Preclinical deep learning model for diabetic retinopathy screening in low-resource settings

Photo by Elena Mozhvilo / Unsplash

medRxiv Published April 23, 2026 Medically reviewed April 24, 2026 Study authors: SHI, M.; Afolabi, S. O. DOI ↗ By Dr. Lars van Dijk, PhD · Surgical, Procedural & Diagnostic

Key Takeaway

Consider this preclinical model for DR screening in low-resource settings, but recognize it requires clinical validation.

This is a preclinical study evaluating a deep learning model using an imbalance-aware optimal transport (OT) learning approach for diabetic retinopathy (DR) detection from smartphone-acquired images in low-resource settings. The model was compared against empirical risk minimization (ERM), Prototype OT, and Sinkhorn OT methods.

The authors report that the proposed model achieved an AUC of 87% (95% CI approximately 84% to 89%), a sensitivity of 89% (95% CI 81% to 96%), and a specificity of 95% (95% CI 93% to 96%). These results represent an increase over the comparator methods.

The study acknowledges limitations, noting that this is a preclinical evaluation with bootstrapping and not a clinical trial. No adverse events, safety data, or follow-up periods were reported. The authors caution against making causal claims or inferring generalizability beyond the reported smartphone image domain.

The authors suggest the framework shows promising results for low-resource DR screening, which could potentially benefit less-advantaged groups and developing nations. However, the evidence is early and requires clinical validation before any practice changes.

Study Details

EvidenceLevel 5

PublishedApr 2026

View Original Abstract ↓

Abstract Background Diabetic Retinopathy (DR) is one of the leading cause of vision loss and blindness. AI models have been instrumental in providing an alternative solution to real-life medical treatment which are costly and sometimes not readily available in developing and underdeveloped nations. However, most of the existing AI models are developed with high-quality clinical images that makes it difficult to use such models in low-resource settings. For this reason, this research focus on bridging this gap by developing a low-resource, mobile-friendly, and deployable deep learning (DL) model for the detection of DR using an imbalance-aware optimal transport (OT) learning approach. Methods We trained our proposed framework with both high-quality hospital- grade images and low-resource smartphone-acquired images, and evaluated with the original test set from the smartphone domain. We also curated three levels of smart- phone image-degradation quality and reported results from multiple experiments with bootstrapping. All model evaluations were assessed using the AUC, Sensitivity, and Specificity. Our results were compared with empirical risk minimization (ERM), Prototype OT, and Sinkhorn OT methods. Results We used four strong backbone architectures in the assessment. With our framework, Mobilevit-s achieved the best performance: an AUC of 87%, sensitivity of 89%, and specificity of 95%. Meanwhile, the statistical significance performance test (95% CI) shows that the AUC results are in the range of approximately 84% to 89%. For sensitivity, the range is 81% to 96%, and for specificity, 93% to 96%. This result indicated a performance increase of more than 3-5% compared to baseline methods. Conclusion Our framework shows promising results for low-resource DR screening, which has a potential to benefit less-advantaged groups and developing nations. Keywords Diabetic retinopathy, cost-effective AI, optimal transport, smartphone screening, deep learning.