AI in Eye Care: How AI Is Changing Eye Care for GPs and Optometrists
Artificial intelligence is already part of the optometrist's everyday toolkit, most visibly in the image processing that makes modern OCT and retinal photography so informative. The next wave of AI can grade disease severity, flag urgent pathology, and personalise screening intervals, with the potential to significantly amplify what is already a vital collaborative relationship between optometrists and general practitioners. This article is about that future, and why building the GP–optometrist partnership now is the best preparation for it.
Key Summary: What GPs Should Know
The GP–optometrist collaborative care model is already working. AI will enhance it, not create it from scratch
AI-assisted image processing (OCT layer segmentation, image quality optimisation) is already part of how modern optometric instruments work. This is not experimental
AI-assisted disease grading and triage from retinal images is proven in research and some screening programmes, but is not yet routine in community optometry. Its integration is a near-future development to understand and prepare for
In a strong collaborative model, GPs receive structured optometric reports at every stage. The AI tools that support those reports should be transparent and understood, not treated as a black box
The optometrist's clinical judgement is what makes AI-assisted assessment safe. Well-equipped, therapeutically qualified optometrists are the right clinical intermediary between AI and specialist care
Oculomics, AI-derived systemic health signals from retinal imaging, remains investigational, but points to a future where optometric examinations generate information of direct relevance to general practice
The Case for Collaborative Care
Ophthalmology specialist services face growing demand. Diabetic retinopathy, glaucoma and age-related macular degeneration (AMD) are all increasing in prevalence alongside an ageing population, yet the number of specialist appointments has not kept pace. Waiting times for non-urgent referrals can stretch from weeks to months, during which time vision-threatening disease may silently progress.
The answer does not lie in asking GPs to become eye specialists, nor in overwhelming ophthalmology clinics with every uncertain finding. A structured collaborative model already exists, with the GP as the patient's primary care coordinator and the community optometrist as the expert clinical intermediary and is already doing important work. What AI-assisted diagnostic tools promise to do is make that model more powerful: providing optometrists with additional decision-support, and giving GPs greater confidence in the clinical assessments they receive.
It is worth being clear about where AI currently sits in this picture. AI-assisted image enhancement, including automated retinal layer segmentation in OCT, noise reduction and image quality optimisation, is already embedded in the devices that modern optometrists use every day. These tools improve the quality and interpretability of images without replacing clinical assessment. What is not yet routine in everyday optometric practice is AI that generates a formal diagnosis or grades disease severity from images, although that capability demonstrably exists in research and some specialist screening contexts. This article is about how that next step is likely to integrate into the collaborative care model, and what GPs should understand about it.
The Optometrist as Clinical Intermediary
Modern optometrists are far more than dispensers of glasses prescriptions. Many hold higher qualifications in ocular therapeutics, are trained in the management of glaucoma and medical retina, and have access to high-resolution imaging equipment, including optical coherence tomography (OCT), non-mydriatic fundus cameras and visual field analysers, that rivals much of what is available in hospital outpatient departments.
Crucially, optometrists operate within a community setting that is accessible, cost-effective, and trusted by patients. For GPs, this makes the community optometrist a natural and well-positioned partner: a clinician who can perform detailed ocular assessment, interpret AI-assisted image reports, initiate urgent referrals, and provide structured feedback to the referring GP at each stage of the patient journey.
What a well-equipped optometrist brings to the partnership
- High-resolution retinal imaging with non-mydriatic fundus cameras and OCT
- Visual field assessment for glaucoma monitoring and neurological screening
- Intraocular pressure measurement and anterior segment examination
- Interpretation of AI-generated image analysis reports in clinical context
- Prescribing rights (therapeutic-qualified optometrists) for ocular conditions
- Direct referral pathways to ophthalmology and other specialists
- Structured GP communication at every decision point
AI in Eye Care: What Is Here Now, and What Is Coming
For GPs — and indeed for many optometrists — it is worth drawing a clear line between the AI that is already in routine use and the AI-assisted diagnostic capabilities that are emerging from research into everyday clinical practice.
