Journal of Theoretical Biology 307 (2012) 160–167

Contents lists available at SciVerse ScienceDirect

Journal of Theoretical Biology

journal homepage: www.elsevier.com/locate/yjtbi

Do the chaotic features of gait change in Parkinson’s disease?

Yashar Sarbaz a,n, Farzad Towhidkhah b, Ayyoob Jafari c, Shahriar Gharibzadeh b
a
Electrical Engineering Faculty, Sahand University of Technology, Tabriz, Iran
b
Neuromuscular Systems Laboratory, Biomedical Engineering Faculty, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran
c
Biomedical Engineering Department, Islamic Azad University, Qazvin Branch, Qazvin, Iran

H I G H L I G H T S

c Showing the difference of chaotic behaviour in gait of normal and PD persons.

c Modelling the gait in normal and PD persons.
c Analysis of gait before and after using modelling tool to show the long time behaviour of gait.
c Showing that short time recording cannot be able to explain the behaviour of gait.

a r t i c l e i n f o a b s t r a c t

Article history: Some previous studies have focused on chaotic properties of Parkinson’s disease (PD). It seems that
Received 2 February 2012 considering PD from dynamical systems perspective is a relevant method that may lead to better
Received in revised form understanding of the disease. There is some ambiguity about chaotic nature in PD symptoms and
21 April 2012
normal behaviour. Some studies claim that normal gait has somehow a chaotic behaviour and disturbed
Accepted 25 April 2012
Available online 12 May 2012
gait in PD has decreased chaotic nature. However, it is worth noting that the basis of this idea is the
difference of fractal behaviour in gait of normal and PD patients, which is concluded from Long Range
Keywords: Correlation (LRC) indices. Our primary calculations show that a large number of normal persons and
Dynamical system patients have similar LRC. It seems that chaotic studies on PD need a different view. Because of short
Lyapunov exponent
time recording of symptoms, accurate calculation of chaotic features is tough. On the other hand, long
Stride time intervals
time recording of symptoms is experimentally difficult. In this research, we have first designed a
Model
physiologically plausible model for normal and PD gait. Then, after validating the model with neural
network classifier, we used the model for extracting long time simulation of stride in normal and PD
persons. These long time simulations were then used for calculating the chaotic features of gait.
According to change of phase space behaviour and alteration of three largest lyapunov exponents, it
was observed that simulated normal persons act as chaotic systems in stride production, but simulated
PD does not have chaotic dynamics and is stochastic. Based on our results, it may be claimed that
normal gait has chaotic nature which is disturbed in PD state. Surely, long time real recordings from
gait signal in normal persons and PD patients are necessary to warranty this hypothesis.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction for the disease is pharmacotherapy with L-Dopa [Factor and

Parkinson’s disease (PD) is the second wide-spread neurode-
generative progressive disease. The main cause of PD is the early
destruction of dopaminergic neurons in Substantia–Nigra Pars
compacta (SNC) of basal ganglia (BG). The main reason of this
destruction is not yet known, but it may be related to genetic
factors, environment, aging, and free radicals. The major treatment
n
Corresponding author. Tel.: þ98 914 314 4373.
E-mail addresses: yashar22c@gmail.com, yashar22c@yahoo.com (Y. Sarbaz).
Weiner, 2008].
The major symptoms of PD are tremor, rigidity, akinesia and
gait disturbance. Gait disorders are characterised by small, shuf-
fling steps resulting in reduced speed of walking. Gait disturbance
is progressive in all stages of the disease. This disturbance is
caused by muscle rigidity, bradykinesia, decreased force genera-
tion, abnormal rhythmicity, asymmetry of the left and right parts
of the body, and abnormal scaling of pace length. The gait
disorder in PD patients includes slowed gait, shortened length
of stride, decreased rhythm and cadence, increased time of double
support in the stance phase, shuffling and festinating gait,
decreased swing of the arms, and disturbed regulation of the

