Artificial intelligence and deep learning in ophthalmology

AI Generated Summary*

A narrative review maps deep learning across ophthalmic imaging, covering fundus photographs, OCT and visual fields. For referable diabetic retinopathy, headline AUCs run from roughly 0.94 to 0.99, while one system's ten external sets (40,752 images, six countries) ranged from 0.889 to 0.983; a separately approved US system reached 87.2% sensitivity and 90.7% specificity in prospective testing. Other examples include AMD classification on AREDS photographs (AUC 0.94 to 0.96), OCT segmentation with triage, glaucoma-like disc grading, visual field archetype analysis and ROP plus disease detection (AUC 0.98). Obstacles include narrow training populations, shaky reference labels, absent power calculations, weak explainability, two-dimensional imaging, medicolegal variation and patient trust, with telemedicine-linked community screening presented only as a possibility pending cost-effectiveness work.

*This summary was generated by AI and is published unedited. Oku does not alter these summaries. It may contain errors or omissions and is provided for general informational purposes only. Oku does not guarantee its accuracy, completeness, or reliability. For authoritative information, please refer to the original, peer-reviewed article.