AI-assisted training improves ophthalmologists’ diagnostic accuracy in retinal diseases

Incorporating an AI-assisted reading label system significantly improves ophthalmologists’ diagnostic accuracy for retinal diseases, with continued training leading to better performance, according to a study.

The study assessed 16 ophthalmologists with 1 to 9 years of experience, analyzing their diagnostic accuracy across multiple annotation rounds.

Researchers evaluated 7,777 pairs of optical coherence tomography (OCT) and color fundus photography (CFP) images labeled with 9 common retinal diseases, including diabetic retinopathy, retinal detachment, and macular degeneration. There was a significant improvement in diagnostic accuracy over 5 rounds (P = 0.013), with stronger correlations for more experienced ophthalmologists (P = 0.007). Both OCT (P = 0.028) and CFP (P = 0.021) accuracy improved with repeated training, with OCT performing better in diagnosing retinal detachment and macular abnormalities. At the same time, CFP was more reliable for diabetic retinopathy and retinal vein occlusion.

Reference
Wang M, Zhang X, Li D, et al. The potential of artificial intelligence reading label system on the training of ophthalmologists in retinal diseases, a multicenter bimodal multi-disease study. BMC Med Educ. 2025;https://doi.org/10.1186/s12909-025-07066-1.