Systematic quantification of health parameters from UK Biobank abdominal MRI using deep learning

Cardiometabolic diseases are an increasing population health burden and while several well established socioeconomic, environmental, behavioural, and genetic risk factors have been identified, our understanding of their drivers and mechanisms remains incomplete. Thus, a better understanding of these factors is required for the development of more effective interventions. Magnetic resonance imaging (MRI) has been used to assess organ health in a number of studies, but large-scale population-based studies are still in their infancy. Using deep learning to segment individual organs from up to 38,683 abdominal MRI scans in the UK Biobank, we demonstrate that image derived phenotypes such as volume, fat and iron content reflect overall organ health. We further show that these traits have a substantial heritable component which is enriched in organ-specific cell types. We also identify several novel genome-wide significant associations. Overall our work demonstrates the feasibility and power of high-throughput MRI for the multi-organ study of cardiometabolic disease, health, and ageing.