applied

sciences
Article
A Brunnstrom Stage Evaluation System
with Fuzzy-Accelerometer
Te-Jen Su 1,2, *, Kun-Liang Lo 1 , Jason Sheng-Hong Tsai 3 , Wen-An Yeh 1 and Cho- Feng Kuo 1
1 Department of Electronic Engineering, National Kaohsiung University of Sciences and Technology,
Kaohsiung City 80778, Taiwan
2 Ph.D Program in Biomedical Engineering, Kaohsiung Medical University, Kaohsiung City 80708, Taiwan
3 Department of Electrical Engineering, National Cheng Kung University, Tainan City 70101, Taiwan
* Correspondence: sutj@nkust.edu.tw




Received: 29 June 2019; Accepted: 28 August 2019; Published: 6 September 2019


Abstract: Advances in medical care has reduced the rate of mortality from strokes, but the incidence
of stroke has remained stable while the incidence of ministrokes has increased. Most stroke victims
require long-term care, imposing a heavy financial and emotional burden on families while incurring
a heavy cost to society. Thus, strokes are a key issue in the context of health care in Taiwan. This paper
proposes using VBA (Visual Basic for Applications) to build a system for assessing Brunnstrom stages
based on the observation of several obvious rehabilitation features The system calculates features
for accelerometer readings, which are then used as input parameters for a fuzzy algorithm to obtain
the Brunnstrom action level. Experimental results show the proposed approach effectively assesses
Brunnstrom level, and that the approach can be used to assist physical therapists in performing
longitudinal assessments of stroke victim progress, thus improving evaluation efficiency.

Keywords: fuzzy algorithm; accelerometer; Brunnstrom stages; features

1. Introduction
In recent years, the World Health Organization (WHO) has repeatedly issued a “1 in 6” warning
referring to the likelihood that 1 in 6 people can expect to suffer a stroke sometime during their lifetime,
regardless of age, gender or ethnicity [1]. Stroke incidents not only entail high medical costs for acute
treatment proximate to the stroke incident, but also require intensive, long-term physical rehabilitation
to restore mobility, resulting in huge costs to families and society at large. Limited human resources
create an important need for, information technology and sensing devices that can be used to monitor
and evaluate the rehabilitation status of hemiplegia patients.
During WWII, Signe Brunnstrom noted that hemiplegia patients seem to follow a stereotyped
recovery process. Between 1954 and 1956, observation of 100 hemiplegia patients was used to divide the
recovery process into seven distinct periods, referred to as Brunnstrom Stages [2], but a clear evaluation
model was lacking. Later, in 1974, Leanne M. LaVigne established a clear stage evaluation model based
on Brunnstrom’s staging test, and made suggestions for appropriate treatment, and these methods
are still used by clinicians today as the basis for diagnosis of degree of hemiplegia [3]. To develop an
appropriate rehabilitation strategy, physicians need to first evaluate the current degree of hemiplegia,
using methods including the Brunnstrom Approach, the Fugl-Meyer Assessment (FMA) [4] and the
Ueda 12-Stage Hemiplegic Function Test [5].
Human body posture recognition is often used in rehabilitation evaluation, sports and film
production. The main methods include the use of an Oqus camera, Kinect sensors, and Inertial
Measurement Units (IMU). An optical high-speed camera is used to record the movement of reflective
markers attached at specific points on the surface of the human body, creating a series of dot trajectory

Appl. Sci. 2019, 9, 3718; doi:10.3390/app9183718 www.mdpi.com/journal/applsci

Appl. Sci. 2019, 9, 3718 2 of 15

data which describe the movement of each limb. This approach is often used to produce animation
or to analyze the movement of athletes [6]. While this measurement method can accurately describe
human body movement, it requires data collection equipment to be fixed in space along with significant
high-speed computing resources for movement analysis. Somatosensory detectors use an infrared
depth camera to capture skeletal images for motion capture and analysis [7,8]. This method does
not require the subject to wear sensors, and thus does not impede subject movement. However this
approach is relatively expensive, and is subject to limitations imposed by the recording venue and the
number of subjects, making it inapplicable to the current case.
With the rapid development of sensing technologies in recent years, inertial sensors have
gradually emerged as the mainstream method for capturing motion signals, including accelerometers,
gyroscopes, and magnetic sensors. Of these, accelerometers provide good performance with low power
consumption and low cost [9]. In addition, the development of Micro Electro Mechanical Systems
(MEMS) allows for such sensors to be worn directly in various places on the human body, with data
collected, processed and transmitted in real time anywhere and anytime. Acceleration sensors are
now widely used in various types of human motion recognition research, including common daily
motion activities such as walking [10], running [11], and ascending/descending stairs [12]. Real-time
human motion recognition is useful for fall detection in home care research [13]. Machine learning
techniques can be applied to longitudinal recording of human motion patterns to allow for more
precise classification and recognition of various movement types [14]. Some studies have focused on
the evaluation of hemiplegia patients [15,16] using machine learning techniques to train samples for
the development of classification and evaluation models. While such approaches can provide highly
accurate identification results, they require very large learning samples and complex mathematical
operations. While fuzzy theory [17] is a relatively simple experience algorithm, it does not require a
large sample set, and can be adjusted according to the experience of the physical therapist. Although
the Brunnstrom sub-stages are still in use today, the evaluation method still needs to be modified
according to the region in question. Applying machine learning techniques will require retraining
with a large sample set. In addition, fuzzy calculations only require parameters adjustment to meet
the requirements.
This paper combines the Brunnstrom method with the fuzzy algorithm to obtain more accurately
and objectively characterize motion data and analyze the patient’s motion features, thus assisting
physical therapists in diagnosis and treatment. Through action analysis, this paper sets 7 main features
and performs a two-stage fuzzy assessment. In the first stage, the original accelerometer data for
the Flexion and Pronation actions are smoothed using low-pass filtering [18,19], and the features are
used as the input parameters for the fuzzy algorithm to calculate the Flexion and Pronation action
scores. The second stage uses the Flexion and Pronation outputs as input Fuzzy parameters, and
then adds the Lap to Chin action score to calculate a Brunnstrom grade.. For hemiplegia patients,
this entails adjusting personalized parameters, such as applying the fuzzy parameters for degree of
concealment to improve the training stability. Thus, fuzzy theory provides considerable value for
rehabilitation practices.

