AI will be used on routine eye scans to detect early signs of illnesses including diabetes and heart disease, under a scheme at the Northern Ophthalmic Research and Innovation hub — NORI — at the University of Sunderland.
The hub has received £12.2m from the government through UK Research and Innovation, with the aim of diagnosing patients faster. It is one of five new UKRI-backed healthcare hubs using technology to find illness earlier.
Why the eye, for diseases that are not about sight
The reasoning is anatomical rather than clever. The retina is the only place in the body where a clinician can look directly at blood vessels and nerve tissue without cutting anything open. A camera pointed through the pupil gives an unobstructed view of microvasculature.
Diabetes and cardiovascular disease both damage small blood vessels, and they do it early — often before a person notices any symptom at all. The changes show up in the retina as subtle alterations in vessel width, branching pattern and leakage.
Those changes are hard for a human to grade consistently, and they are exactly the kind of pattern a trained model reads well. The scan is already being taken. The proposal is to get a second diagnosis out of an image that was captured for a different reason.
What the money is for
Matthew Campbell, interim pro vice-chancellor for health and co-director of NORI, said the funding would help the team develop the technology and take it into international markets.
That second clause is worth noticing. This is framed from the outset as an export, not only as an NHS tool — which shapes what gets built. Software intended to travel has to work on scans from equipment it was not trained on, in populations it did not learn from.
The initial focus is diabetes and cardiovascular disease. The approach could later extend to dementia and strokes.
Why this one could actually matter here
Retinal screening for diabetic eye disease is already established practice in many health systems, including in Bangladesh, where diabetes is common and specialist time is scarce. The infrastructure — a camera, a trained operator, a referral pathway — partly exists.
What this research proposes is not new hardware but more information from the same photograph. For a system where the hard constraint is the number of specialists available to interpret images, a tool that triages which scans need expert attention is worth more than one that produces a diagnosis nobody has time to act on.
The caution that belongs here
This is funding for research, not a product in clinics. Five hubs have been announced, the technology is to be developed, and no accuracy figures have been published.
Screening tools fail in a particular way worth remembering: a test that flags too many healthy people generates anxiety, appointments and cost, while a test that misses cases is worse than none at all because it reassures. Whether this one lands in the useful middle is precisely what £12.2m is meant to find out.
