Short-term Classification of Neurodegenerative Diseases using Time-series Features Based on Analysis of Gait Dynamics

Neurodegenerative Diseases (NDDs) refer to disorders, leading to the gradual degeneration of neurons in the human nervous system and profoundly affecting individuals' motor and cognitive functions. Early diagnosis and monitoring of the progression of NDDs are of critical importance.

The primary aim of this study is to present an automatic diagnostic method using time-domain analysis of short-term gait signals of two consecutive steps, for the classification of Parkinson's Disease (PD), Huntington's Disease (HD), Amyotrophic Lateral Sclerosis (ALS), and a Healthy Control Group (HC).

This analytical study used datasets from PhysioNet, including 64 five-minute recordings from 13 individuals with ALS, 15 with Parkinson's, 20 with Huntington's, and 16 healthy individuals. To reduce noise and enhance signal quality, a median filter was applied. The preprocessed gait signals were then systematically segmented into consecutive two-step windows, which were used to construct the feature vectors. Finally, the constructed feature vectors were used as inputs for both linear and nonlinear classifiers. The proposed method was evaluated by calculating the accuracy index.

The quaternary classification of ALS vs. PD vs. HD vs. HC using the bag-tree classifier achieved an accuracy of 82.8%. Furthermore, for the binary classifications of HC vs. HD, HC vs. ALS, HC vs. PD and NDD vs. HC, the accuracies were 88.4%, 97.6%, 90% and 89.1%, respectively.

Gait dynamics analysis has a high potential for the rapid diagnosis of neurodegenerative diseases. Short term analysis in two consecutive stages enables the use of the Kinect sensor for gait analysis.