DR. SWATI UPADHYAYA

DR.DIVYA RAO, DR. RENGARAJ VENKATESH, DR. KAVITHA S.

Abstract

A prospective study on 215 subjects in a Glaucoma center evaluated the diagnostic ability of a novel, AI-based screening tool for referable glaucoma integrated offline on a smartphone-based fundus camera. The AI was challenged with fundus images alone and was compared against the final diagnosis provided by masked glaucoma specialists following a thorough glaucoma work-up (clinical assessment, SD-OCT, HVF). Disc-centered fundus images were captured by nurses. Images were categorized as normal, glaucoma or suspect (predefined criterion). Referable glaucoma included certain glaucoma and suspects. Sensitivity of the AI was 85.91% (95% CI 79.27% to 91.06%), specificity 96.97% (95% CI 89.48% to 99.63%), PPV 98.46%, NPV 75.29% and accuracy 89.3%. The AI showed robust performance in detecting referable glaucoma with minimal overcall of normal cases. It has the potential to make glaucoma screening accessible, affordable and effective.

Full Text