A Knowledge-based Approach for Segmenting Cerebral Vasculature in Neuroimages

In this paper, we present a novel vasculature segmentation algorithm that incorporates the knowledge of both vascular anatomy and imaging modality. In particular, emphasis is put on the segmentation of main cerebral vessels such as the Circle of Wills. The algorithm segments cerebral vasculature in two major steps. One is vasculature candidate calculation using local intensity distribution, where the knowledge of image properties is used to derive possible vascular voxels. The other is a knowledge-based region growing process, where the knowledge of the vascular anatomy is used in the selection of parameters for region growing including starting seeds, size of neighborhood, and resultant topology. The algorithm is tested on real SPGR MRA images. Experiments have shown that the topology of the tree extracted with our algorithm matched reliably with that of the tree extracted manually by experienced radiologist.