iPonder: A Novel Approach to Multimodal Mental Health Diagnosis and Therapy for Teens via Deep Transfer Learning and Computer Vision

Suicide is the second leading cause of death for teenagers, tarnishing the future potential of the world. More teenagers die from suicide than from cancer, heart disease, stroke, and many more diseases, combined. Although the greatest mental health improvement occurs from a conversation, over 50% of suicidal teenagers stay quiet about their depression due to shame. Thus, iPonder, a cross-platform accessible mobile application for teenagers to privately discuss, confront, and overcome their depression, was created. Using a four-level diagnosis approach based on user data, responses to psychological questionnaires, choice specification, and free-response answers, iPonder constructs a hyper-personalized user profile, which can then be used to pair teenagers anonymously online with similar profiles to discuss and vent about their issues. Questionnaires include the Patient Health Questionnaire-9, General Anxiety Disorder- 7, UCLA Loneliness Scale, and Myers-Briggs Personality Type Indicator test. Using Reddit's Self-Reported Mental Health Diagnosis text-corpus with over 200,000 data points, 11 artificial neural networks were constructed to predict user risk of various mental illnesses based on their free-text-responses, using transfer learning from a token-based text embedding neural network with over 48 million parameters trained on the English Google News 7B corpus with over 4 billion data points. Additionally, two other artificial neural networks were created, trained on the Distress Analysis Interview Corpus with tremendous labeled video data regarding facial landmark tracking and audio analysis for depression-diagnosed people, to predict the risk of depression from a user-submitted video. All neural networks were evaluated on their accuracy, precision, recall, and f1-score. Besting state-of-the-art metrics on similar tasks in the relevant literature, iPonder achieves over 95% accuracy, recall, precision, and f1-score per neural network, pointing to a new age of accessible, effective, and anonymous mental health therapy in teens.