INDEPENDENT COMPONENT ANALYSIS OF ELECTROMYOGRAPHIC SIGNAL ABSTRACT

Transcription

1 ISCA Archive Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA) 2 nd International Workshop Florence, Italy September 13-15, 2001 INDEPENDENT COMPONENT ANALYSIS OF ELECTROMYOGRAPHIC SIGNAL Ajay somkuwar* Sujoy K. Guha** and Sudhir Atreya* * Instrument Designing and Development Center Indian Institute of Technology, New Delhi ** Center for Biomedical Engineering Indian Institute of Technology, New Delhi and All India Institute of Medical Sciences, New Delhi. ABSTRACT Electromyography is a valuable tool in many clinical analyses, as it can give the clinician an accurate representation of what the muscles are doing to contribute to the desired task. For functional EMG the surface electrodes have the advantage of convenience and comfort. The major disadvantage to surface electrodes are cross talk and low level signal reception. Their adverse effects complicate the definition of muscle timing and relative intensity of the activity. During period of low muscle activity there is the possibility that the EMG signals may include signals from musculature other than the muscle of interest. A recently developed linear transformation method is independent component analysis (ICA), in which the desired representation is the one that minimizes the statistical dependence of the components of the representation. In order to define suitable search criteria the approximation of nagentropy is used as a contrast function for minimizing the statistical dependence between the component. The fast ICA algorithm given by Aapo Hyvarinen identify the independent source by maximizing the joint entropy of a set of output signals derived form the Rectus femori (RF) and semimembranous (S) muscles of lower limb during pedaling action of leg. The signal of RF severely affected by sartorius and Tensor facia lata and the S muscle includes signals from Sartorius and Semitendenus together with electrical noise is separated via fast ICA algorithm. Result illustrating the good performance of the method. INTRODUCTION When detecting and recording the Electromyographic (EMG) signal there is main issues of concern that influence the fidelity of the signal is the cross talk and noise. In general noise and cross talk is defined, as the electrical signals that is not part of the wanted EMG signal. It is well established that the electromyographic signal is stochastic in nature. The amplitude of the signal can range to 0 10 mv (peak to peak) or 0 to 1.5 mv (rms.). The usable energy of the signal is limited to Hz frequency range, with the dominant energy being in Hz range. Recently Independent Component Analysis (ICA) has acquired attention because of its potential application in medical signal processing. The goal of ICA is to recover independent sources given only sensor observation that are unknown linear mixture of the unobserved independent source signals. More methodically ICA, is a statistical technique that represents a multidimensional random vector as a linear combination of non-gaussian random variables that are as independent as possible. ICA has many applications in data analysis, source separation, MAVEBA 2001, Firenze, Italy 217

2 and feature extraction. Suppose we are given two linear mixtures of two source signals, which we know to be independent of each other. The object is then to determine the source signals given only the mixtures. Putting this into mathematical notation, we model the problem by x = As where s is a two-dimensional random vector containing the independent source signals, A is the two-by-two mixing matrix, and x contains the observed (mixed) signals. A first step in ICA is to perform centering (subtracting the mean) and whiten the data. This means that we remove any correlation in the data. Again putting the words in mathematical terms, we seek a linear transformation V such that when y = Vx we now have E{yy'} = I. This is easily accomplished by setting V = C -1/2, where C = E{xx'} is the correlation matrix of the data, since then we have E{yy'} = E{Vxx'V'} = C -1/2 CC -1/2 = I. After sphering, the separated signals can be found by an orthogonal transformation of the whitened signals y (this is simply a rotation of the joint density). The appropriate rotation is sought by maximizing the non-normality of the marginal densities (shown on the edges of the density plot). There are many algorithms for performing ICA, but the most efficient to date is the fast ICA (fixed-point) algorithm which was developed by Aapo Hyrenen (Finland, 1999). The concept of ICA may actually be seen as an extension of PCA, which can only impose independence up to the second order and consequently define direction that are orthogonal. Potential application of this method includes data analysis, localization of sources, blind signal separation and deconvolution. METHOD Surface electromyograms were recorded from RF and S muscles of the lower limb. The myoelectric signal was recorded using a laboratory made and well-calibrated electromyogram amplifier. The electrodes were positioned over the distal half of the muscle belly such that contact surface were aligned longitudinally to muscle fibers. Electrode sites were prepared by cleaning the skin with isopropyl alcohol and shaving the hair, when necessary to insure good contact centrally punched electrodes (inter-electrode distance 25 mm., diameter 10 mm) were attached using adhesive pads and electrolytic gel. A common reference electrode was placed on the distal end of the muscle belly. The experimental protocol conducted for one-hour period for each subject, consisted of measurement of electromyogram during pedaling of rickshaw against an applied load of 80 kg. Subject was instructed to maintain a 40-rpm cadence by following a metronome. Once a steady cadence was achieved, 15 second of electromyogram were collected and stored on an audiocassette using TEAC-R-60 data recorder. The recorded signal was converted into 16 bit mono digital signal at 8000 sampling rate. The Cool-Edit software developed by m/s Syntrillium software company, MATLAB 5.3 and fastica_2.1 software were used to process the signal. RESULT The plot below shows the result after ten step of the Fast ICA algorithm. The source signals in this example were a signal from desired muscles (RF or S) with contamination from line frequency, and the muscle neighboring to the above muscles. MAVEBA 2001, Firenze, Italy 218