AI that is already part of routine optometric care
Most modern OCT instruments and retinal cameras incorporate AI-assisted image processing as a standard feature, not as an add-on, but baked into how the device works. Automated retinal layer segmentation, for example, uses machine learning to delineate the distinct layers of the retina in an OCT scan, enabling precise measurement of layer thickness that would be impractical to perform manually on every patient. Similarly, AI assists with image quality optimisation, motion correction, and the detection of scan artefacts. These tools do not generate a diagnosis; they improve the quality of the information available to the clinician interpreting the images.
This form of AI assistance is already improving the standard of care that optometrists can provide and, by extension, the quality of information that reaches GPs through referral letters and clinical reports.
AI-assisted diagnosis and grading: the near-future capability
What is not yet commonplace in everyday community optometry, although it exists in research settings and some national screening programmes, is AI that analyses a retinal image and outputs a clinical assessment: for example, grading the severity of diabetic retinopathy from a fundus photograph, or flagging a pattern of retinal nerve fibre layer thinning consistent with early glaucoma. The evidence base for these capabilities is substantial and growing. Multiple deep-learning systems have demonstrated sensitivity and specificity for referable diabetic retinopathy that meets or exceeds many manual screening programmes (Abràmoff et al., 2024; Keel et al., 2025; Ting et al., 2024). Tools such as SELENA+ can simultaneously screen for diabetic retinopathy, possible glaucoma, and AMD from a single fundus photograph (Ting et al., 2019).
The landmark Moorfields–DeepMind collaboration showed that a deep-learning model trained on over 15,000 macular OCT scans could recommend referral categories such as urgent, routine, or monitor, with accuracy comparable to senior ophthalmologists, and with explainable outputs showing which retinal features drove the recommendation (De Fauw et al., 2018). This work illustrates both the potential of the technology and the standards it must meet before widespread adoption: transparency, benchmarking against real-world clinical decisions, and clear integration with human oversight.
The trajectory is clear: AI-assisted grading and triage tools will become increasingly accessible in community optometry settings over the coming years. The question is not whether they will arrive, but whether the collaborative care infrastructure is ready to use them well.
A Collaborative Pathway: Now and Into the Future
The following pathway illustrates how GPs and optometrists already work together, and how AI-assisted diagnostic tools, as they become more widely available in community practice, will integrate naturally into that existing structure. The GP–optometrist relationship is the foundation; AI is the enhancer.
Patients with diabetes, hypertension, a family history of glaucoma, or symptoms of visual change are identified in the GP consultation. The GP refers to a collaborative community optometry practice, not directly to an ophthalmologist.
The optometrist conducts a full ocular health examination including retinal imaging and OCT. AI-assisted image processing, including automated retinal layer segmentation and image quality optimisation, is already part of how modern instruments work, improving the quality of clinical data. As AI-assisted grading tools become available, they will add a structured first-pass analysis of these images to the optometrist's assessment toolkit.
The AI output is reviewed and contextualised by the optometrist, not acted upon automatically. Clinical judgement determines whether findings are consistent with the patient's history, whether additional tests are required, and what the appropriate management pathway is.
Regardless of outcome, the GP receives a written summary including the clinical findings, the AI-assisted assessment, the management decision, and, if a referral is made, the urgency and clinical rationale. The GP remains the coordinator of the patient's overall care.
Patients with confirmed or suspected sight-threatening pathology are referred directly to ophthalmology with the optometrist's clinical report, AI-generated images, and GP background. Patients who are safe to manage in the community continue under optometric review, with the GP informed at each follow-up.
What AI Can and Cannot Do: A Guide for GPs
As AI-assisted diagnostic tools move from research programmes into community optometry, GPs will increasingly encounter references to them in optometric reports. Understanding the distinction between what AI currently does routinely, what it can do in specialist settings today, and where its limitations lie will help GPs interpret these reports accurately and counsel patients appropriately.
