AI Analysis for Retinal Imaging

Artificial intelligence (AI) is transforming the interpretation of retinal images, with implications for improving disease diagnosis, management, and prognosis.

“Macular imaging has a critical role in routine practice for the diagnosis and management of retinal disease, but image review can be time intensive, and subtle findings that are needed to guide referrals to specialists and treatment may be easily missed,” said Katherine E Talcott MD. “In addition, we are facing a rising burden of retinal disease, and our clinics are overloaded with patients receiving frequent intravitreal injections. Together, these issues make our field ripe for applying AI to retinal imaging, where it can help in clinical care and research for identifying new disease biomarkers and the effectiveness of next-generation therapies.”

Screening for diabetic retinopathy using point-of-care fundus cameras coupled with autonomous AI diagnosis currently represents the most prominent clinical application of AI in macular imaging.

“These systems eliminate the need for image interpretation by an ophthalmologist,” Dr Talcott said. “And their deployment into offices of primary care physicians and endocrinologists, or even within pharmacies, can help address low screening rates for diabetic retinopathy and allow for referral of patients with earlier stage disease.”

Artificial intelligence-powered home OCT imaging devices for monitoring macular fluid levels in patients with neovascular age-related macular degeneration (nAMD) are also available for real-world use. This technology, which is intended to be used on a daily basis, automatically generates and transmits an alert to the retina specialist if the macular fluid level reaches or exceeds the clinician’s preset threshold.

“Home OCT has the potential to facilitate personalised treatment, improve patient outcomes by allowing earlier detection of disease activity, and reduce treatment burden for patients and physicians. The ongoing 104-week DRCR Retina Network Protocol AO comparing home OCT-guided treatment of nAMD with treat-and-extend management is evaluating if the AI-powered tool results in better visual acuity outcomes and/or fewer number of injections,” she said.

Artificial intelligence-powered tools that evaluate OCT image quality, detect various biomarkers, and identify abnormalities are also commercially available. Their programs label abnormal B-scans and recommend an interval for referral to a more specialised provider, if necessary.

Future applications

Other applications of AI to retinal imaging encompass an active area of research. These projects include research aimed at identifying new biomarkers that can serve as better efficacy endpoints in clinical trials. For example, a change in lesion area on fundus autofluorescence has been the gold standard for assessing the efficacy of treatments for geographic atrophy, but there is interest in applying AI to OCT images for quantitative analysis of the ellipsoid zone that reflects photoreceptor health.

“Currently, an individual reader has to look at every B-scan and trace every layer of the retina. The transition that is being developed using AI offers the potential for improving efficiency, reliability, and accuracy,” she noted.

Artificial intelligence-based analysis of ultra-widefield angiography imaging is another area of ongoing research.

“Ultra-widefield angiography images contain a wealth of information that remains untapped. AI-based algorithms that could automatically count microaneurysms and assess and map out areas of leakage and ischemia could overcome limitations of current analysis that relies on subjective clinician interpretation,” she predicted.

Dr Talcott spoke on this topic at the 2026 ASCRS annual meeting in Washington, DC.

Katherine E Talcott MD is a retinal surgeon and associate professor of ophthalmology at Cole Eye Institute, Cleveland Clinic, Cleveland, Ohio, US. talcotk@ccf.org