High Density Surface EMG Decomposition Based On Blind Source Separation

The motor unit(MU)is the smallest organizational and functional element of the neuromuscular control system.After the filtering effect by muscles,subcutaneous fat and skin tissues,the motor unit action potential trains(MUAPTs)of recruited motor units during muscle excitation sum up to form the surface electromyogram(SEMG)at the collecting electrodes on the surface of the skin.SEMG decomposition is the pro-cess of breaking down the multiunit surface EMG signal into the contributions of the underlying MUAP trains.Surface EMG decomposition makes it possible to acquire the recruitment and firing time information and the MUAP waveform information of indi-vidual motor unit,which can facilitate observing the active status of individual motor unit in the central nervous system and understanding the neural control strategy.There-fore,SEMG decomposition is of great significance and value for clinical diagnosis and the study of neural control mechanism of the neuromuscular system.Due to the relative low SNR level and heavy superposition,the decomposition of surface EMG signal is a very difficult task.The rise and application of high-density flexible surface electrode array make it possible to decompose the surface EMG signal.Based on blind source separation(BSS)technology,this study proposed a novel frame-work for the high-density surface EMG decomposition and then validates the proposed framework.The main research contents and contributions of this study are summarized as follows:(1)A new FastICA based framework for high-density surface EMG decomposition called Progressive FastICA Peel-off(PFP)was proposed.The whole decomposition process of the framework can be considered as the process of gradually expanding the set of motor units firing spike trains.The proposed framework is based on the linear convolution model.FastICA is used to initially search motor unit spike trains.Then a peel-off procedure is employed to estimate the MUAP waveforms of the identified motor units and peel them off from the original signal,which can mitigate the effects of the already identified motor units on the FastICA convergence,so more motor units can emerge by repeatedly applying FastICA on the residual signal.In order to ensure the accuracy of each peel-off procedure,Constrained FastICA is applied to assess the reliability of extracted spike trains and correct possible erroneous or missed spikes.All these factors together result in a good decomposition performance.(2)An automatic implementation of the PFP framework was proposed,called Automatic Progressive FastICA Peel-off(APFP).APFP improves the non-automatic part of the PFP framework.Specifically,skewness combined with iterative threshold method is proposed to automatically set the threshold to extract the motor unit firing spike trains from the output of FastICA.Meanwhile,an unsupervised valley-seeking clustering method is introduced to cluster the extracted spikes from output of FastICA,which facilitate increasing the decomposition yield and accuracy.Some effective pa-rameter constraints and a pending strategy are used to final screen the extracted motor units firing spike trains.All the improvements in automation details are well integrated with the PFP framework and achieve similar performance to manual decomposition,re-sulting in an automatic?efficient and accurate surface EMG decomposition framework.(3)A series of comprehensive and detailed validation of the PFP/APFP framework were proposed by this study.The performance and accuracy of the proposed framework were quantitatively analyzed and validated using simulated surface EMG signals;the two-source validation of the APFP framework was performed on the signals of normal subjects and patients with neuromuscular diseases;the decomposition performance was compared between PFP framework and the wide-used Convolution Kernel Compensa-tion(CKC)algorithm.All the experimental results demonstrate that the PFP/APFP framework is a reliable and accurate surface EMG decomposition method.This study systematically completed the design and validation of automatic surface EMG decompos.ition framework and provided some new ideas for the study(including algorithm design,validation approaches and applications)of surface EMG signal de-composition.This study was supported by the National Natural Science Foundation of China under grant 81271658 and 61771444.