Researchers fine-tuned the RETFound vision foundation model on color retinal fundus images from 71,343 UK Biobank participants to predict chronological age, according to a study published in Nature Communications on Aug. 26, 2026. The deep learning model achieved a mean absolute error of 2.85 years in predicting age from the scans.
The difference between predicted age and actual chronological age, designated as the retinal age gap, linked to several systemic health indicators. The study found associations between the gap and cardiometabolic traits, inflammation, cognitive performance, dementia, cancer, incident cardiovascular disease, and all-cause mortality. Genome-wide analyses also connected the retinal age gap to genes governing longevity, metabolism, neurodegeneration, and age-related eye conditions.
Sex-stratified analyses revealed distinct biological patterns between male and female participants despite consistent prediction accuracy across groups. Male participants displayed stronger associations between the retinal age gap and metabolic syndrome. In female participants, model attention maps and genetic data highlighted a larger role for retinal vasculature.
Retinal aging patterns in female participants also shifted across the menopausal transition. Postmenopausal females exhibited higher retinal age gap values and developed clinical association patterns that resembled those seen in males.
The study, led by researchers from the University of Lausanne and the Swiss Institute of Bioinformatics alongside collaborating clinical institutions, utilized UK Biobank data under Application Number 90947. Funding came from the Swiss National Science Foundation under the "VascX" Sinergia project alongside support for the Rotterdam Study cohort.
