Non-invasive biopsy diagnosis of diabetic kidney disease via deep learning applied to retinal images: a population-based study

• DOI: 10.1016/j.landig.2025.02.008
• Publication Date: 2025-05-01T05:13:56Z
• Authors: Ziyao Meng; Zhouyu Guan; Shujie Yu; Yilan Wu; Yaoning Zhao; Jie Shen; Cynthia Ciwei Lim; Tingli Chen; Dawei Yang; An Ran Ran; Feng He; Haslina Hamzah; Sarkaaj Singh; Anis Syazwani Abd...; Jian Wen Samuel L...; Soo-Kun Lim; Xufang Sun; Shuwang Ge; Gang Xu; Hua Su; Yang Cheng; Feng Lu; Xiaofei Liao; Hai Jin; Chenxin Deng; Lei Ruan; Cuntai Zhang; Chan Wu; Rongping Dai; Yixiao Jin; Wenxiao Wang; Tingyao Li; Ruhan Liu; Jiajia Li; Jia Shu; Yuwei Lu; Xiangning Wang; Qiang Wu; Yiming Qin; Jin Tang; Xiaohua Sheng; Qiong Jiao; Xiaokang Yang; Minyi Guo; Gareth J McKay; Ruth E Hogg; Gerald Liew; Evelyn Yi Lyn Chee; Wynne Hsu; Mong Li Lee; Simon Szeto; Andrea O Y Luk; Juliana C N Chan; Carol Y Cheung; Gavin Siew Wei Tan; Yih-Chung Tham; Ching-Yu Cheng; Charumathi Sabana...; Lee-Ling Lim; Weiping Jia; Huating Li; Bin Sheng; Tien Yin Wong

Improving the accessibility of screening diabetic kidney disease (DKD) and differentiating isolated nephropathy from non-diabetic (NDKD) are two major challenges in field diabetes care. We aimed to develop validate an artificial intelligence (AI) deep learning system detect DKD retinal fundus images. In this population-based study, we developed a image-based AI-deep system, DeepDKD, pretrained using 734 084 First, for detection, used 486 312 images 121 578 participants Shanghai Integrated Diabetes Prevention Care System development internal validation, ten multi-ethnic datasets China, Singapore, Malaysia, Australia, UK (65 406 participants) external validation. Second, differentiate NDKD, 1068 267 three (244 Finally, conducted proof-of-concept studies: prospective real-world study with 3 months' follow-up evaluate effectiveness DeepDKD DKD; longitudinal analysis NDKD on renal function changes 4·6 years' follow-up. For detecting DKD, achieved area under receiver operating characteristic curve (AUC) 0·842 (95% CI 0·838-0·846) validation dataset AUCs 0·791-0·826 across datasets. AUC 0·906 (0·825-0·966) 0·733-0·844 compared metadata model, could higher sensitivity (89·8% vs 66·3%, p<0·0001). identified by had significant difference outcomes (proportion estimated glomerular filtration rate decline: 27·45% 52·56%, p=0·0010). Among diverse populations diabetes, showed its potential clinical practice. National Key R & D Program Natural Science Foundation Beijing Foundation, Municipal Clinical Specialty, Research Centre Endocrine Metabolic Diseases, Innovative research team high-level local universities Shanghai, Noncommunicable Chronic Diseases-National Technology Major Project, Special Health Commission, three-year action plan strengthen construction public health Shanghai.