A Brazilian multilabel ophthalmological dataset (BRSET)

AI Generated Summary*

BRSET is a credentialed-access collection of 16,266 color fundus photographs from 8,524 patients at three outpatient ophthalmology services in São Paulo, Brazil. Metadata include camera, image center, age, sex, comorbidities, insulin use and diabetes duration, and one retina specialist graded optic disc, vessel and macular abnormalities, image quality, and pathology, with diabetic retinopathy staged on ICDR and Scottish scales. Female patients accounted for 61.8%, mean age was 57.6 years, and 15.8% had diabetes. A ConvNext V2 benchmark reached AUC 0.97 for binary retinopathy detection (macro F1 0.89) and for three-class grading (F1 0.82), while diabetes and sex prediction yielded AUCs of 0.87 to 0.91. Saliency maps highlighted localized lesions for retinopathy and broader retinal regions for sex.

*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.