Parkinson’s disease detection using a novel weighted ensemble of CNN models

Parkinson’s disease is a common and chronic neu-rodegenerative disorder that results mainly from the death of dopaminergic neurons in the substantia nigra part of the brain. Early and correct diagnosis of Parkinson’s disease is crucial for clinical research, prognostication, and therapeutic purposes. Different clinical methods and tests like INSIT, ISHA articulation assessment, olfactory testing, MRI, etc, can be used to diagnose and verify Parkinson’s disease in any patient. But these tests are costly, tedious, and complex. In recent years, deep learning techniques have shown promising results in various medical applications, including disease diagnosis. In this study, we propose an efficient ensemble learning-based model to assist doctors in detecting Parkinson’s disease effectively and correctly through figure-drawing task images which is a literacy-independent, cost-friendly, simple, and less time-consuming process. Preprocessing techniques are applied to standardize the images and enhance relevant features. We have employed deep transfer learning and have designed a weighted ensemble of three convolutional neural network models, wherein the weights given to three base learners are determined by a novel method. The proposed model has achieved an accuracy rate of about 95% and 90% for spiral and wave figure drawing tasks respectively. The results given by our proposed model are better than individual models and state-of-the-art deep learning methods.