2. System Development
Brunnstrom Stages are used for evaluation using three actions, as follows: Flexion, pronation and
lap to chin (Table 1). These actions and assessed for correctness based on three concepts, as follows:
Accuracy, stability and consistency. Accuracy refers to whether the action fulfills the requirements, while
stability reflects the arm not bending during the action, and consistency reflects uniformity between
the healthy limb and the affected limb. Flexion and Pronation emphasize accuracy and stability based
on these three motion characteristics, while Lap to Chin does not require stability and performance is
determined as a function of speed. Raw accelerometer data collects seven characteristic values.
Through action analysis, our paper employs a two-stage fuzzy method with seven main features.
In the first stage, the original accelerometer data for the Flexion and Pronation actions are smoothed
Appl. Sci. 2019, 9, 3718 3 of 15
Appl. Sci. 2019, 9, x FOR PEER REVIEW 3 of 15

using
using low-pass
low-pass filtering
filtering [18,19],
[18,19], and
and the
the features
features are
are used
used asas the
the input
input parameters
parameters for
for the
the fuzzy
fuzzy
algorithm to
algorithm to calculate the Flexion and Pronation action scores. The second stage uses the
the Flexion and Pronation action scores. The second stage uses the Flexion andFlexion
and Pronation
Pronation outputs
outputs as input
as input parameters,
parameters, andand then
then addsthe
adds theLap
LaptotoChin
Chinaction
action score
score to calculate
calculate aa
Brunnstrom
Brunnstromgrade.
grade.The
Thesystem
systemdevelopment
developmentisisdescribed
describedin
indetail
detailas
asfollowing
followingsections.
sections

2.1.
2.1. Measurement
Measurement System
System
This
Thissystem
systemisisbuilt onon
built an an
Arduino Nano
Arduino development
Nano boardboard
development (microcontroller: ATmega328,
(microcontroller: clock
ATmega328,
speed: 16 MHz, size: 18 × 45 mm; weight: 7 g), using an acceleration sensor (type MPU-6050,
clock speed: 16MHz, size: 18x45mm; weight: 7g), using an acceleration sensor (type MPU-6050, 16bit 16 bit
A/D
A/Dconverter,
converter, size: 21 × 16 mm,
size: 21x16mm, range:
range: ±16
±16g) tog) to measure
measure actionaction acceleration
acceleration in dimensions
in three three dimensions
(x,y,z).
(x,y,z). The analog signal of the obtained acceleration value is transmitted to the development
The analog signal of the obtained acceleration value is transmitted to the development board via an board
via an XBee-S2C
XBee-S2C wireless
wireless communication
communication modulemodule (indoor/urban:
(indoor/urban: 60m; 60 m; outdoor
outdoor RF line-of-sight,
RF line-of-sight, RF
RF data
data
rate:rate: 250 Kbps)
250 Kbps) to Communication
to the the Communication PortPort
(COM (COM Port)
Port) for subsequent
for subsequent analysis.
analysis. Figure
Figure 1 shows
1 shows the
the measuring device of the experimental system mounted on the
measuring device of the experimental system mounted on the user’s wrist. user’s wrist.

Figure1.1. Schematic
Figure Schematicdiagram
diagramof
ofmeasurement
measurementsystem.
system.

2.2.
2.2. Accelerometer
Accelerometer Data
DataProcessing
Processing
Table
Table11shows
showsthe thefeatures
featuresssandanddescriptions
descriptionsof ofthe
thethree
threeactions
actionsininthe
theexperiment.
experiment.When Whenthe the
Flexion
Flexion action
action isis performed,
performed, the the y-axis
y-axisvalues
valueswillwillclearly
clearlyrise
riseand
andfall,
fall,producing
producingaawaveform-like
waveform-like
motion
motionas asshown
shownin inFigure
Figure2.2. The
The maximum
maximum value value (peak)
(peak) is
is scaled
scaled and
and averaged
averaged to to produce
producethethefirst
first
features
features value FAA (Flexion Average Angle), and Flexion Angle Standard Deviation (FASD) isisthe
value FAA (Flexion Average Angle), and Flexion Angle Standard Deviation (FASD) the
standard
standarddeviation
deviationof ofthe
theobtained
obtainedpeakpeakvalue.
value. Finally,
Finally,Flexion
FlexionLateral
LateralStability
Stability(FLS)
(FLS)calculates
calculatesthethe
standard deviation based on the x-axis data for the Flexion. If the Flexion
standard deviation based on the x-axis data for the Flexion. If the Flexion is performed with high is performed with high
stability, thex-axis
stability,the x-axisdatadatawill
willshow
showlittle
littlevariability.
variability.However,
However,high highvariability
variability in thex-axis
inthe x-axiswill
willalso
also
result
resultininparameter
parametervalue valuevariability
variabilityin thez-axis
inthe z-axisduring
duringaction
actionperformance.
performance. However,
However, givengiventhat
that
the Flexion action is mainly performed in the vertical and horizontal axes, it is
the Flexion action is mainly performed in the vertical and horizontal axes, it is not discussed in this not discussed in this
experiment.
experiment.Pronation
Pronationisissimilar
similarto to
Flexion
Flexionin that it uses
in that z-axisz-axis
it uses data to obtain
data the average
to obtain angle, where
the average angle,
the PAAthe
where (Pronation AverageAverage
PAA (Pronation Angle) and the and
Angle) PASD the(Pronation Angle Standard
PASD (Pronation Deviation)
Angle Standard are usedare
Deviation) to
calculate the standard deviation of the peak angle, while standard deviation
used to calculate the standard deviation of the peak angle, while standard deviation of the PVS of the PVS (Pronation
Vertical Stability)
(Pronation Verticalis calculated
Stability) isusing y-axis using
calculated data. Thus
y-axisadata.
stableThus
pronation
a stableaction will result
pronation actioninwill
minimal
result
y-axis variability, and vice versa. Pronation will also produce x-axis variability,
in minimal y-axis variability, and vice versa. Pronation will also produce x-axis variability, but the but the action’s before
and after before
action’s positionand are after
mostly level, and
position arethus is not
mostly discussed
level, here.isLap
and thus nottodiscussed
Chin variability
here. Lapis based
to Chinon
inconsistency
variability is between
based onthe patient’s affected
inconsistency between andthe healthy sides,
patient’s and neither
affected accuracy
and healthy nor stability
sides, are
and neither
considered, thus the z-axis data can be taken as the action peak.
accuracy nor stability are considered, thus the z-axis data can be taken as the action peak.
We first tested the simulation data for different levels and conditions, confirming that the
simulation algorithm can correctly classify Table 1.the various
Define dataFeatures.
motions as belonging to ST2–ST6 (stage 2 to stage 6)
and Healthy. ST1(stage 1) is omitted because it is defined by Brunnstrom as “Flaccidity” or lack of
movement Motion
and thusName cannot beFeatures
detected by use of the accelerometer. Description
The definition of ST2–ST6 are list
in Table 2.
FAA Average angle of peak values for y-axis
Flexion FASD Standard deviation of peak values for y-axis
FLS Standard deviation of values for x-axis
Appl.Appl.
Sci. Sci.
2019, 9, 3718
2019, 9, x FOR PEER REVIEW 4 of 15 4 of 15

