![ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010
EMG Diagnosis using Neural Network Classifier
with Time Domain and AR Features
*Er. Gurmanik Kaur1, Dr. A. S. Arora2, and Dr. V. K. Jain2
EIE Deptt., SLIET Longowal, Distt. Sangrur, Punjab, INDIA, 148106.
Email: Gurmanik Kaur- mannsliet@gmail.com
Dr. A.S.Arora- ajatsliet@yahoo.com
Dr.V.K.Jain- vkjain27@yahoo.com
Abstract—The shapes of motor unit action potentials recruited. Different MUAPs will overlap, causing an
(MUAPs) in an electromyographic (EMG) signal interference pattern in which the neurophysiologist cannot
provide an important source of information for the detect individual MUAP shapes reliably.
diagnosis of neuromuscular disorders. To extract this Traditionally, neurophysiologists assess MUAPs from
information from the EMG signals, the first step is their shape using an oscilloscope and listening to their
identification of the MUAPs composed by the EMG audio characteristics. Thus an experienced electro-
signal, second step is clustering of MUAPs with similar physiologist can detect abnormalities with reasonable
shapes, third step is extraction of the features of MUAP accuracy. However, subjective MUAP assessment,
clusters and last step is classification of MUAPs. In this although satisfactory for the detection of unequivocal
work, the MUAPs are identified by using a data driven abnormalities, may not be sufficient to delineate less
segmentation algorithm, statistical pattern recognition obvious deviations or mixed patterns of abnormalities [5].
technique is used for clustering of MUAPs. Followed by These ambiguous cases call for quantitative MUAP
the extraction of time domain and autoregressive (AR) analysis. With the aid of computer technology, today it is
features of the MUAP clusters. Finally, a neural possible to analyze EMG signals quantitatively that helps
network (NN) classifier is used for classification of in saving time, standardizes the measurements and enables
MUAPs. A total of 12 EMG signals obtained from 3 the extraction of additional features which cannot be easily
normal (NOR), 5 myopathic (MYO) and 4 motor calculated manually.
neuron diseased (MND) subjects were analyzed. The To further the development of quantitative EMG
success rate for the segmentation technique is 95.90% techniques, the need has emerged for adding automated
and for the statistical technique is 93.13%. The decision making support to these techniques so that all data
classification accuracy of NN is 66.72% with time is processed in an integrated environment. Towards this
domain parameters and 75.06 % with AR parameters. goal, Blinowska [6] proposed the use of discriminant
analysis for the evaluation of MUAP findings, Coatrieux
Index Terms—Electromyography, motor unit action and associates [7]-[9] applied cluster analysis techniques
potentials, neural networks, classification. for the automatic diagnosis of pathology based on MUAP
records. Andreassen and co-workers [10]-[12] developed
I. INTRODUCTION the MUNIN (Muscle and Nerve Inference Network) which
employs a causal probabilistic network for the
Electromyography is the study of the muscular function interpretation of EMG findings, Fuglsang-Frederiksen and
based on the analysis of electromyographic (EMG) signals
his group [13], [14] developed a rule-based EMG expert
that are electrical activities generated by muscles during system named KANDID, and Jamieson [15], [16]
voluntary, involuntary or stimulated contractions [1]. EMG developed an EMG processing system based on augmented
signals are composed of motor unit action potentials transition networks. In most of these systems, the
(MUAPs) generated by different motor units (MUs). The generation of the input pattern assumes a probabilistic
term motor unit refers collectively to one motor neuron and
model, with the matching score representing the likelihood
the group of muscle fibers it innervates and is the smallest that the input pattern was generated from the underlying
unit of skeletal muscle that can be activated by volitional class [17]. In addition, assumptions are typically made
effort. MUAPs from different MUs tend to have distinct
concerning the probability density function of the input
shapes, which remain almost the same for each discharge data. Recently, artificial neural networks (ANNs) have
[2-4]. When a patient maintains low level of muscle been proposed as an alternative tool to pattern recognition
contraction, individual MUAPs can be easily recognized.
and classification problems. One of their major advantages
As contraction intensity increases, more motor units are is that ANN models make no assumption about the
underlying probability density functions of the input data,
*Corresponding author: Er. Gurmanik Kaur, Research Scholar, EIE thus possibly improving the performance of classifiers,
Deptt., SLIET, Longowal, Distt. Sangrur, Punjab, INDIA. 148106.
