Statistical Methods and Artificial Neural Networks Techniques in Electromyography

Abstract

This work represents a comparative study for the activity of the masseter muscle for patients before trial base denture insertion and the activity of the same muscle after trial denture base insertion for both right and left masseter muscles. The study tried to find if there were significant differences in the activity of the masseter muscle before and after patients wearing their trial denture base using two approaches: parametric statistical methods and a Neural Network Classifier. Statistical analysis was performed on three feature vectors extracted from autoregressive AR modeling, Discrete Wavelet Transform WT, and from Wavelet Packet Transform WP. The least significant difference test and the student t-test have not proved significant differences in the masseter muscle activity before and after wearing denture. However, using the same feature vectors, a neural network classifier has proved that there are significant differences in the masseter muscle activity before and after patients wearing trial denture base.

Full text

International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012 39
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Keywords: Dentistry,DiscreteWaveletTransform(DWT),ElectrmyographyEMG,FeatureExtraction,
NeuralNetworkClassication,WaveletPacketTransform(WP)
1. INTRODUCTION
Structural reorganization of the motoer unit,
the smallest functional unit of muscle, takes
place because of disorders affecting peripheral
nerve and muscle. Motor unit morphology can
be studied by recording its electrical activity.
The procedure is known as clinical electrmyog-
raphy (EMG). In clinical EMG motor unit ac-
tion potentials (MUAP’s) are recorded using a
needle electrode at mild voluntary contraction.
The MUAP reflects the electrical activity of a
single anatomical motor unit. It represents the
compound action potential of those muscle
fibers within the recording range of the elec-
trode (Kallenberg & Hermens, 2006). Features
of MUAP’s extracted in the time domain such
as duration, amplitude, and phases proved to
be very valuable in differentiating between
muscle and nerve diseases with the duration
measure being the key parameter used in
clinical practice. However, the measurement
of the duration parameter is a difficult task
depending on the neurophysiologist and/or the
computer aided method used. The definitions
of widely accepted criteria that will allow the
Statistical Methods and Articial
Neural Networks Techniques
in Electromyography
AhmadTaherAzar,ModernScienceandArtsUniversity,Egypt
ValentinaE.Balas,AurelVlaicuUniversityofArad,Romania
ABSTRACT
Thisworkrepresentsacomparativestudyfortheactivityofthemassetermuscleforpatientsbeforetrialbase
dentureinsertionandtheactivityofthesamemuscleaftertrialdenturebaseinsertionforbothrightandleft
massetermuscles.Thestudytriedtondifthereweresignicantdifferencesintheactivityofthemasseter
musclebeforeandafterpatientswearingtheirtrialdenturebaseusingtwoapproaches:parametricstatistical
methodsandaNeuralNetworkClassier.Statisticalanalysiswasperformedonthreefeaturevectorsextracted
fromautoregressive(AR)modeling,DiscreteWaveletTransform(WT),andfromWaveletPacketTransform
(WP).Theleastsignicantdifferencetestandthestudentt-testhavenotprovedsignicantdifferencesinthe
massetermuscleactivitybeforeandafterwearingdenture.However,usingthesamefeaturevectors,aneural
networkclassierhasprovedthattherearesignicantdifferencesinthemassetermuscleactivitybeforeand
afterpatientswearingtrialdenturebase.
DOI: 10.4018/ijsda.2012010103

40 International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
computer-aided measurement of this parameter
are still lacking (Calder et al., 2008; Dimitrova
& Dimitrov, 2003).
On the other hand, frequency domain
features of MUAP’s like mean, or median fre-
quency, bandwidth, and quality factor provide
additional information in the assessment of
neuromuscular disorders and it has recently
been shown that the discriminative power
of the MUAP mean or median frequency is
comparable to the duration measure or spike
duration measure (Pfeiffer & Kunze, 1993).
