Automated Algorithm for Neurodegenerative Disorder Detection Using Gait-Based Features

Abstract

Neurodegenerative disorders (NDD) are a set of illnesses marked by the gradual degradation in the functioning of the central or peripheral nervous system. In recent works, aberrant gait patterns have been explored for detecting NDD like Huntington’s disease (HD), Amyotrophic lateral sclerosis (ALS), and Parkinson’s disease (PD) using quantitative evaluation techniques. In this study, we present a three-step automated technique for identifying ALS, PD, and HD subjects using their gait patterns. The first step is to examine the suitability of various time-domain statistical features extracted from the gait rhythm data for the identification of the disorders, the second step involves identifying the most favorable features, and finally, the third step is to test different machine learning classifiers to achieve the best accuracy in distinguishing different NDD patients from control participants. Validation of the proposed algorithm’s efficacy was done by using the PhysioNet dataset available in the public domain which includes gait rhythm patterns for ALS, PD, HD, and control subjects. Using LOOCV (Leave one out cross-validation) for accessing the performance of different available classifiers, maximum accuracy obtained was $96.55\%$ with Decision Tree (DT) for healthy subjects vs ALS, $93.55\%$ with Support Vector Machine (SVM) for control objects (CO) vs PD, and $94.44\%$ using kNN for CO vs HD.