Parkinson's Disease Detection Using CNN Architectures withTransfer Learning

Abstract

Nowadays the most common and incurable neurological disorder disease is Parkinson's disease (PD). This incurable disease is growing terribly. This study determines PD patients on the basis of fine motor symptoms using sketching. We proposed a system where we use spiral and wave sketching that can identify either the sketch is from a PD patient or not. Our experiment was done on a dataset consisting PD patient and Healthy (without PD) control group. We applied a deep learning approach Convolutional Neural Network (CNN) to determine PD infected patients and healthy (without PD) control group. We experimented on two CNN models - Inception v3 and ResNet50, with transfer learning method. The proposed system achieved 96.67% accuracy on the Inception-v3 model with spiral sketching.