However, it will be possible to rely on the
evaluation of a spectral parameter only if each
power spectrum estimate does not suffer from
large mean square error (MSE), otherwise, the
estimates may mislead us in our understanding
of the physiology.
1.1. Problem Definition
Clinical measurements of the vertical dimension
of occlusion often rely on the determination of
mandibular rest position. Recent research efforts
have shown that the postural position is a range
of positions rather than a single and absolute one
(Michelloti et al., 1997). Some authors insist
that interocclusal vertical dimension is highly
variable throughout life (Amorim et al., 2010).
It is therefore not surprising that many proce-
dures for its assessment have been the subject
of many investigations (Krivickas et al., 1998).
In general interocclusal distance at the clinical
rest position is less than introcclusal distance at
the electromyographically silent (physiologic)
rest position of the mandible. Variations in
the interocclosal distance at the physiological
rest position of the mandible determined by
masticatory electromyographic silency are ap-
parently due to emotional stress during clinical
trials and/or the biologic incompatibility of the
instrumentation used to measure mandibular
movement. The question remaining is-do the
jaw muscle plays any part either continuously
or intermittently, in determining mandibular
rest position? Numerous attempts have been
made to record activity from the jaw muscles
of relaxed human subjects, the majority using
surface electromyographic electrodes (Wang et
al., 2007). Some reports claim to demonstrate
continuous activity, particularly of the temporal
muscle, others report intermittent activity or
no activity.
However, in general the methods employed
provided an inadequate basis for objective
conclusion. A methodical approach of study of
intra-muscular activation patterns is provided
by the spectral analysis EMG (Zuccolotto et
al., 2007). However, the Fourier methods often
used in the analysis of EMG usually suffer
from several limitations. One of these is the
underlying assumption that the original signal
is stationary (Karlsson et al., 1999). As a result
Fourier Methods are not generally appropriate
for the analysis of EMG signals. Kihwan and
Minamitani (Barisci, 2008) reported the validity
of the autoregressive (AR) model for surface
EMG. However, the main problem with the
parametric modeling methods is the selection of
the model order (Karlsson et al., 1999). Recently,
time-scale methods (wavelet transforms) were
proposed in an effort to overcome the limitations
of the traditional time-frequency methods (Go-
swami & Chan, 1999). The time-scale methods
act as a mathematical microscope in which one
can observe different parts of the signal by just
adjusting the focus. This allows the detection
of short-lived time components of signals. As a
generalization of the wavelet transform (WT),
the wavelet packet has been recently introduced
and developed, which allows a best adapted
analysis of the signal.
Wavelets were introduced as a signal repre-
sentation in which an analysis window, whose
size is chosen to be short at high frequencies
and long at low frequencies (to pick up all the
abrupt changes), is passed through the signal.
This corresponds to having the frequency
response logarithmically scaled along the
frequency axis, as opposed to the short–time
Fourier transform (STFT) or Gabor represen-
tation (Feichtinger & Strohmer, 1997). Many
researchers have reported the application of
wavelets to the surface EMG (Salvador & de
Bruin, 2006; Berger et al., 2006). In this paper,
we propose a new approach for determining the
physiologic rest position of the mandible. The

International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012 41
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
rest positions were compared to one another with
respect to trial denture base presence or absence
during testing. The approach is based on WT
and WP. The main objective is to investigate
the performance of the WT and WP methods
and compare them to the AR coefficients. Two
methods were utilized to detect if there were
significant differences in the masseter muscle
activity before and after wearing trial denture
base. These are parametric statistical tests and
a neural network classifier.
