Peer Reviewed

1

DOI

10.17077/omia.1041

Conference Location

Athens, Greece

Publication Date

October 2016

Abstract

Qualitative evaluation of stereo retinal fundus images by experts is a widely accepted method for optic nerve head evaluation (ONH) in glaucoma. The quantitative evaluation using stereo involves depth estimation of the ONH and thresholding of depth to extract optic cup. In this paper, we attempt the reverse, by estimating the disc depth using supervised and unsupervised techniques on a single optic disc image. Our depth estimation approach is evaluated on the INSPIRE-stereo dataset by using single images from the stereo pairs, and is compared with the OCT based depth ground truths. We extract spatial and intensity features from the depth maps, and perform classification of images into glaucomatous and normal. Our approach is evaluated on a dataset of 100 images and achieves an AUC of 0.888 with a sensitivity of 83% at specificity 83%. Experiments indicate that our approach can reliably estimate depth, and provide valuable information for glaucoma detection and for monitoring its progression.

Rights

Copyright © 2016 the authors

Included in

Ophthalmology Commons

Share

COinS
 
Oct 21st, 12:00 AM Oct 21st, 12:00 AM

A Depth Based Approach to Glaucoma Detection Using Retinal Fundus Images

Athens, Greece

Qualitative evaluation of stereo retinal fundus images by experts is a widely accepted method for optic nerve head evaluation (ONH) in glaucoma. The quantitative evaluation using stereo involves depth estimation of the ONH and thresholding of depth to extract optic cup. In this paper, we attempt the reverse, by estimating the disc depth using supervised and unsupervised techniques on a single optic disc image. Our depth estimation approach is evaluated on the INSPIRE-stereo dataset by using single images from the stereo pairs, and is compared with the OCT based depth ground truths. We extract spatial and intensity features from the depth maps, and perform classification of images into glaucomatous and normal. Our approach is evaluated on a dataset of 100 images and achieves an AUC of 0.888 with a sensitivity of 83% at specificity 83%. Experiments indicate that our approach can reliably estimate depth, and provide valuable information for glaucoma detection and for monitoring its progression.