Optic cup segmentation: type-II fuzzy thresholding approach and blood vessel extraction

• Published in: Clinical Ophthalmology
• Main Authors: Almazroa,Ahmed, Alodhayb,Sami, Raahemifar,Kaamran, Lakshminarayanan,Vasudevan
• Format: Article in Journal/Newspaper
• Language: English
• Published: Dove Press 2017
• Subjects: Clinical Ophthalmology; Canada; sami
• Online Access: https://www.dovepress.com/optic-cup-segmentation-type-ii-fuzzy-thresholding-approach-and-blood-v-peer-reviewed-fulltext-article-OPTH

Ahmed Almazroa,1 Sami Alodhayb,2 Kaamran Raahemifar,3 Vasudevan Lakshminarayanan1 1School of Optometry and Vision Science, University of Waterloo, Canada; 2Binrushd Ophthalmic Center, Riyadh, Saudi Arabia; 3Department of Electrical Engineering, Ryerson University, Toronto, Canada Abstract: We introduce here a new technique for segmenting optic cup using two-dimensional fundus images. Cup segmentation is the most challenging part of image processing of the optic nerve head due to the complexity of its structure. Using the blood vessels to segment the cup is important. Here, we report on blood vessel extraction using first a top-hat transform and Otsu’s segmentation function to detect the curves in the blood vessels (kinks) which indicate the cup boundary. This was followed by an interval type-II fuzzy entropy procedure. Finally, the Hough transform was applied to approximate the cup boundary. The algorithm was evaluated on 550 fundus images from a large dataset, which contained three different sets of images, where the cup was manually marked by six ophthalmologists. On one side, the accuracy of the algorithm was tested on the three image sets independently. The final cup detection accuracy in terms of area and centroid was calculated to be 78.2% of 441 images. Finally, we compared the algorithm performance with manual markings done by the six ophthalmologists. The agreement was determined between the ophthalmologists as well as the algorithm. The best agreement was between ophthalmologists one, two and five in 398 of 550 images, while the algorithm agreed with them in 356 images. Keywords: optic cup, image segmentation, glaucoma, blood vessel kinks, fuzzy type-II thresholding, retinal fundus images, optic disk