A team of researchers from the University of Copenhagen, the Technical University of Denmark, and Imperial College London has shown that contrastive deep learning can determine an individual’s age from skin biopsy images alone, and that the visual features learned by the model constitute a novel biomarker for aging that predicts mortality and chronic disease risk.
The study, published on arXiv and currently under review at npj Digital Medicine, used histopathological slides of skin biopsies, tissue samples routinely collected in clinical practice. By applying contrastive deep learning, a technique that learns to distinguish between similar and dissimilar examples without explicit labels, the model identified visual features in the tissue that correlate with chronological age.
More significantly, the researchers linked these visual features to comprehensive Danish health registries covering mortality outcomes and chronic disease prevalence. The resulting biomarker predicted not just how old a person was, but how quickly they were aging, distinguishing individuals with faster biological aging from those aging more slowly.
As global life expectancy increases, the burden of chronic diseases associated with aging continues to grow, yet individuals exhibit considerable variability in how quickly they age. Identifying biomarkers that capture this variability is crucial for understanding aging biology, enabling earlier disease detection, and improving prevention strategies.
“What this means is that routinely collected health data can provide additional value when used together with deep learning,” the authors note, “by creating a new biomarker for aging which can be actively used to determine mortality over time.”
The approach is notable for using data already collected in standard clinical practice, skin biopsies are commonly taken to investigate suspicious lesions, rashes, and other dermatological concerns. If validated in broader populations, the technique could provide a low-cost method for assessing biological age without requiring dedicated tests or specialized equipment.
The study’s use of Danish health registries, which track nearly the entire population’s health outcomes over decades, gave the research team unusually rich longitudinal data to validate their biomarker against real-world mortality statistics.
Sources: Contrastive Deep Learning Reveals Age Biomarkers in Histopathological Skin Biopsies (arXiv, revised July 1, 2026; under review at npj Digital Medicine)