Classification of Myopathy and Normal Electromyogram (EMG) Data with a New Deep Learning Architecture

Abstract

Electromyograms (EMG) are recorded movements of nerves and muscles that help diagnose muscles and nerve-related disorders. It is frequently used in the diagnosis of neuromuscular diseases such as myopathy, which causes many changes in EMG signal properties. The most useful auxiliary test in the diagnosis of myopathy is EMG. Therefore, it has become imperative to identify computer-assisted anomalies with full accuracy and to develop an efficient classifier. In this study, a new machine learning method with a deep learning architecture that can score normal and myopathy EMG from the EMGLAB database is proposed. Using the discrete wavelet transform Coiflets 5 (Coif 5) wavelet, the EMG signals are decomposed into subbands and various statistical features are obtained from the wavelet coefficients. The success rates of the decision tree C4.5 algorithm, which is one of the traditional learning architectures, and the Long Short-term Memory (LSTM) algorithm, which is one of the deep learning architectures, were compared. Unlike the studies in the literature, with the LSTM algorithm, a 100% success rate was achieved with the proposed model. In addition, a real-time approach is presented by analyzing the test data classification time of the model.