PAATable 1. Define
Average angle
motions of peak values for z-axis
Features.
Pronation PASD Standard deviation of peak values for z-axis
Motion Name PVSFeatures Standard deviation of values for y-axis
Description
FAA Average angle of peak values for y-axis
Lap to Chin LTCMR Times of the peak value for z-axis
Flexion FASD Standard deviation of peak values for y-axis
FLS Standard deviation of values for x-axis
We first tested the simulation PAA data for different levelsangle
Average and of
conditions,
peak values confirming
for z-axis that the
simulation algorithm can correctly classify
Pronation PASD the various data deviation
Standard as belonging to ST2–ST6
of peak values (stage 2 to stage
for z-axis
6) and Healthy. ST1(stage 1) is omitted because it is defined
PVS Standardbydeviation
Brunnstrom as “Flaccidity”
of values for y-axisor lack of
movement and Lap to Chin
thus LTCMR
cannot be detected Times of the peak
by use of the accelerometer. Thevalue for z-axis
definition of ST2–ST6 are
list in Table 2.
Table 2. Definition of Brunnstrom Stages.
Table 2. Definition of Brunnstrom Stages.
Flexion Pronation Simulate Level
Simulate
0–30Flexion Pronation
0–30 Level ST2
30–60 0–30 30–60
0–30 ST2 ST3
60–90 60–90 ST4
30–60 30–60 ST3
90–120 90–120 ST5
60–90 60–90 ST4
120–150 120–150 ST6
90–120
150–180 90–120
150–180 ST5 Healthy
120–150 120–150 ST6
150–180 150–180 Healthy
Figure 2 shows the raw acceleration data of ST2–ST6 along the y-axis for the Flexion action
Figure 2 shows the raw acceleration data of ST2–ST6 along the y-axis for the Flexion action
performed by the health limb. The horizontal access shows a data sequence with 20 data points per
performed by the health limb. The horizontal access shows a data sequence2with 20 data points per
second, and the acceleration value of the vertical axis is (g, g = 9.81 × s ), thus the time required to
second, and the acceleration value of the vertical axis is (g, g 9.81 𝑠 ), thus the time required to
perform the operation differs with stage, so Figure 2 extracts the acceleration signal for five movements.
perform the operation differs with stage, so Figure 2 extracts the acceleration signal for five
movements.

Figure 2. Flexion motion’s y-axis’s raw data from top to bottom ST2, ST3, ST4, ST5, ST6, and Healthy.

From Figure 2 we can see that some micro-vibrations produced a jagged signal during the
action, causing system errors when identifying feature values. Therefore, this paper used low-pass
 Figure 2. Flexion motion’s y-axis’s raw data from top to bottom ST2, ST3, ST4, ST5, ST6, and
Healthy.

From Figure 2 we can see that some micro-vibrations produced a jagged signal during the action,
Appl. Sci. 2019, 9, 3718 5 of 15
causing system errors when identifying feature values. Therefore, this paper used low-pass filtering
to filter out noise prior to identification. The characteristic values of each action are shown in Equation
(1) [20]. Figure
filtering to filter3out
shows
noisethe filtered
prior accelerationThe
to identification. value, where thevalues
characteristic original sawtooth
of each actionsignal is now
are shown in
smoothed.
Equation (1) [20]. Figure 3 shows the filtered acceleration value, where the original sawtooth signal is
now smoothed.
𝑦𝑖 𝑦𝑖 1 𝑎 𝑥𝑖 𝑦𝑖 1 , (1)
y[i] = y[i − 1] + a × (x[i] − y[i − 1]), (1)
where 𝑦 𝑖 is the filtered data, 𝑎 is the weighting degree between 0–1, 𝑦 𝑖 1 is the previous
where y[i] is the filtered data, a is the weighting degree between 0–1, y[i − 1] is the previous filtered
filtered data, and 𝑥 𝑖 is the original data. In this paper, parameter 𝑎 is set to 0.4. Repeated
data, and x[i] is the original data. In this paper, parameter a is set to 0.4. Repeated experiment finds
experiment finds that excessively high settings resulted in a poor filtering effect, resulting in
that excessively high settings resulted in a poor filtering effect, resulting in identification errors for
identification errors for feature values. Excessively low settings filter out too much, producing feature
feature values. Excessively low settings filter out too much, producing feature values inconsistent with
values inconsistent with the original data, thus the parameter value of 0.4 best meets system
the original data, thus the parameter value of 0.4 best meets system requirements.
requirements.

Figure
Figure 3.
3.Flexion
Flexionmotion’s
motion’sy-axis’s
y-axis’sfiltered
filtered data
data from
from top
top to
tobottom
bottomas
asST2,
ST2,ST3,
ST3,ST4,
ST4,ST5,
ST5,ST6,
ST6,Healthy.
Healthy.

2.3.
2.3. Features
Features Analysis
Analysis
Figure
Figure 33 shows
shows that, during the Flexion action, the acceleration along the y-axis is is about
about −1-1 gg
during
during preparation,
preparation, and
and the
the acceleration
acceleration along
along the
the y-axis
y-axis increases as the arm begins
begins toto rise
rise and
and then
then
decreases
decreases as the
the arm
arm begins
begins to fall.
fall. Figure
Figure 33 shows
shows that
that each
each action
action has
has aa maximum
maximum (peak)
(peak) value,
value, and
and
this
this peak is used in Equation (2) to calculate the angle, which is then averaged as shown in Figure4.4.
peak is used in Equation (2) to calculate the angle, which is then averaged as shown in Figure
As shown in in the
the diagram
diagram ofof Equation
Equation (2),
(2), rather
rather than
than taking
taking the
the maximum
maximum or or minimum
minimum angle angle of
of the
the
user’s
user’s motion,
motion, averaging
averaging the
the angles
angles provides
provides aa more
more objective
objective representation
representation ofof the
the angle
angle of
of user
user
motion, because there may be discrepancies between the angle of each action, thus the average angle
better matches the value of a specific level.
Pn
i=1 ( yi + 1) × 90
yANG = , (2)
n
Appl. Sci.Sci.
Appl. 2019, 9, 9,
2019, x FOR PEER
x FOR REVIEW
PEER REVIEW 6 of 1515
6 of

motion,
motion,because there
because may
there maybebe
discrepancies
discrepancies between the
between angle
the ofof
angle each action,
each thus
action, the
thus average
the angle
average angle
Appl.
betterSci.matches
2019, 9, 3718
the
value of a specific level. 6 of 15
better matches the value of a specific level.
∑∑
where yANG is the average angle, n is the total
𝑦
𝑦 number of peaks, ,and yi is the ith peak. , (2)(2)

y y= =0 0g,g,angle
angle= =9090° °

y y= =-0.3
-0.3g,g,angle
angle= =6363° °
Figure
Figure 4. 4.
4.
Figure Schematic diagram
Schematic ofof
diagram Equation(2).
Equation (2).
Equation(2).

