Dr. MANAVI DEOKRISHNA SINDAL

DR. NIKITA SONAWANE, DR.DIVYA RAO

Abstract

Retrospective, cross-sectional study of performance of automated screening tool for referable DR (RDR) using AI deployed offline on smartphone-based fundus camera at 12 Primary eye centres (PEC). Disc and macula centred images captured using a smartphone-based fundus camera were analysed for RDR (>moderate NPDR +/- CSME) instantly by AI & graded by blinded PEC graders . Patients with RDR diagnosed were referred to base hospital. A total of 279 patients were analysed. The sensitivity of AI vs PEC doctors was 95.92% (95% CI 86.02% to 99.50%), specificity 75.65% (95% CI 69.58% to 81.05%), positive predictive value was 45.63% (95% CI 39.89% to 51.49%) & negative predictive value was 98.86% (95% CI 95.72% to 99.71%). Sub-analysis of 50 referred patients showed AI sensitivity of 100% vs 75% of VC doctors for RDR. Diabetic retinopathy screening using an AI deployed on a smartphone-based fundus camera provides a robust, fast & easy-to-use triaging tool in communities with limited resource

Full Text