0022-5193/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jtbi.2012.04.032
 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167
stride length. 5 min walking can exhibit the disturbances in
patients. Gait analysis is usually used as a part of routine clinical
test for the assessment of walking performance. Some qualitative
tests such as up-and-go test and 180-degree U-turn test are used
for diagnosing the early PD [Factor and Weiner, 2008]. Since gait
disorders are progressive along the disease and their recordings
are simple, many researchers have focused on them.
Different studies have tried to analyse the PD quantitatively:
one group tried to model the PD, some studies wanted to analyse
the PDsymptoms quantitatively, and some of them studied the
disease with dynamical systems view. Some studies have focused
on chaotic properties of the disease but are to some extent
unclear and have postponed the concrete results to future works.
The main studies in these fields are as follows:
Sugiura et al. have recorded finger tapping signal for 15 s and
after putting these in phase space, calculated correlation dimen-
sion and showed that correlation dimension is globally bigger in
PD patients compared to normal subjects. It seems that the chosen
signal in this study cannot show the disturbances of the disease
well and other better signals like gait may have more convincing
results [Sugiura et al., 1998]. Timmer et al. have studied the tremor
of PD and compared it with the response of a random Vanderpol
oscillator and a chaotic Lorenz system and showed that the tremor
behaviour is similar to Vanderpol oscillator. However, they have
not investigated whether the normal physiological tremor has
chaotic or random behaviour [Timmer et al., 2000].Pascolo et al.
have evaluated time series of posture variations in normal and PD
patients, four case in each group. They showed that normal and PD
patients have some differences in chaotic features. They have
emphasised that more studies are needed to show these chaotic
differences [Pascolo et al., 2005].
Among PD symptoms, gait has been studied more with the
computational and fractal view. Since 1980s, quantitative differ-
ences of normal subjects and PD patients in gait were discussed
[Murray, 1979; Handford, 1986]. In 1990s, the stride length
variability was shown in PD patients [Blin et al., 1990; Zijlmans
et al., 1996; Vieregge et al., 1997]. Hausdorff et al. have observed a
fractal structure in gait and showed a stable long-range correla-
tion in stride time interval of normal human gait [Hausdorff et al.,
1995; 1996] and claimed that the fractal structure is destroyed by
neurodegenerative diseases [Goldberger et al., 2002]. In 1998,
Hausdorff et al. used stride time and length variability to separate
PD, Huntigton’ disease and normal persons [Hausdorff et al.,
1998]. Sekine et al. showed that the fractal dimension differs in
the normal, PD, and elderly persons. More dimensions were
observed in PD patients than elder and young healthy subjects
[Sekine et al., 2002]. In 2006 Liao and Wang,. studied the
similarity of stance and stride times between left and right legs
and showed that the symmetry is decreased in PD patients [Liao
and Wang, 2006]. Jeon et al. studied the classification of PD and
healthy persons using spatial-temporal image of plantar pressure.
They also used support vector machine classifier by kernel
function and achieved 91.73% accuracy [Jeon et al., 2008]. In
2009, Henmi et al. presented a spectral analysis of stride signal.
They claimed that power spectra may be useful for analysing PD
[Henmi et al., 2009]. Aziz and Arif studied stride rate variability.
They used two features called ‘‘threshold based acceleration
change index’’ and ‘‘symbolic entropy complexity measure nor-
malised corrected Shannon entropy’’ to calculate the differences
between the healthy and neurodegenerative diseases [Aziz and
Arif, 2006]. The feature values were almost equal for different
groups. In 2009, Hausdorff reviewed fractal features of gait in PD
patients and its relation with gait, stride length, and gait varia-
bility [Hausdorff, 2009].
Besides the analysis of PD symptoms (e.g., gait), modelling
approach is also used for PD evaluation and some of them presented
161
good findings or hypotheses. 1n 1999, Edwards et al. presented an
artificial neural network model with a parameter for attenuating the
connections of neurons to simulate the decreased dopamine level in
PD [Edwards et al., 1999]. In 2005, Haeri et al. focused on BG
structure and presented a mathematical model for tremor. While
being a simple model and accepting some assumptions as consider-
ing the tremor to be simple sinusoidal signal, the role of drugs and
DBS treatments were simulated fairly suitable and clinically plau-
sible [Haeri et al., 2005]. In 2004, Terman and Rubin presented a
neural network containing four substructures of basal ganglia. They
postulated a basic model based on neuronal conduction relations
and compared normal, PD patients without treatment, and PD
patients with DBS treatment [Rubin and Terman, 2004]. In 2008
MashhadiMalek et al. presented a model of BG structures based on
CPG in each block. They showed that rigidity and tremor are
correlated. However, the presence of oscillations in the BG internal
parts contradicts this hypothesis [MashhadiMalek et al., 2008].
Based on CPG theory, some simple models are introduced for PD
gait. These are usually black box models and designed for normal
persons, not for PD patients [Ashkenazy et al., 2002; West and Latka,
2005; West and Scafetta, 2003].
According to the presented studies, it seems that approaching
to PD with dynamical systems view is a relevant method and may
lead to better understanding of the disease. However, there is
some ambiguity about chaotic nature in PD symptoms and
normal behaviour. Some studies claim that the PD symptoms
have chaotic nature and others contradict it; yet there is no
agreement on this point. There is more or less an agreement on
the existence of chaotic nature in normal gait and its disturbance
in PD states. However, it is worth noting that the basis of this idea
is the difference of fractal behaviour in gait of normal and PD
patients, which is concluded from Long Range Correlation (LRC)
indices. Our primary calculations that will be presented in results
section, show that a large number of normal persons and patients
have similar LRC. Therefore, by calculating LRC, we cannot
separate the PD and normal persons with high accuracy.
It seems that chaotic studies on PD need a different view.
Because of short time recording of symptoms, accurate calcula-
tion of chaotic features is tough. On the other hand, long time
recording of symptoms is tough and even impossible because
of patient’s incompliance, muscle fatigue, and environmental
changes during the test. Therefore, designing a perfect model of
PD symptoms may be a good alternative for solving this problem.
In this research, we have first designed a physiologically plau-
sible model for normal and PD gait. Then, after validating the model,
we use the model for extracting long time simulation of symptoms.
This long time simulation will be used for calculating the chaotic
features of gait.
2. Material and method
2.1. Clinical data
We used the data presented in www.physionet.org [Hausdorff
et al., 1998]. This database includes 14 PD patients and 16 healthy
persons (controls). In this database, there are time intervals of
stride, swing, and stand for both legs. Objects were asked to walk
5 min in a 77 m direct path. Patients did not show falling or
freezing of gait (FOG). Normal subjects had no previous neural
disease or gait disorder. The gait data of the first 20 s were deleted
in order to omit the effect of movement beginning. For measuring
time intervals, force sensors were used in the plantar portion
of the foot. The force signal was sampled with the frequency
of 300 Hz.
162 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167