E-mail: mannsliet@gmail.com
especially when the data depart significantly from
normality. Other features of artificial neural networks that
12
© 2010 ACEEE
DOI: 01.IJEPE.01.03.70](https://cdn.statically.io/img/image.slidesharecdn.com/70-120921231853-phpapp02/85/EMG-Diagnosis-using-Neural-Network-Classifier-with-Time-Domain-and-AR-Features-1-320.jpg)
![ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010
make them so attractive to investigate are that: 1) they B. Segmentation
exhibit adaptation or learning, 2) they pursue multiple
hypotheses in parallel, 3) they may be fault tolerant, 4) they The next step is to cut the EMG signal into segments of
may process degraded or incomplete data, and 5) they may possible MUAP waveforms and eliminate areas of low
create complex classification boundaries [18]. activity. We have evaluated three different segmentation
technique and it is concluded that the segmentation of
In this work, the EMG signals recorded from (NOR),
EMG signals by finding the peaks of the MUAPs, yielded
myopathic (MYO) and motor neuron diseased (MND)
the best results [19]. The segmentation algorithm calculates
subjects were segmented using an algorithm that
a threshold depending on the maximum value and the mean
automatically detects the areas of low activity and
absolute value of the whole EMG signal, The Threshold
( )
candidate MUAPs. The possible MUAPs are then clustered
using a statistical pattern recognition technique. T is calculated as:
Furthermore, the time domain and autoregressive (AR)
30 L 5 L
features of MUAP clusters are extracted and are given to a if max{x i } > ∑ xi , then T = ∑ xi
neural network (NN) classifier for their classification into i L i =1 L i =1
NOR, MYO and MND classes. Finally, the performance of
NN classifier with time domain and AR parameters is else T = max{xi } 5
i
compared.
Where xi are the discrete input values and L is the
II. MATERIAL AND METHODOLOGY number of samples in the EMG signal. Peaks over the
calculated threshold are considered as candidate MUAPs.
A. Data Acquisition and Pre-Processing Then a window of 120 sampling points (i.e. 6ms at 20
KHz) is centered at the identified peak. If a greater peak is
Our data contain real time EMG signal obtained from the found in the window, the window is centered at the greater
Department of Computer Science, University of Cyprus, peak; otherwise the 120 points are saved as MUAP
Cyprus. All the EMG signals were acquired from the waveform [20]. The segmented EMG signals of NOR,
biceps brochii muscle at upto 30% of the maximum MYO and MND subjects in segments of 6ms and centered
voluntary contraction (MVC) level under isometric at the maximum peak, are shown in Figure 4, Figure 5 and
conditions., for 5 seconds, using the standard concentric Figure 6 respectively.
needle electrode, from NOR, MYO and MND subjects.
400
Figure 1, Figure.2 and Figure 3 shows the typical EMG
300
recordings. The EMG signals were analogue band pass
filtered at 3-10 KHz, sampled at 20 KHz with 12-bit 200
M g itu eµ )
an d( V
resolution and then low pass filtered at 8KHz. 100
0
300
-100
200
-200
0 20 40 60 80 100 120
Sample No.
M g itu e )
a n d (µV
100
0
Figure 4. Segmented EMG signal of a NOR subject in segments of 6ms
-100
and centered at the maximum peak.
-200 500
0 1000 2000 3000 4000 5000
Sample No. 400
300
200
Figure 1. Raw EMG signal of a NOR subject.
M g itu eµ )
100
an d( V
0
1000
-100
-200
500
-300
-400
Mg itu eµ )
an d( V
0
-500
0 20 40 60 80 100 120
Sample No.
-500
-1000
Figure 5. Segmented EMG signal of a MYO subject in segments of 6ms
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
and centered at the maximum peak.
Sample No.
1000
Figure 2. Raw EMG signal of a MYO subject.
800
600
Mg itu eµ )
an d( V
1500
400
1000
200
500
0
Mg itu eµ )
an d( V
0
-200
-500
-400
0 20 40 60 80 100 120
Sample No.