Figure 5a showsthe theaverage

resultsangle,
of the𝑛peaks for the ST4 acceleration data 𝑦 𝑦in Figure 3. Since ST4 falls
where𝑦 𝑦 is is
where the average angle, 𝑛 is is
the total
the number
total number ofofpeaks,
peaks, and
and is is
the 𝑖𝑡ℎ𝑖𝑡ℎ
the peak。
peak。
between ST3 and ST5, the signal easily falls below or above the stage, thus ST4 is used as an example.
Figure
Figure 5(a) shows the results of the peaks for the ST4 acceleration data in Figure3. 3.Since
5(a) shows the results of the peaks for the ST4 acceleration data in Figure SinceST4ST4
Although the average angle more easily fits a specific level, the average angle disregards the peak
falls
falls between ST3 and ST5, the signal easily falls below or above the stage, thus ST4 is usedasasanan
between ST3 and ST5, the signal easily falls below or above the stage, thus ST4 is used
levels for each action, and thus the precise angle of each action is unknown. It can be seen that the five
example.
example.Although
Althoughthe theaverage
averageangleanglemoremoreeasily
easilyfits
fitsa aspecific
specificlevel,
level,the
theaverage
averageangle
angledisregards
disregards
peaks in Figure 5a are small and the motion is relatively stable, but the motion angle of 62–85.3◦ in
the peak levels for each action, and thus the precise angle of each
the peak levels for each action, and thus the precise angle of each action is unknown. It canaction is unknown. It can bebe seen
seen
Figure 5b represents a significant drop from the 71.1–79.2◦ angle in Figure 5a. The yANG of both are
that
thatthe five
the peaks
five peaks in Figure
in Figure 5(a) are
5(a) aresmall
small and
andthe motion
the motion is relatively
is relatively stable,
stable,but the
but motion
the motion angle
angle ofof
73.62◦ , thus the average angle alone cannot objectively represent the user’s movement, and the peak is
62–85.3 ° in Figure 5(b) represents a significant drop from the 71.1–79.2
62–85.3 ° in Figure 5(b) represents a significant drop from the 71.1–79.2 ° angle in Figure 5(a). The ° angle in Figure 5(a). The
incorporated into Equation (3) to calculate the standard deviation of the angle, where a smaller value
𝑦 𝑦 ofofboth bothareare73.62
73.62°, °,thus
thusthetheaverage
averageangleanglealonealonecannot cannotobjectively
objectivelyrepresent
representthe theuser’s
user’s
indicates a small degree of variation.
movement,
movement, and
andthe
thepeak
peakis is
incorporated
incorporated into
intoEquation
Equation (3)(3)
toto calculate
calculate thethestandard
standard deviation
deviation ofofthethe
angle, where a smaller value indicates a
angle, where a smaller value indicates a smallPdegreesmall s degree of variation.
 of variation. 2
n
i=1 yi − yANG
yASD = ∑∑ , (3)
𝑦𝑦 n , , (3)(3)

where yASD is the angular standard deviation, yANG is the average angle, n is the total number of peaks,
where 𝑦 𝑦 is is the angular standard deviation, 𝑦 is is
the average angle,𝑛 𝑛is is
angle, the total number
is the ith peak.the angular standard deviation, 𝑦
and yi where the average the total number
ofof
peaks, and 𝑦 is the 𝑖𝑡ℎ peak.
peaks, and 𝑦 is the 𝑖𝑡ℎ peak.

100
100 100
100
85.3
85.3
79.2
79.2 7777
8080 7272 74.7
74.7 8080 7373
71.1
71.1 71.1
71.1 70.8
70.8
6262
Angel( °)

6060
Angel( °)

6060
Angel( °)
Angel( °)

4040 4040

2020 2020

00 00
11 22 33 44 55 11 22 33 44 55
Number
Numberofof
Peak
Peak Number
Numberofof
Peak
Peak

(a)(a) (b)(b)

Figure 5. Flexion average angle comparison (a) Stability peak value and (b) Non stability peak value.
Appl. Sci. 2019, 9, x FOR PEER REVIEW 7 of 15

Appl. Sci. Figure

2019, 9, 5. Flexion average angle comparison (a) Stability peak value and (b) Non stability peak
3718 7 of 15
value.

Figure
Figure 66 shows
showsthethex-axis
x-axis acceleration
acceleration datadata for Flexion
for the the Flexionmotion.motion. Normally
Normally thiswill
this data data will
remain
remain
relativelyrelatively static the
static during during the motion.
motion. Figure 6a Figure
shows6(a)thatshows
x-axisthat x-axis acceleration
acceleration dataabove
data was both was both
and
above
below and below the
0 g during 0 g motion.
during the motion.
If the If the
arm does notarm
benddoes not the
during bend duringthe
motion, thedata
motion, the dataa will
will produce line
produce a line
close to zero close
with to zerovariation,
minimal with minimal variation,
as shown as shown
in Figure 6b. Figure in Figure
6b also6(b). Figure
shows that,6(b) alsoFlexion,
during shows
that, during
the x-axis Flexion, the
acceleration x-axis
data acceleration
varies greatly withdata variestogreatly
respect Figurewith respect
6a. Since thetoarm
Figure 6(a).
is also Since
bent the
during
arm is also bent
the action, during
the value the action,
is not as smooththe as
value is not as
in Figure 6a.smooth
Therefore,as inthe
Figure 6(a).calculates
system Therefore, thethe system
standard
calculates
deviation theusingstandard
Equation deviation
(4) fromusing Equation
the x-axis (4) fromdata,
acceleration the x-axis acceleration
and uses data, and
this parameter uses this
to assess the
parameter to assess the
degree of horizontal degree
stability withof which
horizontal stability
the subject with which
performs the subject
the Flexion action,performs the Flexion
with a smaller value
action, withhigher
indicating a smaller value indicating higher stability.
stability.
s
Pn 2
=1 (xi − x)̅
i∑
𝑥 =
xASD
n
,, (4)
(4)

where xASD𝑥is the

where standard
is the deviation
standard of the
deviation x-axis,
of the 𝑥 xaxis,
is the𝑥̅ average amplitude
is the average of theofx-axis,
amplitude the 𝑥 n isaxis,
the
total number of data points on the x-axis, and x is the ith data.
𝑛 is the total number of data points on the 𝑥 iaxis, and 𝑥 is the 𝑖𝑡ℎ data.

(a) (b)
Figure 6. Flexion
Figure 6. Flexion x-axis
x-axisstandard
standarddeviation
deviation comparison
comparison (a) (a) stable
stable x-axis
x-axis acceleration
acceleration data,data, (b) less-
(b) less-stable
stable
x-axis x-axis acceleration
acceleration data. data.

