Automatic Diagnosis of Parkinson’s Disease Using EMG Signals from Different Hand Movements

Abstract

Parkinson's disease (PD) is the second most common neurodegenerative disease that affects wide range of productive individuals worldwide. It is neurological disorder characterized by muscle rigidity, tremors, uncontrolled movement, change in speech and sleep disorders. These problems arise because of loss of substance called dopamine that act as a messenger between two brain areas, the substantia nigra and the corpus striatum to produce smooth and controlled movements. The common approach to diagnose PD is through clinical assessment of the patient, which is highly subjective and time consuming. Electromyography (EMG) recordings can be used for diagnosis of PD. However, highly experienced experts are required to interpret the signals, which is complex and time-consuming procedure. These manual procedures are prone to error and may lead to misdiagnosis. Many researchers designed automated systems to solve this problem but they have their own pitfalls such as, achieving limited accuracy, using small number of data sets and sticking with binary classification. In this research, a reliable, accurate and automatic system for early detection and classification of PD using EMG signals is developed. A total of 1000 EMG signal data were collected from flexor carpi radialis and biceps muscles of 15 PD patients and 10 healthy control subjects at JUMC using SCU-7 EMG system. And the signal was analyzed using MATLAB 2018. Data augmentation for collected signals was performed by adding white noise with SNR value of 90 and 100. The raw EMG signal was denoised by applying an infinite impulse response notch and Butterworth filters. Then features in time and frequency were extracted and feature reduction algorithm was applied to discard irrelevant features. Finally, the selected features were sent to the model to train the fourclass classification. Support vector machine was used to classify the features of the signal into four (normal, early, moderate and advanced) classes for each hand movements. The performance of the system was evaluated at testing phase and a promising result has been found. 90%, 91.7%, 95% and 96.6% overall classification accuracies were obtained for elbow flexion by 900 without load, elbow flexion by 900 with load, touching shoulder and wrist pronation, respectively. The proposed system will be used as a decision support system for physicians, especially those in low resource setting by detecting PD at early stage and classifying it’s level. This will have a great impact in reducing the disease progression and the mortality rate due to PD.