Differential and directional body coupling analysis in Parkinsonian gait

Parkinson's disease (PD) is a progressive neurodegenerative disorder primarily characterized by motor impairments and altered whole-body coordination. Objective and accessible approaches for quantifying inter-limb and cross-directional coupling during gait remain limited in clinical settings. Using a prospectively collected dataset of individuals with PD and healthy controls (HC), we propose a comprehensive analytical framework based on human pose estimation to extract time-series kinematic signals from key body landmarks. Coupling strengths between body segments were quantified to characterize whole-body coordination patterns. Statistical filtering identified discriminative body coupling factors that significantly differentiated PD from HC. The identified factors demonstrated clear and consistent alterations in coordination patterns across groups. Furthermore, multiple machine learning classifiers validated their discriminative capacity, achieving robust classification performance. Correlation analyses between the top 10 pre-specified directional body coupling factors and MDS-UPDRS III scores demonstrated modest but statistically significant associations in the full PD cohort (n = 121), with 8 of 10 factors surviving FDR correction (ρ=0.27-0.31). These associations suggest that directional body coupling metrics capture clinically relevant variation in motor impairment severity beyond binary PD versus HC discrimination and may serve as potential digital biomarkers of PD-related motor dysfunction. Subtype-stratified analyses were conducted only as exploratory supplementary analyses, and larger, more balanced subtype cohorts are required before subtype-specific conclusions can be drawn. Overall, body coupling metrics provide reliable biomechanical markers of PD-related gait impairment and support the clinical potential of this non-invasive and cost-effective pipeline for early diagnosis and longitudinal monitoring.