Skip to content
Open access

MMDF-LR: A Multimodal Contrastive Learning Model for Noninvasive Depression Risk Assessment in College Students

2026 · IEEE Access · Vol 14, pp. 157476-157494 · 0 citations · 45 references

Abstract

Depression is highly prevalent among college students. Current large-scale mental health screenings in universities still rely on self-report scales, which are inherently subjective and associated with high underdiagnosis rates. Existing multimodal depression assessment models generally suffer from suboptimal fusion performance, weak generalization on small samples, and insufficient interpretability. This study designed a $2\times 3$ (task type $\times $ emotional valence) experimental paradigm. By integrating facial video, speech, and demographic features, we proposed the multimodal deep fusion with contrastive learning and regularization (MMDF-LR) model. The model adopted adaptive weighted fusion and supervised contrastive loss combined with maximum mean discrepancy (MMD) regularization to optimize feature learning. We also conducted interpretability analysis using the SHAP method and established a three-level risk stratification system. We performed 5-fold cross-validation on data from 90 college students. The model achieves an AUC of 0.810 (95% CI: 0.702-0.914) on the test set, with an accuracy of 0.778 and an F1-score of 0.724, showing competitive performance against eight mainstream baseline models. Ablation experiments and SHAP interpretability analyses demonstrate that the video modality acts as the primary information source for depression recognition, and high-contribution fused features can be traced back to raw facial expressions. Meanwhile, the three-level risk stratification scheme offers preliminary objective support for differentiated psychological intervention in universities. This work represents a proof-of-concept study. The MMDF-LR framework is developed for preliminary screening. Further validation based on larger multi-center cohorts remains essential before real-world clinical deployment. With accessible interpretability and potential for campus depressive tendency screening, MMDF-LR could serve as an objective, lightweight candidate framework for university mental health assessment.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.