Analysis And Application Of Motor Unit Activities Derived From Surface Electromyography

Electromyogram(EMG)is an electrophysiological signal generated with muscle contraction.It carries important information of neuromuscular system(NMS)concerning neural control and muscle contractions formulating various movements.Thus it has been widely applied in motor health evaluation and motor intention decoding.Motor unit(MU),which consists of an alpha motor neuron as well as multiple muscle fibers innervated by this neuron,is the basic functional unit and final pathway of NMS.Essentially,EMG signal is composed of action potentials generated by a series of activiated MUs,and these MU activities are spatially and temporally superimposed at a recording electrode site.Surface EMG(sEMG)refers to a noninvasive EMG technique that records signals from the surface of the skin through surface electrodes,and it posses the advantages of non-invasive measurement,safety,ease of use and capability of reflecting the global activities of the muscles.Therefore,it becomes the most commonly used technique in the decoding of motor neural control information.Although sEMG can be an appropriate measurement to detect the whole group of MU activities,it has limited selectivity to examine individual MU activities due to severe noise interference and its high superimposition of MU activities.Consequently,it is quite challenging to accurately decode the motor neural control information from sEMG signal.Therefore,how to extract MU characteristics-related information from sEMG signals becomes the most important challenge and key problem for the promotion and application of the sEMG technique.The objective of this study is to develop non-invasive sEMG approaches for analyzing the MU characteristics.A series of methods were proposed towards two application directions.On one hand,methods for non-invasive examination of complex neuromuscular changes were developed,providing quantitative evaluation indicators and approaches for exploring relavant pathologies towards diagnoses at an individual subject level.On the other hand,methods for accurate interpretation of individual MU activities after the cutting-edge sEMG decomposition were developed to achieve precise MU-based muscle force estimation.This provides useful solution to continuous and proportional myoelectric control,towards establishment of natural and effective human-machine interfaces.The main contributions and innovations of this dissertation are summarized as follows:(1)A series of quantitative diagnostic methods and indicators are presented for examining complex neuromuscular changes at the MU level.Three primary methods were reported,including clustering index(CI),sample entropy(SampEn),and a method based on the percentage of determination(DET%)in recurrence quantification analysis(RQA)and median frequency(MDF)named DET%-MDF method.These methods were applied to stroke data for non-invasive diagnosis of complex neuromuscular changes after stroke.Through the sEMG collection and signal analysis,the indicators exhibited abnormal distribution patterns in subjects with stroke compared to healthy control subjects,and each method was able to produce a quantitative decision for an individual subject.The findings helped to reveal complex neuromuscular alterations at the single MU level in various types and degrees in muscles following stroke.This work provides quantitative methods for the clinical diagnosis of neuromuscular disease in a non-invasive manner,offering novel sEMG tools for exploring the pathological mechanism of MU alterations across individuals and disease progress.(2)In view of the current research status that the diagnostic efficacy and sensitivity of sEMG indicators to specific MU alterations remains unknown,a neuromuscular changes simulation model coupled with various MU alterations was established.This model was modified from an existing sEMG simulation model with additional simulated approaches regarding specific neuromuscular changes.Through this model,the diagnostic efficacy and sensitivity of the CI indicator,the SampEn indicator and the DET%-MDF indicator to different MU alterations were examined.It was found that these indicators exhibited different distribution patterns for different MU alterations,and they presented different sensitivity degrees to different MU alterations.These three indicators presented considerable different diagnostic decisions on various cases and showed different sensitivity degrees for specific types of MU alterations.Such differences can be utilized to provide information complementation in the clinical diagnosis using sEMG techniques,thus enhancing its ability to discriminate specific MU alterations.Generally,this study investigated the diagnostic principles of sEMG indicators through a forward model simulation approach,and confirmed its diagnostic efficacy and sensitivity to neuromuscular changes especially for MU alterations,thus providing valuable information to interpret sEMG diagnostic result for clinical use.(3)A method was proposed for decoding microscopic neural drive information represented by individual MU activities.It was designed to overcome the bottleneck of applying exsiting sEMG decomposition technique towards practical use.On this basis,a novel muscle force estimation method was established by incorporating a physiological model with the machine learning approaches.Specifically,a non-strict/general cross-trial MU tracking strategy was designed based on the MU clustering and classification.In this process,the consistency between motor unit action potential(MUAP)waveforms was measured and MUs with similar spatial distribution of MUAP waveforms were recognized and classified into several predefined categories.After various MUs were categorized,the MU category distribution pattern of each analysis segment was calculated.The force level of each muscle contraction segment can be further identified through a quantification procedure of the MU category distribution pattern using support vector machine algorithm.Through this way,the gain factor to derive the actual muscle force intensity was obtained,which helped to overcome the partial loss of neural drive information resulted from incomplete decomposition.The above processes facilitated interpretation of microscopic neural drive information.In order to further decode the muscle force,the twitch force model was employed to convert the MU activities to twitch force trains,and the twitch force trains were re-organized according to the aforementioned MU categories.The spatial information of each MU category and MU category information was characterized by the encoder of a deep network with a specific structure,and the long-short-term memory block was integrated to characterize the temporal correlation for estimating the normalized force curves.The proposed method achieved high-precision muscle force estimation,and it provided a feasible solution for the decoding of MU activities after sEMG decomposition facing the difficulty in cross-trial MU tracking and incomplete decomposition,thus contributing to the establishment of more advanced human-machine interfaces.