. 2012). Nevertheless, the main purpose of these cumbersome personalization
methods was generally focussed on bone and muscle geometry.
The introduction of skin markers used for motion analysis within the procedure of
the BSIPs assessment was more widely adopted with photogrammetry (Davidson
et al. 2008; Pillet et al. 2010; Verriest 2012). This means that calibrated photographs
are taken during a static posture (just before or after motion analysis) allowing to
build a 3D volume model of the subject, straightforwardly registered with the rigid
body biomechanical model.
Moreover, without additional calibrated photographs and not performed during a
static posture but a dynamic movement, the BSIPs can be estimated by identification
methods (Ayusawa et al. 2014; Jovic et al. 2016; Vaughan et al. 1982). In this case, a
rigid body biomechanical model and the equations of motion are directly used to
compute the BSIPs that minimize the errors between model-derived and measured
ground reaction forces and moments. However, as for the functional calibration of
the joint centers and axes, these methods may require dedicated movements. With
that respect, a method that simultaneously identifies BSIPs, joint centers, and
segment lengths has been recently proposed (Bonnet et al. 2017). This method,
based on extended Kalman filters, minimizes the errors between model-derived and
measured ground reaction forces and moments as well as skin marker trajectories.
Other advanced methods for inverse dynamics (i.e., residual elimination/reduction
algorithms) typically alter the generalized accelerations of the model but may also
include either some (Delp et al. 2007) or all of the BSIPs (Jackson et al. 2015) as
design variables of the minimization process.
Most of the abovementioned methods for BSIPs assessment involve numerous
skin markers, especially when all the segments of the human body (e.g., the
16 segments previously mentioned) are of interest. Therefore, another direction is
to reduce the number of skin markers at its minimum while estimating the center of
mass of the human body, typically for the analysis of the dynamic stability. Adapted
segment definition, regression equations, and marker set have been dedicated for
such specific applications (Tisserand et al. 2016; Yang and Pai 2014).
References
Ackland TR, Blanksby BA, Bloomfield J (1988) Inertial characteristics of adolescent male body
segments. J Biomech 21(4):319–327. doi:10.1016/0021-9290(88)90261-8
Ayusawa K, Venture G, Nakamura Y (2014) Identifiability and identification of inertial parameters
using the underactuated base-link dynamics for legged multibody systems. Int J Robot Res
33(3):446–468. doi:10.1177/0278364913495932
Bauer JJ, Pavol MJ, Snow CM, Hayes WC (2007) MRI-derived body segment parameters of
children differ from age-based estimates derived using photogrammetry. J Biomech
40(13):2904–2910. doi:10.1016/j.jbiomech.2007.03.006
Bonnet V, Dumas R, Cappozzo A, Joukov V, Daune G, Kulić D, Fraisse P, Andary S, Venture G
(2017) A constrained extended Kalman filter for the optimal estimate of kinematics and kinetics
of a sagittal symmetric exercise. J Biomech. doi:10.1016/j.jbiomech.2016.12.027
Chandler RF, Clauser CE, McConville JT, Reynolds HM, Young JW (1975) Investigation of inertial
properties of the human body. Aerospace Medical Research Laboratory, Wright-Patterson Air
Force Base, Dayton
Cheng C-K, Chen H-H, Chen C-S, Lee C-L, Chen C-Y (2000) Segment inertial properties of
Chinese adults determined from magnetic resonance imaging. Clin Biomech 15(8):559–566.
doi:10.1016/S0268-0033(00)00016-4
Clauser CE, McConville JT, Young JW (1969) Weight, volume, and center of mass of segments of
the human body. Aerospace Medical Research Laboratory, Wright–Patterson Air Force Base,
Dayton
Dao TT, Marin F, Pouletaut P, Charleux F, Aufaure P, Ho Ba Tho MC (2012) Estimation of accuracy
of patient-specific musculoskeletal modelling: case study on a post polio residual paralysis
subject. Comput Methods Biomech Biomed Eng 15(7):745–751. doi:10.1080/
10255842.2011.558086
Davidson PL, Wilson SJ, Wilson BD, Chalmers DJ (2008) Estimating subject-specific body
segment parameters using a 3-dimensional modeller program. J Biomech 41(16):3506–3510.
doi:10.1016/j.jbiomech.2008.09.021
Davis RB III, Õunpuu S, Tyburski D, Gage JR (1991) A gait analysis data collection and reduction
technique. Hum Mov Sci 10(5):575–587. doi:10.1016/0167-9457(91)90046-Z
de Leva P (1996a) Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J Biomech
29(9):1223–1230. doi:0.1016/0021-9290(95)00178-6
de Leva P (1996b) Joint center longitudinal positions computed from a selected subset of Chan-
dler’s data. J Biomech 29(9):1231–1233. doi:10.1016/0021-9290(96)00021-8
OpenSim: Open-source software to create and analyze dynamic simulations of movement. IEEE
Trans Biomed Eng 54(11):1940–1950. doi:10.1109/TBME.2007.901024
Dempster WT (1955) Space requirements for the seated operator. Wright Air Development Center,
Wright-Patterson Air Force Base, Dayton
Drillis R, Contini R, Bluestein M (1964) Body segment parameters: a survey of measurement
techniques. Artif Limbs 8(1):44–66
Dumas R, Aissaoui R, Mitton D, Skalli W, de Guise JA (2005) Personalized body segment
parameters from biplanar low-dose radiography. IEEE Trans Biomed Eng 52(10):1756–1763.
doi:10.1109/TBME.2005.855711
Dumas R, Cheze L, Verriest JP (2007a) Adjustments to McConville et al. and Young et al. body
segment inertial parameters. J Biomech 40(3):543–553. doi:10.1016/j.jbiomech.2006.02.013
Dumas R, Cheze L, Verriest JP (2007b) Corrigendum to “Adjustments to McConville et al. and
Young et al. body segment inertial parameters”. J Biomech 40(7):1651–1652. doi:10.1016/j.
Dumas R, Robert T, Cheze L, Verriest J-P (2015) Thorax and abdomen body segment inertial
parameters adjusted from McConville et al. and Young et al. Int Biomech 2(1):113–118.
doi:10.1080/23335432.2015.1112244
Durkin JL, Dowling JJ, Andrews DM (2002) The measurement of body segment inertial parameters
using dual energy X-ray absorptiometry. J Biomech 35(12):1575–1580. doi:10.1016/S0021-
9290(02)00227-0
Ganley KJ, Powers CM (2004) Determination of lower extremity anthropometric parameters using
dual energy X-ray absorptiometry: the influence on net joint moments during gait. Clin Biomech
19(1):50–56. doi:10.1016/j.clinbiomech.2003.08.002
Hinrichs RN (1990) Adjustments to the segment center of mass proportions of Clauser et al. (1969).
J Biomech 23(9):949–951. doi:10.1016/0021-9290(90)90361-6
Ho Hoang K-L, Mombaur K (2015) Adjustments to de Leva-anthropometric regression data for the
changes in body proportions in elderly humans. J Biomech 48(13):3732–3736. doi:10.1016/j.
Jackson JN, Hass CJ, Fregly BJ (2015) Residual elimination algorithm enhancements to improve
foot motion tracking during forward dynamic simulations of Gait. J Biomech Eng
137(11):111002. doi:10.1115/1.4031418
Jensen RK (1978) Estimation of the biomechanical properties of three body types using a photo-
grammetric method. J Biomech 11(8-9):349–358. doi:10.1016/0021-9290(78)90069-6
Jovic J, Escande A, Ayusawa K, Yoshida E, Kheddar A, Venture G (2016) Humanoid and human
inertia parameter identification using hierarchical optimization. IEEE Trans Robot
32(3):726–735. doi: 10.1109/TRO.2016.2558190
during level walking. J Orthop Res 8(3):383–392. doi:10.1002/jor.1100080310
McConville JT, Churchill TD, Kaleps I, Clauser CE, Cuzzi J (1980) Anthropometric relationships
of body and body segment moments of inertia. Aerospace Medical Research Laboratory,
Wright-Patterson Air Force Base, Dayton
Mungiole M, Martin PE (1990) Estimating segment inertial properties: comparison of magnetic
resonance imaging with existing methods. J Biomech 23(10):1039–1046. doi:10.1016/0021-
9290(90)90319-X
Pearsall DJ, Reid G (1994) The study of human body segment parameters in biomechanics. Sports
Med 18(2):126–140. doi:10.2165/00007256-199418020-00005
Pearsall DJ, Reid JG, Livingston LA (1996) Segmental inertial parameters of the human trunk as
determined from computed tomography. Ann Biomed Eng 24(2):198–210. doi:10.1007/
BF02667349
Pillet H, Bonnet X, Lavaste F, Skalli W (2010) Evaluation of force plate-less estimation of the
trajectory of the centre of pressure during gait. Comparison of two anthropometric models. Gait
Plagenhoef S (1971) Patterns of human motion: a cinematographic analysis. Prentice-Hall, Engle-
wood Cliffs
Reid JG, Jensen RK (1990) Human body segment inertia parameters: a survey and status report.
Exerc Sport Sci Rev 18(1)
Sreenivasa M, Chamorro CJG, Gonzalez-Alvarado D, Rettig O, Wolf SI (2016) Patient-specific
bone geometry and segment inertia from MRI images for model-based analysis of pathological
gait. J Biomech 49(9):1918–1925. doi:10.1016/j.jbiomech.2016.05.001
Taddei F, Martelli S, Valente G, Leardini A, Benedetti MG, Manfrini M, Viceconti M (2012)
Femoral loads during gait in a patient with massive skeletal reconstruction. Clin Biomech
27(3):273–280. doi:10.1016/j.clinbiomech.2011.09.006
Tisserand R, Robert T, Dumas R, Chèze L (2016) A simplified marker set to define the center of
mass for stability analysis in dynamic situations. Gait Posture 48:64–67. doi:10.1016/j.
Valente G, Pitto L, Testi D, Seth A, Delp SL, Stagni R, Viceconti M, Taddei F (2014) Are subject-
specific musculoskeletal models robust to the uncertainties in parameter identification? PLoS
One 9(11):e112625. doi:10.1371/journal.pone.0112625
Vaughan CL, Andrews JG, Hay JG (1982) Selection of body segment parameters by optimization
methods. J Biomech Eng 104(1):38–44. doi:10.1115/1.3138301
Verriest JP (2012) Automatic anthropometric personalization of a digital human model from a set of
subject’s photographs. Work 41(Suppl 1):4061–4068. doi:10.3233/WOR-2012-0071-4061
joint coordinate system of various joints for the reporting of human joint motion – part I: ankle,
hip, and spine. J Biomech 35(4):543–548. doi:10.1016/S0021-9290(01)00222-6
Wu G, van der Helm FCT, Veeger HEJ, Makhsous M, Van Roy P, Anglin C, Nagels J, Karduna AR,
McQuade K, Wang X, Werner FW, Buchholz B (2005) ISB recommendation on definitions of
joint coordinate systems of various joints for the reporting of human joint motion – part II:
shoulder, elbow, wrist and hand. J Biomech 38(5):981–992. doi:10.1016/j.jbiomech.2004.05.042
Yang F, Pai Y-C (2014) Can sacral marker approximate center of mass during gait and slip-fall
recovery among community-dwelling older adults? J Biomech 47(16):3807–3812. doi:10.1016/
j.jbiomech.2014.10.027
Yeadon MR, Morlock M (1989) The appropriate use of regression equations for the estimation of
segmental inertia parameters. J Biomech 22(6):683–689. doi:10.1016/0021-9290(89)90018-3
Young JW, Chandler RF, Snow CC, Robinette KM, Zehner GF, Lofberg MS (1983) Anthropometric
and mass distribution characteristics of the adults female. FAA Civil Aeromedical Institute,
Oklaoma City
Zatsiorsky VM, Seluyanov VN, Chugunova LG (1990) Methods of determining mass-inertial
characteristics of human body segments. In: Chernyi GG, Regirer SA (eds) Contemporary
problems of biomechanics. CRC Press, Massachusetts, pp 272–291
3D Musculoskeletal Kinematics Using
Dynamic MRI
Frances T. Sheehan and Richard M. Smith
Abstract
Until the early 1990s, the tools available to measure musculoskeletal motion were
typically highly invasive. Thus, knowledge of musculoskeletal system dynamics
was primarily derived through cadaver and modeling experiments. The rapid
development of dynamic magnetic resonance (MR) imaging techniques changed
this and opened vast new opportunities for the study of 3D musculoskeletal
dynamics during volitional activities. Today, dynamic MR methodologies remain
the only techniques that can noninvasively track in vivo 3D musculoskeletal
movement.
One difficulty in applying these dynamic MR techniques to the study of
musculoskeletal motion is the complex interplay of parameters that affect the
spatial/temporal resolution, accuracy, and precision. The purpose of this chapter
is to first provide an explanation of the fundamental principles behind two of
these dynamic imaging techniques, cine and cine phase-contrast MR. Tagged cine
MR is another technique that has been primarily used to track muscle motion and
strain but will not be addressed. In doing so, this will create a platform for future
experimental designs using dynamic MR. This will be followed by a review of the
accuracies, the advantages, and disadvantages of the these dynamic MR methods.
Finally, several previously published studies will be highlighted to provide an
explanation of how these techniques can be applied and what main challenges
must be considered for future experiments using dynamic MR.
Keywords
MR • Magnetic resonance imaging • Cine • Cine phase contrast • Fastcard • CPC •
Fast-PC • Musculoskeletal • Kinematics • Strain • Moment arms • Tendon
F.T. Sheehan (*) • R.M. Smith

Rehabilitation Medicine Department, Functional & Applied Biomechanics Section, National
Institutes of Health, Bethesda, MD, USA
e-mail: fsheehan@cc.nih.gov; gavellif@cc.nih.gov; richard.smith5@nih.gov

DOI 10.1007/978-3-319-30808-1_155-1
2 F.T. Sheehan and R.M. Smith
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Static MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Dynamic MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Cine MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Fastcard (Fast Cine) MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Fast Cine: Phase-Contrast (Fast-PC) MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Advanced Applications of Dynamic MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Combining Fastcard Imaging with 3D Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Applications of CPC and Fast-PC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Introduction
Until the early 1990s, the tools available to measure musculoskeletal motion were
typically highly invasive (Regev et al. 2011; Edsfeldt et al. 2015; Lafortune et al.
1994; van Kampen and Huiskes 1990; Manal et al. 2000). Thus, knowledge of
musculoskeletal system dynamics was primarily derived through cadaver and
modeling experiments. The rapid development of dynamic magnetic resonance
(MR) imaging techniques changed this and opened vast new opportunities for the
study of 3D musculoskeletal dynamics during volitional activities (Burnett et al.
1987; Drace and Pelc 1994; Sheehan et al. 1998; Sheehan and Drace 2000; Asakawa
et al. 2002; Pappas et al. 2002; Fujiwara et al. 2004; Finni et al. 2006; Cheng et al.
2008). Today, dynamic MR methodologies remain the only techniques that can
noninvasively track in vivo 3D musculoskeletal movement.
One difficulty in applying these dynamic MR techniques to the study of muscu-
loskeletal motion is the complex interplay of parameters that affect the spatial/
temporal resolution, accuracy, and precision. The purpose of this chapter is to first
provide an explanation of the fundamental principles behind two of these dynamic
imaging techniques, cine (Glover and Pelc 1988) and cine phase-contrast (Feinberg
et al. 1984; Wedeen et al. 1985; Pelc et al. 1991b, 1994) MR. Tagged cine MR is
another technique that has been primarily used to track muscle motion and strain but
will not be addressed (Moerman et al. 2012). In doing so, this will create a platform
for future experimental designs using dynamic MR. This will be followed by a
review of the accuracies, the advantages, and disadvantages of the these dynamic
MR methods. Finally, several previously published studies will be highlighted to
provide an explanation of how these techniques can be applied and what main
challenges must be considered for future experiments using dynamic MR.
3D Musculoskeletal Kinematics Using Dynamic MRI 3
State of the Art
The earliest use of MR technology was isolated to spectroscopy (NRM or nuclear

magnetic resonance). Once researchers realized how to manipulate the MR signal to
produce in vivo images of the human tissue, the technology was rapidly transferred
to the clinical domain in the early 1980s, providing a wealth of new information. A
primary drawback of this new technology was the acquisition time for a single image
prohibited its use in imaging moving tissue, particularly the heart. Cine MR was
developed to acquire anatomic images of moving tissue during a cyclic movement
sequence by synchronizing the data collection to the cardiac cycle. This was rapidly
followed by the development of tagged MR and cine phase-contrast (CPC) MR,
which allowed the analytical tracking of tissue. Tagged MR has remained a tool
primarily used for cardiac imaging, but both cine and CPC MR have been widely
applied to the noninvasive study of 3D musculoskeletal kinematics. The original
applications of cine and CPC MR to the study of musculoskeletal motions required
long acquisition times with a high number of repeated cycles. Currently, the accu-
racy of using CPC to track musculoskeletal motion has dropped to less than 0.3 mm.
In addition, the imaging times have dramatically dropped, enabling acquisitions of
less than a minute.
Static MR Imaging
MR imaging is distinct from other imaging techniques in two key ways. First, the
measured signal is generated from the tissue being imaged. MR imaging is based on
the principle that atoms with an odd number of neutrons and protons will precess
about an external magnetic field at a frequency directly proportional to the field
strength. Clinical MR imaging typically focusses on hydrogen atoms, often referred
to as “spins.” With the introduction of a perpendicular magnetic field at the same
frequency of precession, the spins will “tip” into the transverse plane, creating a
transverse magnetization. The MR signal is the transverse magnetization of all spins
within a voxel (the smallest volumetric unit of the image). As the signal for all voxels
is collected together in a single summed signal, the location of a voxel’s signal is
encoded in the phase and frequency of its signal using magnetic gradients (Riederer
1993). This encoding process results in another unique feature of MR imaging; the
data collected during a scan is not spatial but spatial frequency data (k-space data,
Fig. 1). More importantly, numerous data acquisitions are required to provide a
complete spatial frequency map (k-space map) that can be converted into an image
using inverse Fourier transforms. Various methods for collecting data in k-space
have been developed (Foo et al. 1995; Markl et al. 2003; Pike et al. 1994; Thompson
and McVeigh 2004; Thunberg et al. 2003), but the simplest, collecting a single line in
k-space (phase encode) per acquisition (Fig. 2), will be used as the example for this
chapter.
Fig. 1 K-space map to MR image. K-space map (left) of a 2D GRE sagittal image of the knee
(right). The majority of the data in the k-space map resides at the lower frequencies, which represent
the portions of the image where the contrast is not changing rapidly from one pixel to the next. For
example, the magnitude at kx = ky = 0 represent the parts of the 2D spatial image where the
contrast is not changing from pixel to pixel. One such region of low spatial frequency is shown on
the left by the number 1. The number 2 denotes a region of high spatial frequency in the anterior-
posterior direction
Fig. 2 K-space map

representing a linear phase
encode collection: in this
example, the frequency
direction is kx (anterior-
posterior) and the phase
direction is ky (superior-
inferior). For visual clarity, the
acquisition is represented with
very low frequency and phase
resolution (41 41). For this
example, 41 individual phase
encodes (each represented by
a unique color) are required to
produce one 2D spatial image
The speed at which a single phase encode can be captured is referred to as the
repetition rate (TR). Thus, the time needed to acquire an entire image is:
Imaging time ¼ #phases TR (1)
#phases = number of required phase encodes to create a single image

If data averaging is used, to reduce the effect of random noise, then the imaging
times increase linearly with the number of averages (NA) acquired:
Imaging time ¼ NA #phases TR (2)
As noted above (Eqs. 1 and 2), the imaging time is dependent on the number of
phase encodes acquired. For a fixed field of view, #phases directly determines the
spatial resolution of the final image. To put numbers to this, let us assume a
256 256 (frequency resolution #phases) acquisition. This would result in a
square image 256 pixels wide by 256 high. Assuming a TR = 5 ms and no data
averaging, the imaging time (Eq. 2) would be 1.28 s for a single 2D image. If we set
the spatial field of view at 200 200 mm, the spatial resolution becomes (200 mm/
256) = 0.78 mm/pixel.
Dynamic MR Imaging
Cine MR
When the first scanners became commercially available in the early 1980s, it was not
feasible to image the beating heart, because its period of motion was significantly
greater than the image acquisition time. As noted in the example above for a
256 256 acquisition with a TR of 5 ms, the imaging time (1.28 s) is longer than
the typical cardiac period of 1 s. A TR of 5 ms has only been feasible recently, when
cine MR was first being developed TRs above 20 ms were typical (Keegan et al.
1994; Sheehan et al. 1998). Cine MR (Waterton et al. 1985; Burnett et al. 1987;
Glover and Pelc 1988; Cadera et al. 1992) was developed to overcome this limita-
tion, with a particular focus on cardiac imaging. The underlying assumption of cine
MR is that the tissue being imaged is moving in a repeatable, cyclic motion path. By
gating the data collection to the cardiac cycle, the required phase encodes in k-space
(256 in our example) can be acquired over numerous cardiac cycles. Retrospective
gating with cine MR compensates for variations in the period of motion by repeat-
edly acquiring the same k-space data line every TR during a single cycle. A trigger is
used to detect when a new cycle begins, signaling a change in the k-space data line
being acquired (Fig. 3). After #phases cycles, the data are retrospectively interpo-
lated so that images are created at specified time intervals. Although any number of
image time frames can be created through the interpolation process, the true temporal
resolution is TR, whereas, the spatial resolution is dependent on #phases:
Fig. 3 Cine MR data capture: when the scanner receives a trigger that a new cycle has begun (black
line on red signal), the ky data line (phase encode) being acquired is incremented. In the above
example, all cycles are of even length and a true 16 frames is captured. The temporal resolution is
equal to the TR (time between data captures). Compensating for variations in the periods of each
cycle is accomplished with retrospective gating and interpolation to align the data to the correct
temporal location in the motion cycle
cine temporal resolution ¼ TR (3)
spatial resolution ¼ FOV=#phases (4)
Thus, with a TR of 5 ms, 200 true frames of data are acquired with a period of 1 s.
Other types of gating, including prospective, are available for cine MR. One issue
with prospective gating is that the entire cycle is typically not captured. The imaging
time for cine MR depends on the #phases and motion period (T) of the moving tissue
being imaged:
cine imaging time ¼ T #phases (5)
Following the example above, and assuming T = 1 s, the cine imaging

time = 1 s*256 = 4.3 min (Eq. 5), requiring 256 cycles of motion.
Although the cine techniques were motivated by a desire to image cardiac motion,
its applicability to musculoskeletal motion was quickly realized. In 1987, a study
using cine MR to visualize passive temporomandibular joint mechanics (Burnett
et al. 1987) was published. This was quickly followed by a qualitative study of active
ankle joint motion (Melchert et al. 1992) and a quantitative study of active knee joint
motion (Brossmann et al. 1993). More recently, cine MR has been used to evaluate
tongue movement (Stone et al. 2001), fetal motion (Verbruggen et al. 2016), and
scaphoid-lunate motion (Langner et al. 2015).
Fastcard (Fast Cine) MR
Fastcard (Foo et al. 1995), or fast cine MR, was developed to enable a trade-off
between temporal resolution and the number of cycles required to capture a full
image set. Instead of repeatedly capturing a single phase encode line per motion
cycle, fastcard collects multiple phase encodes during a single motion cycle. This
reduces the overall imaging time, with a degradation in temporal resolution:
fastcard imaging time ¼ T #phases=#views (6)
fastcard temporal resolution ¼ #views TR (7)
#views = the number of unique phase encode lines acquired during a single
motion cycle
For most joint motion studies using cine MR, the period of motion is longer than
that of a beating heart. Using the above example, but assuming a typical knee joint
extension/flexion type movement (Brossmann et al. 1993), with a motion period of
2 s, the 256 motion cycles would require 8.5 min to capture a cine MR image set.
Yet, 400 temporal frames (a temporal resolution of 5 ms) would be produced to
represent an arc of motion of approximately 40 . Using fastcard and assuming
20 views are collected during each motion cycle, the fastcard imaging time is just
26 s, requiring only 13 motion cycles. Here, only 20 true data frames are collected,
with a temporal resolution equal to 100 ms. This trade-off between temporal
resolution and number of required motion cycles afforded by fastcard is fundamental
to the design of experiments that can be repeatedly performed by both healthy
volunteers and individuals with musculoskeletal impairments/pathologies. Fastcard
also enables a trade-off between motion period and the number of required motion
cycles. Using the same example, if the period is increased by a factor of 2 (T = 4),
the number of views acquired during each motion cycle could be doubled
(40 views). Although the temporal resolution would increase by a factor of 2, the
speed of motion would reduce by the same amount. As such, no increase in temporal
blurring would be expected. This would reduce the number of required motion
cycles to just 7, but the overall acquisition time would still remain at ~26 s. It is
important to note that this example is assuming an acquisition matrix of 256 256.
Using the spatial parameters from a previous study (Carlson et al. 2016) of
patellofemoral kinematics (FOV = 180 mm and using a 75% phase acquisition),
the acquisition matrix can be dropped to 180 135 pixels and still maintain a
resolution of 1 1 mm. This would require just four motion cycles with an imaging
time of 16 s. Such a reduced number of required cycles could potentially allow for
studies with increased joint loading and is crucial to acquiring data during volitional
joint motion in individuals with musculoskeletal pathologies/impairments (Fig. 4).
Fast Cine: Phase-Contrast (Fast-PC) MR
CPC MR (Pelc et al. 1991b) combines the ability of phase-contrast imaging to

produce three-dimensional quantitative velocity data with cine’s ability to produce
a series of images throughout a gated motion cycle. Fast-PC MR (Foo et al. 1995) is
identical to CPC MR but inherits fastcard’s ability to capture multiple views, or
phase encode lines, during a single motion/cardiac cycle. Phase-contrast imaging
Fig. 4 Fastcard data acquisition: for this figure, five phase encode lines (#views = 5) are captured
per motion cycle. Thus, phase encode lines ky = 16 through ky = 20 are captured during the first
motion cycle. Temporal resolution is increased to 5*TR (the time between acquiring the same
k-space line is 5*TR). Yet, on the positive side, the imaging time and the number of cycles required
are also reduced by a factor of 5
(Pelc 1995; Pelc et al. 1991a) manipulates the concept that the MR signal from spins
moving in the direction of a magnetic gradient will accumulate phase proportional to
the first moment of that gradient. In MR imaging, magnetic gradient fields are used
in three perpendicular directions to isolate the slice being imaged (slice selection
gradient) and to encode signal’s location into the phase and frequency of the signal
(phase and frequency gradients) (Riederer 1993). Thus, for PC imaging, each of
these gradients is sequentially modified to enable data an acquisition that is sensitive
to velocity in the three perpendicular directions, frequency, phase, and slice (Fig. 5).
In addition, a fourth acquisition is used that is insensitive to motion, as a reference.
Thus, for each phase encode line in k-space, four acquisitions are needed to capture a
complete (velocity in three perpendicular direction) CPC data line. This increases the
overall temporal resolution by a factor of 4, with no reduction in imaging time or
number of cycles required:
FastPC temporal resolution ¼ 4 TR #views (8)
FastPC imaging time ¼ T #phases=#views (9)
The complete acquisition produces a temporal series of images representing the

anatomy within the imaging plane, plus images representing the velocity in the
frequency, phase, and slice (x, y, and z) directions (Figs. 5 and 6). Thus, following
from above, using a FOV = 180 mm, a 75% phase acquisition, a T = 2 s, a TR = 5 ms,
and 20 views, the temporal resolution would be 400 ms, allowing for only 5 true
frames of data. By reducing the #views to 5, the temporal resolution returns to
100 ms with 20 true frames of data. Yet, this comes at a cost in imaging time, which
would increase to 54 s.
Fig. 5 Acquiring fast-PC data. To capture the full fast-PC dataset, four separate data acquisitions
are needed. This provides measures of velocity in the frequency (x), phase (y), and slice
(z) directions. The temporal resolution is reduced by a factor of 4 (temporal resolution = 4 * TR)
and 5 true frames of data are available. If the #views was increased from 1 to 2, then the temporal
resolution would increase to 8*TR, with a reduction by a factor of 2 in the number of motion cycles
and the imaging time
A final key parameter when using fast-PC MR to track musculoskeletal motion is

the maximum velocity encoding, venc. As the velocity is encoded in the phase of the
signal, a phase shift of 180 is designated as the venc. Any velocity greater than 180
will create velocity aliasing. For example, a velocity producing a phase shift of 270
could not be distinguished from a negative velocity that produced a 90 phase
shift. Thus, keeping venc low will improve the velocity resolution, but potentially
could create aliasing. Further, any reduction in venc results in increases in TR, which
will negatively affect the temporal resolution.
For both cine and CPC imaging, any tissue that is moving in the imaging plane
that is not synchronized to the data capture will result in destructive noise (ghosts).
The largest sources of such noise when using dynamic MR to track motion are
inconsistent movement and blood flow within the image. The noise from the latter
can be greatly reduced using spatial pre-saturation of the blood (Im et al. 2015; Wood
and Wiang 1993). This pre-saturation destroys the MR signal in the blood prior to it
entering the imaging plane, thus eliminating (or greatly reducing) the noise it
generates.
Advanced Applications of Dynamic MR Imaging
Combining Fastcard Imaging with 3D Modeling
One disadvantage of cine MR is that quantifying musculoskeletal movement from

2D spatial images becomes less accurate when there is out-of-plane motion, as the
points of reference cannot be directly tracked throughout the entire movement,
which could lead to larger errors (Shibanuma et al. 2004, 2005). Yet, with 2D
fastcard imaging times approaching just 13 s and newer low-resolution cine imaging
becoming available (Kaiser et al. 2013), it is quite possible to capture a multiplane
Fig. 6 Fast-PC acquisition of the knee (Sheehan et al. 2012): from top to bottom, the rows
represent the anatomic images and the right-left (RL), the anterior-posterior (AP), and the
superior-inferior (SI) velocity images. In the velocity images, pure white represents +30 cm/s and
black represents 30 cm/s. The following parameters were used: TR = 5 ms, motion rate =
30 cycles/min; phase direction = AP; #views = 4; number of averages = 2; percent phase fov =
72%; 256 256 pixels; 200 mm fov, acquired resolution = 0.94 0.94 8 mm; reconstructed
resolution = 0.78 0.78 8 mm; maximum velocity encoding = 30 cm/s. The temporal resolu-
tion = 80 ms = 25 frames of data (Eq. 8). Imaging time = 1 min 30 s, 45 motion cycles (Eq. 9).
Note, the scan time and number of required cycles could easily be reduced by 50% if average was
not used, but all subjects could tolerate the scan time and the data averaging reduced random noise
in the images. The black edges seen on the right and left side of the images are due to the 75% phase
fov, which reduced the imaging time by 25%, without reducing the spatial resolution
fastcard (MPC) image set with limited scanning time. Scan times from 1.5 min
(45 cycles) (Borotikar et al. 2012) to 5 min (150 cycles) (Kaiser et al. 2016) have
been reported. From each time frame, a low-resolution model of the bone being
tracked can be created and fit to a static 3D model of the bone, providing a method to
consistently track the 3D motion of the bone (Fig. 7). This model fitting methodol-
ogy also provides a method for quantifying dynamic cartilage contact parameters
(Borotikar et al. 2012; Kaiser et al. 2016).
Model fitting using dynamic MR data has been applied in two ways. The first is
based on fitting the sparse dynamic model from a single time frame to a 3D static
model of the same bone (Borotikar et al. 2012). From a single time frame in the MPC
image set, a sparse model of the bone is created. Next, a high-resolution model of the
same bone is created from static images. The optimal rotation and translation that
minimize errors between the surfaces of the two models of the same bone are found.
Fig. 7 3D kinematics derived from combining fastcard imaging and 3D modeling. This example
set is based on a previous study (Borotikar et al. 2012), where fastcard imaging was used to produce
a 4D image set representing 24 evenly spaced temporal increments throughout the motion (first
column). Each 3D image (representing the anatomy at a single time frame) is comprised of seven
images. A sparse dynamic model of the bone (2nd column) is extracted from each of the 24, 3D
image sets. The femur is shown segmented, but this can be done for the patella and tibia as well.
From a high-resolution 3D image static image set, a high-resolution model of the bone is created
(third column). Using registration, typically iterative closest point algorithm, the rotation and
translation of the rough dynamic model that places the dynamic and static model into the best
alignment are calculated. This is independently accomplished at each time frame. These rotation
and translation are expressed as a matrix (STT1 = transformation required to bring the dynamic
model from time frame 1 into alignment with the static model). When this is completed for all time
frames, the 3D kinematic motion of the bone is known (T1TTi). Applying these transforms to rigid
models of the bone that include the cartilage enables an analysis of cartilage contact (fourth column)
Thus, the alignment of the 3D static model is now known for one dynamic time
frame. The 3D displacement and rotation of the bone, derived from the fast-PC data,
are then applied to the static model, so that its position is known for all time frames.
This was done for the patellofemoral joint with an accuracy of less than 0.9 mm
(average absolute error) reported (Borotikar et al. 2012). In addition, it is possible to
create the sparse model for every time frame of interest (Kaiser et al. 2016; Borotikar
et al. 2012). By fitting each sparse dynamic bone model to the 3D static bone model,
the 3D kinematics of the bone can be backed out from the transformation matrices
describing the sparse dynamic to high-resolution static model. This has been done
for the tibiofemoral joint, with an accuracy of than 0.60 mm (RMS error) (Kaiser
et al. 2016), and for the patellofemoral joint with an accuracy less than 1.3 mm
(average absolute error) (Borotikar et al. 2012). Ultimately, the final accuracy is
affected by the signal-to-noise ratio, the spatial resolution of the images, the size and
shape of the bone being tracked, and the type of motion being analyzed.
Applications of CPC and Fast-PC
The patellofemoral and tibiofemoral joint dynamics were the first to be studied using
CPC MR. With the addition of fast-PC, and continuously decreasing TRs, the time
for acquiring a full CPC dataset of the knee has dropped from over 7 min (Sheehan
et al. 1998) to just 1.5 min (Behnam et al. 2011). If data averaging had not been used,
both of these times would be halved. This reduction in time opens up the use of CPC,
as the number of required motion cycles dropped from just over 200 to 45 (motion
rate = 30 cycles/min). During this time, improvements in scanner strength and coil
design greatly enhanced the signal-to-noise ratio, which allowed an improved
accuracy. Based on a nearly identical phantom experiment, the average absolute
error for tracking both in- and out-of-plane motion dropped from a maximum
1.48 mm (Sheehan et al. 1998) to 0.33 mm (Behnam et al. 2011). A recent study
by Jensen and colleagues (Jensen et al. 2015) did an excellent job exploring the
various potential sources of error in tracking muscle strain with CPC, finding that the
bias (average error) in the velocity measures was below 1.3 mm/s.
In terms of tracking muscle, researchers have focused primarily on two types of
studies. One is to track a single point on the muscle in order to calculate moment
arms (Finni et al. 2006; Im et al. 2015; Sheehan 2012; Westphal et al. 2013; Wilson
and Sheehan 2009), tendon paths (Wilson and Sheehan 2010), and muscle excur-
sions (Wen et al. 2008). The other is to evaluate muscle deformation, particularly as a
marker of pathology (Finni et al. 2006; Kinugasa et al. 2008; Pappas et al. 2002;
Silder et al. 2010; Sinha et al. 2012; Zhou and Novotny 2007). A recent group has
shown the feasibility of evaluating three-dimensional strain (Jensen et al. 2016).
Unfortunately, the scan time for the 3D CPC images required for 3D strain measures
is unrealistically long to be applied to the evaluation of in vivo muscle strain during a
volitional activity. Yet, as scanners further improve, the available TRs continue to
drop, and new 3D fast MR data acquisition algorithms are developed; the scan time
will likely reduce enough to allow in vivo experiments.
Future Directions
CPC and fast-PC MR remain the only methodologies available to noninvasively

measure in vivo 3D musculoskeletal kinematics (Bey et al. 2008; Shih et al. 2003;
Fregly et al. 2005; Manal et al. 2000; Moro-oka et al. 2007; Yamashita et al. 2007).
The reported accuracy of tracking rigid motion (0.33 mm) is better than other
noninvasive techniques that measure in vivo skeletal kinematics (e.g., fluoroscopy
and motion capture) and muscle dynamics (e.g., ultrasound). It has an advantage
over x-ray-based techniques (e.g., fluoroscopy and 4D ultrasound) in that it does not
expose subjects to ionizing radiation. Quantifying kinematics by combining model
fitting with fastcard data is less accurate (Kaiser et al. 2016; Borotikar et al. 2012)
than fast-PC imaging but does provide the opportunity to quantify cartilage contact.
The accuracy of tracking musculoskeletal motion from a fast-PC acquisition is
dependent on the strength of the magnet, the consistency of movement, the quality of
the scanner, the TR, the number of data averages, the fov, the venc, and the signal-to-
noise ratio within the imaging plane. The integration algorithms used to track the
motion can help compensate for some of the errors due to noise and systematic errors
from the MR scanner (Jensen et al. 2015; Pelc et al. 1995; Zhu et al. 1996). Thus,
relying on past validation studies does provide a general framework for the accuracy
of tracking musculoskeletal motion, but a validation based on the specific parameters
being used for a particular study should be done in order to insure that the combi-
nation of parameters used leads to accurate tracking.
As with all measurement techniques, dynamic MR imaging has its limitations.
First, it relies on a costly imaging modality that is not readily available to all
clinicians and researchers. For the most accurate data, the majority of studies use
high field strength, closed-bore, MR units, which limits the types of movements that
can be studied. Open-bore units are available and some cine studies have used this
technology, but currently the images produced are of an inferior quality. Thus, as
open-bore technology improves, the types of functional movements that can be
evaluated will expand. Lastly, the dynamic MR techniques rely on repetitive move-
ments. As acquisition methods and scanners continue to improve, this limitation is
quickly being removed, and real-time dynamic MR imaging is becoming a reality.
Another limitation is that the integration routines to track musculoskeletal motion are
not widely available for all researchers. Yet, this will likely rapidly change in the next
few years.
Future advancements in various aspects of MRI technology will help expand the
application of dynamic MR in the quantification of 3D musculoskeletal dynamics.
Improvements in scanner and coil design will support a wider range of tasks that can
be studied. Eventually, this will likely include the full range of motion for joints such
as the shoulder, as well as the ability to study the kinematics of multiple joints during
dynamic tasks. The development and sharing of integration algorithms and packages
for CPC will support its expanded use in both the research and clinical setting. As the
accuracy of real-time MR imaging reaches the level of the current CPC techniques,
the types of motions and pathologies that can be studied will rapidly expand.
Acknowledgments This work was funded by the Intramural Research Program of the National
Institutes of Health Clinical Center, Bethesda, MD, USA. This research was also made possible
through the NIH Medical Research Scholars Program, a public-private partnership (http://fnih.org).
We thank Judith Welsh for her help and support in the work.
Cross-References
▶ Cross-Platform Comparison of Imaging Technologies for Measuring Musculo-

skeletal Motion
References
Asakawa DS, Blemker SS, Gold GE, Delp SL (2002) In vivo motion of the rectus femoris muscle
after tendon transfer surgery. J Biomech 35(8):1029–1037
Behnam AJ, Herzka DA, Sheehan FT (2011) Assessing the accuracy and precision of musculo-
skeletal motion tracking using cine-PC MRI on a 3.0T platform. J Biomech 44(1):193–197.
doi:10.1016/j.jbiomech.2010.08.029
Bey MJ, Kline SK, Tashman S, Zauel R (2008) Accuracy of biplane x-ray imaging combined with
model-based tracking for measuring in-vivo patellofemoral joint motion. J Orthop Surg Res
3:38. doi:10.1186/1749-799x-3-38
Borotikar BS, Sipprell WH 3rd, Wible EE, Sheehan FT (2012) A methodology to accurately
quantify patellofemoral cartilage contact kinematics by combining 3D image shape registration
and cine-PC MRI velocity data. J Biomech 45(6):1117–1122
Brossmann J, Muhle C, Schroder C, Melchert UH, Bull CC, Spielmann RP, Heller M (1993)
Patellar tracking patterns during active and passive knee extension: evaluation with motion-
triggered cine MR imaging. Radiology 187(1):205–212. doi:10.1148/radiology.187.1.8451415
Burnett KR, Davis CL, Read J (1987) Dynamic display of the temporomandibular joint meniscus by
using “fast-scan” MR imaging. AJR Am J Roentgenol 149(5):959–962. doi:10.2214/
ajr.149.5.959
Cadera W, Viirre E, Karlik S (1992) Cine magnetic resonance imaging of ocular motility. J Pediatr
Ophthalmol Strabismus 29(2):120–122
Carlson VR, Boden BP, Sheehan FT (2016) Patellofemoral kinematics and tibial tuberosity-
trochlear groove distances in female adolescents with patellofemoral pain. Am J Sports Med.
doi:10.1177/0363546516679139
Cheng S, Butler JE, Gandevia SC, Bilston LE (2008) Movement of the tongue during normal
breathing in awake healthy humans. J Physiol 586(17):4283–4294. doi:10.1113/
jphysiol.2008.156430
Drace JE, Pelc NJ (1994) Skeletal muscle contraction: analysis with use of velocity distributions
from phase-contrast MR imaging. Radiology 193(2):423–429. doi:10.1148/
radiology.193.2.7972757
Edsfeldt S, Rempel D, Kursa K, Diao E, Lattanza L (2015) In vivo flexor tendon forces generated
during different rehabilitation exercises. J Hand Surg Eur Vol 40(7):705–710. doi:10.1177/
1753193415591491
Feinberg DA, Crooks L, Hoenninger J 3rd, Arakawa M, Watts J (1984) Pulsatile blood velocity in
human arteries displayed by magnetic resonance imaging. Radiology 153(1):177–180.
doi:10.1148/radiology.153.1.6473779
Finni T, Hodgson JA, Lai AM, Edgerton VR, Sinha S (2006) Muscle synergism during isometric
plantarflexion in achilles tendon rupture patients and in normal subjects revealed by velocity-
encoded cine phase-contrast MRI. Clin Biomech (Bristol, Avon) 21(1):67–74. doi:10.1016/j.
Foo TK, Bernstein MA, Aisen AM, Hernandez RJ, Collick BD, Bernstein T (1995) Improved
ejection fraction and flow velocity estimates with use of view sharing and uniform repetition
time excitation with fast cardiac techniques. Radiology 195(2):471–478. doi:10.1148/
radiology.195.2.7724769
Fregly BJ, Rahman HA, Banks SA (2005) Theoretical accuracy of model-based shape matching for
measuring natural knee kinematics with single-plane fluoroscopy. J Biomech Eng 127
(4):692–699
Fujiwara T, Togashi K, Yamaoka T, Nakai A, Kido A, Nishio S, Yamamoto T, Kitagaki H, Fujii S
(2004) Kinematics of the uterus: cine mode MR imaging. Radiogr Rev Publ Radiol Soc N Am
Inc 24(1):e19. doi:10.1148/rg.e19
Glover GH, Pelc NJ (1988) A rapid-gated cine MRI technique. In: Magnetic resonance annual.
Raven Press, New York, NY, pp 299–333
Im HS, Goltzer O, Sheehan FT (2015) The effective quadriceps and patellar tendon moment arms
relative to the tibiofemoral finite helical axis. J Biomech 48(14):3737–3742. doi:10.1016/j.
Jensen ER, Morrow DA, Felmlee JP, Odegard GM, Kaufman KR (2015) Error analysis of cine
phase contrast MRI velocity measurements used for strain calculation. J Biomech 48(1):95–103.
doi:10.1016/j.jbiomech.2014.10.035
Jensen ER, Morrow DA, Felmlee JP, Murthy NS, Kaufman KR (2016) Characterization of three
dimensional volumetric strain distribution during passive tension of the human tibialis anterior
using cine phase contrast MRI. J Biomech 49(14):3430–3436. doi:10.1016/j.
Kaiser J, Bradford R, Johnson K, Wieben O, Thelen DG (2013) Measurement of tibiofemoral
kinematics using highly accelerated 3D radial sampling. Magn Reson Med 69(5):1310–1316.
doi:10.1002/mrm.24362
Kaiser J, Monawer A, Chaudhary R, Johnson KM, Wieben O, Kijowski R, Thelen DG (2016)
Accuracy of model-based tracking of knee kinematics and cartilage contact measured by
dynamic volumetric MRI. Med Eng Phys 38(10):1131–1135. doi:10.1016/j.
medengphy.2016.06.016
Keegan J, Firmin D, Gatehouse P, Longmore D (1994) The application of breath hold phase velocity
mapping techniques to the measurement of coronary artery blood flow velocity: phantom data
and initial in vivo results. Magn Reson Med 31(5):526–536
Kinugasa R, Shin D, Yamauchi J, Mishra C, Hodgson JA, Edgerton VR, Sinha S (2008) Phase-
contrast MRI reveals mechanical behavior of superficial and deep aponeuroses in human medial
gastrocnemius during isometric contraction. J Appl Physiol (Bethesda, MD: 1985) 105
(4):1312–1320. doi:10.1152/japplphysiol.90440.2008
Lafortune MA, Cavanagh PR, Sommer HJ 3rd, Kalenak A (1994) Foot inversion-eversion and knee
kinematics during walking. J Orthop Res 12(3):412–420. doi:10.1002/jor.1100120314
Langner I, Fischer S, Eisenschenk A, Langner S (2015) Cine MRI: a new approach to the diagnosis
of scapholunate dissociation. Skelet Radiol 44(8):1103–1110. doi:10.1007/s00256-015-2126-4
Manal K, McClay I, Stanhope S, Richards J, Galinat B (2000) Comparison of surface mounted
markers and attachment methods in estimating tibial rotations during walking: an in vivo study.
Markl M, Alley MT, Pelc NJ (2003) Balanced phase-contrast steady-state free precession
(PC-SSFP): a novel technique for velocity encoding by gradient inversion. Magn Reson Med
49(5):945–952. doi:10.1002/mrm.10451
Melchert UH, Schroder C, Brossmann J, Muhle C (1992) Motion-triggered cine MR imaging of
active joint movement. Magn Reson Imaging 10(3):457–460
Moerman KM, Sprengers AM, Simms CK, Lamerichs RM, Stoker J, Nederveen AJ (2012)
Validation of continuously tagged MRI for the measurement of dynamic 3D skeletal muscle
tissue deformation. Med Phys 39(4):1793–1810. doi:10.1118/1.3685579
Moro-oka TA, Hamai S, Miura H, Shimoto T, Higaki H, Fregly BJ, Iwamoto Y, Banks SA (2007)
Can magnetic resonance imaging-derived bone models be used for accurate motion measure-
ment with single-plane three-dimensional shape registration? J Orthop Res 25(7):867–872.
doi:10.1002/jor.20355
Pappas GP, Asakawa DS, Delp SL, Zajac FE, Drace JE (2002) Nonuniform shortening in the biceps
brachii during elbow flexion. J Appl Physiol (Bethesda, MD: 1985) 92(6):2381–2389.
doi:10.1152/japplphysiol.00843.2001
Pelc NJ (1995) Flow quantification and analysis methods. Magn Reson Imaging Clin N Am 3
(3):413–424
Pelc NJ, Bernstein MA, Shimakawa A, Glover GH (1991a) Encoding strategies for three-direction
phase-contrast MR imaging of flow. J Magn Reson Imaging: JMRI 1(4):405–413
Pelc NJ, Herfkens RJ, Shimakawa A, Enzmann DR (1991b) Phase contrast cine magnetic resonance
imaging. Magn Reson Q 7(4):229–254
Pelc NJ, Sommer FG, Li KC, Brosnan TJ, Herfkens RJ, Enzmann DR (1994) Quantitative magnetic
resonance flow imaging. Magn Reson Q 10(3):125–147
Pelc NJ, Drangova M, Pelc LR, Zhu Y, Noll DC, Bowman BS, Herfkens RJ (1995) Tracking of
cyclic motion with phase-contrast cine MR velocity data. J Magn Reson Imaging: JMRI 5
(3):339–345
Pike GB, Meyer CH, Brosnan TJ, Pelc NJ (1994) Magnetic resonance velocity imaging using a fast
spiral phase contrast sequence. Magn Reson Med 32(4):476–483
Regev GJ, Kim CW, Tomiya A, Lee YP, Ghofrani H, Garfin SR, Lieber RL, Ward SR (2011) Psoas
muscle architectural design, in vivo sarcomere length range, and passive tensile properties
support its role as a lumbar spine stabilizer. Spine 36(26):E1666–E1674. doi:10.1097/
BRS.0b013e31821847b3
Riederer SJ (1993) Spatial encoding and image reconstruction. In: Bronskill MJ, Sprawls P (eds)
The physics of MRI: 1992 AAP< summer school proceedings. American Instiutes of Physics,
Woodbury, pp 135–165
Sheehan FT (2012) The 3D in vivo Achilles’ tendon moment arm, quantified during active muscle
control and compared across sexes. J Biomech 45(2):225–230. doi:10.1016/j.
Sheehan FT, Drace JE (2000) Human patellar tendon strain. A noninvasive, in vivo study. Clin
Sheehan FT, Zajac FE, Drace JE (1998) Using cine phase contrast magnetic resonance imaging to
non-invasively study in vivo knee dynamics. J Biomech 31(1):21–26
Sheehan FT, Borotikar BS, Behnam AJ, Alter KE (2012) Alterations in in vivo knee joint
kinematics following a femoral nerve branch block of the vastus medialis: implications for
patellofemoral pain syndrome. Clin Biomech (Bristol, Avon) 27(6):525–531. doi:10.1016/j.
Shibanuma N, Sheehan FT, Lipsky PE, Stanhope SJ (2004) Sensitivity of femoral orientation
estimates to condylar surface and MR image plane location. J Magn Reson Imaging: JMRI 20
(2):300–305. doi:10.1002/jmri.20106
Shibanuma N, Sheehan FT, Stanhope SJ (2005) Limb positioning is critical for defining
patellofemoral alignment and femoral shape. Clin Orthop Relat Res 434:198–206
Shih YF, Bull AM, McGregor AH, Humphries K, Amis AA (2003) A technique for the measure-
ment of patellar tracking during weight-bearing activities using ultrasound. Proc Inst Mech Eng
H J Eng Med 217(6):449–457
Silder A, Reeder SB, Thelen DG (2010) The influence of prior hamstring injury on lengthening
muscle tissue mechanics. J Biomech 43(12):2254–2260. doi:10.1016/j.jbiomech.2010.02.038
Sinha S, Shin DD, Hodgson JA, Kinugasa R, Edgerton VR (2012) Computer-controlled,
MR-compatible foot-pedal device to study dynamics of the muscle tendon complex under
isometric, concentric, and eccentric contractions. J Magn Reson Imaging: JMRI 36
(2):498–504. doi:10.1002/jmri.23617
Stone M, Davis EP, Douglas AS, NessAiver M, Gullapalli R, Levine WS, Lundberg A (2001)
Modeling the motion of the internal tongue from tagged cine-MRI images. J Acoust Soc Am
109(6):2974–2982
Thompson RB, McVeigh ER (2004) Flow-gated phase-contrast MRI using radial acquisitions.
Magn Reson Med 52(3):598–604. doi:10.1002/mrm.20187
Thunberg P, Karlsson M, Wigstrom L (2003) Accuracy and reproducibility in phase contrast

