Detection of Parkinson’s Disease from Hand-Drawn Images Using Deep Transfer Learning

Abstract

Parkinson’s disease mainly occurs in older people and unfortunately no specific cure is available till date. With early detection of this disease and with proper medication, a patient can lead a better life. This imparts the importance for early detection of this disease. In this paper, our aim is to process hand-drawn images such as spiral, wave, cube, and triangle shapes drawn by the patients using deep learning architectures. For computer-based diagnosis of Parkinson’s disease in early stage, deep convolutional neural networks are investigated. In this paper, three approaches are considered. In first approach, all types of images are fed into various pretrained models VGG19, ResNet50, MobileNet-v2, Inception-v3, Xception, and Inception-ResNet-v2 which are trained from the scratch. In second approach, exactly same techniques are being repeated with the exception that fine-tuning has been performed using transfer learning. In third approach, two shallow convolutional neural networks have been proposed. For all the three approaches mentioned above, the experimental work is conducted on two different datasets and the results reflect that the fine-tuned networks VGG19, ResNet50, and MobileNet-v2 from second approach perform better than the rest of the models with accuracy of 91.6% and 100% for dataset 1 and dataset 2, respectively.