Artificial Intelligence in Eye Care

Devika Tandon

Introduction

Artificial intelligence (AI) is rapidly reshaping modern medicine, with ophthalmology standing at the forefront of this technological revolution. AI refers to the ability of machines to perform tasks traditionally requiring human intellect such as learning, problem solving and decision making. With the development of deep learning, advanced neural networks can learn complex patterns from large datasets without step-by-step programming. AI systems have reached impressive levels of accuracy in analysing and processing complex datasets and images. Ophthalmology is particularly well suited to benefit from AI, due to its reliance on high-resolution imaging for diagnosis and treatment planning. Fundus photographs, optical coherence tomography (OCT), and fluorescein angiography generate large amounts of visual data for pattern recognition by machine learning algorithms.

Regulatory bodies have begun approving AI systems for clinical use, recognising their potential to transform ophthalmic care. For example, the NHS in the United Kingdom (UK) is involved in AI trials in areas such as cataract follow-up and monitoring of age-related macular degeneration (AMD). Despite this momentum, AI remains confined to research and limited clinical settings, with widespread adoption yet to be achieved. This review outlines the current clinical applications of AI in ophthalmology, identifies ongoing challenges and discusses future directions and ethical considerations for global implementation.

Current clinical applications of AI in Ophthalmology

AI, particularly deep learning, has already demonstrated significant clinical value in ophthalmology, most notably in the screening and diagnosis of diabetic retinopathy (DR). In 2018, the FDA approved IDx-DR, the first autonomous AI diagnostic tool in healthcare, capable of detecting more-than-mild DR directly in primary care settings without clinician interpretation. Subsequent studies confirmed its high sensitivity and specificity in detecting DR and diabetic macular oedema, highlighting its potential as a significant tool in AI-driven ophthalmic diagnostics (1). A large meta-analysis of over 134,000 patients across ten years further demonstrated the strong performance of AI-based DR screening, reporting an overall accuracy of 81%, sensitivity of 94%, and specificity of 89% compared to human graders (2). Several commercial AI platforms such as EyeArt, RetmarkerDR, and iGradingM have since been deployed across North America, Europe, and Australia, reducing specialist workload and expanding screening coverage (3).

Scotland has successfully integrated automated grading algorithms into its national DR programme, where images confidently classified as “no retinopathy” by AI are excluded from further human review, significantly improving efficiency (4). These applications highlight how AI models, trained to identify retinal features such as haemorrhages, microaneurysms, and exudates, can achieve diagnostic performance comparable to or exceeding human graders (5).

Beyond DR, AI applications are expanding to other posterior segment diseases, including AMD and glaucoma. In glaucoma, deep learning models have been trained on OCT and visual field data to detect early structural changes, predict progression, and support risk stratification (5,6,7). In AMD, AI systems can quantify biomarkers such as central retinal thickness and intraretinal or subretinal fluid from OCT scans, supporting monitoring and personalised treatment planning (7).

AI is also emerging as a valuable tool in the screening and management of retinopathy of prematurity (ROP), a vasoproliferative retinal disorder that has become a leading cause of childhood blindness in middle-income countries. Increasing neonatal survival rates, unregulated oxygen therapy, and limited access to screening have contributed to a growing ROP burden. Telemedicine offers a valuable and innovative approach to ROP care in middle-income countries, enabling retinal images captured by non-ophthalmologists to be remotely interpreted, thereby reserving specialist paediatric ophthalmologists for diagnosis and treatment rather than screening.

A 2025 study demonstrated that a machine learning algorithm using smartphone captured retinal videos of premature neonates achieved high sensitivity but exhibited lower specificity when compared to paediatric ophthalmologists grading‑ images. This usable process may substantially increase screening for ROP in low resource settings. As screening is one of the major barriers to receiving ROP treatment, offloading this task to a less scarce clinician with less specialised training could free paediatric ophthalmologists to focus on activities that require their specific expertise (8).

Despite these promising results, challenges persist. Algorithms trained on limited datasets may not recognise advanced stages of ROP, where image quality is poor and small variations between imaging devices can significantly alter AI outputs, raising concerns about reproducibility. However, when integrated with telemedicine as a hybrid approach, AI could automatically triage infants and flag high-risk cases for urgent review. Although challenges remain, AI-supported telemedicine offers considerable potential to expand ROP screening in middle-income countries, where the alternative is often no screening at all.

