Authors:
Aswana Lal, Chettiyar Vani Vivekanand

Addresses:
Department of Computer Science and Engineering, Rajalakshmi Engineering College, Chennai, Tamil Nadu, India.

Abstract:

According to some major mental diseases, MDD is estimated to impact around 332 million people globally, and it contributes to about 727,000 cases of suicide annually. HAMD-17 and PHQ-9 are examples of traditional methods of clinical assessment of patients, which involve subjective, episodic assessment and can lead to inter-rater differences, memory recall bias, and cultural challenges to disclosure. This study combines speech acoustics, natural language processing, facial dynamics, physiological biosignals, and behavioral biomarkers gathered via passive digital phenotyping to provide an organized analysis of the multimodal computational methods for automatic depression assessment. Taxonomies of five axes input modality, feature extraction technique, learning architecture, fusion strategy, and operational context are adopted in classifying forty papers published from 2020 to 2025. Comparing cross-modal attention fusion techniques early, late, and hybrid, quantifying cross-dataset generalization capabilities for DAIC-WOZ, D-Vlog, LMVD, and BlackDog challenges, concentrating on behavioral biomarkers like psychomotor function, sleep physiology, gait kinematics, typing rhythms, and passive smartphone measurements, and delineating the future directions for explainable artificial intelligence, demographically biased models, and federated architectures for privacy preservation. The main aspects of data extraction and classification are always expressed analytically. While generalization gaps of 15% to 22% across datasets remain a hurdle for clinical application, multimodal hybrid fusion models outperform unimodal models, achieving F1 scores up to 0.84.

Keywords: Multimodal Depression Detection; Speech-Based Assessment; Behavioral Biomarkers; Affective Computing; Digital Phenotyping; Federated Learning; Mental Health Monitoring.

Received on: 01/06/2025, Revised on: 06/08/2025, Accepted on: 19/10/2025, Published on: 07/05/2026

DOI: 10.69888/FTSHSL.2026.000639

FMDB Transactions on Sustainable Health Science Letters, 2026 Vol. 4 No. 2, Pages: 141-153

• Views : 243
• Downloads : 12

Download PDF