2.2. Model
BG are involved in motor control. When the person decides
to do a certain movement, this willing is converted to motor
commands, mainly by cerebellum, BG and cortex. Then these
commands pass from spinal cord and reach the motor end plate.
Meanwhile, BG is responsible for regulating the quality of motion.
It is suggested that BG has key roles in timing, initiation of
voluntary movement, controlling the speed and acceleration
of movement, regulation of muscle tone, etc. BG receive their
input from cortex and give their output through thalamus to
supplementary motor area (SMA).
The relations among internal blocks of BG and their inhibitory
or excitatory nature as well as the kind of neurotransmitters are
depicted in Fig. 1 [Kandel et al., 2000]. This model have been
explained in more details in Sarbaz et al., 2011. As it is shown, BG
inputs and outputs are related to each other by direct and indirect
pathways. These two pathways are controlled by dopamine signal
via the modulatory effect of SNc. The direct pathway has a simple
and fast processing on the BG input, but the indirect pathway
fulfils more complex processing on the BG input. The balance
between these two pathways is regulated by SNc. In PD, because
SNc is destructed, dopamine output is reduced and the balance
between the two pathways is disrupted. Therefore, the symptoms
of the disease appear.
In our presented model, the roles of SNc and dopamine signal
are in the core of attention, because of their importance in PD.
Elman structure is considered for this part of BG (because of its
feedback, similar to SNc modulatory effect). Elman network is
powerful for producing and modelling of periodic and semi
periodic signals. In our model, since the final output is stride
time intervals, we are prone to semi periodic signals. Therefore,
Elman network is useful for producing the final output signal. For
modelling the other parts of BG, we used feed-forward neural
networks to produce a physiologically plausible model. Therefore,
we finally used an Elman network with one layer and a feed-
forward network with two layers. Inhibitory or excitatory beha-
viour of layers were implemented by negative or positive weights,
respectively. The role of dopamine was considered in the feed-
back of first layer (Elman network). The input of the model
(cortex input to BG) was chosen as a constant signal which was
not different between normal and disease states, because the
input of cortex to BG is intact in PD patients. However, the
parameters of the model (BG) are changed in PD in comparison
with normal persons. The model structure is shown in Fig. 2.
In our model, the numbers of neurons are 20, 20, and 10 in the
first, second and third layers, respectively. The nonlinear func-
tions of all the layers are sigmoid. The third layer output of the
network passes through a pure line function and produces the
final network output. There were 20 inputs to the system, which

