"Eye age": when an AI reads ageing in the back of the eye

Why measure our "biological" age?

Two 60-year-olds do not always age at the same pace. Biological age tries to capture the real state of the body, as opposed to the age on an ID card. "Ageing clocks" already exist, based on blood tests for example, but they are not very accurate when you want to follow changes over short periods.

Yet to test treatments that might slow ageing one day, you would need a way to tell quickly whether they have any effect.

What did the researchers do?

The eye is a window onto the body: the back of the eye, photographed during an eye examination, lets you see small blood vessels directly. The team trained deep-learning models (AI that learns from examples) on retinal photos from the EyePACS dataset: 217,289 images from 100,692 patients for training, then 54,292 images from 25,238 patients for tuning.

The models were then tested on a second dataset, the UK Biobank (119,532 images), whose population is very different. The researchers also looked for the genes linked to the gap between eye age and real age, and tested one of them in fruit flies.

What did the study find?

eyeAge estimates real age more accurately than other ageing clocks, with an average error of 2.86 years on EyePACS and 3.30 years on the UK Biobank (after filtering for good-quality images).

The gap between eye age and real age is linked to death from any cause. The notable point is that this link remains (hazard ratio of 1.026) even when blood-based biological age is taken into account. The eye therefore seems to carry different information.

Several regions of the genome are tied to this gap. For the strongest one, the researchers lowered the activity of the matching gene in flies, which slowed the age-related decline in their vision.

What does this change for patients?

For now, nothing in everyday practice. The aim is research: a fast measure of ageing could help evaluate sooner whether treatments designed to protect against ageing are working.

In the longer term, a photo of the back of the eye, a simple and inexpensive exam, could help study ageing and the diseases linked to it.

What are the limits?

Like other ageing clocks, eyeAge does less well than an age calculated from blood markers that were chosen for their link with mortality when it comes to estimating the risk of death. So it does not replace those measures.

The study finds statistical links: it does not prove that an "older eye" causes a shorter life. And the fly experiment does not yet tell us what would happen in humans.