Automated classification system for normal and ALS EMG signals based on iterative filtering

Abstract

Electromyogram (EMG) signals are proved very useful in identification of neuromuscular diseases. In proposed work, we came up with a new method for the analysis and classification of normal and abnormal EMG signals to identify neuromuscular diseases. First, we have obtained all motor unit action potentials (MUAPs) from EMG signals. Extracted MUAPs are then decomposed using iterative filtering decomposition method. Intrinsic mode functions (IMF) obtained from iterative filtering method, are considered for analysis and classification purpose. Features like Euclidean distance quadratic mutual information (ED-QMI), Cauchy-Schwartz quadratic mutual information (CS-QMI), cross information potential (CIP) and correntropy (COR) are computed for each level of IMFs separately. For the analysis of EMG signals, statistical analysis has been performed by the Kruskal-Wallis statistical test. From the results obtained after analysis process, we have observed that the iterative filtering decomposition method is better and provides statistical significant difference in normal and ALS EMG signals. For classification, the calculated features are given as an input to the three different classifiers: repeated incremental pruning to produce error reduction (JRip) rules classifier, reduces error pruning (REP) tree classifier and random forest classifier for the classification of normal and ALS EMG signals. The results obtained from classification process show that this classification method is very efficient and provided very accurate classification of normal and ALS EMG signals.