2. DATA ACQUISITION
Twenty-four patients (fourteen males and ten
females) were selected among the completely
edentulous patients for complete denture con-
struction. Maxillary and mandibular acrylic
trial denture base were constructed for every
patient and tested for adequate retention (Gross
& Ormianer, 1994; Sato et al., 1998). The sur-
face electromyographic activity of the masseter
muscle before and after wearing trial denture
base was recorded bilaterally by bipolar surface
electrodes for 30 seconds. EMG signals were
recorded bilaterally using Biopac acquisition
system which consists of the MP100 data ac-
quisition unit, an EMG module, and an Apple
Macintosh computer. Electromyography was
made while the patient was setting upright in
a chair, his head unsupported, and he was told
to relax, looking straight forward and swallow-
ing of 5 ml of water repeated four times. Two
electrodes were placed over the middle belly of
the right or left masseter muscles. A common
ground electrode was attached to the forehead
of the patient. The signal output of the MP100
is a digital signal and the sampling rate is 2
kHz. The digital output was obtained using the
Acknowledge II software and stored in a com-
puter file for further analysis. Figure 1 shows
a typical example of the acquired EMG signal.
3. FEATURE EXTRACTION
3.1. Autoregressive Modeling
Before applying any model to EMG signals, the
stationarity of the signal was examined. This
test was used to determine if there were any
trends present in the data, such as increasing
or decreasing amplitudes. The amplitude of the
signal was tested using a “Run Test” (Bendat &
Piersol, 2010). It has been found that the data was
non-stationary for 5% level of significance and a
sample size of 100 samples. Nonstationarity has
been removed using best fit curves of variable
order according to the different EMG segments.
In the autoregressive (AR) model each sample
x(n) of the EMG signal is described as a linear
combination of previous samples plus an error
term e(n) which is independent of past samples
(Dhanjal, 1997; Vaidyanathan, 2008).
Figure1.AtypicalexampleofEMGsignals

42 International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
x n a k x n k e n
k
p
( ) ( ) ( ) ( )= − − +
=
∑
1
(1)
Where x(n) are samples of the modeled signal,
a(k) are AR coefficients, e(n) is residual or error
sequence, and p is the model’s order.
In order to determine the order of the au-
toregressive model, the power of the residual
signal (difference between the original signal
and the AR model) was calculated and compared
with that of the original EMG signal. The ratio
which was less or equal to 0.05 was accepted.
It has been found that the optimal model order
which gave minimum ratio between residual
power and original signal power was 7. To
confirm the whiteness of the residual signal,
the power was tested using a Kolmogorov-
Smirnov one sample test at 5% level of sig-
nificance. The maximum absolute difference
(D) between cumulative theoretical normalized
distribution and cumulative observed normal-
ized error spectrum was computed. The critical
value was P N= =1 35 59662. / .0 0 which
corresponded to a confidence level of 95% as
indicated by (K-S) - Tables (Wasserman, 2005).
3.2. Discrete Wavelet
Transform (DWT)
Discrete wavelet decomposition was performed
on 1024 samples of the EMG signal to the
tenth level of resolution using “Haar” mother
function (Debnath, 2003). The DWT feature
sets were computed using two methods. The
first is based on applying principal component
analysis (PCA) to reduce the data dimensional-
ity (Englehart et al., 2001) and the second is
to calculate the average energy content of the
wavelet coefficients at each resolution as fol-
lows. There were a total of 10 sub-bands from
which features were extracted. The ith element
of a feature vector was given by Goswami and
Chan (1999).
vnw i
i
dwt
i
i j
j
ni
= =
=
∑
11 2 10
2
1
,, ,..... (2)
Where
n n n n vi
dwt
1
9
2
8
3
7
10
0
2 2 2 2= = = =, , ,......, ;
was the ith feature element in a DWT feature
vector; ni was the number of samples in an
individual subband; and wi j,
2was the jth coef-
ficient of the ith subband.