imaging using SENSE. Magn Reson Med 50(5):1061–1068. doi:10.1002/mrm.10634
van Kampen A, Huiskes R (1990) The three-dimensional tracking pattern of the human patella.
J Orthop Res 8(3):372–382. doi:10.1002/jor.1100080309
Verbruggen SW, Loo JH, Hayat TT, Hajnal JV, Rutherford MA, Phillips AT, Nowlan NC (2016)
Modeling the biomechanics of fetal movements. Biomech Model Mechanobiol 15
(4):995–1004. doi:10.1007/s10237-015-0738-1
Waterton JC, Jenkins JP, Zhu XP, Love HG, Isherwood I, Rowlands DJ (1985) Magnetic resonance
(MR) cine imaging of the human heart. Br J Radiol 58(692):711–716. doi:10.1259/0007-1285-
58-692-711
Wedeen VJ, Rosen BR, Chesler D, Brady TJ (1985) MR velocity imaging by phase display.
J Comput Assist Tomogr 9(3):530–536
Wen H, Dou Z, Finni T, Havu M, Kang Z, Cheng S, Sipila S, Sinha S, Usenius JP, Cheng S (2008)
Thigh muscle function in stroke patients revealed by velocity-encoded cine phase-contrast
magnetic resonance imaging. Muscle Nerve 37(6):736–744. doi:10.1002/mus.20986
Westphal CJ, Schmitz A, Reeder SB, Thelen DG (2013) Load-dependent variations in knee
kinematics measured with dynamic MRI. J Biomech 46(12):2045–2052. doi:10.1016/j.
Wilson NA, Sheehan FT (2009) Dynamic in vivo 3-dimensional moment arms of the individual
quadriceps components. J Biomech 42(12):1891–1897. doi:10.1016/j.jbiomech.2009.05.011
Wilson NA, Sheehan FT (2010) Dynamic in vivo quadriceps lines-of-action. J Biomech 43
(11):2106–2113. doi:10.1016/j.jbiomech.2010.04.002
Wood M, Wiang Q (1993) Motion artifacts and remedies. In: Bronskill MJ, Sprawls P (eds) The
physics of MRI: 1992 AAP< summer school proceedings. American Instiutes of Physics,
Woodbury, pp 383–411
Yamashita S, Isoda H, Hirano M, Takeda H, Inagawa S, Takehara Y, Alley MT, Markl M, Pelc NJ,
Sakahara H (2007) Visualization of hemodynamics in intracranial arteries using time-resolved
three-dimensional phase-contrast MRI. J Magn Reson Imaging: JMRI 25(3):473–478.
doi:10.1002/jmri.20828
Zhou H, Novotny JE (2007) Cine phase contrast MRI to measure continuum Lagrangian finite
strain fields in contracting skeletal muscle. J Magn Reson Imaging: JMRI 25(1):175–184.
doi:10.1002/jmri.20783
Zhu Y, Drangova M, Pelc NJ (1996) Fourier tracking of myocardial motion using cine-PC data.
Magn Reson Med 35(4):471–480
Ultrasound Technology for Examining
the Mechanics of the Muscle, Tendon,
and Ligament
Glen Lichtwark
Abstract
Ultrasound imaging provides a means to look inside the body and examine how
tissues respond to mechanical stress or muscle contraction. As such, it can
provide a valuable tool for understanding how muscle, tendon, and ligament
mechanics influence the way we move, or vice versa, in health and disease, or to
understand how and why these tissues might get injured due to chronic or acute
loading. This chapter explores the basic concepts of ultrasound and how it can be
used to examine muscle, tendon, and ligament structure and mechanical function.
It introduces different techniques, like conventional B-mode imaging, three-
dimensional ultrasound, and various forms of elastography that can be used to
quantify geometrical and mechanical properties of the muscle, tendon, and
ligament. Furthermore, methods to quantify muscle and tendon mechanical
function during dynamic human movement are explored, and recommendations
provided on which techniques are most suitable for different biomechanical
investigations. Finally, some predictions about how new ultrasound imaging
technologies might continue to advance our understanding of human motion
are proposed and explored.
Keywords
Biomechanical imaging • Stress • Strain • 3D ultrasound • Tissue tracking •
Elastography
G. Lichtwark (*)
Centre for Sensorimotor Performance, School of Human Movement and Nutrition Sciences, The
University of Queensland, St Lucia, QLD, Australia
e-mail: g.lichtwark@uq.edu.au

DOI 10.1007/978-3-319-30808-1_156-1
2 G. Lichtwark
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Static Measurement of Muscle, Tendon, or Ligament Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
The Basics of Conventional B-Mode Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Measuring Relevant Biomechanical Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Extended Field of View Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Three-Dimensional Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
In Vivo Mechanical Properties of the Muscle, Tendon, and Ligament . . . . . . . . . . . . . . . . . . . . . . . . . 9
Dynamic Tissue Response to Forces and/or Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Elastography and Tissue Strain Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Shear-Wave Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
In Vivo Determination of the Muscle and Tendon Length Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Dynamic Imaging of Muscle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Dynamic Imaging of the Tendon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Dynamic Imaging of Ligaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Other Applications for Ultrasound in Human Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Introduction
A key area of basic, clinical, and applied biomechanics is the quantification of soft
tissue mechanical properties and deformation of such tissues in response to force or
motion (see “▶ Simulation of Soft Tissue Loading from Observed Movement
Dynamics”). Of particular interest are the mechanical properties of the muscle,
tendon, and ligament, as these tissues play a critical role in enabling efficient,
powerful, or highly precise movement. In addition, these tissues are also often
injured in response to either acute or chronic loads. Ultrasound imaging provides a
noninvasive and in vivo method for examining the biomechanical properties and
function of the muscle, tendon, and ligament in humans.
The dynamic function of the muscle, tendons, and ligament in human motion has
traditionally been inferred based on the mechanical properties of tissue that is
harvested from human or animal specimens and the estimated forces that these
tissues experience during movement. For instance, in a classic study Alexander
and Bennet-Clark (1977) estimated the elastic energy that is stored and returned
from the Achilles tendon during running by estimating the forces applied to the
tendon (through inverse dynamics and estimates of tendon moment arm) and stress-
strain properties of the tendon. This approach reveals that much of negative work in
early stance is actually stored in the tendon (rather than dissipated by the muscle) and
then returned in late stance (rather than being generated by the muscle). Therefore
understanding the mechanical interaction of the muscle, tendons, and ligaments is
essential to understanding the energetics of human movement.
In general, muscles, ligaments, and tendons are commonly characterized by
similar compositions across different people (e.g., fiber-type composition in muscle,
collagen type and content in the tendon and ligament). However, different
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . . 3
individuals do display variations in the precise material compositions and structural

makeup of the muscle, tendon, and ligament, which can ultimately influence the
mechanical function of these tissues. Variation in mechanical properties of soft
tissues in the human body are likely to depend on extrinsic factors, like loading
commonly experienced by tissues during activities of daily living or athletic training,
intrinsic factors like gender, age, body size/structure, and various clinical patholo-
gies or genetic factors. Techniques that can help determine which of these extrinsic
and intrinsic factors are important for performance or injury risk is a key area of
research in human motion.
Ultrasound imaging provides a relatively affordable and low-risk method to look
inside the body and examine the mechanical properties of soft tissues like the
muscle, tendon, and ligament.
State of the Art
Ultrasound imaging can be used for three broad purposes in investigating the
mechanics of the muscle, tendon, and ligaments: (1) measurement of tissue archi-
tecture to infer mechanical properties, (2) direct (in vivo) determination of tissue
mechanical properties, and (3) direct measurement of tissue function during force
production or movement. Advances in ultrasound technology, including improve-
ments in image collection and processing, extended field of view measurements, 3D
ultrasound imaging, elastography, shear-wave imaging, and advanced methods to
track tissue deformation will be discussed below with respect to achieving the above
broad purposes.
Static Measurement of Muscle, Tendon, or Ligament Architecture
A primary determinant of the basic mechanical properties of the muscle, tendon, and
ligament is the architecture or geometry of the tissue. For instance, the maximum
force generating capacity of muscle is strongly related to its physiological cross-
sectional area (PCSA, muscle volume divided fiber length (Powell et al. 1984)) and
the peak stress of the tendon and ligaments is directly related to its cross-sectional
area (Pollock and Shadwick 1994). Ultrasound imaging allows for low-risk, in vivo
measurements of the geometry of such tissue, which can be useful for understanding
potential performance of such tissues.
The Basics of Conventional B-Mode Ultrasound
Ultrasound imaging has been used as a medical diagnosis tool for characterizing
tissue dimensions since the 1940s and was quickly adopted for use in musculoskel-
etal assessment (Kane et al. 2004). Ikai and Fukunaga (1968) used ultrasound to
report the relationship between human biceps brachii muscle cross-sectional area
4 G. Lichtwark
Sagittal plane Transverse plane

(side view) (top view)
Image plane Transducer

Transducer
Fig. 1 Conceptual diagram of scanning plane viewed from the sagittal plane (left) and transverse
plane (right), where the transducer is held horizontally to produce a transverse plane image of the
underlying tissues (grayscale image overlaid on transverse view). Note that depth is not sufficient to
view through the leg and the width is constrained by the size of the transducer
and strength in 1968. However, the systematic use of ultrasound to study the
geometry of muscles, tendon, or ligaments for the purpose of understanding
human motion and performance really only became prominent in the 1990s
(Kallinen and Suominen 1994; Kawakami et al. 1993).
Brightness mode, or B-mode, imaging is the most common ultrasound imaging
modality, as it can generate an image which is equivalent to a cross section through
the tissues in which the ultrasound transducer is imaging (Fig. 1a, b). Tissues that
have higher echogenicity (the ability to reflect a sound wave), such as tendinous
connective tissue, show up as white on B-mode scans, while tissues with low
echogenicity show up as black or gray. As such, contrasts between tissues of
different echogenicity are easily identifiable. Muscle fibers typically have low
echogenicity, due to their high water content, while the connective tissue that
binds the muscle (fascia/aponeurosis) and binds the fibers as fascicles (perimyosim)
has high echogenicity (Fig. 2a, b). As such, muscle fascicles can be seen as striated
patterns within muscles when the fascicles are viewed within the plane of the image
(Fig. 2b). Tendons and ligaments typically have high collagen content, which is
relatively highly echogenic; therefore this tissue, and particularly the borders,
typically shows up as white pixels in the image (Fig. 2c, d). Ligaments are often
less echogenic than tendons, partially because of the less regular filament pattern and
the difficulty in aligning to these structures through an image plane. The
echogenicity is also dependent on the frequency of the sound waves emitted by
the transducer, with superficial structures like the tendon and ligaments optimally
imaged with frequencies between 10–12 MHz and deeper muscle at lower frequen-
cies (e.g., 6–8 Hz).
The reflection of sound waves depends not only on the composition of the tissue
but also on the orientation of the tissue relative to the direction of travel of the sound
waves. The “angle of incidence” is the angle at which the sound waves encounter the
a Muscle – transverse plane c Tendon – transverse plane

Skin
Stand-off pad
Subcutaneous fat Skin/tissue
Muscle tissue
Achilles tendon
Muscle fascia Muscle tissue
b Muscle – sagittal plane d Tendon – sagittal plane

Tendon
Muscle
Achilles tendon
Muscle – tendon
junction Muscle tissue
Muscle fascia
aponeurosis
Fig. 2 Ultrasound images of the muscle (a–b) and tendon (c–d) in both transverse and sagittal
plane relative to the leg. Bright white regions indicate tissues with high echogenicity (e.g., muscle
fascia in (a), muscle fascicles in (b)). To ensure contact with the skin with a flat transducer, a stand-
off pad can be used between the transducer and the tissue, which deforms around the region and
allows sound waves to be transmitted, as used in (c)
surface of the tissue of interest (Ihnatsenka and Boezaart 2010) and is optimum when
the direction of the structure is approximately perpendicular to the direction of travel
of the sound waves. As the angle of incidence increases, so that the sound waves
become more and more parallel to the surface of the tissue, the amount of reflection
decreases and therefore reduces the definition of the structure within the tissue, as
more of the sound waves are scattered and less reflected (Fig. 3a). Therefore to
enhance the clarity of structures within an image, the transducer should be positioned
such that it is as perpendicular to the structures of interest as possible, or additionally
in many modern ultrasound machines the angle at which the sound waves are
transmitted can be changed relative to the transducer to enhance the image quality
(Fig. 3b).
Understanding how the angle of incidence influences image quality is important,
because in some instances the surfaces of tissues may have a circular shape (e.g.,
imaging a tendon cross section, Fig. 3a) or the tissues may move (e.g., during muscle
contraction) and hence it is unavoidable that the image quality may not be optimum
across the image or over time. Therefore the correct placement of the transducer to
maximize the image quality for the purposes of the measurement is essential for
measuring muscle, tendon, or ligament architecture or function. Because ligaments
6 G. Lichtwark
a Transverse view of Achilles

tendon (cross section)
Clear echo Dispersed echo

b
Conventional Steering Angle = -10 ° Steering Angle = 10 °
Fig. 3 (a) Conceptual diagram of effect of the angle of incidence or angle of reflecting tissue
relative to the sound wave transmission direction. When the tissues are less orthogonal to the sound
waves, the reflected sound is dispersed and is not detected by the ultrasound transducers receiving
crystals. This can decrease the clarity of borders of tissues that effectively increase the angle of
incidence, like tendons in cross section. (b) Pennate muscle fascicles act at an angle to the direction
of the sound waves, which reduces the signal strength. Changing the angle of the sound waves
through changing the steering angle or by adjusting the direction of the transducer can improve or
degrade the clarity of the fascicles, depending on the direction of the fascicles
often lie between bones, it is often only possible to image superficial ligaments (e.g.,
collateral ligaments of the knee) or small part of ligaments between bones. Therefore
a critical limiting factor in using ultrasound is its dependence on a good imaging site
of the structures relative to the skin, which limits the structures that can be accurately
imaged.
Measuring Relevant Biomechanical Parameters
The main architectural parameters of interest that can be measured statically in the
muscle, tendons, and ligaments are measures of thickness, cross-sectional area, and
length (e.g., fascicle or ligament length) (see “▶ Cross-platform Comparison of
Imaging Technologies for Measuring Musculoskeletal Motion”). The borders of a
muscle, tendon, and ligament contain connective tissue that is reasonably echogenic
and therefore relatively easily identifiable. Therefore it is often possible to visualize
clear cross sections through these structures in different planes (Fig. 2). However, the
size of the cross section that is visible is limited by the length of the imaging
transducer, which defines the maximum width of the image. Therefore for larger
muscles or longer tendons (e.g., Achilles) or ligaments (e.g., plantar fascia), it is hard
to capture the entire cross section in either the sagittal or transverse planes. While the
field of view of ultrasound can limit the potential structures that can be measured,
new technologies have provided solutions to overcome this limitation.
Extended Field of View Ultrasound
One method to overcome the limitation of the field of view of the transducer is to use
‘extended field of view’ or ‘panoramic’ imaging methods that require the user to
move the transducer along a straight line to image consecutive regions of the muscle,
which can be stitched back together using image processing techniques (Cooperberg
et al. 2001). Such imaging methods have been shown to have good validity and
reliability for measuring lengths of muscle fibers and cross-sectional area in various
large human muscles and in different planes of imaging (Noorkoiv et al. 2010a, b);
however the valid reconstruction of the plane is highly dependent on the ability to
move the transducer in a single plane on the surface of the structure.
Three-Dimensional Ultrasound
Three-dimensional (3D) imaging of soft tissues makes it possible to make geomet-

rical measurements in 3D space without the limitations of measurements made in a
single plane. Freehand 3D ultrasound is a technique that uses conventional
two-dimensional imaging (e.g., B-mode), however utilizes multiple image slices
along a structure to reconstruct the area of interest, much in the same way that
magnetic resonance imaging works. To overcome the issue of the image planes not
being perpendicular from one image to the next, it is necessary to track the orien-
tation and position of the transducer and apply a known calibration of the position of
the image relative to the transducer, so that images can be correctly projected into the
3D space and accurate voxel information generated (Treece et al. 2003). This method
is particularly useful as it allows small spaces between image slices and hence has
good spatial resolution, although this is subject to the resolution of the instruments
used to track the position and orientation of the transducer (e.g., magnetic or optical
systems). Freehand 3D ultrasound also makes it possible to make volumetric
measures (e.g., muscle volume; Fig. 4) as well as examine the geometry of large
tissue structures that may not be imaged within the field of view of the transducer in
conventional B-mode ultrasound [e.g., muscle aponeurosis (Raiteri et al. 2016),
tendon length, and cross-sectional area (Obst et al. 2014)].
Three-dimensional ultrasound imaging of smaller volumes can also be performed
in real time using specialized transducers. The first method involves capturing a
small volume by mechanically sweeping a linear array through an angular motion
and reconstructing the volume in a similar manner to freehand 3D ultrasound. This
8 G. Lichtwark
a Transvers scan image b 3D image location

Sagittal
reconstruction
plane
Transverse image scan
Tibialis
anterior
muscle
border
Reconstructed
muscle volume
c Sagittal reconstruction
Current image plane
Tibialis anterior
muscle border
Fig. 4 Freehand 3D ultrasound uses conventional B-mode images that are collected sequentially
along the length of the tissue of interest while the position and orientation of the image is recorded
so that the images can be stacked together to generate a 3D volume. (a) Transverse scan of the
tibialis anterior muscle in mid-region of muscle. (b) Position of transverse scan relative to muscle
and a 3D reconstruction of the muscle volume created from segmenting the muscle borders in
sequential images along the muscle. (c) Sagittal plane reconstruction of the muscle through the
mid-region
method is typically used in obstetrics and gynecology, but has found limited use at
the present for biomechanics, possibly because of the limited capture volume.
However, for imaging small ligamentous structures or sites of muscle injury, this
method may be useful because it can collect and reconstruct data in near real time.
The second method developed was real-time 3D ultrasound, which uses a matrix of
ultrasound elements (instead of an array) to reconstruct volumes at discrete time
points; hence this method is often known as four-dimensional (4D) ultrasound (three
spatial dimensions and a time dimension). Because of the large amount of data that
must be transmitted at high frequencies, the size of the matrix is currently limited and
this technology is primarily used to examine cardiac function (e.g., valve mechan-
ics); however there has been some investigation on the function of the pelvic floor
muscles (Braekken et al. 2009). This technique should be considered further in both
static and dynamic measurement of small muscle, tendon, and ligaments where it is
well suited.
In Vivo Mechanical Properties of the Muscle, Tendon,

and Ligament
While static measures of soft tissue geometry are valuable for biomechanical
assessment of musculoskeletal capacity, the true value of ultrasound comes from
being able to determine subject-specific material properties of tissues (see
“▶ Induced Acceleration and Induced Power Analyses of Human Motion; Optimal
Control Modeling of Human Movement; Physics-based Models for Human Gait
Analysis”). Such imaging allows for the characterization of how soft tissues adapt in
the event of different loading or exercise, various clinical conditions, and also across
the life span. This information is important for understanding capacity to perform
movements, requirements for rehabilitative interventions, or prevention of injury.
Dynamic Tissue Response to Forces and/or Movement
B-mode ultrasound is the most accessible tool to assess deformation or changes in

geometry of human muscle, tendon, and ligaments. Conventional ultrasound
machines have good temporal (time) resolution, with frame rates that vary from
10–100 frames per second, depending on imaging parameters, scan depth, and
computer processing power. This allows users to record changes in geometry during
tasks where tissues are deformed (e.g., muscle contraction, passive stretching).
One of the first demonstrations of this capacity was measurement of muscle
fascicle length changes during passive length changes (Herbert and Gandevia
1995; Narici et al. 1996) and during isometric contractions (Fukunaga et al. 1997;
Fukashiro et al. 1995). These studies clearly showed that human fascicle length
changes are not necessarily concomitant with the whole muscle-tendon unit length.
This is because the muscle fibers connect to the skeleton via elastic tendons which
stretch when force (either passive or active) is applied by the muscle. Utilizing this
knowledge, the first in vivo estimates of tendon stiffness in muscles of the lower limb
(e.g., gastrocnemius, tibialis anterior) were reported through measuring the shorten-
ing of muscle or movement of the muscle-tendon junction during isometric contrac-
tions, which was assumed to be the equivalent of the stretch of the elastic tendons
(Fukashiro et al. 1995; Maganaris and Paul 1999). Providing that adequate estimates
of forces applied to tissues can be determined, typically through measurement of
external forces/torques and estimates or direct measurement of muscle moment arms
(Maganaris 2005), then these measurements can be used to provide estimates of
stress versus strain relationships of both the muscle and tendon.
There are numerous limitations to using fascicle length changes during isometric
contractions to infer strain or material properties of external tendons. One of the
main limitations is that it is difficult to prevent rotation of joints during isometric
contractions, which induce fascicle shortening independent of tendon strain
(Maganaris 2005; Karamanidis et al. 2005). Another difficulty is the ability to
accurately estimate muscle forces using external force measurement techniques
(Lichtwark et al. 2013) or ability to accurately synchronize the force and length
10 G. Lichtwark
change data (Finni et al. 2013). Detailed methods to correct for joint rotation errors
have been developed (Karamanidis et al. 2005), along with methods to track the
movement of the ultrasound transducer to limit errors in displacement measurements
and more accurately synchronize signals; however it remains uncertain whether
strain measured at one end of a tissue is representative of strain occurring throughout
the tissue. In some tendons, like the patella tendon, it is possible to image the entire
tendon within the field of the tendon, and hence this has served as a good model to
understand tendon adaptation to exercise (Hansen et al. 2006; Onambele et al. 2007;
Pearson et al. 2007). However, recent advancements in ultrasound technology have
concentrated on methods to examine local strains or material properties of the
muscle, tendon, and ligament.
Elastography and Tissue Strain Measurement
Ultrasound elastography was first developed as a method to distinguish strain of soft

tissues in response to compression. Tissues that strain more for a given compressive
force are considered more compliant or elastic. Ultrasound is an ideal medium for
examining this mechanical response because the radio frequency (RF) data received
from the return sound wave is in the same line of action as the line of compression of
the tissue and hence will also compress in response to the tissue deformation.
Changes in the compression across a region are indicative of different tissue stiffness
values and can be quantified visually using a map overlaid across the image.
Methods to compress the tissue are varied and include quasi-static or hand-driven
compression, as well as mechanical compression or vibrations or an acoustic radi-
ation force (force generated using the transducer itself) (for reviews of these methods
see Varghese 2009; Treece et al. 2011; Nightingale 2011). While all of these methods
can be used to quantify the strain of tissues in response to the force, it is difficult to
quantify the material properties of the tissue (e.g., Young’s modulus), and these
methods only quantify the stiffness of the tissue in the direction of the sound-wave
beams. For tissues such as muscles and ligaments, where the interesting material
properties are often orthogonal to the sound-wave beams (or parallel with the skin
where the line of action of these tissues is usually most prominent), various other
exciting methods have been developed.
Shear-Wave Imaging
Shear-wave imaging is a form of elastography that utilizes a mechanical perturbation

to the tissue, but instead of determining the compression of the tissue, the speed of
the resulting propagation of the mechanical shear wave along the tissue is examined
through speckle tracking (effective movement of the tissues) across the image. The
technique most commonly used to examine the muscle, ligament, and tendon is
known as “super-sonic shear imaging” or SSI (Hug et al. 2015). This technique uses
an acoustic radiation force and ultrafast imaging to quantify the wave propagation
Fig. 5 Example of supersonic shear imaging (SSI) technique. The bottom image is a conventional
B-mode image of the muscle, and the top image has the shear modulus map of a particular region of
interest. In this case the muscle is contracting slightly, which increases the shear modulus of the
muscle tissue
speed in real time. Making some assumptions about the tissue density enables a
region-specific quantification of shear modulus of the tissue (for full review see Hug
et al. (2015), example image in Fig. 5).
Quantification of the shear modulus of tissues has been shown to have many
biomechanical applications that are useful for understanding human motion. For
instance, there is considerable evidence showing that the average shear modulus
across an area of muscle is highly related to muscle isometric force across a range of
forces that can be assessed (Ates et al. 2015; Bouillard et al. 2012). Therefore, this
technique may be a valuable method to understand which muscles contribute to
forces that generate human motion (Hug et al. 2015). However, at present the low
acquisition rate and low saturation level limit the potential to look at dynamic muscle
contractions (Hug et al. 2015). SSI has also been used extensively to examine
differences in material properties in the tendon (Helfenstein-Didier et al. 2016;
Hug et al. 2013). For instance, the shear modulus of young and old Achilles tendon
12 G. Lichtwark
seems to be different (Slane et al. 2016); however, due to the low saturation level, it
is only possible to look at the tendon at very low (often passive) forces. Currently the
technology has limitations in being able to understand dynamic muscle, tendon, or
ligament function during human motion; however, as this technology improves, it
will likely become an invaluable tool for understanding the material properties of the
muscle, ligament, and tendon in both healthy and clinical populations.
In Vivo Determination of the Muscle and Tendon Length Changes
Ultrasound has a relatively high temporal resolution for characterizing strains of the
tissue, and the imaging location is not constrained because the transducer is freely
moveable. As such, it is possible to use ultrasound imaging to examine the muscle
and tendon length changes during muscular contractions and movement and there-
fore assess muscle mechanical function of individual muscles.
Dynamic Imaging of Muscle
The early measures of muscle fascicle length changes in response to changes in joint
angle or isometric contractions clearly demonstrated that the muscle fascicles appar-
ently changed length in a manner that wasn’t consistent with the whole muscle-
tendon unit (Fukashiro et al. 1995; Fukunaga et al. 1997; Kawakami et al. 1998;
Narici et al. 1996, Herbert and Gandevia 1995). Since this time, it has become
apparent that the length changes of muscle fascicles may be very different to the
length changes of the muscle-tendon unit, particularly in the lower limb muscles like
the gastrocnemius or soleus. For instance, during the early to mid-stance phase of
human walking, it has been demonstrated that muscle fascicles operate relatively
isometrically while the muscle-tendon unit is lengthened (Fukunaga et al. 2001;
Lichtwark and Wilson 2006). As such, it can be concluded that the tendon tissue
must be stretching to store elastic energy that can later be used to help power
propulsion.
Because muscle fascicles are relatively simple to image through the skin, it is
possible to measure dynamic length changes during human motion; however there
are numerous technical considerations that must be considered. Firstly, for such
imaging to be successful, one must ensure that the ultrasound transducer remains in a
similar plane to that in which the fascicles lengthen and shorten. Even in static
measures of muscle length, it can be difficult to do this in a way that accurately
measures muscle fascicle length. Firstly, this requires that most (if not all) of the
length of the muscle fascicle is imaged within the field of view. This limits the size of
the muscles that are capable of being imaged. Many of the human lower limb
muscles have fiber lengths that are less than the width of the transducer and hence
are suitable for dynamic imaging. Secondly, the image plane must be aligned with
the line of the fascicles. Recent comparisons between ultrasound imaging in different
planes and magnetic resonance imaging (more specifically diffusion tensor imaging)
in the gastrocnemius muscle concluded that errors in fascicle length measurement of

up to 20% could be found with misalignment of the transducer (Bolsterlee et al.
2016a, b). These studies have also found that the best location was approximately
perpendicular to the skin, parallel to the tibia, and in the mid-region of the muscle.
Whether this alignment remains consistent through a dynamic task is difficult to
determine.
The general rule of thumb in imaging muscle fascicle lengths during dynamic
tasks is that the lines that constitute the connective tissue around the fascicles should
remain continuous and clear, as should the fascia to which the muscle connects,
throughout the movement (see Fig. 2b). During dynamic contraction the muscle
fascicles shorten and their pennation angle increases. As such the image quality
decreases due to the change in the angle of incidence (Fig. 3b). This can make it
difficult to interpret whether the fascicles are still in the plane of the image. A secure
attachment to the site of imaging, such that there is little movement or rotation
relative to the skin, is a key requirement. Various flat-shaped ultrasound transducers
that can be strapped to the leg seem to be best suited to such tasks (e.g., T-shaped or
veterinary rectal transducers). However, as secure attachment requires pressure, this
can lead to artificial changes in muscle geometry which need to be considered as they
may impact results (Wakeling et al. 2013).
Temporal resolution is an important factor when assessing ultrasound data. As is
the case with any analysis of human motion, the capture rate must be sufficient to
detect the event of interest (typically greater than twice the Nyquist frequency). The
rate at which ultrasound machines can collect a complete frame of B-mode ultra-
sound data depends on many factors including the machine and transducers (pro-
cessing power, size, and resolution of the transducer) and the imaging settings (e.g.,
depth of image, image processing techniques). The range of acquisition speeds
varies from 5 to 10, 000 frames per second. For slow movements (e.g., passive
rotations of joints) a slow frame rate is sufficient, but for activities that require high
temporal resolution (e.g., high speed running, impact during landing), a high frame
rate is essential. A novel application of ultrahigh speed ultrasound has been the
ability to accurately measure the electromechanical delay – the delay between an
electrical impulse transmitted along a muscle to induce contraction and the time at
which force is transmitted. By examining the timing of length changes of the muscle
fibers, which represents force transmission due to stretch of the connective tissue,
Nordez and colleagues (Nordez et al. 2009) were able to determine that the electro-
mechanical delay is as small as 6 milliseconds in the ankle plantar flexor muscles,
although the time taken to transmit the force along a long tendon like the Achilles
may contribute to further delays in force transmission.
Automatic tracking of muscle fascicle length changes during dynamic human
motion is one major advance that has reduced the time required to assess dynamic
length changes. There have been numerous different approaches. The most common
approach has been to attempt to track homologous structures from one image frame
to the next, either through cross-correlation or optic flow techniques (Korstanje et al.
2010; Lee et al. 2008; Loram et al. 2006). These techniques are useful especially for
slower movements or those where it is possible to visualize the same structures
14 G. Lichtwark
across consecutive frames. Another method that is commonly used is an optic flow
algorithm with an affine fit to estimate deformation across a region of the muscle
(Cronin et al. 2011; Farris and Lichtwark 2016; Gillett et al. 2013). Optic flow
algorithms create a vector field estimate of the displacement of multiple regions
across an image, from one frame to the next. An affine transformation (horizontal
and vertical displacement, rotation, dilation, shear in horizontal and vertical direc-
tion) can then be fit to the vector field so as to create a smoothed distortion map from
one image to the next based on movement across the entire image or area of interest.
The distortion map can then be applied to any points within or outside the image, for
example, the end points of fascicles, fascia, or other regions of interest. This has been
shown to be a useful method for tracking dynamic tasks like walking or isometric
contraction. However, the major limitation to these frame-by-frame approaches are
that small errors in frame-by-frame estimations of movement can accumulate over
time and cause the lengths to drift. High frame rates (such that the movement
between frames is small, as is the potential error) can alleviate tracking errors,
while other methods have been proposed to correct drift (Farris and Lichtwark
2016); however there are cases where it may be more suitable to treat individual
frames as separate problems and to identify structural measurements directly from
individual images (e.g., aponeurosis locations, average pennation angle). For
instance, there have been attempts to quantify pennation angle automatically in
individual frames (Rana et al. 2009) or machine learning algorithms that can detect
the length of regions of fascicles (Darby et al. 2012). However such approaches are
still subject to noise and can be highly time consuming (e.g., having to train a data set
for subsequent analysis) and hence the algorithm of choice is very much dependent
on the question being asked and accuracy required.
Dynamic Imaging of the Tendon
There has been less focus on examining dynamic function of the tendon when
compared to the muscle. As was the case described earlier, much of the literature
examining dynamic function of the tendon (e.g., strain responses during movement)
have been deduced based on measures of muscle fascicle length changes. Estimates
of muscle-tendon length changes are determined using kinematics and various
models based on cadaveric data (Hawkins and Hull 1990; Grieve et al. 1978) or
geometrical models (Delp et al. 2007), and the difference between length changes of
the muscle-tendon unit and the fascicles (often corrected for by the pennation angle)
is attributed to the strain of the tendinous tissues (Fukunaga et al. 2001). While this
approach gives a global understanding of the function of tendinous tissue, its
application to understanding differences between or within different populations is
limited because it is hard to assess where potential differences lie within a tendon
(regionally within external tendon or within muscle tendinous tissues like
aponeurosis).
One method to determine the length changes of tendons is to examine the
movement of the end points. As previously mentioned, in tendons like the patella
tendon, that are reasonably short, it may be possible to image the end points of the
tendon to assess overall strain (Hansen et al. 2006), although this author is unaware
of any publications where this has been done outside of isometric contractions. An
alternative for longer tendons, like the Achilles tendon, is to image the muscle-
tendon junction (one end point of the tendon, Fig. 2b) and track the position of this
junction within the image while also tracking the position and orientation of the
image using conventional motion capture techniques (e.g., motion capture markers
mounted on the transducer) (Lichtwark and Wilson 2005). This provides a dynamic
measure of where the proximal end of the tendon is located, and this can be
combined with information about where the tendon inserts onto the skeleton (e.g.,
marker placed on a bony landmark) and a length between the two points determined.
While this is a useful technique for examining tendon strains during dynamic tasks
like walking, running, or hopping, it is still limited in that it ignores factors like
curvature and/or spiral twists (Obst et al. 2014) in tendons and does not give an
indication of regions of high strain along the tendon.
One of the most exciting new developments in tendon research has been the
development of new elastography methods that use the natural deformation of the
tendon during dynamic tasks (i.e., when forces are applied or removed from the
tendon) to estimate strain in local regions. Using speckle tracking algorithms that
require the raw radio frequency data and which have been customized and validated
for estimating tendinous tissue strain (Chernak Slane and Thelen 2014), it has now
become possible to examine dynamic tendon strain in the Achilles tendon during
dynamic tasks like eccentric contracts (Slane and Thelen 2014) or walking (Franz
et al. 2015) (Fig. 6). The potential for this technology to look at region-specific
strains (including regions attaching to the gastrocnemius vs. soleus muscles) in
various populations (Franz and Thelen 2015) and in different tendons is likely to
provide the next major advance in understanding the relationships between move-
ment and tissue strain. However, as was the case with the dynamic muscle imaging,
there needs to be numerous considerations about the potential accuracy of measures
depending on factors like transducer movement and misalignment. Dynamic
elastography techniques where tissue strain is induced by movement or muscle
contraction also has potential for assessing local muscle and ligament strains;
however there is currently little or no research in this area.
Dynamic Imaging of Ligaments
There is very little literature using ultrasound to examine the dynamic strain of
human ligaments. Ultrasound is routinely used as a diagnostic tool to assess liga-
mentous damage; however there is a dearth of literature on the mechanical function
of ligaments in dynamic activities. This may partially be because many ligaments are
difficult to image because they reside in locations with high concentrations of bone,
which can cause large distortions and unwanted reflections within the image. It may
also be because it is difficult to maintain the position of an ultrasound transducer
when attached to a joint. Finally, ligaments often twist and turn around bones and
16 G. Lichtwark
Fig. 6 Example of dynamic speckle tracking (elastography) of the Achilles tendon during dynamic
contractions (eccentric). The transducer is placed over along the length of the tendon (imaging in the
sagittal plane), and the displacement of individual nodes is tracked using the radio frequency
content from the raw ultrasound signal. Note the difference in regional movement from superficial
to deep parts of the Achilles tendon (Figure from Slane and Thelen 2014. Permission from Elsevier
(License Number 3976471248270))
hence it may be difficult to align a transducer well with the ligaments. However, the
potential to use methods already established in tendon research, including tracking
the end points of the bones (Hansen et al. 2006) or using elastography, provides a
large avenue of research for understanding ligament function in human motion.
Other Applications for Ultrasound in Human Motion
While ultrasound has traditionally been used as an imaging modality to assess soft
tissue, it is also increasingly being used to assess bone geometry and bone motion.
For instance, 3D ultrasound techniques are being used to assess the location of bony
landmarks or bone surfaces (Jia et al. 2016; Passmore and Sangeux 2016) or to
assess the position of the joint center of rotation (Peters et al. 2010). Alternatively,
measuring the position and orientation of an ultrasound transducer while imaging
bony landmarks within the plane of the image can also be used to determine bony
translation and/or rotation beneath the surface of the skin (Telfer et al. 2014). This
method could potentially be used as a way to correct for soft tissue movement artifact
associated with conventional 3D motion analysis techniques (see “▶ 3D Dynamic
Probablistic Pose Estimation from Data Collected Using Cameras and Reflective
Markers”).The current limitation to these techniques is the small field of view and
the fact that only the surface can be imaged due to the fact that ultrasound does not
penetrate the bone.
Future Directions
Ultrasound imaging has clearly provided a significant advance in the ability to assess
the material properties or mechanical function of human muscle, tendon, and
ligaments. However, there are numerous limitations to current approaches that either
limits the capacity to make accurate biomechanical measurements or to fully under-
stand how muscle, tendon, or ligament mechanical properties influence or adapt to
human motion.
The future will see continued improvement in image resolution (both space and
time), potential to track strain or displacement of tissues and methods to perturb
tissues for assessment of mechanical properties. Developments in elastography
techniques are likely to have the greatest potential in biomechanics. For instance,
increases in the rate at which SSI can be collected and the range over which it can
measure the shear modulus will allow tissues to be examined while contracting in
more realistic conditions and therefore potentially allow researchers to better assess
when and how much individual muscles contribute to various tasks. Examining
dynamic and localized strains in tendinous and ligamentous tissues using speckle
tracking techniques also has great potential at examining injury mechanics in these
tissues.
One of the biggest limitations of ultrasound is that it is highly localized and only
represents a single plane of the tissue of interest. Advancements in 3D technology
will be a major area of advance in ultrasound imaging. For instance, 3D elastography
has already been developed (Lindop et al. 2006) and may be used to precisely
determine areas of tissue “weakness” or structural difference. 4D ultrasound will
continue to be developed with larger acquisition volumes that will enable researchers
to examine precise muscle, tendon, or ligament strains across multiple planes with
good time resolution. A final possibility is that transducers will also advance so that
they may also deform along with the tissues of interest to enable more accurate 3D
representations with less influence of pressure on the transducer.
Despite the relentless advancement of such technology, it is always necessary to
validate the ability of these technologies in actually quantifying the measurement of
interest. Such validation is often difficult to do in humans in vivo, and hence this is
likely to be time-limiting factor in advancing this area of biomechanics and human
motion.
References
Alexander RM, Bennet-Clark HC (1977) Storage of elastic strain energy in muscle and other
tissues. Nature 265:114–117
Ates F, Hug F, Bouillard K, Jubeau M, Frappart T, Couade M, Bercoff J, Nordez A (2015) Muscle
shear elastic modulus is linearly related to muscle torque over the entire range of isometric
contraction intensity. J Electromyogr Kinesiol 25:703–708
Bolsterlee B, Gandevia SC, Herbert RD (2016a) Effect of transducer orientation on errors in
ultrasound image-based measurements of human medial gastrocnemius muscle fascicle length
and Pennation. PLoS One 11:e0157273
18 G. Lichtwark
Bolsterlee B, Gandevia SC, Herbert RD (2016b) Ultrasound imaging of the human medial gastroc-
nemius muscle: how to orient the transducer so that muscle fascicles lie in the image plane.
J Biomech 49:1002–1008
Bouillard K, Hug F, Guevel A, Nordez A (2012) Shear elastic modulus can be used to estimate an
index of individual muscle force during a submaximal isometric fatiguing contraction. J Appl
Physiol (1985) 113:1353–1361
Braekken IH, Majida M, Engh ME, Bo K (2009) Test-retest reliability of pelvic floor muscle
contraction measured by 4D ultrasound. Neurourol Urodyn 28:68–73
Chernak Slane L, Thelen DG (2014) The use of 2D ultrasound elastography for measuring tendon
motion and strain. J Biomech 47:750–754
Cooperberg PL, Barberie JJ, Wong T, Fix C (2001) Extended field-of-view ultrasound. Semin
Ultrasound CT MR 22:65–77
Cronin NJ, Carty CP, Barrett RS, Lichtwark G (2011) Automatic tracking of medial gastrocnemius
fascicle length during human locomotion. J Appl Physiol (1985) 111:1491–1496
Darby J, Hodson-Tole EF, Costen N, Loram ID (2012) Automated regional analysis of B-mode
ultrasound images of skeletal muscle movement. J Appl Physiol (1985) 112:313–327
OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE
Trans Biomed Eng 54:1940–1950
Farris DJ, Lichtwark GA (2016) UltraTrack: software for semi-automated tracking of muscle
fascicles in sequences of B-mode ultrasound images. Comput Methods Prog Biomed
128:111–118
Finni T, Peltonen J, Stenroth L, Cronin NJ (2013) Viewpoint: on the hysteresis in the human
Achilles tendon. J Appl Physiol (1985) 114:515–517
Franz JR, Slane LC, Rasske K, Thelen DG (2015) Non-uniform in vivo deformations of the human
Achilles tendon during walking. Gait Posture 41:192–197
Franz JR, Thelen DG (2015) Depth-dependent variations in Achilles tendon deformations with age
are associated with reduced plantarflexor performance during walking. J Appl Physiol (1985)
119:242–249
Fukashiro S, Itoh M, Ichinose Y, Kawakami Y, Fukunaga T (1995) Ultrasonography gives directly
but noninvasively elastic characteristic of human tendon in vivo. Eur J Appl Physiol Occup
Physiol 71:555–557
Fukunaga T, Ichinose Y, Ito M, Kawakami Y, Fukashiro S (1997) Determination of fascicle length
and pennation in a contracting human muscle in vivo. J Appl Physiol (1985) 82:354–358
Fukunaga T, Kubo K, Kawakami Y, Fukashiro S, Kanehisa H, Maganaris CN (2001) In vivo
behaviour of human muscle tendon during walking. Proc Biol Sci 268:229–233
Gillett JG, Barrett RS, Lichtwark GA (2013) Reliability and accuracy of an automated tracking
algorithm to measure controlled passive and active muscle fascicle length changes from
ultrasound. Comput Methods Biomech Biomed Engin 16:678–687
Grieve DW, Pheasant S, Cavanagh PR (1978) Prediction of gastrocnemius length from knee and
ankle joint posture. Biomechanics vi-a. University Park Press, Baltimore
Hansen P, Bojsen-Moller J, Aagaard P, Kjaer M, Magnusson SP (2006) Mechanical properties of
the human patellar tendon, in vivo. Clin Biomech (Bristol, Avon) 21:54–58
Hawkins D, Hull ML (1990) A method for determining lower extremity muscle-tendon lengths
during flexion/extension movements. J Biomech 23:487–494
Helfenstein-Didier C, Andrade RJ, Brum J, Hug F, Tanter M, Nordez A, Gennisson JL (2016) In
vivo quantification of the shear modulus of the human Achilles tendon during passive loading
using shear wave dispersion analysis. Phys Med Biol 61:2485–2496
Herbert RD, Gandevia SC (1995) Changes in pennation with joint angle and muscle torque: in vivo
measurements in human brachialis muscle. J Physiol 484(Pt 2):523–532
Hug F, Lacourpaille L, Maisetti O, Nordez A (2013) Slack length of gastrocnemius medialis and
Achilles tendon occurs at different ankle angles. J Biomech 46:2534–2538
Hug F, Tucker K, Gennisson JL, Tanter M, Nordez A (2015) Elastography for muscle biomechan-
ics: toward the estimation of individual muscle force. Exerc Sport Sci Rev 43:125–133
Ihnatsenka B, Boezaart AP (2010) Ultrasound: basic understanding and learning the language. Int J
Shoulder Surg 4:55–62
Ikai M, Fukunaga T (1968) Calculation of muscle strength per unit cross-sectional area of human
muscle by means of ultrasonic measurement. Int Z Angew Physiol 26:26–32
Jia R, Mellon S, Monk P, Murray D, Noble JA (2016) A computer-aided tracking and motion
analysis with ultrasound (CAT & MAUS) system for the description of hip joint kinematics. Int J
Comput Assist Radiol Surg 11:1965–1977
Kallinen M, Suominen H (1994) Ultrasonographic measurements of the Achilles tendon in elderly
athletes and sedentary men. Acta Radiol 35:560–563
Kane D, Grassi W, Sturrock R, Balint PV (2004) A brief history of musculoskeletal ultrasound:
‘from bats and ships to babies and hips’. Rheumatology (Oxford) 43:931–933
Karamanidis K, Stafilidis S, Demonte G, Morey-Klapsing G, Bruggemann GP, Arampatzis
A (2005) Inevitable joint angular rotation affects muscle architecture during isometric contrac-
tion. J Electromyogr Kinesiol 15:608–616
Kawakami Y, Abe T, Fukunaga T (1993) Muscle-fiber pennation angles are greater in hypertrophied
than in normal muscles. J Appl Physiol (1985a) 74:2740–2744
Kawakami Y, Ichinose Y, Fukunaga T (1998) Architectural and functional features of human triceps
surae muscles during contraction. J Appl Physiol (1985b) 85:398–404
Korstanje JW, Selles RW, Stam HJ, Hovius SE, Bosch JG (2010) Development and validation of
ultrasound speckle tracking to quantify tendon displacement. J Biomech 43:1373–1379
Lee SS, Lewis GS, Piazza SJ (2008) An algorithm for automated analysis of ultrasound images to
measure tendon excursion in vivo. J Appl Biomech 24:75–82
Lichtwark GA, Cresswell AG, Ker RF, Reeves ND, Maganaris CN, Magnusson SP, Svensson RB,
Coupe C, Hershenhan A, Eliasson P, Nordez A, Foure A, Cornu C, Arampatzis A, Morey-
Klapsing G, Mademli L, Karamanidis K, Vagula MC, Nelatury SR (2013) Commentaries on
viewpoint: on the hysteresis in the human Achilles tendon. J Appl Physiol (1985) 114:518–520
Lichtwark GA, Wilson AM (2005) In vivo mechanical properties of the human Achilles tendon
during one-legged hopping. J Exp Biol 208:4715–4725
Lichtwark GA, Wilson AM (2006) Interactions between the human gastrocnemius muscle and the
Achilles tendon during incline, level and decline locomotion. J Exp Biol 209:4379–4388
Lindop JE, Treece GM, Gee AH, Prager RW (2006) 3D elastography using freehand ultrasound.
Ultrasound Med Biol 32:529–545
Loram ID, Maganaris CN, Lakie M (2006) Use of ultrasound to make noninvasive in vivo
measurement of continuous changes in human muscle contractile length. J Appl Physiol
(1985) 100:1311–1323
Maganaris CN (2005) Validity of procedures involved in ultrasound-based measurement of human
plantarflexor tendon elongation on contraction. J Biomech 38:9–13
Maganaris CN, Paul JP (1999) In vivo human tendon mechanical properties. J Physiol 521
(Pt 1):307–313
Narici MV, Binzoni T, Hiltbrand E, Fasel J, Terrier F, Cerretelli P (1996) In vivo human gastroc-
nemius architecture with changing joint angle at rest and during graded isometric contraction.
J Physiol 496(Pt 1):287–297
Nightingale K (2011) Acoustic radiation force impulse (ARFI) imaging: a review. Curr Med
Imaging Rev 7:328–339
Noorkoiv M, Nosaka K, Blazevich AJ (2010a) Assessment of quadriceps muscle cross-sectional
area by ultrasound extended-field-of-view imaging. Eur J Appl Physiol 109:631–639
Noorkoiv M, Stavnsbo A, Aagaard P, Blazevich AJ (2010b) In vivo assessment of muscle fascicle
length by extended field-of-view ultrasonography. J Appl Physiol (1985) 109:1974–1979
Nordez A, Gallot T, Catheline S, Guevel A, Cornu C, Hug F (2009) Electromechanical delay
revisited using very high frame rate ultrasound. J Appl Physiol (1985) 106:1970–1975
20 G. Lichtwark
Obst SJ, Newsham-West R, Barrett RS (2014) In vivo measurement of human achilles tendon
morphology using freehand 3-D ultrasound. Ultrasound Med Biol 40:62–70
Onambele GN, Burgess K, Pearson SJ (2007) Gender-specific in vivo measurement of the structural
and mechanical properties of the human patellar tendon. J Orthop Res 25:1635–1642
Passmore E, Sangeux M (2016) Defining the medial-lateral axis of an anatomical femur coordinate
system using freehand 3D ultrasound imaging. Gait Posture 45:211–216
Pearson SJ, Burgess K, Onambele GN (2007) Creep and the in vivo assessment of human patellar
tendon mechanical properties. Clin Biomech (Bristol, Avon) 22:712–717
of the hip joint centre. Gait Posture 31:530–532
Pollock CM, Shadwick RE (1994) Allometry of muscle, tendon, and elastic energy storage capacity
in mammals. Am J Phys 266:R1022–R1031
Powell PL, Roy RR, Kanim P, Bello MA, Edgerton VR (1984) Predictability of skeletal muscle
tension from architectural determinations in guinea pig hindlimbs. J Appl Physiol Respir
Environ Exerc Physiol 57:1715–1721
Raiteri BJ, Cresswell AG, Lichtwark GA (2016) Three-dimensional geometrical changes of the
human tibialis anterior muscle and its central aponeurosis measured with three-dimensional
ultrasound during isometric contractions. PeerJ 4:e2260
Rana M, Hamarneh G, Wakeling JM (2009) Automated tracking of muscle fascicle orientation in
B-mode ultrasound images. J Biomech 42:2068–2073
Slane LC, Thelen DG (2014) Non-uniform displacements within the Achilles tendon observed
during passive and eccentric loading. J Biomech 47:2831–2835
Slane LC, Martin J, Dewall R, Thelen D, Lee K (2016) Quantitative ultrasound mapping of regional
variations in shear wave speeds of the aging Achilles tendon. Eur Radiol 27(2):474–482
Telfer S, Woodburn J, Turner DE (2014) An ultrasound based non-invasive method for the
measurement of intrinsic foot kinematics during gait. J Biomech 47:1225–1228
Treece GM, Gee AH, Prager RW, Cash CJ, Berman LH (2003) High-definition freehand 3-D
ultrasound. Ultrasound Med Biol 29:529–546
Treece G, Lindop J, Chen L, Housden J, Prager R, Gee A (2011) Real-time quasi-static ultrasound
elastography. Interface Focus 1:540–552
Varghese T (2009) Quasi-static ultrasound Elastography. Ultrasound Clin 4:323–338
Wakeling JM, Jackman M, Namburete AI (2013) The effect of external compression on the
mechanics of muscle contraction. J Appl Biomech 29:360–364
Three-Dimensional Human Kinematic
Estimation Using Magneto-Inertial
Measurement Units
Andrea Cereatti, Ugo Della Croce, and Angelo M. Sabatini
Abstract
This chapter deals with the estimation of human kinematics using magneto and
inertial sensing technology. A magneto-inertial measurement unit typically
embeds a triaxial gyroscope, a triaxial accelerometer, and a triaxial magnetic
sensor in the same assembly. By combining the information provided by each
sensor within a sensor fusion framework, it is possible to determine the unit
orientation with respect to a common global coordinate system. Recent advances
in the construction of microelectromechanical system devices have made possible
the manufacturing of small and light devices. These advances have widened the
range of possible applications to include areas such as human movement. This
chapter aims at providing the reader with a picture of the state of the art in the
measurement and estimation methods for the description of human joint kine-
matics using magneto-inertial sensing technology. In the first section, fundamen-
tal concepts of rigid body kinematics are introduced with special reference to
magneto-inertial measurements. Then a short description of the operational
A. Cereatti (*)
Department POLCOMING, University of Sassari, Sassari, Italy
Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, University
of Sassari, Sassari, Italy
Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy
e-mail: acereatti@uniss.it
U. Della Croce
Department POLCOMING, University of Sassari, Sassari, Italy
e-mail: dellacro@uniss.it
A.M. Sabatini
The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
e-mail: angelo.sabatini@sssup.it