-1000
0 1000 2000 3000
Sample No.
4000 5000 Figure 6. Segmented EMG signal of a MND subject in segments of 6ms
and centered at the maximum peak.
Figure 3. Raw EMG signal of a MND subject.
13
© 2010 ACEEE
DOI: 01.IJEPE.01.03.70](https://cdn.statically.io/img/www.slideshare.net/data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7)
EMG Diagnosis using Neural Network Classifier with Time Domain and AR Features
- 1. ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010 EMG Diagnosis using Neural Network Classifier with Time Domain and AR Features *Er. Gurmanik Kaur1, Dr. A. S. Arora2, and Dr. V. K. Jain2 EIE Deptt., SLIET Longowal, Distt. Sangrur, Punjab, INDIA, 148106. Email: Gurmanik Kaur- mannsliet@gmail.com Dr. A.S.Arora- ajatsliet@yahoo.com Dr.V.K.Jain- vkjain27@yahoo.com Abstract—The shapes of motor unit action potentials recruited. Different MUAPs will overlap, causing an (MUAPs) in an electromyographic (EMG) signal interference pattern in which the neurophysiologist cannot provide an important source of information for the detect individual MUAP shapes reliably. diagnosis of neuromuscular disorders. To extract this Traditionally, neurophysiologists assess MUAPs from information from the EMG signals, the first step is their shape using an oscilloscope and listening to their identification of the MUAPs composed by the EMG audio characteristics. Thus an experienced electro- signal, second step is clustering of MUAPs with similar physiologist can detect abnormalities with reasonable shapes, third step is extraction of the features of MUAP accuracy. However, subjective MUAP assessment, clusters and last step is classification of MUAPs. In this although satisfactory for the detection of unequivocal work, the MUAPs are identified by using a data driven abnormalities, may not be sufficient to delineate less segmentation algorithm, statistical pattern recognition obvious deviations or mixed patterns of abnormalities [5]. technique is used for clustering of MUAPs. Followed by These ambiguous cases call for quantitative MUAP the extraction of time domain and autoregressive (AR) analysis. With the aid of computer technology, today it is features of the MUAP clusters. Finally, a neural possible to analyze EMG signals quantitatively that helps network (NN) classifier is used for classification of in saving time, standardizes the measurements and enables MUAPs. A total of 12 EMG signals obtained from 3 the extraction of additional features which cannot be easily normal (NOR), 5 myopathic (MYO) and 4 motor calculated manually. neuron diseased (MND) subjects were analyzed. The To further the development of quantitative EMG success rate for the segmentation technique is 95.90% techniques, the need has emerged for adding automated and for the statistical technique is 93.13%. The decision making support to these techniques so that all data classification accuracy of NN is 66.72% with time is processed in an integrated environment. Towards this domain parameters and 75.06 % with AR parameters. goal, Blinowska [6] proposed the use of discriminant analysis for the evaluation of MUAP findings, Coatrieux Index Terms—Electromyography, motor unit action and associates [7]-[9] applied cluster analysis techniques potentials, neural networks, classification. for the automatic diagnosis of pathology based on MUAP records. Andreassen and co-workers [10]-[12] developed I. INTRODUCTION the MUNIN (Muscle and Nerve Inference Network) which employs a causal probabilistic network for the Electromyography is the study of the muscular function interpretation of EMG findings, Fuglsang-Frederiksen and based on the analysis of electromyographic (EMG) signals his group [13], [14] developed a rule-based EMG expert that are electrical activities generated by muscles during system named KANDID, and Jamieson [15], [16] voluntary, involuntary or stimulated contractions [1]. EMG developed an EMG processing system based on augmented signals are composed of motor unit action potentials transition networks. In most of these systems, the (MUAPs) generated by different motor units (MUs). The generation of the input pattern assumes a probabilistic term motor unit refers collectively to one motor neuron and model, with