Eigenvalue decomposition based analysis and classification of electromyogram signals

Abstract

In the rst proposed work, motor unit action potentials (MUAPs) are extracted from electromyogram (EMG) signals. The MUAPs are then decomposed using improved eigenvalue decomposition of Hankel matrix (IEVDHM) technique and the correlation based features, correntropy (CORR) and cross information potential (CIP) are applied on it. Kruskal-Wallis statistical test is applied to see the di erence in the computed values of features from both the classes of signals. The results obtained after the analysis process shows that this decomposition technique is better and provides statistical signi cant di erence in the normal and amyotrophic lateral sclerosis (ALS) EMG signals. The second methodology proposed in this thesis is to classify the ALS and normal EMG signals. This classi cation is achieved by representing the MUAP in the timefrequency (TF) plane. The eigenvalue decomposition of Hankel matrix and Hilbert transform (IEVDHM-HT) technique has been used to convert the MUAP signals into TF domain. Further the mean has been computed by running the window over the TF distribution of signal. This computed mean has been used as a feature to use into classi ers. Three di erent classi ers - random forest, random tree and J48 have been used to classify ALS and normal EMG signals. The proposed method has provided good classi cation performance in the classi cation of normal and ALS EMG signals.