DOI 10.1007/978-3-319-30808-1_162-1
2 A. Cereatti et al.
characteristics of accelerometers, gyroscopes, and magnetometers is provided.

The third section reports theory and methods for the estimation of the orientation
and position of magneto-inertial measurement units along with the implementa-
tion of a Kalman filter for 3D orientation estimate as an example. In the last
section, a critical review of the most common methodologies for the joint
kinematic estimation is reported.
Keywords
Joint mechanics • Acceleration • Angular velocity • Orientation • Position • Multi-
segmental model • Multibody • Anatomical coordinate system • Joint kine-
matics • Wearable sensors • Kalman filter • Pose
Abbreviations
ALI Anatomical landmark identification
ARW Angle Random Walk
ACS Anatomical coordinate system
BCS Body-fixed coordinate system
CoR Center of rotation
CS Coordinate system
DoFs Degree of freedom
EKF Extended Kalman filter
FUN Functional
KF Kalman filter
GCS Global coordinate system
IMU Inertial measurement unit
MCS MIMU coordinate system
MEMS Microelectromechanical systems
(M)IMU (Magneto)-inertial measurement unit
MUL Manual Unit Alignment
NEMS Nano-electromechanical systems
VRW Velocity Random Walk
h, i Dot product between vectors
Quaternion multiplication
[q] Skew-symmetric operator
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Rigid Body Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Magneto-Inertial Measurement Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Orientation and Position Estimates Using MIMU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Three-Dimensional Human Joint Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Joint Positional Kinematic Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial. . . 3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Introduction
Human movement kinematics requires the description of the displacements, veloc-

ities, and accelerations, with respect to a global coordinate system (GCS), of every
bony segment modeling the portion of skeletal system under analysis. In general, the
body segments are assumed to be perfectly rigid and therefore to constitute a single
rigid body with the underlying bone. Adjacent body segments are constrained
through ideal joint models. Since the majority of the human joints allow only
small relative linear displacements, whose amplitude is comparable to the errors
associated to their estimate, only joint angular displacements are generally
considered.
This chapter deals with the estimation of the human kinematics using magneto
and inertial sensing technology. The term inertial navigation refers to a set of
techniques that exploit measurements of linear accelerations and angular velocities,
with the aim of estimating the position and orientation (pose) of an object in the
three-dimensional (3D) space relative to a known starting point, orientation, and
velocity (Titterton and Weston 2004). An inertial measurement unit (IMU) is
commonly used to measure linear accelerations and angular velocities. Typically,
an IMU embeds a triaxial gyroscope and a triaxial accelerometer; another type of
sensor that is commonly integrated in an IMU is a triaxial magnetic sensor, which
measures the strength and direction of the local magnetic field, allowing the north
direction to be found (Barbour and Schmidt 2001). Triaxial means that the sensor
sensitivity axes are three, and they are mutually orthogonal to span the whole space
and define the IMU-based coordinate system. The term MIMU (magneto-inertial
measurement unit) is popularly used to denote a device that integrates accelerometer,
gyroscope, and magnetic sensor in the same assembly (Bergamini et al. 2014).
Inertial navigation was first developed for applications including navigation of
aircraft, tactical and strategic missiles, submarines, and ships. Recent advances in
the construction of microelectromechanical system (MEMS) devices have made
possible the manufacturing of small and light IMUs. These advances have widened
the range of possible applications to include areas such as human motion. Several
key factors are behind the success of these sensing technologies. First, (M)IMUs are
self-contained, in the sense that they do not rely on any external infrastructure to be
operational. Second, since inertial and magnetic sensors are heavily used in the
consumer electronics market, their price keeps dropping, while their performance
improves. Lastly, the move from wearable measurement systems to pervasive
systems made possible by the MEMS/NEMS (nano-electromechanical systems:
NEMS) technology opens up new perspectives for motor performance assessment
and monitoring.
Fig. 1 Rigid body motion
This chapter aims at providing the reader with a picture of the state of the art in the
measurement and estimation methods for the description of the human joint kine-
matics using (M)IMUs. In the first section of the chapter, fundamental concepts of
rigid body kinematics are introduced with special reference to magneto-inertial
measurements. Then a short description of the operational characteristics of accel-
erometers, gyroscopes, and magnetometers is provided. The third section reports
theory and methods for the estimation of the orientation and position of (M)IMUs
along with the implementation of a Kalman filter for 3D orientation estimate as an
example. In the final section, a critical review of the most common methodologies
for joint kinematics estimation is reported.
State of the Art
Rigid Body Kinematics
Let B be a rigid body and P a point of B (Fig. 1).

Two coordinate systems are introduced: the global (earth-fixed) coordinate sys-
tem (GCS), specified by the origin OG and the right-handed orthonormal basis
G = {Gx Gy Gz}, and the body-fixed coordinate system (BCS), specified by the
origin OB and the right-handed orthonormal basis B = {Bx By Bz}. The motion
of the body B is described by the translation of the origin OB and the rotation of
!
BCS with respect to GCS. The translation is given by the vector G b ¼ OG OB of the
!
origin OB relative to GCS. The vector G p ¼ OG P gives the coordinates of P relative
!
to GCS; finally, the vector G r ¼ OB P is the vector from the origin OB to P (as seen
from GCS):
G
p ¼ G b þ G r, (1)
With explicit notation, all vectors in Eq. 1 are measured in GCS. The vector Bω
represents the angular velocity, resolved in the BCS, and it describes the rotational
speed of B and its axis of rotation (Fig. 1). Since the body is rigid, the magnitude of
G
r is constant during the body motion. The vector Gr can be expressed in terms of the
time-independent vector Br from OB to P as seen in BCS:
B
r ¼ BG CG r, (2)
where C ¼ BG C is called the rotation matrix from GCS to BCS (Shuster 1993). An
alternative representation is provided by the quaternion q ¼ BG q. The quaternion q is
defined by:
q ¼ ½q q4 T : (3)
It is composed of the scalar component q4 and the vector component given by

q = [qx qy qz]T.
For the quaternion to be a valid parameterization of rotation, the following
normalization constraint must be enforced:
jqj2 þ q24 ¼ q2x þ q2y þ q2z þ q24 ¼ 1: (4)
The quaternion multiplication between two generic quaternions q and h is defined

as:
O
q4 h þ h4 q q h
q h¼ : (5)
q 4 h 4 qT h
A generic vector Gp can be transformed from one coordinate system (e.g., GCS)
to another (e.g., BCS), by pre- and post-multiplying its quaternion (with scalar part
equal to zero) by the rotation quaternion BG q and its inverse (i.e., the same quaternion
with the vector part changed by sign), respectively:
O Gp O
B B B 1
p¼ Gq Gq (6)
0
According to the Euler’s theorem, the most general motion of a rigid body with
one point fixed is a rotation by an angle θ (rotation angle) about some axis n (rotation
axis). This yields another representation of the rigid body orientation in terms of a
rotation vector:
θ ¼ θn: (7)
The rotation vector is related to the quaternion as follows:


T θ θ
q ¼ qx qy qz ¼ sin n, q4 ¼ cos (8)
2 2
The rotation matrix C can be expressed as a function of the quaternion q (Shuster

1993)

C ¼ 2q24 1 I33 ½q 2q4 þ qqT 2, (9)
where In n is the n n identity matrix (n = 3) and the skew-symmetric operator:

2 3
0 qZ qy
½q ¼ 4 qz 0 qx 5, (10)
qy qx 0
and denotes the matrix notation for the cross product. In other words, the skew-
symmetric matrix can be used to represent the cross product between the vector
q and a generic vector t in terms of the matrix-vector multiplication [q]t.
When the BCS is moving with respect to the GCS, the rotation matrix C ¼ G BC
can be shown to be the solution to the matrix differential equation:
C_ ¼ ½ω C, (11)
where ω = Bω. A triaxial gyroscope with the sensitivity axes aligned along the
directions of BCS is customarily used in (M)IMUs to provide the (noisy) measure-
ments of the angular velocity vector.
Equation 11 can be reformulated in an equivalent form involving the quaternion
q¼G B q and its first-order time derivative:

1 1 ½ω ω
q_ ¼ ΩðωÞq ¼ q: (12)
2 2 ωT 0
The solution to this system of first-order linear differential equations from known
initial conditions provides therefore the orientation of the rigid body relative to
the GCS.
Furthermore, time differentiation of (11) yields:

€ ¼ ½ω
C _ þ ½ω2 C: (13)
Using Eqs. 1 and 2, the velocity and the acceleration of the point P (relative to
GCS) can then be written:
p ¼ G b_ þ G
G_ _B G_ G
B C r ¼ b þ ½ω r

G€
p ¼ Gb€ þ GC € B r ¼ Gb
€ þ ½ω
_ þ ½ω2 G r,
B
A triaxial accelerometer located in P senses, resolved in BCS, the so-called

specific force B a ¼ B p€ B g , where Bg is the constant-orientation gravity vec-
tor (Ligorio and Sabatini 2016); for example, when the basis vector Gz is aligned
parallel to the gravity vector, Gg = [0 0 g]T, with g = 9.81 m/s2. Let u = Br and
θi , i = x , y , z denote the location and sensing directions of the accelerometer with
respect to BCS, which are all time independent. Using (14), the components of the
specific force (accelerometer output) a = [ax ay a z]Tcan also be expressed with all
quantities resolved in the BCS as:
D E
B€
ai ðu, θi Þ ¼ _ þ ½ω2 u, θi , i ¼ 1, 2, 3
b B g þ ½ω (15)
where h, i denotes the dot product between vectors. The first term within the dot
product expresses the specific force as the additive combination of linear accelera-
tion, gravity vector, and angular acceleration; the angular acceleration includes the
_
tangential acceleration ½ωu and the centripetal acceleration [ω][ω]u. Suppose
that the location of the accelerometer is where the origin of the BCS lies, i.e., u = 0.
The angular acceleration is therefore null; without loss of generality, we can also
assume that the sensitive axes
of the triaxial accelerometer are oriented along the
directions of Bx By Bz , namely, θ1 = [1 0 0]T , θ2 = [0 1 0]T , θ3 = [0 0 1]T.
As for tracking the origin of the BCS, it is necessary, first, to rotate the (noisy)
measured specific force from the BCS to the GCS using the rotational matrix
computed from the integration of Eq. 11, or equivalently Eq. 12; second, the gravity
vector contribution must be canceled by adding the known expression of Gg to the
rotated measured specific force:
G G B
€¼
p BC a þ G g: (16)
When available in the GCS, the linear acceleration can be integrated once to
obtain velocity and again to obtain displacement (strap-down approach to inertial
navigation).
Determination of the Center of Rotation of a Rigid Body

Let us consider the rigid body B constrained to a rigid frame through a spherical
joint; the body can then only experience a pure rotational motion around the center of
rotation (CoR). According to Eq. 14, assuming OBOG and coinciding with the
CoR, the acceleration of a point P can be expressed as:

G
_ þ ½ω2 G r,
€ ¼ ½ω
p (17)
After some algebraic manipulation, Eq. 2 can be rearranged as:
KG r ¼ G p
€, (18)
where
2

3
ω2y ω2z ωx ωy ω_ z ω_ y þ ωx ωz
6
2

7
K¼6
4 ω_ z þ ωx ωy
ωx ω2z 7
ωy ωz ω_ x 5 (19)

ωx ωz ω_ y ω_ x þ ωy ωz ω2x ω2y
Equation 18 is linear in the unknown vector Gr, which represents the position
vector of CoR in the GCS. The vector G p € can be obtained from the accelerometric
measurements according to (16), and ω can be obtained from the gyroscope read-
ings. The position vector of CoR can be then expressed in the BCS using either the
rotation matrix or the rotation quaternion from GCS to BCS.
In the presence of noise, a more reliable estimate of Gr can be obtained by
computing (18), for each of the N sampled instants of time, recorded during a pure
rotational motion of the rigid body, to obtain an over determined linear system which
can be solved using a least-square technique.
Determination of the Axis of Rotation of a Rigid Body

Let us consider the rigid body B constrained to a rigid frame through a revolute joint,
which is rotating with an angular velocity ω around the single axis of rotation n. The
most straightforward solution to compute the direction of n is from its angular
velocity:
ω
n¼ , (20)
kωk
Alternatively, n can be also obtained from Eq. 8:
q
n¼ , θ ¼ 2cos1 ðq4 Þ, (21)
θ
sin
2
In the presence of noise, a more reliable estimate of r can be estimated either by

selecting only angular velocity above a given threshold or by averaging the quater-
nion over the N observations (Prentice 1986).
Magneto-Inertial Measurement Technology
Measurement System Description

(M)IMUs for applications in the areas of human motion fall in the category of
so-called strap-down systems. Since inertial sensors are rigidly mounted on the
device, output quantities are measured in the BCS rather than the GCS (Titterton
and Weston 2004). To track the (M)IMU orientation, the signals from the gyroscopes
are time integrated. To track position, the signals from the accelerometer must be
Fig. 2 Strap-down approach to inertial navigation
resolved into earth-fixed coordinate system (GCS) using the computed orientation
and then integrated from known initial conditions. This procedure is shown in Fig. 2.
In strap-down IMUs, the signals produced by the inertial sensors are resolved
mathematically, prior to the calculation of navigational information. This reduces the
mechanical complexity of the inertial navigation system, as it is implemented in the
classical applications of inertial navigation technology, i.e., stable platform tech-
nique, thus decreasing the cost and size of the system and consequently increasing its
reliability. Nowadays, the processing speed and low cost of modern computers and
microcomputers allow for the implementation of wearable strap-down IMUs for
applications in human motion.
Measurement Characteristics of State-of-the-Art MEMS

It is commonplace to distinguish different categories, or grades, of inertial sensors,
which group them according to their expected performance, namely, in decreasing
order of performance, “marine and navigation,” “tactical,” “industrial,” and “auto-
motive and consumer” grades (Yazdi et al. 1998). Table 1 depicts the expected
performance in terms of parts per million (ppm) of scale-factor stability (i.e., how
well the sensor reproduces the sensed angular velocity or acceleration) and /h or
m/s2/h of inherent bias stability (i.e., the error independent of angular velocity or
acceleration). While these performance factors are not the only ones that influence
sensor selection, they are useful for comparison purposes. Quite invariably, the
MEMS technologies that, for reasons of cost, complexity, size, and weight, are
compatible with the requirements of human motion studies fall within the “automo-
tive and consumer” grade. It is noted that in the absence of rotation (acceleration),
the gyroscope (accelerometer) output is the sum of white noise and a slowly varying
function (bias). The parameters Angle Random Walk (ARW) and Velocity Random
Walk (VRW) reported in Table 1 are customarily used to quantify the white noise
strength (in alternative to the RMS per square root of measurement bandwidth).
These noise specifications describe the average deviation occurring when signals
from the gyroscope and the accelerometer are integrated and when their actual
estimation is based on Allan variance computation (El-Sheimy et al. 2008).
Although the bias drift could be defined in different ways, the values reported in
Table 1 are intended as the peak-to-peak value of the bias (Yazdi et al. 1998).
Table 1 IMU accuracy specifications

Marine and
Automotive and consumer Tactical navigation
Gyroscopes
Angle random walk [ /h/√Hz] >1 0.1–0.5 <0.002
Bias drift [ /h] >100 5–50 <0.01
Scale factor stability [ppm] >2000 500–1500 <30
Accelerometers
Velocity random walk [g/√Hz] >100 50–75 <20
Bias drift [m/s2/h] >1500 500–1,000 <50
Scale factor stability [ppm] >5000 1000–3000 <150
The accuracy specifications of inertial sensors reported in Table 1 cannot be

directly translated into pose error estimates. Qualitatively, longer time horizons
and higher accuracy are achievable when higher-grade inertial sensors are used in
a given application. However, other problems exist, which can be particularly critical
because of the low performance requirements of MEMS/NEMS technologies. First,
the difficulty of correctly interpreting the acceleration signals, when the component
due to the gravity field (vertical reference) coexists with the component related to the
motion of the object. Hence, the vertical reference is reliable only for static or slowly
moving objects. Second, nearby ferromagnetic materials are critically disturbing
sources when attempts are made to interpret the signals from a triaxial magnetic
sensor as the horizontal reference; this problem becomes especially acute within
man-made indoor environments (Ligorio and Sabatini 2016). Hence, it is difficult to
achieve high accuracy in determining pose by strap-down (M)IMUs, since their
stand-alone accuracy and run-to-run stability are poor, which motivates the growing
interest for the development of highly sophisticated algorithms of sensor fusion
(Sabatini 2011).
Orientation and Position Estimates Using MIMU
This section describes in detail the strap-down navigation algorithm, schematically

depicted in Fig. 2 and outlines how errors stemming from accelerometers and
gyroscopes propagate through the algorithm.
If the angular velocity is constant, then the closed-form solution to Eq. 12 with
known initial conditions is given by:
qðtk Þ ¼ exp½ðtk t0 ÞΩðωðtk ÞÞqðt0 Þ, (22)

and the matrix exponential can be written:

t t
k 0
t t sin jωj
exp½ðtk t0 ÞΩðωðtk ÞÞ ¼ cos jωj
k 0 2
I44 þ t t ΩðωÞ: (23)
2 jωj
k 0
2
The integration of Eq. 12 can be performed using discrete-time angular velocity
measurements, under the assumption that the angular velocity is constant over the
integration period Δt = tk+1 tk:
2 3
Δt
sin jωðtk Þj
6 Δt 2 7
qðtkþ1 Þ ¼ 6
4 cos jωðtk Þj 2 I44 þ Ωðωðtk ÞÞ7
5qðtk Þ: (24)
Δt
jωðtk Þj
2
The quaternion can then be used to rotate the sensed acceleration into the GCS, in
preparation for the following steps of gravity cancelation and time integration that
are needed to obtain velocity and position.
As shown in Eq. 19, errors in the gyroscope signals propagate through to the
calculated quaternion. White noise and uncompensated bias are the main causes of
any error in the orientation (Woodman 2007). White noise is responsible for an
ARW, whose standard deviation grows proportionally to the square root of integra-
tion time (α t1/2). An uncompensated bias causes an error in orientation that grows
linearly with the integration time (α t). Errors due to the accelerometers propagate
through the double integration, giving rise to a quadratic growing position error
(α t2) (due to uncompensated bias) and a second-order random walk (due to white
noise), the standard deviation of which grows proportionally to the square root of the
third power of integration time (α t3/2). This is the obvious cause of drift in the
computed position. However, errors in the angular velocity signals also cause drift,
since the quaternion from Eq. 20 is used to project the acceleration signals into GCS.
The consequence of the orientation error is thus twofold: (a) the accelerations are
integrated in the wrong direction, especially in the presence of uncompensated
heading errors, and (b) acceleration due to gravity can no longer be correctly
removed. Suppose that a tilt error θ has incurred, then a component of the gravity
with magnitude g sin θ is projected onto the horizontal plane, where it will appear as
a residual bias; a residual bias of magnitude g(1 cos θ) will be present in the
vertical direction. Since for small θ we have cosθ 1 and sinθ θ, the vertical
error is generally much lower than the horizontal error, where heading drift further
add to the position error. The propagation of gyroscope errors through to the
calculated position is the critical error path in nearly all strap-down systems, giving
rise to a position error growing with the third power of time (α t3).
In the light of the previous considerations, the information provided by inertial
sensors can be suitably combined within a sensor fusion framework to increase the
quality of the relevant estimates. Two main sensor fusion approaches have been
proposed in the literature (Bergamini et al. 2014). The first is stochastic filtering,
often implemented in the form of Kalman filters (KFs) or their extension to the
nonlinear case, i.e., extended Kalman filters (EKFs). Given a model of the time
evolution of the state of the system under analysis (the MIMU orientation) and of its
noisy observations (the MIMU output signals), Kalman-based methods provide an
estimate of that state. The second approach is complementary filtering, which fuses
multiple noisy measurements that have complementary spectral characteristics. For
each measurement, the complementary filtering exploits only the part of the signal
frequency spectrum that contains useful information. In this case, the characteristics
of the noise present in the process are not modeled for incorporation in the algorithm.
As highlighted by the extensive literature dealing with the development of sensor
fusion algorithms and their applications in human movement analysis, the main
differences between them are (a) how gyroscope bias drift, inertial acceleration, and
magnetic disturbances are modeled/accounted for; (b) how the orientation is
described (e.g., rotation matrix, quaternion); (c) how magnetic sensor data are
employed, when available in the estimation of the heading and attitude angles; and
(d) the number of parameters included in the model of the state of the system for the
stochastic filtering approach. Stochastic estimation algorithms use a model for
predicting aspects of the time behavior of a system (dynamic model) and a model
of the sensor measurements (measurement model), in order to produce the most
accurate estimate possible of the system state. (E)KF algorithms lend themselves
perfectly to this task. An exemplary local EKF algorithm for estimating the orien-
tation will be now considered, partly based on the development presented in Sabatini
(2011), Fig. 3.
The system’s state is the quaternion associated to the rotation from the GCS to the
BCS, the time propagation of which is described by (18), where the noisy gyroscope
measurements are used as input variables to the system (prediction step). Acceler-
ometers and magnetic sensors, or accelerometers alone, are used to implement the
measurement equations according to the following nonlinear model:
E
a
aðtk Þ Cðqðtk ÞÞ 0 g eðtk Þ
¼ E þ m , (25)
mðtk Þ 0 Cðqðtk ÞÞ m eðtk Þ
where the acceleration and magnetic measurements (i.e., ak and mk , respectively)

are explained by rotations of their respective reference vectors, the gravity Eg and the
earth’s magnetic field Em, with measurement noises aek and mek that are modeled as
white Gaussian stochastic processes with zero mean and known covariance matrices,
which can be combined together, to form the measurement noise covariance matrix
R:

I aσ2 0
R ¼ 33 , (26)
0 I33 m σ 2
where aσ 2 and mσ 2 are the measurement noise variances of the accelerometer and
magnetic sensor, respectively (isotropic model). When accelerometers and magnetic
Fig. 3 Flow chart associated to the operation of a generic EKF
sensors are used to prevent the drift resulting from the numerical integration of
Eq. 12, they are both involved in the estimation of the heading angle. The measure-
ment equations are nonlinear; however, the computations for the linearization
process to be carried out are neither algebraically difficult nor computationally
demanding. The derivation of the measurement noise covariance matrix is not
difficult either, since it can be expressed directly in terms of the statistics of the
measurement noise affecting each sensor.
Before performing the Kalman update step, measurements are verified in terms of
expected fields magnitudes and directions, procedure which is commonly referred to
as vector selection. Different strategies can be implemented by the vector selection
block to cope with conditions of fast motion (when the accelerometer output is not
dominated by the gravity) and perturbed magnetic conditions (when the knowledge
of the local magnetic field is, in itself, an estimation problem to be solved). The
Kalman equations, which are not reported here in full for the sake of brevity, are used
to update the computation of the Kalman gain and the second-order statistics of the
error prediction state; then, the quaternion estimate is updated, in a way that
optimally reflects the information available in the incoming measurements and the
knowledge acquired so far about the quaternion itself (Sabatini 2011). If only
accelerometers are considered in the development of the sensor fusion algorithm,
the blocks regarding the use of the magnetic sensors are simply canceled without any
further modification to the Kalman filter equations (prediction and update).
While the tilt error can be mitigated by using accelerometer outputs under quasi-
static conditions, there is no way for a triaxial accelerometer alone to tame the effect
of the gyroscope drift on the heading estimate accuracy and hence on the errors in the
directions where accelerations are integrated. Fortunately, in some applications, the
determination of attitude (roll and pitch angles) is what is really needed; hence the
sensor fusion algorithm can also be implemented without magnetic sensors (Ligorio
and Sabatini 2015). In the context of specific interest in this chapter, for example, a
number of kinematic constraints are enacted in multi-IMU systems, so as to stabilize
the heading estimate “outside” the sensor fusion algorithms running locally at the
level of each unit.
Three-Dimensional Human Joint Kinematics
Quantitative description of human body movement in terms of segmental and joint

kinematic estimation requires preliminarily the implementation of the following
steps: (i) definition of the multibody kinematic model, (ii) calibration of the multi-
body kinematic model, and (iii) determination of the reference “zero” joint
configuration.
Definition of the Multibody Kinematic Model

The skeleton is generally modeled as an assembly of rigid bodies connected by joints
(multibody system). In principle, the relative movement between two adjacent body
segments is described through six degrees of freedom (DoFs), three angular and
three linear displacements (Grood and Suntay 1983). Since for the majority of
human joints the magnitude of some DoFs is comparable to the errors associated
to the estimate, these are not included in the description of the joint motion.
Consequently, the main joints of the human body are described using ideal joint
models. For instance, the knee can be modeled by a revolute joint (R), which
constrains the relative rotation between the femur and tibia about a fixed rotation
axis (1 DoF). Other joints, such as the elbow, ankle, and wrist, can be modeled using
either a universal joint or by skew-oblique universal joint (two non-perpendicular
and non-bisecting revolutes) (2 DoFs) (van den Bogert et al. 1994; Biryukova et al.
2000). Finally, both hip and glenohumeral joints are well described by a spherical
joint (3 DoFs) (Cereatti et al. 2010; Veeger 2000). Both universal and skew-oblique
universal joints can be represented as a three-link kinematic chain connected by two
revolutes (the link between the proximal and the distal one can be imagined to have
zero length) (Grood and Suntay 1983; van den Bogert et al. 1994; Sommer and
Miller 1980), whereas the spherical joint can be replaced by an equivalent compound
joint consisting of four links and three revolutes. One revolute is fixed with the
proximal segment, the second fixed with the distal segment, and the third mutually
orthogonal to both. By considering two imaginary links of zero length between the
proximal and distal segments, the three rotation axes result concurrent (Fig. 4).
The total number of DoFs of the multibody model determines the independent
generalized coordinates, which uniquely define the configuration of the system
relative to the reference configuration. A complete spatial description of the multi-
body kinematic model requires the definition of the length and geometry of the
single bodies. Different formalism can be followed to develop and to mathematically
describe the multibody kinematic model (Denavit-Hartenberg convention, natural
coordinates, twist and product of exponential map) (Dumas and Chèze 2007; Corke
2007; Bregler and Malik 1998). For further details on the mathematical formulation
of the joint constraint and the description of body segment pose, we refer the reader
to specialized texts (Waldron and Schmiedeler 2008).
Calibration of the Multibody Kinematic Model

Once a convenient multibody kinematic model is selected, this has to be calibrated to
the specific subject under analysis (Cereatti et al. 2015). Let us consider a multibody
Fig. 4 Equivalent mechanisms used to model human joints. (a) 1 DoF revolute joint; (b) 2 DoFs
skew-oblique universal joint (link 1 has zero length); (c) 3 DoFs spherical joint (link 1 and 2 have
zero length)
system and a (M)IMU attached to the skin on each of the body segments. The unit
location and the fixing techniques are chosen to minimize the soft tissue artifact and
to avoid restricting the range of movement (for instance, areas close to joints should
be avoided due to the large skin sliding, and fixations should reduce, when possible,
the wobbling of the soft tissues surrounding the bone) (Camomilla et al. 2015) . Let
us also assume that the orientation of the (M)IMU coordinate system (MCS) with
respect to the GCS is made available following one of the sensor fusion approach
previously described. A calibration procedure is needed to determine (1) the relative
position and orientation of the MCS with respect to a bone-embedded CS and (2) the
joint model parameters, which is direction and position of the joint axes with respect
to the (M)IMU.
Typically, in human movement analysis, the bone pose is described through the
position and orientation of an anatomical set of three axes (ACS) which results in a
repeatable relationship with the morphology of the bone. This operation is known as
“anatomical calibration” (Cappozzo et al. 1995) or more generically as “sensor-to-
segment alignment” (Roetenberg et al. 2009), which results in the determination of
the time-invariant transformation between the MCS and the ACS of each segment
and hence between the ACS and the GCS for each instant of time. The latter
transformation can be either represented in the matrix or quaternion formulation.
Subsequently, for each joint, out of the six axes associated with the two bones
involved, three non-orthogonal axes are chosen as joint axes. The rotation angles
about these axes are used to describe joint kinematics (Grood and Suntay 1983). It is
crucial, for a clinically and meaningful interpretation of joint kinematics, that the
direction of each rotation axis are defined according to the joint functional anatomy
in a repeatable and standardized fashion (Wu et al. 2002, 2005).
In some cases, it can be convenient to estimate the joint kinematics as the
composition of rotations around the axes representing the DoFs of the joint model.
This approach may simplify the analysis and the experimental methods since it does
not require the explicit definition of an ACS for each of two body segments forming
the joint (six axes). In this case, the problem of the ACS identification can be reduced
to the determination of the joint model parameters along with the definition of the
reference “zero” joint configuration. The complexity of the joint model calibration
depends on the number of parameters defining the specific model employed (Fig. 4).
For instance, the revolute joint model is completely defined by the direction and the
position of the rotation axis, which is constant with respect to the MCSs of the (M)
IMU attached to joint segments.
Differently, to calibrate the universal joint is necessary to identify the direction,
and when required, the position of two rotation axes fixed with the proximal and
distal segment, respectively. Similarly, the spherical model calibration requires to
identify two rotation axes fixed with the proximal and distal segment (the third axis
is commonly assumed to be perpendicular to both). In addition, when required, the
3D coordinates of its center of rotation must be also determined.
Since methods and solutions proposed for the identification of the ACS and for
the joint axes are the same, no distinction is made in the forthcoming.
Identification of Anatomical/Functional Axes

Different methodologies can be implemented for the identification of joint/segment
axes from the quantities measured or estimated using (M)IMUs. The most straight-
forward solution is to manually align the direction of the axes of the (M)IMU case
with the observational detected anatomical directions/joint rotation axes (“Manual
Unit Alignment” – MUL approach) (Bouvier et al. 2015). However, this approach is
operator dependent, and care should be taken for guaranteeing repeatable outcomes,
given the small size of the (M)IMU case and the different and often nonplanar
surfaces of the body segments.
A second solution is to exploit the joint function (“Functional approach” – FUN)
(Frigo et al. 1998). The use of the FUN requires the existence of a unique rotation
axis fixed with the relevant body and the capability of the subject to generate,
actively or passively, a pure joint rotation about the latter. The direction of this
axis is assumed to coincide with the direction of either the mean attitude vector or the
mean angular velocity vector (Fig. 5a) (see section “Determination of the Axis of
Rotation of a Rigid Body”). For instance, the direction of the knee joint flexion-
extension axis can be determined by recording the movement of the shank with
respect to the femur during a flexion-extension movement (Favre et al. 2009;
Fig. 5a). Similarly, by modeling the elbow joint as an oblique universal joint, the
relevant axes direction can be calibrated by recording arm and forearm movement
data during a pure flex-extension and pro-supination (Luinge et al. 2007; Cutti et al.
2008). The application of the FUN approach may also be extended to joints such as
Fig. 5 Joint axes identification; (a) Functional approach (FUN), identification of the mean knee
flexion-extension axis; (b) Anatomical landmark identification approach (ALI): the direction of the
lines joining the greater trochanter (GT) with the femoral lateral epicondyle (LE) and LE with the
femoral medial epicondyle (ME) are used to define the longitudinal axis of the femur and its frontal
plane
the hip and the shoulder which allow free rotations in all anatomical planes. By
rotating a bone, relative to its adjacent counterpart, in an anatomical plane, the
direction of the orthogonal anatomical axes can be obtained. It should be noted
that the repeatability of the calibration procedure is expected to be higher for those
joints with one dominant and well-defined rotational DoF.
A third solution is to derive the direction of the body/joint axes from information
about the bone morphology and, specifically, from the position of few selected
palpable anatomical landmarks (bone prominence of relatively easy identification)
(Cappozzo et al. 1995) (anatomical landmark identification approach – ALI). To this
purpose, a specifically designed caliper is required. The device carries a (M)IMU
aligned with the axis passing through the tips of the two pointers. By pointing two
palpable anatomical landmarks, an axis can be determined with respect to the LF of
the MIMU attached to the body segment (Fig. 5b; Picerno et al. 2008). This axis can
be used to directly identify the joint axis of rotation or to define anatomical planes
(unit vector orthogonal two nonparallel lines). For instance, the knee flexion-
extension axis can be approximated with the transepicondylar axis (line passing
through the apexes of the medial and lateral femoral epicondyles) or as the unit
vector perpendicular to the transepicondylar axis and the line joining the great
trochanter with the lateral epicondyle (Churchill et al. 1998). The ALI approach,
relying only on subject-specific bone morphology, does not require the subject to
perform any joint movement. The main disadvantage is that a calibration device,
carrying an extra MIMU, is required.
Joint Center Identification

Some methodologies for the joint kinematic estimation require the determination of
the joint center/axis position of the multibody model with respect to the MCSs and
the knowledge of the segment length (Roetenberg et al. 2009; Koning et al. 2013;
Slajpah et al. 2014). The position of the CoR can be estimated from the coordinate
acceleration and angular velocities of the body, while the subject performs a cali-
bration exercise according to Eq. 18. However, since the (M)IMU measures the
specific force in relation to the MCS, the linear acceleration must be obtained by
removing the gravitational acceleration from the specific force once the (M)IMU
orientation with respect to the GF is known (see Eq. 16) (McGinnis and Perkins
2013; Seel et al. 2012). The reliability of the CoR calibration procedure is consid-
erably affected by the joint angular velocity, and hence, slow joint movements
should be avoided (Crabolu et al. 2016). Furthermore, an accurate estimation of
the (M)IMU orientation is fundamental for accurately subtracting the contribution
owing to gravity. This technique can be employed to functionally estimate the CoR
of the hip or shoulder joints as commonly done using stereophotogrammetry. A
similar approach can be employed for determining the joint axes position (Seel et al.
2012). The segment length can be derived from the position of the joint centers and
joint axes once these are defined in the same MCS.
Determination of the Reference “Zero” Joint Configuration

When an ACS is available for each body segment, then the “zero” joint configuration
is defined as the joint configuration for which the ACSs of the distal body the
proximal body coincide. Differently, when only the direction the joint axes of
rotation are defined, the “zero” joint configuration must be assigned. This operation
can be accomplished by imposing a selected posture, commonly the neutral standing
posture (Cooper et al. 2009; Taetz et al. 2016). In some pathological populations,
with irreducible joint flexion or deformities, different postures can be used (Cutti
et al. 2010). It is evident that the information on the joint angles, during the
calibration posture, are lost. This is not the case when the ACS definition relies on
the specific bone functional/anatomical information.
Methods for Joint Angular Kinematic Estimation

The methods proposed in the literature can be divided in different classes based on
different taxonomies. For this short and not exhaustive survey, we focused on the
approach followed to define the joint kinematics from the data provided as output by
the (M)IMU. Methods are described according to the following categories: single
segmental methods (Picerno et al. 2008; Cutti et al. 2010), joint constraint-based
methods (Luinge et al. 2007; Koning et al. 2013; Cooper et al. 2009; Wells et al.
2015; Seel et al. 2014), and methods based on multibody kinematic relations
(Slajpah et al. 2014; Taetz et al. 2016).
Single Segmental Methods

These methods deal with each body segment orientation independently and assume
that each (M)IMU provides highly accurate, driftless, and complete (roll, pitch, and
yaw) estimates of its orientation with respect to a common GCS. To this end, typical
a nine-axis configuration (triaxial accelerometer, gyroscope, and magnetometer) is
employed. The orientation of the MCS can be obtained using either a Kalman or a
complementary filtering approach (Madgwick et al. 2011). Given a proximal and a
distal body segment equipped with a (M)IMU each, their orientation is described by
L p q and L p q. Following one of the general approaches (MUL, FUN, ALI) previously
G G
described, the rotation between the MCS of the (M)IMU and the anatomical ACS for
M M
each segment can be computed (A pp q and A dd q) (Picerno et al. 2008; Cutti et al. 2010).
The orientation of the ACS with respect to the GCS is determined as:
G G O Mp G O Ap 1
Ap
q¼ M p q A p q¼ M pq Mp q , (27)
G G O Md G O Ad 1
Ad
q¼ M d q A d q¼ M d q Md q : (28)
The relative orientation between the CSA of the proximal and the distal body
segments:
Ap G O G 1
Ad
q¼ A dq Ap q : (29)
The joint angular kinematics can then be obtained by decomposing the relative
motion between body segments (inverse kinematic problem). In human movement
analysis, this is commonly accomplished using joint coordinate system angles or a
selected Euler angles sequence (Grood and Suntay 1983).
Joint Constraint-Based Methods

This family of methods exploits the presence of joint constraints between adjacent
body segments to estimate of the joint kinematics. In particular, some of the methods
proposed in the literature require a complete description of the (M)IMU orientation.
The relative orientation between the MCS of the proximal and distal segments can be
expressed as:
Mp
O
G G 1
M d q¼ M d q Mp q , (30)
The relative orientation between the MCS of the proximal and distal segments
during the reference “zero” joint configuration can be expressed as:
O
M p qð0Þ
Mp
¼ MG
d qð0Þ
G 1
M p q ð0Þ: (31)
It follows that the joint angular configuration is represented as:
Mp M 1
O Mp
M d qJ ¼ M dp qð0Þ M d q: (32)
By assuming that the joint model is properly calibrated, the joint angles
representing the generalized coordinates of the system can be determined using
either an analytical or approximated solution. Optimization can be performed to
estimate the model configuration that best fits the kinematic model to a subject’s
movement data. Approximated solutions are convenient to exploit redundancy for
overdetermined joint and multibody models (van den Bogert et al. 1994; Sommer
and Miller 1980; Koning et al. 2013; Wells et al. 2015).
When movement analysis is performed in perturbed magnetic field environments,
the use of the magnetometer for the yaw determination can be critical. However, by
excluding the magnetometer, roll and pitch can be determined from the data recorded
by the triaxial accelerometers and gyroscope, but no correction for the yaw can be
implemented, thus affecting a thorough description of the joint angular kinematics.
Several methods proposed to overcome the latter limitation by imposing joint
constraints. For instance, Cooper and colleagues represented the knee as a revolute
joint, and having specified the rotation axis of the knee, they used a least-square
approach to determine the knee flexion-extension angle (Cooper et al. 2009). Based
on a similar knee joint model, Seel and colleagues, first, identified the direction and
position of the knee joint axis using a functional calibration and then combined, by
means of a sensor fusion algorithm, the knee joint angle estimates obtained by
integrating the angular velocity about the knee axis with those obtained from the
measured acceleration (Seel et al. 2014). In general, the exploitation of joint con-
straints can be only applied for those joints characterized by a number of DoFs<3.
The joint kinematics is computed by first identifying the average axis/axes about
which the relative motion between segments occurs and secondly by decomposing
the motion into one/two rotations about the abovementioned axis/axes.
Methods Based on Multibody Kinematic Relations

