Early Detection of Parkinson’s Disease Through a Combination of Facial Expression and Hand Movements and Asymmetry-Related Features Using Machine Learning

Abstract

Parkinson’s disease (PD) ranks among the most prevalent neurodegenerative disorders worldwide. Despite substantial research focused on identifying early motor symptoms in PD, accurate detection remains challenging. This study introduces a novel multimodal framework that integrates facial expression analysis with hand movement kinematics and uniquely incorporates quantitative assessment of hand movement asymmetry for early PD detection and disease staging. We utilized data from 109 participants, including 93 PD patients across Hoehn and Yahr stages 1–3 and 16 healthy controls, resulting in 549 total recordings. Our findings demonstrate that a combined feature set incorporating facial expressions, hand movements, and asymmetry-related features consistently outperforms unimodal approaches across all classification tasks. The framework achieved balanced accuracy scores of 90.1% for early PD detection, 92.2% for general PD detection, and 66.4% for distinguishing between early and moderate PD stages. Additionally, three-class classification (healthy controls vs. early PD vs. moderate PD) attained 74.0% balanced accuracy. All results were validated using three-fold nested cross-validation.