Investigation Into the Nature Of Subtle Features of Emg Signals in Healthy and Disease Conditions

Abstract

A bioelectric signal is a collective electrical signal acquired from any biological event newlineof a living being that represents a physical variable of interest. This signal is usually newlinea function of time and is describable in terms of its amplitude, frequency and phase. newlineElectromyographic (EMG) signals are one of the best known bioelectric signals, newlinewhich are extensively applied in the estimation of neurophysiological characteristics newlineof skeletal muscles. The EMG signal is a biomedical signal that measures electric newlinepotentials generated in muscles during its polarization, representing neuro-muscular newlineactivities. newlineThe primary aim of this thesis is to investigate the application of time domain, non- newlinelinear, fractal and multifractal feature extractors from EMG, to characterise EMG and newlineto study the nature of newly extracted features to extend it to the classification of EMG newlinesignals for various neuromuscular conditions. These proposals combine handcrafted newlinediscriminative features with powerful Machine Learning (ML) algorithms. In the newlinesecond part of the thesis, the strength of learned features towards the characterisation newlineof EMG in the Deep Learning domain are explored. newlineIn the first section of the thesis, a systematic feature extraction technique for newlineclassifying neuro-muscular diseases is proposed. This proposed scheme introduces newlinemodified CENTRIST (CENsus TRansform hISTogram) for 1-dimensional signals newlineas the feature descriptor for EMG signals. The proposed algorithm encodes dis- newlinecriminative structural properties of the EMG signal of different classes and is vigil newlinein accounting for abrupt transitions in the signal. k-nearest neighbour (k-NN) and newlineSupport Vector Machine (SVM) classifiers are employed for classification. The crux newlineof this work is to formulate a feature extraction technique to distinguish the structural newlineand topological properties of pathological and non-pathological EMG signals. The proposed algorithm is validated using three publicly available datasets.