Fig. 1. The structure of human basal ganglia [Kandel et al., 2000].

Fig. 2. A schematic representation of the presented model.

 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167 163

were supposed to be constant in healthy and PD state. Since the Table 1

nature of this signal is complicated and unclear, we used a Results of classifier with real recorded data.
random 20 input time series as an initial assumption. Model
Classifier With 70% With remaining 30%
was trained in healthy and PD states. The weights of the network training signals (%) testing signals (%)
were obtained and this was taken as the base of our model.
Accuracy 100 89
Specificity 100 83
2.3. Classifier
Sensitivity 100 100

Gate is a complex signal and its time wave form does not show
the differences between the gait of PD and normal persons. On the
other hand, our vision to PD is dynamical systems view, and Table 2
changes in the parameters of dynamical systems lead to change in Mean of three largest Lyapunov exponents in real recorded data.
gait states. Hence, for comparing the gait of different persons and
First LE Second LE Third LE
diagnosing its membership to normal or PD groups, it is necessary
to extract proper features of gait. Then, a good classifier is needed Health record  0.2668 70.0425  0.8501 7 0.0702  1.0236 7 0.2058
to determine the states of gait (normal and PD states) based on Parkinson record  0.2621 70.0397  0.6771 7 0.1463  0.8519 7 0.4825
stride features. So in our study, after extracting the proper
features, a classifier was designed to separate gait behaviour in
PD and normal groups. This classifier was also used to validate the
results of our gait model. The classifier is a simple feed forward
neural network with two hidden layers.

