International Journal of Science and Technology (IJST)

MindGuard AI: A Human-Centred Multimodal Framework for Early Mental Health Risk Assessment

Identifying mental health crises at an early stage requires significant effort. An individual can appear healthy despite suffering internally for weeks, and by the time he or she consults a professional, a situation that could easily have been solved using basic help mechanisms transforms into something much more complicated to solve. The following paper introduces MindGuard AI-a multi-channel screening application designed to address that essential phase-the period spanning from initial signs of a psychological problem to the consultation with a specialist. Three distinct signals inform the model simultaneously: a 15-item self-report questionnaire following PHQ-9 and GAD-7 scoring, a BERT powered chat interface capable of assessing emotional tone throughout an informal discussion, and a MobileNetV2 convolutional neural network processing a short video stream captured by a webcam. A weighted fusion function combines all channels' results into one value, and the independent crisis flag continuously monitors for signs of suicide regardless of the resulting score. Our system proved efficient on the set of 240 sessions, showing 91% overall accuracy-an improvement of nearly ten percentage points compared to any single-channel alternative. In addition to describing our algorithm, this paper will also outline its limitations. Matching patterns in the emotional cues does not necessarily imply knowing the patient, and the development of MindGuard AI technology aims at early diagnosis but not as a substitute for the actual physician's judgment.