Realistic fundus photograph generation for improving automated disease classification
British Journal of Ophthalmology, vol. 109 (7), pp. 791–798, 2025
Diffusion models generated retinal photographs realistic enough that ophthalmologists told real from synthetic only 61% of the time; adding them to training improved a disease classifier’s accuracy by 5.8%.
@article{pandey2025realistic,
title = {Realistic fundus photograph generation for improving automated disease classification},
author = {Prashant U. Pandey and Jonathan A. Micieli and Stephan Ong Tone and Kenneth T. Eng and Peter J. Kertes and Jovi C. Y. Wong},
journal = {British Journal of Ophthalmology},
volume = {109},
number = {7},
pages = {791--798},
year = {2025},
doi = {10.1136/bjo-2024-326122}
}