Swiss researchers develop AI test for biological age from retinal images

A retinal photograph could provide a new warning signal for Swiss patients. Researchers at the University of Lausanne have developed an artificial intelligence system that estimates biological age from images of the back of the eye. Their study links an older-looking retina with elevated risks of cardiovascular and respiratory disease, cancer and dementia.

The system does not diagnose a disease. It identifies whether the retina appears older or younger than expected for a person's chronological age, a gap that researchers associate with broader health risks. The work was published in Nature Communications and reported by the University of Lausanne on September 16, 2026.

The approach draws on a large dataset. Scientists trained the algorithm using retinal images from more than 70,000 people aged 40 to 79. On average, the model estimated retinal age within less than three years of a person's actual age.

That performance gives the technology potential as a screening tool, particularly because retinal imaging is already used in medical settings. For Swiss healthcare providers, the attraction lies in a relatively accessible image that may reveal information about blood vessels and systemic health before symptoms become obvious. Clinical use remains some distance away, and the researchers stress that the findings show associations rather than proof of causation.

The algorithm detects details that human observers can barely see. It analyses subtle changes in retinal colour and brightness, along with the density and condition of the blood vessels visible in an image. Those features can change as people age and may also reflect processes affecting organs elsewhere in the body.

The retina offers a direct view of small blood vessels without the need for invasive sampling. That makes it an attractive source of data for researchers studying vascular health, metabolic disorders and the effects of ageing. A photograph can be collected quickly, then assessed by software trained on tens of thousands of examples.

The Lausanne study links a higher retinal age to several major disease categories. The associations include cardiovascular and respiratory illnesses, cancer and dementia. They describe risk patterns across a population, however, rather than a prediction for a particular patient. An older-looking retina does not mean that someone has a specific disease or will necessarily develop one.

This distinction will matter if the technology enters Swiss clinical practice. Doctors would need to combine the AI result with medical history, blood tests, lifestyle information and established diagnostic examinations. The retinal estimate could support earlier assessment, but it cannot replace clinical judgment or confirm a diagnosis on its own.

The study found different patterns in men and women. In men, an older retinal appearance was more strongly associated with characteristics of metabolic syndrome, including high blood pressure, diabetes and being overweight. In women, retinal age showed a closer relationship with vascular factors, including an increased risk of thrombosis.

The researchers also observed a shift around menopause. Before menopause, women had retinas that appeared younger on average than those of men. After menopause, that difference was reversed. The finding raises questions about the role of hormonal changes in vascular ageing and the way biological age markers behave across the life course.

The authors do not claim that menopause causes retinal ageing or that retinal age directly causes disease. Their results point to associations that require further investigation. Age, sex, medication, family history and health behaviours could all influence the patterns detected by the algorithm.

The sex-specific findings are relevant to Switzerland's health system, where screening tools must work across diverse patient groups. A model trained on a large population still needs careful validation in other populations and clinical settings. Researchers will need to test whether the system performs consistently across different cameras, hospitals, ethnic backgrounds and stages of disease before doctors can rely on it in routine care.

Smoking emerged as the strongest factor associated with accelerated retinal ageing. The finding gives the research a practical dimension because several of the health factors linked to an older-looking retina can be monitored or treated. The study also points to the possible importance of early treatment for cardiovascular problems.

A retinal age estimate could eventually help doctors identify patients who need a closer assessment of blood pressure, blood sugar, weight, smoking habits or vascular health. In Switzerland, such information could fit into prevention programmes delivered through family practices, ophthalmology clinics or specialised hospitals. The technology could be particularly useful when a patient already receives a retinal examination for another reason.

The researchers have not established that changing a single behaviour will make a retina appear younger or eliminate disease risk. Their results show relationships in the data, not a guaranteed pathway from a lifestyle change to a health outcome. Patients would also need clear explanations of what an AI estimate means and what it cannot say.

Privacy will be another consideration. Retinal images are biometric health data, and any Swiss deployment would require secure storage, transparent consent procedures and clear rules about access. Patients should know whether an image is being used for diagnosis, research or both.

The Lausanne system is promising research, not yet a clinical test. Further studies are needed before doctors can use it to guide treatment or predict an individual's future illness. Researchers must establish how the model performs outside the original dataset and whether its estimates improve outcomes for patients.

Validation will require more than technical accuracy. A useful clinical tool must show that it identifies people who benefit from earlier intervention without creating unnecessary anxiety, tests or costs. Doctors will also need to understand how the algorithm reaches its estimate and how often it produces misleading results.

The University of Lausanne study arrives as Swiss medical experts continue to examine how artificial intelligence can support diagnosis and monitoring. Retinal imaging offers a practical route because it is non-invasive and already familiar to healthcare professionals. The next stage will involve connecting the AI estimate to established care pathways rather than treating it as a standalone verdict.

For Swiss patients, the immediate consequence is limited. No one should interpret an older-looking retina as proof of cancer, dementia or cardiovascular disease. The research does, however, strengthen the case for using eye images as a source of broader health information. With further evidence, a routine retinal scan could become one part of a more personalised picture of ageing and preventable risk.