The
The Pronation
Pronation action uses the same same features
features as thethe Flexion
Flexion action.
action. Pronation
Pronation stretches
stretches and
and rotates
rotates
the
the hand
hand 180 degrees, and is used to determine the degree of write write joint
joint rotation.
rotation. This
This action
action is
is
calculated usingacceleration
calculated using accelerationdatadataalong
alongthethe z-axis.
z-axis. TheThe average
average angleangle
and and its standard
its standard deviation,
deviation, along
along with
with the the acceleration
acceleration data along
data along the y-axis
the y-axis is usedis used to assess
to assess vertical
vertical stability
stability when when performing
performing this
this action
action because
because the the wrist
wrist is kept
is kept straight
straight during
during Pronation,
Pronation, and and
thethe acceleration
acceleration data
data in in
thethe y-axis
y-axis is
is
notnot due
due to motion
to motion of the
of the wrist.
wrist. When When rotation
rotation is performed,
is performed, the numerical
the numerical valuevalue fluctuates
fluctuates in a
in a large
large
range,range,
and ifand if the fluctuation
the value value fluctuation
increasesincreases as a function
as a function of wrist of wrist rotation,
rotation, this indicates
this indicates unstable
unstable wrist movement
wrist movement during Pronation.
during Pronation. The Lap The Lapaction
to Chin to Chin is action is only
only tested in tested
terms ofin speed
terms of to speed
assess
to assess Brunnstrom
Brunnstrom stage, comparing
stage, comparing the peaks
the peaks taken from taken from the
the healthy andhealthy andsides
hemiplegia hemiplegia sides
over a five over
second
aperiod.
five second period.used
This paper Thisthe
paper
ratioused the ratio ofside
of hemiplegia hemiplegia
actions to side actions
healthy to healthy
side side
actions as theactions as
basis for
the basis for
assessing theassessing
differencethe difference
between betweenhealthy
the subjects the subjects healthy andsides.
and hemiplegia hemiplegia sides.

2.4. Fuzzy
2.4. Fuzzy System
System Design
Design
According to
According to the
the Brunnstrom
Brunnstrom method,
method, aa higher
higher average
average angle
angle and lower angular
angular standard
standard
deviation indicates
deviation indicates higher
higher action
action stability
stability and
and aa higher
higher level.
level. A
A set
set of
of fuzzy
fuzzy rules
rules was
was accordingly
accordingly
established, and the fuzzy rule design
established, and the fuzzy rule design was divided
divided into establishing
establishing input and output language
language
variables to create fuzzy rules, fuzzy inferences and defuzzification.
variables defuzzification. This
This rule
rule was
was used
used to
to design
design A
A
Brunnstrom Stage Evaluation System (BSES) with the fuzzy system design, as shown in Figure 7.
Brunnstrom 7.
Appl. Sci. 2019, 9, 3718 8 of 15
Appl. Sci. 2019, 9, x FOR PEER REVIEW 8 of 15

Appl. Sci. 2019, 9, x FOR PEER REVIEW 8 of 15

Establish input and output variables
Establish input and output variables

Establish linguistic terms

Establish linguistic terms

Membership Function
Membership Function

Fuzzy Rule Database

Fuzzy Rule Database

Defuzzification
Defuzzification

Evaluation Brunnstrom Level

Evaluation Brunnstrom Level
Figure
Figure 7. 7.Schematic
Schematicdiagram
diagram of
ofBSES
BSESFuzzy
FuzzySystem.
System.
Figure 7. Schematic diagram of BSES Fuzzy System.
2.4.1.2.4.1.
Input andand
Input Output
Output
2.4.1.This
This Input
paper and
paper Output
usesuses two-order
two-order fuzzyjudgment.
fuzzy judgment. The
The fuzzy
fuzzyinference
inferenceis is
thethe
min-min-max
min-min-max operations
operations
proposed
proposed This by
by MamdaniMamdani
paper uses in 1975,
intwo-order
1975, andfuzzyand defuzzification
defuzzification uses
usesfuzzy
judgment. The the centroid
the centroid method
inferencemethod to solve the explicit
to solve the operations
is the min-min-max explicit output
output
(Crisp).
proposed (Crisp).
Before by Before
making
Mamdani themaking
1975,the
injudgment,andjudgment,
we we first
first confirm
defuzzification confirm
the
uses input
the theand
centroidinput and to
output
method output
solve parameters.
parameters. Figure 8a
the explicit
Figure
output 8(a) shows
(Crisp). the Flexion
Before making motion fuzzy judgment
the judgment, we first diagram.
confirm FAA, FASD and FLS are used to
shows the Flexion motion fuzzy judgment diagram. FAA, FASDthe andinput
FLSandare output
used toparameters.
calculate FMR.
calculate
Figure FMR. Figure
8(a) shows 8(b) shows
the Flexionmotion the Pronation
motionfuzzy motion
fuzzy judgment fuzzy
judgment diagram. judgment diagram. PAA, PASD and
Figure 8b shows the Pronation diagram.FAA,PAA, FASD andand
PASD FLSPVS
are used to to
are used
PVS are
calculate used
FMR.to calculate
Figure 8(b)PMR, and
shows Figure
the 8(c)
Pronationshows the
motion Brunnstrom
fuzzy level
judgment fuzzy judgment
diagram. PAA, diagram.
PASD and
calculate
FMR PMR,
andused andare
PMR Figure
used as8cinput
shows the Brunnstrom
parameters, level the
fuzzy
Lapjudgment diagram. FMR andtoPMR
PVS are to calculate PMR, and Figure 8(c)and added
shows theto
Brunnstrom to level
Chin fuzzy
Motion Rate (LTCMR)
judgment diagram.
are used as inputBrunnstrom
calculate parameters, and added to the Lap to Chin Motion Rate (LTCMR) to calculate the
FMR andthe PMR are used asLevel.input parameters, and added to the Lap to Chin Motion Rate (LTCMR) to
Brunnstrom Level.
calculate the Brunnstrom Level.

(a)
(a)

(b)
(b)

(c)
(c)

Figure 8. Schematic diagram of Fuzzy inference (a) Flexion (b) Pronation, and (c) Brunnstrom Level.
Appl. Sci. 2019, 9, 3718 9 of 15

2.4.2. Linguistic Terms and Membership Function

In the fuzzy judgment, the related linguistic terms are used according to the input variables.
Each term is judged by at least three points (small, medium and large) from the polarized value to the
intermediate value. The use of too many terms can easily lead to judgment problems and increase
calculation time.
The first-order input parameters FAA and PAA are divided into five levels, as follows: very
high, high, medium, low and very low. FASD and PASD are divided into three levels, as follows:
low, medium and high. FLS and PVS are divided into five levels, as follows: very stable, moderate,
acceptable, unstable and very poor. The outputs FMR and PMR are divided into nine levels, as follows:
best, excellent, good, slightly better, normal, slightly worse, worse, very poor and worst.
Mathematics
The input 2019, xx, x
parameters of the second stage, FMR and PMR are divided into three levels, as1 offollows:
7

poor, normal and good. LTCMR is divided into three levels, as follows: low, medium and high, and
the output of the Brunnstrom grade is divided into six levels, as follows: ST2, ST3, ST4, ST5, ST6
and Healthy.
After establishing related terms, an attribution function must be defined for each term.
The membership function (MF) inputs and outputs for this paper all use triangular and trapezoidal
attribution functions. The various central values of the attribution function and their value range affect
the fuzzification of the input variable. Figure 9 shows the attribution function of first-order Flexion
and Pronation.