As a result, a DWT feature vector was
formed as given by:
v v v v
dwt dwt dwt dwt t
=
{}
1 2 10
, , ........., (3)
3.3. Wavelet Packet
Transform (WP)
A wavelet packet multiresolution analysis was
performed on 1024 samples of data to the third
level of resolution using “Sym5” wavelets to
obtain 8 subbands (Debnath, 2003). Each sub-
band contained a total of 128 wavelet packet
coefficients. Here, PCA can also be used to
reduce the data dimensionality or to compute the
average energy content. From each subband at
the third level of resolution, the average energy
content in the wavelet packet coefficients was
computed such that (Goswami & Chan, 1999):
(4)
Where vi
wp was the ith feature in a wavelet
packet feature vector, ni was the number of
samples in each subband and pi j,. was the jth
wavelet packet coefficient in the ith subband.
The WP feature vector was represented as fol-
lows:
vnp i and n i
i
wp
i
i j
j
n
i
i
= = = ∀
=
∑
11 2 8 128
2
1
,
, ,.....

International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012 43
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
v v v v
wp wp wp wp t
=
{}
1 2 8
, , ........., (5)
4. STATISTICAL ANALYSIS
Two-tailed Student t-test at 5% level of sig-
nificance and F-test at 5% level of significance
have been performed on each of the three
feature vectors: 7 autoregressive prediction
coefficients, 10 wavelet transform feature ele-
ments, and 8 wavelet packet feature elements.
Least significant difference test at 5% and 1%
levels of significance have been performed on
both 10 DWT feature elements and 8 WP fea-
ture elements. The tests have shown that there
are no significant differences before and after
wearing trial denture base. Table 1 depicts the
results obtained from the t-test and F-test for
the three feature sets.
There are no statistically significant dif-
ferences between the feature elements before
and after patients wearing trial denture base for
both right and left sides. The 7 AR prediction
coefficients do not show significant differ-
ences at 5% level of significance for both tests.
However, using the 8 WP features set, or 10
DWT feature set, few coefficients show sig-
nificant differences.
Table 2 illustrates the results obtained from
the least significant difference test at 5% and
1% levels of significance for both the 10 DWT
feature set and the 8 WP feature set. The results
have proved that there were no statistically sig-
nificant differences between feature elements
before and after trial denture base insertion.
5. NEURAL NETWORK
CLASSIFIER
In an attempt to discriminate between the mas-
seter muscle activity before and after inserting
trial denture base, a multi-layer feed forward
network trained by back propagation algorithm
was utilized (Gurney, 1997). The designed
feed-forward neural network had three layers;
an input layer, an output layer, and a hidden
layer (Anthony & Bartlett, 2009; Luo & Unbe-
hauen, 1997). The input layer has been varied
by varying the feature vector. The hidden layer
was thought for each vector for obtaining best
Table1.Comparisonbetween7ARpredictioncoefficients,8WPfeatureelements,10WTfeature
elementsforbothrightandleftsidesbeforeandafterwearingdenturesusingt-testandF-test
at5%levelofsignificance
Feature
Element
WT WP AR
T-Test F-Test T-Test F-Test T-Test F-Test
R L R L R L R L R L R L
1 0.68 -0.94 5.59* 0.28* 1.24 -0.28 5.15* 0.92 0.41 -0.77 1.26 0.43
2 0.67 -0.46 6.69* 0.58 0.23 -0.82 2.12 0.25* 0.77 0.59 1.86 0.63
3 0.54 -0.83 6.10* 0.26* 0.41 -0.26 4.64* 0.83 -0.68 -0.84 1.36 1.05
4 -0.64 0.88 0.48 9.14* 0.99 -0.76 19.2* 0.15* 1.23 -0.08 0.51 0.26*
5 1.48 0.67 18.7* 4.0* 0.79 -1.17 4.36* 0.13* 1.12 -0.02 0.72 1.09
6 1.42 1.12 15.8* 15.0* 0.49 -1.24 5.97* 0.07* -1.78 -1.23 2.24 1.07
7 1.21 0.76 43.3* 12.1* 0.44 -0.38 6.44* 0.87 0.39 2.50* 2.01 1.03
8 1.35 0.75 5.25* 4.42* 0.56 -0.27 1.42 0.45
9 1.03 0.78 42.5* 20.5*
10 0.73 0.16 21.1* 2.49
R=Right side L=Left side t tab.(5%) = 2.021 f tab.(5%) = 2.51 *significantly differ

44 International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
results. The output layer has been consisted of
1 neuron corresponding to 2 different classes of
the EMG signal; “before” and “after” wearing
trial denture base classes. The back-propagation
algorithm was utilized for the training procedure
using MATLAB 2009 library. Weights were
initially set to small random values (Anthony
& Bartlett, 2009).
