Abstract: FR-PO0009
Retinal Biomarkers Demonstrate Strong Diagnostic Performance for CKD Stage 3B Detection in UK Biobank Participants
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Thakur, Sahil, Mediwhale Inc, Seoul, Korea (the Republic of)
- Rukmini, Annadata V., Mediwhale Inc, Seoul, Korea (the Republic of)
- Park, Junseok, Mediwhale Inc, Seoul, Korea (the Republic of)
- Nam, Dongjin, Mediwhale Inc, Seoul, Korea (the Republic of)
- Cho, Jungkyung, Mediwhale Inc, Seoul, Korea (the Republic of)
- Park, Tae Hyun, Mediwhale Inc, Seoul, Korea (the Republic of)
- Seo, Jaewon, Mediwhale Inc, Seoul, Korea (the Republic of)
- Nusinovici, Simon, Mediwhale Inc, Seoul, Korea (the Republic of)
- Rim, Tyler Hyungtaek, Mediwhale Inc, Seoul, Korea (the Republic of)
Background
Chronic kidney disease (CKD) is a major global health issue with high morbidity and mortality. Early detection is challenging. Given the shared microvascular features of the retina and kidney, retinal imaging has emerged as a potential non-invasive biomarker for CKD. This study assessed the diagnostic accuracy of deep learning-derived retinal biomarkers for detecting CKD stage 3B in a large population-based cohort.
Methods
Retinal images from UK Biobank participants (N=20,322) were analyzed to derive three biomarker scores: Reti-ACR (predicting albumin-to-creatinine ratio), Reti-CKD (predicting CKD status), and Reti-eGFR (predicting estimated glomerular filtration rate). Diagnostic performance for detecting CKD stage 3B (eGFR <45 ml/min/1.73m^2 and ACR >3 mg/mmol) was evaluated using ROC curve analysis.
Results
All three biomarkers showed strong diagnostic performance. Reti-ACR had the highest accuracy with an AUC of 0.805. Reti-eGFR and Reti-CKD also performed well with AUCs of 0.796 and 0.772, respectively. These results demonstrate that retinal biomarkers can effectively detect reduced kidney function before clinical symptoms appear.
Conclusion
Deep learning-derived retinal biomarkers offer promising non-invasive tools for identifying CKD stage 3B in general populations. The Reti-ACR score, in particular, showed excellent discrimination, underscoring the link between retinal microvasculature and albuminuria. These findings support the integration of retinal imaging in CKD screening and risk stratification to enable earlier intervention.
Figure: ROC curves showing the performance of Reti-ACR, Reti-eGFR, and Reti-CKD scores in predicting CKD stage 3B using UK Biobank data.