Prediction of future brain disorder risk using imaging data in 46,090 adults

The advent of biobank level datasets offers unprecedented opportunities to discover novel imaging markers and develop prediction models for brain diseases. In this study, we first performed a prospective cohort study and derived a comprehensive atlas of 692 brain image-derived phenotypes (IDPs) associated with 22 brain diseases in 46,090 individuals from the UK Biobank, with 2,494 cases and a follow-up time of 10.8 years. This study revealed 1,746 IDP-disease associations and over 558 IDPs were shared among more than one disease. We then developed 9 machine learning models that combine imaging markers and easily accessible clinical demographic variables to identify individuals at high risk of developing disease up to 8.4 years prior to the official diagnosis. The models incorporating imaging profiles and clinical predictors effectively predicted the onset of brain diseases, including glioma, Alzheimer's disease (AD), Parkinson's disease (PD), and multiple sclerosis (MS) (C-index > 0.6). Additionally, multi-omics genomic integration analysis (ranging from 9,725 to 807,553 individuals) revealed shared genetic structure between IDPs and brain diseases. The IDP-disease association and disease prediction models are deposited as an encyclopedia, a publicly available imaging-phenome resource (https://bsbd.hapyun.com/home). This work sheds light on the biological mechanisms of IDP-disease association and contributes to the identification of novel imaging markers and the development of prediction models for brain diseases. Our findings strongly emphasize that IDPs are the prospective biomarkers for predicting brain diseases, even at the time that is more than eight years prior to diagnosis. This study demonstrates the strong potential of these IDPs to enhance high-risk screening and early intervention of brain diseases.