Anatomy of aging through organ-resolved multi-modal imaging and deep learning

Abstract While aging manifests differently across organs and individuals, existing approaches to measure it lack the spatial resolution to capture this complexity. Here, we develop an approach that applies multi-modal imaging, segmentation algorithms, and deep-learning to assess organ-specific aging across 39 anatomical regions in a total of 134K individuals in the UK Biobank. Our analysis reveals significant organ aging heterogeneity across and within individuals and a remarkable prevalence of organ-specific extreme aging. We validate that our imaging measures capture pathophysiologically meaningful aging through correlation with organ-specific biomarkers, revealing biologically coherent patterns. We find that accelerated organ aging is robustly predictive of corresponding organ disease. We identify the cerebrum as one of the strongest predictors of organismal aging. We investigate organ aging patterns underlying disease risk and find that each disease is linked to aging of highly distinct subsets of organs. Exploring lifestyle factors and interventions reveals a range of divergent organ-specific effects. Our work establishes a powerful paradigm for noninvasively evaluating human aging at anatomical resolution and population scale.