the matching score representing the likelihood the group of muscle fibers it innervates and is the smallest that the input pattern was generated from the underlying unit of skeletal muscle that can be activated by volitional class [17]. In addition, assumptions are typically made effort. MUAPs from different MUs tend to have distinct concerning the probability density function of the input shapes, which remain almost the same for each discharge data. Recently, artificial neural networks (ANNs) have [2-4]. When a patient maintains low level of muscle been proposed as an alternative tool to pattern recognition contraction, individual MUAPs can be easily recognized. and classification problems. One of their major advantages As contraction intensity increases, more motor units are is that ANN models make no assumption about the underlying probability density functions of the input data, *Corresponding author: Er. Gurmanik Kaur, Research Scholar, EIE thus possibly improving the performance of classifiers, Deptt., SLIET, Longowal, Distt. Sangrur, Punjab, INDIA. 148106. E-mail: mannsliet@gmail.com especially when the data depart significantly from normality. Other features of artificial neural networks that 12 © 2010 ACEEE DOI: 01.IJEPE.01.03.70
- 2. ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010 make them so attractive to investigate are that: 1) they B. Segmentation exhibit adaptation or learning, 2) they pursue multiple hypotheses in parallel, 3) they may be fault tolerant, 4) they The next step is to cut the EMG signal into segments of may process degraded or incomplete data, and 5) they may possible MUAP waveforms and eliminate areas of low create complex classification boundaries [18]. activity. We have evaluated three different segmentation technique and it is concluded that the segmentation of In this work, the EMG signals recorded from (NOR), EMG signals by finding the peaks of the MUAPs, yielded myopathic (MYO) and motor neuron diseased (MND) the best results [19]. The segmentation algorithm calculates subjects were segmented using an algorithm that a threshold depending on the maximum value and the mean automatically detects the areas of low activity and absolute value of the whole EMG signal, The Threshold ( ) candidate MUAPs. The possible MUAPs are then clustered using a statistical pattern recognition technique. T is calculated as: Furthermore, the time domain and autoregressive (AR) 30 L 5 L features of MUAP clusters are extracted and are given to a if max{x i } > ∑ xi , then T = ∑ xi neural network (NN) classifier for their classification into i L i =1 L i =1 NOR, MYO and MND classes. Finally, the performance of NN classifier with time domain and AR parameters is else T = max{xi } 5 i compared. Where xi are the discrete input values and L is the II. MATERIAL AND METHODOLOGY number of samples in the EMG signal. Peaks over the calculated threshold are considered as candidate MUAPs. A. Data Acquisition and Pre-Processing Then a window of 120 sampling points (i.e. 6ms at 20 KHz) is centered at the identified peak. If a greater peak is Our data contain real time EMG signal obtained from the found in the window, the window is centered at the greater Department of Computer Science, University of Cyprus, peak; otherwise the 120 points are saved as MUAP Cyprus. All the EMG signals were acquired from the waveform [20]. The segmented EMG signals of NOR, biceps brochii muscle at upto 30% of the maximum MYO and MND subjects in segments of 6ms and centered voluntary contraction (MVC) level under isometric at the maximum peak, are shown in Figure 4, Figure 5 and conditions., for 5 seconds, using the standard concentric Figure 6 respectively. needle electrode, from NOR, MYO and MND subjects. 400 Figure 1, Figure.2 and Figure 3 shows the typical EMG 300 recordings. The EMG signals were analogue band pass filtered at 3-10 KHz, sampled at 20 KHz with 12-bit 200 M g itu eµ ) an d( V resolution and then low pass filtered at 8KHz. 100 0 300 -100 200 -200 0 20 40 60 80 100 120 Sample No. M g itu e ) a n d (µV 100 0 Figure 4. Segmented EMG signal of a NOR subject in segments of 6ms -100 and centered at the maximum peak. -200 500 0 1000 2000 3000 4000 5000 Sample No. 400 300 200 Figure 1. Raw EMG signal of a NOR subject. M g itu eµ ) 100 an d( V 0 1000 -100 -200 500 -300 -400 Mg itu eµ ) an d( V 0 -500 0 20 40 60 80 100 120 Sample No. -500 -1000 Figure 5. Segmented EMG signal of a MYO subject in segments of 6ms 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 and centered at the maximum peak. Sample No. 1000 Figure 2. Raw EMG signal of a MYO subject. 