Cross Spectrum Aided Deep Feature Extraction Based Neuromuscular Disease Detection Framework

Abstract

In this letter, a novel neuromuscular disease detection framework employing cross-wavelet spectrum (XWS) based deep feature extraction is proposed. In joint time–frequency analysis, cross-wavelet transform (XWT) is an important tool which provides the degree of correlation between two time series in both time scale and time–frequency plane. In this present contribution, XWT of healthy, myopathy, and amyotrophic lateral sclerosis electromyography signals was initially done with a reference healthy signal to obtain their respective XWS. The obtained cross-spectrum images of different electromyography signals were fed to a pretrained deep residual network for the purpose of feature extraction. Then, we selected the most discriminative feature set using one way analysis of variance test and false discovery rate correction. Finally, using the selected deep features, classification of Electromyography (EMG) signals was done using different benchmark machine learning classifiers. Four (three binary and one multiclass) classification tasks were performed and we observed that the proposed method delivered very high classification accuracy for all cases, which can be implemented for real time neuromuscular disease detection.