The learning rate has been changed for
each feature vector. As the number of cases is
limited, so, another method called “leave-one
out” method was utilized to overcome this
problem (Mumford & Desolneux, 2010). The
complete set of data is used as the training set
after isolation of one case. Thus, the training is
done with 23 cases instead of 12 cases using the
“hold out” method. After finishing the network
training phase, testing was done with the iso-
lated sample. The procedure was repeated until
all samples have been classified individually.
Although the training phase using this approach
takes longer time, however, it indicates that as
the number of cases increases, the performance
of the network improves and the percentage of
correct classified signals becomes larger. The
neural network has been trained and tested
using the 3 feature vectors separately; the 7
AR prediction coefficients, the 8 WP feature
set using “Sym5” basis function, and the 10
DWT feature set using “Haar” basis function.
The principal component analysis has been per-
formed on DWT and WP coefficients matrices to
reduce feature dimensionality. After extracting
the principal components for both WT and WP
coefficients matrices, the neural network was
trained again. Higher classification rate was
obtained from WP using “Sym5” basis func-
tion and applying the PCA. Table 3 illustrates
the percentage of correct classification of the
neural network for 10 WT feature elements, 10
WT using PCA, 8 WP feature elements, 8 WP
using PCA, and 7 AR prediction coefficients for
the 24 patients for right (R) and left (L) sides.
The correct classification rate reaches 66.7%
for the right side and 58.7% for the left side
when using WP (sym5) and PCA.
6. DISCUSSION AND
CONCLUSION
The main objective of the present study was
to compare the masseter muscle surface elec-
tromyographic activity of patients before and
Table2.Comparisonbetween10WTfeatureelements,8WPfeatureelementsforbothrightand
leftsidesbeforeandafterwearingdenturesusingleastsignificantdifferencetestat5%and1%
levelsofsignificance
Feature Element WT WP
LSD (5%) LSD (1%) LSD (5%) LSD (1%)
1 0.0011* 0.0015* 0.0258* 0.0379*
2 0.0007* 0.0010* 0.0006* 0.0009*
3 0.0008* 0.0010* 0.0006* 0.0008*
4 0.0010* 0.0013* 0.0009* 0.0011*
5 0.0015* 0.0020* 0.0010* 0.0014*
6 0.0052* 0.0070* 0.0018* 0.0023*
7 0.0315* 0.0419* 0.0009* 0.0012*
8 0.0653* 0.0869* 0.0021* 0.0028*
9 1.1485* 1.5275*
10 0.9503* 1.2639*
*not significantly differ

International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012 45
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
after wearing trial denture base for both right
and left sides. For this study twenty-four healthy
edentulous patients were selected from patients
for complete denture construction. All patients
had to meet the following criteria: no history
suggesting neuromuscular disease and no signs
and symptoms of stomatognathic dysfunction.