The portion of the locomotor system under analysis is modeled as a kinematic chain
of rigid bodies. This includes information about the segment length, the location of
joint axes/centers with respect to the MCS, and the definition of the maximum range
of motion. Exploiting the rigid body kinematics Eq. 14, the kinematics of the
multibody system, is then optimally determined, for each data frame, by minimizing
the difference between the measured and the estimated accelerations and angular
velocities. The optimization can be pursued either following a Kalman filter
approach (Slajpah et al. 2014) or a constrained weighted least-square approach
(Taetz et al. 2016). The calibration of the model can be implemented outside or
inside the optimization process. In the latter case, the motion of the multibody
system and the relevant calibration parameters are determined simultaneously.
These methods are quite attractive since they have the capability of incorporating
modeling constraints to improve the quality of the kinematic estimates; however,
several unresolved questions about their robustness over different experimental
conditions (limited range of motion, slow motions, the absence of stationary body
points, inaccuracy of the model parameters, etc.) require to be investigated.
Joint Positional Kinematic Estimation
Since position errors grow with the third power of time, (M)IMUs do not provide
reliable positional information. Hence, with the performance of the low-cost inertial
sensors currently available, free-inertial navigation is only feasible for durations in
the range of a few seconds. A type of information commonly used for the purpose of
extending the duration is knowledge about the time epochs when the system is in a
stationary phase, i.e., when the system has a constant position and attitude (Skog
et al. 2010). Using this information to bound the error growth is referred to as using
zero-velocity updates. Zero-velocity updates are well suited for bounding the error
growth of a foot-mounted IMU, because during ordinary gait, the foot returns to a
stationary state on a regular basis. In summary, the position in the 3D space of the
multibody model can be derived using a forward kinematic approach once the length
and geometry of the bodies are provided, the independent generalized coordinates
defined, and the instantaneous position of at least one point of the multibody model
is known. This is the case, for instance, when that point is stationary, as when the
base of support is fixed.
Conclusion
• The measurement characteristics of state-of-the-art inertial/magnetic sensors

make the development of sophisticated sensor fusion methods mandatory to
achieve good accuracy in reconstructing a human body pose, especially for
long time scales of observation.
• Kalman filter-based filtering methods are usually designed with the aim to fuse
inertial/magnetic sensor data; accurate pose estimates are possible, in spite of few
difficulties: (a) numerical drift occurring in the integration of measured angular
velocities and accelerations, (b) conditions of fast motion of the involved body
segments, and (c) distortions of the local magnetic field.
• Model-based sensor fusion is a relatively new idea for human motion capture,
where additional information, i.e., biomechanical constraints enacted by the way
a human body moves in the 3D space, is incorporated in the design of sensor
fusion methods for inertial body sensor networks.
• Quantitative description of human body movement in terms of segmental and
joint kinematic estimation requires the implementation (implicit or explicit) of the
following steps: definition of an appropriate multibody kinematic model, a
procedure for the calibration of the multibody kinematic model on the specific
subject under analysis, and the determination of the reference “zero” joint
configuration.
• Some methods for joint kinematic estimates assume a time-invariant anatomical
calibration, but if an undesired movement of the (M)IMU with respect to the body
segment occurs, the calibration is no longer valid. This can be particular critical
when recording the human movement during daily living under unsupervised
conditions.
• In the last years, several original solutions for joint kinematic estimate using (M)
IMUs have been proposed; however, very often the proposed methods lack of
adequate validation in different experimental scenarios and across different path-
ological populations. Future research in this direction is needed to extend the use
of magneto-inertial technologies to different human movement applications.
Cross-References
▶ 3D Dynamic Pose Estimation Using Cameras and Reflective Markers

▶ 3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras
▶ Optimal Control Modeling of Human Movement
References
Barbour N, Schmidt G (2001) Inertial sensor technology trends. IEEE Sensors J 1:332–339
Bergamini E, Ligorio G, Summa A, Vannozzi G, Cappozzo A, Sabatini AM (2014) Estimating
orientation using magnetic and inertial sensors and different sensor fusion approaches: accuracy
assessment in manual and locomotion tasks. Sensors 14:18625–18649
Biryukova EV, Roby-Brami A, Frolov AA, Mokhtari M (2000) Kinematics of human arm
reconstructed from spatial tracking system recordings. J Biomech 33:985–995
van den Bogert AJ, Smith GD, Nigg BM (1994) In vivo determination of the anatomical axes of the
ankle joint complex: an optimization approach. J Biomech 27:1477–1488
Bouvier B, Duprey S, Claudon L, Dumas R, Savescu A (2015) Upper limb kinematics using inertial
and magnetic sensors: comparison of sensor-to-segment calibrations. Sensors 15:18813–18833
Bregler C, Malik J (1998) Tracking people with twists and exponential maps. In: Proceedings of
1998 I.E. computer society conference on computer vision pattern recognition (Cat.
No.98CB36231), pp 8–15
Camomilla V, Bergamini E, Fantozzi S, Vannozzi G (2015) In-field use of wearable magneto-
inertial sensors for sports performance evaluation. In: 33rd international conference on biome-
chanics in sport, pp 1–4
Cappozzo A, Catani F, Della Croce U, Leardini A (1995) Position and orientation in space of bones
during movement. Clin Biomech 10:171–178
Cereatti A, Margheritini F, Donati M, Cappozzo A (2010) Is the human acetabulofemoral joint
spherical? J Bone Joint Surg Br 92:311–314
Cereatti A, Trojaniello D, Della Croce U (2015) Accurately measuring human movement using
magneto-inertial sensors: techniques and challenges. In: 2015 I.E. international symposium on
inert sensors system proceedings, pp 1–4
Churchill DL, Incavo SJ, Johnson CC, Beynnon BD (1998) The transepicondylar axis approximates
the optimal flexion axis of the knee. Clin Orthop Relat Res 356:111–118
Cooper G, Sheret I, McMillian L, Siliverdis K, Sha N, Hodgins D et al (2009) Inertial sensor-based
knee flexion/extension angle estimation. J Biomech 42:2678–2685
Corke PI (2007) A simple and systematic approach to assigning Denavit-Hartenberg parameters.
IEEE Trans Robot 23:590–594
Crabolu M, Pani D, Raffo L, Cereatti A (2016) Estimation of the center of rotation using wearable
magneto-inertial sensors. J Biomech 49:3928–3933
Cutti AG, Giovanardi A, Rocchi L, Davalli A, Sacchetti R (2008) Ambulatory measurement of

shoulder and elbow kinematics through inertial and magnetic sensors. Med Biol Eng Comput
46:169–178
Cutti AG, Ferrari A, Garofalo P, Raggi M, Cappello A, Ferrari A (2010) “Outwalk”: a protocol for
clinical gait analysis based on inertial and magnetic sensors. Med Biol Eng Comput 48:17–25
Dumas R, Chèze L (2007) 3D inverse dynamics in non-orthonormal segment coordinate system.
Med Biol Eng Comput 45:315–322
El-Sheimy N, Hou H, Niu X (2008) Analysis and modeling of inertial sensors using Allan variance.
IEEE Trans Instrum Meas 57:140–149
Favre J, Aissaoui R, Jolles BM, de Guise JA, Aminian K (2009) Functional calibration procedure
for 3D knee joint angle description using inertial sensors. J Biomech 42:2330–2335
Frigo C, Rabuffetti M, Kerrigan DC, Deming LC, Pedotti A (1998) Functionally oriented and
clinically feasible quantitative gait analysis method. Med Biol Eng Comput 36:179–185
dimensional motions: application to the knee. J Biomech Eng 105:136–144
Koning BHW, van der Krogt MM, Baten CTM, Koopman BFJM (2013) Driving a musculoskeletal
model with inertial and magnetic measurement units. Comput Methods Biomech Biomed Eng
5842:37–41
Ligorio G, Sabatini AM (2015) A novel Kalman filter for human motion tracking with an inertial-
based dynamic inclinometer. IEEE Trans Biomed Eng 62:2033–2043
Ligorio G, Sabatini AM (2016) Dealing with magnetic disturbances in human motion capture: a
survey of techniques. Micromachines 7:1–17
Luinge HJ, Veltink PH, Baten CTM (2007) Ambulatory measurement of arm orientation.
Madgwick SOH, Harrison AJL, Vaidyanathan R (2011) Estimation of IMU and MARG orientation
using a gradient descent algorithm. IEEE Int Conf Rehabil Robot 2011:5975346. doi:10.1109/
ICORR.2011.5975346.
McGinnis RS, Perkins NC (2013) Inertial sensor based method for identifying spherical joint center
of rotation. J Biomech 46:2546–2549
Picerno P, Cereatti A, Cappozzo A (2008) Joint kinematics estimate using wearable inertial and
magnetic sensing modules. Gait Posture 28:588–595
Prentice MJ (1986) Orientation statistics without parametric assumptions. J R Stat Soc Ser
B 48:214–222
Roetenberg D, Luinge H, Slycke P (2013) Xsens MVN: Full 6DOF Human Motion Tracking Using
Miniature Inertial Sensors. In: Xsens Technologies, whitepaper, version 3:1–9.
Sabatini AM (2011) Estimating three-dimensional orientation of human body parts by inertial/
magnetic sensing. Sensors 11:1489–1525
Seel T, Schauer T, Raisch J (2012) Joint axis and position estimation from inertial measurement data
by exploiting kinematic constraints. In: 2012 IEEE international conference on control appli-
cations (CCA). Part of 2012 IEEE multi-conference on systems and control, Dubrovnik, 3–5 Oct
2012, pp 45–49
Seel T, Raisch J, Schauer T (2014) IMU-based joint angle measurement for gait analysis. Sensors
14:6891–6909
Shuster MD (1993) A survey of attitude representations. J Astronaut Sci 41:439–517
Skog I, Händel P, Nilsson J-O, Rantakokko J (2010) Zero-velocity detection – an algorithm
evaluation. IEEE Trans Biomed Eng 57:2657–2666
Slajpah S, Kamnik R, Munih M (2014) Kinematics based sensory fusion for wearable motion
assessment in human walking. Comput Methods Prog Biomed 116:131–144
Sommer HJ, Miller NR (1980) A technique for kinematic modeling of anatomical joints. J Biomech
Eng 102:311–317
Taetz B, Bleser G, Miezal M (2016) Towards self-calibrating inertial body motion capture. 19th Int
Conf Informa Fusion 1751–1759
Titterton DH, Weston JL (2004) Strapdown inertial navigation technology. Editor AIAA Educ. Ser.,
Reston
Veeger HEJ (2000) The position of the rotation center of the glenohumeral joint. J Biomech
33:1711–1715
Waldron K, Schmiedeler J (2008) Kinematics. Springer Handbook of Robotics, pp 9–33
Wells D, Cereatti A, Camomilla V, Donnelly C (2015) A calibration procedure for mimu sensors
allowing for the calculation of elbow angles. In: 33rd international conference on biomechanics
in sports, Poitiers
Woodman OJ (2007) An introduction to inertial navigation. Cambridge University Press,
Cambridge
Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D et al (2002) ISB recommendation
on definitions of joint coordinate system of various joints for the reporting of human joint
motion – part I: ankle, hip, and spine. J Biomech 35:543–548
Wu G, Van Der Helm FCT, Veeger HEJ, Makhsous M, Van Roy P, Anglin C et al (2005) ISB
recommendation on definitions of joint coordinate systems of various joints for the reporting of
human joint motion – part II: shoulder, elbow, wrist and hand. J Biomech 38:981–992
Yazdi N, Ayazi F, Najafi K (1998) Micromachined inertial sensors. Proc IEEE 86:1640–1658
Gait Parameters Estimated Using Inertial
Measurement Units
Ugo Della Croce, Andrea Cereatti, and Martina Mancini
Abstract
Gait temporal and spatial parameters are effective indicators of the quality of
mobility. They are usually estimated in a controlled and dedicated space using
relatively expensive instrumentation. The development of wearable technology
allowed for the use of inertial measurement units to estimate various gait param-
eters. The level of accuracy for their determination, required in clinical contexts,
can be achieved by carefully processing the data recorded by the sensors. In this
chapter, a survey of approaches and methods proposed in the literature for
estimating temporal and spatial gait parameters is presented. They differ in the
sensor configuration and in the data processing and are applied to healthy and/or
pathologic subject groups.
Moreover, an overview of the use of gait temporal and spatial parameters in
both straight-ahead walking and turning is presented in a clinimetric context.
Applications in the laboratory or clinic are presented as well as in real-life
environments.
U. Della Croce (*)

Department of POLCOMING, University of Sassari, Sassari, Italy
e-mail: dellacro@uniss.it
A. Cereatti
Department of POLCOMING, University of Sassari, Sassari, Italy
Department of Electronics and Telecommunications, Politecnico di Torino, Torino, Italy
e-mail: acereatti@uniss.it
M. Mancini
Department of Neurology, Oregon Health & Science University, Portland, OR, USA
e-mail: mancinim@ohsu.edu

DOI 10.1007/978-3-319-30808-1_163-1
2 U. Della Croce et al.
Keywords
Human movement analysis • Gait • Inertial measurement unit • Angular velocity •
Acceleration • Temporal and spatial gait parameters • Turning • Clinimetrics •
Foot clearance • Interfoot distance
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Technology and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
The “Traditional” Temporal and Spatial Gait Parameters, Their Determination,
and the Relevance of the IMU Locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Additional Gait Spatial Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Temporal and Spatial Parameters in the Clinic: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Clinimetrics of Gait Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Is Straight-Ahead Gait Enough? Importance of Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Objective Measures of Gait and Turning in Daily Life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Potential for Real-Time IMU-Based Biofeedback and Smartphone Applications . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Introduction
Inertial measurement units (IMUs) often include triaxial accelerometers and gyro-
scopes. In some cases triaxial magnetometers are also integrated. Accelerometers,
gyroscopes, and magnetometers are complementary sensors since they measure
different physical quantities, which, if properly combined, allow for the estimation
of derived physical quantities, such as position and orientation of the sensor, thus
making possible a complete kinematic analysis of the observed phenomenon. With
the advances in microelectromechanical systems (MEMS) technology, miniaturized
IMUs are nowadays often employed as wearable devices in human movement
analysis applications. Typically, the IMU is built so that the accelerometer, gyro-
scope, and magnetometer share a single coordinate system.
The accurate determination of the orientation in space of an IMU from the
measured quantities is crucial and not effortless. In particular, the determination of
the direction of gravity in the IMU coordinate system and of the IMU orientation in
the horizontal plane is very critical for an accurate kinematic description of the IMU
motion. Sensor fusion algorithms, such as Kalman filters and complementary and
particle filters, have been proposed to improve orientation estimates as provided by
the measured quantities (Sabatini 2011).
The determination of the position in space of an IMU requires to be handled with
extreme care. The information that can be extracted from an IMU allows for the
determination of its own position with respect to itself at a different time, i.e., its
displacement. To obtain the IMU displacement in the three directions of space, it is
necessary to determine first its orientation in space. This allows to identify the
Gait Parameters Estimated Using Inertial Measurement Units
Fig. 1 White circles represent gait temporal parameter estimation methods (SB signal-based analysis, ML machine learning); gray circles represent gait spatial
parameter estimation methods (DI direct integration, GM human gait model, ML machine learning, OA other approaches) (From Cereatti et al. 2015)
3
direction of the gravity line with respect to the IMU reference system. Then, the
gravitational component can be subtracted from the proper acceleration measured by
the accelerometer to obtain the actual IMU acceleration. Next, the latter can be
integrated twice and the IMU displacement can be determined if the velocity at the
beginning of the integration interval is known. Unfortunately, the residual presence
of the gravitational component in the acceleration signals causes the error in deter-
mining the displacement to grow dramatically as the integration time increases.
Moreover, if the known value of the initial velocity is affected by an error, the
estimation of the displacement will be also affected by an error that grows propor-
tionally to the integration time. Finally, accelerometric and gyroscopic intrinsic drifts
introduce nonlinear additional errors to the displacement estimation.
Typical countermeasures to limit orientation errors include the fine-tuning of the
filters employed, while errors in the known value of the initial velocity can be limited
when the integration is made to begin at a time of zero velocity as in the zero velocity
update techniques (Foxlin 2005; Peruzzi et al. 2011). When the displacement at the
end of the integration time is known, additional techniques to limit the error in
determining the time history of the IMU displacement can be employed (Zok et al.
2004).
In analyzing gait, the integration time can be made to coincide with the cycle
duration, and a combination of optimal filtering and weighted direct and reverse
integration techniques can be employed to determine the IMU displacement (Kose
et al. 2012). When positioned on the foot or in its proximity, the IMU can be
effectively used to determine some of the classical clinical gait parameters with a
satisfactory accuracy both in normal and pathological gait (Trojaniello et al. 2014a).
State of the Art
In recent years, body-worn IMUs have been frequently presented as an affordable

solution to assess gait in a variety of environments. Two reviews pointed out gait
analysis of temporal and spatial parameters in the clinic and daily activity monitoring
as the most common and promising areas of application for IMUs (Kavanagh and
Menz 2008; Iosa et al. 2016).
As mentioned above, gait is normally described as repetitions of cycles. The most
common event used to segment gait into cycles is the initial foot contact (in some
cases called heel or foot strike). The time difference between two consecutive initial
contacts of the same foot is the cycle (or stride) duration. Often, a second time event
is considered of interest and, therefore, observed: the foot final contact (in some
cases called toe off), occurring just before the beginning of the leg swing. The two
events for each foot define all gait phases: swing and stance phase or single and
double support phases (Perry and Burnfield 1992). As a consequence, the accuracy
and precision of the determination of duration of all gait phases rely on the quality of
the determination of the two gait events.
The literature offers a large variety of methods to determine the duration of the
gait cycle and the relevant phases using IMUs. IMU location, selected signal(s), and
Gait Parameters Estimated Using Inertial Measurement Units 5
data analysis are the main factors differentiating the proposed methods. Some
methods propose the use of a single IMU located usually on the low back or chest,
while others require the use of two IMUs located on the feet or on the lower limbs
(Fig. 1). Acceleration signals are the preferred choice in single IMU-based methods,
while gyroscopic signals are more suitable when the IMUs are positioned on both
lower limbs. It was noted that single IMU methods can be accurate and robust in
detecting gait phases in normal gait (Trojaniello et al. 2014b) while their use in
analyzing pathologic gait may be critical (Trojaniello et al. 2015a).
Technology and Methodology
The “Traditional” Temporal and Spatial Gait Parameters, Their

Determination, and the Relevance of the IMU Locations
Some spatial gait parameters, such as stride and step length (i.e., the distance
traversed in full and in half a gait cycle, respectively), are closely related to the
corresponding temporal parameters: stride and step duration. The determination of
any of such spatial parameters requires the prior determination of the relevant
temporal parameters and, ultimately, of the gait events. In general, the more reliable
is the determination of the time parameter, the more reliable is the determination of
the relevant spatial parameter. Approaches proposed in the literature vary based on
the location of the IMU. When a single IMU is applied on the lower back, the
acceleration along the direction of progression can be double integrated to provide
the step and stride length and consequently the traversed distance (Kose et al. 2012).
Alternatively, a geometric model based on the inverted pendulum can be employed
(Gonzalez et al. 2007). When the IMUs are applied bilaterally on the lower limbs or
feet, the determination of the stride length can be more accurate than using a single
IMU, while the determination of the step length requires the management of signals
from two different IMUs, thus having different drift and error characteristics.
An alternative approach for the determination of both temporal and spatial
parameters of gait takes advantage of machine learning algorithms operating on a
number of IMU raw and derived signals (Fig. 1).
Additional Gait Spatial Parameters
Foot Clearance
The foot clearance during gait can be defined as the minimum height of the forefoot
from the walking surface during in the middle of the swing phase of gait (Khandoker
et al. 2010) or, more generally, if no specific part of the foot is specified, as the
minimum vertical distance during mid swing between the ground and the swinging
limb (Best and Begg 2008). The foot clearance is considered of great clinical
relevance since it is associated to the risk of fall due to tripping (Winter 1992;
Barrett et al. 2010). Maximum vertical displacements after toe off and during the
swing phase have been also extensively explored (Nagano et al. 2011), since they are
related to the subject’s ability to negotiate an obstacle. The minimum distance
between the foot and the obstacle often occurs when the foot passes over the highest
point of the obstacles, and it may occur when either the toes or the heel passes the
obstacle. However, in some pathological gait, this might not be the case.
The foot clearance during swing and its minimum value can be estimated with
acceptable accuracy using an IMU located on the foot, possibly in the proximity of
the toes (Lai et al. 2008; McGrath et al. 2011; Mariani et al. 2012b; Dadashi et al.
2013). These methods provide an estimate of the minimum toe clearance using a
simple double integration (Lai et al. 2008) or the maximal foot height (both at toe
and heel level) using a de-drifted double-integration technique (Mariani et al. 2012b)
reaching an accuracy precision, computed as the mean and standard deviation
value of the error (i.e., the difference between the values of IMU-based estimated
parameters and the gold standard ones), of 13 9 mm at the toe and 41 23 mm at
the heel. Other studies explored the relationship between the mean absolute value of
the medio-lateral angular velocity or vertical acceleration and minimum ground
clearance (McGrath et al. 2011) showing r-values ranging from 0.63 to 0.77. An
alternative solution is to place the IMU on the shank. This location has been proven
to show lower signal variability across subjects and lower dependency on pathologic
foot postures (Goldstein and Harper 2001). In addition, the ankle kinematics can be
estimated using the signals provided by shank-mounted IMUs when the foot can be
assumed to be flat on the ground.
The abovementioned studies limited the analysis of the foot clearance to over-
ground walking and did not specifically address the analysis during obstacle nego-
tiation. Trojaniello and colleagues proposed and evaluated a method for estimating
the maximum foot clearance during overground walking and obstacle negotiation
using IMUs attached above the ankles (Trojaniello et al. 2015a, b). The method was
applied to the gait of both healthy young subjects and subjects with various abnormal
gaits showing mean absolute errors associated with maximum clearance estimates of
10% of its range of variation. In addition, a mean absolute error < 5% was reached
for the maximum clearance during obstacle passing.
Interfoot Distance
The interfoot distance (IFD) is the projection along the medio-lateral direction of the
vector identified by corresponding points of the feet. In straight walking, its value
varies with time and exhibits a quasi-period of half a gait cycle if gait is symmetric.
Its pattern in time may be used as an indicator of the stability of gait (Krebs et al.
2002). Its value in double support is equivalent to the step width (SW), which is used
in clinical gait analysis to evaluate stability and risk of falls (Riley et al. 1995; Brach
et al. 2005). The SW may be measured in laboratory settings with instrumented
treadmills or instrumented gait mats (Brach et al. 2001; Owings and Grabiner 2004).
Conversely, to measure the entire IFD pattern during the gait cycle (including SW),
optoelectronic stereophotogrammetric (SP) systems are used (Krebs et al. 2002).
However, as for most gait-related measurements, IFD or SW measurements obtained
in laboratory settings may not represent the subject-specific gait characteristics in
real-life conditions. It is important to highlight that IMUs alone do not allow to

estimate neither the IFD nor the SW. For their estimation, it is necessary to integrate
a proximity sensors and an inertial-magnetic platform. In this respect, a few works
proposed the measurement of IFD using wearable sensors, such as those based on
ultrasounds (US) and infrared light (IR), often integrated into the shoes (Bamberg
et al. 2008; Mariani et al. 2012a; Wada et al. 2012; Hung and Suh 2013). In
particular, a system composed by IMUs and US was proposed for gait analysis,
but resulted to be bulky and awkward to use routinely. It also required a frequent
recalibration (Bamberg et al. 2008; Wada et al. 2012). IMUs were also integrated
with IR-based systems, including a microcamera and a panel with LEDs, the size of
which could affect walking patterns (Mariani et al. 2012a; Hung and Suh 2013).
Infrared range sensors (IRR) employ the single-point optical triangulation principle
for measuring the distance from a target object and are competitive in terms of
response time, resolution, beam width, power consumption, and size. Once inte-
grated with IMUs, they have shown their potential in estimating IFD in two instances
during the gait cycle (Trojaniello et al. 2014c; Valeri et al. 2016). More recently, time
of flight-IR technology was tested to estimate the IFD with promising results
(Bertuletti et al. 2016). Once integrated with IMUs, the resulting device would be
a fully wearable stand-alone device, which could be used for estimating the gait
parameters described so far.
Direction of Progression and Turning

Most of the current definitions of gait spatial parameters have been based on the
premise that walking occurs along a straight line. However, in analyzing the gait of
subjects with mobility impairments and due to the functional nature of turning, the
quantification of changes of direction becomes of paramount importance (see
clinical applications for details). Stride-by-stride changes of the direction of pro-
gression (DoP) may effectively quantify the change of direction in gait for quanti-
fying either deviations from the straight path or turnings. The changes of the DoP
can be described as the angles between the stride vectors identified from one point of
the foot at initial contact to the same point of the foot at the following initial contact.
Several IMU-based methods have been proposed for the estimate of changes in the
DoP (Mariani et al. 2010; Trojaniello et al. 2014d) or of turning parameters (Salarian
et al. 2009; El-Gohary et al. 2013). While the former are designed to describe mainly
deviations from a straight path and require the IMUs to be positioned on feet or
shanks, the latter are designed to analyze curved paths, and they are generally based
on the angular velocity signals recorded on the trunk or lower back. Some additional
methods were developed exclusively to identify turns in walking. In some cases a
combination of angular velocity and angle around the vertical direction was used
(Novak et al. 2014; Nguyen et al. 2015) or on local magnetic field signals (Fleury
et al. 2007). The reliability of some of these methods in identifying turning occur-
rences was evaluated on the gait of subjects with different pathologies walking at
comfortable and faster speed. It was shown that all tested methods identified
correctly all the turns performed by healthy elderly and Parkinsonians, failed a few
times in identifying the turns performed by choreic subjects and stroke survivors
(Bertoli et al. 2016).
Temporal and Spatial Parameters in the Clinic: Introduction
The locomotor system relies upon a complex musculoskeletal system; it integrates

inputs from various regions of the brain (premotor cortex, motor cortex, cerebellum,
basal ganglia, and brain stem) as well as feedback from the visual, somatosensory,
and vestibular systems to produce and control a stable and highly consistent walking
pattern (Jahn et al. 2008; Takakusaki et al. 2008).
The state of global health, or even age, of a person is conveyed by how they walk.
In fact, gait has been recognized as a marker of global health, and it is a predictor of
health status in older adults (Hausdorff et al. 2001; Studenski et al. 2011). People
tend to walk slower as they get older, those who suffered a stroke may drag one leg,
and those with a knee or hip injury may walk asymmetrically.
Gait performance is also a predictor of survival (Studenski et al. 2011), cognitive
decline (Verghese et al. 2007), and fall status (Beauchet et al. 2009) and is affected
by frailty status (Gill et al. 2001) and the presence of neurological diseases (Fasano
and Bloem 2013; Moon et al. 2016). Moreover, gait disturbances are among the most
common causes of falls in older adults. They often lead to injury, disability, loss of
independence, and markedly impaired quality of life (Jankovic et al. 2001; Sudarsky
2001a, b; Jahn et al. 2010; Salzman 2010).
Among gait parameters, speed is generally used to characterize gait because of its
robust clinimetric properties (Fritz and Lusardi 2009; Bohannon and Glenney 2014).
However, gait disturbances are usually multifactorial in origin and require a com-
prehensive assessment to determine contributing factors and targeted interventions
(Alexander 1996; Sudarsky 2001a). Although gait speed is sensitive to pathological
conditions, it is neither discriminative nor reflective of subtle and selective alter-
ations of gait expressed in response to change in neuropathology in aging and
disease (Stolze et al. 2001; Verghese et al. 2007; Lord et al. 2014). Selective
identification of gait characteristics is therefore critical for discriminating pathology,
identifying specific features of disease progression, and discerning the effect of age
for detection of shared neural correlates. For example, stride time variability (stride-
to-stride fluctuations) has been found to be more sensitive than gait speed in
predicting falls (Lord et al. 2011) and discriminating for detection of genetic
mutation precursor of Parkinson’s disease (Mirelman et al. 2011, 2013).
In addition to clinical trials, physical therapists need innovative technologies to
quickly and efficiently quantify gait abnormalities. IMU-based gait evaluation tools
allow rehabilitation therapy to be targeted on specific gait impairments and provide
evidence that therapy is working (Horak et al. 2015).
Validity Reliability Discriminative Sensitivity to

•Refer to how well •Establish if your ability change
a test measures measures are •Ability to correctly •Ability to measure
what is supposed reliable and can classify subjects changes in time or
to measure. be trusted into different treatment.
•Example: Relative (ICC) categories when Example:
criterior, Example: Intra- true group is Standardized
construct, rater, test-retest known. Response mean
content, and Absolute (SEM, Example: (SRM)
face. SDD) Sensitivity and
Specificity, ROC
curve
Fig. 2 Summary of clinimetric properties
Clinimetrics of Gait Parameters
As in any quantitative clinical application, the clinimetric properties of measures

must be addressed also for the objective measures of gait (de Vet et al. 2003;
Moe-Nilssen et al. 2008; Fig. 2), including those obtained with IMUs. First, validity
is an essential property (i.e., a measurement is valid if it measures what it is intended
to measure); then adequate reliability is required for a consistent and accurate
parameter estimation. Third, the ability to correctly classify subjects into different
categories (discriminative ability) is important for screening or selecting subjects for
inclusion in a study. Fourth, sensitivity to change (can be seen also as longitudinal
validity) is essential to evaluate the effect on the parameter of the stage of the disease,
the intervention, and the environment.
As here we are referring to validity of gait parameter measures with IMUs, and
considering that a gold standard is available (either stereophotogrammetry, pressure
mats, or force platform), criterion validity is the kind of validity to be investigated
(Streiner and Norman 1995; de Vet et al. 2003). Criterion validity examines the
extent to which a measurement instrument provides the same results as the gold
standard. Validity, reliability, discriminative ability, and sensitivity to change are
measurement properties which should be therefore considered for gait parameters to
be used in clinical trials and longitudinal studies and to evaluate the responsiveness
to drugs and or rehabilitation intervention (Fig. 3).
Is Straight-Ahead Gait Enough? Importance of Turning
Gait research has been focused primarily on straight-ahead walking and mostly
restricted to a clinical environment. However, turning is ubiquitous during activities
of daily living. Nearly every task performed during the day requires some amount of
10
Fig. 3 Summary of the literature about which gait parameters, location of sensors, and four clinimetric characteristics for a variety of neurological diseases.
Only papers investigating at least 2 clinimetric characteristics were reported
U. Della Croce et al.
turning (Glaister et al. 2007). Difficulty in turning during gait is a major contributor
to mobility limitations, falls, and reduced quality of life in older people and people
with movement disorders. Turning gradually becomes more difficult as we age due
to increasing sensorimotor impairments. The ability to modify our locomotor trajec-
tory by turning safely is important for functional independence but, surprisingly,
much more difficult for the nervous system to control than straight-ahead walking.
Moreover, falls during turning are particularly dangerous because they usually result
in contact of the femur with the ground, which results in eight times more hip
fractures compared with falls during straight walking (Cummings and Nevitt 1994;
Feldman and Robinovitch 2007).
Laboratory analysis of turns in gait has been focused on stride parameters
(Huxham et al. 2006), muscle activity (Courtine and Schieppati 2003), and coordi-
nation strategies used during turning (Crenna et al. 2007) using cumbersome and
expensive motion analysis systems, including EMG and force platforms. Clinically,
research has focused on assessing turning ability with the timed up and go (TUG)
test (Thigpen et al. 2000; Dite and Temple 2002) that includes a 180 turn or with the
Berg balance scale or Tinetti balance and gait test that includes a 360 turn. Other
clinicians use a stopwatch to time a 360 turn and count the number of steps during a
turn (Visser et al. 2007). Still, a critical barrier to neurologists and therapists is the
lack of a measure of turn quality in natural, functional settings and across long
periods of time to monitor risk for falls, fluctuations across the day, response to
interventions, etc. since single event mobility measures in the clinic do not accu-
rately reflect functional mobility.
A growing body of literature has been focusing on characterizing turning while
walking with IMUs in different neurological population and in healthy elderly
(Zampieri et al. 2010; Herman et al. 2011; Spain et al. 2012; Mirelman et al. 2014;
Fino et al. 2016). It has been suggested that neural systems related with turning may
be more vulnerable to impairments than those related with straight-ahead gait since
turning involves more interlimb coordination, more coupling between posture and
gait, and modifications of locomotor patterns requiring frontal lobe cognitive and
executive function that plays a role in postural transitions (Herman et al. 2011; King
et al. 2012; Mirelman et al. 2014).
Few studies, using IMUs in the clinic, have shown that patients with Parkinson
disease (PD) or multiple sclerosis (MS) exhibit abnormal turning characteristics even
when their straight-ahead walking is unimpaired (Zampieri et al. 2010; Spain et al.
2012). In addition, it has been found that turning parameters measurable with IMUs
in the clinic can distinguish between mild PD and control subjects, even when
clinical measures of balance from the Berg balance score or the Tinetti balance
and gait test do not (King et al. 2012). In addition, a novel method to characterize
spin or step turns by using three IMUs has been proposed in a clinical setting (Fino
et al. 2015); such findings are of particular interest considering that higher preva-
lence of spin turns may be associated to falls in an elderly population (Yamada et al.
2012).
Fig. 4 IMU-based continuous monitoring of mobility
Objective Measures of Gait and Turning in Daily Life
The assessment of mobility in the clinic might not adequately reflect mobility
function during daily life. In fact, patients appear to walk better in the clinical facility
compared to what the caregivers report about their daily lives, and this is known as
the “white coat effect.” It has been found that even an identical, instrumented test of
mobility conducted in the home environment resulted in slower turns and walking
than when conducted in a laboratory (Zampieri et al. 2011; Weiss et al. 2013).
A single, or sparsely spaced, measure of mobility cannot assess diurnal, day-to-
day, or other clinically relevant windows of change, such as daily medication-
induced motor fluctuations or effects of fatigue (Horak et al. 2015). The ambulatory
assessment of mobility during daily life activities can provide measures of mobility
function outside the clinic, similar to how a heart rate Holter monitor evaluates
cardiac function over days and weeks.
The use of wearable, light-weight IMUs allows the characterization of quality of
mobility outside the laboratory environment, for example, at home (Fig. 4). Novel
measures determined from both accelerometers and gyroscopes enable a detailed
analysis of gait bouts and turnings over multiple days of continuous recording
(El-Gohary et al. 2013; Weiss et al. 2013, 2014, 2015; Schwenk et al. 2014; Mancini
et al. 2015, 2016; van Schooten et al. 2015; Del Din et al. 2016). Similarly to the use
in the laboratory or clinic, also for continuous monitoring, if a single IMU is used, it
is generally placed on the lower back, while, if multiple IMUs are used, they are
placed either on the shanks or on the feet (Mancini et al. 2015, 2016). However,
continuous monitoring of gait calls for a compromise between accuracy and practi-
cality; therefore, the use of multiple IMUs should be carefully evaluated.
When measuring mobility in the home environment, it is important to consider
the added value of “quality” of straight-ahead walking and turning to predict fallers
and distinguish subtle impairments in neurological diseases, compared to the amount
or quantity of mobility (Del Din et al. 2016). For instance, greater step-to-step
variability was found in retrospective fallers (healthy and subjects with PD), com-
pared to non-fallers, over 3 days of continuous monitoring, although the number of
steps per day was similar between the two groups (Weiss et al. 2013, 2014). In
addition, among the subjects with PD who did not report falls in the previous year,
IMU-derived parameters predicted the time to first fall (Weiss et al. 2014). Generally,
the added value of objective measure of mobility over multiple days compared to
clinical tests has been recognized a powerful tool in predicting falls in healthy or
pathological populations (Schwenk et al. 2014; van Schooten et al. 2015). However,
a consensus on a specific set of measures and on a definition of gait bout at home is
still missing (Del Din et al. 2016).
Quality of turning (quantified by turn duration, peak velocity, and number of steps
to complete a turn) is also significantly compromised in recurrent fallers, compared
to non-fallers, as recently reported (Mancini et al. 2016). Specifically, recurrent
fallers (two or more falls in the previous year) showed a significantly longer mean
turn duration, a slower mean turn peak speed, and a larger number of steps to turn
compared to non-fallers. Unlike quality of turning, quantity of turning quantified by
active rate (% of time of walking or turning compared to full time of monitoring) and
mean number of turns/hour were not significantly different across the three groups.
In addition, future falls in this cohort were related to an increased variability of
number of steps to turn (Fig. 5).
Future Directions
Potential for Real-Time IMU-Based Biofeedback and Smartphone

Applications
An advantage of IMUs is that they can be used to provide real-time information to

the user and to augment feedback while walking or performing other functional
tasks. Indeed, in rehabilitation, an important role of the therapist is to provide
patients with feedback about their performance during the rehabilitation session.
Biofeedback is a technique that provides information about physiological func-
tions in real time that would otherwise be unknown to patients, complementing the
internal feedback already present and reinforcing the patient’s weaker or absent
sensory signals, in some cases acting as an additional sense (Tate and Milner
2010). An IMU-based biofeedback approach for gait retraining has the potential to
detect gait abnormalities in real time, to respond with immediate instructions to
modify specific gait patterns, and to quantify response and progress to both the
therapist and the patient. Biofeedback systems use electronic equipment to provide
Fig. 5 Variability of mean number of steps/turns in a representative non-faller and faller among
hours and the 7 days of monitoring (left panel) versus group (right panel). Non-fallers and fallers are
defined based on prospective falls in 6 months following the 7-day monitoring with IMUs (From
Mancini et al. 2016)
individuals with visual, auditory, and/or tactile signals regarding internal physiolog-
ical events, both normal and abnormal. Using a movable graph on a computer screen
or an audio signal, the system provides individuals with visual or audio feedback
with instructions to enhance their movement. An audio-biofeedback system, com-
bining two IMUs mounted on the feet and a smartphone application, was designed
and tested for gait training in subjects affected by PD (CuPiD) (Ginis et al. 2016).
Results from a pilot randomized controlled trial (n = 40) showed that the use of the
audio-feedback system over 6 weeks at home provided superior results over the
training assisted by a physical therapist, in enhancing gait and balance control, as the
active group (PD group using the proposed system) trained less but improved
equally or more than the control group (PD group following rehabilitation advices),
suggesting that real-time feedback stimulates corrective actions and promotes self-
efficacy in achieving optimal performance. Smartphone-based applications in gait
rehabilitation, both in monitoring mobility and in delivering training, may have a
profound impact for patients with movement disorders and elderly at risk for falls
(Mellone et al. 2012; Mannini and Sabatini 2015).
Cross-References

▶ Gait During Real World Challenges: Gait Initiation, Gait Termination, Accelera-
tion, Deceleration, Turning, Slopes and Stairs
▶ Gait Retraining for Balance Improvements
Analysis
▶ Three-Dimensional Human Kinematics Estimation Using Magneto-Inertial Mea-
surement Units
References
Alexander NB (1996) Gait disorders in older adults. J Am Geriatr Soc 44:434–451
Aminian K, Najafi B, Bula C, Leyvraz PF, Robert P (2002) Spatio-temporal parameters of gait
measured by an ambulatory system using miniature gyroscopes. J Biomech 35:689–699
Bamberg SJ, Benbasat AY, Scarborough DM, Krebs DE, Paradiso JA (2008) Gait analysis using a
shoe-integrated wireless sensor system. IEEE Trans Inf Technol Biomed 12:413–423
Barrett RS, Mills PM, Begg RK (2010) A systematic review of the effect of ageing and falls history
on minimum foot clearance characteristics during level walking. Gait Posture 32:429–435
Beauchet O, Allali G, Annweiler C, Bridenbaugh S, Assal F, Kressig RW, Herrmann FR (2009) Gait
variability among healthy adults: low and high stride-to-stride variability are both a reflection of
gait stability. Gerontology 55:702–706
Bertoli M, Cereatti A, Trojaniello D, Ravaschio A, Della Croce U (2016) The identification of
multiple U-turns in gait: comparison of four trunk IMU-based methods. Proceedings
BODYNETS Submitted
Bertuletti S, Cereatti A, Caldara M, Della Croce U (2016) A proximity sensor for the measurement
of the inter-foot distance in static and dynamic tasks. Gait Posture 49(Suppl 1):S15
Best R, Begg R (2008) A method for calculating the probability of tripping while walking.
J Biomech 41:1147–1151
Bohannon RW, Glenney SS (2014) Minimal clinically important difference for change in comfort-
able gait speed of adults with pathology: a systematic review. J Eval Clin Pract 20:295–300
Brach JS, Berthold R, Craik R, VanSwearingen JM, Newman AB (2001) Gait variability in
community-dwelling older adults. J Am Geriatr Soc 49:1646–1650
Brach JS, Berlin JE, VanSwearingen JM, Newman AB, Studenski SA (2005) Too much or too little
step width variability is associated with a fall history in older persons who walk at or near
normal gait speed. J Neuroeng Rehabil 2:21
Bregou Bourgeois A, Mariani B, Aminian K, Zambelli PY, Newman CJ (2014) Spatio-temporal gait
analysis in children with cerebral palsy using, foot-worn inertial sensors. Gait Posture
39:436–442
Cereatti A, Trojaniello D, Della Croce U (2015) Accurately measuring human movement using
magneto-inertial sensors: techniques and challenges. In: IEEE international symposium on
Inertial Sensors and Systems (ISISS) proceedings, pp 1–4
Courtine G, Schieppati M (2003) Human walking along a curved path. II. Gait features and EMG
patterns. Eur J Neurosci 18:191–205
Crenna P, Carpinella I, Rabuffetti M, Calabrese E, Mazzoleni P, Nemni R, Ferrarin M (2007) The

association between impaired turning and normal straight walking in Parkinson’s disease. Gait
Posture 26:172–178
Cummings SR, Nevitt MC (1994) Non-skeletal determinants of fractures: the potential importance
of the mechanics of falls. Study of Osteoporotic Fractures Research Group. Osteoporos Int
4(Suppl 1):67–70
Curtze C, Nutt JG, Carlson-Kuhta P, Mancini M, Horak FB (2015) Levodopa is a double-edged
sword for balance and gait in people with Parkinson’s disease. Mov Dis 30:1361–1370
Dadashi F, Mariani B, Rochat S, Bula CJ, Santos-Eggimann B, Aminian K (2013) Gait and foot
clearance parameters obtained using shoe-worn inertial sensors in a large-population sample of
older adults. Sensors 14:443–457
Dalton A, Khalil H, Busse M, Rosser A, van Deursen R, Olaighin G (2013) Analysis of gait and
balance through a single triaxial accelerometer in presymptomatic and symptomatic
Huntington’s disease. Gait Posture 37:49–54
de Bruin ED, Hubli M, Hofer P, Wolf P, Murer K, Zijlstra W (2012) Validity and reliability of
accelerometer-based gait assessment in patients with diabetes on challenging surfaces. J Aging
Res 2012:954378
de Vet HC, Terwee CB, Bouter LM (2003) Current challenges in clinimetrics. J Clin Epidemiol
56:1137–1141
Del Din S, Godfrey A, Rochester L (2015) Validation of an accelerometer to quantify a compre-
hensive battery of gait characteristics in healthy older adults and Parkinson’s disease: toward
clinical and at home use. IEEE J Biomed Health Inform
Del Din S, Godfrey A, Mazza C, Lord S, Rochester L (2016) Free-living monitoring of Parkinson’s
disease: lessons from the field. Mov Dis 31:1293–1313
Dite W, Temple VA (2002) Development of a clinical measure of turning for older adults. Am
J Phys Med Rehabil/Assoc Acad Physiatrists 81:857–866; quiz 867–858
El-Gohary M, Pearson S, McNames J, Mancini M, Horak F, Mellone S, Chiari L (2013) Continuous
monitoring of turning in patients with movement disability. Sensors 14:356–369
Esser P, Dawes H, Collett J, Feltham MG, Howells K (2011) Assessment of spatio-temporal gait
parameters using inertial measurement units in neurological populations. Gait Posture
34:558–560
Esser P, Dawes H, Collett J, Feltham MG, Howells K (2012) Validity and inter-rater reliability of
inertial gait measurements in Parkinson’s disease: a pilot study. J Neurosci Methods
205:177–181
Fasano A, Bloem BR (2013) Gait disorders. Continuum 19:1344–1382
Feldman F, Robinovitch SN (2007) Reducing hip fracture risk during sideways falls: evidence in
young adults of the protective effects of impact to the hands and stepping. J Biomech
40:2612–2618
Fino PC, Frames CW, Lockhart TE (2015) Classifying step and spin turns using wireless gyro-
scopes and implications for fall risk assessments. Sensors 15:10676–10685
Fino PC, Nussbaum MA, Brolinson PG (2016) Locomotor deficits in recently concussed athletes
and matched controls during single and dual-task turning gait: preliminary results. J Neuroeng
Rehabil 13:65
Fleury A, Noury N, Vuillerme N (2007) A fast algorithm to track changes of direction of a person
using magnetometers. In: Annual international conference of the IEEE engineering in medicine
and biology – proceeding, pp 2311–2314
Foxlin E (2005) Pedestrian tracking with shoe-mounted inertial sensors. IEEE Comput Graph Appl
25:38–46
32:46–49
Gill TM, McGloin JM, Gahbauer EA, Shepard DM, Bianco LM (2001) Two recruitment strategies
for a clinical trial of physically frail community-living older persons. J Am Geriatr Soc
49:1039–1045
Ginis P, Nieuwboer A, Dorfman M, Ferrari A, Gazit E, Canning CG, Rocchi L, Chiari L, Hausdorff
JM, Mirelman A (2016) Feasibility and effects of home-based smartphone-delivered automated
feedback training for gait in people with Parkinson’s disease: a pilot randomized controlled trial.
Parkinsonism Relat Disord 22:28–34
Glaister BC, Bernatz GC, Klute GK, Orendurff MS (2007) Video task analysis of turning during
activities of daily living. Gait Posture 25:289–294
Goldstein M, Harper DC (2001) Management of cerebral palsy: equinus gait. Dev Med Child
Neurol 43:563–569
Gonzalez RC, Alvarez D, Lopez AM, Alvarez JC (2007) Modified pendulum model for mean step
length estimation. In: 2007 29th annual international conference of the IEEE engineering in
medicine and biology society, pp 1371–1374
Gonzalez RC, Lopez AM, Rodriguez-Uria J, Alvarez D, Alvarez JC (2010) Real-time gait event
detection for normal subjects from lower trunk accelerations. Gait Posture 31:322–325
Greene BR, McGrath D, O’Neill R, O’Donovan KJ, Burns A, Caulfield B (2010) An adaptive
gyroscope-based algorithm for temporal gait analysis. Med Biol Eng Comput 48:1251–1260
Han J, Jeon HS, Yi WJ, Jeon BS, Park KS (2009) Adaptive windowing for gait phase discrimination
in parkinsonian gait using 3-axis acceleration signals. Med Biol Eng Comput 47:1155–1164
Hanlon M, Anderson R (2009) Real-time gait event detection using wearable sensors. Gait Posture
30:523–527
Hartmann A, Luzi S, Murer K, de Bie RA, de Bruin ED (2009) Concurrent validity of a trunk
tri-axial accelerometer system for gait analysis in older adults. Gait Posture 29:444–448
Hausdorff JM, Rios DA, Edelberg HK (2001) Gait variability and fall risk in community-living
older adults: a 1-year prospective study. Arch Phys Med Rehabil 82:1050–1056
Herman T, Giladi N, Hausdorff JM (2011) Properties of the ‘timed up and go’ test: more than meets
the eye. Gerontology 57:203–210
Horak F, King L, Mancini M (2015) Role of body-worn movement monitor technology for balance
and gait rehabilitation. Phys Ther 95:461–470
Houdijk H, Appelman FM, Van Velzen JM, Van der Woude LH, Van Bennekom CA (2008) Validity
of DynaPort GaitMonitor for assessment of spatiotemporal parameters in amputee gait.
J Rehabil Res Dev 45:1335–1342
Howell DR, Oldham JR, DiFabio M, Vallabhajosula S, Hall EE, Ketcham CJ, Meehan WP 3rd,
Buckley TA (2016) Single-task and dual-task Gait among collegiate athletes of different sport
classifications: implications for concussion management. J Appl Biomech 1–25
Hundza SR, Hook WR, Harris CR, Mahajan SV, Leslie PA, Spani CA, Spalteholz LG, Birch BJ,
Commandeur DT, Livingston NJ (2014) Accurate and reliable gait cycle detection in
Parkinson’s disease. IEEE Trans Neural Syst Rehabil Eng 22:127–137
Hung TN, Suh YS (2013) Inertial sensor-based two feet motion tracking for gait analysis. Sensors
13:5614–5629
Huxham F, Gong J, Baker R, Morris M, Iansek R (2006) Defining spatial parameters for non-linear
walking. Gait Posture 23:159–163
Iosa M, Picerno P, Paolucci S, Morone G (2016) Wearable inertial sensors for human movement
analysis. Expert Rev Med Devices 13:641–659
Jahn K, Deutschlander A, Stephan T, Kalla R, Hufner K, Wagner J, Strupp M, Brandt T (2008)
Supraspinal locomotor control in quadrupeds and humans. Prog Brain Res 171:353–362
Jahn K, Zwergal A, Schniepp R (2010) Gait disturbances in old age: classification, diagnosis, and
treatment from a neurological perspective. Deutsches Arzteblatt Int 107:306–315; quiz 316
Jankovic J, Nutt JG, Sudarsky L (2001) Classification, diagnosis, and etiology of gait disorders.
Adv Neurol 87:119–133
Kavanagh JJ, Menz HB (2008) Accelerometry: a technique for quantifying movement patterns
during walking. Gait Posture 28:1–15
Khandoker AH, Lynch K, Karmakar CK, Begg RK, Palaniswami M (2010) Toe clearance and
velocity profiles of young and elderly during walking on sloped surfaces. J Neuroeng Rehabil
7:18
King LA, Mancini M, Priest K, Salarian A, Rodrigues-de-Paula F, Horak F (2012) Do clinical scales
of balance reflect turning abnormalities in people with Parkinson’s disease? J Neurol Phys Ther:
JNPT 36:25–31
Kose A, Cereatti A, Della Croce U (2012) Bilateral step length estimation using a single inertial
measurement unit attached to the pelvis. J Neuroeng Rehabil 9:9
Krebs DE, Goldvasser D, Lockert JD, Portney LG, Gill-Body KM (2002) Is base of support greater
in unsteady gait? Phys Ther 82:138–147
Lai DTH, Begg R, Charry E, Palaniswami M, Hill K (2008) Measuring toe clearance using a
wireless inertial sensing device. In: ISSNIP 2008 – proceedings of the 2008 international
conference on intelligent sensors, sensor networks and information processing, pp 375–380
Lau H, Tong K (2008) The reliability of using accelerometer and gyroscope for gait event
identification on persons with dropped foot. Gait Posture 27:248–257
Lee JK, Park EJ (2011) Quasi real-time gait event detection using shank-attached gyroscopes. Med
Biol Eng Comput 49:707–712
Lopez-Meyer P, Fulk GD, Sazonov ES (2011) Automatic detection of temporal gait parameters in
poststroke individuals. IEEE Trans Inf Technol Biomed 15:594–601
Lord S, Howe T, Greenland J, Simpson L, Rochester L (2011) Gait variability in older adults: a
structured review of testing protocol and clinimetric properties. Gait Posture 34:443–450
Lord S, Galna B, Coleman S, Yarnall A, Burn D, Rochester L (2014) Cognition and gait show a
selective pattern of association dominated by phenotype in incident Parkinson’s disease. Front
Aging Neurosci 6:249
Mancini M, El-Gohary M, Pearson S, McNames J, Schlueter H, Nutt JG, King LA, Horak FB
(2015) Continuous monitoring of turning in Parkinson’s disease: rehabilitation potential.
NeuroRehabilitation 37:3–10
Mancini M, Schlueter H, El-Gohary M, Mattek N, Duncan C, Kaye J, Horak FB (2016) Continuous
monitoring of turning mobility and its association to falls and cognitive function: a pilot study.
J Gerontol A Biol Sci Med Sci 71:1102–1108
Mannini A, Sabatini AM (2012) Gait phase detection and discrimination between walking-jogging
activities using hidden Markov models applied to foot motion data from a gyroscope. Gait
Posture 36:657–661
Mannini A, Sabatini AM (2014) Walking speed estimation using foot-mounted inertial sensors:
comparing machine learning and strap-down integration methods. Med Eng Phys
36:1312–1321
Mannini A, Sabatini AM (2015) A smartphone-centered wearable sensor network for fall risk
assessment in the elderly. In: BodyNets '15 proceedings of the 10th EAI international conference
on body area networks, pp 167–172
Mariani B, Hoskovec C, Rochat S, Bula C, Penders J, Aminian K (2010) 3D gait assessment in
young and elderly subjects using foot-worn inertial sensors. J Biomech 43:2999–3006
Mariani B, Lisco G, Aminian K (2012a) New gait analysis method based on wiimote technology
and fusion with inertial sensors. In: Proceedings 1st joint world congress ISPGR & Gait and
mental function
Mariani B, Rochat S, Bula CJ, Aminian K (2012b) Heel and toe clearance estimation for gait
analysis using wireless inertial sensors. IEEE Trans Biomed Eng 59:3162–3168
Mariani B, Rouhani H, Crevoisier X, Aminian K (2013) Quantitative estimation of foot-flat and
stance phase of gait using foot-worn inertial sensors. Gait Posture 37:229–234
McCamley J, Donati M, Grimpampi E, Mazza C (2012) An enhanced estimate of initial contact and
final contact instants of time using lower trunk inertial sensor data. Gait Posture 36:316–318
McGrath D, Greene BR, Walsh C, Caulfield B (2011) Estimation of minimum ground clearance
(MGC) using body-worn inertial sensors. J Biomech 44:1083–1088
Mellone S, Tacconi C, Chiari L (2012) Validity of a smartphone-based instrumented timed up and
go. Gait Posture 36:163–165
Mirelman A, Gurevich T, Giladi N, Bar-Shira A, Orr-Urtreger A, Hausdorff JM (2011) Gait
alterations in healthy carriers of the LRRK2 G2019S mutation. Ann Neurol 69:193–197
Mirelman A, Heman T, Yasinovsky K, Thaler A, Gurevich T, Marder K, Bressman S, Bar-Shira A,

