Classification of Electromyogram (EMG) Signal using Time-Frequency Representation for Diagnosis of Amyotrophic Lateral Sclerosis (ALS) Disorders

Abstract

The objective of this ongoing study is to introduce electromyography signal (EMG) in time-frequency representation (TFR) applying spectrogram with optimized window size and continuous wavelet transform (CWT) with optimal daubechies order where four features were extracted from each analysis. In order to qualify the capability of TFR features in separating healthy and amyotrophic lateral sclerosis (ALS) pathology, four useful classifiers namely radial basis function-support vector machine (RBF-SVM), particle swarm optimization-support vector machine (PSO-SVM), linear discriminate analysis (LDA), K-Nearest Neighbor (KNN) were implemented to classify EMG signals. AS result, spectrogram and CWT with optimized window size of 512 ms and 4^th daubechies (db4) order respectively using RBF-SVM classifier presented the highest classification accuracy of 92.3% Followed by PSO-SVM with optimized window size of 512 ms for spectrogram and 7^th daubechies (db7) order in case of CWT, LDA and KNN are the two classifier recommended for db4 because they accurate 87.5% and 81.3% respectively. But in case of spectrogram with optimize window size of 256 ms, LDA was recommended in opposite of KNN where the appropriate window size of 512 ms must considered. Also, the proposed TFR is able to show the nonstationary variations of EMG signals. the features exhibit statistically significant difference in healthy muscle and neuropathic conditions. The combination of RBF based SVM is found to be most accurate in classifying healthy and unhealthy conditions with the extracted features mentioned in this work.