Mixture model-based approach for optic cup segmentation

Abstract

Glaucoma is a leading cause of blindness with permanent damage to optic nerve head. ARGALI is an automated computer-aided diagnosis system designed for glaucoma detection via optic cup-to-disc ratio assessment. It employs several methods to determine the optic cup and disc from retinal images. Optic disc detection and segmentation works have been widely reported with high success rate. However, the task of segmenting the optic cup in a non-stereo fundus photograph is more difficult due to the presence of retinal vessels and the inconsistent intensities of the optic cup. In this paper we propose an approach for optic cup segmentation based on Gaussian mixture models. The algorithm is tested on 71 images from the SiMES database. The optic cup boundaries in these images are manually segmented by a senior ophthalmologist as our clinical ground truth. In our experiments, we show that our approach is able to achieve an improvement of 8.1% in cup area overlap and 14.1% in relative area difference from the ARGALI cup segmentation. This demonstrates the capability of this model of have a closer segmentation to the clinical ground truth.