(a) (b)

(c) (d)

Figure 9. Flexion and Pronation MFs (a) FAA, PAA (b) FASD, PASD (c) FLS, PVS (d) FMR, PMR.

The first stage output results for FMR and PMR plus LTCMR are used as input parameters for the
second stage. Figure 10 shows the attribution function design.
 Mathematics 2019, xx, x 2 of 7

Appl. Sci. 2019, 9, 3718 10 of 15

(a) (b)

(c) (d)

Figure 10. Brunnstrom Level MFs (a) FMR (b)PMR (c) LTCMR, and (d) Brunnstrom Level.

2.4.3. Fuzzy Inference Rules

Observations of the subject’s rehabilitation data can be combined with his Brunnstrom level and
summarized for the fuzzy rule base. The user can then determine the output terms according to
the input term. Tables 3 and 4 list the 75 Flexion and Pronation action rules, while Table 5 lists the
27 Brunnstrom Level rules.

Table 3. Stage 1 rules: Some part of Flexion motion’s rules.

IF Then
FAA FASD FLS FMR
Very High Low Very Good Best
Very High Low Good Very Good
Very High Low SoSo Good
Very High Low Bad Not Bad
Very High Low Very Bad SoSo
Very High Middle Very Good Very Good
Very High Middle Good Good
Very High Middle SoSo Not Bad
Very High Middle Bad SoSo
Very High Middle Very Bad Not Good
Very High High Very Good Good
Very High High Good Not Bad
Very High High SoSo SoSo
Very High High Bad Not Good
Very High High Very Bad Bad
Appl. Sci. 2019, 9, 3718 11 of 15

Table 4. Pronation motion’s rules of Stage 1.

IF Then
PAA PASD PVS PMR
Very High Low Very Good Best
Very High Low Good Very Good
Very High Low SoSo Good
Very High Low Bad Not Bad
Very High Low Very Bad SoSo
Very High Middle Very Good Very Good
Very High Middle Good Good
Very High Middle SoSo Not Bad
Very High Middle Bad SoSo
Very High Middle Very Bad Not Good
Very High High Very Good Good
Very High High Good Not Bad
Very High High SoSo SoSo
Very High High Bad Not Good
Very High High Very Bad Bad

Table 5. Some part of Level rules.

IF Then
FMR PMR LTCMR Level
Bad Bad Low Worst
Bad Bad Middle Worst
Bad Bad High Very Bad
Bad SoSo Low Worst
Bad SoSo Middle Very Bad
Bad SoSo High Bad
Bad Good Low Very Bad
Bad Good Middle Bad
Bad Good High SoSo

3. Experimental Design and Results

This study is designed by the researchers who are not clinicians but instructed by a doctor. Based
on Brunnstrom Stages, ten healthy male research subjects were recruited to perform simulations (age
34 ±11 years, height 171.5 ±6.5 cm, weight 70 ±20 kg). Subjects first went through a 30 min training
routine to familiarize them with the different movement ranges for each Brunnstrom Stage. As the use
of authentic data would not meet the sampling requirement for the experiment, ten healthy subjects
were used to simulate the actions of Brunnstrom Stage II (ST2) to VI (ST6) along with healthy action.
In each simulation, 60 data points were obtained for Flexion, Pronation and Lap to Chin to analyze
variation at different levels.
Table 6 provides a reference for the simulations tested in this experiment. Experimental subjects
E1–E10 were instructed to repeatedly simulate Flexion and Pronation at Brunnstrom Stage-specific
angles, along with Lap to Chin to obtain additional different Lap to Chin scores. Lap to Chin was
simulated five times on the Good Side, and the Bad Side was then adjusted based on the specified level.
Appl. Sci. 2019, 9, 3718 12 of 15

Table 6. Simulation Motions references.

Experimenter Number of Motion Flexion Pronation Lap to Chin Simulate Level

Bad Side Good Side
E1–E10 10 0–30 0–30 0–1 5 ST2
E1–E10 10 30–60 30–60 1–2 5 ST3
E1–E10 10 60–90 60–90 2–3 5 ST4
E1–E10 10 90–120 90–120 2–4 5 ST5
E1–E10 10 120–150 120–150 3–4 5 ST6
E1–E10 10 150–180 150–180 5 5 Healthy

To meet experimental requirements, this paper recruited 10 students to simulate the actions shown
in Table 7 in six different scenarios. The simulation table is based on the “Ueda 12-stage paralysis
function” which divides the angle of action into six levels. To confirm the feasibility of the fuzzy
algorithm in this situation.

Table 7. Brunnstrom Level mapping table.

Range Brunnstrom Level

0% > Level < 17% ST2
17% ≥ Level < 34% ST3
34% ≥ Level < 51% ST4
51% ≥ Level < 68% ST5
68% ≥ Level < 85% ST6
85% ≥ Level < 100% Healthy

Table 7 shows the corresponding Brunnstrom Level for each level simulated in this experiment.
The Stage value calculated by the fuzzy system is mapped to Table 7.
Table 8 shows the simulation results for ST2, ST3, and ST4. Two of the results (E4, 20 %; E10,
23.5 %) in ST2 are classified as ST3, and the FMR score for E4 is slightly higher at 42.4 %. Since the
FASD parameter is lower than that of the other data, the FMR score is higher and is thus classified in
ST3. E10 has slightly higher FMR (38 %) and PMR (42.5 %) scores. Due to high action stability, the FMR
and PMR scores are slightly higher and are thus classified in ST3. In the results for ST3, E7 (34.3 %) is
classified in ST4, and the FMR of E7 is slightly over 50 %. As its FASD parameter is lower than that of
the other data, it is classified in ST4. In the ST4 results, E3 (24.7 %) is classified in ST3. The FMR and
PRM scores for E3 are slightly lower at 34.6 % and 32.6 %, and thus fall into ST3. Even if FAA has a
range that conforms to the reference table, the action parameters FASD, FLS, PASD and PVS all show
that E3 is insufficiently stable, and is thus determined by the system as ST3.

Table 8. ST2–ST4 Simulations results.

