Anthony Loizides,
Foundation Doctor, Royal Sussex County Hospital
Introduction
Oculomics is the study of ocular biomarkers as windows into systemic health (1). By combining high resolution retinal imaging with artificial intelligence (AI) analytics, the eye is moving from an organ specific diagnostic subject to something closer to a non-invasive systemic health sensor (1,2). A 2025 review in Progress in Retinal and Eye Research consolidated over a decade of this evidence, marking oculomics as a maturing field rather than an emerging one (3). The question for clinicians is no longer really whether oculomics works. The evidence increasingly says it does. The more pressing question is whether ophthalmology, and medicine more broadly, is ready to use it responsibly.
Why the Eye
Traditional teaching treats ocular signs such as diabetic retinopathy, hypertensive arteriolar changes, and papilloedema as consequences of systemic disease that appear after damage is already underway (1). Oculomics inverts this logic, asking whether subtle microvascular and neurostructural changes in the eye can flag systemic disease before symptoms appear (1,2).
The retina and optic nerve develop as embryological outgrowths of the diencephalon, so the retinal nerve fibre layer offers direct, in vivo viewing of unmyelinated central nervous system axons (1). It is also the only human microcirculatory bed that can be imaged directly, non-invasively, and at cellular resolution. With Optical Coherence Tomography (OCT) and widefield fundus imaging now routine across NHS eye clinics and high street optometry, millions of high dimensional scans are generated every year as a byproduct of standard care (1,2).
What the Evidence Shows
Several UK-led datasets have driven this field forward. The AlzEye dataset, linking ophthalmic imaging and hospital admissions data for over 350,000 patients from Moorfields Eye Hospital with NHS Hospital Episode Statistics, has been particularly influential (4). Findings from this and related work include the following.
Inner retinal layer thinning on OCT correlates with cerebral atrophy, allowing AI models to flag subclinical structural markers of neurodegenerative disease before cognitive decline is clinically apparent (1,5).
Using AlzEye alongside UK Biobank data, thinner inner retinal layers were associated with both prevalent and incident Parkinson’s disease, detectable years before a clinical diagnosis would typically be made (6).
Deep learning models analysing routine colour fundus photographs can predict cardiovascular risk factors, including systolic blood pressure, biological age, and major adverse cardiac events, directly from the image (7).
Large, retina specific foundation models trained on over a million unlabelled images are now being adapted efficiently to new diagnostic and predictive tasks, suggesting this pattern of finding systemic signal in routine ophthalmic images is likely to accelerate rather than plateau (3,8).
None of this requires a new test. The signal is sitting inside scans that are already being taken.
Implications for Screening
The opportunity here is access rather than novelty. In the UK, far more people attend routine optometry sight tests than attend general health checks in primary care (1). If AI driven oculomic screening were integrated into that existing footfall, a patient booking in for new glasses could be risk stratified for cardiovascular disease or early neurodegeneration, with low friction referral into primary or specialist care where needed (2,7). This represents a genuinely different model of population screening, built on an appointment people are already attending, rather than one more thing competing for a GP slot.
What Needs to Happen First
Oculomics is not yet ready for routine deployment. Four issues need addressing.
First, the explainability problem. Many models flag risk from visual features that are not straightforwardly interpretable to clinicians. Before oculomic screening informs real referral decisions, clinicians need to understand, at least in broad terms, what a model is actually responding to (2,9).
Second, a plan for incidental findings. Flagging elevated dementia risk during a routine optometry visit is not a small thing to hand a patient with no counselling infrastructure in place. This needs a proper pathway rather than a message on a screen (1).
Third, validation across genuinely diverse populations before deployment, not after. Work outside ophthalmology has shown that imbalanced training data can produce clinically meaningful performance gaps between patient groups once a model is deployed (10). Models trained on unrepresentative cohorts risk encoding exactly the kind of bias that could widen existing health inequalities (7).
Fourth, clarity over who owns the follow-up. A referral triggered by an optometrist from an AI flag needs a clear receiving pathway in primary care, or the result is anxiety and demand with nowhere for it to land.
Conclusion
Oculomics is redefining what a routine eye examination can be for (1,2,3). As imaging quality improves and the underlying models mature, retinal scans are on track to become population level health signals, not just assessments of vision. Getting there responsibly means building explainability, an incidental findings pathway, and bias testing in from the outset, rather than retrofitting them once the technology is already embedded in high street optometry. Achieve that, and the eye becomes not just the organ of vision, but a genuine gateway into preventive medicine.
References
1. Wagner SK, Fu DJ, Faes L, et al. Insights into Systemic Disease through Retinal Imaging-Based Oculomics. Transl Vis Sci Technol. 2020;9(2):6.
2. Keane PA, Topol EJ. AI-facilitated health care requires education of clinicians. Lancet. 2021;397(10281):1254.
3. Zhu Z, Wang Y, Qi Z, et al. Oculomics: Current Concepts and Evidence. Prog Retin Eye Res. 2025; Article 101350.
4. Wagner SK, Hughes F, Cortina-Borja M, et al. AlzEye: longitudinal record-level linkage of ophthalmic imaging and hospital admissions of 353,157 patients in London, UK. BMJ Open. 2022;12(3):e058552.
5. Wisely CE, Wang D, Henao R, et al. Convolutional neural network to identify symptomatic Alzheimer’s disease using multimodal retinal imaging. Br J Ophthalmol. 2022;106(3):388-395.
6. Wagner SK, Romero-Bascones D, Cortina-Borja M, et al, for the UK Biobank Eye & Vision Consortium. Retinal optical coherence tomography features associated with incident and prevalent Parkinson disease. Neurology. 2023;101(16):e1581-e1593.
7. Poplin R, Varadarajan AV, Blumer K, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018;2(3):158-164.
8. Zhou Y, Chia MA, Wagner SK, et al. A foundation model for generalizable disease detection from retinal images. Nature. 2023;622(7981):156-163.
9. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44-56.
10. Larrazabal AJ, Nieto N, Peterson V, Milone DH, Ferrante E. Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proc Natl Acad Sci U S A. 2020;117(23):12592-12594.