1. Project Overview and Scientific Rationale
Major psychiatric disorders – MPD, including major depressive disorder (MDD), bipolar disorder (BD), and schizophrenia (SCZ)-are heterogeneous neuropsychiatric mental illnesses that are leading causes of morbidity around the world. The current MPD diagnoses rely primarily on behavioral evaluations and clinical symptoms, which lack biological objectivity and medical accuracy, often resulting in misdiagnosis and delayed treatment.
By leveraging the power of artificial intelligence (AI) tools like foundation models and integrating multimodal data of neuroimages, clinical records, and genotypic variations, we aim to develop a multimodal approach to integrate clinical, genetic, and neuroimaging data for objective and accurate diagnosis and subtyping of MPDs to help inform personalized therapy of the disorders and uncover complex pathogenesis mechanisms behind diseases.
2. Research Questions
Q1. What is the neuropathological and genetic mechanism behind the morbidity of MPDs, and what are critical biomarkers?
Q2. How can multimodal data be effectively integrated using foundation models to enable objective and reproducible diagnosis, subtyping, and therapy of psychiatric disorders?
3. Objectives
By integrating large-scale multimodal data and leveraging the great power of AI and foundation model tools, we aim to achieve the following research objectives.
1) Establish a foundation model that facilitates precision diagnosis of multiple MPDs by integrating multimodal patient data, thereby bridging gaps in establishing an objective and unified diagnosis pipeline.
2) Identify typical subtypes and biomarkers through disease-specific multimodal feature clustering, differential expression analysis, and multi-level neuroimage analysis.
3) Explain disease mechanisms through utilizing model interpretability study approaches and the LLM’s reasoning capability.
