Non-invasive biopsy diagnosis of diabetic kidney disease via deep learning applied to retinal images: a population-based study
Problem
Screening for diabetic kidney disease and distinguishing isolated diabetic nephropathy from non-diabetic kidney disease are major access-limited challenges in diabetes care.
Why it matters
Early, accessible detection of kidney complications could improve outcomes for millions with diabetes.
What the AI did
DeepDKD, a deep learning system pretrained on 734,084 retinal fundus images, detects DKD and isolated diabetic nephropathy from retinal images.
Results
not stated in source
Impact
Abstract describes development and external validation across ten multi-ethnic datasets, but performance metrics are not visible.
Record
| Venue | The Lancet Digital Health |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Ziyao Meng, Zhouyu Guan, Shujie Yu, Yi-Lan Wu, Yaoning Zhao, Jie Shen, Cynthia Ciwei Lim, Ting‐Li Chen et al. |
| Citations | 47 (OpenAlex, at import time) |
| References | 40 |
| Model or method | DeepDKD (deep learning system) |
| Organization | Shanghai Integrated Diabetes Prevention and Care System (data source) |
| Primary source | https://doi.org/10.1016/j.landig.2025.02.008 |
| Open access | gold |
| Keywords | deep learning, retinal images, diabetic kidney disease, screening, external validation |
Importance rating
significant — thousands to millions affected
DKD screening and differential diagnosis are called major access-limited challenges in diabetes care; retinal-based AI screening could reach large diabetic populations, with multi-country validation.