Orr-Urtreger A, Giladi N, Hausdorff JM, Consortium LAJ (2013) Fall risk and gait in
Parkinson’s disease: the role of the LRRK2 G2019S mutation. Mov Dis 28:1683–1690
Mirelman A, Weiss A, Buchman AS, Bennett DA, Giladi N, Hausdorff JM (2014) Association
between performance on timed up and go subtasks and mild cognitive impairment: further
insights into the links between cognitive and motor function. J Am Geriatr Soc 62:673–678
Mizuike C, Ohgi S, Morita S (2009) Analysis of stroke patient walking dynamics using a tri-axial
accelerometer. Gait Posture 30:60–64
Moe-Nilssen R, Helbostad JL (2004) Estimation of gait cycle characteristics by trunk
accelerometry. J Biomech 37:121–126
Moe-Nilssen R, Nordin E, Lundin-Olsson L, Work Package 3 of European Community Research
Network Prevention of Falls Network E (2008) Criteria for evaluation of measurement proper-
ties of clinical balance measures for use in fall prevention studies. J Eval Clin Pract 14:236–240
Moon Y, Sung J, An R, Hernandez ME, Sosnoff JJ (2016) Gait variability in people with
neurological disorders: a systematic review and meta-analysis. Hum Mov Sci 47:197–208
Nagano H, Begg RK, Sparrow WA, Taylor S (2011) Ageing and limb dominance effects on foot-
ground clearance during treadmill and overground walking. Clin Biomech 26:962–968
Nguyen HP, Ayachi F, Lavigne-Pelletier C, Blamoutier M, Rahimi F, Boissy P, Jog M, Duval
C (2015) Auto detection and segmentation of physical activities during a timed-up-and-go
(TUG) task in healthy older adults using multiple inertial sensors. J Neuroeng Rehabil 12:36
Novak D, Gorsic M, Podobnik J, Munih M (2014) Toward real-time automated detection of turns
during gait using wearable inertial measurement units. Sensors 14:18800–18822
Owings TM, Grabiner MD (2004) Step width variability, but not step length variability or step time
variability, discriminates gait of healthy young and older adults during treadmill locomotion.
Perry J, Burnfield JM (1992) Gait analysis: normal and pathological function
Peruzzi A, Della Croce U, Cereatti A (2011) Estimation of stride length in level walking using an
inertial measurement unit attached to the foot: a validation of the zero velocity assumption
during stance. J Biomech 44:1991–1994
Rebula JR, Ojeda LV, Adamczyk PG, Kuo AD (2013) Measurement of foot placement and its
variability with inertial sensors. Gait Posture 38:974–980
Riley PO, Benda BJ, Gill-Body KM, Krebs DE (1995) Phase plane analysis of stability in quiet
standing. J Rehabil Res Dev 32:227–235
Sabatini AM (2011) Estimating three-dimensional orientation of human body parts by inertial/
magnetic sensing. Sensors 11:1489–1525
Sabatini AM, Martelloni C, Scapellato S, Cavallo F (2005) Assessment of walking features from
foot inertial sensing. IEEE Trans Biomed Eng 52:486–494
Salarian A, Russmann H, Vingerhoets FJ, Dehollain C, Blanc Y, Burkhard PR, Aminian K (2004)
Gait assessment in Parkinson’s disease: toward an ambulatory system for long-term monitoring.
IEEE Trans Biomed Eng 51:1434–1443
Salarian A, Zampieri C, Horak FB, Carlson-Kuhta P, Nutt JG, Aminian K (2009) Analyzing
180 degrees turns using an inertial system reveals early signs of progression of Parkinson’s
disease. In: Conference proceedings: annual international conference of the IEEE engineering in
medicine and biology society IEEE engineering in medicine and biology society annual
conference 2009, pp 224–227
Salarian A, Horak FB, Zampieri C, Carlson-Kuhta P, Nutt JG, Aminian K (2010) iTUG, a sensitive
and reliable measure of mobility. IEEE Trans Neural Syst Rehabil Eng 18:303–310
Salzman B (2010) Gait and balance disorders in older adults. Am Fam Physician 82:61–68
Sant’Anna A, Salarian A, Wickstrom N (2011) A new measure of movement symmetry in early
Parkinson’s disease patients using symbolic processing of inertial sensor data. IEEE Trans
Biomed Eng 58:2127–2135
Schwenk M, Hauer K, Zieschang T, Englert S, Mohler J, Najafi B (2014) Sensor-derived physical
activity parameters can predict future falls in people with dementia. Gerontology 60:483–492
Selles RW, Formanoy MA, Bussmann JB, Janssens PJ, Stam HJ (2005) Automated estimation of
initial and terminal contact timing using accelerometers; development and validation in trans-
tibial amputees and controls. IEEE Trans Neural Syst Rehabil Eng 13:81–88
Shin SH, Park CG (2011) Adaptive step length estimation algorithm using optimal parameters and
movement status awareness. Med Eng Phys 33:1064–1071
Spain RI, St George RJ, Salarian A, Mancini M, Wagner JM, Horak FB, Bourdette D (2012) Body-
worn motion sensors detect balance and gait deficits in people with multiple sclerosis who have
normal walking speed. Gait Posture 35:573–578
Stolze H, Kuhtz-Buschbeck JP, Drucke H, Johnk K, Illert M, Deuschl G (2001) Comparative
analysis of the gait disorder of normal pressure hydrocephalus and Parkinson’s disease. J Neurol
Neurosurg Psychiatry 70:289–297
Streiner DL, Norman GR (1995) Health measurement scales. In: A practical guide to their
development and use. Oxford University Press, Oxford
Studenski S, Perera S, Patel K, Rosano C, Faulkner K, Inzitari M, Brach J, Chandler J, Cawthon P,
Connor EB, Nevitt M, Visser M, Kritchevsky S, Badinelli S, Harris T, Newman AB, Cauley J,
Ferrucci L, Guralnik J (2011) Gait speed and survival in older adults. JAMA 305:50–58
Sudarsky L (2001a) Gait disorders: prevalence, morbidity, and etiology. Adv Neurol 87:111–117
Sudarsky L (2001b) Neurologic disorders of gait. Curr Neurol Neurosci Rep 1:350–356
Takakusaki K, Tomita N, Yano M (2008) Substrates for normal gait and pathophysiology of gait
disturbances with respect to the basal ganglia dysfunction. J Neurol 255(Suppl 4):19–29
Tate JJ, Milner CE (2010) Real-time kinematic, temporospatial, and kinetic biofeedback during gait
retraining in patients: a systematic review. Phys Ther 90:1123–1134
Thigpen MT, Light KE, Creel GL, Flynn SM (2000) Turning difficulty characteristics of adults aged
65 years or older. Phys Ther 80:1174–1187
Trojaniello D, Cereatti A, Pelosin E, Avanzino L, Mirelman A, Hausdorff JM, Della Croce
U (2014a) Estimation of step-by-step spatio-temporal parameters of normal and impaired gait
using shank-mounted magneto-inertial sensors: application to elderly, hemiparetic, parkinsonian
and choreic gait. J Neuroeng Rehabil 11:152
Trojaniello D, Cereatti A, Della Croce U (2014b) Accuracy, sensitivity and robustness of five
different methods for the estimation of gait temporal parameters using a single inertial sensor
mounted on the lower trunk. Gait Posture 40:487–492
Trojaniello D, Cereatti A, Bourke A, Aminian K, Della Croce U (2014c) A wearable system for the
measurement of the inter-foot distance during gait. In: 20th IMEKO T4 international sympo-
sium, pp 765–769.
Trojaniello D, Cereatti A, Ravaschio A, Bandettini M, Della Croce U (2014d) Assessment of gait
direction changes during straight-ahead walking in healthy elderly and Huntington disease
patients using a shank worn MIMU. In: Conference proceedings: annual international confer-
ence of the IEEE engineering in medicine and biology society IEEE engineering in medicine
and biology society annual conference 2014, pp 2508–2511
Trojaniello D, Ravaschio A, Hausdorff JM, Cereatti A (2015a) Comparative assessment of different
methods for the estimation of gait temporal parameters using a single inertial sensor: application
to elderly, post-stroke, Parkinson’s disease and Huntington’s disease subjects. Gait Posture
42:310–316
Trojaniello D, Cereatti A, Della Croce U (2015b) Foot clearance estimation during overground
walking and vertical obstacle passing using shank-mounted MIMUs in healthy and pathological
subjects. In: Conference proceedings: annual international conference of the IEEE engineering
in medicine and biology society IEEE engineering in medicine and biology society annual
conference 2015, pp 5505–5508
Valeri N, Trojaniello D, Cereatti A, Aminian K, Della Croce U (2016) Inter-foot distance measured
during gait with wearable IMU and IRR sensors. In: Proceedings of the GNB2016 conference
van Schooten KS, Pijnappels M, Rispens SM, Elders PJ, Lips P, van Dieen JH (2015) Ambulatory
fall-risk assessment: amount and quality of daily-life gait predict falls in older adults. J Gerontol
A Biol Sci Med Sci
Veltink PH, Slycke P, Hemssems J, Buschman R, Bultstra G, Hermens H (2003) Three dimensional
inertial sensing of foot movements for automatic tuning of a two-channel implantable drop-foot
stimulator. Med Eng Phys 25:21–28
cognitive decline and dementia. J Neurol Neurosurg Psychiatry 78:929–935
Visser JE, Voermans NC, Oude Nijhuis LB, van der Eijk M, Nijk R, Munneke M, Bloem BR (2007)
Quantification of trunk rotations during turning and walking in Parkinson’s disease. Clin
Wada C, Ikeda S, Wada F, Hachisuka K, Ienaga T, Kimuro Y, Tsuji T (2012) Improvement study for
measurement accuracy on wireless shoe-type measurement device to support walking rehabil-
itation. In: Proceedings ICME, pp 471–474
Weiss A, Brozgol M, Dorfman M, Herman T, Shema S, Giladi N, Hausdorff JM (2013) Does the
evaluation of gait quality during daily life provide insight into fall risk? A novel approach using
3-day accelerometer recordings. Neurorehabil Neural Repair 27:742–752
Weiss A, Herman T, Giladi N, Hausdorff JM (2014) Objective assessment of fall risk in Parkinson’s
disease using a body-fixed sensor worn for 3 days. PLoS One 9:e96675
Weiss A, Herman T, Giladi N, Hausdorff JM (2015) New evidence for gait abnormalities among
Parkinson’s disease patients who suffer from freezing of gait: insights using a body-fixed sensor
worn for 3 days. J Neural Transm 122:403–410
Winter DA (1992) Foot trajectory in human gait: a precise and multifactorial motor control task.
Phys Ther 72:45–53; discussion 54–46
Yamada M, Higuchi T, Mori S, Uemura K, Nagai K, Aoyama T, Ichihashi N (2012) Maladaptive
turning and gaze behavior induces impaired stepping on multiple footfall targets during gait in
older individuals who are at high risk of falling. Arch Gerontol Geriatr 54:e102–e108
Yang CC, Hsu YL, Shih KS, Lu JM (2011) Real-time gait cycle parameter recognition using a
wearable accelerometry system. Sensors 11:7314–7326
Yang S, Zhang JT, Novak AC, Brouwer B, Li Q (2013) Estimation of spatio-temporal parameters
for post-stroke hemiparetic gait using inertial sensors. Gait Posture 37:354–358
Yuwono M, Su SW, Moulton BD, Nguyen HT (2012) Gait cycle spectrogram analysis using a torso-
attached inertial sensor. In: Conference proceedings: annual international conference of the
IEEE engineering in medicine and biology society IEEE engineering in medicine and biology
society annual conference 2012, pp 6539–6542
Zampieri C, Salarian A, Carlson-Kuhta P, Aminian K, Nutt JG, Horak FB (2010) The instrumented
timed up and go test: potential outcome measure for disease modifying therapies in Parkinson’s
disease. J Neurol Neurosurg Psychiatry 81:171–176
Zampieri C, Salarian A, Carlson-Kuhta P, Nutt JG, Horak FB (2011) Assessing mobility at home in
people with early Parkinson’s disease using an instrumented timed up and go test. Parkinsonism
Relat Disord 17:277–280
Zijlstra W, Hof AL (2003) Assessment of spatio-temporal gait parameters from trunk accelerations
during human walking. Gait Posture 18:1–10
Zok M, Mazza C, Della Croce U (2004) Total body Centre of mass displacement estimated using
ground reactions during transitory motor tasks: application to step ascent. Med Eng Phys
26:791–798
Physics-Based Models for Human Gait
Analysis
Petrissa Zell, Bastian Wandt, and Bodo Rosenhahn
Abstract
This chapter deals with fundamental methods as well as current research on
physics-based human gait analysis. We present valuable concepts that allow
efficient modeling of the kinematics and the dynamics of the human body. The
resulting physical model can be included in an optimization-based framework. In
this context, we show how forward dynamics optimization can be used to
determine the producing forces of gait patterns.
To present a current subject of research, we provide a description of a 2D
physics-based statistical model for human gait analysis that exploits parameter
learning to estimate unobservable joint torques and external forces directly from
motion input. The robustness of this algorithm with respect to occluded joint
trajectories is shown in a short experiment. Furthermore, we present a method that
uses the former techniques for video-based gait analysis by combining them with
a nonrigid structure from motion approach. To examine the applicability of this
method, a brief evaluation of the performance regarding joint torque and ground
reaction force estimation is provided.
Keywords
Computer vision • Human motion analysis • Physics-based simulation • Forward
dynamics optimization • Data-driven regression • 3D motion reconstruction •
Video-based force estimation
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Models and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
P. Zell (*) • B. Wandt • B. Rosenhahn

Institut für Informationsverarbeitung, Leibniz Universität Hannover, Hannover, Germany
e-mail: zell@tnt.uni-hannover.de; wandt@tnt.uni-hannover.de; rosenhahn@tnt.uni-hannover.de

DOI 10.1007/978-3-319-30808-1_164-1
2 P. Zell et al.
Kinematics and Dynamics of Physical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Optimization-Based Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Learning Force Patterns in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Video-Based Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Introduction
The human locomotor system is a complex construction consisting of the skeleton,

the nervous system, muscles, tendons, and ligaments. Its functionality is a basic
human need and the focus of numerous biomechanical studies. In the course of these
studies, researchers require a measure to quantify healthiness of movement. A
mathematically straightforward approach to describing the action of the neuromus-
cular system is to compute the net joint moments acting on the segments through
inverse dynamics analysis. These net moments form a first approximation to the
analysis of stress at the skeletal joint and are the foundation for classifying human
motion. Let us consider an isolated joint in the human body. It might be effected by
the pressure of neighboring bone segments, muscular exertion, and the strain of
tendons and ligaments. In order to facilitate research, all of these forces are summa-
rized in one vector, the joint torque.
Based on inverse dynamics, researchers develop and evaluate rehabilitation
techniques, e.g., prosthetics alignment (Schmalz et al. 2002) and patient-specific
gait modification (Fregly et al. 2007). Others investigate interdependencies in the
locomotor system and their effect, such as the influence of abnormal hip mechanics
on knee injury (Powers 2010). Many of these studies focus on analyzing gait
patterns, as the basic form of human movement. A well-balanced walking style is
an essential feature of a healthy locomotor system.
Unfortunately joint torques are not directly measurable but have to be assessed by
interpreting their external effect, that is, the ground reaction force (GRF), which is
exerted by the feet on the ground. The clinical standard to estimate joint torques is to
inversely calculate them from the GRF vector, the geometry of the skeleton, and the
recorded motion of the body. For this purpose, the subject’s gait pattern has to be
recorded by a motion capture (MoCap) system, and the GRF has to be measured with
force plates, which restricts this method to a laboratory setup.
For more flexibility, researchers use the profound knowledge introduced by
physical models to support gait analysis. These models yield a comprehensive
description and control over the correlation between acting forces and resulting
motion. Based on a physical model, joint torques can be estimated in the context
of an optimization problem. There exist several established optimization formula-
tions for this particular problem, namely, forward, inverse, and predictive dynamics
optimization. A comparison of these approaches and a detailed description of
forward dynamics optimization can be found in section “Optimization-Based
Methods.”
Physics-Based Models for Human Gait Analysis 3
Optimization-based torque estimation entails several challenges: A representation

of the entire body rather than just isolated segments (e.g., only one leg) is often
required, and model properties such as the inertial parameters and the formulation of
foot-ground contact model have to be defined with care. Furthermore, optimization
approaches are usually connected with high computational cost and the necessity for
considerable stabilization of the dynamic system. These issues might be addressed
by the use of sophisticated optimization algorithms and refined constraints, but
researchers also propose an alternative route, circumventing the optimization prob-
lem and instead relying on machine learning techniques. In such a framework, the
connection between a motion pattern and the underlying forces is learned on the
basis of a training set of motion sequences and associated optimization results. This
way, the gait pattern of a new subject can be assessed using the knowledge gained in
the training phase, i.e., unobservable joint torques are directly inferred from the
motion data.
With the ever-increasing amount of easily accessible video data, the demand for
motion analysis based on image sequences is inevitable. In this context, a combina-
tion of 3D motion reconstruction from 2D landmarks (nonrigid structure from
motion) and physics-based motion analysis is sought for and determines the direc-
tion of research in this field.
This chapter is structured in the following way: First of all, we present state-of-
the-art methods for physics-based modeling and motion analysis. Then, we describe
fundamental concepts to formulate the kinematics and dynamics of physical models
and introduce optimization-based methods for the estimation of effective forces. On
the basis of this methodology, we present two approaches and related experimental
results in more detail: a two-dimensional statistical model and a video-based frame-
work for human gait analysis. Both approaches are data-driven, i.e., the respective
algorithm learns parameter correlations on a training set and uses this knowledge to
directly infer underlying forces from input motion data. Finally, we point to possible
future research directions in this field.
State of the Art
A wide range of state-of-the-art methods related to human motion simulation and

analysis rely on physics-based modeling as a tool to control movement or to gain
valuable insides into the dynamics of human motion. In the field of computer
graphics, physical models enable researchers to synthesize realistic looking human
motions (Fang and Pollard 2003; Liu et al. 2005; Safonova et al. 2004; Sok et al.
2007; Wei et al. 2011; Zordan et al. 2005). The naturalness of a generated motion is
either ensured by the closeness to motion capture (MoCap) data or by an efficiency
measure. The latter case is based on the assumption that humans perform low
energetic movement. Both approaches can be applied in an optimization framework.
General issues of physics-based optimization are computational expense concerning
calculation of objectives, constraints, and the associated derivatives and a high-
dimensional search space, which complicates the optimization process. Fang and
4 P. Zell et al.
Pollard (2003) introduce efficient physical constraints and objective functions for a
fast optimization-based character animation. The authors formulate constraints and
objectives in such a way that first derivatives can be computed analytically in linear
time. The reduction of a high-dimensional parameter space to an appropriate sub-
space is addressed in Safonova et al. (2004). In this work, the authors use MoCap
data to transfer the optimization problem to a low-dimensional subspace that
includes the desired motion characteristics, exclusively.
While the minimization of an energy function to simulate movement corresponds
to the overall tendency of humans to avoid energy expenditure, this technique fails to
catch subject-specific motion traits and is not suitable for expressive motions, such
as dancing or acting. In those cases, an approach based on MoCap data that provides
more extensive information is sensible. Following this observation, Liu et al. (2005)
propose a physics-based dynamical model that incorporates specific features of a
persons’ motion style, e.g., the tendency to strain certain muscles more than others.
These characteristics are learned on a training set of MoCap sequences using an
introduced method called nonlinear inverse optimization.
A physical background is also suitable to support statistical approaches by
providing necessary constraints on the utilized model, e.g., (Wei et al. 2011)
combines statistical motion priors with physical constraints in a probabilistic frame-
work. The authors use a maximum a posteriori approach to synthesize a wide range
of physically realistic motions and motion interactions.
Apart from these graphic applications, physics-based models can be found,
facilitating typical computer vision tasks, like robust person tracking and 3D pose
estimation (Bhat et al. 2002; Brubaker and Fleet 2008; Vondrak et al. 2008; Wren
and Pentland 1998). Brubaker and Fleet (2008) use a simple planar model to estimate
biomechanical characteristics of gait and combine it with a 3D kinematic model for
monocular tracking. Due to the underlying physics, the model is able to handle large
occlusions. While this approach solely focuses on walking motions, a related work
by Vondrak et al. (2008) considers a wide range of motion types. The authors
introduce a full-body 3D physical prior that integrates the corresponding dynamics
into a Bayesian filtering framework. In addition to the motion dynamics, the
algorithm is able to model ground contact and environment interaction.
The previously described works concentrate on human pose tracking with
physics-based constraints but do not analyze the resulting force patterns. This
biomechanical objective, i.e., the estimation of inner joint torques and external
contact forces, is referred to as motion analysis and has been extensively treated in
various fields (Blajer et al. 2007; Brubaker et al. 2009; Johnson and Ballard 2014;
Stelzer and von Stryk 2006; Xiang et al. 2010; Zell and Rosenhahn 2015). In the
following, we will list a selection of recent works.
Brubaker et al. (2009) use an articulated body model to infer joint torques and
contact dynamics from motion data. They accelerate the optimization procedure by
introducing additional root forces and effectively decoupling the problem at different
time frames. This way an optimization step does not include the integration of EOM.
A similar goal is pursued by Xiang et al. (2010). The authors introduce predictive
dynamics as an approach for human motion simulation. Recently, researchers tried to
learn a direct mapping from a motion parametrization (joint angles) to the acting
forces (joint torques) on the basis of MoCap data: Johnson and Ballard (2014)
investigate sparse coding for inverse dynamics regression and Zell and Rosenhahn
(2015) introduces a two dimensional statistical model for human gait analysis.
The transfer of motion analysis to a 2D setting, i.e., a video-based framework, is a
relatively novel subject of research (Brubaker et al. 2009). By combining physics-
based modeling with standard nonrigid structure-from-motion techniques (e.g.,
(Bregler et al. 2000)), it is possible to infer the unobservable physical parameters
from monocular image sequences. State-of-the-art human pose estimation methods
rely either on anthropometric constraints (Akhter and Black 2015; Ramakrishna
et al. 2012; Wang et al. 2014) or on temporal bone length constancy (Wandt et al.
2015, 2016). These assumptions are relatively weak compared to a kinematic model.
Therefore, the next step is a combination of pose estimation with a physics-based
model of humans. This will not only improve the 3D reconstruction but also allow
for the estimation of joint torques from an image sequence. An example implemen-
tation of a joint model for pose estimation and physical parameter regression is
described in section “Combining Physical Models with 3D Reconstruction.”
Models and Methodology
Kinematics and Dynamics of Physical Models
The groundwork of physics-based human motion analysis is a physical model of the

human body. The literature offers a large amount of different models with varying
complexity from which many have been developed in the scope of robotics appli-
cations. A frequently used type are mass-spring models that consist of bone seg-
ments and connected by joints which are provided with torsional springs to create
joint torques. Beyond that, muscular skeletal models are applied (especially in
biomechanical research) for a more realistic presentation of the human locomotor
system, modeling the interaction between bone segments, joints, and an elaborate
structure of muscles. Since the methods described in this chapter are more suitable
for simple skeletal models, we focus on the portrayal of mass-spring models in this
section.
The modeled skeleton comprises a number of bone segments with associated
inertial properties (mass and inertial tensor) and connecting torsional springs for
each joint degree of freedom (DOF). Examples of a 2D and a 3D model can be seen
in Fig. 1. An essential part of these models is the kinematic chain that provides the
linkage between all segments and predefines the space of possible motions. In the
following part, we show how a kinematic chain can be defined using the Denavit-
Hartenberg notation (Steinparz 1985), well-known from robotics.
Denavit-Hartenberg Convention
In 1955, Jacques Denavit and Richard Hartenberg introduced a convention to
characterize a kinematic chain by a minimal set of four parameters for each interlink
6 P. Zell et al.
transformation. A link refers to a rotational or translational degree of freedom. In this

presentation, coordinate frames are attached to every link of the chain, and the
Denavit-Hartenberg parameters specify the transformation between successive link
coordinate systems.
Let Hj1
j be the transformation matrix from link j 1 to link j and [θj,dj,αj,aj] the
constituting parameters. Then the following operations are performed consecutively:
A rotation of the coordinate system around the zj1-axis by the angle θj, a translation
along the zj1-axis by the distance dj, a second rotation around the new xj-axis by the
angle αj, and a final translation along the xj-axis by the distance aj. The
corresponding transformation matrix Hj1 j is composed of the total rotation Rj1 j
and total translation Tj1
j , with
ð1Þ
When considering a kinematic chain consisting of several links, the position and
orientation of every link are determined by its predecessor transformations. There-
fore, the transformation from the coordinate system at link j to the world coordinate
frame L0 is obtained by the product of all transformations in the sub-chain from the
root link to the respective link:

j
Ri 0 Tj 0
Hj ¼ ∏ Hi
0 i1
¼ (2)
i¼1 0T 1
The linear coordinates p0ci of the center of mass (COM) of segment i can be
derived from the total translation and the total rotation of the coordinate system
attached to link j with respect to L0:
p0ci ¼ T0j þ R0j pjci , (3)
where pjci are the coordinates of the considered segment COM in the coordinate
system of the preceding link.
TMT Method
The dynamics of the considered mass-spring model are specified by a set of
equations of motion (EOM), which we will formulate using the TMT method
(Schwab and Delhaes 2009). This approach combines advantages of both the
Newton-Euler method and the Lagrange-Euler formulation by basically trans-
forming from constrained to unconstrained dynamics. In the Newton-Euler
approach, the EOM have to be extended to differential algebraic equations by
additional constraints that model the interconnection (kinematic chain) between
the model coordinates. This often leads to poor accuracy of solutions provided by
numerical solvers. In contrast to that, the Lagrange-Euler approach deals with a
minimum set of independent generalized coordinates, and the EOM are formulated
based on an energetic view point. For this purpose, all arising energies have to be
identified and their derivatives have to be calculated analytically, which might be
difficult in the case of complex, large DOF models.
To exploit the benefits of both methods, the TMT method uses a force approach,
but incorporates the kinematic constraints in a transformation T from dependent
coordinates x (segment COM position and orientation) to independent generalized
coordinates q (joint angles):
x ¼ TðqÞ (4)
Derivation supplies the corresponding linear and angular velocities and

accelerations:
@T
x_ ¼ q_ (5)
@q

@T @ @T
x€ ¼ q€ þ q_ q_ (6)
@q @q @q
The second term in Eq. (6) is referred to as convective acceleration

G ¼ @ q @@ Tq qÞ
@ _ q:
_
Furthermore the relation
@T
δx ¼ δq (7)
@q
between virtual displacements of the coordinates is valid and can be inserted into
D’Alembert’s principle of virtual work:
δxðF Mx€Þ ¼ 0 (8)

@T @T
δq F M €
qþG ¼0 (9)
@q @q
8 P. Zell et al.
Here F summarizes all applied forces and torques and M is the system’s inertia
matrix. Because of the independence of δq and the validity of virtual displacements
larger than zero, we can write Eq. (9) as
T
@T @T
FM q€ þ G ¼0 (10)
@q @q
Rearranging of Eq. (10) yields the EOM, formulated in independent generalized

coordinates:
T T
@T @T @T
M q€ ¼ ðF MGÞ (11)
@q @q @q
@T
This equation can be simplified by introducing the Jacobian J ¼ @q and the
T
generalized inertia matrix M ¼ @@ Tq M @@ Tq , so that
q ¼ J T ðF MGÞ:
M€ (12)
Formulating the Equations of Motion

For a clearer insight into the described methods, we will now provide an outline of
the derivation of EOM following Eq. (12) and using the 3D mass-spring model
depicted in Fig. 1 on the right-hand side. The model has 23 joint DOF and 6 DOF for
the global position and orientation of the root joint, resulting in a total of 29 DOF.
Therefore, we will receive a set of 29 coupled EOM.
The kinematic interconnection between the DOF q, i.e., between the generalized
coordinates, is defined via the Denavit-Hartenberg transformation matrices. In this
context, the coordinates q appear either as a translation a or d (prismatic joints) or as
a rotation α or θ (revolute joints). The Jacobian J of the according coordinate
transformation T (q) from generalized coordinates to Cartesian coordinates (position
and orientation of segment COM) is composed of a linear and a rotational part:

J lin
x_ ¼ J rot q_ (13)
@ Ti
Instead of formulating T (q) and calculating every differential @ qj to determine
the Jacobian, we choose a geometric approach. In the Denavit-Hartenberg conven-
tion, the rotation and translation axis associated with a DOF qj is always equal to the
z-axis of the respective coordinate frame Lj1. This leads to the following equations
for prismatic joints (Spong et al. 2005):
ij ¼ zj1
J lin 0
(14)
ij ¼ 0,
J rot
y
segment COM
x
joint DOF with torsional
spring
contact point
global DOF
y
x
z
(0,0)
Fig. 1 Physical mass-spring-models in 2D (left) and 3D (right). Joint degrees of freedom (DOF)
are depicted in blue and global DOF in green. Each joint degree of freedom is equipped with a
torsional spring and each segment is assigned with a mass and a tensor of inertia.
and for revolute joints:

ij ¼ zj1 pci Tj1
0
J lin 0 0
(15)
ij ¼ zj1 ,
J rot 0
with T0j1 and p0ci calculated according to Eqs. (2) and (3), respectively.
Apart from the Jacobian, we also need an expression for the convective acceler-
ation G, which can be derived in a similar fashion with

Glin
G¼ Grot
_
q: (16)
The corresponding sectional matrices for prismatic joints are
ij ¼ 0
Glin
(17)
Grot
ij ¼ 0,
and for revolute joints, we get

10 P. Zell et al.
X
i1
Glin
ij ¼ z0minfj1, kg z0maxfj1, kg p0ci T0maxfj1, kg q_ k
k¼0
(18)
X
i1
Grot
ij ¼ z0j1 z0k q_ k :
k¼j
The Jacobian together with the convective acceleration describes the kinematic of
our model. To characterize the dynamics of the system, i.e., formulate the EOM, we
need further properties, more precisely, the inertia matrix M and the active forces and
torques F (cf. Eq. (12)).
The inertia matrix is composed of mass values and inertia tensors for all model
segments. Let Mci and Ici be diagonal matrices containing masses and moments of
inertia in the segment coordinate frames, i.e., the values for the inertial properties
correspond to the rotation axes in the local frames attached to the individual segment
COM. Then the inertia matrix related to the coordinates x in the global frame L0 is

Mc i
M¼ T , (19)
R0ci Ici R0ci
with the rotation R0ci given by Eq. (2). Finally, the generalized inertia matrix required
for the EOM is obtained by M = JTMJ.
The motion of the 3D mass-spring model is driven by spring torques and outer
contact forces. The spring torques τj are applied at every joint DOF of the model
(represented by blue arrows in Fig. 1). A commonly used parametrization is

ð0Þ
τj ¼ κ j qj qj dj q_ j , (20)
with a linear resetting torque depending on the spring constant κj and the resting
ð0Þ
angle qj and an additional damping term proportional to the angular velocity of the
joint DOF with damping constant dj (Brubaker and Fleet 2008). To simulate
increasingly dynamic and natural-looking motions, additional nonlinear terms
might be added, e.g., Zell and Rosenhahn (2015):

ð0Þ
τj ¼ κ j qj qj ðqÞ d j q_ j ,
ð0Þ ð0Þ
X
4 (21)
qj ðqÞ ¼ qj þ ck xkCoM ðqÞ:
k¼1
Here, the resting angle is characterized by a fourth-order polynomial of the

position of the whole body COM. It is noteworthy that this representation is only
sensible for motions like walking or running that exhibit an even forward movement
of the COM. The single joint torques are summarized in the vector τ with τj = 0 for
j 6 (six global DOF that are not actuated by a joint torque).
When the model is in contact with its environment, e.g., standing on the ground or
lifting objects, it is effected by contact forces, denoted by Fc in the following. For the
majority of human motion types, our feet have to touch the ground and so-called
ground reaction forces (GRF) act on the sole of the foot. These forces mostly
accelerate the body in vertical direction, balancing the gravitational force. They act
as a braking force when the foot contacts the ground and a propulsive force at push off
from the ground. Furthermore, a horizontal friction force, depending on the properties
of the touching materials and the vertical force, prevents a sliding of the feet on the
ground. There are several possibilities to model the ground contact: The 2D model
illustrated in Fig. 1 is equipped with circular feet that enables a rolling motion of the
feet on the ground during gait. The GRF is concealed in a model constraint, by
defining the contact point as the root of the kinematic chain and effectively setting
its vertical coordinate to zero (Zell and Rosenhahn 2015). This approach is not
advisable for a 3D model with a significantly larger DOF, since it would result in
intolerably long kinematic chains that entail high simulation uncertainties.
As an alternative, Brubaker et al. (2009) suggest a spring contact force that
accelerates points on the sole of the foot, when approaching the ground plane. The
force is modulated by two sigmoid functions, which on the one hand ensure that a
contact point p is only effected when it is very close to the ground surface S and on
the other hand prevent an acceleration toward the ground. The force is defined by the
following equations:
_ θS Þ ¼ hð60dS ðpÞÞhð5nc ðpÞÞ½nc ðpÞnS ðpÞ þ tc ðpÞ,

Fc ðp,p,
nc ðpÞ ¼ κN ðdS ðpÞ 1Þ δN p_ T nY ðpÞ, (22)

tc ðpÞ ¼ δT p_ nS ðpÞT pÞn
_ S ð pÞ ,
with the sigmoid hðxÞ ¼ 12 ð1 þ tanhðxÞÞ. The total force is composed of the normal
force ncnS and the tangential attenuation tc, where nc denotes the magnitude and nS is
the surface normal. The sigmoidal modulation depends on the shortest distance d(p)
between the point p and the ground surface S. Finally, model parameters are included
in the vector θS, namely, the position and orientation of the ground plane, the spring
stiffness κN, the normal damping constant δN, and the tangential damping constant δT.
The last force term we are still missing for a complete description of the 3D model
with our EOM is the gravitational force:
Fg ¼ Mg, (23)
where g is the gravitational acceleration with respect to the dependent coordinates x.

Since the acceleration g = 9.81 m/s2 only affects the linear coordinates of the
segment COM in vertical direction, all other components are set to zero.
Finally, we insert all discussed terms into Eq. (12) and receive the model EOM:

M€
q ¼ J T τ þ Fc þ Fg MG : (24)
12 P. Zell et al.
This equation determines the dynamics of our model; in other words, it predicts
the development originating from an initial configuration of joint angles and angular
velocities. The acting forces and therefore the accelerations of the generalized
coordinates are completely defined by the model parameters θ and the current state
of the model ½q,q_ T . The model parameters include all spring and contact parameters,
e.g., stiffness and damping constants, resting angles and the position and orientation
of the contact plane.
Optimization-Based Methods
On the basis of the previously described physical models, we can estimate the
underlying forces and torques of motion data in an optimization framework. The
corresponding approaches found in the literature can be divided into three different
categories, inverse dynamics, forward dynamics, and predictive dynamics.
The inverse dynamics approach considers the motion parameters (e.g., joint
angles q) as optimization variables. A suitable objective function, e.g., the distance
to a target motion or a human performance measure, is minimized and joint torques
are calculated inversely. This way, the expensive integration of EOM is avoided and
the optimization problem is easily controlled, i.e., no sophisticated constraints are
necessary. A disadvantage of this method is the indirect derivation of forces that is
affected by the characteristic inaccuracy of joint trajectories and their accelerations.
In contrast to that, forward dynamics treats the forces and joint torques as design
variables that generate a motion through integration of EOM. As a result, the
optimization problem induces large computational cost and accurate boundaries
are crucial for the convergence of the optimization algorithm. Advantages of this
approach are the directly optimized forces and a natural conclusion of motion
parameters via EOM. Inverse as well as forward dynamics depends on the accuracy
of the inertial properties to achieve a sound torque estimation.
A rather novel approach, called predictive dynamics, was introduced by Xiang
et al. (2010). Both the joint torques and the model states are optimization parameters
and the EOM are treated as equality constraints. The method is closely related to
direct collocation methods (Stelzer and von Stryk 2006). It is computationally
efficient since no integration of EOM is required but because of the large number
of design variables an optimization algorithm suitable for large-scale problems that
has to be used.
Optimization Problem Setup

In this section, we will focus on forward dynamics optimization, since it is based on
the natural relation between forces and motion and provides a convenient frame to
introduce optimization principles. We will formulate the problem and present appro-
priate objective functions, regularization, and constraint terms. Let us consider a
problem of the form
min ff ðθÞg
θ
s:t:gi ðθÞ 0, i ¼ 1, . . . , m (25)
hi ðθÞ ¼ 0, i ¼ 1, . . . , n,
where θ are model parameters, f denotes the objective function, and gi and hi are
inequality and equality constraints, respectively. The forward dynamics step, i.e., the
integration of the EOM, is represented by the function D and results in the temporal
development of the model states
½qðtÞ, q_ ðtÞTmod ¼ Dðt, θ, qð0Þ, q_ ð0ÞÞ, (26)
with the initial state ½qð0Þ, q_ ð0ÞT. For the approximate solution, an ordinary differ-
ential equation (ODE) solver has to be used. These solvers are generally based on a
Runge-Kutta method (Mayers and Sli 2003).
In our problem statement, the motion data exists as input information, e.g.,
recorded by a MoCap system and we want to estimate appropriate joint torques.
Therefore, it is reasonable to define an objective function that includes the distance
between modeled states and target states: ½qðtÞ, q_ ðtÞTtarg,e.g., the quadratic L2-norm of
their difference:
ðT 2
T
f ð θÞ ¼ Dðt, θ, qð0Þ, q_ ð0ÞÞ ½qðtÞ, q_ ðtÞtarg : (27)
0
Since the optimization algorithm generally has to deal with a high-dimensional

parameter space and the subspace that creates realistic motions is non-convex,
careful regularization and constraints on the system are indispensable. A number
of human performance measures can be used to support the optimization routine.
These measures penalize motions with high energetic cost which are considered to
be unnatural. Commonly used quantities are the dynamic effort:
ðT
f ¼ jjτjj2 dt, (28)
0
and the jerk:

ðT
f ¼ _ 2 dt:
jjτjj (29)
0
Minimizing the jerk increases the smoothness of the simulated motion.

