Graph based optic nerve head segmentation

In this paper, we propose a fast and accurate method to locate and segment ONH boundary. This method integrates both the gradient and the local intensity cues to locate the ONH. It formulates the ONH segmentation problem into tracing an optimal path under graph theory framework. The optimal graph search technic considers both local and global information and makes our algorithm work well under poor ONH border contrast. Experiments over two public datasets show that the proposed method can accurate segment ONH boundary under blurred border or different pathological lesions. The average automatic ONH segmentation accuracies are 88.7% for MESSIDOR dataset and 88.6% for ARIA dataset. It can reach 90.2% for MESSIDOR dataset and 91.0% for ARIA dataset if use manual determined center.