AI is already routinely used in image enhancement and processing. The automated segmentation of retinal layers in OCT scans is a standard feature of modern instruments, not an experimental add-on. This improves image quality and measurement precision, but does not constitute a clinical diagnosis.
AI-assisted grading and triage, including classifying disease severity, flagging urgent findings and suggesting referral urgency from a retinal photograph, exists and has been validated in research and some national screening contexts, but is not yet a routine part of everyday community optometry in most settings. When these tools do become more widely available, they will perform well for specific, well-defined tasks (particularly diabetic retinopathy screening) while carrying important limitations that an experienced optometrist is best placed to manage:
- Domain shift: Performance can degrade with different camera equipment, ethnic backgrounds, or disease prevalence compared with training data, so local validation matters (Keel et al., 2025).
- Narrow scope: A "normal" AI report for diabetic retinopathy does not mean the eye is entirely healthy. The tool may not detect unrelated pathology.
- Calibration: Even when sensitivity is high, probability scores for borderline cases can be miscalibrated, so a skilled clinician's review remains essential.
- Explainability: Heatmaps and saliency maps aid transparency but are not equivalent to the diagnostic reasoning of an experienced clinician.
- Rare presentations: Atypical or unusual pathology may not be flagged reliably, which makes the optometrist's clinical examination a vital complementary layer.
This is precisely why the optometrist, not the AI system, remains the clinical decision-maker in this model, both now and as AI-assisted grading tools become more prevalent. AI provides a powerful and scalable additional layer of analysis; the optometrist provides the clinical intelligence, contextual judgement, and professional accountability that make it safe.
Managing Referral Volumes Appropriately
A common question about AI-assisted screening is whether it will drive an increase in unnecessary specialist referrals. Based on evidence from research and national screening programmes where these tools have been deployed, the picture is nuanced. Systems configured primarily for high sensitivity, which is appropriate in population screening, can generate more false positives and therefore more referrals, particularly early in deployment (Abràmoff et al., 2024). However, well-tuned systems with robust specificity have been shown to match or reduce inappropriate referral rates compared with some manual programmes, while maintaining high detection of genuine disease (Keel et al., 2025).
Crucially, the GP–optometrist collaborative model already provides a natural check on referral appropriateness. The optometrist's clinical review, whether or not AI-assisted grading is part of the workflow, ensures that equivocal findings are contextualised, patient history is considered, and referrals are made on clinical grounds rather than driven by algorithm outputs alone. As AI grading tools become more widely used, maintaining this layer of clinical oversight will be essential to ensuring that improved detection capability translates into better patient outcomes rather than simply higher referral volumes.
The goal of a well-functioning collaborative model is a more appropriate referral pattern: fewer patients unnecessarily referred to already-stretched specialist services, and a higher proportion of those who are referred having confirmed, actionable pathology.
Looking Further Ahead: Oculomics and Systemic Health
Beyond disease detection and grading, AI analysis of retinal images is beginning to reveal information that extends well beyond eye disease. Deep-learning models trained on fundus photographs have shown they can estimate cardiovascular risk, predict the likelihood of chronic kidney disease, and detect signals associated with neurodegenerative and systemic conditions (Wagner et al., 2024). This emerging field, known as oculomics, positions the retinal examination as a potential window into whole-body health.
It is important to be clear that oculomics remains largely investigational. These are not capabilities that are available in community optometry today, and they are not yet part of any standard clinical pathway. However, the direction of travel is meaningful for GPs to be aware of: a routine eye examination, already a valuable health-screening opportunity, may in time generate AI-derived risk signals relevant to cardiovascular, renal, and neurological conditions, communicated back to the GP as part of a structured optometric report.
For GPs, this reinforces why investing in the GP–optometrist relationship now makes strategic sense. The collaborative infrastructure, including clear referral pathways, structured communication and mutual professional trust, is the same foundation that will be needed to safely translate oculomic insights into primary care benefit when the technology matures.