800 600 Mg itu eµ ) an d( V 1500 400 1000 200 500 0 Mg itu eµ ) an d( V 0 -200 -500 -400 0 20 40 60 80 100 120 Sample No. -1000 0 1000 2000 3000 Sample No. 4000 5000 Figure 6. Segmented EMG signal of a MND subject in segments of 6ms and centered at the maximum peak. Figure 3. Raw EMG signal of a MND subject. 13 © 2010 ACEEE DOI: 01.IJEPE.01.03.70
- 3. ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010 C. MUAP Clustering 500 150 400 In this stage, MUAP clusters are automatically detected 300 200 100 50 and for each cluster the average or template shape is 100 0 determined. We have used statistical pattern recognition 0 -100 -50 technique for clustering of similar MUAPs. In this -200 -100 -300 technique the euclidian distance is used to identify and -400 -150 group similar MUAP waveforms. The group average is -500 0 20 40 60 80 100 120 -200 0 20 40 60 80 100 120 continuously calculated and is used for the classification of 200 MUAPs using a constant threshold [21]. The 150 implementation steps are: 100 Step 1: Start with the first waveform x as input (the first 50 member of the class). 0 Step 2: Calculate the vector length of x and the distance -50 between it and all the other segmented waveforms y as: -100 -150 0 20 40 60 80 100 120 N lx = ∑ i =1 xi2 where N = 120 Figure 8. Clustered EMG signals of a MYO subject. and 300 250 300 250 N ∑ (x 200 200 d xy = i − yi ) 2 150 100 150 100 i =1 50 50 0 0 Step 3: Find the waveform y with the minimum -50 -50 -100 -100 distance d min . The waveform y having minimum distance -150 -200 -150 -200 0 20 40 60 80 100 120 0 20 40 60 80 100 120 with the x has the greatest similarity with x and remove it from the input data. Step 4: if d min / l x < 0.3 then group, calculate group 250 200 150 average and go to step 1 with group average as input. 100 else if number of group members > 2, then form a new 50 0 class. -50 else waveform is superimposed, go to step 1 with y as -100 -150 input. -200 0 20 40 60 80 100 120 This process continues where it stopped comparing the Figure 9. Clustered EMG signal of a MND subject. last encountered waveform with all the remaining until all waveforms are processed. The threshold value of 0.3 in Step 4 is critical because a smaller value may split a D. Feature Extraction MUAP class with high waveform variability in two or In this work, we have extracted the time domain and AR more subclasses, whereas a greater threshold value may features of the MUAP clusters. To extract time domain merge resembling MUAP classes. Figure 7, Figure 8 and features, all clustered MUAP waveforms are expanded to Figure 9 illustrates the clustered EMG signals of NOR, 25ms on the original signal where the position of identified MYO and MND subjects respectively. peak was marked during segmentation. The rationale is that the MUAP duration is in most of the cases longer than 6ms 150 300 and the signal expansion is therefore, necessary for a 250 100 200 correct measurement of parameters [22]. The following 50 150 100 time domain parameters are computed from the MUAP 0 50 waveforms: 0 -50 -50 Spike duration: measured from the first to the last -100 0 20 40 60 80 100 120 -100 0 20 40 60 80 100 120 positive peak. 140 Amplitude: Amplitude difference between minimum 120 100 positive and maximum negative peak. 80 Area: Rectified MUAPs integrated over the calculated 60 duration 40 20 Phases: Number of baseline crossings where amplitude 0 exceeds ±25µV, plus one. -20 0 20 40 60 80 100 120 Turns: Number of positive and negative peaks where the Figure 7. Clustered EMG signal of a NOR subject. difference from the preceding and following turn exceeds 25µV. In the autoregressive (AR) model a current signal x(n) 14 © 2010 ACEEE DOI: 01.IJEPE.01.03.70
