AI eye scans move from vision care to early disease detection
Routine eye examinations may soon do far more than assess sight. At the Northern Ophthalmic Research and Innovation hub at Sunderland University, researchers will use AI to examine standard eye scans for early signs of diabetes and cardiovascular disease, backed by £12.2m from UK Research and Innovation.
The approach rests on a simple clinical premise with broad implications: the retina offers a direct view of small blood vessels and nerve tissue, structures that often mirror damage occurring elsewhere in the body. Because those changes can emerge before symptoms are obvious, an eye scan can become a window into systemic disease rather than a narrow ophthalmic test.
The funding gives the project scale, but also a clear mandate. The immediate goal is earlier identification of diabetes and heart disease, with future expansion into dementia and stroke detection if the underlying models prove robust enough across real-world care.
Retinal microvasculature as a marker of systemic disease
The back of the eye is uniquely suited to this kind of analysis. Tiny blood vessels and nerve tissue there can show subtle changes linked to metabolic and vascular injury long before patients feel unwell or receive a diagnosis through traditional channels.
AI is well suited to this task because retinal abnormalities can be too fine for the human eye to classify consistently at speed across large populations. In practice, the system will analyze routine eye scans alongside NHS health records and community data, looking for patterns that may signal disease earlier than current workflows allow.
A single eye image may not diagnose diabetes on its own. The strength of the method lies in pattern recognition across multiple data sources, where an algorithm can identify combinations of microvascular and neurologic change that suggest elevated risk and prompt follow-up testing.
Early detection matters because diabetes and cardiovascular disease often progress silently for years. By the time symptoms appear, vascular injury may already be established. Detecting risk through a familiar outpatient test could pull diagnosis forward into a window where treatment has a better chance of preventing downstream harm.
A new role for routine ophthalmic testing
The technology will be embedded in routine eye scans rather than built as a separate, stand-alone screening program. That matters because eye testing already sits within established care pathways, making it a practical entry point for opportunistic screening if the performance remains strong outside a research environment.
Matthew Campbell, interim pro vice-chancellor for health and co-director of NORI, said the investment would allow the team to develop the technology, test it in real-world healthcare settings, and take it into international markets. The project also aims to build a health innovation ecosystem in the North East, linking research capacity with translational delivery.
Science minister Chris McDonald, who visited NORI on Thursday, framed the effort around faster treatment access and reduced pressure on the NHS. Earlier diagnosis can do both: it can move patients into therapy sooner and reduce the burden created by late presentation, emergency care, and avoidable complications.
The work arrives at a moment when health systems are under pressure to do more with existing clinical contact points. If an eye scan can trigger earlier recognition of diabetes, the downstream effect could include earlier glucose-lowering treatment, cardiovascular risk reduction, and closer surveillance for complications such as retinopathy, nephropathy, and neuropathy.
From diabetes and heart disease to dementia and stroke
NORI’s initial focus is diabetes and cardiovascular disease, but the team says the same general strategy could later extend to dementia and stroke. The logic is biologically plausible. Both conditions can leave vascular and tissue-level signatures that may appear in imaging data long before major neurological events or cognitive decline become clinically obvious.
A broader extension will depend on whether the algorithm can separate disease signals from background variation caused by age, blood pressure, smoking, or other comorbidities. In a real clinical setting, those factors are not noise to be ignored; they are part of the risk profile the model must learn to interpret correctly.
International expansion is part of the stated plan as well. For that to succeed, the technology will need to travel across different health systems, population groups, and imaging protocols without losing accuracy. Generalizability will matter as much as raw model performance in a single center.
One sentence captures the clinical stakes: if the scan reliably flags disease before symptoms begin, the encounter changes from vision care to prevention.
The NHS pathway may be where impact is decided
The project’s value will be measured less by technical novelty than by whether it fits into everyday NHS practice. That means the system must identify risk quickly, avoid excessive false alarms, and lead to action that patients and clinicians can actually complete.
AI analysis of retinal images could help triage who needs confirmatory blood tests, cardiovascular review, or closer monitoring. In a busy service, that may improve efficiency by directing resources to people most likely to benefit from escalation, rather than treating every abnormality as equally urgent.
For patients, the attraction is straightforward: a routine test already being performed for eye health could also reveal a hidden systemic disorder. For clinicians, the challenge is maintaining trust in the output while preserving the judgment needed to interpret borderline or ambiguous findings.
The project still leaves important questions unanswered in the public record. The article does not provide diagnostic accuracy data, validation cohorts, or implementation timelines, and those details will determine whether the tool becomes an adjunct to care or remains a promising research platform. Before routine use becomes credible, the next steps will need to show how well the system performs across diverse patients and whether it can reliably alter clinical decisions in ordinary practice.