Hybrid deformable model for aneurysm segmentation

Automatic extraction of aortic aneurysm thrombus is a non-trivial challenge for existing segmentation algorithms. Due to similar intensity, the boundary to surrounding tissue is characterized by a small gradient. On the other hand, the aneurysm contains calcification spots that introduce wrong gradients. Therefore, purely intensity- or gradient-based methods fail to give optimal results. In this paper, we present a hybrid deformable model approach that integrates local and global image information and combines it with shape constraints. By the use of NURBS surfaces and distance functions, segmentation leakage into adjacent structures is prevented. The results of several experiments were evaluated by standard measures and expert inspection.