Most current medical AI models focus on a single organ, disease, or data type, despite the reality that long-term health and multimorbidity reflect interacting processes across the body. This project will use de-identified UK Biobank data to develop and evaluate multi-organ, multi-disease, multi-modal AI analytics for longitudinal disease risk prediction, risk-factor profiling, and survival stratification. We will integrate imaging-derived phenotypes from multiple organ systems with linked health outcomes and key non-imaging variables (e.g., demographics, clinical measures, lifestyle factors) to model who develops disease and when.
Research questions: (1) How much does combining multiple organs and modalities improve prediction compared with single-organ/single-modal models? (2) Which organ measurements and modifiable risk factors most strongly contribute to future disease onset and adverse outcomes? (3) Can we identify population subgroups with distinct risk and survival trajectories across major disease groups (e.g., cardiometabolic and neurological conditions)?
Objectives: curate harmonised cohorts and endpoints; build a reproducible modelling pipeline; benchmark models for incident-disease prediction and time-to-event outcomes; provide transparent model interpretation to rank key risk factors; and disseminate findings as aggregate statistics and methods to support future prevention and stratified care research. All analyses will be conducted in secure environments in line with UK Biobank governance, with no attempt to identify or contact participants.