- 4. ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010 is described as linear combination of previous samples x(n- technique is 93.13%. Table 2 presents the results of k) weighted by a coefficient [23]. A common strategy to clustering for each of the three MUAP classes. calculate the AR coefficients is to use the Burg’s algorithm. Table II It provides an iterative and fast method to figure out the MUAP Clustering Success Rate parameters of the AR-model adaptively. We have used this method in our work to find the AR coefficients ao to a2 of a MUAP classes Success rate (%) 3rd order AR model. E. Classification NOR 93.13 (192/204) In order to classify the clustered MUAPs into NOR, MYO and MND classes, a NN classifier with back propagation (BP) training algorithm is employed. The back MYO 93.10 (270/290) propagation algorithm works in much the same way as the name suggests: After propagating an input through the MND 92.10 (175/190) network, the error is calculated is propagated back through the network while the weights are adjusted in order to make the error smaller. Taking into account the problem under Total 93.13 (637/684) consideration, NN architecture with three layers is used, as this model is documented as able to draw the boundaries of After clustering of MUAPs, time domain and AR arbitrarily complex decision regions [17]. features of the MUAPs are extracted. It is observed that standard deviation (SD) of all the time domain parameters III. RESULTS of the MYO 1 signal is zero, because in that case we have only one class of MUAPs. Moreover, the SD of all the AR EMG data recorded from 3 NOR, 5 MYO and 4 MND parameters of a signal will be zero, if the total numbers of subjects were analyzed using the methodology described in classes are less than two. Finally the extracted features are section II. Following the pre-processing, EMG signals are given to a NN classifier for training and classification of segmented by identifying the peaks of the MUAPs. Table 1 MUAPs. Table 3 presents the results of classifications. presents the success rate for each of the three MUAP classes. Table III Comparison of the Results of NN Classifier with Time Domain and AR Table I Features. MUAP Detection Success Rate Features % Classification accuracy MUAP classes Success rate (%) Time domain features 66.72 NOR 96.07 (196/204) AR features 75.06 MYO 95.86 (278/290) It is observed from the results that the classification of MUAPs is better with AR features than time domain features. Moreover, AR parameters have two MND 95.78 (182/190) important advantages over time domain parameters: 1) variations in the positioning of the electrodes on the surface Total 95.90 (656/684) of the muscle do not severely affect the AR-coefficients. 2) the amount of information to be presented to the classifier is greatly reduced. Therefore, the total processing time is The success rate is the percentage ratio of the also reduced. correctly identified MUAPs by the segmentation algorithm and the number of true MUAPs identified by manual CONCLUSIONS observation. The lowest success rate (95.78%) for the MND group is attributed to the more complex and variable In conclusion, the methodology described in this waveform shapes. The segmented MUAPs are clustered by work make possible the development of a fully automatic using statistical pattern recognition technique. Sometimes EMG signal analysis system which is accurate, simple, fast due to waveform variability, MUAP classes coming from and reliable enough to be used in routine clinical the same MU, although they looked simiIar, were not environment. This work can provide a good understanding grouped together. Merging of these classes can be achieved of EMG analysis procedures to the researchers to identify using a pattern recognition technique with a greater neuromuscular diseases. Future work will evaluate the constant threshold and the averaged class waveforms as algorithms developed in this study on EMG data recorded input. The total success rate obtained by using this from more muscles and more subjects. In addition, this 15 © 2010 ACEEE DOI: 01.IJEPE.01.03.70