2.4. Features of strides

in order to find proper features of stride, we studied the

behaviour of stride signals. The previous studies of the researchers
usually have focused on statistical analysis of mean and variance.
Significant differences were usually seen between the mean and
variance of stride in normal and patient cases [Hausdorff et al.,
1998]. Therefore, we used these two features in our study. Because
of the semi-periodic behaviour of gait, and therefore stride, power
spectra studies were also done on the cases. These studies show
that normal persons have more regular behaviours and their
spectra have high energy in definite ranges. However, the patients
had irregular behaviours and their spectra were distributed irre-
gularly in all of the range. Therefore, the features extracted from
spectral analysis may be useful for making the model response
similar to the real signal. For this purpose, the power spectra range
in each recording was divided into 5 equal parts. The amount of
energy in each part (the area under the curve) was calculated and
chosen as the features of the classifier.
In the disease situation, the state of the person is changed with
respect to the normal conditions. Chaotic features such as fractal
dimension are proper for showing the state changes. Therefore,
we used these features in our study. One method for its comput-
ing is Petrossian’s Algorithm. Petrossian uses a quick estimate of
the Fractal Dimension, which can be done by extracting a time
series from the main signal with different methods. Then this
relation is used [Petrossian, 1998]:
log10 n
D¼
log10 n þ log10 ðn=n þ 0:4ND Þ
where N is the length of the sequence (number of points), and ND
is the number of sign changes (number of dissimilar pairs) in the
binary sequence generated.
The chosen features for stride signal included mean and
variance of the strides, Petrossian Dimension of strides and power
spectra features.
3. Results
Analysis of the recorded signals: at first, we tried to design a
proper classifier on the basis of features extracted from the stride
signal. We trained the neural network classifier with two hidden
layers, with 70% of dataset. Then, we tested our classifier with the
Fig. 3. LRC for the stride of normal cases (black points) and PD patients (white
points).
remainder 30% of dataset. The accuracy, specificity, and sensitiv-
ity of the test and train are shown in Table 1, which show that the
trained classifier has an acceptable accuracy for real recorded
data. Then, we calculated 3 Largest Lyapunov Exponents (LLE)
using the method presented in Shintani and Linton, 2004. This
method was tested on some benchmark systems such as Logistic
map, Lorenz system and Henon map. The mean of these LLEs in
PD and normal persons are shown in Table 2. LLEs can be used as
powerful tools to determine the existence of chaotic behaviour in
the stride time intervals. Since different researches have studies
the known feature ‘‘LRC’’, we plotted its value in normal and PD
persons in Fig. 3, in order to see whether it is a suitable criterion
for separating normal and PD persons and whether it can show
fractal behaviour changes from normal to PD persons. Although
the data were obtained from physionet, LRC values for used data
sets were not reported in prior publications to give us a compar-
ison measure. As shown in Fig. 4, LRC is not good measure for
separation of our dataset. This was expectable, because LRC
measure is not designed for such analyses.
Training the model: the abovementioned model was trained for
normal and PD persons. The recorded signal for a healthy person
and the response of the trained model for the same person are
shown in Fig. 4. Similarly, the recorded signal for a patient and the
response of the model for the same patient are shown in Fig. 5.
164 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167

Fig. 4. Real recorded data and the response of the model for a healthy case.

Fig. 5. Real recorded data and the response of the model for a Parkinsonian case.

Because of the high power of the model and its proper ability in
training the signals, the response of the model has little error.
Validating the model: we let the weights of our model to
change in a 10% range around their initial values. This work was
done 160 times for normal and 160 times for PD models and the
outputs were saved. These simulated signals were divided into
train and test groups. Then, we used 200 signals (100 healthy and
100 PD simulated signals) for training a new feed forward ANN
with two hidden layers. Remaining 120 signals were used for test
and the error was lower than 0.01. This very low error shows that
the model is producing two different states. For showing whether
these two are normal and PD states, we used the previous
classifier (which was trained with the recorded real data). All
320 simulated signals were entered to this classifier. It was able to
diagnose them truly with an error of less than 0.005. This shows
that the two different states are normal and PD groups.
Analysing the simulated signals: based on the validity of the
model, we used the model to produce long time simulated signals.
These signals can be used to configure the attractors in phase
space. Please see [Jafari and Almasganj, 2010] for more details on
embedding approach from a 1-dimensional signal to multidimen-
sional phase space.At first, we study the behaviour of normal and
PD cases in phase space of real data (Fig. 6). A similar phase space
was studied for the results of the model in a long time simulation
(6 times the length of the real data). Fig. 7 depicts the simulated
long time response for a normal person and its phase space, and
Fig. 8 shows the simulated long time response of PD patient and
its phase space.
4. Discussion
Some researchers have studied the PD with dynamical systems
view. Some of them have tried to investigate the disease beha-
viour on the basis of chaos theory. These researchers have studied
various symptoms of PD like tremor, gait disorders, and postural
instability and have reported more or less different results. Some
other researchers had not report a definite conclusion about
chaotic nature of gait and have claimed that for determining the
behaviour of normal and PD persons, more studies are necessary
 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167 165

Fig. 6. Phase space of healthy and normal cases from real recorded data.

