Deep Learning Based Parkinson's Disease Prediction System

Abstract

Parkinson's disease is one among the chronic neurodegenerative (loss of structure or function of neurons) disorders which may result in continuous degeneration of bodily functions that leads to serious disabilities in people performing the simple day-to-day activities, due to defect in dopamine mechanism of the brain cells. A few symptoms of Parkinson's disease are stated here, namely, tremor, slowed movements, impaired posture, loss of automatic movements, change in speech and changes in handwriting. Symptoms of the diseases may vary from one patient to another. Out of the many symptoms, the change in speech is a vital sign for detecting Parkinson's as identified in the patient history. Given these symptoms can be recognized in the patient by investigating his voice data, and where deep learning is more powerful in investigating unstructured data which includes speech and audio signals, therefore, DL is familiar in its use for picking up these signals of the disease. Deep learning algorithm for unstructured data is constructed by multi-layer neurons which are stacked together for classification and feature selection. In order to predict the Parkinson's disease based on patient voice data, the proposed Parkinson's disease prediction system uses Restricted Boltzmann Machine (RBM) algorithm with low resource and large scale speech task. For predicting the disease severity, Restricted Boltzmann Machine (RBM) algorithm makes use of UCI dataset.