Furthermore, we can reduce the searched parameter space by incorporating prior
knowledge as constraints gi and hi. Obviously, it is advisable to set boundaries on the
model parameters, such as
14 P. Zell et al.
0 κ j uκ j ,
ð0Þ
lqj qj uq j , (30)
0 d j ud j ,
with the respective lower and upper bounds l and u for the joint spring parameters. In
addition to that, we might have knowledge about the motion of certain points of our
model, e.g., the contact points during walking are not supposed to slide over the
ground, but have to stay fixed. This information can be taken into account by the
equality constraint
ðT
hc ¼ _ 2 dt ¼ 0,
jjJ c qjj (31)
0
where Jc is the section of the Jacobian that is associated with the coordinates of the
contact points.
In some scenarios, GRF have been recorded with force plates and can be used as
ground truth data Ftarg. Then we can constrain the modeled contact force with
ðT
hf ¼ jjFc Ftrag jj2 dt ¼ 0: (32)
0
In practice, it is often beneficial to formulate equality constraints as soft inequality

constraints by introducing an upper threshold or to include them as regularization
terms. This way the convergence of the optimization algorithm is more likely.
Learning Force Patterns in 2D
In this section, we will present a statistical approach for the analysis of human gait
(Zell and Rosenhahn 2015). The general idea of this work is to learn model
parameters on a set of MoCap walking sequences from 115 subjects and then use
this knowledge to infer the underlying joint torques of a new gait pattern directly
from the motion data. The proposed model considers the 2D projection of walking in
the sagittal plane (cf. Fig. 1 on the left-hand side). It consists of ten segments and
nine DOF for the joint angles. The corresponding EOM can be derived as described
in section “Formulating the Equations of Motion” and have the form
MðqÞ€ _ θÞ:
q ¼ Fðq,q, (33)
Based on this, movement can be simulated by solving the related initial value
problem according to Eq. (26).
Combined Statistical Model

For the direct regression of joint torques from motion, a statistical model has to be
learned that combines motion characteristics with a physical representation. Follow-
ing Troje (2002a), the authors represent walking as a linear combination of principle
component postures pi with sinusoidal variation of coefficients:
pðtÞ ¼ p0 þ p1 sin ðωtÞ þ p2 sin ðωt þ ϕ2 Þ

(34)
þ p3 sin ð2ωt þ ϕ3 Þ þ p4 sin ðωt þ ϕ4 Þ:
Here, ω is the fundamental frequency of the gait and (ϕ2,ϕ3,ϕ4) are phase shifts.
In Troje (2002a), a posture consists of 15 three-dimensional joint positions, resulting
in a 45-dimensional vector p. For a full motion representation, all characterizing
parameters are summarized in one vector:
u ¼ ½p0 , p1 , p2 , p3 , p4 , ω, ϕ2 , ϕ3 , ϕ4 T : (35)
To include information about physical properties, the gait patterns from the
training set are approximated using a forward dynamics optimization approach
(cf. section “Optimization-Based Methods”). The EOM are integrated, resulting in
a simulated motion which is compared to the target motion at a set of key times {tk}k.
hThe authors optimize the physical model parameters θ together with the initial state
q0 , q_ 0 T by minimizing the sum of squared reconstruction errors and constraining
the simulated GRF Fc to lie in the vicinity of ground truth data Ftarg. This way,
realistic force patterns are ensured. The optimization problem is formulated as
follows:
( )
X 2
ðq0 , q_ 0 , θÞ ¼ arg min
Dðtk , q0 , q_ 0 , θÞ ½qðtk Þ, q_ ðtk Þtarg ,
T
qo , q_ 0 , θ
k

s:t: Fc qðtk Þ, q_ ðtk Þ, q€ðtk ÞÞ Ftarg ηk ,
with thresholds ηk. The simulated GRF vectors are normalized with respect to body
weight and compared to a normalized mean ground truth force Ftarg , obtained from
force plate measurements. The effective GRF is calculated via
X mi
_ qÞ ¼
Fc ðq,q,€ _ qÞ gÞ,
ðai ðq,q,€ (37)
i
M
where g is the gravitational acceleration and ai is the Cartesian acceleration of body

segment i with respective to mass mi. The total body mass is denoted by M.
The optimization problem can be solved by sequential quadratic programming
(SQP) (Powell 1978) and the resulting physics-based model parameters are
16 P. Zell et al.
v ¼ ½q0 , q_ 0 , θ, MT : (38)
Based on the combined parametrization of u and v, it is possible to infer joint

torques from joint angle trajectories and vice versa. The corresponding parameter
vectors are optimized for each walking sequence in the training set and then stacked
in one matrix:
h i
u1 ...u115 ℝ229 ðmotion Eq:ð35ÞÞ
W¼ (39)
|fflfflfflfflffl{zfflfflfflfflffl} ℝ ðphysical model Eq:ð38ÞÞ
v1 ...v115 52
115 subjects
The combined statistical model encompasses geometrical properties, dynamical

behavior, and the physical description of a walking motion. All of these features
contribute to the characteristics of a gait pattern, and their mutual dependence can be
exploited to infer missing information based on an incomplete parameter set. Two
different regression methods are applied: a k-nearest neighbor (k-NN) regression and
an asymmetrical projection into the principal component space (aPCA), as intro-
duced by Al-Naser and Söderström (2012).
Force Estimation with Occluded Input

The proposed direct regression methods are evaluated regarding the deviation of
estimated 2D GRF and stance knee torques from ground truth data and inversely
calculated torques, respectively. For this purpose, MoCap sequences of three differ-
ent subjects with synchronized force plate measurements were recorded. Based on
the motion parameters u of the considered sequence, the physical parameters v are
inferred and used as input to the physical model simulation. This yields the estimated
GRF via Eq. (37) and the estimated joint torques through the respective spring torque
equation, e.g., Eqs. (20) or (21).
In order to evaluate the robustness of the direct regression methods with respect to
incomplete input information, joint trajectories are iteratively removed from the
considered gait sequence. Starting with missing left-hand trajectory, the authors
successively remove the trajectories of the left elbow, ankle, and knee, resulting in
a final regression based solely on the right-hand side of the human body. The number
of missing joint trajectories is denoted by N. Results for one example subject are
illustrated in Fig. 2. For a quantitative evaluation, the symmetric mean absolute
percentage errors (SMAPE) eF and eτ for GRF Fc and stance knee torques τK1 ,
respectively, are listed in Table 1 together with the arising computation times. To
compare the performance of the direct regression methods to an optimization
procedure, a third method (OPT) following the forward dynamics optimization in
Eq. (36) was implemented. In the case of missing joint trajectories, an additional
term, based on the dynamic effort, is minimized to compensate for the incomplete
information.
With zero missing input trajectories, the best approximation is achieved by the
optimization-based method OPT, as expected. Under these circumstances, the phys-
ical parameters are optimized, to create a simulated motion as close to the target
motion as possible with no additional energy minimization that affects the results. As
the input information is reduced, the direct regression methods (k-NN, aPCA)
increasingly outperform OPT. In particular, the estimates based on the k-NN method
show consistently low SMAPE values. In regard to the computational cost, the
optimization requires computation times that are at least three orders of magnitude
higher than those for the direct regression methods. All of these values were
generated using unoptimized Matlab code.
In summary, the combination of motion and force parameters in one statistical
representation allows a fast and robust regression of joint torques from MoCap data.
In this work, a limiting factor of the approximation quality is the rather simple 2D
physical model.
a k-NN
2 11 0.6
aPCA
τK1/ M [Nm/kg]
OPT 0.4
1 ground truth
Fx[N/kg]
10
Fy[N/kg]
0.2
0 0
9
-0.2
-1
8 -0.4
-2 -0.6
20 40 60 80 20 40 60 80 20 40 60 80
Stance Phase [%] Stance Phase [%] Stance Phase [%]
b
2 11 0.6
τK1/ M [Nm/kg]
0.4
1
Fx [N/kg]
10
Fy [N/kg]
0.2
0 0
9
-0.2
-1
8 -0.4
-2 -0.6
20 40 60 80 20 40 60 80 20 40 60 80
Stance Phase [%] Stance Phase [%] Stance Phase [%]
Fig. 2 Comparison between different regression and optimization methods concerning estimated
GRF components Fy, Fx and knee torques τK1. Positive values correspond to flexor torques. The
results are based on full joint trajectory information (in (a)) and on partial information with N = 4
(in (b)), respectively. In the case of GRF components, the black line illustrates ground truth data,
and in the case of joint torques, it represents torques calculated via inverse dynamics. A quantitative
evaluation can be found in Table 1
Table 1 SMAPE values EF and Et for GRF and knee torque estimates with related computation
times tc
k-NN aPCA OPT
N eF eτ tc [s] eF eτ tc [s] eF ετ tc [s]
0 1.504 1.624 2.994 1.726 2.019 0.881 1.483 1.420 8103
1 1.504 1.586 2.117 1.780 2.052 0.934 1.811 1.851 4358
2 1.496 1.582 2.112 2.005 2.083 0.913 1.890 1.793 3096
3 1.524 1.562 2.583 1.606 2.169 0.619 1.843 2.069 3229
4 1.528 1.565 2.538 1.594 2.072 0.669 1.911 2.188 2092
N indicates the number of missing input joint trajectories
18 P. Zell et al.
Video-Based Gait Analysis
As shown in the previous sections, methods based on motion capture records

combined with additional force measurement achieve high-quality results. However,
they require expensive and complex laboratory setups. With the increasing ease of
capturing videos, for example, with mobile devices, the demand for purely video-
based motion capture rises. In the combination of structure-from-motion methods
with complex physical models, researchers found a tool to infer physical parameters
directly from monocular input videos.
A Short Introduction to NRSfM

Structure-from-motion (SfM) describes techniques to infer a 3D structure of an
object from its 2D projections into an image plane. Early work concerning the 3D
reconstruction of rigid objects was done in the 90s by Tomasi and Kanade (1992). In
their seminal work, they proposed to decompose a 2D measurement matrix W2D into
camera matrices K and a shape matrix S:
W2D ¼ KS: (40)
With observations from multiple viewpoints, they were able to reconstruct the
rigid 3D structure of the object. In 2000, Bregler et al. (2000) generalized Tomasi
and Kanades work to the nonrigid case, where the observed object is allowed to
deform over time. Their method rests upon the idea that a deformable shape S can be
represented by a linear combination of K-weighted base shapes Qi:
X
K
S5 d i Qi , (41)
i¼1
where di is the weighting factor for the i-th base shape. This leads to a generalization
of Eq. (40):
W2D ¼ KDQ, (42)
where D contains the weighting factors di for the base shapes in Q. A common
approach is to minimize a reprojection error er such as
er ¼ kW2D KDQk (43)
for the different variable sets with respect to orthogonality constraints on the camera
matrix. For further detail, the reader is referred to Bregler et al. (2000).
Since the base shapes in this mathematical representation are nonunique in the
following years, numerous authors (e.g., Dai and Li 2012; Hamsici et al. 2011; Park
et al. 2010; Torresani et al. 2003, 2008) proposed several constraints to convert this
problem into a more feasible one.
NRSfM for 3D Reconstruction of Human Motion

Although the methods briefly introduced in section “A Short Introduction to
NRSfM” were developed for arbitrary deformable objects, they also showed great
success in the 3D reconstruction of human motion. However, they require a suffi-
cient camera or object motion to achieve acceptable results. Therefore, they are not
applicable for more realistic scenarios with limited camera motion such as in
surveillance situations. This leads to the idea of including prior knowledge about
the observed object into the problem formulation.
In 2002, N. Troje (2002a, b) showed that a single human pose during a gait
motion can be described by the linear combination of specific weighted vectors.
These vectors (named eigenpostures) are obtained by performing a PCA on multiple
3D gait sequences. It was shown that the first four eigenpostures were sufficient to
cover more than 98% of the variance of a walking motion. N. Troje also discovered
that for a walking motion, the weights for the eigenpostures describe a sine curve.
This leads to the conclusion that the human gait can be described by just the four
eigenpostures and the corresponding sine functions which are defined by their
amplitude, frequency, and phase shift. In Troje (2002a, b), this discovery was used
to identify and create gait patterns for different gait characteristics. Since Troje’s
formulation is in close proximity to Eqs. (41) and (42), the applicability for 3D
reconstruction is obvious.
Multiple researchers used the idea of representing human poses in a PCA basis for
the 3D reconstruction from single images, e.g., Akhter and Black (2015), Ramakrishna
et al. (2012), and Wang et al. (2014). Since these learned bases also include many
nonhuman poses, further anthropometric priors are applied, such as symmetry and
predefined bone length relations. Wandt et al. (2015, 2016) were the first to combine
temporal information of the skeletal structure with the learning of subspaces by
employing a bone length constancy term. They also exploited the periodic behavior
of the eigenpostures coefficients for periodic motions as proposed by Troje (2002a).
Thus, their method appears to be well suited for gait reconstruction tasks in general,
but not for a detailed analysis of idiosyncratic gait (e.g., pathological gait caused by
neurological disorders) due to the limited number of used principle components.
Combining Physical Models with 3D Reconstruction

Here an algorithm is presented to jointly infer the 3D pose and physical parameters
from monocular input data. The presented algorithm first performs the 3D pose
reconstruction followed by the physical simulation to eliminate any ambiguities
between camera and object motion as well as enforcing a physically plausible
reconstruction. The dynamical description is based on a 3D physical mass-spring
model (cf. Fig. 1 on the right-hand side) with spring and contact parameters θ that
result from the following optimization procedure:
20 P. Zell et al.
( )
T 2 T w X
w0 X w1 X 2
T

½q,q_ mod, t ½q,q_ targ, t kτ t k þ kτ_t k þ J c q_mod, t k ,
T T 2
θ5 arg min þ 2 2
θ T t¼1 T t¼1 T t¼1
(44)
with regularization weights (w0,w1,w2). A description of the individual terms has
been given in section “Optimization Problem Setup.”
A joint parameter space is built consisting of the weighting matrix D from
Eq. (43) and the physical parameters θ. As shown in Eq. (42), the 3D shape can be
written as S = DQ. For known 3D shapes S from the training data, D can be directly
calculated via
D ¼ SQþ , (45)
where Q+ denotes the Moore-Penrose pseudoinverse of Q. Finally, a vector vk

composed of the vectorized weighting matrix D and the physical parameters θ is
assigned to each sequence k in the training set:

vecðDÞ
vk ¼ : (46)
θ
Here, vec() is the vectorization operator which stacks the columns of the matrix
into a vector. We assume that a newly observed motion lies in the space spanned by
the vectors vk for each sequence k.
In the first step, the algorithm performs a 3D reconstruction of the observation
matrix W2d alternatingly minimizing Eq. (43) for the camera matrix K and coefficient
matrix D. The consequent 3D shapes are calculated by S = DQ. Since a linear model
is used to represent nonlinear deformations, the estimated 3D motion is expected to
differ from the real motion. This can be easily seen when analyzing the temporal
behavior of the bone lengths, shown on the left in Fig. 3. After the 3D reconstruction,
the bone lengths fluctuate heavily due to the above mentioned linear model. This
issue is addressed by limiting the parameter space to physically valid motions; in
other words, the weighting coefficients and physical parameters are inferred by
means of a k-nearest neighbor (k-NN) regression in the space spanned by the vectors
vk. As suggested by Zell and Rosenhahn (2015), a local k-NN regression can be used,
which outperforms global approaches like PCA or asymmetric PCA for this partic-
ular problem.
The recovered physical parameters θ are now employed to simulate a 3D motion
by integrating the corresponding set of EOM. This step provides the resulting joint
torques and contact forces and converts the rough 3D pose estimation to a physically
feasible 3D reconstruction of the observed motion. Comparing the bone length
variation before (cf. Fig. 3 left) and after physical simulation (cf. Fig. 3 right)
indicates an improvement regarding plausibility.
Additionally, the use of the physical model allows for resolving the ambiguity
between camera and object motion. Based on the knowledge gained from the
physical simulation of the observed object, e.g., the forward movement during
Physics-Based Models for Human Gait Analysis
Fig. 3 Comparison of the temporal behavior of bone lengths after 3D reconstruction (left) and after applying the physical model (right). Obviously, the desired
bone length constancy is assured
21
22 P. Zell et al.
walking, a standard camera calibration technique with known 2D-3D point corre-
spondences can be used to reconstruct the camera parameters.
Experiments on Force Estimation

The joint model is evaluated on a training set consisting of 45 MoCap walking
sequences with synchronized force plate data. For each reconstruction, the consid-
ered sequence is excluded from the training set. First of all, the estimation of knee
torques from video data is assessed. The torque profiles are generated as described in
the previous section. Figure 4 shows the mean value of estimated knee torques for all
reconstructed gait sequences together with the related standard deviation. The graph
on the right-hand side of Fig. 4 displays the inversely calculated knee torque of an
example sequence for comparison.
The torque was computed based on force plate data using a bottom-up procedure,
i.e., calculating all acting lever arms, joint forces, and torques from the center of
pressure (COP) on the ground to the knee joint. Because of this, only the torque
during the stance phase is shown. The kinematic chain from the COP to the knee at
the swing leg would be too long and consequently the accumulated error would
become too high. It can be seen that the estimated torques are consistent for all
reconstructed 3D motions and the absolute values are similar to inverse dynamics
torques. The maximal extension torque is reached during the second half of the
swing phase and not during double support.
This discrepancy is mainly due to the estimated contact model, that is very
sensitive to the distance of contact points to the ground. Further error sources are
model inaccuracies, e.g., concerning mass distribution, imprecision of the fitted
skeleton, and of course the reconstruction error of θ.
To further analyze the joint model regression, vertical GRF are compared to
ground truth data in Fig. 5. The absolute values of the extremal points are slightly
high, but the overall curve progression resembles ground truth reaction forces. The
experiment shows that the joint model provides a sound estimation of unobservable
3D torques from monocular videos without the need for tedious optimization.
To demonstrate the applicability of the algorithm to real-world scenarios, a
sequence from the KTH football database (Kazemi et al. 2013) has been
reconstructed. This dataset contains multiview sequences of a challenging noisy
outdoor scene that shows a football player walking over a playfield. The 3D
reconstruction from camera 1 with estimated torques is illustrated in Fig. 6. As
expected, the reconstructions from the remaining two cameras yield very similar
results with a maximal reconstruction error of 0.05 m.
Future Directions
In this chapter, we gave you an overview over motion, in particular gait analysis
based on a physical simulation. We presented basic concepts of physical modeling,
parameter optimization, and 3D reconstruction and showed how these methods can
1.5 1.5
1 1
joint torque [Nm/kg]
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
double sup single sup. double sup. single sup. double sup. single sup. stance
swing stance
-2 -2
0 20 40 60 80 100 50 60 70 80 90 100
% gait period % gait period
Fig. 4 Consistency of torques: Estimated knee torques for all reconstructed walking sequences
(left). The torques are shown for a full gait cycle with mean value in blue and standard deviation in
green. On the right-hand side, an example of the knee torque for the stance phase, calculated via
inverse dynamics, is displayed
25 25
vertical contact force [N/kg]
20 20
15 15
10 10
5 5
0 0
double sup. single sup. stance
double sup. single sup. stance
-5 -5
50 60 70 80 90 100 50 60 70 80 90 100
% gait period % gait period
Fig. 5 Consistency of forces: Modeled vertical contact forces (left) with mean value in blue and
standard deviation in green for the whole set of gait sequences. The curve on the right-hand side
displays the vertical component of a measured GRF vector
be extended to data-driven learning approaches for force estimation. At this point,

there are multiple promising directions for future research.
Considering the video-based gait analysis presented in section “Video-Based Gait
Analysis,” it is a straightforward conclusion to search for a method that does not
combine the 3D reconstruction and the physical simulation in a pipeline as separate
steps, but joins them in one model. A further goal in this context is the development
of algorithms for handheld devices that would be able to jointly estimate the 3D
motion and 3D forces from an integrated monocular camera view. Based on this,
short video clips could be directly analyzed using the smartphone that recorded
them. For this objective, we require a very robust algorithm concerning camera
24 P. Zell et al.
Fig. 6 3D reconstruction and estimated torques of the KTH football dataset. The reconstructions
and torques (red spheres) appear to be plausible compared to the corresponding images and torques
in Fig. 4
motion and an implementation that enables real-time processing, in order to ensure

the applicability in everyday life.
Another interesting, sparsely treated research direction is the fusion of different
sensor types to facilitate human motion analysis. For example, prior knowledge
about the acceleration and the orientation of a subset of body segments could be
supplied by including inertial measurement units (IMU). This way, the search space
for underlying forces could be reduced significantly. In addition, when considering a
muscular skeletal model, the fusion between video data and electromyograph (EMG)
measurements might provide vital constraints on the high-dimensional system.
References
Akhter I, Black MJ (2015) Pose-conditioned joint angle limits for 3D human pose reconstruction.
In: IEEE Conference on computer vision and pattern recognition (CVPR 2015). IEEE, pp
1446–1455
Al-Naser M, Söderström U (2012) Reconstruction of occluded facial images using asymmetrical
principal component analysis. Integrated Comput Aided Eng 19(3):273–283
Bhat KS, Seitz SM, Popović J, Khosla PK (2002) Computer vision – ECCV 2002: 7th European
conference on computer vision copenhagen, Denmark, 2002. In: Proceedings, Part I, chapter
computing the physical parameters of rigid-body motion from video, Springer, Berlin/Heidel-
berg, pp 551–565, 28–31 May 2002
Blajer W, Dziewiecki K, Mazur Z (2007) Multibody modeling of human body for the inverse
dynamics analysis of sagittal plane movements. Multibody Sys Dyn 18(2):217–232
Bregler C, Hertzmann A, Biermann H (2000) Recovering non-rigid 3D shape from image streams.
In: IEEE Conference on computer vision and pattern recognition (CVPR). IEEE, pp 690–696
Brubaker MA, Fleet DJ (2008) The kneed walker for human pose tracking. In: IEEE conference on,
computer vision and pattern recognition, 2008 (CVPR 2008). pp 1–8, June 2008
Brubaker MA, Sigal L, Fleet DJ (2009) Estimating contact dynamics. In: IEEE 12th international
conference on Computer vision 2009. IEEE, pp 2389–2396
Dai Y, Li H (2012) A simple prior-free method for non-rigid structure-from-motion factorization.
In: Conference on computer vision and pattern recognition (CVPR), CVPR’12, IEEE Computer
Society, Washington DC, pp 2018–2025, 2012
Fang AC, Pollard NS (2003) Efficient synthesis of physically valid human motion. ACM Trans
Graph 22(3):417–426
Fregly BJ, Reinbolt JA, Rooney KL, Mitchell KH, Chmielewski TL (2007) Design of patient-
specific gait modifications for knee osteoarthritis rehabilitation. IEEE Trans Biomed Eng 54
(9):1687–1695
Hamsici O, Gotardo P, Martinez A (2011) Learning spatially-smooth mappings in non-rigid
structure from motion. In: European conference on computer vision (ECCV). Springer,
Berlin/Heidelberg
Johnson L, Ballard DH (2014) Efficient codes for inverse dynamics during walking. In: Proceedings
of the twenty-eighth AAAI press conference on artificial intelligence, AAAI’14. AAAI Press,
pp 343–349
Kazemi V, Burenius M, Azizpour H, Sullivan J (2013) Multi-view body part recognition with
random forests. In: British machine vision conference (BMVC). BMVC Press, Bristol
Liu CK, Hertzmann A, Popović Z (2005) Learning physics-based motion style with nonlinear
inverse optimization. ACM Trans Graph 24(3):1071–1081
Mayers D, Sli E (2003) An introduction to numerical analysis. Cambridge University Press,
Cambridge
Park HS, Shiratori T, Matthews I, Sheikh Y (2010) 3D reconstruction of a moving point from a
series of 2D projections. In: European conference on computer vision (ECCV). Springer, Berlin/
Heidelberg
Powell MJD (1978) Numerical analysis. In: Proceedings of the Biennial Conference held at
Dundee, chapter A fast algorithm for nonlinearly constrained optimization calculations,
Springer, Berlin/Heidelberg, pp 144–157, June 28–July 1 1977
Powers CM (2010) The influence of abnormal hip mechanics on knee injury: a biomechanical
perspective. J Orthop Sports Phys Ther 40(2):42–51
Ramakrishna V, Kanade T, Sheikh YA (2012) Reconstructing 3D human pose from 2D image
landmarks. In: European conference on computer vision (ECCV). Springer, Berlin/Heidelberg
Safonova A, Hodgins JK, Pollard NS (2004) Synthesizing physically realistic human motion in
low-dimensional, behavior-specific spaces. ACM Trans Graph 23(3):514–521
of lower limb amputee gait: the influence of prosthetic alignment and different prosthetic
components. Gait Posture 16(3):255–263
Schwab AL, Delhaes GMJ (2009) Lecture notes multibody dynamics B, wb1413
Sok KW, Kim M, Lee J (2007) Simulating biped behaviors from human motion data. ACM Trans
Graph 26(3):107:1–107:9
Spong M, Hutchinson S, Vidyasagar M (2005) Robot modeling and control. Wiley
Steinparz F (1985) Co-ordinate transformation and robot control with denavit-hartenberg matrices.
J Microcomput Appl 8(4):303–316
Stelzer M, von Stryk O (2006) Efficient forward dynamics simulation and optimization of human
body dynamics. ZAMM – J Appl Math Mech/Zeitschrift fr Angewandte Mathematik und
Mechanik 86(10):828–840
Tomasi C, Kanade T (1992) Shape and motion from image streams under orthography: a factori-
zation method. Int J Comput Vis 9:137–154
Torresani L, Hertzmann A, Bregler C (2003) Learning non-rigid 3D shape from 2D motion. In:
Thrun S, Saul LK, Schölkopf B (eds) Neural information processing systems (NIPS). MIT
Press, Cambridge, MA
Torresani L, Hertzmann A, Bregler C (2008) Nonrigid structure-from-motion: estimating shape and
motion with hierarchical priors. In: IEEE Transactions pattern analysis and machine intelli-
gence, IEEE, 21 March 2008
Troje NF (2002a) Decomposing biological motion: a framework for analysis and synthesis of
human gait patterns. J Vis 2(5):371–387
Troje NF (2002b) The little difference: Fourier based synthesis of gender-specific biological
motion. AKA Press, Berlin, pp 115–120
26 P. Zell et al.
Vondrak M, Sigal L, Jenkins OC (2008) Physical simulation for probabilistic motion tracking. In:
IEEE conference on computer vision and pattern recognition, 2008 (CVPR 2008), pp 1–8, June
2008. IEEE
Wandt B, Ackermann H, Rosenhahn B (2015) 3d human motion capture from monocular image
sequences. In: IEEE conference on computer vision and pattern recognition workshops, June
2015. IEEE
Wandt B, Ackermann H, Rosenhahn B (2016) 3d reconstruction of human motion from monocular
image sequences. IEEE Trans Pattern Anal Mach Intell 38(8):1505–1516
Wang C, Wang Y, Lin Z, Yuille A, Gao W (2014) Robust estimation of 3d human poses from a
single image. In: IEEE Conference on computer vision and pattern recognition (CVPR). IEEE
Wei X, Min J, Chai J (2011) Physically valid statistical models for human motion generation. ACM
Trans Graph 30(3):19:1–19:10
Wren CR, Pentland AP (1998) Dynamic models of human motion. In: Proceedings of the third
IEEE internatonal conference on automatic face and gesture recognition, Nara, April 1998.
Xiang Y, Chung H-J, Kim JH, Bhatt R, Rahmatalla S, Yang J, Marler T, Arora JS, Abdel-Malek K
(2010) Predictive dynamics: an optimization-based novel approach for human motion simula-
tion. Struct Multidiscip Optim 41(3):465–479
Zell P, Rosenhahn B (2015) Pattern recognition: 37th German conference, GCPR 2015. In: Pro-
ceedings, chapter A physics-based statistical model for human gait analysis, Springer Interna-
tional Publishing, Aachen, Germany, October 7–10, 2015, pp 169–180.
Zordan VB, Majkowska A, Chiu B, Fast M (2005) Dynamic response for motion capture animation.
ACM Trans Graph 24(3):697–701
Simulation of Soft Tissue Loading from
Observed Movement Dynamics
Scott C. E. Brandon, Colin R. Smith, and Darryl G. Thelen
Abstract
An understanding of in vivo soft tissue loading is essential for investigating the
causes of traumatic injury, progression of degenerative joint disease, and options
for orthopedic treatment. Modern motion analysis technologies can be used to
observe movement dynamics and characterize resultant joint loads. However,
technologies for directly measuring the loads on in vivo soft tissues (e.g.,
ligament, tendon, and cartilage) generally require invasive techniques and thus
are impractical for widespread use. In this chapter, we review two computational
modeling approaches for characterizing soft tissue loads from observed move-
ment dynamics. In the traditional sequential simulation approach, linked-segment
dynamic musculoskeletal models are first solved for muscle and resultant joint
loading. These loads are then applied as boundary conditions on a more detailed
model of the joint to estimate loading and deformation of ligament and cartilage
tissues. The primary limitation of the sequential approach is that it decouples
movement dynamics and joint mechanics, making it infeasible to predict how
muscle coordination may adapt to changes in soft tissue behavior. To overcome
this limitation, concurrent simulation approaches have been introduced, which
enable simultaneous solution of muscle forces and soft tissue mechanics under-
lying human movement. We demonstrate the advantage of the concurrent
approach to directly probe inherent coupling between muscle coordination,
joint kinematics, cartilage contact pressure, and ligamentous behavior that can
arise with soft tissue damage. We conclude with suggestions for further develop-
ment and use of concurrent simulation approaches that could greatly extend our
ability to investigate both surgical and rehabilitative treatments of musculoskel-
etal pathologies.
S.C.E. Brandon • C.R. Smith • D.G. Thelen (*)

Department of Mechanical Engineering, University of Wisconsin-Madison, Madison, WI, USA
e-mail: scott.brandon@gmail.com; crsmith25@wisc.edu; dgthelen@wisc.edu

DOI 10.1007/978-3-319-30808-1_172-1
2 S.C.E. Brandon et al.
Keywords
Joint Mechanics • Cartilage Pressure • Contact Force • Musculoskeletal Model •
Ligament Force • Simulation • Secondary Kinematics • Concurrent Simulation •
COMAK • Optimization • Inverse Dynamics • Force Dependent Kinematics •
Knee • Multibody
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Invasive Measurement of In Vivo Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
External Measurement of Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Status Quo: Sequential Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Concurrent Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Articular Contact and Ligament Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Simulating Soft Tissue Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Musculoskeletal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Body-Segment Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
COMAK: Concurrent Optimization of Muscle Activations and Secondary Kinematics . . . 18
Example Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Introduction
Since a casual observer can often distinguish healthy movement patterns from those
of an individual with injured muscle, ligament, or cartilage, why simulate soft tissue
loading? A primary motivation is that estimates of soft tissue loads can help us
understand the causes of injuries and provide an objective basis for evaluating
treatment options for soft tissue injuries. For example, if surgery is required, an
analysis of soft tissue loading could help establish the required strength of graft
materials and engineered components. Further, models can be used to predict how
soft tissues will be loaded during rehabilitative exercises. More fundamentally, if we
desire to understand how and why humans move the way they do, we often need to
characterize the loading and energetics of soft tissues during movement.
In the musculoskeletal system, soft tissues both mobilize and constrain skeletal
motion. Muscles generate forces that are transmitted to the skeleton by tendons that
can elastically deform. Ligaments, connective tissues, and articular cartilage connect
the bones, distribute loading, and guide joint motion. Although the relative motion of
bones can be measured and analyzed with increasing precision (Miranda et al. 2013;
Farrokhi et al. 2014), (▶ Measurement of 3D Dynamic Joint Motion Using Biplane
Videoradiography), soft tissue loading is not easily measured in vivo. Therefore,
simulations are often required to investigate the mechanical loading and behavior of
soft tissues during movement and the resulting effects on other structures within the
Simulation of Soft Tissue Loading from Observed Movement Dynamics 3
Fig. 1 Simulation of soft tissue loading from observed dynamics. Experimental measurements of
whole-body motion and external forces are supplied as inputs to a musculoskeletal simulation,
which yields estimates of the internal soft tissue (i.e., muscle, ligament, and articular cartilage) loads
musculoskeletal system (Fig. 1). One of the earliest musculoskeletal models assessed
the contributions of muscle loading to joint injury and bone fractures (Pauwels
1935). Since then, increasingly complex models have been introduced to character-
ize musculoskeletal tissue mechanics at various scales, including lower-extremity
models to assess the contributions of muscles to hip, knee, and ankle joint loads
during walking (Heller et al. 2001; D’Lima et al. 2006; Herzog and Longino 2007;
Kutzner et al. 2010) and multiscale models to assess cartilage tissue and cellular
deformations under functional loads (Erdemir et al. 2007; Halloran et al. 2012).
Scope
In biomechanical models of human movement, passive soft tissues are often omitted
and simplified joints (e.g., hinge, ball-and-socket) are assumed. These reduced
degree-of-freedom (DOF) joints impose kinematic constraints to account for the
action of soft tissues at each joint (▶ 3D Dynamic Pose Estimation using Cameras
and Reflective Markers; ▶ 3D Dynamic Probabilistic Pose Estimation using Cam-
eras and Reflective Markers; ▶ 3D Dynamic Pose Estimation Using Cameras and
No Markers). While simplified multibody musculoskeletal models are useful for
investigating muscle coordination and resultant joint loading, they do not allow
estimation of the soft tissue loads needed to constrain joint motion. To address this
limitation, high resolution finite element (FE) models have been used to elucidate the
complex interactions of morphology, micromotion, and soft tissue loading for many
of the joints in the human body. However, due to their computational complexity, FE
simulations are typically driven at the joint-level using a combination of prescribed
kinematics and external loads. Therefore, they inherently cannot account for the
interrelationship between whole-body dynamics, muscle forces, and joint-level
mechanics. Improved algorithms and computational hardware will likely facilitate
multiscale and multidomain simulations in the future. However, with current tech-
niques, it remains exceedingly computationally demanding to incorporate detailed
finite element joints into a muscle driven simulation framework.
In this chapter, we first review measurement techniques to assess movement
dynamics and internal tissue mechanics. We show that in vivo measurements of
soft tissue loads remain inherently limited for practical use, necessitating a modeling
approach to characterize and predict in vivo tissue behavior. We then review
traditional sequential simulation approaches where multibody movement dynamics
are solved independently of joint mechanics. Finally, we introduce a state-of-the-art
concurrent simulation approach to simultaneously solve for muscle forces and
internal joint mechanics that generate observed whole body movement (Fig. 2).
The concurrent approach ensures dynamic consistency between the simulated full
body movement and joint-level mechanics, resulting in joint kinematics that evolve
naturally from the soft tissue forces. Concurrent simulations avoid the complexity of
the FE approaches without requiring the oversimplified joint representations of
Fig. 2 Sequential vs. Concurrent modeling of soft tissue loading. The traditional “Sequential
Solution” first resolves the muscle forces using a simplified multibody model, then applies these
forces to a detailed joint model. While this approach enables the use of complex, high-fidelity joint-
level models to assess ligament and cartilage loading, it de-couples the muscle forces from joint-
level mechanics. An alternative “Concurrent Solution” approach has recently been proposed,
wherein the muscle forces and joint mechanics are solved simultaneously such that the joint
kinematics, ligament forces, and cartilage pressures are directly coupled with the muscle forces.
In each image, a padlock symbol is superimposed to indicate the presence or absence of kinematic
joint constraints. In the “Sequential Solution,” the musculoskeletal model is the OpenSim Gait 2392
model, and the joint model is the Open Knee finite element model (Erdemir 2013, 2016). The
“Concurrent Solution” model is described by Lenhart et al. (2015a)
traditional musculoskeletal models. The ability to describe joint motion completely

using soft tissue loads and contact between articulating joint surfaces is retained by
using strand-based ligaments and elastic foundation contact models. This approach
is approximately two orders of magnitude faster than a comparable FE simulation
(Guess et al. 2013), enabling many more simulations to be performed with various
individuals, activities, or sets of model parameters. While a FE approach may be
used to analyze only a few discrete instants during a dynamic motion (Adouni and
Shirazi-Adl 2014), a concurrent approach can be used to simulate the entire motion
(Smith et al. 2016b). Hence, the computational efficiency of the concurrent approach
enables stochastic (e.g., Monte Carlo) simulations to more fully probe the sensitivity
of model predictions to inherent uncertainty in measurements and model parameters.
State of the Art
Invasive Measurement of In Vivo Loading
Current technologies enable two methodologies to quantify soft tissue loading:

either measure the soft tissue loads directly (which can require invasive procedures)
or take indirect measurements and use a mathematical model to estimate the
unknown quantities. While this chapter will focus on the second option, we will
briefly survey direct in vivo measurements of human soft tissue loads, as these
measurements are critical for establishing model validity.
In vivo ligament and tendon forces have been measured in humans (Fleming and
Beynnon 2004) using a variety of sensors such as “buckle-gage” transducers
(Fukashiro et al. 1995), arthroscopically implantable force probes (Roberts et al.
1994), or, more recently, optical fiber transducers (Finni et al. 2000). In each case,
soft tissue tension causes a measureable deflection of the sensor. Unfortunately,
buckle transducers require a relatively large incision for implantation, and subjects
may not fully recover for up to 3 weeks (Fleming and Beynnon 2004). Implantable
force probes and optic fiber methods are much less invasive, but sensor mis-
alignment and migration can induce large measurement errors (Fleming and
Beynnon 2004). Alternatively, implantable displacement sensors (Beynnon et al.
1997; Ravary et al. 2004) and noninvasive image-based methods (MRI, fluoroscopy,
ultrasound) have been used to measure in vivo soft tissue strain (Li et al. 2005; Nissi
et al. 2007; Chernak et al. 2014) (▶ Cross-platform Comparison of Imaging Tech-
nologies for Measuring Musculoskeletal Motion; ▶ Ultrasound Technology for
Examining the Mechanics of Muscle, Tendon and Ligament). When combined
with estimates of tissue properties from ex vivo cadaver experiments, these tissue
strain measurements can be used to approximate soft tissue loading.
Intra-articular joint contact pressure has only been measured in vivo in a handful
of studies. Anderson et al. (2003) performed a quasi-static experiment during
surgery where thin-film resistive sensors were interposed between the femoral and
tibial cartilage while subjects were asked to stand with and without knee braces.
Similarly, Rikli et al. (2007) temporarily inserted a thin-film capacitive sensor into
the radioulnocarpal joint to measure contact pressure during active wrist motion.
Both studies reported that it was not feasible to implant a sensor large enough to
cover the entire joint surface; therefore, some of the force was shunted around the
sensor. Hodge et al. (1986) measured contact pressures in the hip joints of two
human subjects after hemiarthoplasty. Contact pressure was measured at 14 locations
between an engineered metal femoral head and the natural cartilage of the acetabular
cup. Interestingly, the in vivo measurements revealed substantially higher local
contact pressures (up to 18 MPa) than computational models and cadaveric mea-
surements had reported at the time.
Perhaps the most reliable, and oft-cited, in vivo data are actually an indirect
measure of soft-tissue loading: strain-gauge measurements from instrumented
replacement joints (D’Lima et al. 2013). Rydell (1966) was the first to transduce
dynamic, in vivo joint loads in humans using a hip prosthesis that was instrumented
with strain gauges. Subsequent investigations introduced wireless measurements
(English and Kilvington 1979), synchronized these internal force measurements
with external motion capture data (Bergmann et al. 1993), and extended the
in vivo measurements to other joints (Bergmann et al. 2007; D’Lima et al. 2013).
In vivo joint load measurements indirectly reflect the net contribution of muscle,
ligament, and connective tissues. However, the redundancy of the soft tissue contri-
butions permits an infinite set of feasible soft tissue loads that could induce the net
joint loads. Furthermore, in vivo data are only available for participants who have
undergone invasive procedures where soft tissues may have been surgically altered,
and whose joints inherently no longer possess the same morphology or material
properties as natural joints. Thus, while valuable, these data should be extrapolated
to the natural human condition with caution.
In vivo measurements of soft tissue loading have provided invaluable data for:
(a) establishing loading thresholds for the design and testing of surgical devices and
(b) for validating the predictions of computational models. However, these experi-
mental measurements are invasive and difficult to obtain during dynamic activities
and can be highly sensitive to measurement technique, calibration, and sensor
misalignment. Thus, there remains a need for using computational models to esti-
mate unmeasurable quantities and, perhaps more importantly, to predict how loading
may change in response to injury, disease, and treatment.
External Measurement of Movement
The first step toward simulating soft tissue loading during observed motion is to
obtain experimental measures of movement kinematics and external forces (Fig. 2).
Despite recent advances in wearable sensors (Wong et al. 2015), optical motion
capture using skin-mounted markers remains the standard for laboratory measure-
ment of human kinematics (Cappozzo et al. 2005; Chiari et al. 2005; Leardini et al.
2005). Cameras record the three-dimensional trajectories of optical markers attached
to the subject’s skin, and these markers are used to infer the motion of the underlying
bones. Joint angles can be computed either directly from the 3D positions of the
markers (Wu and Cavanagh 1995; Wu et al. 2002) or by optimizing the joint
coordinates of a skeletal model to minimize error between model-fixed and exper-
imental marker positions through a process called Inverse Kinematics (Lu and
O’Connor 1999; Kainz et al. 2016). (▶ 3D Dynamic Pose Estimation Using Reflec-
tive; ▶ 3D Kinematics of Human Motion; ▶ 3D Dynamic Probabilistic Pose Esti-
mation From Data Collected Using Cameras and Reflective Markers).
Dynamic imaging techniques such as fluoroscopy (You et al. 2001; Gerber et al.
2007) and magnetic resonance imaging (Logan 2004; Kaiser et al. 2013) enable
precise measurement of 6 degree of freedom (DOF) kinematics for a joint of interest.
While these data are not influenced by soft tissue artifact (Leardini et al. 2005),
constraints on their field of view enable measurement of only a single joint.
Nevertheless, dynamic imaging data can be quite valuable for validation of joint
level models (Lenhart et al. 2015a), and the field will likely see broader use of such
data as availability of dynamic imaging systems becomes more widespread.
(▶ Cross-platform Comparison of Imaging Technologies for Measuring Musculo-
skeletal Motion; ▶ 3D Musculoskeletal Kinematics using Dynamic MRI; ▶ Mea-
surement of 3D Dynamic Joint Motion Using Biplane Videoradiography).
External loads applied to the human body may be measured using a variety of
sensors (Wong et al. 2015). In a laboratory setting, and for studies of human gait, a
ground-embedded force platform is commonly used to record six degree-of-freedom
interaction forces between the foot and the ground (Grundy et al. 1975). Addition-
ally, electromyogram (EMG) data (▶ Surface Electromyography to Study Muscle
Coordination) can be acquired to either define (Lloyd and Besier 2003; Buchanan
et al. 2004) or validate the simulated muscle excitations (Erdemir et al. 2007).
Status Quo: Sequential Modeling
Traditionally, soft tissue loads are obtained through the sequential solution of one or
more models at differing physiological scales (Fig. 2). First, a whole-body muscu-
loskeletal model is used to estimate the musculotendon forces and resulting net joint
loads necessary to generate the observed motion. (▶ Rigid Body models of the
Musculoskeletal System; 3D Kinetics of Human Motion; ▶ Physics-based Models
for Human Gait Analysis) This whole-body model typically contains a simplified set
of coordinates and constraints for each joint (e.g., ball and socket at the hip, hinge at
knee and ankle). Then, a finer-resolution joint-level model uses musculotendon
loads and joint pose as boundary conditions and predicts the deformation and
internal loading of ligaments, connective tissue, and articular cartilage. This joint-
level model may include a rigid (Schipplein and Andriacchi 1991), elastic founda-
tion (Blankevoort et al. 1991), or finite element representation (Besier et al. 2005) of
articular contact, and varying degrees of kinematic and kinetic constraints. The
challenges associated with both of these modeling stages are detailed in the follow-
ing sections.
Muscle Force Distribution

Perhaps the hardest and oldest challenge in modeling the human musculoskeletal
system during functional movement is the distribution of muscle forces among
agonist, and antagonist, muscles. Each joint of the human body is spanned by
more muscles than required to control the joint motion, leading to the “muscle
redundancy problem.” Furthermore, antagonist muscles that act in opposite direc-
tions at a given joint are often simultaneously activated; thus, for each joint there is
an infinite set of muscle forces that will generate an observed motion.
Early mathematical models resolved muscle redundancy through physiologically
informed simplification. Morrison (1970) outlined a series of hypotheses about knee
mechanics which formed the basis for his musculoskeletal model of soft tissue
loading in the knee. For example, any anterior muscle or external force applied to
the knee joint was solely resisted by the anterior cruciate ligament (ACL), torsional
action about the proximal-distal axis of the joint was neglected, and redundancy of
knee flexor muscles was resolved by comparing experimental muscle excitation
(EMG) data to decide whether hamstrings or gastrocnemii muscles should be
activated at a given instant. These simplifying assumptions captured salient features
of knee mechanics and resulted in a model that provided a unique solution for
muscle, joint contact, and primary ligament loading during observed movements.
However, the accuracy of soft tissue load estimates, and implications regarding
relationships between soft tissue loading and joint kinematics, was limited.
Subsequent models attempted to include greater flexibility in muscle coordina-
tion, by optimizing the force distribution at discrete instants in time to minimize an a
priori objective function (Seireg and Arvikar 1975; Hardt 1978; Pedotti et al. 1978;
Crowninshield and Brand 1981). This process is known as static optimization, where
the muscle forces needed to generate the dynamic joint moments are estimated. Note
that the “static” terminology refers to independent solution of the optimization
problem at each time step, as opposed to dynamic optimization which solves all
time steps simultaneously. Motor control studies have shown that humans appear to
exhibit varying degrees of optimality in motor planning and execution (Todorov and
Jordan 2002; Bertram 2005; Ota et al. 2016); therefore, using optimization to
emulate physiological distribution of muscle forces is not just mathematically
convenient, but also desirable (Hardt 1978). Unfortunately, it remains very chal-
lenging to establish a true “objective” of the human motor control system in
performing functional movements. Many investigations have shown that the mini-
mization of the sum of squares or cubes of muscles forces (Pedotti et al. 1978;
Crowninshield and Brand 1981) sometimes weighted by muscle volume (Happee
and Van der Helm 1995) can provide muscle force estimates that are temporally
similar to experimental EMG data in healthy normal subjects. However, subtle
changes to the objective function can lead to dramatic changes in estimated joint
loads (Demers et al. 2014).
Many other approaches have been implemented to resolve the muscle force
distribution problem for observed movements, including EMG-driven modeling
(Lloyd and Besier 2003; Buchanan et al. 2004), synergy-based optimization (Sartori
et al. 2013), dynamic optimization (Davy and Audu 1987; Yamaguchi and Zajac
1989; Anderson and Pandy 2001), and various tracking algorithms including com-
puted muscle control (Thelen et al. 2003), neuromusculoskeletal tracking (Seth and
Pandy 2007), and PID control of joints (Guess et al. 2014; Kłodowski et al. 2016)
(▶ Optimal Control Modeling of Human Movement). These approaches may pro-
vide advantages over static optimization if, for example, the subject of interest
exhibits abnormal muscle activation (EMG-driven modeling) or is moving rapidly
such that muscle contraction dynamics are significant (dynamic optimization). Any
of these methods can be suitable for distributing muscle forces in a two-stage
sequential modeling approach.
Ligament and Cartilage Loading

After the muscle forces are resolved in the first stage of the sequential modeling
workflow, they are applied to a detailed joint model to calculate soft tissue loading.
With advancements in algorithms and computational hardware, these joint level
models have become increasingly sophisticated in the number of structures modeled
and the fidelity of their representations.
The most rudimentary models of passive tissue loading employ rigid assumptions
for articular contact, where the centers of joint rotation, points of contact, and
moment arms of each muscle and soft tissue are geometrically defined (Morrison
1970; Schipplein and Andriacchi 1991). Using these models, given a known set of
joint kinematics, external loads, and muscle forces, it is possible to estimate a limited
set of ligament and net contact forces. However, in order to formulate a mathemat-
ically determinate system, such models include only a small subset of the passive
structures crossing a joint and typically fail to capture any elastic deformation of soft
tissues. This approach is computationally efficient and may yield reasonable results
for the overall magnitude of joint contact forces (Winby et al. 2009; Brandon et al.
2014). However, it fails to yield insight into the interaction of kinematics and soft
tissue loads because kinematics are prescribed prior to the soft tissue analysis.
A more nuanced description of joint mechanics is enabled by the use of a rigid
body, penetration-based “elastic foundation” model for articular contact
(Blankevoort et al. 1991; Bei and Fregly 2004) and strand-based ligaments
(Blankevoort and Huiskes 1991). In the elastic foundation model, each articular
surface is discretized into a surface mesh of small polygonal elements with inde-
pendent springs acting normal to each face. When contact occurs, the interpenetra-
tion of rigid mesh surfaces is used to compute spring deflection, and thereby
pressure, for each surface element. Strand-based ligament models use bundles of
nonlinear springs to represent the elastic contribution of ligaments to joint mechan-
ics. Using these approaches, joint models where all six degrees of freedom are
controlled by the combined action of muscle, ligament, and articular contact loads
have been developed (Essinger et al. 1989; Halloran et al. 2005). However, an elastic
foundation model only provides surface contact pressures and cannot resolve inter-
nal tissue-level stresses and strains.
Finite element (FE) analysis is a powerful method for quantifying the mechanics
of ligament, capsule, and cartilage tissues (Kazemi et al. 2013; Zheng et al. 2016).
The fundamental principal of FE analysis is to discretize complex tissue geometries
into a set of small three-dimensional elements that can be analyzed using numerical
solution techniques. Finite element modeling enables studies such as the sensitivity
of passive knee joint mechanics to ACL reconstruction graft parameters (Peña et al.
2005; Dhaher et al. 2016), the effect of a meniscectomy on knee cartilage stress
(Tanska et al. 2015), the relationship between focal cartilage defects and osteoar-
thritis (Papaioannou et al. 2010), or evaluations of the stability of the hip (Ferguson
et al. 2000) or glenohumeral (Ellis et al. 2010) joints. Finite element modeling can
even be used to investigate the importance of depth-wise changes in mechanical
structure within a thin layer of articular cartilage (Halonen et al. 2013) or to simulate
changes in cartilage stiffness and strength due to variation in nanoscale cross-link
density between tropocollagen molecules (Adouni and Dhaher 2016). The primary
limitation on the use of FE models in biomechanical studies is computational cost.
For example, a finite element representation of the knee, including poroviscoelastic
properties for the cartilage, may take days or even weeks to solve under quasi-static
conditions (Kazemi et al. 2013). Thus, in the sequential modeling workflow, joint
mechanics are often resolved only at specific instances in a movement.
Limitations of Sequential Approach

