Contactless depression screening via facial video-derived heart rate variability

Key findings from this study

• The study found that facial video-derived HRV combined with demographic factors achieved moderate discrimination (AUROC 0.64) for depressive symptoms in 1453 participants.
• The researchers report that smoking status, sex, and medical comorbidities were stronger predictive contributors than HRV features alone.
• The authors demonstrate that contactless video-based physiological measurement can facilitate large-scale depression screening despite modest individual discriminative capacity.

Overview

Contactless facial video analysis derived heart rate variability (HRV) features combined with demographic information to screen for depressive symptoms in a large cohort. A stacking ensemble classifier incorporating four base learners and an SVM meta-learner achieved moderate discrimination performance (AUROC 0.64) on 1453 individuals assessed via the Patient Health Questionnaire-9.

Methods and approach

Researchers developed a stacking ensemble model using HRV measurements extracted from facial video recordings alongside demographic variables. Four base learners (logistic regression, gradient boosting, XGBoost, and SVM) fed into an SVM meta-learner. Model evaluation employed 5-fold cross-validation. Feature importance analysis identified predictors contributing most substantially to classification decisions.

Results

The stacking ensemble achieved an AUROC of 0.64 with AUPRC of 0.45 and Matthews correlation coefficient of 0.21. Including demographic features with HRV metrics yielded superior performance relative to HRV features alone. Feature importance analysis identified smoking status, sex, and medical comorbidities as the strongest predictive contributors, substantially outweighting HRV-derived parameters in the model's decision boundary.

The contactless video-based approach successfully distinguished individuals displaying depressive symptoms, though with modest predictive strength. This moderate discriminative capacity indicates potential utility despite performance limitations. The relative importance of demographic and behavioral factors over physiological metrics suggests complex interactions between multiple data modalities in depression-related classification.

Implications

Facial video-derived HRV presents a scalable, non-invasive pathway for accessible depression screening across distributed populations. Elimination of sensor contact requirements reduces barriers to implementation in remote or resource-limited settings. The integration of demographic factors alongside physiological markers indicates that depression screening via automated video analysis may require multimodal risk stratification rather than reliance on single physiological signatures.

Modest performance metrics constrain direct clinical deployment as a standalone diagnostic tool. However, the contactless methodology supports application within preliminary screening workflows that subsequently route individuals toward clinical assessment. Future developments should investigate whether enhanced HRV feature extraction, larger training cohorts, or integration of additional video-derived biomarkers improve discrimination capacity. Temporal dynamics of HRV measurement or incorporation of facial expression analysis may strengthen predictive signals.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.