Automatic detection of the optic cup using vessel kinking in digital retinal fundus images

Detection of the optic cup, which is an excavation in the optic disc, is an essential step in glaucoma assessment. In digital fundus photography, which captures 2D images of the retina, identification of the optic cup can be challenging. Kinks are bendings of vessels as they traverse the optic cup boundary, and are used clinically to determine the location of the optic cup. In this paper, we propose a method to automatically detect the optic cup boundary through the use of kinks. Patches are extracted within the optic disc, from which segment-based wavelet, edge and color features are generated for vessel candidates. A SVM technique is subsequently used to classify these candidates. To detect and localize kinking, a shifting multi-scale window interval is used to probe along the vessels. The obtained kinks are combined with pallor-based information to determine the optic cup. Experiments on a sample set of data show promising results for the proposed method, close to variability between the reference and a second observer segmentations.