Maxillary and mandibular acrylic trial denture
base were constructed for every patient and
tested for adequate retention. The surface
Electromyographic activity of the masseter
muscles were recorded bilaterally by bipolar
surface electrodes for 30 seconds. The sampling
rate was 2 kHz. Autoregressive modeling of
the recorded surface EMG signals was carried
out to obtain 7 prediction coefficients. Discrete
wavelet decomposition to the tenth resolution
level and discrete wavelet packet decomposition
to the third resolution level have been performed
to compute feature vectors corresponding to
each transform. Results have shown that no
statistically significant differences can be de-
tected for the autoregressive features set using
the student t-test and the F-test at the 5% level
of significance. However, few coefficients
showed significant differences when using the
WP feature set, or the DWT feature set. The
results of the least significant difference test
have showed that there were no statistically
significant differences between feature elements
before and after inserting trial denture base at
5% and 1% levels of significance for both the
10 WT and 8 WP feature sets. The results of
the 3 parametric statistical tests did not show
any significant electromyographic recording
activity difference along the right and left
masseter muscles, these results confirm the
proper patient’s selection as all have healthy
stomatognathic system. Moreover, the fact that
the tests do not show significant differences in
the masseter muscle electromyographic record-
ing with or without trial denture base insertion
did not violate silency during testing but may
suppose that; the electrical activity pattern was
not an approximate significant difference level
to be recorded. The neural network classifier was
able to detect significant differences before and
after wearing dentures using each of the three
features set: the 7 AR prediction coefficients,
10 DWT feature elements, and 8 WP feature
elements for both right and left sides.
The principal components for both WT and
WP coefficient matrices have been extracted.
The percentage of correct classification using
24 patients reached to 66.7% for right side and
58.7% for left side by using 8 WP principal
components. These results are in contrast to
the reports of investigations whose reported
that muscle activity was minimal or absent at
clinical rest position (Akihisa, 2003; Gomes et
al., 2010). The significant differences detected
by the neural network and found in 66.7% of the
patients may be due to wide variation between
individuals as each subject has its particular
intra-muscular activity pattern that differs
from one individual to another. The difference
between the two approaches statistical and to
neural networks that about 66.7% patients had
significant differences may be attributed pa-
tients differences in their times of dental extrac-
Table3.Percentageofcorrectclassificationfor10WTfeatureelements,10WTprincipalcom-
ponents,8WPfeatureelements,8WPprincipalcomponents,and7ARpredictioncoefficients
using24patients(%)
Feature Vectors Right Left
WT feature elements 47.9 50
WT Using PCA 58.3 56.5
WP feature elements 43.8 58.7
WP Using PCA 66.7 58.7
AR prediction coefficients 50 39.1

46 International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
tion; and associated preferred side of chewing
which may be simulate the masseter muscles to
have abnormal muscle tone and activity; or the
other explanation due to anatomical variations
of facial skeleton and muscle attachment; which
related to familial genesis.
All these reflexes cannot reach a diagnostic
level by conventional methods. However, the
wavelet methodology as an accurate decom-
position technique can give more accurate
and diagnostic details for subclinical abnormal
muscle reflex activity which denotes it to be
more valuable and considerable diagnostic
tools for patients follow; particularly for those
implants, bone grafting and occlusal analysis.
In conclusion, it has been found that the sug-
gested scheme provides useful information
that can assist the physician in diagnosis. The
present work is the first step to the way of
constructing a complete automatic system for
measuring, processing, analyzing, classifying
and diagnosing for all muscle disorders.
REFERENCES
Akihisa, S. (2003). Stability of mandibular position
during pronunciation on determination of occlusal
vertical dimension. Ohu University DentalJournal,
30(4), 281–287.
Amorim, C. F., Giannasi, L. C., Ferreira, L. M., Magini,
M., Oliveira, C. S., & de Oliveira, L. V. (2010). Behavior
analysis of electromyographic activity of the masseter
muscle in sleep bruxers. Journal of Bodywork and
MovementTherapies, 14(3), 234–238. doi:10.1016/j.
jbmt.2008.12.002
Anthony, M., & Bartlett, P. L. (2009). Neuralnetwork
learning:Theoreticalfoundations (1st ed.). Cambridge,
UK: Cambridge University Press.
Barisci, N. (2008). The adaptive ARMA analysis of
EMG signals. Journal of Medical Systems, 32(1),
43–50. doi:10.1007/s10916-007-9106-8
Bendat, J. S. M., & Piersol, A. G. (2010). RandomData:
Analysisandmeasurementprocedures (4th ed.). New
York, NY: John Wiley & Sons.
