A deep learning model trained on skin biopsy slides can estimate how fast a person is aging — and predict their mortality risk.
Researchers used contrastive deep learning to show that histopathological images of skin biopsies contain enough visual information to determine a person's chronological age. They then built a novel aging biomarker from those visual features and validated it against comprehensive health registers in Denmark. The model's predictions correlated with mortality rates and the prevalence of chronic age-related diseases across the population studied. The key point: none of this required new tests — only slides already collected for other clinical reasons.
The finding matters because biological age and chronological age diverge significantly between individuals, and clinicians currently lack cheap, scalable ways to measure that gap. If a biomarker this accessible holds up under broader validation, it could flag high-risk patients years before symptoms appear — using data hospitals are already sitting on.
The approach won't reach a clinic soon; external validation across more diverse populations is the obvious next hurdle. But it fits a pattern of researchers wringing unexpected diagnostic signal out of routine imaging, a trend that's been quietly outpacing purpose-built aging tests for the past several years.