Performance Evaluation of Optic Disc Segmentation Algorithms in Retinal Fundus Images: an Empirical Investigation

Careful evaluation of Optic nerve head structure and its documentation is extremely important for diagnosis of Glaucoma, an eye disease which leads to vision loss. This work focuses on automatic segmentation of Optic disc from fundus images, which is an important parameter for disease diagnosis. We investigate and compare performance of five methods used for Optic disc segmentation. These five methods are based on use of algorithms namely; distance regularized level set, Otsu thresholding, region growing, particle Swarm optimization, generalized regression neural network. For ease of comparison all the methods were implemented and tested on a single database. The method using generalized regression neural network best suits the said application. It outperforms the other four due to highest region agreement, lowest non overlap ratio, lowest relative absolute area difference and low execution time.