Calder, K. M., Stashuk, D. W., & McLean, L. (2008).
Physiological characteristics of motor units in the bra-
chioradialis muscle across fatiguing low-level isometric
contractions. JournalofElectromyographyandKinesi-
ology, 18(1), 2–15. doi:10.1016/j.jelekin.2006.08.012
De A Berger, P., Nascimento, F. A., do Carmo, J. C.,
& da Rocha, A. F. (2006). Compression of EMG
signals with wavelet transform and artificial neural
networks. PhysiologicalMeasurement, 27(6), 457–465.
doi:10.1088/0967-3334/27/6/003
Debnath, L. (2003). Wavelets & signal processing
(1st ed.). Boston, MA: Birkhauser. doi:10.1007/978-
1-4612-0025-3
Dhanjal, S. (1997). OSLP: a new technique in linear
prediction of speech. In ProceedingsoftheIEEEPacific
RimConferenceonCommunications,Computersand
SignalProcessing (Vol. 2, pp. 711-714).
Dimitrova, N. A., & Dimitrov, G. V. (2003). Interpreta-
tion of EMG changes with fatigue: facts, pitfalls, and fal-
lacies. JournalofElectromyographyandKinesiology,
13(1), 13–36. doi:10.1016/S1050-6411(02)00083-4
Englehart, K., Hudgins, B., & Parker, P. A. (2001). A
wavelet-based continuous classification scheme for
multifunction myoelectric control. IEEE Transac-
tions on Bio-Medical Engineering, 48(3), 302–311.
doi:10.1109/10.914793
Feichtinger, H. G., & Strohmer, T. (1997). Gabor
analysis and algorithms: Theory and applications
(1st ed.). Boston, MA: Birkhauser. doi:10.1007/978-
1-4612-2016-9
Gomes, S. G., Custodio, W., Faot, F., Del Bel Cury, A.
A., & Garcia, R. C. (2010). Masticatory features, EMG
activity and muscle effort of subjects with different
facial patterns. JournalofOralRehabilitation, 37(11),
813–819. doi:10.1111/j.1365-2842.2010.02075.x
Goswami, J. C., & Chan, A. K. (1999). Fundamentals
ofWavalets:theory,algorithms,andapplications. New
York, NY: John Wiley & Sons.
Gross, M. D., & Ormianer, Z. (1994). A preliminary
study on the effect of occlusal vertical dimension in-
crease on mandiblar postural rest position. International
JournalofProsth, 7, 216–220.
Gurney, K. (1997). An introduction to neural
ne tworks. B oc a R at on , F L: C RC P ress.
doi:10.4324/9780203451519
Kallenberg, L. A., & Hermens, H. J. (2006). Behaviour
of motor unit action potential rate, estimated from
surface EMG, as a measure of muscle activation level.
JournalofNeuroengineeringandRehabilitation, 17,
3–15.

International Journal of System Dynamics Applications, 1(1), 39-47, January-March 2012 47
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Karlsson, S., Yu, J., & Akay, M. (1999). Enhancement
of spectral analysis of myoelectric signals during static
contractions using wavelet methods. IEEETransac-
tions on Bio-Medical Engineering, 46, 670–684.
doi:10.1109/10.764944
Krivickas, L. S., Taylor, A., Maniar, R. M., Mascha,
E., & Reisman, S. S. (1998). Is spectral analysis of
the surface electromyographic signal a clinically
useful tool for evaluation of skeletal muscle fatigue?
JournalofClinicalNeurophysiology, 15(2), 138–145.
doi:10.1097/00004691-199803000-00006
Luo, F. L., & Unbehauen, R. (1997). Applied neu-
ral networks for signal processing. IEEE Signal
Processing Magazine, 14(6), 38–39. doi:10.1109/
MSP.1997.637305
Michelloti, A., Farella, M., & Vollaro, S. (1997).
