Detection and Predictive analytics of Parkinson’s disease employing Tremor analysis and Deep Learning Algorithm

Abstract

Among medical diseases affecting human races and communities globally neurological diseases are dominant one. Latest investigations have revealed statistics that more than seven million globally and one million in United States which is a developed country are suffering from Parkinson’s disease. In fact studies reveal that predicted statistics is so large that medical communities are underprepared to treat these patients affected in near future .Parkinson’s disease (PD) is  a disease of nervous system. This disease is caused due to paucity of dopamine in human brain and effect being on the daily routine activity profile of affected person. Symptoms abstractly remain same among communities but may vary slightly among genders and age groups. Onset of Parkinson’s disease detection and predictive analytics is an innovation in medical research. Clinical diagnosis of the disease as practiced by clinicians globally is by examination of tremor signals. Tremor quantification methods encompass clinical analysis of drawn figures, computerized signal analysis in time/frequency domains and using tremor rating scales. Objectively to complement clinical research, varieties of Tremor rating scales are standardized for assessment of intensity of disease among neurological patients. Primitive scale employed being Fahn-Tolosa-Marin Tremor Rating Scale (TRS). It is a scale rated from 4 (serious tremor) to 0 (no tremor) based on amplitude of tremor which happens to be a 5-point scale. This study scope is development of a hardware tremor detection module using Raspberry Pi processor interfaced with flex and accelerometric sensors attached on wrist of subjects from which dataset of tremor is built. To complement same machine learning algorithms namely Random Forest and Convolutional neural network (CNN) algorithms are developed in Python for Predictive analytics of onset of disease. Comparison of algorithms indicates that prediction accuracy is 96.97% of CNN algorithm.