Peer Reviewed

1

DOI

10.17077/omia.1038

Conference Location

Munich, Germany

Publication Date

October 2015

Abstract

Cup-to-disc ratio is commonly used as an important parameter for glaucoma screening, involving segmentation of the optic cup on fundus images. We propose a novel polar map representation of the optic disc, using a combination of supervised and unsupervised cup segmentation techniques, for detection of glaucoma. Instead of performing hard thresholding on the segmentation output to extract the cup, we consider the cup confidence scores inside the disc to construct a polar map, and extract sector-wise features for learning a glaucoma risk probability (GRP) for the image. We compare the performance of GRP vis-à-vis the cup-to-disc ratio (CDR). On an evaluation dataset of 100 images from the publicly available RIM-ONE database, our method achieves 82% sensitivity at 84% specificity, and 96% sensitivity at 60% specificity (AUC of 0.8964). Experiments indicate that the polar map based method can provide a more discriminatory glaucoma risk probability score compared to CDR.

Rights

Copyright © 2015 Akshaya Ramaswamy, Keerthi Ram, Niranjan Joshi, Mohanasankar Sivaprakasam

Included in

Ophthalmology Commons

Share

COinS
 
Oct 9th, 12:00 AM Oct 9th, 12:00 AM

A Polar Map Based Approach Using Retinal Fundus Images for Glaucoma Detection

Munich, Germany

Cup-to-disc ratio is commonly used as an important parameter for glaucoma screening, involving segmentation of the optic cup on fundus images. We propose a novel polar map representation of the optic disc, using a combination of supervised and unsupervised cup segmentation techniques, for detection of glaucoma. Instead of performing hard thresholding on the segmentation output to extract the cup, we consider the cup confidence scores inside the disc to construct a polar map, and extract sector-wise features for learning a glaucoma risk probability (GRP) for the image. We compare the performance of GRP vis-à-vis the cup-to-disc ratio (CDR). On an evaluation dataset of 100 images from the publicly available RIM-ONE database, our method achieves 82% sensitivity at 84% specificity, and 96% sensitivity at 60% specificity (AUC of 0.8964). Experiments indicate that the polar map based method can provide a more discriminatory glaucoma risk probability score compared to CDR.