Parkinson’s disease (PD) is one of the fastest-growing neurodegenerative disorders, characterized by a constellation of motor and non-motor symptoms (1). The cardinal motor features of PD include bradykinesia, rigidity, and resting tremor. Hypomimia, or reduced facial expressivity, constitutes another significant feature of PD. Given that facial expression involves not merely facial muscular movements but also emotional and motivational aspects, hypomimia represents a symptom bridging both motor and non-motor domains (2). In addition, a wide range of non-motor symptoms—including gastrointestinal, genitourinary, sleep, and cognitive disturbances—are increasingly recognized as central characteristics of PD (3).

Sialorrhea, or excessive drooling, is a common non-motor manifestation in PD, affecting approximately 50% of patients, typically in the later stages of the disease (4, 5). The etiology of sialorrhea in PD is multifactorial, likely involving impaired oral motor control and diminished swallowing frequency, rather than hypersecretion of saliva (4). Clinically, sialorrhea is associated with adverse outcomes such as poor oral hygiene, increased risk of aspiration pneumonia, and significant psychosocial burden, such as that caused by embarrassment and social withdrawal (6). Despite its detrimental impact on quality of life, sialorrhea often goes unrecognized or unreported, partly due to its subtle presentation and the reluctance of patients to discuss it with healthcare providers. One survey revealed that fewer than half of affected individuals had ever discussed the symptom with their clinical team (6), highlighting a need for improved recognition and proactive screening.

Like sialorrhea, hypomimia represents a multifaceted symptom with both motor components (impaired facial muscle control and coordination) and potential non-motor elements (altered emotional expression and social communication). However, while sialorrhea typically remains hidden until patients voluntarily disclose it, facial expressions are externally visible and can be readily observed in the clinic (7). Emerging evidence suggests a potential association between hypomimia and sialorrhea in PD (4, 8). This relationship raises the possibility that facial expressivity—or its absence—may serve as a surrogate marker for detecting drooling. However, objective and validated tools to assess these symptoms, particularly sialorrhea, remain underutilized in clinical and research contexts. While semi-quantitative scales such as the Movement Disorder Society–sponsored Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) are routinely used to assess hypomimia, these measures lack the granularity required for automated analyses. For sialorrhea assessment, the Radboud Oral Motor Inventory for Parkinson’s Disease–Saliva subscale (ROMP-saliva) represents the most validated instrument available, yet it entails similar limitations (9).

Technological advances in computer vision and artificial intelligence (AI) have enabled more objective and quantitative assessment of facial movement in PD (10). Automated video-based analysis can provide more precise and reproducible measurements of facial movements compared to conventional clinical scale-based methods (7). Given the reported link between hypomimia and sialorrhea, objective facial movement quantification may serve as a useful tool to indirectly detect drooling and enhance awareness of this symptom.

With such background, the present study aimed to (i) examine the association between hypomimia and sialorrhea using validated clinical scales, and (ii) further characterize their relationship through a quantitative facial video analysis.

This study investigated the association between hypomimia and sialorrhea in patients with PD using an integrated approach combining quantitative facial expression analysis and a validated clinical drooling scale. By applying PCA and CCA to the extracted facial features, we identified moderate canonical correlations between ROMP-saliva scores and both lower facial (mean r = 0.600) and upper facial (mean r = 0.641) features. The CCA revealed five statistically significant canonical correlations across the lower, upper, and combined facial regions. For the lower face, coefficients ranged from 0.799 (p < 0.001) for the first dimension to 0.414 (p = 0.002) for the fifth, possibly reflecting hypomimia-related restrictions in lower-face expressivity and diminished oral motor control, both associated with sialorrhea severity. The upper face showed a comparable CC1 (0.799, p < 0.001) but a slightly higher mean correlation, suggesting a more distributed variance across dimensions, potentially linked to hypomimia-related brow and eye aperture changes. The combined facial region yielded CC1 = 0.722 (p < 0.001) but a lower mean (0.575), indicating that mixing upper- and lower-face variables may dilute region-specific effects. Lower-order canonical correlations (CC4–CC5) explained minimal additional variance, likely reflecting subtle or non-specific associations. Overall, these results support the potential utility of facial motion patterns as surrogate markers for the screening of sialorrhea in clinical PD settings.

