Automatic optic cup segmentation using Kåsa's circle fitting technique

We present a technique for optic cup segmentation and outlining based on Kasa's circle fit model. The outlining problem is posed as a task of fitting a circle to the sparse set of optic cup boundary points. For automatic localization of the optic disc, we use the matched filtering technique. We clear-off the non-optic disc area by drawing a circle with point of optic disc localization as the coordinates of the center and diameter just above the normal optic disc diameter to overcome the problem of optic cup overestimation due to any retinal pathology. We report validation results on three publicly available fundus image databases, amounting to a total of 1411 fundus images for automatic optic disc localization, and 300 fundus images randomly selected for optic cup segmentation and outlining. The proposed method results in an optic disc localization accuracy of 94.06%, 94.17%, and 95.45%, and an average Dice similarity index of 0.7302, 0.7050, and 0.7120 on DRISHTI-GS, MESSIDOR, and DRIONS-DB fundus image databases, respectively. The average computation times for optic disc localization and optic cup segmentation are 3.83 and 5.33 seconds, respectively.