Level ST2 ST3 ST4

Experimenter FMR PMR LTCMR Level FMR PMR LTCMR Level FMR PMR LTCMR Level
E1 29.8% 10.3% 20.0% 15.8% 39.5% 41.1% 20.0% 18.8% 56.4% 51.2% 40.0% 41.0%
E2 31.0% 27.2% 0.0% 16.0% 37.3% 27.0% 40.0% 24.2% 48.0 50.8% 60.0% 47.2%
E3 27.5% 17.2% 20.0% 15.1% 36.7% 30.3% 40.0% 25.0% 34.6% 32.6% 40.0% 24.7%
E4 42.4% 22.1% 20.0% 20.0% 25.5% 45.5% 40.0% 24.8% 46.5% 53.0% 40.0% 35.7%
E5 27.7% 23.9% 20.0% 15.2% 36.1% 41.0% 40.0% 27.0% 58.0% 48.4% 40.0% 40.0%
E6 32.5% 33.5% 20.0% 16.6% 45.9% 32.6% 20.0% 21.1% 45.8% 54.8% 60.0% 49.6%
E7 21.0% 30.0% 0.0% 14.8% 50.0% 41.2% 40.0% 34.3% 49.1% 51.3% 40.0% 34.4%
E8 12.3% 24.6% 0.0% 10.0% 32.5% 34.0% 20.0% 17.1% 43.0% 48.8% 60.0% 40.8%
E9 24.0% 17.0% 20.0% 14.3% 44.0% 40.8% 20.0% 20.7% 49.7% 48.0% 60.0% 45.5%
E10 38.0% 42.5% 20.0% 23.5% 43.9% 30.8% 20.0% 27.5% 50.5% 46.7% 40.0% 34.5%

Table 9 shows the simulation results for ST5, ST6 and Healthy. It shows that E6 (60 %) is classified
as ST5, despite its FMR score of 37 % being similar to ST4 results but is correctly classified as ST5
Appl. Sci. 2019, 9, 3718 13 of 15

because its PMR score of 72.6 % is similar to that of ST6. In the ST6 results, the action assessment results
of all subjects match the reference table, thus all ten subjects are correctly classified as ST6. The Healthy
results of all subjects match the reference table, so all ten subjects are correctly classified as Healthy.

Table 9. ST5–ST6 and Healthy Simulations results.

Appl. Level
Sci. 2019, 9, x FOR PEER REVIEW
ST5 ST4 Healthy 13 of 15
Experimenter FMR PMR LTCMR Level FMR PMR LTCMR Level FMR PMR LTCMR Level
From
E1 the60.3%
experimental
57.1% results,65.3%
60.0% the errors
69.2% were
64.2%concentrated
80.0% between
78.9% 87.1% ST2 and 100.0%
93.1% ST4 in Table
93.3% 5
because
E2 the movement
65.6% 67.6% of the test subjects
40.0% 51.4% differs
63.7% from
62.0% that of actual
80.0% 77.3%patients,
71.7% corresponding
83.8% 100.0% to85.8% a low
E3 61.6% 49.1% 80.0% 65.2% 60.7% 75.5% 60.0% 69.8% 74.0% 91.3% 100.0% 89.6%
Brunnstrom
E4
level due
52.6% 53.0%
to low stability
60.0%
and
58.0%
angle of
62.1% 65.7%
action. There
60.0%
is
68.5%
no way to confirm
86.6% 80.1% 100.0%
the standard
84.2%
deviation,
E5 thus during
51.3% 56.7%simulations,
60.0% subjects
58.0% 62.1%trying
65.7% to 60.0%
mimic low 70.5% action
73.6%stability result in88.3%
85.2% 100.0% more
E6
exaggerated 37.0% 72.6% making
movements 80.0% it 60.0% 74.9%
difficult 71.0% a 60.0%
to obtain standard 70.5% 73.6%
for the action.85.2%Thus100.0% 87.5%
the resulting
E7 60.9% 54.6% 60.0% 62.5% 63.8% 74.1% 60.0% 69.4% 73.3% 79.2% 100.0% 85.3%
value E8
of the action
51.0% may
52.3%differ from 53.9%
60.0% the expected
62.2% result.
67.9% ST5–Healthy
80.0% 78.1% results
86.5% are closer
68.3% to the actions
100.0% 86.7%
of healthy
E9 subjects,
53.8% and
56.1%thus ST2–ST4
60.0% miscategorizations
61.8% 61.0% 68.3% are wavied.
60.0% 68.2% 72.3% 90.8% 100.0% 89.0%
E10 61.0% 63.6% 40.0% 52.1% 70.8% 63.8% 60.0% 68.3% 90.1% 82.5% 100.0% 91.3%
The experimental results show that the fuzzy algorithm can effectively distinguish the features
of different conditions and make correct classifications. Unlike Artificial Neural Networks (ANN) or
Figure
Support 11 shows
Vector Machinethat(SVM)
a comparison of thesystem
the proposed actual allows
experimental
physicalresults with to
therapists preset
makeresults. ST2
immediate
shows eight subjects
adjustments successfully
to accommodate theclassified, while two
actual condition areneeds
and classified as ST3. Nine
of hemiplegic subjectsAlthough
patients. are correctly
the
classified as ST3, while one is classified as ST4. Likewise, nine subjects are correctly
fuzzy algorithm used in this paper is a subjective and simple method, it does not require the classified as ST4,
and one is misclassified
construction of complexasmathematical
ST3. For ST5–Healthy,
models, and all 30 datarules
using results
of are correctly
thumb it canclassified.
be adjusted Thus, for
at any
the entire simulation, 56 results were correctly classified and four were misclassified,
time in response to actual conditions. As the determination method is based on the observation of for a success rate
of 93.3 %. data, errors are limited to ± 1 and thus will not significantly differ from actual results.
simulation

Figure 11.
Figure 11. Default
Default results
results comparison
comparison with actual results.
with actual results.

4. Conclusions
From the experimental results, the errors were concentrated between ST2 and ST4 in Table 5
because the movement of the test subjects differs from that of actual patients, corresponding to a
Based on fuzzy theory, this paper identifies features for discrete actions and evaluates the
low Brunnstrom level due to low stability and angle of action. There is no way to confirm the
corresponding Brunnstrom levels, providing an improvement over previous evaluation methods,
standard deviation, thus during simulations, subjects trying to mimic low action stability result in
reducing the degree of subjectivity and facilitating recording of longitudinal tracking data. Although
more exaggerated movements making it difficult to obtain a standard for the action. Thus the resulting
fuzzy theory provides a simple algorithm, it is more practical and simpler to use than other
value of the action may differ from the expected result. ST5–Healthy results are closer to the actions of
applications, such as SVM or ANN. In physical rehabilitation, the fuzzy algorithm is based on
healthy subjects, and thus ST2–ST4 miscategorizations are wavied.
personal experience, thus allowing the physical therapist to adjust the parameters in response to
The experimental results show that the fuzzy algorithm can effectively distinguish the features
differing requirements, such as changing the weight of action accuracy and stability.
of different conditions and make correct classifications. Unlike Artificial Neural Networks (ANN) or
Experimental results show that the acceleration data generated by the system can effectively
Support Vector Machine (SVM) the proposed system allows physical therapists to make immediate
filter out environmental noise by low-pass filtering, allowing the system to accurately capture feature
adjustments to accommodate the actual condition and needs of hemiplegic patients. Although the fuzzy
values. From observing the feature values, we can determine the degree of stability with which the
algorithm used in this paper is a subjective and simple method, it does not require the construction of
subject performs the action, and the numerical change for each action. The proposed BESE can
complex mathematical models, and using rules of thumb it can be adjusted at any time in response to
improve on traditional physical rehabilitation methods with effective classifying and storing in
databases. However, the present experiment used healthy subjects to simulate hemiplegia patients
and their performance may differ from actual patients, with implications for the fuzzy system
established here. For example, consider the simulated FLS parameter. During evaluation, the subject
would deliberately increase the amplitude of the action and increase the FLS value. However, actual
hemiplegic patients may not produce the same data. Verifying system validity and adjust the
Appl. Sci. 2019, 9, 3718 14 of 15

actual conditions. As the determination method is based on the observation of simulation data, errors
are limited to ± 1 and thus will not significantly differ from actual results.

