A Supervised Bayesian Method for Cerebrovascular Segmentation

In this paper, we present a supervised statistical-based cerebrovascular segmentation method from Time-Of-Flight MRA. The novelty of this method is that rather than model the dataset over the entire intensity range, we at first use a low threshold to eliminate the lowest intensity region, and then use two uniform distributions to model the middle and high intensity regions, respectively. Subsequently, in order to overcome the intensity overlap between subcutaneous fat and arteries, a high order multiscale features based energy function is introduced to enhance the segmentation. Comparing with those sole intensity based segmentation method the newly proposed algorithm can solve the problem of the regional intensity variation of TOF–MRA well and improve the quality of segmentation. The experimental results also show that the proposed method can provide a better quality segmentation than sole intensity information used method. Key-Words: Statistical segmentation, Bayesian method, Maximum a posteriori (MAP) estimation, Markov Random field, High-order multiscale features.