Abstract

Parkinson’s disease can present with both physical and emotional symptoms, which can vary widely between individuals. These symptoms may include prob- lems with walking, shaking or tremors, difficulty maintaining posture, slowed movement, instability while walking, feeling tired or fatigued, experiencing feelings of sadness or hopelessness and have trouble in sleeping. It is impor- tant to diagnose Parkinson’s disease as early as possible.This can greatly aid in managing the disease effectively, particularly in the healthcare industry.In this study, we used Convolutional Neural Networks (CNN) to diagnose PD and CNN-Long Short-Term Memory (LSTM) network using Hoehn & Yahr rat- ing scale leads to severity rating prediction. Our dataset was obtained from Physionet which consists gait patterns of 93 PD patients and 73 healthy con- trols. We captured gait patterns using eight Force-Sensitive Resistors (FSR) located under the foot, which measured the Vertical Ground Reaction Forces (VGRF). To extract spatiotemporal features, including the swing phase, stance phase, and gait phase (a combination of swing and stance phase), we have used proposed classifier. We used spatial features which are measured across spatio-temporal features to predict gait abnormality. We had implemented the feature extractor using deep learning during the training process, which is a more efficient approach than manual implementation. We used the CNN for PD classification and the CNN-Long Short Term Memory for severity scale prediction based on the widely used Hoehn and Yahr scale.