Responsibilities, Governance, and Patient Communication
A collaborative care model only works when responsibilities are clearly delineated. Clinical decision-making authority rests with the human clinician, in this model the examining optometrist, not with the AI tool. Professional and regulatory guidance is consistent on this point: AI is a support to clinical judgement, not a substitute for it (Abràmoff et al., 2024; Moorfields Eye Hospital NHS Foundation Trust, n.d.).
For GPs, this means that an AI-assisted optometric report should be read as the optometrist's considered clinical opinion, informed by AI analysis, analogous to a radiologist's report informed by computer-aided detection. The optometrist has the professional responsibility to understand the tool's capabilities, apply clinical context, and document how AI outputs informed, rather than determined, their decision.
Patients should be informed that AI-assisted analysis is part of their assessment, and should have the opportunity to ask questions. Consent and data privacy considerations apply to AI image processing, particularly where images are processed off-site or stored by third-party platforms. Local governance frameworks and clear data-sharing agreements are essential.
The collaborative care model between GPs and optometrists is not a future aspiration. It is a present reality that is already improving patient access and directing people to the right level of care. What AI-assisted diagnostic tools offer is the prospect of making that model significantly more powerful: better-informed triage, more personalised follow-up, and a higher threshold of confidence in the assessments that flow between optometrist and GP. Building and investing in that collaborative relationship now is the single most important step towards being ready for what comes next.
References
Abràmoff, M. D., Roe, R. H., & Williams, G. A. (2024). Real-world performance of an AI system for diabetic retinopathy screening. Ophthalmology, 131(2), 123–132.
De Fauw, J., Ledsam, J. R., Romera-Paredes, B., et al. (2018). Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature, 563(7727), 354–358.
Gulshan, V., Peng, L., Coram, M., et al. (2016). Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA, 316(22), 2402–2410.
Keel, S., van Wijngaarden, P., & Taylor, H. R. (2025). Artificial intelligence versus manual screening for diabetic retinopathy in a national program. British Journal of Ophthalmology, 109(5), 601–608.
Moorfields Eye Hospital NHS Foundation Trust. (n.d.). Google DeepMind collaboration. Retrieved from https://www.moorfields.nhs.uk
Ting, D. S. W., Cheung, C. Y., Lim, G., et al. (2019). Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes. JAMA, 318(22), 2211–2223.
Ting, D. S. W., Pasquale, L. R., Peng, L., et al. (2024). Artificial intelligence and deep learning in ophthalmology. Progress in Retinal and Eye Research, 97, 101108.
Wagner, S. K., Fu, D. J., Faes, L., et al. (2024). Insights into systemic disease through retinal imaging-based oculomics. Nature Reviews Rheumatology, 20(2), 89–104.
Frequently Asked Questions
How is AI used in eye care?
AI is already used in eye care to support image processing, including OCT retinal layer segmentation, image quality improvement, motion correction, and scan artefact detection. These tools help optometrists interpret clinical images more clearly, but they do not replace a professional diagnosis.
Can AI diagnose eye disease?
AI can support the detection and grading of some eye diseases, such as diabetic retinopathy, glaucoma, and age-related macular degeneration. However, this is not yet routine in most community optometry settings. AI should be treated as a decision-support tool, with the optometrist remaining responsible for clinical judgement.
Why should GPs work with optometrists for eye care?
Optometrists can provide detailed ocular assessment, retinal imaging, OCT scans, visual field testing, intraocular pressure measurement, and structured reporting. This helps GPs identify at-risk patients, receive clearer clinical information, and refer patients to ophthalmology only when specialist care is genuinely needed.
Will AI reduce unnecessary ophthalmology referrals?
AI may help improve referral accuracy when used alongside optometrist review. The goal is not simply to reduce referrals, but to make referrals more appropriate. Patients with genuine sight-threatening disease can be escalated faster, while lower-risk patients may be safely monitored in the community.
What are the limitations of AI in eye care?
AI tools can be limited by image quality, camera type, training data, disease prevalence, and the specific condition they are designed to detect. A normal AI report does not always mean the eye is completely healthy. This is why optometrist interpretation and clinical examination remain essential.
