DR.DIVYA RAO

DR. SUJANI SHROFF, DR. ZIA PRADHAN, DR. SHRUTHI S.

Abstract

Aim: Vertical cup-to-disc ratio (vCDR) is a key feature to identify glaucoma. We evaluated the performance of an automated optic disc and cup segmentation and vCDR measurement tool using AI running offline on a portable fundus camera.

Methods: 268 consecutive subjects (536 eyes) from a glaucoma clinic were enrolled. Automated vCDR measurement was compared to SD-OCT and automated analysis by high-end Kowa camera (VK-2S WX) after a thorough glaucoma evaluation.

Results: The automated vCDR had a mean absolute error (MAE) of 0.053 and P-correlation of 0.85 against OCT. Kowa had a MAE of 0.096 with P-correlation of 0.46. A majority (265 images, 57.6%) had an absolute error < 0.05 vCDR versus OCT. The model identified thinning, excavation and notches. Conclusion: The vCDR tool has good accuracy when compared to OCT and is within clinically acceptable differences. It performs better than Kowa automated analysis. It allows for remote screening and usage as a clinical decision support tool.