Work

Using AI-generated ultra-widefield-like fundus images to assist colour fundus photography in screening for chronic kidney disease: a multicentre cross-sectional study

Year: 2026
Type: article
Study design: Observational Study
Abstract: Objective To develop a generative artificial intelligence (AI) model that derives ultra-widefield-like (UWF-like) fundus images from conventional colour fundus photographs (CFP) and to assess whether these images improve chronic kidney disease (CKD) screening.Methods and analysis This multicentre cross-sectional diagnostic study included 3776 fundus images from 2056 patients at four centres in China and Japan between 2023 and 2024. The generative model was developed and validated using 1865 images from 1049 patients. CKD screening models were developed using 1750 UWF images and their corresponding UWF-derived CFP-like central crops from 909 patients at Ruijin Hospital with patient-level fivefold cross-validation and externally validated using 161 native CFP images from 98 patients at Jiading Hospital of Traditional Chinese Medicine. Performance was assessed by area under the receiver operating characteristic curve (AUC) and specificity at fixed sensitivity of 0.80.Results In internal validation, AUCs were 0.911 for real UWF images, 0.773 for AI-generated UWF-like images and 0.689 for UWF-derived CFP-like central crops. At fixed sensitivity of 0.80, specificities were 0.879, 0.573 and 0.459, respectively. In external validation, AI-generated UWF-like images outperformed native CFP images, with an AUC of 0.707 versus 0.609 and specificity of 0.458 versus 0.344. Retinal vascular features from AI-generated UWF-like images remained associated with renal function indicators.Conclusion AI-generated UWF-like images improved CKD screening performance compared with CFP alone. This approach may extend retinal imaging-based CKD screening where UWF imaging is unavailable, but prospective real-world validation is needed before clinical implementation.
Language: English
Field: Medicine
Open Access status: gold
Estimated APC paid: $2,207
OpenAlex ID: W7225815138
FWCI:
Cites: 41
Cited by:
Related to: 10