While the sequential approach has yielded valuable insight into the magnitudes of
soft tissue loading and provides a straightforward methodology to simulate soft
tissue loading across multiple scales, several limitations exist. In the sequential
approach, muscle forces are often computed to equilibrate only a subset of the
available degrees of freedom. However, a different solution will be obtained if
additional degrees of freedom are considered (Jinha et al. 2006; Marouane et al.
2016). Perhaps more importantly, the sequential modeling approach solves the full-
body and joint-level dynamics separately, neglecting the interaction between scales
and thereby failing to account for inherent coupling. As a result, it does not allow
muscle forces to be affected by subtle changes in joint kinematics which arise via
deformation of ligament, capsular, and cartilage tissues. The human neuromuscular
system modulates muscle forces to control joint rotations and translations; muscle
forces therefore arise in proportion with their capability to induce accelerations along
each degree of freedom. But the contribution of a muscle to accelerating each degree
of freedom is not static; it changes dynamically with the pose of a joint (▶ Induced
Acceleration and Induced Power Analyses of Human Motion). Using the knee for
example (Lin et al. 2010b), quadriceps forces are directly affected by the position of
the patella. However, patellar position is influenced by translation of the tibiofemoral
joint as well as strain in the patellar tendon, and these parameters vary with
quadriceps force. Thus, to analyze soft tissue loading during dynamic motions, it
is imperative to ensure that joint kinematics are coupled with soft tissue loads. The
coupling between dynamic motion, muscles, and soft tissue loads is likely amplified
in cases where soft tissue injury or disease exists and compromises the native
function of the joint.
Concurrent Methods
To capture the interactions between body scale musculoskeletal behavior and joint
scale soft tissue mechanics, the dynamics of each scale must be coupled. For
example, coupled simulations are needed to investigate how one may alter neuro-
muscular coordination to adapt for ligament damage that compromises joint stability.
However, concurrent solution of the neuromuscular control and joint behavior
necessitates different modeling and simulation techniques than are required for
sequential simulation approaches.
Articular Contact and Ligament Models
Ideally, a soft tissue FE representation of the joint mechanics would be solved within
a multibody muscle-driven simulation, thereby allowing for characterization of the
tissue deformation and stress patterns. However, the practical challenges and com-
putational demands of simultaneously solving multibody dynamic and finite element
models have generally precluded this approach (Koolstra and Van Eijden 2005). To
avoid the computational bottle neck, investigators have either used surrogate
approaches to limit the number of FE model calls, or opted for simpler elastic
foundation models and strand-based structures to represent articular cartilage contact
and ligaments, respectively (see Table 1 for summary of 3D knee models that use
such approaches).
Surrogate approaches can be used to generate a numerical input-output relation-
ship for a complex finite element model that, following calibration, can be queried
efficiently within a dynamic simulation. For example, Lin et al. (2006, 2010a) were
able to pre-compute a regression relationship between joint kinematics and contact
loads for a finite element knee model over a large range of anatomically reasonable
kinematic poses. In a subsequent multibody dynamic simulation, they showed that
the regression equations could be queried in place of solving the FE model, with
negligible computational cost and relatively close adherence to the FE solutions. An
alternative “adaptive surrogate modeling,” or “lazy-learning,” scheme has also been
proposed (Halloran et al. 2009), wherein a FE model is still used to generate a
surrogate regression model, but the FE model is only solved when the input states
differ significantly from previously encountered states within a dynamic simulation.
For situations where the model parameters are well defined, surrogate modeling
approaches can afford significant computational savings. However, any alteration in
finite element model parameters or geometry would require re-calibration of the
surrogate regression equations, which can make it computationally demanding to
fully assess sensitivities of the model outcomes.
The alternative approach used to represent joint mechanics within a dynamic
simulation uses strand-based ligaments and an elastic foundation articular contact
model. As described above (Ligament and Cartilage Loading), the elastic foundation
model solves for the contact pressure on each face in a surface mesh based on the
local penetration depth. In doing so, the pressure on each face is solved
12
Table 1 Summary of three-dimensional knee models in literature. Included models contained: all three bones (Tibia, Femur, Patella), both tibiofemoral
(TF) and patellofemoral (PF) joint contact, a six-degree-of-freedom (DOF) tibiofemoral joint, both muscle and contact forces, and high-fidelity contact surfaces
Solution
Year(s) Authors Geometry Meniscus Contact Model Method Muscle Control Motions
1989 Essinger et al. TKA n/a Elastic Foundation Sequential Passive Springs Flexion-
Extension
1998a, b Kim Native – Elastic Foundation (TF) / Sequential Constant Force Flexion-
Rigid (PF) Extension
1998a, b Pandy et al. NativeT,P – Elastic Foundation (PF) / Sequential Passive; Constant Flexion-
Rigid (PF) Force Extension
2001 Piazza and Delp TKA n/a Rigid Surfaces Sequential EMG-driven Step-up
2004 Caruntu and Hefzy NativeT – Elastic Foundation Sequential Constant Force Flexion-
Extension
2004, 2005, Shelburne et al. Native Posterior Shear Elastic Foundation Sequential Dynamic Gait, Vertical
2006 Force Optimization Jump
2007, 2009, Shin et al. Native – Elastic Foundation Concurrent Passive Springs Single Leg
2011 Landing
2010 Dhaher et al. Native Finite Element Finite Element Sequential Constant Force Extension
2010, 2012, Guess et al. Native Multibody Elastic Foundation Concurrent PD Controller Squat, Gait
2013 Deformable
2013 Hast and Piazza TKA n/a Elastic Foundation Sequential Dual Joint Gait
Optimization
2013, 2014 Stylianou et al., TKA n/a Elastic Foundation Concurrent PID Controller Squat,
Guess et al. Toe-Rise, Gait
2014, 2016, Adouni et al., Native Finite Element Finite Element Concurrent Static Optimization Gait
2016 Marouane et al. (6 instants)
2014 Thelen et al. TKA – Elastic Foundation Concurrent Computed Muscle Gait
Control
S.C.E. Brandon et al.
(continued)
Table 1 (continued)
Solution
Year(s) Authors Geometry Meniscus Contact Model Method Muscle Control Motions
2015a Lenhart et al. Native – Elastic Foundation Concurrent Computed Muscle Flexion-
Control Extension
2015, 2016 Marra et al., Chen TKA n/a Elastic Foundation Concurrent Force-Dependent Gait
et al. Kinematics
2016 Halonen et al. Native Finite Element Finite Element Sequential PID Controller Gait
2016a, b Smith et al. Native/ – Elastic Foundation Concurrent COMAK Gait
TKA
2016 Eskinazi et al. TKA n/a Surrogate Finite Element Concurrent PD Controller Flexion-
Extension
T
Geometry assumed planar tibia surface; PGeometry assumed planar patella surface
Simulation of Soft Tissue Loading from Observed Movement Dynamics
13
independently, making the problem more computationally tractable within the con-
text of a whole body dynamic simulation. Advanced algorithms have recently been
introduced which use bounding volume hierarchy techniques developed by the
computer graphics field to further improve the speed of elastic foundation
implementations (Smith et al. 2016c). Strand-based ligaments also provide vastly
improved speeds compared to FE models by representing ligaments as bundles of
independent one-dimensional nonlinear springs.
Simulation Techniques
Six degree of freedom, multibody joint models can greatly enhance the information
obtained in a musculoskeletal simulation of movement. Traditionally, movement
simulation models use simplified joints (e.g., hinge or ball-and-socket) in which
artificial mathematical constraints restrict secondary motion. In contrast, a six degree
of freedom joint model directly includes representations of the soft tissues (liga-
ments, cartilage) that span a joint and thus can provide estimates of the physiological
loads needed to constrain motion. A number of investigators have demonstrated the
viability of simulating dynamic six degree of freedom joint behavior during passive
and simple movement conditions (Table 1, concurrent solution method). However,
six DOF joints introduce significant complexity into calculation of the neuromuscu-
lar control needed to generate active, multijoint movement. Conventional tracking
control algorithms are predicated on linked-segment models with constrained joints.
With constrained joints, the capacity of individual muscles to transmit joint loads
and induce whole body accelerations is readily calculated, and this information is
then used to compute controls that generate a desired movement (Thelen et al. 2003;
Thelen and Anderson 2006; Seth and Pandy 2007). In a six degree of freedom joint
with soft constraints, muscle loads are no longer instantaneously transmitted across a
joint, complicating the calculation of a muscle’s capacity to induce acceleration
along a specific DOF. This makes it challenging to use static optimization
approaches to resolve muscle redundancy at a discrete time step within a dynamic
simulation. Dynamic optimization is a powerful alternative for determining a muscle
activation patterns that generate a desired movement (Davy and Audu 1987; Ander-
son and Pandy 1999, 2001; Lee and Umberger 2016), but dynamic optimization is
much more computationally expensive and as a result not widely used to simulate
subject-specific movement.
Recently, two different research groups have introduced approaches for comput-
ing muscle coordination patterns and six degree of freedom joint mechanics that are
dynamically consistent with observed movements (Andersen et al. 2011; Thelen
et al. 2014; Smith et al. 2016b). These studies define a set of primary degrees of
freedom (DOF), which can be readily and reliably measured using standard gait
analysis techniques (e.g., pelvis translations and rotations, hip rotations, tibiofemoral
flexion, ankle flexion) (▶ 3D Dynamic Pose Estimation Using Cameras and Reflec-
tive Markers). A set of secondary DOF is also defined, which consist of kinematics
(e.g., tibiofemoral translations and non-sagittal rotations) that are poorly measured
with conventional motion capture (Leardini et al. 2005; Li et al. 2012). A multibody
musculoskeletal simulation is then performed where the primary degrees of freedom
track observed movement, and the secondary DOF evolve naturally from the soft
tissues acting within a six degree of freedom joint. Muscle force distribution within
these algorithms (Andersen et al. 2011; Thelen et al. 2014) is resolved using
numerical optimization at each time frame to minimize an assumed cost function,
such as the sum of weighted muscle activations squared (Crowninshield and Brand
1981; Happee and Van der Helm 1995). For relatively slow activities such as human
gait, these approaches have been shown to yield patterns of muscle activation that
are similar to experimental EMG and yield reasonable predictions of in vivo knee
contact forces (Marra et al. 2015; Smith et al. 2016b). Furthermore, these methods
are not restricted to knee joint analysis. The method of Andersen et al. (2011), called
force-dependent-kinematics (FDK), has been recently used to examine joint
mechanics at the spine (Ignasiak et al. 2016), shoulder (Sins et al. 2015), wrist
(Eschweiler et al. 2016), and hip (Zhang et al. 2015) joints. In the following section,
we will present our algorithm, Concurrent Optimization of Muscle Activations and
Kinematics (COMAK), to simulate full six degree of freedom tibiofemoral and
patellofemoral mechanics during gait (Lenhart et al. 2015b; Lenhart et al. 2016;
Smith et al. 2016a, 2016b).
Simulating Soft Tissue Loads
Musculoskeletal Model
Multibody Musculoskeletal Model

A physics-based computational model is essential for the simulation of soft tissue
loading. The development of such a model requires the anatomic structure and
dynamic function of the musculoskeletal system to be translated into mathematical
representations. Traditionally, the topology of the musculoskeletal system is
described as a chain of rigid segments, each with unique inertial properties,
connected via idealized joints such as a ball and socket hip joint, or revolute
talocrural joint. A set of generalized coordinates (q) is defined to quantify the
individual rotational and translational degrees of freedom (DOF) of each joint. The
state of the resulting multibody model is fully determined by the generalized
coordinates (q) and generalized speeds (q).
Muscle-tendon units are most often modeled as line-segment actuators that
originate on one bony segment and insert onto another. To accurately capture their
lines of action and prevent penetration into the bone geometries, muscle paths are
often constrained to wrap over discrete points or primitive shapes. Muscle-tendon
dynamics are often described by Hill type muscle models, in which muscle force is
dependent on activation, length, and rate of contraction (Hill 1938; Thelen 2003;
Millard et al. 2013) (▶ Hill-Based Muscle Modeling). In this chapter, we will use a
Fig. 3 A 12 degree of freedom knee is incorporated into a lower extremity musculoskeletal model,
allowing for analysis of ligament and cartilage loading during gait
simplified version where each muscle’s force, Fmuscle

i , scales linearly with its
activation level ai,
Fmuscle
i ¼ ai Fmax
i (1)
Here, Fmax
i is the assumed maximal force of a given muscle i.
It is possible to generate subject-specific musculoskeletal geometries from imag-
ing data (Scheys et al. 2005; Blemker et al. 2007; Scheys et al. 2008; Hausselle et al.
2012; Valente et al. 2014) (▶ Cross-platform Comparison of Imaging Technologies
for Measuring Musculoskeletal Motion). However, subject-specific modeling
remains an intensely laborious process. Hence, a more common approach is to
scale a generic model based on subject-specific anthropometric measures. There
are numerous generic models available for the lower (e.g., Arnold et al. 2010; Brand
et al. 1982; Carbone et al. 2015; Delp et al. 1990; Klein Horsman 2007; Modenese
et al. 2010; Rajagopal et al. 2016) and upper (e.g., Christophy et al. 2012; Seth et al.
2016; Van der Helm et al. 1992) extremities. In the methods described below, we
adapt the generic musculoskeletal geometry of the lower-limb as described by
Arnold et al. (2010) (Fig. 3).
Detailed Joint Model

Traditional musculoskeletal models represent the links between segments as ideal-
ized kinematic joints (e.g., ball-and-socket hip and a hinge knee joint). In doing so,
the contributions of passive structures (ligaments, joint capsule and connective
tissue) and articular contact are implicitly accounted for by the kinematic constraints.
However, to analyze the loading in these soft tissues, the definition of the joint of
interest must be refined to remove the artificial constraints and include explicit
representation of the soft tissues.
To study soft tissue loading and internal joint mechanics of the knee, we replaced
the one DOF knee joint in the Arnold model with a knee that has six DOF
tibiofemoral and patellofemoral joints. Removal of the artificial kinematic con-

straints requires that the force contributions of the passive structures and articular
contact be explicitly modeled and calculated. To obtain the geometries of these
structures, we segmented the bone, ligament, and cartilage of a healthy young adult
subject from high resolution MRI images (Lenhart et al. 2015a). Each ligament was
represented in the model as a bundle of nonlinear springs. The ligament force-strain
relationship was assumed quadratic at low strains and linear at high strains to capture
the nonlinear effects of collagen crimp straightening and fiber elongation
(Blankevoort and Huiskes 1991).
8 1 9
>
> 2
0 e et >
>
< 2e ke =
Fspring ¼ t
et (2)
>
> k e e > et >
>
: 2 ;
0 e<0
Fdamping ¼ kce_ (3)
Fligament ¼ Fspring þ Fdamping (4)
In these equations, e is the strain, et is the strain where the model transitions from
the quadratic toe region to the linear region, k is the linear stiffness, and c is the
damping coefficient.
The articular contact forces are computed using an elastic foundation model (Bei
and Fregly 2004). The articular surfaces are represented by high resolution triangular
meshes that do not deform, but are allowed to interpenetrate. Contact pressure ( p) on
each triangle face is computed based on the local penetration depth (d), cartilage
thickness (h), and material properties.

ð1 vÞE d
p¼ ln 1 (5)
ð1 þ vÞð1 2vÞ h
The material properties of the cartilage are defined by the elastic modulus (E) and
Poisson’s ratio (v).
Body-Segment Dynamics
Multibody dynamic equations of motion can be derived to relate applied forces to

accelerations of the generalized coordinates. For any model of reasonable complex-
ity, computational dynamics engines such as SD/Fast (Hollars 1994) and Simbody
(Sherman et al. 2011) can readily generate computational implementations of the
equations of motion. These software packages enable each of the generalized
accelerations to be defined as prescribed or unconstrained. At any point in time,
the generalized accelerations can be computed by solving the equations of motion of

the form

MðqÞ€ _ Fext , Fmusle
q þ AT ðqÞλ ¼ f q,q, (6)
where q, q,_ and q€ are the generalized coordinates, speeds, and accelerations of the
model, respectively. M is the system mass matrix, A is the constraint coefficient
matrix, λ is the constraint multiplier array used to enforce a subset of prescribed
accelerations, and f is an array of generalized forces that includes the effects of
external forces, muscle forces, ligament forces, contact, and gravity. The computa-
tional implementation of the dynamics of the musculoskeletal system allows simu-
lations of functional movement to be readily performed.
COMAK: Concurrent Optimization of Muscle Activations

and Secondary Kinematics
We recently introduced the Concurrent Optimization of Muscle Activations and

Kinematics (COMAK) algorithm (Fig. 4) to simultaneously solve for muscle and
soft tissue loading during functional movement. In COMAK, inverse kinematic
measurement techniques (Lu and O’Connor 1999) are first used to compute the
_ qÞ of the primary model degrees of
coordinates, speeds and accelerations ðq,q,€
freedom. Then, numerical optimization is performed to simultaneously solve for
the secondary kinematics, muscle, ligament, and articular contact forces that gener-
ate the primary joint accelerations while minimizing a cost function that resolves
inherent muscle redundancy.
Observable Kinematics
Optical motion capture enables the measurement of segment kinematics during
functional movement (▶ 3D Dynamic Pose Estimation Using Reflective Markers
or Electromagnetic Sensors; ▶ 3D Kinematics of Human Motion; ▶ 3D Dynamic
Probabilistic Pose Estimation From Data Collected Using Cameras and Reflective
Markers). However, because skin and soft tissue motion prevent direct observation
of the underlying bones, motion capture is limited in its ability to quantify degrees of
freedom which undergo small excursions during full body movement (Benoit et al.
2006; Li et al. 2012). The differentiation between measurable DOFs of high (pri-
mary) and low (secondary) confidence is a key concept in COMAK. The algorithm
solves for the muscle and soft tissue loads necessary to generate the measured
motion of the primary DOFs while simultaneously predicting a dynamically con-
sistent set of secondary kinematics. In our studies of the lower extremity, we have
defined three-dimensional hip rotations, tibiofemoral flexion and ankle flexion as the
primary DOFs. The tibiofemoral translations, nonsagittal rotations, and all
patellofemoral DOFs cannot be as reliably measured (Leardini 2005) and thereby
form the secondary DOFs. Pelvis translations and rotations are also measurable, but
classified as prescribed DOFs such that their accelerations are prescribed within the
multibody model to ensure consistency with observed multibody dynamics.
The joint kinematics of the primary and prescribed DOFs are calculated from the
measured motion capture marker trajectories using inverse kinematics. The inverse
kinematics routine is formulated as a global optimization to determine the general-
ized coordinates of the primary and prescribed DOFs that minimize the sum of
squared differences between model marker positions and measured marker positions
at each time step (Lu and O’Connor 1999). To enable this calculation, the secondary
DOFs must be also determined, but are unknown at this stage. To account for this,
the secondary generalized coordinates are constrained to be functions of the primary
generalized coordinates during inverse kinematics optimization (Walker et al. 1988;
Gerus et al. 2013). These constraints are then removed when later performing the
optimization for dynamically consistent soft tissue loads and secondary kinematics
(Fig. 2). Fundamentally, this approach assumes that the differences between the
constrained secondary kinematics and load-dependent secondary kinematics subse-
quently predicted by COMAK have negligible influence on the primary coordinates
calculated by inverse kinematics. For the knee joint, internal rotation, varus-valgus
rotation, and all translations can be defined as functions of the knee flexion angle
which have been reported in literature (Walker et al. 1988). In practice, we calculate
secondary kinematics that are consistent with the articular geometry of the model by
performing passive (i.e., minimal muscle activation) forward simulations where a
primary DOF is prescribed to travel through its range of motion and the secondary
kinematics evolve as a result of the contact, ligament and passive muscle forces
(Lenhart et al. 2015a).
Simultaneous Optimization
After inverse kinematics, an optimization problem (COMAK) is solved to simulta-
neously predict the muscle and soft tissue loading and secondary kinematics
required to generate the measured primary accelerations. The optimization is for-
mulated to solve for the muscle activations and secondary coordinates that minimize
an objective function while satisfying overall dynamic constraints (Fig. 4). The
dynamic constraints require that optimized muscle forces and internal joint loads
(ligament and contact forces) resulting from the optimized secondary kinematics
generate the measured primary accelerations, while inducing equilibrium (zero
accelerations) in the secondary DOFs.
At the first time step in COMAK, the prescribed coordinates, speeds, and
_ qÞprescribed and primary coordinates and speeds ðq,q_ Þprimary are
accelerations ðq,q,€
set to their observed values, and a forward simulation is performed with minimal
muscle activations (ai =0.01) to allow the secondary coordinates, qsecondary, to settle
into an initial pose. At each subsequent time step, ðq,q_ Þprescribed and ðq,q_ Þprimary are set
to their observed values, while qsecondary and ai are determined by the optimization.
The secondary speeds, q_ secondary , are determined from the difference between
qsecondary at the current and previous time steps. After setting the states of the
model, the generalized forces are computed and applied. The contact forces are
Fig. 4 (continued)
calculated using the elastic foundation model, Fcontact(q) (Eq. 5), while the ligament
forces are computed using the nonlinear spring model, Fligament ðq,q_ Þ (Eqs. 2–4).
Viscous damping forces are applied to each DOF to ensure minimal changes in
_ . The muscle forces are computed from the
kinematics between time steps Fdamping ðqÞ
activations Fmuscle(ai) (Eq. 1) and the measured external forces,Fext , are applied to
their corresponding segments. The q€prescribed are then prescribed to their measured
values and the equations of motion are solved for q€primary and q€secondary .
Three constrains are enforced during the optimization for qsecondary and ai. Muscle
activations are constrained to vary between 0 and 1, to ensure the resulting muscle
forces are physiologically reasonable.
0 < ai < 1 (7)
Consistency with measured gait dynamics is ensured by satisfying the constraint

that the simulated accelerations of the primary degrees of freedom matched the
observed values,
nX
muscles
q€j observed primary ¼ ai Fmax ^€ muscle

ðqÞ þ q€other ðq,q_ Þ (8)
i q j, i j
i¼1
while the accelerations of secondary DOFs are constrained to be zero.
nX
muscles
q€secondary ¼0¼ ai Fmax ^€ muscle

ðqÞ þ q€other ðq,q_ Þ (9)
k i q k, i k
i¼1
In these equations, q^€ muscle

is the acceleration along coordinate j due to a unit force
j, i
in muscle i, while q€jother
constitute the accelerations due to all other forces in the
multibody system (contact, ligament, damping, external, gravitational, centripetal
and Coriolis). The third constraint (Eq. 9) assumes that inertial effects due to
accelerations in the secondary degrees of freedom are negligible. During gait, this
assumption is justified given the small mass of the patella and the small magnitudes
of rotational and translational excursion in secondary degrees of freedom. The
redundancy of the musculoskeletal system allows multiple combinations of qsecondary
and ai to fulfill these constraints, thus static optimization must be performed to
minimize an objective function and identify a unique solution.
Fig. 4 COMAK: Concurrent Optimization of Muscle Activations and Kinematics. The COMAK
algorithm is a concurrent simulation method that integrates a multibody musculoskeletal with a
detailed knee joint representation and external observations of full body movement to predict soft
tissue loading. The optimization is formulated to solve for the muscle activations and secondary
coordinates that minimize an objective function while satisfying overall dynamic constraints. The
dynamic constraints require that optimized muscle forces and internal joint loads (ligament and
contact forces) resulting from the optimized secondary kinematics generate the measured primary
accelerations, while inducing equilibrium (zero accelerations) in the secondary DOFs
The objective function (J ) used by COMAK is generalizable, allowing any user-

defined quantity to be minimized. We have found a common cost function proposed
for static optimization performs well for COMAK in most applications:
nX
muscles
J¼ W i V i a2i (10)
i
where Wi is a weighting term, Vi is muscle volume, and ai is the muscle activation.

The weighting term enables the activation of individual muscles to be penalized
within the optimization. Consistent with Demers et al. (2014), we have found
penalizing biarticular muscles (i.e., gastrocnemii and rectus femoris) necessary to
predict tibiofemoral contact forces consistent with measurements from instrumented
implants during walking. This redistributes the hip flexor moment and ankle
plantarflexor moments to the uniarticular muscles during late stance, reducing the
loading in the muscles crossing the knee and thus the compressive contact force.
Introducing an additional term in the cost function to minimize articular contact
energy also produces similar contact force predictions, largely by similarly
redistributing the muscle loading to uniarticular muscles (Smith et al. 2016b).
However, the errors in the numerical calculation of the gradient of contact energy
with respect to the optimized secondary kinematics can cause convergence issues
within the optimization. Thus, we have found the cost function presented above
(Eq. 10) to be more robust with similar results.
Example Application
The value of the concurrent simulation approach can be illustrated by simulating the
effects of an injury to the anterior cruciate ligament (ACL) (▶ ACL Injuries and
Gait; ▶ Objectifiying (sic) Surgical Treatment by Musculosceletal (sic) Modelling).
Within the human knee joint, the ACL functions to restrain anterior tibial translation
and internal tibial rotation (Cabaud 1983; Isaac et al. 2005). The ACL is often
injured during single-leg landing events (Shin et al. 2009; Sanders et al. 2016). After
injury, a frequent clinical observation is increased anterior translation when an
anterior load is applied to the tibia (Katz and Fingeroth 1986), and it has been
proposed that small kinematic changes within the joint could contribute to postinjury
incidence of osteoarthritis by altering the location of contact pressure during
dynamic activities (Andriacchi et al. 2009; Koo et al. 2011). Therefore, we demon-
strate the application of COMAK by comparing ligament and cartilage loads esti-
mated throughout an entire gait cycle in two conditions: (i) Intact, with a healthy,
normal joint and (ii) Injured, with a compromised ACL. For the purposes of this
demonstration, the ACL was “injured” by reducing its stiffness to 5% of the nominal
value, approximating a near-complete tear (Fig. 5).
Each COMAK simulation took approximately 10 minutes to run on a desktop
computer and revealed coupled changes in knee kinematics, contact pressure, and
Translation
20 Intact
Anterior
(mm)
Injured
10
0
0 50 100
Posterior Force
1
Tibia Contact
(BW)
0.5
0
0 50 100
0.4
Force
(BW)
MCL
0.2
0
0 50 100
Anterior
Translation Gait Cycle (%)
12
10
8
MPa
6
4
2
0
Fig. 5 Example simulation: prediction of changes in knee kinematics, ligament force, and contact
pressure due to simulated anterior cruciate ligament (ACL) injury. In this example, the COMAK
approach was used to simulate a single gait cycle for a representative subject with the knee joint
intact, and with ACL stiffness reduced to 5% of its nominal value. When the ACL was injured, there
was a dramatic increase in anterior tibial translation, with a posterior shift in tibiofemoral contact.
The MCL force, as well as other ligaments and muscles (not shown) increased to compensate for the
ACL injury
ligament and muscle forces between intact and injured conditions. When the ACL
was injured, the tibia shifted anteriorly relative to the femur, with a peak translation
early in the gait cycle at the time of quadriceps muscle activation (Winter and
Scott 1991). This anterior shift resulted in an increased posteriorly directed compo-
nent of the contact force and an increased force in the medial collateral ligament
(MCL). An examination of the pressure distribution on the tibial plateau revealed
both increased contact pressure magnitudes and a posterior shift in the contact
location for the injured condition (Fig. 5).
The simulations were obtained without altering the underlying objective function
that was used to resolve muscle redundancy. Further analysis could involve modi-
fying the cost function to ascertain if it is feasible to modulate muscle coordination to
restore normative tibiofemoral kinematics and/or tibiofemoral cartilage contact

locations. Such analyses are relevant in rehabilitation, where there has long been
interest in understanding the neuromuscular coping strategies of ACL-deficient
individuals (Beard et al. 1996; Houck and Yack 2003; Torry et al. 2004). The
model could also be used to simulate an ACL reconstruction procedure and predict
how graft placement and properties may affect knee mechanics after reconstruction.
These analyses are relevant to orthopedics where there is interest in understanding
the effects of orthopedic surgical factors on cartilage loading and potential for early
onset osteoarthritis. Another avenue of further study is the inclusion of a deformable
multibody meniscus (Guess et al. 2010) in order to better understand observed
relationships between meniscal damage, abnormal knee mechanics and osteoarthri-
tis. This example and proposed follow-up studies demonstrate the potential clinical
relevance of concurrent simulation approaches for understanding dynamic interac-
tions between soft tissues and joint kinematics during functional movement.
Future Directions
Soft tissue loading during functional movements plays an important role in the cause
of traumatic injuries, progression of degenerative joint diseases, and the success of
orthopedic treatments. In this chapter, we reviewed modeling approaches to quan-
tifying soft tissue loading from observed movement. Traditional sequential simula-
tion approaches inherently decouple movement dynamics and joint scale tissue
mechanics and thus cannot predict how muscle coordination may change in response
to localized joint tissue damage. More recent concurrent simulation approaches, e.g.,
COMAK, address this limitation by simultaneously solving for the secondary joint
kinematics and musculotendon, ligament, and articular contact loading that are
consistent with observed movement. To demonstrate the capabilities of concurrent
simulation approaches, we simulated an anterior cruciate ligament tear in a knee
model and predicted injury-induced changes in muscle, ligament, and articular
cartilage loading that were generally consistent with clinical observations.
Concurrent simulation methods enable the use of musculoskeletal modeling to
answer a variety of new clinically motivated research questions. For example, the
effects of ligament damage and treatments can be simulated during functional
movements. In our own work, we have studied the effect of ligament properties on
joint mechanics during walking in healthy (Smith et al. 2016a) and reconstructed
(Smith et al. 2016b) knees. Additionally, we have studied crouch gait treatments by
simulating surgical procedures to the knee joint and predicting the effects on knee
extensor function and cartilage contact pressures (Lenhart et al. 2016). We have also
used models to begin to investigate the influence of the alignment of knee replace-
ment components on load distribution on the tibial component (Smith et al. 2016b).
Looking forward, researchers will need to continue to balance the objective of
physiological realism with model simplifications needed for computational imple-
mentation and interpretation. In our concurrent simulation framework (COMAK),
we currently strike this balance by distributing muscle forces with static
optimization, modeling ligaments as bundles of strands, and computing contact

pressures with an elastic foundation model. While we have shown that our current
concurrent simulation models yield results that are consistent with in vivo data
(Lenhart et al. 2015a; Smith et al. 2016b), there are certainly opportunities for
improved fidelity. Strand-based ligaments do not allow for shear load transmission
and cannot resolve three-dimensional tissue deformation patterns. Simulation of
highly deformable tissues such as the meniscus in the knee remains challenging
when using discretized multibody representations (Guess et al. 2010). Novel algo-
rithms are beginning to bridge the gap between multibody and finite element models
(Lloyd et al. 2012), such that it may soon be possible to embed more complex,
volumetric soft tissue models within the concurrent simulation framework.
There remain numerous opportunities to expand the use of musculoskeletal
simulations to address clinical questions. Current musculoskeletal software packages
allow one to generate simulations that closely replication subject-specific measure-
ments of either normal or pathological movement. However, most simulations still
rely on generic musculoskeletal geometries, due in part to the arduous task of
generating subject-specific models from medical images. Emerging advances in
segmentation algorithms and statistical shape modeling may enable the customiza-
tion of model geometries in an efficient manner. However, quantification of subject-
specific soft tissue constitutive properties and muscle model parameters remains
troublesome and a source of uncertainty in many models. To address this challenge,
enhanced use of probabilistic simulation techniques is critically needed. In our own
lab, we leverage high throughput computing resources to regularly perform Monte
Carlo-type simulations to assess uncertainty propagation (Thelen and Smith 2016).
We find these probabilistic analyses critical to understand the robustness of model
predictions, to explore variations in muscle coordination, and to assess sensitivities
in light of uncertainty in tissue material properties and geometry.
When using musculoskeletal simulation tools, it is always important to recognize
that models are approximations designed to address questions that are difficult or
impossible to answer experimentally (Miyazaki et al. 2002; Arnold et al. 2006;
Hamner et al. 2010; Lenhart et al. 2016; Smith et al. 2016a). Hence, a common use of
musculoskeletal models is to estimate unmeasurable in vivo tissue loads from
measurable kinematics and external forces. Indeed, tissue load estimates have
proven extremely useful for assessing injury thresholds and designing orthopedic
treatment procedures. However, the ultimate goal of modeling should be prediction,
and it is exciting to see models used to assess sensitivities and predict the ramifica-
tions of injuries, surgical procedures, and rehabilitation on musculoskeletal function.
We are hopeful that concurrent simulation approaches can be coupled with advances
in subject-specific modeling, constitutive models, high performance computing and
computational algorithms to accurately simulate soft tissue behavior and to predict
changes in function that can arise from injury, disease and treatment.
References
Adouni M, Dhaher YY (2016) A multi-scale elasto-plastic model of articular cartilage. J Biomech
49:2891–2898. doi:10.1016/j.jbiomech.2016.06.031
Adouni M, Shirazi-Adl A (2014) Partitioning of knee joint internal forces in gait is dictated by the
knee adduction angle and not by the knee adduction moment. J Biomech 47:1696–1703.
doi:10.1016/j.jbiomech.2014.02.028
Andersen MS, Damsgaard M, Rasmussen J (2011) Force-dependent kinematics: a new analysis
method for non-conforming joints. 13th Biennial International Symposium on Computer Sim-
ulation in Biomechanics, Leuven, 2–3
Anderson FC, Pandy MG (2001) Dynamic optimization of human walking. J Biomech Eng
123:381–390
Anderson FC, Pandy MG (1999) A dynamic optimization solution for vertical jumping in three
dimensions. Comput Biomech Biomed Engin 2:201–231
Anderson IA, MacDiarmid AA, Harris ML et al (2003) A novel method for measuring medial
compartment pressures within the knee joint in-vivo. J Biomech 36:1391–1395. doi:10.1016/
S0021-9290(03)00158-1
Andriacchi TP, Koo S, Scanlan SF (2009) Gait mechanics influence healthy cartilage morphology
and osteoarthritis of the knee. J Bone Joint Surg Am 91(Suppl 1):95–101. doi:10.2106/JBJS.
H.01408
Arnold AS, Liu MQ, Schwartz MH et al (2006) The role of estimating muscle-tendon lengths and
velocities of the hamstrings in the evaluation and treatment of crouch gait. Gait Posture
23:273–281. doi:10.1016/j.gaitpost.2005.03.003
Arnold EM, Ward SR, Lieber RL, Delp SL (2010) A model of the lower limb for analysis of human
movement. Ann Biomed Eng 38:269–279. doi:10.1007/s10439-009-9852-5
Beard DJ, Soundarapandian RS, O’Connor JJ, Dodd CAF (1996) Gait and electromyographic
analysis of anterior cruciate ligament deficient subjects. Gait Posture 4:83–88. doi:10.1016/
0966-6362(95)01036-X
Bei Y, Fregly BJ (2004) Multibody dynamic simulation of knee contact mechanics. Med Eng Phys
26:777–789
Benoit DL, Ramsey DK, Lamontagne M et al (2006) Effect of skin movement artifact on knee
kinematics during gait and cutting motions measured in vivo. Gait Posture 24:152–164.
doi:10.1016/j.gaitpost.2005.04.012
Bergmann G, Graichen F, Rohlmann A, et al (2007) Loads acting on orthopaedic implants.
Measurements and practical applications. Orthopade 36:195–196, 198–200, 202–204.
doi: 10.1007/s00132-007-1055-x
Bergmann G, Graichen F, Rohlmann A (1993) Hip joint loading during walking and running ,
measured in two patients Comparison of intestinal absorption of cholesterol with different plant
sterols in man Cite Discreet indiscretions: The social organization of gossip Quantized periodic
orbits in lar. J Biomech 26:969–990
Bertram JEA (2005) Constrained optimization in human walking: cost minimization and gait
plasticity. J Exp Biol 208:979–991. doi:10.1242/jeb.01498
Besier TF, Gold GE, Beaupré GS, Delp SL (2005) A modeling framework to estimate
patellofemoral joint cartilage stress in vivo. Med Sci Sports Exerc 37:1924–1930.
doi:10.1249/01.mss.0000176686.18683.64
Beynnon BD, Johnson RJ, Fleming BC et al (1997) The strain behavior of the anterior cruciate
ligament during squatting and active flexion-extension. A comparison of an open and a closed
kinetic chain exercise. Am J Sports Med 25:823–829. doi:10.1177/036354659702500616
Blankevoort L, Huiskes R (1991) Ligament-bone interaction in a three-dimensional model of the
knee. J Biomech Eng 113:263–269. doi:10.1115/1.2894883
Blankevoort L, Kuiper JH, Huiskes R, Grootenboer HJ (1991) Articular contact in a three-
dimensional model of the knee. J Biomech 24:1019–1031. doi:10.1016/0021-9290(91)90019-J
Blemker SS, Asakawa DS, Gold GE, Delp SL (2007) Image-based musculoskeletal modeling:
applications, advances, and future opportunities. J Magn Reson Imaging 25:441–451.
doi:10.1002/jmri.20805
Brand RA, Crowninshield RD, Wittstock CE et al (1982) A model of lower extremity muscular
anatomy. J Biomech Eng 104:304–310
Brandon SCE, Miller RH, Thelen DG, Deluzio KJ (2014) Selective lateral muscle activation in
moderate medial knee osteoarthritis subjects does not unload medial knee condyle. J Biomech
47:1409–1415. doi:10.1016/j.jbiomech.2014.01.038
Buchanan TS, Lloyd DG, Manal K, Besier TF (2004) Neuromusculoskeletal modeling: estimation
of muscle forces and joint moments and movements from measurements of neural command. J
Appl Biomech 20:367–395. doi:10.1016/j.bbi.2008.05.010
Cabaud HE (1983) Biomechanics of the anterior cruciate ligament. Clin Orthop Relat Res 26–31.
doi: 10.1097/00003086-198301000-00007
stereophotogrammetry. Part 1: theoretical background. Gait Posture 21:186–196. doi:10.1016/
j.gaitpost.2004.01.010
Carbone V, Fluit R, Pellikaan P et al (2015) TLEM 2.0 – a comprehensive musculoskeletal
geometry dataset for subject-specific modeling of lower extremity. J Biomech 48:734–741.
doi:10.1016/j.jbiomech.2014.12.034
Caruntu DI, Hefzy MS (2004) 3-D anatomically based dynamic modeling of the human knee to
include Tibio-femoral and Patello-femoral joints. J Biomech Eng 126:44. doi:10.1115/1.1644565
Chen Z, Zhang Z, Wang L, Li D, Zhang Y, Jin Z (2016) Evaluation of a subject-specific
musculoskeletal modeling framework for load prediction in total knee arthroplasty. Med Eng
Phys 38(8):708–716. https://doi.org/10.1016/j.medengphy.2016.04.010
Chernak L, Thelen DG, Slane LC, Thelen DG (2014) Non-uniform displacements within the
Achilles tendon observed during passive and eccentric loading. J Biomech 47:2831–2835.
doi:10.1016/j.jbiomech.2014.07.032
Chiari L, Della Croce U, Leardini A, Cappozzo A (2005) Human movement analysis using
stereophotogrammetry. Part 2: instrumental errors. Gait Posture 21:197–211. doi:10.1016/j.
Christophy M, Faruk Senan NA, Lotz JC, O’Reilly OM (2012) A musculoskeletal model for the
lumbar spine. Biomech Model Mechanobiol 11:19–34. doi:10.1007/s10237-011-0290-6
Crowninshield RD, Brand RA (1981) A physiologically based criterion of muscle force prediction
in locomotion. J Biomech 14:793–801
D’Lima DD, Fregly BJ, Colwell CW Jr, Colwell CW (2013) Implantable sensor technology:
measuring bone and joint biomechanics of daily life in vivo. Arthritis Res Ther 15:203.
doi:10.1186/ar4138
D’Lima DD, Patil S, Steklov N et al (2006) Tibial forces measured in vivo after total knee
arthroplasty. J Arthroplast 21:255–262. doi:10.1016/j.arth.2005.07.011
Davy DT, Audu ML (1987) A dynamic optimization technique for predicting muscle forces in the
swing phase of gait. J Biomech 20:187–201. doi:10.1016/0021-9290(87)90310-1
Delp SLL, Loan JPP, Hoy MGG et al (1990) An interactive graphics-based model of the lower
extremity to study orthopedic surgical procedures. IEEE Trans Biomed Eng 37:757–767.
doi:10.1109/10.102791
Demers MS, Pal S, Delp SL (2014) Changes in tibiofemoral forces due to variations in muscle
activity during walking. J Orthop Res 32:769–776. doi:10.1002/jor.22601
Dhaher YY, Kwon T-H, Barry M (2010) The effect of connective tissue material uncertainties on
knee joint mechanics under isolated loading conditions. J Biomech 43:3118–3125. doi:10.1016/
j.jbiomech.2010.08.005
Dhaher YY, Salehghaffari S, Adouni M (2016) Anterior laxity, graft-tunnel interaction and surgical
design variations during anterior cruciate ligament reconstruction: a probabilistic simulation of
the surgery. J Biomech 49:3009–3016. doi:10.1016/j.jbiomech.2016.07.019
Ellis BJ, Drury NJ, Moore SM et al (2010) Finite element modelling of the glenohumeral capsule
can help assess the tested region during a clinical exam. Comput Methods Biomech Biomed
Engin 13:413–418. doi:10.1080/10255840903317378
English TA, Kilvington M (1979) In vivo records of hip loads using a femoral implant with
telemetric output (a prelimary report). J Biomed Eng 1:111–115. doi:10.1016/0141-5425(79)
90066-9
Erdemir A (2016) Open knee: open source modeling & simulation to enable scientific discovery and
clinical care in knee biomechanics. J Knee Surg 19:107–116. doi:10.1016/j.jocn.2013.06.015.
C-reactive
Erdemir A (2013) Open knee: a pathway to community driven modeling and simulation in joint
biomechanics. J Med Device 7:40910. doi:10.1115/1.4025767
Erdemir A, McLean S, Herzog W, van den Bogert AJ (2007) Model-based estimation of muscle
forces exerted during movements. Clin Biomech 22:131–154
Eschweiler J, Stromps J-P, Fischer M et al (2016) Development of a biomechanical model of the
wrist joint for patient-specific model guided surgical therapy planning: part 1. Proc Inst Mech
Eng Part H J Eng Med 230:310–325. doi:10.1177/0954411916632791
Eskinazi I, Fregly BJ (2016) An open-source toolbox for surrogate modeling of joint contact
mechanics. IEEE Trans Bio-Med Eng 63(2):269–277. https://doi.org/10.1109/
TBME.2015.2455510
Essinger JR, Leyvraz PF, Heegard JH, Robertson DD (1989) A mathematical model for the
evaluation of the behavior during flexion of condylar-type knee prostheses. J Biomech
22:1229–12241
Farrokhi S, Voycheck CA, Klatt BA et al (2014) Altered tibiofemoral joint contact mechanics and
kinematics in patients with knee osteoarthritis and episodic complaints of joint instability.
Clin Biomech (Bristol, Avon) 29:629–635. doi:10.1016/j.clinbiomech.2014.04.014
Ferguson SJ, Bryant JT, Ganz R, Ito K (2000) The influence of the acetabular labrum on hip joint
cartilage consolidation: a poroelastic finite element model. J Biomech 33:953–960. doi:10.1016/
S0021-9290(00)00042-7
Finni T, Komi PV, Lepola V (2000) In vivo human triceps surae and quadriceps femoris muscle
function in a squat jump and counter. Eur J Appl Physiol 83:416–426. doi:10.1007/
s004210000289
Fleming BC, Beynnon BD (2004) In vivo measurement of ligament/tendon strains and forces: a
review. Ann Biomed Eng 32:318–328
Fukashiro S, Komi PV, Järvinen M, Miyashita M (1995) In vivo achilles tendon loading’ during
jumping in humans. Eur J Appl Physiol Occup Physiol 71:453–458. doi:10.1007/BF00635880
Gerber H, Foresti M, Zihlmann M, Stussi E (2007) Method to simultaneously measure kinetic and
3D kinematic data during normal level walking using KISTLER force plates, VICON system
and video-fluoroscopy. J Biomech 40:S405. doi:10.1016/S0021-9290(07)70400-1
Gerus P, Sartori M, Besier TF et al (2013) Subject-specific knee joint geometry improves pre-
dictions of medial tibiofemoral contact forces. J Biomech 46:2778–2786. doi:10.1016/j.
Grundy M, Tosh PA, RD ML, Smidt L (1975) An investigation of the centres of pressure under the
foot while walking. J Bone Joint Surg Br 57:98–103
Guess TM (2012) Forward dynamics simulation using a natural knee with menisci in the multibody
framework. Multibody Syst Dyn 28:37–53. doi:10.1007/s11044-011-9293-4
Guess TM, Liu H, Bhashyam S, Thiagarajan G (2013) A multibody knee model with discrete
cartilage prediction of tibio-femoral contact mechanics. Comput Methods Biomech Biomed
Engin 16:256–270. doi:10.1080/10255842.2011.617004
Guess TM, Stylianou AP, Kia M (2014) Concurrent prediction of muscle and tibiofemoral contact
forces during treadmill gait. J Biomech Eng 136:21032. doi:10.1115/1.4026359
Guess TM, Thiagarajan G, Kia M, Mishra M (2010) A subject specific multibody model of the knee
with menisci. Med Eng Phys 32:505–515. doi:10.1016/j.medengphy.2010.02.020
Halonen KS, Mononen ME, Jurvelin JS, Töyräs J, Klodowski A, Kulmala J-P, Korhonen RK
(2016) Importance of patella, quadriceps forces, and depthwise cartilage structure on knee joint
motion and cartilage response during gait. J Biomech Eng 138(7):71002. https://doi.org/
10.1115/1.4033516
Halloran JP, Easley SK, Petrella AJ, Rullkoetter PJ (2005) Comparison of deformable and elastic
foundation finite element simulations for predicting knee replacement mechanics. J Biomech
Eng 127:813–818. doi:10.1115/1.1992522
Halloran JP, Erdemir A, van den Bogert AJ (2009) Adaptive surrogate modeling for efficient
coupling of musculoskeletal control and tissue deformation models. J Biomech Eng
131:11014. doi:10.1115/1.3005333
Halloran JP, Sibole S, van Donkelaar CC et al (2012) Multiscale mechanics of articular cartilage:
potentials and challenges of coupling musculoskeletal, joint, and microscale computational
models. Ann Biomed Eng 40:2456–2474. doi:10.1007/s10439-012-0598-0
Halonen KS, Mononen ME, Jurvelin JS et al (2013) Importance of depth-wise distribution of
collagen and proteoglycans in articular cartilage-a 3D finite element study of stresses and strains
in human knee joint. J Biomech 46:1184–1192. doi:10.1016/j.jbiomech.2012.12.025
Hamner SR, Seth A, Delp SL (2010) Muscle contributions to propulsion and support during
running. J Biomech 43:2709–2716
Happee R, Van der Helm FCT (1995) The control of shoulder muscles during goal directed
movements, an inverse dynamic analysis. J Biomech 28:1179–1191. doi:10.1016/0021-9290
(94)00181-3
Hardt DE (1978) Determining muscle forces in the leg during normal human walking - an
application and evaluation of optimization methods. J Biomech Eng 100:72–78
Hast MW, Piazza SJ (2013) Dual-joint modeling for estimation of total knee replacement contact
forces during locomotion. J Biomech Eng 135:21013. doi:10.1115/1.4023320
Hausselle J, Assi A, El Helou A et al (2012) Subject-specific musculoskeletal model of the lower
limb in a lying and standing position. Comput Methods Biomech Biomed Engin 17:37–41.
doi:10.1080/10255842.2012.693173
Heller MO, Bergmann G, Deuretzbacher G et al (2001) Musculo-skeletal loading conditions at the
hip during walking and stair climbing. J Biomech 34:883–893. doi:10.1016/S0021-9290(01)
00039-2
Herzog W, Longino D (2007) The role of muscles in joint degeneration and osteoarthritis. J
Biomech 40(Suppl 1):S54–S63. doi:10.1016/j.jbiomech.2007.03.001
Hicks JL, Uchida TK, Seth A, Rajagopal A, Delp S (2015) Is my model good enough? Best
practices for verification and validation of musculoskeletal models and simulations of human
movement. J Biomech Eng 137(February):20905. https://doi.org/10.1115/1.4029304
Hill AV (1938) The heat of shortening and the dynamic constants of muscle. Proc R Soc B Biol Sci
126:136–195. doi:10.1098/rspb.1938.0050
Hodge WA, Fijan RS, Carlson KL et al (1986) Contact pressures in the human hip joint measured
in vivo. Proc Natl Acad Sci USA 83:2879–2883. doi:10.1073/pnas.83.9.2879
Hollars MG, Rosenthal DE, Sherman MA (1994) SD/FAST user’s manual, Ver. B.2. Symbolic
Dynamics, Mountain View
Houck J, Yack HJ (2003) Associations of knee angles, moments and function among subjects that
are healthy and anterior cruciate ligament deficient (ACLD) during straight ahead and crossover
cutting activities. Gait Posture 18:126–138. doi:10.1016/S0966-6362(02)00188-1
Ignasiak D, Dendorfer S, Ferguson SJ (2016) Thoracolumbar spine model with articulated ribcage
for the prediction of dynamic spinal loading. J Biomech 49:959–966. doi:10.1016/j.
Isaac DL, Beard DJ, Price AJ et al (2005) In-vivo sagittal plane knee kinematics: ACL intact,
deficient and reconstructed knees. Knee 12:25–31. doi:10.1016/j.knee.2004.01.002
Jinha A, Ait-Haddou R, Herzog W (2006) Predictions of co-contraction depend critically on
degrees-of-freedom in the musculoskeletal model. J Biomech 39:1145–1152. doi:10.1016/j.
Kainz H, Modenese L, Lloyd DG et al (2016) Joint kinematic calculation based on clinical direct
kinematic versus inverse kinematic gait models. J Biomech 49:1658–1669. doi:10.1016/j.
Kaiser J, Bradford R, Johnson K et al (2013) Measurement of tibiofemoral kinematics using highly
accelerated 3D radial sampling. Magn Reson Med 69:1310–1316. doi:10.1002/mrm.24362
Katz JW, Fingeroth RJ (1986) The diagnostic accuracy of ruptures of the anterior cruciate ligament
comparing the Lachman test, the anterior drawer sign, and the pivot shift test in acute and
chronic knee injuries. Am J Sports Med 14:88–91. doi:10.1177/036354658601400115
Kazemi M, Dabiri Y, Li LP (2013) Recent advances in computational mechanics of the human knee
joint. Comput Math Methods Med 2013:718423. doi:10.1155/2013/718423
Kim S (1998a) Muscle-ligament interactions at the human knee (I) – maximum isometric leg
exercises. KSME Int J 12:1079–1089
Kim S (1998b) Three-dimensional dynamic model of the knee. KSME Int J 12:1041–1063
Klein Horsman MD (2007) The Twente lower extremity model: consistent dynamic simulation of
the human locomotor apparatus, University of Twente
Kłodowski A, Mononen ME, Kulmala JP et al (2016) Merge of motion analysis, multibody
dynamics and finite element method for the subject-specific analysis of cartilage loading
patterns during gait: differences between rotation and moment-driven models of human knee
joint. Multibody Syst Dyn 37:271–290. doi:10.1007/s11044-015-9470-y
Koo S, Rylander JH, Andriacchi TP (2011) Knee joint kinematics during walking influences the
spatial cartilage thickness distribution in the knee. J Biomech 44:1405–1409. doi:10.1016/j.
Koolstra JH, Van Eijden TMGJ (2005) Combined finite-element and rigid-body analysis of human
jaw joint dynamics. J Biomech 38:2431–2439. doi:10.1016/j.jbiomech.2004.10.014
Kutzner I, Heinlein B, Graichen F et al (2010) Loading of the knee joint during activities of daily
living measured in vivo in five subjects. J Biomech 43:2164–2173
Leardini A, Chiari L, Della CU, Cappozzo A (2005) Human movement analysis using stereo-
photogrammetry. Part 3. Soft tissue artifact assessment and compensation. Gait Posture 21:212–225
Lee L, Umberger BR (2016) Generating optimal control simulations of musculoskeletal movement
using OpenSim and MATLAB. Peer J 4:e1638. doi:10.7717/peerj.1638
Lenhart RL, Brandon SCE, Smith CR et al (2016) Influence of patellar position on the knee extensor
mechanism in normal and crouched walking. J Biomech. doi:10.1016/j.jbiomech.2016.11.052
Lenhart RL, Kaiser J, Smith CR, Thelen DG (2015a) Prediction and validation of load-dependent
behavior of the tibiofemoral and patellofemoral joints during movement. Ann Biomed Eng
43:2675–2685. doi:10.1007/s10439-015-1326-3
Lenhart RL, Smith CR, Vignos MF et al (2015b) Influence of step rate and quadriceps load
distribution on patellofemoral cartilage contact pressures during running. J Biomech
48:2871–2878. doi:10.1016/j.jbiomech.2015.04.036
Li G, Defrate LE, Rubash HE, Gill TJ (2005) In vivo kinematics of the ACL during weight-bearing
knee flexion. J Orthop Res 23:340–344. doi:10.1016/j.orthres.2004.08.006
Li K, Zheng L, Tashman S, Zhang X (2012) The inaccuracy of surface-measured model-derived
tibiofemoral kinematics. J Biomech 45:2719–2723. doi:10.1016/j.jbiomech.2012.08.007
Lin Y-C, Farr J, Carter K, Fregly BJ (2006) Response surface optimization for joint contact model
evaluation. J Appl Biomech 22:120–130
Lin Y-C, Haftka RT, Queipo NV, Fregly BJ (2010a) Surrogate articular contact models for
computationally efficient multibody dynamic simulations. Med Eng Phys 32:584–594.
doi:10.1016/j.medengphy.2010.02.008
Lin YC, Walter JP, Banks SA et al (2010b) Simultaneous prediction of muscle and contact forces in
the knee during gait. J Biomech 43:945–952. doi:10.1016/j.jbiomech.2009.10.048
Lloyd DG, Besier TF (2003) An EMG-driven musculoskeletal model to estimate muscle forces and
knee joint moments in vivo. J Biomech 36:765–776
Lloyd JE, Stavness I, Fels S (2012) ArtiSynth: a fast interactive biomechanical modeling toolkit
combining multibody and finite element simulation. In: Soft Tissue Biomechanical Modeling
for Computer Assisted Surgery, Springer-Verlag Berlin Heidelberg. pp 355–394. doi:10.1007/

