Abstract
Sialorrhea, or excessive drooling, is a prevalent yet frequently under-recognized non-motor symptom of Parkinson's disease (PD). Hypomimia, or reduced facial expressivity, constitutes another significant feature of PD. Although previous studies have suggested a potential clinical association between these two disease features, this relationship has seldom been quantified using artificial intelligence (AI) methodologies. In this study, we sought to characterize the association between hypomimia and sialorrhea in PD using both traditional clinical scales and AI-based video analysis. We conducted a cross-sectional study involving 52 individuals diagnosed with PD. Sialorrhea severity was assessed using the Radboud Oral Motor Inventory for Parkinson's Disease-Saliva subscale (ROMP-saliva), while hypomimia was evaluated via the Unified Parkinson's Disease Rating Scale (UPDRS). Facial video recordings were acquired and analyzed using AI algorithms to extract key facial landmarks. These landmarks were processed into 20 quantitative features representing the mouth, eyes, and combined facial regions. To assess the relationship between facial expressivity and sialorrhea severity, we employed Principal Component Analysis, Canonical Correlation Analysis, and bootstrapping. Clinical rating scales demonstrated a modest correlation between hypomimia and drooling severity (r = 0.368, p = 0.007). In contrast, video analysis revealed moderate correlations between ROMP-saliva scores and features derived from the mouth (mean r = 0.600), eyes (mean r = 0.641), and combined facial regions (mean r = 0.575). These findings support a quantifiable association between hypomimia and sialorrhea in PD and underscore the utility of quantitative facial analysis for the automated detection of under-recognized non-motor symptoms such as drooling.