Study On Decomposition Of Surface EMG Into Motor Unit Action Potential Trains Based On Prior Templates

Surface electromyography (sEMG) signal is the result of composite superpositionof motor unit action potentials (MUAP), which are generateed by all recruited motorunits when the muscle is excited, it contains lots of information regarding MUrecruitment and MUAP firing. The decomposition of an sEMG signal is the process ofdecompose sEMG signals into their constituent motor unit action potential trains(MUAPTs), the acquired information of MUAP firing contributes to the research of theneuromuscular regulation mechanism, has a good application prospect in the fields ofclinical medicine and prosthesis control, rehabilitation, sports medicine and so on.At present, the technology of the sEMG signal decomposition can be roughlydivided into two types: One is Blind source separation algorithm or SystemIdentification, the other is based on MUAP shape. Due to the basic assumptions of theformer are not always satisfied when applied to sEMG signal decomposition, and Thecurrent decomposition effect is not ideal, so we designed an algorithm of sEMG signaldecomposition based on the shape of MUAP.Based on the analyzing and summarizing of the related literature, in combinationwith the characteristics of MUAPs that they are common for dual phase or three phasewaveform,4kinds of MUAP waveforms with variable time during and amplitude werefitted using Hermite-Rodriguez function; In order to reduce the effects of sEMG signalsequence segmentation along the time on stacked waveform recognition, with the“bigger first” principle, MUAP was automatically extracted from the whole sEMGsignal one by one based on the “size principle” of MU recruitment. It is noticing that,there is no adding assumption of MUAP firing rule during the MUAP waveformrecognition and just limited the scope of MUAP firing frequency. In addition, theproposed sEMG signal decomposition algorithm can decompose the single channelsignal independently. It overcomes the dependence on the information of the otherchannels.In order to obtaining high SNR ratio sEMG signal for decomposition, three stepsof preprocessing of raw sEMG signals were adopted. Firstly, the elliptical bandpassdigital filter was adopted to remove the low and high frequency noise outside of thesEMG signal main frequency zone. Second, the power frequency interference wasseparated based on the fast independent component analysis (FastICA) algorithm. Lastly,coif2mother wavelet that has the advantage of orthogonality, compactly supportting and approximate symmetry was used for wavelet packet denoising. The decompositionresult of real sEMG signals shows that, the proposed preprocessing algorithm can notonly remove the power frequency interference and other noise, but also keep thesharpness of MUAP waveform.Due to the lack of priori knowledge of constituent motor unit action potential trainsin sEMG signals, specially designed accuracy validation protocol is often needed.Therefore, simple sEMG signal model is constructed for the proposed algorithmaccuracy validation. Simulated sEMG signals (5s) with different SNR(5dB/10dB/15dB/20dB) and different degree of superposition (0%/10%/20%/30%)were decomposed,20groups for each case. The decomposition result of SimulatedsEMG signals shows that, the accuracy (90.94%1.27%) of this decompositionalgorithm is higher when the SNR is higher (SNR=20dB) and degree of superposition islower (10%).In order to verify if there is some relation between the MUAPTs and thecorresponding neuromuscular activity, the proposed algorithm was applied to thedecompose in the multi-channel (12channels)sEMG recorded from8subjects (3groupsper subject) flexor digitorum superficialis muscle in different fingers activity mode. Thedecomposition result of single channel sEMG indicate that,the main MUAPts of highforce level sEMG can be effectively distinguished and classified; Statistics show that，with the increasing of force level，the number of MUAP increases; The proportion ofdifferent size of MUAP varies significantly with active finger and force level. Theexperimental results preliminary proved that it is feasible to gradually extract MUAPsfrom sEMG signals using prior templates，it also provides a new method for sEMGdecomposition and the evaluation of MU firing patterns.