8415_2012_126
Logan MC (2004) Tibiofemoral kinematics following successful anterior cruciate ligament recon-
struction using dynamic multiple resonance imaging. Am J Sports Med 32:984–992.
doi:10.1177/0363546503261702
Lu T-W, O’Connor JJ (1999) Bone position estimation from skin marker co-ordinates using
global optimisation with joint constraints. J Biomech 32:129–134. doi:10.1016/S0021-9290
(98)00158-4
Marouane H, Shirazi-Adl A, Adouni M (2016) 3D active-passive response of human knee joint in
gait is markedly altered when simulated as a planar 2D joint. Biomech Model Mechanobiol.
doi:10.1007/s10237-016-0846-6
Marra MA, Vanheule V, Fluit R, Koopman BH, Rasmussen J, Verdonschot N, Andersen MS (2015)
A subject-specific musculoskeletal modeling framework to predict in vivo mechanics of total
knee arthroplasty. J Biomech Eng 137(2):20904. https://doi.org/10.1115/1.4029258
Marra MA, Vanheule V, Rasmussen J et al (2015) A subject-specific musculoskeletal modeling
framework to predict in vivo mechanics of Total knee arthroplasty. J Biomech Eng 137:20904.
doi:10.1115/1.4029258
Millard M, Uchida T, Seth A, Delp SL (2013) Flexing computational muscle: modeling and
simulation of musculotendon dynamics. J Biomech Eng 135:21005. doi:10.1115/1.4023390
Miranda DL, Fadale PD, Hulstyn MJ et al (2013) Knee biomechanics during a jump-cut maneuver:
effects of sex and ACL surgery. Med Sci Sports Exerc 45:942–951. doi:10.1249/
MSS.0b013e31827bf0e4
Miyazaki T, Wada M, Kawahara H et al (2002) Dynamic load at baseline can predict radiographic
disease progression in medial compartment knee osteoarthritis. Ann Rheum Dis 61:617–622
Modenese L, Phillips ATM, Bull AMJ (2010) A state of the art 3D model of the lower limb:
application to muscle force estimation and validation. (WCB 2010) Aug 1–6, 2010. pp 981–984
Morrison JB (1970) The mechanics of the knee joint in relation to normal walking. J Biomech
3:51–61. doi:10.1016/0021-9290(70)90050-3
Nissi MJ, Rieppo J, Töyräs J et al (2007) Estimation of mechanical properties of articular cartilage
with MRI - dGEMRIC, T2 and T1 imaging in different species with variable stages of
maturation. Osteoarthr Cartil 15:1141–1148. doi:10.1016/j.joca.2007.03.018
Ota K, Shinya M, Kudo K (2016) Sub-optimality in motor planning is retained throughout 9 days
practice of 2250 trials. Sci Rep 6:37181. doi:10.1038/srep37181
Pandy MG, Sasaki K, Kim S (1998a) A three-dimensional musculoskeletal model of the human
knee joint. Part 1: Theoretical construct. Comput Biomech Biomed Engin 1:87–108.
doi:10.1080/01495739708936697
Pandy MG, Sasaki K, Taylor P et al (1998b) A three-dimensional musculoskeletal model of the
human knee joint. Part 2: Analysis of ligament function. Comput Methods Biomech Biomed
Engin 1:265–283. doi:10.1080/01495739808936707
Papaioannou G, Demetropoulos CK, King YH (2010) Predicting the effects of knee focal articular
surface injury with a patient-specific finite element model. Knee 17:61–68. doi:10.1016/j.
knee.2009.05.001
Pauwels F (1935) Der schenkelhalsbruch em mechanisches problem: Grundlagen des Heilungs-
vorganges Prognose und kausale Therapie. Stuttgart Ferdinand Enke Verlag. doi:10.1590/1413
Pedotti A, Krishnan VV, Stark L (1978) Optimization of muscle-force sequencing in human
locomotion. Math Biosci 38:57–76. doi:10.1016/0025-5564(78)90018-4
Peña E, Martínez MA, Calvo B et al (2005) A finite element simulation of the effect of graft stiffness
and graft tensioning in ACL reconstruction. Clin Biomech 20:636–644. doi:10.1016/j.
Piazza SJ, Delp SL (2001) Three-dimensional dynamic simulation of total knee replacement motion
during a step-up task. J Biomech Eng 123:599–606. doi:10.1115/1.1406950
Rajagopal A, Dembia CL, DeMers MS et al (2016) Full-body musculoskeletal model for muscle-
driven simulation of human gait. IEEE Trans Biomed Eng 63:2068–2079. doi:10.1109/
TBME.2016.2586891
Ravary B, Pourcelot P, Bortolussi C et al (2004) Strain and force transducers used in human and
veterinary tendon and ligament biomechanical studies. Clin Biomech 19:433–447. doi:10.1016/
j.clinbiomech.2004.01.008
Rikli DA, Honigmann P, Babst R et al (2007) Intra-articular pressure measurement in the radio-
ulnocarpal joint using a novel sensor: in vitro and in vivo results. J Hand Surg Am 32:67–75.
doi:10.1016/j.jhsa.2006.10.007
Roberts CS, Cumming JF, Grood ES, Noyes FR (1994) In vivo measurement of human anterior
cruciate ligament forces during knee extension exercises. In: Transactions of the 40th Ortho-
paedic Research Society, New Orleans, pp 15–84
Rydell NW (1966) Forces acting on the femoral head-prosthesis: a study on strain gauge supplied
prostheses in living persons. Acta Orthop Scand 37:1–132. doi:10.3109/ort.1966.37.
suppl-88.01
Sanders TL, Maradit Kremers H, Bryan AJ et al (2016) Incidence of anterior cruciate ligament tears
and reconstruction: a 21-year population-based study. Am J Sports Med 44:1502–1507.
doi:10.1177/0363546516629944
Sartori M, Gizzi L, Lloyd DG, Farina D (2013) A musculoskeletal model of human locomotion
driven by a low dimensional set of impulsive excitation primitives. Front Comput Neurosci
7:79. doi:10.3389/fncom.2013.00079
Scheys L, Jonkers I, Schutyser F et al (2005) Image based methods to generate subject-specific
musculoskeletal models for gait analysis. Int Congr Ser 1281:62–67. doi:10.1016/j.
ics.2005.03.076
Scheys L, Spaepen A, Suetens P, Jonkers I (2008) Calculated moment-arm and muscle-tendon
lengths during gait differ substantially using MR based versus rescaled generic lower-limb
musculoskeletal models. Gait Posture 28:640–648. doi:10.1016/j.gaitpost.2008.04.010
Schipplein OD, Andriacchi TP (1991) Interaction between active and passive knee stabilizers
during level walking. J Orthop Res 9:113–119. doi:10.1002/jor.1100090114
Seireg A, Arvikar RJ (1975) The prediction of musclular load sharing and joint forces in lower
extremities during walking. J Biomech 8:89–102
Seth A, Matias R, Veloso AP, Delp SL (2016) A biomechanical model of the Scapulothoracic joint
to accurately capture scapular kinematics during shoulder movements. PLoS One 11:e0141028.
Seth A, Pandy MG (2007) A neuromusculoskeletal tracking method for estimating indivi-
dual muscle forces in human movement. J Biomech 40:356–366. doi:10.1016/j.
Shelburne KB, Pandy MG, Anderson FC, Torry MR (2004) Pattern of anterior cruciate ligament
force in normal walking. J Biomech 37:797–805
Shelburne KB, Torry MR, Pandy MG (2006) Contributions of muscles, ligaments, and the ground-
reaction force to tibiofemoral joint loading during normal gait. J Orthop Res 24:1983–1990
Shelburne KB, Torry MR, Pandy MG (2005) Muscle, ligament, and joint-contact forces at the
knee during walking. Med Sci Sports Exerc 37:1948–1956. doi:10.1249/01.
mss.0000180404.86078.ff
Sherman MA, Seth A, Delp SL (2011) SIMBODY: multibody dynamics for biomedical research.
Procedia IUTAM 2:241–261
Shin CS, Chaudhari AM, Andriacchi TP (2009) The effect of isolated valgus moments on ACL
strain during single-leg landing: a simulation study. J Biomech 42:280–285. doi:10.1016/j.
Shin CS, Chaudhari AM, Andriacchi TP (2011) Valgus plus internal rotation moments increase
anterior cruciate ligament strain more than either alone. Med Sci Sports Exerc 43:1484–1491.
doi:10.1249/MSS.0b013e31820f8395
Shin CS, Chaudhari AM, Andriacchi TP (2007) The influence of deceleration forces on ACL strain
during single-leg landing: a simulation study. J Biomech 40:1145–1152. doi:10.1016/j.
Sins L, Tétreault P, Hagemeister N, Nuño N (2015) Adaptation of the AnyBody™ musculoskeletal
shoulder model to the nonconforming Total shoulder arthroplasty context. J Biomech Eng
137:101006. doi:10.1115/1.4031330
Smith CR, Lenhart RL, Kaiser J et al (2016a) Influence of ligament properties on tibiofemoral
mechanics in walking. J Knee Surg 29:99–106. doi:10.1055/s-0035-1558858
Smith CR, Vignos MF, Lenhart RL et al (2016b) The influence of component alignment and
ligament properties on tibiofemoral contact forces in Total knee replacement. J Biomech Eng
138:21017. doi:10.1115/1.4032464
Smith CR, Won Choi K, Negrut D, Thelen DG (2016c) Efficient computation of cartilage contact
pressures within dynamic simulations of movement. Comput Methods Biomech Biomed Eng
Imaging Vis 1–8. doi: 10.1080/21681163.2016.1172346
Stylianou AP, Guess TM, Kia M (2013) Multibody muscle driven model of an instrumented
prosthetic knee during squat and toe rise motions. J Biomech Eng 135:41008. doi:10.1115/
1.4023982
Tanska P, Mononen ME, Korhonen RK (2015) A multi-scale finite element model for investigation
of chondrocyte mechanics in normal and medial meniscectomy human knee joint during
walking. J Biomech 48:1397–1406. doi:10.1016/j.jbiomech.2015.02.043
Thelen DG (2003) Adjustment of muscle mechanics model parameters to simulate dynamic
contractions in older adults. J Biomech Eng 125:70. doi:10.1115/1.1531112
Thelen DG, Anderson FC (2006) Using computed muscle control to generate forward dynamic
simulations of human walking from experimental data. J Biomech 39:1107–1115
Thelen DG, Anderson FC, Delp SL (2003) Generating dynamic simulations of movement using
computed muscle control. J Biomech 36:321–328
Thelen DG, Smith CR (2016) Webinar: enabling stochastic simulations of movement with high
throughput computing. Natl Cent Simul Rehabil Res. http://opensim.stanford.edu/support/
event_details.html?id=169. Accessed 13 Dec 2016
Thelen DG, Won Choi K, Schmitz AM (2014) Co-simulation of neuromuscular dynamics and knee
mechanics during human walking. J Biomech Eng 136:21033. doi:10.1115/1.4026358
Todorov E, Jordan MI (2002) Optimal feedback control as a theory of motor coordination.
Nat Neurosci 5:1226–1235. doi:10.1038/nn963
Torry MR, Decker MJ, Ellis HB et al (2004) Mechanisms of compensating for anterior cruciate
ligament deficiency during gait. Med Sci Sports Exerc 36:1403–1412. doi:10.1249/01.
MSS.0000135797.09291.71
Valente G, Pitto L, Testi D et al (2014) Are subject-specific musculoskeletal models robust to the
uncertainties in parameter identification? PLoS One 9:e112625. doi:10.1371/journal.
pone.0112625
Van der Helm FC, Veeger HE, Pronk GM et al (1992) Geometry parameters for musculoskeletal
modelling of the shoulder system. J Biomech 25:129–144
Walker PS, Rovick JS, Robertson DD (1988) The effects of knee brace hinge design and placement
on joint mechanics. J Biomech 21:965–974. doi:10.1016/0021-9290(88)90135-2
Winby CR, Lloyd DG, Besier TF, Kirk TB (2009) Muscle and external load contribution to knee
joint contact loads during normal gait. J Biomech 42:2294–2300
Winter DA, Scott SH (1991) Technique for interpretation of electromyography for concentric and
eccentric contractions in gait. J Electromyogr Kinesiol 1:263–269
Wong C, Zhang ZQ, Lo B, Yang GZ (2015) Wearable sensing for solid biomechanics: a review.
IEEE Sensors J 15:2747–2760. doi:10.1109/JSEN.2015.2393883
Wu G, Cavanagh PR (1995) ISB recommendations for standardization in the reporting of kinematic
data. J Biomech 28:1257–1261
Wu G, Siegler S, Allard P et al (2002) ISB recommendation on definitions of joint coordinate
system of various joints for the reporting of human joint motion–part I: ankle, hip, and spine.
International Society of Biomechanics. J Biomech 35:543–548. doi:10.1016/S0021-9290

(01)00222-6
Yamaguchi GT, Zajac FE (1989) A planar model of the knee joint to characterize the knee extensor
mechanism. J Biomech 22:1–10. doi:10.1016/0021-9290(89)90179-6
You BM, Siy P, Anderst W, Tashman S (2001) In vivo measurement of 3-D skeletal kinematics from
sequences of biplane radiographs: application to knee kinematics. IEEE Trans Med Imaging
20:514–525. doi:10.1109/42.929617
Zhang X, Chen Z, Wang L et al (2015) Prediction of hip joint load and translation using
musculoskeletal modelling with force-dependent kinematics and experimental validation.
Proc Inst Mech Eng H 229:477–490
Zheng M, Zou Z, Bartolo PJDS et al (2016) Finite element models of the human shoulder complex:
a review of their clinical implications and modelling techniques. Int J Numer Method
Biomed Eng 30:659–680. doi:10.1002/cnm.2777
Induced Acceleration and Power Analyses
of Human Motion
Anne K. Silverman
Abstract
Induced acceleration and power analyses are tools that are used to determine
principles of muscle coordination and the functional roles of muscles during
movement. This chapter describes how these analyses are performed using an
underlying musculoskeletal model and movement simulation. Induced accelera-
tion analyses use the dynamic equations of motion of the musculoskeletal system
to determine the effects of individual muscles, and other modeled actuators, on
the body’s movement, such as the acceleration of specific joints and/or the body’s
mass center. Induced power analyses build upon induced acceleration analyses to
evaluate mechanical power that is generated to, absorbed from, and/or transferred
between body segments and can reveal how muscles work together to achieve a
dynamic task. Examples of application of induced acceleration and power ana-
lyses to human performance and clinical questions are provided. In addition,
limitations of these analyses and potential impacts on the interpretation of the
results are discussed. Future directions include the use of induced acceleration
and power analyses with improved accuracy of musculoskeletal models and
computational approaches to distill large quantities of information into clinical
decision-making.
Keywords
Muscle function • Musculoskeletal model • Movement simulation • Muscle
coordination • Induced acceleration • Segment power
A.K. Silverman, Ph.D. (*)

Functional Biomechanics Laboratory, Department of Mechanical Engineering, Colorado School
of Mines, Golden, CO, USA
e-mail: asilverm@mines.edu

DOI 10.1007/978-3-319-30808-1_175-1
2 A.K. Silverman
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Formulation of Induced Acceleration and Power Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Underlying Musculoskeletal Model and Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Induced Acceleration Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Segment Power Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Joint Contact Force Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Interpretation of Induced Acceleration and Power Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Application to Pathological Movement Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Limitations of Induced Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Validation of Induced Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Introduction
The study of human motion is often directed at identifying principles of neuromus-

cular control that drive movement. Determining how muscles are coordinated to
produce observed kinematics and kinetics is paramount to understanding the
neuromusculoskeletal system. Identifying how movement is coordinated is critical
for various applications such as augmenting human performance, designing
improved assistive devices, and developing effective rehabilitative care for enhanced
mobility. Muscles are precisely orchestrated to perform multiple functional roles to
achieve a wide variety of movements in a smooth and efficient manner. However,
determining principles of movement coordination can be difficult because of the
complexity and redundancy of the neuromusculoskeletal system and because of
measurement limitations in humans. For example, direct measurement of muscle
forces cannot be completed noninvasively, and precisely perturbing the human
system is difficult or impossible to achieve experimentally. Musculoskeletal model-
ing and simulation approaches are often applied to provide insight into dynamic
movement tasks and build upon experimental measurements. Movement simulations
can reproduce movement patterns that were observed in a laboratory or generate
theoretical movement trajectories from a performance- or task-based cost function.
Analyses of muscle and joint mechanics, including induced acceleration and power
analyses, from musculoskeletal models and movement simulations can be used to
determine how muscles produce a coordinated movement pattern. Causal relation-
ships between muscle action and output movement can be thoroughly explored in a
computational environment. Such methods provide a wealth of information regard-
ing how muscles coordinate to achieve a variety of dynamic tasks and how muscle
functional roles may vary across environments, contexts, people, and neuromuscular
health.
Induced Acceleration and Power Analyses of Human Motion 3
State of the Art
Induced acceleration and induced power, or segment power, analyses are used to
identify the effects of individual forces on the resulting motion of a dynamic system.
In the application to human motion, these analyses are often used to derive the function
of individual muscles and devices in producing coordinated movement (Zajac et al.
2002, 2003). These analyses are powerful because often the functional roles of muscles
could not be determined by another method and can be used to identify potential
muscles to be targeted for invasive treatment (e.g., surgical lengthening, botulinum
toxin injection, tendon transfer) or training (e.g., rehabilitation, strengthening, gait
retraining). Identifying muscle functional roles can also be helpful for assistive device
development and design (e.g., prostheses, orthoses, exoskeletons), as muscular roles
in facilitating movement may be augmented or supplanted by these devices. An
underlying dynamic model of the musculoskeletal system is needed in order to
perform induced acceleration and segment power analyses. Musculoskeletal models
have advanced substantially in recent years to improve accuracy in representing the
musculoskeletal system. Advances in imaging techniques and computing power have
resulted in development of models of practically every part of the human musculo-
skeletal system including the lower limbs (Arnold et al. 2010), upper limbs (Saul
et al. 2015), lumbar spine (Christophy et al. 2012), thoracic spine (Bruno et al. 2015),
cervical spine (Suderman and Vasavada 2012; Cazzola et al. 2017), and jaw (Ackland
et al. 2015). Advancements in musculoskeletal models have included the addition of
soft tissue considerations and finite element models of joint contact (▶ Simulation of
Soft Tissue Loading from Observed Movement Dynamics).
Human movement simulations of musculoskeletal models are often the basis for
induced acceleration analyses. In a simulation, the muscle forces that are required to
drive a movement have been estimated through static or dynamic optimization
techniques (▶ Optimal Control Modeling of Human Movement). The movement
simulation may reproduce movement that has been observed in a biomechanical
laboratory setting, or it may represent a potential movement trajectory that achieves a
specified dynamic task (e.g., maximum height jumping). With estimates of forces,
induced acceleration and power analyses describe the role of each force in producing
the simulated movement. These analyses determine how muscles, the actuators of
our musculoskeletal system, accelerate and decelerate body segments, joints, and the
body center of mass. In addition, how mechanical power is distributed throughout
the body provides a description of how movement tasks are achieved by the
coordination of multiple body segments. These analyses have been used to deter-
mine how muscles are orchestrated to coordinate cycling (Fregly and Zajac 1996;
Neptune et al. 2000), walking (Neptune et al. 2001, 2004, 2008; Anderson and
Pandy 2003; Pandy et al. 2010), running (Debaere et al. 2015; Dorn et al. 2012a;
Hamner and Delp 2013; Sasaki and Neptune 2006), turning (Ventura et al. 2015),
pathological gait (Silverman and Neptune 2012; Steele et al. 2010; Peterson et al.
2010; Hall et al. 2011), stair walking (Lin et al. 2015), incline/decline walking
(Pickle et al. 2016), sit-to-stand (Caruthers et al. 2016), wheelchair push and
recovery (Rankin et al. 2011), and other tasks.
4 A.K. Silverman
Formulation of Induced Acceleration and Power Analyses
Underlying Musculoskeletal Model and Equations of Motion
Performing induced acceleration and segment power analyses requires formulating

the dynamic equations of motion of a mechanical system. In the study of human
motion, the dynamic equations of motion describe an underlying musculoskeletal
model (Yamaguchi 2006). Defining a musculoskeletal model that is appropriate to
the research question of interest is critical for obtaining useful information from
induced acceleration and segment power analyses. A musculoskeletal model
includes definitions of degrees of freedom between body segments, body segment
masses and inertial properties, musculotendon paths, musculotendon force-generating
properties, and contact between the body and its environment (▶ Three-Dimensional
Reconstruction of the Human Skeleton in Motion). Musculoskeletal models take
significant time and effort to develop, and generic models are typically based on
measurements from cadavers. Models can be made more specific to an individual
person through scaling of body size as well as advanced imaging techniques of
muscle volume and body segment mass distribution (▶ Estimation of the Body
Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in
Motion Analysis). Musculotendon force-generating properties define the relation-
ship between input neural excitation and the development of muscle force, which is
nonlinear. Many musculoskeletal models implement a Hill-type musculotendon
model that incorporates muscle force length, muscle force velocity, and tendon
force-strain properties (Zajac 1989; Yamaguchi 2006), as they are tractable for a
full-body model that may include hundreds of muscles (▶ Hill-Based Muscle
Models). Advances in muscle modeling approaches continue to improve our under-
standing of muscle responses to stimulation and muscular effects on skeletal motion.
Once a musculoskeletal model has been developed, the dynamic equations of
motion are given by:
MðqÞ€ _ þ Eðq,qÞ
q ¼ RðqÞFmuscles þ Tpassive ðqÞ þ GðqÞ þ Cðq,qÞ _ (1)
where q, q,_ and q€ are the vectors of generalized positions, velocities, and accelerations
defined within a musculoskeletal model, which can and do include both translational
(e.g., pelvis position relative to the ground) and rotational (e.g., joint angles) degrees
of freedom (▶ Physics-Based Models for Human Gait Analysis). These vectors are
size n 1, where n is the total number of generalized coordinates in the model. M(q)
is the n n system mass matrix or inertia matrix. This matrix is a function of
position (i.e., configuration or pose of the body), and it describes the mass distribu-
tion of the musculoskeletal system at each instant in time. The system mass matrix
is non-diagonal, meaning that the off-diagonal elements of this matrix are nonzero.
R(q) is a matrix of the muscle moment arms as applied to each generalized coordi-
nate, which is a function of the muscle path. Fmuscles is a vector of the forces
generated by the muscles in the model. The term R(q)Fmuscles therefore expresses
the contributions of muscles to the net joint moments. Tpassive is a vector including
torques due to the effects of passive structures, such as ligaments and joint capsules.
Like the term including muscular effects, Tpassive may also be broken into a moment
arm matrix multiplied by linear forces, if ligaments are implemented in the model as
linear passive actuators with attachment points on the body segments. The G(q)
n 1 vector includes the effects of gravity. As moments due to gravity will change
with the system configuration, this vector is a function of the generalized position
coordinates. The Cðq,qÞ_ n 1 vector includes Coriolis and centripetal effects, which
are a function of generalized position and velocity coordinates. The Eðq,qÞ _ n1
vector includes model interactions with the external environment. For example,
during walking, interaction with the environment is quantified by the ground reac-
tion force between the foot and the ground. The dynamic equations of motion
become large as the number of generalized coordinates increases. Thus, as applied
to the musculoskeletal system, these equations are generally formulated explicitly
with the aid of a dynamics software package, such as SD/Fast (PTC), Simbody
(SimTK), MotionGenesis (Motion Genesis LLC), or Adams (MSC Software). Note,
a joint torque-driven model would have a similar equation of motion formulation,
although the moment arm matrix multiplied by muscle forces would instead be a
vector of forces/torques as applied to each generalized coordinate.
Induced Acceleration Analysis
The principle of dynamic coupling is the basis for which results from induced
acceleration and segment power analyses are interpreted. Multiplying both sides of
the dynamic equations of motion (Eq. 1) by the inverted system mass matrix will
result in a formulation for acceleration:

q€ ¼ MðqÞ1 RðqÞFmuscles þ Tpassive ðqÞ þ GðqÞ þ Cðq,qÞ
_ þ Eðq,qÞ
_ (2)
The principle of dynamic coupling (Zajac and Gordon 1989; Zajac et al. 2002) is
clearly revealed mathematically in Eqs. 1 and 2. That is, the application of any one
force to a linked dynamic system will accelerate all joints and segments within that
system. As the system mass matrix is non-diagonal, the application of any muscular
force will result in contributions to the overall acceleration of each generalized
coordinate ( q€), which highlights the coupling inherent within the musculoskeletal
system. Physically, the force produced by a muscle is directly applied to the skeleton
at the points of origin and insertion. A muscle contributes to the net joint torque
about the joints it crosses. Muscle action is distributed to other body segments
through joint intersegmental forces induced by that muscle. Uniarticular muscles
will accelerate the joints they cross in the direction that is consistent with their
anatomical classification. Biarticular muscles may accelerate the joints they cross in
the opposite direction of their anatomical classification due to dynamic coupling.
In an induced acceleration analysis, the effect of any one muscle force on
resulting accelerations of the system is determined by firing one muscle in isolation.
That is, in Eq. 2, all terms except for the muscle force of interest in Fmuscles and that
6 A.K. Silverman
muscle’s contribution to the model’s interaction with the environment (i.e., the
_ ) are set to zero. Then, solving the dynamic equations
muscle’s contribution to Eðq,qÞ
of motion will result in the instantaneous induced acceleration of all generalized
coordinates due to that one muscle’s action (Fig. 1; Eq. 3):
q€musi ¼ MðqÞ1 ½½Rmusi Fmusi þ Emusi ðq,qÞ

_ (3)
Equation 3 would be similarly applied to determine the effects of any non-

muscular force of interest on the accelerations of the dynamic system, such as
passive structures or gravity. Performing this analysis at each instant in time
throughout a movement provides insight into each muscle’s functional role in
completing a dynamic task.
Determining the individual muscle contribution to external reaction forces (Emusi
from Eq. 3) can be accomplished using a reaction force decomposition method, as
described for walking in (Neptune et al. 2001). To perform this decomposition, first,
Fig. 1 Demonstration of induced acceleration of the body’s center of mass (COM) during walking
due to the vasti muscle group, which are uniarticular knee extensors. During walking, the vasti are
activated in early stance (position shown above for the left leg). The vasti muscles produce an
extensor moment about the knee, resulting in angular acceleration of the knee into extension. The
muscles contribute to support and braking the body center of mass, as indicated by center of mass
acceleration upward and backward (blue arrow). A segment power analysis further reveals the
transfer of power among body segments due to the vasti. The vasti induce backward acceleration of
the leg segments while their velocity is forward, absorbing power from the leg (foot, shank, and
thigh). Power is delivered to the trunk (torso and pelvis) to accelerate it upward and forward in the
direction of motion. For additional description of the functional roles of major lower limb muscle
groups during walking, see (Neptune et al. 2001, 2004; Pandy et al. 2010)
a baseline movement simulation is generated and a total ground reaction force and
center of pressure is determined throughout the duration of the movement. The ground
_ Then, at time step t-1, all
reaction force is a function of the state of the system (q, q).
muscle forces except for the muscle of interest (Fmusi ) are applied to the musculo-
skeletal model. The equations of motion are integrated forward in time to time step t.
At time step t, a new ground reaction force has resulted from the reduced system that
has different states. The difference in the ground reaction force between the full and
reduced systems is the contribution of muscle i to the ground reaction force at time
step t. This process is repeated for each instant in time throughout the simulation.
An alternative approach is to model ground contact as a kinematic constraint (e.g.,
weld, pin, or rolling constraint at the center of pressure). In this case, an integration
approach to decompose the ground reaction force is not needed because the effects
of kinematic constraints are resolved simultaneously with the equations of motion
(Hamner et al. 2013).
Induced acceleration analyses are often performed from an underlying muscle-
driven simulation of human movement. However, such a simulation is not neces-
sarily required for this analysis, as induced acceleration results are dependent on
pose. That is, placing the model in a pose that is representative of the motion of
study, and then performing the induced acceleration analysis for a given value of
muscle force, will provide the capacity for one muscle to accelerate the system.
Thus, the relationship between force from a muscle and resulting acceleration can be
determined by Eq. 3. The magnitude of the induced acceleration is proportional to
the force developed by that muscle, but the direction of motion does not change with
the magnitude of the force.
Even though a muscle-driven simulation of movement is not required for an
induced acceleration analysis, often movement simulations are generated to esti-
mate the trajectory of muscle force over a movement, which can be validated in
timing with electromyographic measurements (▶ Surface Electromyography to
Study Muscle Coordination). In this case, the magnitude of the induced accelera-
tion from an individual muscle will be scaled by the magnitude of its force. In an
induced acceleration analysis of a movement simulation, the sum of contributions
from all sources equals the total acceleration of the system at every instant in time.
When performing an induced acceleration analysis, it is important to verify that
this principle of superposition holds. That is, the sum of contributions from all forces
acting on the model (i.e., muscles, ligaments, gravity, assistive devices, etc.) equals
the total acceleration observed. Similarly, the sum of individual muscle contributions
to external forces (e.g., the ground reaction force) should equal the total external
force.
An induced acceleration analysis quantifies the changing effects of a single
muscle on each generalized coordinate throughout the duration of a movement and
across various movement tasks. Thus, this analysis provides important information
regarding how muscles accelerate segments and joints throughout the musculoskel-
etal system. However, induced acceleration analyses alone do not reveal how muscles
work together to generate, absorb, and transfer power between body segments.
8 A.K. Silverman
Segment Power Analysis
Segment power analyses build upon induced acceleration analyses to determine how
muscles are orchestrated to achieve a given movement task (Fig. 1). These analyses
can provide a better description of movement coordination of multiple segments and
are best suited for tasks with a clear mechanical energy objective.
The power generated or absorbed by an individual musculotendon actuator is its
force multiplied by its shortening or lengthening velocity. Muscles generate power
when shortening (i.e., muscle force and velocity are in the same direction) and
absorb power when lengthening (i.e., muscle force and velocity are in opposing
directions). During muscle shortening, a muscle can generate power to the system by
accelerating all segments, or it may absorb power from some segments while
generating more power to others. During muscle lengthening, a muscle will absorb
more power from segments than it generates. When acting isometrically (no muscle
length change), muscles can redistribute power among body segments. Note that for
the musculotendon actuator, the total length change includes the sum of the fiber and
tendon length changes. In certain cases, muscle fibers may be shortening (concentric
contraction), while the overall musculotendon unit is isometric or lengthening. In a
Hill-type musculotendon actuator model, the fiber is represented by the contractile
element, and the tendon is represented by the series elastic element (Zajac 1989)
(▶ Hill-Based Muscle Modeling). For a more detailed explanation of dynamic
muscle fiber-tendon interactions and their direct measurement, see (▶ Ultrasound
Technology for Examining the Mechanics of Muscle, Tendon and Ligament). The
total power delivered to all body segments in a musculoskeletal model by a muscle
equals the total musculotendon power produced by that muscle.
Mechanical power is the time derivative of mechanical energy. In a dynamic model,
the power delivered to or absorbed from a segment is the time derivative of the sum
of potential and kinetic energy of a segment over a short duration in time (Eq. 4):
dE dKE dPE
P¼ ¼ þ (4)
dt dt dt
From the terms in the equations of motion, via Kane’s method, the time derivative
of kinetic energy is:
dKE h iT
q Cðq,q_ Þ q_
¼ MðqÞ€ (5)
dt
The time derivative of potential energy is:
dPE
¼ GðqÞT q_ (6)
dt
Thus, the total mechanical power of the system is the sum of Eqs. 5 and 6 and
results in:
h iT
q Cðq,q_ Þ GðqÞ q_
P ¼ MðqÞ€ (7)
The mechanical power of a particular body segment ( j) within a musculoskeletal

model is computed by setting all terms in the system mass matrix (M(q)) to zero
other than the mass and inertia of the segment of interest. The total power of a
segment equals the sum of the power contributions from all muscles and their
external reaction force contributions, passive structures, gravity, velocity, and any
other forces acting on the musculoskeletal system:
Pj ¼ PMuscle
j þ PPassive
j þ PVelcoity
j þ PGravity
j þ POther
j (8)
These contributions expanded in further detail result in:

h iT h iT h iT
qMuscle q_ þ Mj ðqÞ€
Pj ¼ Mj ðqÞ€ qPassive q_ þ Mj ðqÞ€ _ q_
qVelocity Cðq,qÞ
h iT h iT
þ Mj ðqÞ€ _ q_ þ Mj ðqÞ€
qGravity Gi ðq, qÞ qOther q_ (9)
Equations 8 and 9 represent the total power of an individual segment. The

contributions of individual sources to the total segment power are useful in deter-
mining muscle functional roles. Thus, to determine an individual muscle i’s contri-
bution to the mechanical power of body segment j, the induced acceleration due to
that muscle ( q€musi , computed from an induced acceleration analysis using Eq. 3)
is substituted into Eq. 9:
T
qmusi q_
Pjmusi ¼ Mj ðqÞ€ (10)
The contributions to segment power from other forces acting on the musculo-
skeletal model would be similarly determined. As power is a scalar quantity, the
power from multiple body segments may be added together if the net power delivery
to more than one segment is applicable to the research question. For example, during
walking, energy delivered to the entire leg is of interest to evaluate swing initiation.
The leg may include individual segments for the thigh, shank, and foot.
Often, muscles are co-activated so that they work together to complete a move-
ment task. A demonstrative example of muscles co-activating to complete a movement
occurs in cycling (Fregly and Zajac 1996; Neptune et al. 2000), which has a clear
mechanical energy objective of delivering power to the crank. A segment power
analyses of a torque-driven simulation of cycling showed that the hip extensor torque
works in combination with the ankle plantarflexion torque during the downstroke to
deliver power to the crank (Fregly and Zajac 1996). The hip extensor torque
generates a large amount of mechanical power during the downstroke, but it does
not directly deliver this power to the crank. Rather, the ankle plantarflexor torque
then transfers that power from the leg to the crank. A muscle-level analysis of this
task confirmed these functional roles and revealed that the gluteus maximus, a hip
10 A.K. Silverman
extensor, is co-activated with the soleus, an ankle plantarflexor (Neptune et al. 2000)
to deliver power to the crank during the downstroke. The gluteus maximus generates
power to the system as the hip extends and rotates the thigh. The soleus muscle is
activated at this same portion of the pedaling cycle, acting to stiffen the ankle joint
and redistribute the power generated from the gluteus maximus to the crank. Segment
power analyses reveal the functional roles of multiple muscle groups in generating,
absorbing, and transferring mechanical power throughout the dynamic system dur-
ing a movement.
Joint Contact Force Decomposition
While the focus of this chapter has been on induced acceleration and segment power
analyses specifically, induced approaches can also be applied to decompose joint
contact forces into the contributions from individual muscles (Sasaki and Neptune
2010). Joint contact forces are comprised of the joint intersegmental forces (i.e.,
forces transmitted between linked segments as a result of their acceleration) as well
as compressive action from muscles and other biological tissues. In a similar manner
to performing a ground reaction force decomposition as described above, the joint
contact force may be decomposed into contributions from individual muscles. Joint
contact forces are of critical importance in joint degeneration and are important in
understanding the progression of osteoarthritis and joint pain. For example, in
people with a unilateral transtibial amputation, knee joint osteoarthritis in the intact
leg is a secondary condition of high prevalence. Joint contact force decomposition
methods have suggested that greater forces from the biarticular hip extensors in the
intact leg may contribute to elevated peak loading in the intact knee, thus providing
guidelines for potential gait retraining in order to mitigate disease progression
(Silverman and Neptune 2014).
Interpretation of Induced Acceleration and Power Analyses
Application to Pathological Movement Patterns
Many clinical populations are characterized by reduced mobility and abnormal

movement coordination. Induced acceleration and segment power analyses have
been used to determine how muscle functional roles differ in these populations and
how altered muscle force production may result in altered movement patterns.
Several examples of application of induced acceleration and segment power analyses
to understand movement patterns in clinical populations are provided in the follow-
ing paragraphs. In the following examples, the roles of individual muscles have been
identified to provide insight into how to best target treatment.
In children with cerebral palsy, crouch gait is a common movement pattern
characterized by excessive hip and knee flexion. An induced acceleration analysis
has shown that the uniarticular knee extensor demand increases with greater
crouched postures due to an increased need for body support from muscles as the
efficacy of skeletal support from a straighter leg posture is reduced (Steele et al.
2010). In stiff-knee gait, reduced knee flexion has been attributed to greater forces
from the rectus femoris in pre-swing, which provides a basis for recommending
rectus femoris transfer surgery (Fox et al. 2009). In people poststroke, contributions
of the ankle plantarflexors and hip abductors to body propulsion and muscle power
generation were associated with higher functional walking status (Hall et al. 2011).
In addition, smaller contributions from the hip flexors and the biarticular gastrocne-
mius to deliver energy to the leg were found in a simulation of a limited community
walker and may be targets for strengthening and/or retraining in order to increase
walking speed and improve symmetry (Peterson et al. 2010).
In people with transtibial amputations, the lack of ankle plantarflexor muscle
function results in reduced energy delivered to the leg in pre-swing, which contrib-
utes to asymmetric walking mechanics (Silverman and Neptune 2012). In addition,
reduced body propulsion from the functional loss of the ankle plantarflexors in the
prosthetic leg is counteracted by reduced force output from the ipsilateral knee
extensors to reduce braking of the body center of mass to maintain a constant
walking speed. Induced acceleration approaches have demonstrated that “quadriceps
avoidance” movement patterns observed in people who have undergone total knee
arthroplasty result in reduced contributions from the knee extensors to braking and
supporting the body in early stance and that forward lean of the trunk is an adaptation
that results in greater contributions from the trunk muscles to these walking subtasks
(Li et al. 2013).
Clearly, results of induced acceleration and segment power analyses have pro-
vided important insight into pathological movement patterns, and, thus, there are
many potential uses for clinical translation. However, it is important to note that
limitations in accurately modeling pathological muscle mechanics such as muscle
spasticity and contracture can affect conclusions of muscle force predictions and
therefore interpretation of induced acceleration and segment power analyses.
Limitations of Induced Analyses
Induced acceleration and power analyses provide a wealth of information in how

individual muscles contribute to a coordinated human movement. However, many
decisions, including joint definition, muscle force-generating properties and archi-
tecture, and body segment inertial properties, must be made by a modeler in the
development of musculoskeletal models and movement simulations. As described
above, the underlying dynamic model and associated equations of motion are the
basis of induced acceleration and power analyses. Thus, modeling decisions and
model assumptions will affect the dynamics of the system and the results of induced
acceleration and power analyses. Choosing the correct model for the research
question of interest is critical for obtaining meaningful results. For example, many
early modeling studies defined the human body in two dimensions with few body
segments and hinge joints defined at the ankle, knee, and hip. This type of model has
12 A.K. Silverman
been useful in providing insight into sagittal-plane dynamic tasks, but this model
would clearly not be appropriate for examining mediolateral balance or a turning
motion. A model that lumps the pelvis, lumbar spine,

Visual Speech Animation: Lei Xie, Lijuan Wang, and Shan Yang

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Visual Speech Animation: Lei Xie, Lijuan Wang, and Shan Yang

Uploaded by

Copyright:

Available Formats

Visual Speech Animation

Lei Xie, Lijuan Wang, and Shan Yang

# Springer International Publishing AG 2016 1

Hence, visual speech animation (VSA)1 aims to animate the lips/mouth/articulators/

State of the Art

After decades of research, a state-of-the-art visual speech animation system currently

Fig. 1 The building blocks of a typical VSA system

A Typical VSA System

As shown in Fig. 1, a typical visual speech animation system is usually composed of

According to the source of data used for face/mouth/articulator modeling, a VSA

Camera Video Motion Capture EMA

to reconstruct the 3D marker locations on a subject’s face. Although the mocap

The appearance of a visual speech animation system is determined by the underlying

In facial animation world, a great variety of different animation techniques based on

Input and Feature Extraction

learned text/audio-to-visual mapping model that will be introduced in the next

Fig. 4 Several deﬁned visemes from Ezzat and Poggio (2000)

Lower Background Stitched

The hidden Markov model-based statistical parametric speech synthesis (SPSS)

A Deep BLSTM-RNN-Based Approach

Allowing cyclical connections in a feed-forward neural network, a recurrent neural

ht ¼ HðWxh xt þ Whh ht1 þ bh Þ, (1)

Fig. 7 Bidirectional recurrent neural networks (BRNNs)

it ¼ σ ðWxi xt þ Whi ht1 þ Wci ct1 þ bi Þ, (6)

at ¼ τðWxc xt þ Whc ht1 þ bc Þ, (8)

Fig. 8 Long short-term

ot ¼ σ ðWxo xt þ Who ht1 þ Wco ct þ bo Þ, (10)

The Talking Head System

Face Model and Visual Feature Extraction

Applying principal component analysis (PCA) to all J shapes, sj can be given

where j = 1,2,..., J and U is the total number of pixels.

where t0 is the mean texture. Pt contains the eigenvectors corresponding to the

DBLSTM-RNN Model Training

where 0 α 1 is the learning rate, 0 m 1 is the momentum parameter, and w

and according to Eqs. (21) and (23), we can derive

The performances of the DBLSTM-based talking head are evaluated on an A/V

Table 1 Performance RMSE RMSE RMSE

45.7% 28.6% 25.7%

• Reading news and other information to users

corrective feedback in second language learning. Additionally, a number of studies

When such a database is available, a more powerful LTS can be trained

# Springer International Publishing AG 2016 1

clarifying the mathematical framework for blendshapes. We also demonstrate a

State of the Art

In the following sections, we describe the basic practice and mathematical

Or, if we use matrix notation, Eq. (2) can be expressed as:

where B is an m n matrix having bk – b0 as the k-th column vector, and w = (w1,

Examples and Practice

• Target shape construction

model can be registered to each scan in order to obtain vertex-wise correspon-

Techniques for Efficient Animation Production

Fig. 3 Example of direct manipulation interface for blendshapes

manipulation approach would then be a legitimate requirement. To achieve this, we

Use of PCA Models

In performance-driven facial animation, the motion of human actor is used to derive

c5UT ðBwþb0 e0 Þ, (7)

Blendshape Creation, Retargeting, and Transfer

Creating a blendshape model for professional animation requires sculpting on the

Andrew T. Duchowski and Sophie Jörg

A.T. Duchowski (*)

# Springer International Publishing AG 2016 1

Listing’s and Bonders’ Laws

Modeling Physiologically Plausible Eye Rotation

c5UT ðBwþb0 e0 Þ, (7)

N. Sadoughi () • C. Busso ()

Y. Shen () • J. Zhang () • L. Yang () • H.P.H. Shum ()