Prediction of Parkinson’s disease based on feature selection and classification of dopamine transporter scan of brain using deep learning architectures

Abstract

Parkinson’s disease (PD) is a neurological disease that produces uncontrollable movements and a variety of other symptoms. It can be difficult to make an accurate PD diagnosis since the signs and symptoms, especially early on, might be mistaken for other medical diseases or physiological changes associated with normal aging. This research proposed novel technique in predicting PD based on dopamine transporter scan (DaTscan) images of brain using deep learning techniques. Here the aim is to collect the historical data and live DaTscan image of patients with symptoms of PD and predict disease. Initially input data have been pre-processed for image resize, noise removal and smoothening. Then the processed image has been selected based on their features using kernel-based deep convolution neural network (KDCNN). The selected deep features have been classified using reinforcement Q-learning-based neural networks (RQLNNs) to predict the presence of PD. Here experimental results show feature-selected and classified output of DaTscan brain image using the proposed model. For MRI image dataset, the proposed technique obtained accuracy of 97.5%, precision of 93%, recall of 82% and F-1 score of 87%. The proposed technique obtained accuracy of 98%, precision of 93%, recall of 80% and F-1 score of 88% for DaTscan dataset.