Vessel segmentation using prior shape based on tensor analysis for inhomogeneous intensity and weak-edge images

Blood vessel segmentation is a key component for various clinical applications. In this paper, we present a novel method for vessel segmentation by using a prior shape based on tensor analysis and then incorporate it in a level-set-based segmentation method. We firstly introduce the prior shape via tensor analysis, which formulates the fractional anisotropy and anisotropic character of the mechanical tensor. Comparing to conventional statistical prior shape models, the main advantage of the proposed prior shape is that it directly derives from the given clinical images via tensor analysis, instead of statistical shape from a training sample set, leading to a simple and practice method for complex vascular structures. We subsequently explicitly incorporate the prior shape in our hybrid energy function, which enforces the segmentation depending on the joint influence of the region-homogeneity, gradient (edge), and the proposed prior shape. We validate our method both on the synthetic images and multimodal clinical images, which shows that our method outperforms the competing methods.