Parkinsonian Gait Severity in Older Adults with Dementia Using Natural Gait Video Analysis

Abstract

Parkinson's disease presents challenges in assessing severity, especially in older adults with dementia. This study introduces a novel approach using vision-based systems for gait analysis [1]. 2D human pose tracking in video data from a dementia unit extracted gait features. Correlation between these features and parkinsonism severity, assessed by UPDRS, was observed [6]. Five machine learning algorithms were evaluated, with KNN demonstrating superior performance [9]. Random Forest and Extra Trees also performed well [10], while Logistic Regression faced challenges [11], [12]. XGBoost showed exceptional predictive ability [8]. The study underscores the effectiveness of these algorithms in non-invasively identifying gait abnormalities, particularly highlighting KNN's adeptness with high-dimensional features [9]. This research advances healthcare diagnostics by showcasing machine learning's potential in early detection and ongoing monitoring of neurodegenerative disorders, promising improved medical outcomes.