Segmentation of Cerebral Vascular Structures Using an Active Contour Model

Segmentation of brain blood vessels is essential in medical diagnostic applications. In this study, a new active contour model (ACM) implemented by the level-set framework is proposed for segmenting vessels from TOF-MRA data. The energy function of the proposed model, combining region intensity and boundary information, is composed of two region terms, one boundary term and one penalty term. The global threshold embedded into the first term is used to guide the extraction of thick vessels. While, the dynamic intensity threshold in the second term is employed to obtain the tiny ones. The boundary term is used to drive the contours to evolve towards the boundaries with high gradients. And lastly, the penalty term is used to avoid re-initializing the level set function. Compared with the global threshold based method and localized hybrid level-set method, experiments implemented on the segmentation of cerebral vessels present that our method is not only able to achieve better Dice Similarity Coefficient, but also able to extract whole cerebral vessel trees, including the thin vessels.