4. Conclusions
Based on fuzzy theory, this paper identifies features for discrete actions and evaluates the
corresponding Brunnstrom levels, providing an improvement over previous evaluation methods,
reducing the degree of subjectivity and facilitating recording of longitudinal tracking data. Although
fuzzy theory provides a simple algorithm, it is more practical and simpler to use than other applications,
such as SVM or ANN. In physical rehabilitation, the fuzzy algorithm is based on personal experience,
thus allowing the physical therapist to adjust the parameters in response to differing requirements,
such as changing the weight of action accuracy and stability.
Experimental results show that the acceleration data generated by the system can effectively filter
out environmental noise by low-pass filtering, allowing the system to accurately capture feature values.
From observing the feature values, we can determine the degree of stability with which the subject
performs the action, and the numerical change for each action. The proposed BESE can improve on
traditional physical rehabilitation methods with effective classifying and storing in databases. However,
the present experiment used healthy subjects to simulate hemiplegia patients and their performance
may differ from actual patients, with implications for the fuzzy system established here. For example,
consider the simulated FLS parameter. During evaluation, the subject would deliberately increase
the amplitude of the action and increase the FLS value. However, actual hemiplegic patients may not
produce the same data. Verifying system validity and adjust the parameters of the simulated system
requires submission to an Institutional Review Board (IRB) to collect data from actual hemiplegia
patients, which would allow for the parameters to be modified to improve accuracy. This paper focused
mainly on upper limb action, and future work can extend this work to lower limbs and fingers.

Author Contributions: System organization/design, experiment supervisor, paper editing and submitting, T.-J.S.;
making experiments, paper editing, K.-L.L.; system design, paper editing, J.S.-H.T.; making experiments, W.-A.Y.;
figures and tables editing, C.-F.K.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Mackay, G.; Mensah, A.; Mendis, S.; Greenlund, K. The Atlas of Heart Disease and Stroke; World Health
Organization: Geneva, Switzerland, 2004.
2. Brunnstrom, S. Motor testing procedures in hemiplegia: Based on sequential recovery stages. Phys. Ther.
1966, 46, 357–375. [CrossRef]
3. Brunnstrom, S. Movement Therapy in Hemiplegia: A Neurophysiological Approach; Medical Dept., Harper and
Row: New York, NY, USA, 1970.
4. Sanford, J.; Moreland, J.; Swanson, L.R.; Stratford, P.W.; Gowland, C. Reliability of the Fugl-Meyer assessment
for testing motor performance in patients following stroke. Phys. Ther. 1993, 7, 447–454. [CrossRef]
5. Ueda, B.; Fukuya, Y.K. Standardization of hemiplegia functional test-method of 12 stages hemiplegia recovery
grade. Gen. Rehabil. 1977, 5, 749–766.
6. Wang, W.F.; Yang, C.Y.; Wang, D.Y. Analysis of Movement Effectiveness in Badminton Strokes with
Accelerometers. In Genetic and Evolutionary Computing. Advances in Intelligent Systems and Computing;
Springer: Berlin/Heidelberg, Germany, 2015.
7. Zhang, Z.; Liu, Y.; Li, A.; Wang, M. A novel method for user-defined human posture recognition using
Kinect. In Proceedings of the 7th International Congress on Image and Signal Processing, Dalian, China,
14–16 October 2014; pp. 736–740.
8. Wang, W.J.; Chang, J.W.; Haung, S.F.; Wang, R.J. Human Posture recognition based on images captured by
the Kinect sensor. Int. J. Adv. Robot. Syst. 2016, 13, 54–69. [CrossRef]
Appl. Sci. 2019, 9, 3718 15 of 15

9. Lee, G.X.; Low, K.S. A factorized quaternion approach to determine the arm motions using triaxial
accelerometers with anatomical and sensor constraints. IEEE Trans. Instrum. Meas. 2012, 61, 1793–1802.
[CrossRef]
10. Chen, J.R. Design and Implementation of Fall Detection and Triaxial Pedometer Based on Smartphone
Platform. Master’s Thesis, Department of Computer Science and Information Engineering, Tamkang
University, Taipei, Taiwan, 2014.
11. Kostopoulos, P.; Nunes, T.; Salvi, K.; Deriaz, M.; Torrent, J. Increased Fall Detection Accuracy in an
Accelerometer-based Algorithm Considering Residual Movement. In Proceedings of the International
Conference on Pattern Recognition Applications and Methods, Lisbon, Portugal, 10–12 January 2015;
pp. 30–36.
12. Howcroft, J.; Kofman, J.; Lemaire, E. Prospective Fall-Risk Prediction Models for Older Adults based on
Wearable Sensors. IEEE Trans. Neural Syst. Rehabil. Eng. 2017, 25, 1812–1820. [CrossRef]
13. Weng, W.X. Fall Detection Based on Angular Variation and Acceleration. Master’s Thesis, Department of
Computer Science and Information Engineering, Nation Dong Hwa University, Hualien, Taiwan, 2015.
14. Wang, Z.; Wu, D.; Gravina, R.; Fortino, G.; Jiang, Y.; Tang, K. Kernel fusion based extreme learning machine
for cross-location activity recognition. Inf. Fusion 2011, 37, 1–9. [CrossRef]
15. Yu, L.; Xiong, D.; Guo, L.; Wang, J.A. A remote quantitative Fugl-Meyer assessment framework for stroke
patients based on wearable sensor networks. Comput. Methods Programs Biomed. 2016, 128, 100–110.
[CrossRef]
16. Mannini, A.; Trojaniello, D.; Cereatti, A.; Sabatini, A.M. A Machine Learning Framework for Gait Classification
Using Inertial Sensors: Application to Elderly, Post-Stroke and Huntingtons Disease Patients. Sensors 2016,
16, 134. [CrossRef]
17. Zadeh, L.A. Fuzzy Sets. Inf. Control 1965, 8, 338–353. [CrossRef]
18. Paskaš, M.; Reljin, B.; Reljin, I.; Dujković, D. Edge preserved low-pass filtering controlled by local dimension.
In Proceedings of the 20th International Conference on Systems, Signals and Image Processing (IWSSIP),
Bucharest, Romania, 7–9 July 2013; pp. 87–90.
19. Mamdani, E.H.; Assilian, S. An experiment in linguistic synthesis with a fuzzy logic controller. Int. J.
Man-Mach. Stud. 1975, 7, 1–13. [CrossRef]
20. Lee, C.C. Fuzzy logic in control systems: fuzzy logic controller. II. IEEE Trans. Syst. Man Cybern. 1990,
20, 419–435. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).