Fig. 7. Phase space of simulated of healthy person.

166 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167
Fig. 8. Phase space for simulate Parkinsonian person.
Table 3
Mean of three largest Lyapunov exponents in simulated data.
First LE Second LE Third LE
Healthy output þ0.019 7 0.0372  0.0484 7 0.0629  0.1945 70.0527
Parkinson output  0.3522 7 0.0625  0.8490 7 0.2053  0.8821 70.3802
[Pascolo et al., 2005; Sugiura et al., 1998]. Based on previous
studies and the results of present study, we believe that more
evidences show that fractal behaviour is seen in gait of normal
persons, but it is reduced in PD gait. The change of LRC behaviour
has been reported in some investigations which can be a clue of the
fractal behavioural changes in PD stride. However, our initial study
on chaotic features of gait shows that a large number of PD and
normal persons have similar chaotic features (Table 1 and Fig. 3).
It seems that the main problem in analysing the chaotic nature
of PD is that in different studies on PD, short time recordings are
used for feature calculation. For example, it is usual in gait
analysis to execute 5 min recording. On the other hand, long term
recording is practically tough or even impossible. Limited coop-
eration as well as fatigue and disability of subjects do not let the
researchers to record such a long time. Also, it is not acceptable to
disconnect the recording and repeat it in another time to obtain
more clinical data. Therefore, an inexact estimation of chaotic
feature will be extracted in most studies. For solving this problem,
it seems useful to present a suitable model of gait which is able to
produce long time stride data without fatigue.
In this study, we initially hypothesised that fractal structure
exists in normal subjects, but is defective in PD patients. The
presence of fractal structure has been suggested in different parts
of the normal body as arteries, heart, and brain neurons. The
presence of these fractal structures leads to chaotic behaviours of
normal persons. This behaviour will change in disease states.
Different studies have tried to track this hypothesis in stride time
intervals and have been partially successful in this regard. In this
study, based on our physiological model of gait, we tried to
calculate the chaotic criteria (LLE) exactly. Our results show that
the normal subjects have chaotic stride time intervals which are
reduced in PD patients.
Chaos theory states that systems which have chaotic beha-
viour must at least show a positive, a near zero, and a negative
Lyapunov exponent. As it is shown in Table 3, this behaviour is
seen obviously in simulated normal persons but it is not seen in
simulated PD patients. This may indicate that stride is chaotic in
normal persons, but this chaotic nature changed towards a
stochastic behaviour in PD states. Phase space graphs (Figs. 6–8)
confirm our hypothesis. The difference between the phase space
of normal and PD patients, when studied on short time real
recorded signals, is not significant and stochastic-like behaviour is
seen in both of them. Evaluating the response of a trained model
for a long period of time shows that no special change in PD
patients and their stochastic-like behaviour is repeated; however,
in phase space of normal person, an attractor appears gradually.
This shows that stride production in normal persons obeys a
specific pattern. This pattern is disturbed in PD state. Now,
according to change of phase space behaviour from normal to
