考虑对脑血管进行三维分割具有一定难度，提出了一种基于全局LBF (Local Binary Fitting)水平集模型的脑血管层次化粗分割方法。首先，应用定向加权中值(DWM)滤波和各向异性扩散滤波去除脑图像噪声，同时保存血管边缘信息，在多尺度条件下局部梯度最大(LIGM)算法，应用灰度和梯度信息提取备选血管，基本实现脑灰质去除。然后，改进全局信息LBF水平集算法实现最大强度投影(MIP)图像分割，采用形态信息提取备选血管，剔除干扰组织。最后，融合两种方法实现脑血管粗提取。实验表明，层次化的分割方法可去除大部分不相关脑组织，包含直接双高斯统计模型中的所有分割血管信息。本项研究基于时飞磁共振血管造影(TOF_MRA)数据，相关研究结果可扩展到其它相似系统中。

To solve the problem that human brain vessels are difficult to be segmented, a level coarse brain vessel segmentation based on the global Local Binary Fitting(LBF) model was presented in the paper. First, the Directional Weight Median(DWM) filtering and the anisotropic diffusion model were used to reduce the noise and to enhance the vessel edges of brain images. Then, the Local Intensity Gradient Maximum(LIGM) algorithm was implemented based on a multi-scale space. The information of intensity and gradient was used to get the vessel candidate set and remove the influence of gray matter in the brain. At the same time, the improved global LBF level set model was used to segment the Maximum Intensity Projection(MIP) image. The vessel voxels were extracted with the conformation information. The results of these two steps were fused together to get the minimal covering set of the brain vessel. The experimental results show that all most the vessel voxels directly segmented by the double Gauss model can be reserved and most uncorrelated voxels can be removed. This research is based on the Time of Flight Magnetic Resonance Angiography(TOF MRA) and it is easy to expand to the similar system.