Barriers to Implementation, Ethical considerations and future directions

Advancements in AI could fundamentally reshape ophthalmology. Future applications and advances in deep learning could integrate AI into triage, diagnosis, management, and patient education, improving access to care worldwide. However, achieving this vision demands coordinated solutions to practical and ethical obstacles.

One of the concerns is algorithmic bias as most high-quality data, particularly images come from high-income countries, limiting applicability in diverse populations. Infrastructure gaps including limited computing power and internet access in low- and middle-income countries further hinder adoption. AI’s high energy demands also raise environmental concerns for sustainable use (9).

Regulatory and methodological issues persist. Few ophthalmic AI tools have undergone large-scale clinical trials and global standards for validation and safety remain unclear. Ethically, clarity is needed on whether AI should operate independently or as an assistive tool with clinician oversight. Ensuring fairness requires addressing underrepresentation in training data and designing scalable, low-cost AI systems to prevent worsening disparities in eye care.

Conclusion

In conclusion, AI offers transformative potential for ophthalmology, offering solutions to pressing challenges such as the global burden of retinopathy of prematurity and diabetic retinopathy. By combining AI with teleophthalmology, there is the opportunity to extend high-quality screening and diagnosis to underserved regions, addressing specialist shortages and preventing avoidable blindness. But success will depend on ensuring models are generalisable, systems are affordable, and policies are in place to protect patients and promote trust. With careful design AI can move from a promising innovation to a practical, globally accessible tool that reshapes eye care and improves vision health globally.

References

  1. Abràmoff MD,‑ Lavin PT, Birch M, Shah N, Folk JC, Townsend R, et al. Pivotal trial of an autonomous AI based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digit Med. 2018;1:39. doi:10.1038/s41746-018-0040-6
  2. American Academy of Ophthalmology. AI-based Fundus Image Algorithms May Assist With Diabetic Retinopathy Detection. EyeNet Magazine. 2024 Aug 21. Available from: https://www.aao.org/education/editors-choice/ai-based-fundus-image-algorithms-may-assist-with-d
  3. Ong AY, Taribagil P, Sevgi M, Kale AU, Dow ER, Macdonald T, Kras A, Maniatopoulos G, Liu X, Keane PA, Denniston AK, Hogg HDJ. A scoping review of artificial intelligence as a medical device for ophthalmic image analysis in Europe, Australia and America. NPJ Digit Med. 2025 May 29;8(1):323. doi:10.1038/s41746-025-01726-8. PMID:40442400; PMCID:PMC12122805.
  4. Styles CJ. Introducing automated diabetic retinopathy systems: it’s not just about sensitivity and specificity. Eye (Lond). 2019;33:1357–1358. doi:10.1038/s41433-019-0428-1
  5. Martins TGDS, Schor P, Mendes LGA, Fowler S, Silva R. Use of artificial intelligence in ophthalmology: a narrative review. Sao Paulo Med J. 2022 Nov-Dec;140(6):837-845. doi:10.1590/1516-3180.2021.0713.R1.22022022. PMID:36043665; PMCID:PMC9671570.
  6. Anton N, Doroftei B, Curteanu S, et al. Comprehensive review on the use of artificial intelligence in ophthalmology and future research directions. Diagnostics (Basel). 2023;13(1):100. doi:10.3390/diagnostics13010100.
  7. Ting DSW, Pasquale LR, Peng L, Campbell JP, Lee AY, Raman R, Sabanayagam C, Schmetterer L, Keane PA, Wong TY. Artificial intelligence and deep learning in ophthalmology. Br J Ophthalmol. 2019;103(2):167–175‑. doi:10.1136/bjophthalmol-2018-313173.
  8. Ortiz‑ A, Patiño S, Torres J, et al. AI Enabled Screening for Retinopathy of Prematurity in Low ResourceSettings.JAMANetwOpen.2025;8(4):e257831. doi:10.1001/jamanetworkopen.2025.7831
  9. Ueda D, Walston SL, Fujita S, et al. Climate change and artificial intelligence in healthcare: Review and recommendations towards a sustainable future. Diagn Interv Imaging. 2024;105(11):453-9. doi:10.1016/j.diii.2024.06.002

Leave a Reply