3 Figure-1 shows the independent component analysis technique applied to the signal recorded from Rectus femori muscle. The first plot shows the signals mixture and remaining are the line frequency, RF signal, Tensor facia and sartorius signal components. MAVEBA 2001, Firenze, Italy 219

4 Figure-2 shows the independent component analysis technique applied to the signal recorded from Semimembranus muscle. The first plot shows the mixture of the signals and remaining is the line frequency, Semimembranous, Semitendenous and sartorius muscles signal components. DISCUSSION It is desirable to obtain EMG signals that contain the maximum information from the selected muscle and minimum amount of contamination from cross talk and electrical noise. Which complicates the determination of onset and cessation times of muscle action; thereby confusing the precise identification of muscle phasing, which is a common clinical objective. Thus the minimization of the signal to noise ratio should be done with minimal distortion to the EMG signal. Therefore it is important that any detecting and recording devices process the signal linearly. In particular the signal should not be clipped, that is the peaks should not be distorted and no unnecessary filtering should be performed. Because the power line radiation (50-60 Hz) is a dominant source of electrical noise it is tempting to design device that have a notch- filter at this frequency. Theoretically this type of filter would only removes the unwanted power line frequency however practical implementations also removes portions of the adjacent frequency components. Because the dominant energy of the EMG signal is located in the Hz range MAVEBA 2001, Firenze, Italy 220

5 the use of notch filter is not advisable. When there are alternative methods of dealing with power line radiation. Research studies have demonstrated that double differentiation can reduce the cross talk to half or less. The tensor facia and Sartorius muscle surrounds the RF muscle while the semitendenous and Sartorius contributes to Semimembranous muscle signal. Volume conduction allows wide dispersion of myoelectric signals through the tissues. The thin films of fascia between adjacent muscles present no significant barrier to the myoelectric signals from nearby muscles. Also muscle functions in-groups rather than in isolation. As a result the recording from a synergist may indicate activity in the designated muscle when actually it is quiet. We find that the ICA algorithm can successfully isolate the independent component in observed EMG, which is severely affected by sensor noise and additional low level sources.. The advantages of applying this method in separation of EMG are first; the convergence is cubic under the assumption of the ICA data modal. This is in contrast to ordinary ICA algorithms based on gradient descent methods, where the convergence is only linear. This means a very fast convergence, as been confirmed by simulations and experiments on real data, and second is as contrary to gradient-based algorithms; there are no step size parameters to choose. This means that the algorithm is easy to use. Nongaussianity is here measured by the approximation of negentropy. In the ICA model it is easy to see that the following ambiguities will hold: First we cannot determine the variances (energies) of the independent components and we cannot determine the order of the independent components but despite of this limitations we are able to identifies the nature of the independent component which will help in predicting the contribution of the desired muscle in a particular task. This means the nature of the most interesting component may be observed. It is envisaged that this tool might be useful in antenna array processing i. e separation and recognition of sources from unknown array, where Principal component analysis has been already utilized for many years. Recently ICA technique shown its potential to capture the essential structure of the data in biomedical signal as well as vocal and speech signal REFERENCE A.Hyvärinen. (1999), Fast and Robust Fixed-Point Algorithms for Independent Component Analysis, IEEE Transactions on Neural Networks 10(3): , Astrand, P. O. And Rodahl, K. (1986), Textbook of work physiology, 3rd Ed. (New York: McGraw-Hill) De Luca, C.J. and Merletti, R (1988) Surface EMG crosswalk among muscles of the leg. Electroencephalography and Clinical Neurophysiology. 69: De Luca, CJ (1986) Electromyography. Encyclopedia of Medical Devices and Instrumentation (John G. Webster, Ed.) John Wiley Publisher, Pierre Common (1994), Independent component analysis a new concept, signal processing 36, MAVEBA 2001, Firenze, Italy 221