Shape-Specific Adaptations for Level-Set Deformable Model-Based Segmentation

In this paper we present two modifications to the original level set algorithm for implementation of deformable models. The modifications are motivated by difficulties that we have encountered in application of deformable models to segmentation of abdominal aortic aneurysm from computed tomography images. The level set algorithm has some advantages over the classical snake deformable models but it has difficulties with large gaps in the boundary of segmented region. Such boundary gaps may cause inaccurate segmentation that requires manual correction by the user while our goal is to keep user assistance at a minimum level. The proposed modifications have a form of additional stopping criteria. The first modification utilizes shape constraints and is less general than the second modification, which utilizes feature-based tracking of curve segments. These two modifications are developed for our specific application but we believe that they could be utilized in any similar application.