Mandibular rest position and electrical activity of the
masticatory muscles. TheJournalofProstheticDen-
tistry, 78, 48–53. doi:10.1016/S0022-3913(97)70087-8
Mumford, D., & Desolneux, A. (2010). Patterntheory:
Thestochasticanalysisofreal-worldsignals. Natick,
MA: A K Peters.
Pfeiffer, G., & Kunze, K. (1993). Frequency analysis and
duration of motor unit potentials: reliability and diag-
nostic usefulness. ElectroencephalographyandClinical
Neurophysiology, 89(6), 365–374. doi:10.1016/0168-
5597(93)90109-3
Salvador, J., & de Bruin, H. (2006). The use of the
wavelet transform in EMG M-wave pattern classifi-
cation. In ProceedingsoftheIEEE28thInternational
ConferenceoftheEngineeringinMedicineandBiology
Society (Vol. 1, pp. 2304-2307).
Sato, S., Ohta, M., Goto, S., Kawamura, H., & Motegi,
K. (1998). Electromyography during chewing move-
ment in patients with anterior disc displacement of
the temporomandibular joint. International Journal
ofOraland MaxillofacialSurgery, 27(4), 274–277.
doi:10.1016/S0901-5027(05)80613-7
Vaidyanathan, P. P. (2008). Thetheoryoflinearpredic-
tion (1st ed.). UK: Morgan and Claypool.
Wang, K., Lobbezoo, F., Svensson, P., & Arendt-
Nielsen, L. (2007). Influence of jaw gape on EMG
of jaw muscles and jaw-stretch reflexes. Archivesof
OralBiology, 52(6), 562–570. doi:10.1016/j.archo-
ralbio.2006.12.004
Wasserman, L. (2005). Allofnonparametricstatistics.
New York, NY: Springer.
Zuccolotto, M. C., Vitti, M., Nobilo, K. A., Regalo, S. C.,
Siéssere, S., & Bataglion, C. (2007). Electromyographic
evaluation of masseter and anterior temporalis muscles
in rest position of edentulous patients with temporo-
mandibular disorders, before and after using complete
dentures with sliding plates. Gerodontology, 24(2),
105–110. doi:10.1111/j.1741-2358.2007.00152.x
AhmadAzarreceivedaMScdegree(2006)inSystemDynamicsandaPhD degree(2009)in
AdaptiveNeuro-FuzzySystemsfromFacultyofEngineering,CairoUniversity (Egypt).He is
currentlytheVicePresidentoftheEgyptSystemDynamicsChapterandViceChairoftheIEEE
ComputationalIntelligenceSociety (CIS)Egypt Chapter.Dr.Ahmad Azarhasworked inthe
areasofsystemdynamics,softcomputing,intelligentcontrolsystemsandmodellinginbiomedi-
cineandistheauthorofmorethan35papersinthesesubjects.Heholdscopyrightsforsome
novelsoftwareintheeldofmedicalintelligentsystems.Heisaneditorofthreebooksinthe
eldoffuzzylogicsystemsandbiomedicalengineering.Dr.AhmadAzarcurrentlyservesasthe
Editor-in-Chiefofmanyinternationaljournals.Healsoservesasaninternationalprogramme
committeemember inmanyinternationalandpeer-reviewed conferences.Hisbiographywas
selectedtoappearinthe27thEditionofWho’s Who in the World,Marquis Who’s Who,USA,
2010.DrAhmadAzar’sresearchinterestsaremainlyfocusedoncontrolsystemanalysis,sys-
temsengineering,systemdynamics,medical robotics,processcontrol,neuralnetworks,fuzzy
logiccontrollers,neuro-fuzzysystems,systemthinking,mathematicalmodelingandcomputer
simulation,statistical analysis,decision-makinganalysis,researchmethodology,biofeedback
systemsandthemonitoringandcontrollingofhemodialysissystems.