Integrating AI-driven technologies and facial-semantic features for depression detection: A cross-sectional study

Area of Science:

• Artificial Intelligence
• Psychiatry
• Digital Health

Background:

• Depression is a global health issue with challenges in early detection in outpatient settings.
• Limitations include self-report bias, stigma, and underreporting of depressive symptoms.
• AI offers objective, scalable, and unobtrusive methods for early depression detection.

Purpose of the Study:

• To evaluate the predictive accuracy of two AI systems, iSeeME and EDDTW-V2, for identifying depressive symptoms.
• To assess the utility of AI in detecting depression among high-risk outpatients.
• To compare the performance of facial-expression analysis (iSeeME) and narrative analysis (EDDTW-V2).

Main Methods:

• A cross-sectional study involving 62 outpatients from psychiatric and surgical-oncology clinics.
• Standardized depression assessments (HDRS, BDI-II, PHQ-9) were administered.
• Facial expression data analyzed by iSeeME; narrative transcripts by EDDTW-V2; predictive validity assessed via metrics, correlations, and cluster analysis.

Main Results:

• 39 out of 62 participants were clinically depressed (HDRS ≥7).
• iSeeME achieved 0.761 precision, 0.854 recall, 0.805 F1-score, and 0.770 accuracy.
• iSeeME showed stronger correlations with BDI-II (r=0.442) and PHQ-9 (r=0.335) than EDDTW-V2.

Conclusions:

• Both AI systems show potential as supplementary tools for depression assessment.
• iSeeME excels in detecting affective-behavioral symptoms; EDDTW-V2 captures cognitive-linguistic features.
• AI supports earlier detection, monitoring, and targeted interventions for depression in outpatient settings.