 Y. Sarbaz et al. / Journal of Theoretical Biology 307 (2012) 160–167
PD state and alteration of 3 largest lyapunov exponents, it can be
claimed that a normal person acts as a chaotic system in stride
production; however, because of alteration of body fractal struc-
ture (human brain) in PD, chaotic dynamics is changed and
stochastic behaviour is seen in stride production. Surely real long
time recordings in PD patients are needed to prove our hypoth-
esis. It seems that paying attention to this new idea may help the
clinicians to interpret different gait behaviours of PD patients.
This idea that gait has chaotic nature in normal persons and
stochastic behaviour in PD patients can be used in understanding
the origin of disease symptoms as well as the mechanism of
different effective therapies. For example, if we focus on the
chaotic nature of physiological gait, we may be able to diagnose
PD based on a simple gait recording. As another example, it is
worth noting that although deep brain stimulation (DBS) is used
as an effective route of treatment for nearly two decades, its
mechanism is yet unknown. It is usual in chaotic systems that
adding an external frequency can alter the behaviour of the
system. Hence, we believe that chaotic view on gait may elucidate
new horizons for understanding such complex treatments.
References
Ashkenazy, Y., Hausdorff, J.A., Ivanov, P.C., Stanley, H.E., 2002. A stochastic model
of human gait dynamics. Physica A 316, 662–670.
Aziz, W., Arif, M., 2006. Complexity analysis of stride interval time series by
threshold dependent symbolic entropy. Eur. J. Appl. Physiol. 98 (1), 30–40. Jul 14.
Blin, O., Ferrandez, A.M., Serratrice, G., 1990. Quantitative analysis of gait in
Parkinson patients: Increased variability of stride length. J. Neurosci. 98 (1),
91–97.
Edwards, R., Beuter, A., Glass, L., 1999. Parkinsonian tremor and simplification in
network dynamics. Bull. Math. Biol. 61 (1), 157–177.
Factor, S.A., Weiner, W.J., 2008. Parkinson’s disease Diagnosis and Clinical Manage-
ment, Second edition Demos Medical Publishing, New York.
Goldberger, A.L., Amaral, L.A., Hausdorff, J.M., Ivanov, P.C., Peng, C.K., Stanley, H.E.,
2002. Fractal dynamics in physiology: alterations with disease and aging.
PNAS 99 (1), 2466–2472, http://dx.doi.org/10.1073/pnas.012579499.
Haeri, M., Sarbaz, Y., Gharibzadeh, Sh, 2005. Modeling the Parkinson’s tremor and
its treatments. J. Theor. Biol. 236, 311–322.
Handford, F., 1986. Parkinsonian gait. Physiotherapy 72 (7), 341–342.
Hausdorff, J.M., 2009. Gait dynamics in Parkinson’s disease: common and distinct
behaviour among stride length, gait variability and fractal-like scaling. Chaos
19 (2). art.No. 026113.
Hausdorff, J.M., Cudkowicz, M.E., Firtion, R., Wei, J.Y., Goldberger, A.L., 1998. Gait
variability and basal ganglia disorders: stride-to-stride variations of gait cycle
timing in Parkinson’s disease and Huntington’s disease. Movement Disorders
13 (3), 428–437, http://dx.doi.org/10.1002/mds.870130310.
Hausdorff, J.M., Peng, C.K., Ladin, Z., Wei, J.Y., Goldberger, A.L., 1995. Is walking a
random walk? Evidence for long-range correlations in stride interval of human
gait. J. Appl. Physiol. 78, 349–358.
167
Hausdorff, J.M., Purdon, P.L., Peng, C.K., Ladin, Z., Wei, J.Y., Goldberger, A.L., 1996.
Fractal dynamics of human gait: stability of long-range correlations in stride
interval fluctuations. J. Appl. Physiol. 80, 1448–1457.
Henmi, O., Shiba, Y., Saito, T., Tsuruta, H., Takeuchi, A., Shirataka, M., Obuchi, S.,
