Computer-vision based diagnosis of Parkinson's disease via gait: A survey

Abstract

Parkinson's Disease (PD) being the second most hazardous neurological disorder has developed its roots in damaging people's quality of life (QOL). The ineffectiveness of clinical rating scales makes the PD diagnosis a very complicated task. Thus, more efficient systems are required to perform an automated evaluation of PD for its earlier detection and to enhance life expectancy rate. Gait based clinical diagnosis can provide useful indications regarding the presence of PD. From recent years, computer vision-based (VB) analysis is in great demand and seems to be highly effective in PD inspection. The objective of this article is to systematically analyze the applications of computer vision in PD evaluation through gait. This paper surveys the VB PD gait acquisition modalities as well as provides a concise overview of preprocessing techniques. The study presents a description of PD related gait features, extraction and selection methods used for PD analysis. A number of machine learning techniques for classification of PD and healthy gait are also discussed. This article extensively surveys PD gait datasets considering data from 1997 to 2018. Also, several research gaps in existing studies have identified that need to be addressed in the future. At last, an outline of the proposed idea is given that can cope up with the related issues and can lead to quality VB PD gait investigation.