A deep learning algorithm to detect chronic kidney disease from retinal photographs in community-based populations

AI Generated Summary*

A deep learning algorithm was built to identify chronic kidney disease (eGFR below 60) from macula-centred retinal photographs. Development and internal validation used 6485 participants from the Singapore Epidemiology of Eye Diseases study, with external testing in 3735 Singapore Prospective Study Program and 1538 Beijing Eye Study participants. Three models were compared: image only, risk factors, and a hybrid. Image-only AUC was 0.911 internally, 0.733 in Singapore testing and 0.835 in Beijing. Risk factor models scored 0.916, 0.829 and 0.887. Image-only positive predictive value was low externally (14% in Singapore, 9% in Beijing) while negative predictive value stayed above 95%. In the internal validation set, diabetes and hypertension subgroups showed similar image-only performance to the whole group.

*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.