AI in Ophthalmology
Diabetic retinopathy screening from fundus photos is the clearest success. Autonomous systems have been cleared in the US and are used in primary care. AI for OCT, glaucoma and retinopathy of prematurity is growing, and retinal foundation models are in research.
What AI does in Ophthalmology
Reads fundus photos in primary care or diabetes clinics and says whether referral is needed.
Examples: LumineticsCore, EyeArt
Detects and measures fluid in macular disease and tracks response to injections.
Examples: Vendor OCT software
Reads disc photos or OCT to flag likely glaucoma.
Examples: Research and commercial tools
Grades plus disease on infant retinal images, which helps where few specialists exist.
Examples: Research systems
AI on images from low-cost cameras for community screening.
Examples: Research and field programmes
One pre-trained model adapted to many eye diseases.
Examples: RETFound
routine = in everyday use in many centres · early adoption = in use at some sites, evidence still building · research = studies only
Key ideas for this specialty
New to the basics? Start with the theory lessons.
Tools
Autonomous diabetic retinopathy diagnosis in primary care.
Regulatory status: FDA cleared (2018), per Digital Diagnostics
Diabetic retinopathy AI screening from fundus photos.
Regulatory status: FDA cleared, per search sources
Public fundus dataset for DR grading practice.
Regulatory status: Not listed
Large public fundus dataset with DR grades.
Regulatory status: Not listed
A listing here is not an endorsement. Ask any vendor for its CDSCO licence and class before clinical use.
Where to learn
The trial behind the first autonomous AI clearance.
Large multiethnic validation from Singapore.
Good start for residents.
Introduces RETFound.
Indian data.
General medical imaging AI grounding.
Do it yourself
Download the APTOS fundus images from Kaggle, fine-tune a small pretrained image model in a free notebook, and check kappa on a held-out set. Then test on a few images from another source.
Open the dataset or tool →Clone RETFound and read its fine-tuning instructions. Run the example on a public dataset to see how a foundation model needs little labelled data.
Open the dataset or tool →Use public or properly de-identified data only, and get ethics approval before any project on your own patients.
Papers worth reading
- A foundation model for generalizable disease detection from retinal images Nature, 2023
- Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices npj Digital Medicine, 2018
- Performance of a Deep-Learning Algorithm vs Manual Grading for Detecting Diabetic Retinopathy in India JAMA Ophthalmology, 2019
- Artificial intelligence and deep learning in ophthalmology British Journal of Ophthalmology, 2018
Societies and guidance
Indian national society; check for AI sessions and screening programmes.
Links checked 3 October 2026. “Not listed” means we could not confirm the detail, not that it is absent. Educational material, not clinical advice.