← Back to the atlas
No. 1271Problem statementHealth and wellbeing

Explainable AI for Diabetic Retinopathy Screening in Rural India

Source

Source summary

The MATLAB retinal-screening prototype should assess image quality, enhance borderline images and reject ungradeable ones with recapture feedback. It should extract retinal structures and lesions, grade diabetic retinopathy from levels 0–4 and explain results with Grad-CAM, lesion evidence and calibrated confidence for ophthalmologist review. Source targets are over 90% sensitivity and 85% specificity for referable disease, plus review under 30 seconds. A Simulink service model should assess throughput and constrained bandwidth at district scale; these are validation targets, not achieved clinical results.