Our findings revealed a significant positive correlation between hypomimia and drooling severity in PD, consistent with prior observational studies that have suggested a link between these motor and non-motor manifestations. These findings build upon earlier observations linking sialorrhea to impaired orofacial motor control (4), and suggest potential relationship between facial expressivity and drooling in patients with PD (8). Previous research has employed various assessment tools to quantify these symptoms, including the drooling item from the UPDRS Part II, the Drooling Severity and Frequency Scale, and the facial expression subscore from MDS-UPDRS Part III (20, 21). It is noteworthy that the ROMP questionnaire is currently the only scale endorsed by the Movement Disorder Society for evaluating drooling in PD (22). A prior study using the ROMP-saliva scale was limited to group comparisons based on predefined UPDRS hypomimia cutoffs (22). In contrast, our study employed both continuous clinical scores and AI-derived video features to provide a more granular analysis of the hypomimia–sialorrhea relationship, thereby enhancing the robustness and resolution of our findings.

Interestingly, we observed that participants with the most severe hypomimia (MDS-UPDRS item 3.2 score = 4) reported lower ROMP-saliva scores (Figure 1). This finding suggests a potential nonlinear relationship between hypomimia severity and reported drooling symptoms. Several explanations may account for this finding. Patients with more advanced motor impairment may be less aware of or less likely to report drooling, potentially due to reduced insight or concurrent cognitive decline. Alternatively, patients with the most severe drooling may already be under treatment for drooling symptoms. Finally, the relatively small sample size (n = 52) of the study may have limited the reliability of this pattern. Whether this inverted U-shaped relationship remains consistent warrants validation in larger cohorts.

Application of PCA and CCA revealed that facial features from the upper, lower, and combined facial regions were moderately associated with drooling severity. Notably, the strongest correlations were observed in periocular features (r = 0.641), exceeding those from perioral features (r = 0.600). This was somewhat unexpected given the presumed pathophysiological link between reduced perioral motor control and impaired saliva clearance. Sialorrhea in PD is believed to arise from multiple contributing factors, including increased saliva production, impaired oral retention, and diminished clearance due to orofacial bradykinesia (4). Some studies have also implicated cognitive impairment in the exacerbation of drooling severity (23). Our findings suggest that while perioral dysfunction remains important, broader facial motor impairment—potentially reflecting more widespread basal ganglia circuit disruption—may play a more central role in the pathophysiology of drooling in PD.

Drooling remains a common but under-recognized symptom in PD, despite its well-documented impact on quality of life and the availability of effective treatments, including speech-language therapy, botulinum toxin injections, and pharmacologic interventions (6). Our study highlights the potential of quantitative facial analysis with the aid of AI to serve as a non-invasive, automated tool for predicting sialorrhea, thereby improving early detection and clinical monitoring. Recent advances in digital biomarkers for PD, such as wearable sensors and vision-based assessments, have shown promise for evaluating motor features (9, 24). Video-based hypomimia assessment tools, which are already under active development (14, 17), may also be extended to screen for non-motor symptoms like drooling—broadening the scope of digital phenotyping in PD.

This study has several limitations. First, we did not collect detailed demographic or clinical information such as disease duration, medication status, or total MDS-UPDRS Part III scores. This was an intentional decision aimed at simplifying data collection and promoting remote, patient-friendly assessment, aligning with the broader goals of digital healthcare. Second, the study was limited to evaluating associations and did not develop predictive models for clinical deployment. Future work should focus on building AI-based systems that can estimate drooling severity and support therapeutic decision-making, such as determining the need for botulinum toxin treatment or referral to speech therapy. Third, our analysis was based on static apex–neutral frame comparisons and did not capture dynamic features such as tremor or freezing episodes. Incorporating time-series or frequency-based features would require a larger dataset to ensure robust modeling. In addition, while we employed linear methods (PCA and CCA) to ensure interpretability and robustness, we acknowledge that some facial abnormalities in PD may follow nonlinear patterns. Future work employing nonlinear approaches, such as neural networks, may better capture these complex associations, particularly with larger datasets. Fourth, we focused exclusively on happiness as the main target emotion, which may not fully represent the range of facial motor dysfunction in PD. Different emotions engage distinct muscle groups, and incorporating multiple expressions could provide a more comprehensive assessment of facial motor function. Lastly, our analysis was cross-sectional and does not provide insight into the causal or mechanistic links between hypomimia and sialorrhea. Longitudinal studies incorporating multimodal data—including neuroimaging, autonomic testing, and cognitive assessments—will be necessary to elucidate the underlying pathophysiological mechanisms and validate the predictive potential of facial features in PD.

This study provides novel evidence supporting an association between hypomimia and sialorrhea in PD and demonstrates the feasibility of using AI-based video analysis to evaluate these features objectively. By identifying facial motion characteristics as potential digital biomarkers for drooling, our findings lay the groundwork for more accessible, scalable, and non-invasive assessment strategies for this often-overlooked non-motor symptom. Such approaches may ultimately enhance early detection and enable more timely, targeted interventions, thereby contributing to improved quality of life for individuals living with PD.