一种自适应立体脑图像分割方法  

An Approach to Adaptive Stereo Brain Image's Segmentation

作　　者：史勇红[1] 戚飞虎[1] 

机构地区：[1]上海交通大学计算机科学和工程系,上海200030

出　　处：《中国医疗器械杂志》2006年第2期88-89,87,共3页Chinese Journal of Medical Instrumentation

基　　金：国家自然科学基金项目(60271033)

摘　　要：提出一种自动分割单通道磁共振脑图像的新方法．这种方法建立在耦合的马尔科夫模型的基础上。耦合马尔科夫模型使用了两个空间上交织的马尔可夫随机场先验模型．其一对亮度测度建模以实现分段光滑性约束,另一个对非连续性建模以控制相邻体素间的交互作用。依据这一模型,该方法使用贝叶斯理论和领域约束获得了区域和边界的量大后验概率估计。这种方法具有如下属性:①大脑图像被准确地分割成白质、灰质和脑脊髓:②对噪声和亮度不一致性具有较强的鲁棒性。This paper presents a new method for automatically segmenting brain parenchyma and cerebrospinal fluid in routine single-echo MR images. This method is based on the coupled Markov models. They can model intensity measurement at each voxel site to implement piecewise smoothness constraint, and at the same time, model discontinuities to control the interaction between each pair of the neighboring voxel. The method is to derive the maximum a posteriori estimate of the regions and the boundaries by using Bayesian inference and neighborhood constraints based on Markov random fields （MRFs） models. This method has the following desirable properties： ① the brain image can be well classified into white matter, grey matter and cerebrospinal fluid （CSF）, and ②it has a better robustness to noise and intensity inhomogeneity.

关 键 词：非连续性 马尔可夫随机场 耦合马尔科夫模型 图像分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]