AI reads skin biopsies to predict age and mortality risk
A contrastive deep learning model turns routine histopathology slides into a powerful aging biomarker.
A multinational research team led by Kaustubh Chakradeo (University of Copenhagen) and colleagues from Imperial College London, DTU, and Danish hospitals has demonstrated that contrastive deep learning can extract aging biomarkers directly from routine histopathological skin biopsy images. The model, trained on thousands of slides, learns visual features—cellular morphology, tissue architecture—that correlate with chronological age with high accuracy. Instead of relying on blood tests or genetic panels, the algorithm uses the same slides pathologists examine for skin conditions.
Crucially, the team linked these visual features to comprehensive Danish health registers, covering mortality records and chronic disease diagnoses over decades. They found that the AI-derived aging score predicts both all-cause mortality and the prevalence of age-related diseases (e.g., cardiovascular disease, diabetes) beyond chronological age alone. This suggests that microscopic changes in skin tissue capture an individual's biological age—a hidden signature of how fast they are aging. The work, submitted to npj Digital Medicine, highlights how deep learning can unlock latent value from existing medical archives, turning a 20-minute biopsy into a predictive longevity tool.
- Contrastive deep learning on skin biopsy histopathology slides predicts chronological age with high fidelity.
- Visual features extracted by the model correlate with mortality risk and chronic disease prevalence when linked to Danish health registers.
- The approach repurposes routinely collected biopsy data, offering a low-cost aging biomarker without additional testing.
Why It Matters
Enables early disease detection and personalized longevity insights from existing pathology slides, transforming routine clinical data.