- 5. ACEEE Int. J. on Electrical and Power Engineering, Vol. 01, No. 03, Dec 2010 system may be integrated into a diagnostic system for [12] S. Andreassen, F. Jensen, S. K. Andersen, B. Falck, U. neuromuscular diseases based on neural network where Kjaerulff, M. Woldbye, A. R. Sorensen, A. Rosenfalck, and EMG, muscle biopsy, biochemical and molecular genetic J. Frank, “MUNIN-An expert EMG assistant,” in Computer findings, and clinical data may be combined to provide a Aided Electromygraphy and Expert Systems, J. E. Desmedt, Ed. New York : Elsevier Science Publisher B.V., pp. 255- diagnosis. 277, 1989. . [13] A. FugIsang-Frederiksen and S. M. Jeppesen, “A rule-based REFERENCES EMG expert system for diagnosing neuromuscular [1] A. O. Andrade, Decomposition and analysis of disorders,” in Computer Aided Electromyography and Expert electromyographic signals, Ph. D. Thesis, University of Systems, J. E. Desmedt, Ed. New York: Elsevier Science Reading, UK, 2005. Publishers B.V., pp. 289-296, 1989. [2] DeLuca CJ, Towards understanding the EMG signal, 4th ed., [14] A. Fuglsang-Frederiksen, J. Ronager, and S, Vingtoft, “A Baltimore: Williams & Wilkinson; 1978. plan-test-diagnose expert system for EMG: KANDID,” [3] Krarup C, “Pitfalls in electrodiagnosis,” J Neurophysiol, vol. J.Neurolog. Sci., vol.98 (suppl.), p. 150, 1990. 81, pp. 1115-1126, 1999. [15] P. W. Jamieson, “Computerized interpretation of [4] McGill KC, “Optimal resolution of superimposed action electromyographic data.” Electroencephalogr. Clin. potentials,” IEEE Trans Biomed Engg, vol. 49, pp. 640-650, Neurophysiol., vol. 75, p. 392, 1990. 2002. [16] P. W. Jamieson, “A model for diagnosing and explaining [5] Richfield EK, Cohen BA, Albers JW, “Review of multiple disorders,” Comput. Biomed. Res., vol. 24, pp. 307- quantitative and automated needle electromyographic 320, 1991. analyses,” IEEE Trans Biomed Eng, pp. 506-514, 1981. [17] R. P. Lippmann, “An introduction to computing with neural [6] K. J. Blinowska, I. Hausmanowa-Petrusewicz, A. Miller- nets,” IEEE ASSP Mag. pp. 4-22, 1987. Larsson and J. Zachara, “The analysis of single EMG [18] Constantinos S. Pattichis, Christos N. Schizas and Lefkos potentials by means of multivariate methods,” Electromyogr. T. Middleton, “Neural Network Models in EMG Diagnosis,” Clin. Neurophysiol, vol. 20, pp. 105-123, 1980. IEEE Trans Biomed Eng., vol. 42, 1995. [7] J. L. Coatrieux, P. Toulouse, B. Rouvrais, and R. Le Bars, [19] Gurmanik kaur, A. S. Arora and V. K. Jain, “Comparison of “Automatic classification of electromyographic signals,” the techniques used for segmentation of EMG signals,” in EEG Clin. Neurophysiol., vol. 55, pp. 333-341, 1983. Proc. WSEAS Int. Conf. on Mathematical and [8] B. Rouvrais, P. Toulouse, J. L. Coatrieux. and R. Le Bars, Computational Methods, Baltimore, USA, pp. 124-129, “A possible method of automatic electromyographic analysis 2009. and diagnosis on line,” Electmmyogr. Clin. Neurophysiol.. [20] Christodoulos I. Christodoulou and Constantinos S. Pattichis, vol. 23, pp. 457-470, 1983. “Unsupervised Pattern Recognition for the Classification of [9] P. Toulouse, J. L. Coatrieux. and B. Le Marec, “An attempt EMG Signals,” IEEE Transactions on Biomedical Engg., to differentiate female relatives of Duchenne type dystrophy vol. 46, pp. 169-178,1999. from healthy subjects using an automatic EMG analysis,” J. [21] C. I. Christodoulou and C. S. Pattichis, “A new technique for Neurolog. Sci., vol. 67, pp. 45-55, 1985. the classification and decomposition of EMG signals,” in [10] S. K. Andersen, S. Andreassen, and M. Woldbye, Proc. IEEE Int. Conf. on Neural Networks, Perth, Western “Knowledge representation for diagnosis and test planning in Australia, vol. 5, pp. 2303–2308, Nov. 1995. the domain of EMG.” in Proc. 7th Eur. Conf. Artijcicial C.D. Katsis, D.I. Fotiadis, A. Likas and I. Sarmas, Automatic 1ntell., Brighton, U.K., pp. 357-368, 1986. discovery of the number of MUAP clusters and [11] S. Andreassen, S. K. Andersen, F. V. Jensen, M. Woldbye, superimposed MUAP decomposition in electromyograms,” A. Rosen- falck, B. Falck, U. Kjaerluff, and A. R. Sorensen, in Proc. of the 4th Annual IEEE Conf on Information “MUNIN-An expert system for EMG,” Electroenceph,ph. Technology Applications in Biomedicine, UK., pp. 177-180, Clin. Neurophvsiol., vol. 66, 1987. 2003.Marple S. L. Jr., Digital Spectral analysis with applications,Prentice-Hall,USA,1987. 16 © 2010 ACEEE DOI: 01.IJEPE.01.03.70