Retinal Age (Deep-Learning Fundus Photography)

Retinal age is a biological age biomarker computed by a deep-learning model from a single fundus photograph of the retina. The retina is uniquely accessible: it lets researchers directly image microvasculature and neural tissue at high resolution using a non-invasive, low-cost camera that is already standard in optometry and ophthalmology clinics. Older-looking retinas, in models trained to predict chronological age from fundus images, predict all-cause mortality, cardiovascular disease, and diabetes complications independent of chronological age.

The most rigorous recent validation is Wang et al., published in npj Digital Medicine in juni 2025 [1]. They trained a longitudinal, label-distribution learning model on retinal photographs from multiple population cohorts and tested it in independent populations, showing consistent age-acceleration signals across ethnic groups. Compared to earlier retinal-age models like the UK Biobank retinal-age-gap by Zhu et al. 2022, the newer approach corrects for the label-noise problem where chronological age is not a perfect proxy for biological age.

Measurement is a color fundus photograph from a standard fundus camera; no dilation is required for many modern devices. The image is processed through the trained deep-learning model to produce a predicted biological age and an age-gap (biological minus chronological age). Some optometry chains and specialty biological-age testing services offer retinal age as an add-on for around 50-150 USD; a few consumer apps also compute it from smartphone-attachment fundus cameras.

Associations reported across cohorts include: each 1-year increase in retinal age gap associated with roughly 2-3 percent higher all-cause mortality risk, 3-5 percent higher cardiovascular disease risk, and higher incidence of diabetic retinopathy and dementia [1]. Effect sizes vary by cohort and by model version.

No controlled human trial has yet shown that a specific intervention lowers retinal age. Interventioner that improve retinal microvascular health, such as blood-pressure control, glycemic control, smoking cessation, and treatment of diabetic retinopathy, plausibly slow retinal aging based on the underlying vascular biology. This has not been formally tested with retinal age as an RCT endpoint.

Caveats. Retinal age is model-dependent: two different deep-learning models trained on different populations will give somewhat different scores for the same photograph, and models trained on white-European cohorts perform worse on African or East Asian populations unless retrained. The measurement is sensitive to image quality, camera type, and lighting. It is not a diagnostic tool; a high retinal-age gap is a risk signal, not a disease.

The plain takeaway: retinal age is a low-cost, non-invasive biological age biomarker with strong associations to mortality and cardiovascular disease in cohort studies. It is now available at some optometry chains and biological-age testing services. No intervention has been shown to lower it in a controlled trial.