Towards Better Mental Health: Rapid Screening for Mood Disorder using Electroencephalogram and Facial Expression Paired Data with Weakly Supervised Contrastive Learning

Depression has become a growing crisis among teenagers in recent years, with rates of suicide and depression rising annually. Despite the increasing need for more accessible resources to combat this mental health emergency, most available mental health resources remain either too expensive or inaccurate to effectively help those affected by the surge in mental health crises. The method proposed in this paper aims to combat the mental health epidemic by using Electroencephalogram and Facial Expression data to train a multi-modal system that can analyze emotion to detect early signs of depression and suicidal ideations. The proposed system is composed of two modules: an emotion classification module and a mood disorder screening module. The proposed emotion classification module takes both Electroencephalogram data and facial images as inputs and classifies the individual's emotional categories. After predicting the emotional status, the mood disorder screening module detects abnormalities in the individual's mental health condition by comparing the results with those of other individuals. Through extensive experiments, I have demonstrated that the proposed system achieves an accuracy of 95.7% on the Electroencephalogram and Fical Expression dataset which proves its feasibility for real-world application.