Feature Extraction to Classify Parkinsonian Tremor in EMG Signals

Abstract

Accurate early diagnosis of PD remains challenging due to lack of specific biomarkers. Recent literature indicates multimodal signal processing approach to characterize PD markers. PD tremor reflects in lower EMG signal frequencies with some overlap with other types of tremor response. Therefore, frequency alone can’t differentiate various tremors for diagnosis of PD. This work aims to develop robust algorithms for extracting PD related features in EMG. Data are collected from online databases. Results on different EMG signal processing shows certain variation but not very distinct to differentiate PD from the counterpart. Calculated Power of Wavelet cross-scalogram is different in PD subjects compared to normal subjects. Artificial Neural Network based machine learning classifiers are applied to identify PD response from normal based on EMG data with 75% and 92% accuracy during work and rest respectively, which can be used for early PD diagnosis. © 2025 Elsevier B.V., All rights reserved.