Performance of deep learning for detection of chronic kidney disease from retinal fundus photographs: A systematic review and meta-analysis.
Level 1 - systematic review of randomized trials
Systematic review and meta-analysis of diagnostic accuracy studies
PubMed 37671422 · doi:10.1177/11206721231199848
What was done
Authors conducted a systematic review and meta-analysis of studies evaluating deep learning models for detecting chronic kidney disease (CKD) using retinal fundus photographs. Searches were performed across PubMed, Embase, Cochrane Library, and Web of Science up to October 31, 2022. Study risk of bias was assessed using the QUADAS-2 tool.
What was found
Four studies comprising 114,860 subjects were included. Pooled diagnostic metrics reported: - Sensitivity: 87.8% (95% CI: 61.6% to 98.3%) - Specificity: 62.4% (95% CI: 44.9% to 78.7%) - Area under the curve (AUC): 0.864 (95% CI: 0.769 to 0.986)
Why it matters
This review shows that retinal deep learning algorithms have proof-of-concept potential for non-invasive CKD detection, but moderate specificity and wide variation indicate they are not yet ready for standalone clinical practice.
Limits
The meta-analysis included only four studies. Imprecision is substantial, reflected in wide confidence intervals for both sensitivity (61.6%–98.3%) and specificity (44.9%–78.7%). Ground-truth CKD definitions, imaging hardware heterogeneity, and population demographics are not reported in the abstract.
Cited by
- supports AI applied to high-resolution retinal imaging can detect early signs of Alzheimer's disease and kidney dysfunction.