Kojima, M., Ikeda, N., 2009. Spectral analysis of gait variability of strides interval
time series: comparison of young, elderly and Parkinson’s disease patients.
J. Phys. Ther. Sci. 21 (2), 105–111, http://dx.doi.org/10.1589/jpts.21.105.
Jafari, A., Almasganj, F., [2010] Statistical Modeling of Speech Poincaré Sections in
Combination of Frequency Analysis to Improve Speech Recognition Perfor-
mance. Chaos (Woodbury, N.Y.). 09/2010; 20(3):033106.
Jeon, H.S., Han, J., Yi, W.J., Jeon, B.S., Park, K.S., 2008. Classification of Parkinson
Gait and Normal Gait using Spatial-Temporal Image of Plantar Pressure. 30th
Annual International IEEE EMBS Conference Vancouver, British Columbia,
Canada, August 20–24.
Kandel, E.R., Schwartz, J.H., jessell, T.M., 2000. ‘‘Principles of Neural Sciences’’, 4th
ed. McGraw-Hill, New York.
Liao, F., Wang, J., 2006. Effect of Parkinson’s disease on gait symmetry. J. Xi’an
Jiaotong Univ. 40 (2)., pp. 228-230 þ238.
MashhadiMalek, M., Towhidkhah, F., Gharibzadeh, Sh., Daeichin, V., Ahmadi-
Pajouh, M.A., 2008. Are rigidity and tremor two sides of the same coin in
Parkinson’s disease? Comput. Biol. Med. 38, 1133–1139.
Murray, M.P., 1979. Studies of normal and abnormal locomotion. Int. J. Rehabil.
Res. 2 (4), 510–511.
Pascolo, P.B., Marini, A., Carniel, R., Barazza, F., 2005. Posture as a chaotic system
and an application to the Parkinson’s disease. Chaos, Solitons Fractals 24,
1343–1346.
Petrossian, A., 1998. Kolmogorov complexity of finite sequences and recognition of
different preictal EEG patterns, in: Proceedings of the IEEE Symposium
Computer-Based Medical System, pp. 212–217.
Rubin, E., Terman, D., 2004. High frequency stimulation of the subthalamic nucleus
eliminates pathological thalamic rhythmicity in a computational model.
J. Comput. Neurosci. 16, 211–235.
Sarbaz, Y., Gharibzadeh, S., Towhidkhah, F., Jafari, A., 2011. A gray box neural
netowrk model of Parkinson disease using gait signal. Spring 2011 2 (3),
33–42.
Sekine, M., Tamura, T., Akay, M., Fujimoto, T., Togawa, T., Fukui, Y., 2002. Discrimina-
tion of walking patterns using wavelet-based fractal analysis. IEEE Trans. Neural
Syst. Rehabil. Eng. 10 (3)http://dx.doi.org/10.1109/TNSRE.2002.802879.
Shintani, M., Linton, O., 2004. Nonparametric neural network estimation of
Lyapunovexponents and a direct test for chaos. J. Econometrics 120, 1–33.
Sugiura, T., Sugiura, N., Sugiyama, K., Yokoyama, T., [1998]. Chaotic approach to
the quantitative analysis of Parkinson’s disease. Proceedings of the 20th
AnnualInternational Conference of the IEEE Engineering in Medicine and
Biology Society.
Timmer, J., Ha€ ussler, S., Lauk, M., Lucking, C.H., 2000. Pathological tremors:
deterministic chaos or nonlinear stochastic oscillators? Chaos 10 (1).
Vieregge, P., Stolze, H., Klein, C., Heberlein, I., 1997. Gait quantitation in Parkinson’s
disease—Locomotor disability and correlation to clinical rating scales. J. Neural
Transm. 104 (2–3), 237–248, http://dx.doi.org/10.1016/S0065-2458(08)60272-7.
West, B.J., Latka, M., 2005. Fractional Langevin model of gait variability.
J. Neuroeng. Rehabil. 2, 24.
West, B.J., Scafetta, N., 2003. Nonlinear dynamical model of human gait. Phys. Rev.
67, 051917.
Zijlmans, J.C.M., Poels, P.J.E., Duysens, J., Straaten, J., Thien, T., Van’t Hof, M.A.,
Thijssen, H.O.M., Horstink, M.W.I.M., 1996. Quantitative gait analysis in
patients with vascular Parkinsonism. Movement Disorders 11 (5), 501–508,
http://dx.doi.org/10.1002/mds.870110505.