Optic disc segmentation for glaucoma screening system using fundus images

Ahmed Almazroa,1,2 Weiwei Sun,3 Sami Alodhayb,4 Kaamran Raahemifar,5 Vasudevan Lakshminarayanan6 1King Abdullah International Medical Research Center (KAIMRC), Riyadh, Saudi Arabia; 2Ophthalmology and Visual Science Department, University of Michigan, Ann Arbor, MI, USA; 3School of Resource and Environmental Sciences, Wuhan University, Wuchang, Wuhan, Hubei, China; 4Bin Rushed Ophthalmic Center, Riyadh, Saudi Arabia; 5Department of Electrical and Computer Engineering, University of Ryerson, Toronto, ON, 6School of Optometry, University of Waterloo, ON, Canada Abstract: Segmenting the optic disc (OD) is an important and essential step in creating a frame of reference for diagnosing optic nerve head pathologies such as glaucoma. Therefore, a reliable OD segmentation technique is necessary for automatic screening of optic nerve head abnormalities. The main contribution of this paper is in presenting a novel OD segmentation algorithm based on applying a level set method on a localized OD image. To prevent the blood vessels from interfering with the level set process, an inpainting technique was applied. As well an important contribution was to involve the variations in opinions among the ophthalmologists in detecting the disc boundaries and diagnosing the glaucoma. Most of the previous studies were trained and tested based on only one opinion, which can be assumed to be biased for the ophthalmologist. In addition, the accuracy was calculated based on the number of images that coincided with the ophthalmologists’ agreed-upon images, and not only on the overlapping images as in previous studies. The ultimate goal of this project is to develop an automated image processing system for glaucoma screening. The disc algorithm is evaluated using a new retinal fundus image dataset called RIGA (retinal images for glaucoma analysis). In the case of low-quality images, a double level set was applied, in which the first level set was considered to be localization for the OD. Five hundred and fifty images are used to test the algorithm accuracy as well as the agreement among the manual markings of six ophthalmologists. The accuracy of the algorithm in marking the optic disc area and centroid was 83.9%, and the best agreement was observed between the results of the algorithm and manual markings in 379 images. Keywords: optic disc, image segmentation, RIGA dataset, glaucoma, level set, image inpainting