An Intelligent Emo-Care Facial Expression Monitoring System for Mental Health Assessment using Deep Learning

Facial expressions constitute a critical, noninvasive interface to monitor an individual's emotions and psychology and are thus useful in mental health evaluation. Systems for tele-mental health consultations still predominantly depend on manual subjective clinician interpretation without systematic emotion tracking in real-time. This paper presents Emo-Care, an intelligent system for real-time facial expression analysis of micro and macro expressions for mental health assessment, utilizing deep learning techniques. The proposed framework includes a multi-task lightweight CNN with a transformer-based temporal attention module for emotion modelling which is capable of recognizing AUs and deriving six affective dimensions associated with DSM-5 such as: valence, arousal, anhedonia, anxiety, irritability, and fatigue. To enhance clinical usability, the system features a secure, user-friendly dashboard displaying temporal emotion trajectories along with confidence scores. Preliminary tests on the publicly accessible DERM-MH dataset and the newly created MindFace-23 benchmark reveal an F1 score improvement of 7.1% over state-of-the-art performance on facial expression recognition without sacrificing inference latency, which was as low as 24 milliseconds on consumer-grade hardware. Emo-Care fills a critical void in digital mental health by providing objective, real-time insights into emotions during remote consultations, offering a scalable and privacy-preserving solution for remote psychiatric evaluations.