Two-stage binary classifier for neuromuscular disorders using surface electromyography feature extraction and selection

Abstract

If surface electromyography (sEMG) can be used to determine neuromuscular disorders, it can diagnose conditions more easily than needle electromyography. In this study, sEMG during maximum voluntary isometric contraction and repetitive exercise was measured, and normal, myopathy, and neuropathy were classified with high accuracy using these signals. First, a two-stage binary classifier model was constructed to classify the patient group and the normal group and categorize the cases assigned to the patient group into myopathy and neuropathy groups. To this end, features related to muscle activity and muscle fatigue were extracted using activity analysis and frequency analysis of the sEMG signal. Since the features for high performance are different for each classifier, the features with statistical differences in the data of each class were selected for each classifier. The selected features and a two-stage binary classifier were distinguished with an accuracy of 86.9%. This shows an accuracy higher than 82.3%, which was found for the two-stage binary classifier without feature selection and 73.9% of the multi-classifier. Through this, the possibility of using sEMG to diagnose neuromuscular disorders was confirmed.

Keywords: Feature extraction; Feature selection; Machine learning; Neuromuscular disorders; Surface electromyography.