局部信息约束的形状导向水平集算法对下颌管的分割  

Mandibular Canal Segmentation Based on Shape-Driven Level-set Algorithm Restrained by Local Information

作　　者：杨玲[1,2] 侯小叶[2] 王中科[3] 饶妮妮[1] 

机构地区：[1]电子科技大学生命科学与技术学院,成都610054 [2]成都信息工程学院电子工程学院,成都610225 [3]成都信息工程学院网络工程学院,成都610225

出　　处：《中国生物医学工程学报》2012年第2期161-166,共6页Chinese Journal of Biomedical Engineering

基　　金：四川省科技支撑计划项目(2011FZ0034)

摘　　要：针对下颌管在CT成像中相对于周围组织的模糊特性,提出一种基于局部信息约束的形状导向水平集算法,对下颌管进行高精度分割。首先根据下颌管分布的位置,重建出下颌管相应部位的多个侧断面,并进行下颌管提取;然后利用主成分分析方法(PCA),对下颌管进行先验形状统计;之后在形状导向水平集算法的基础上,引入下颌管周围组织的局部信息,对水平集演化的能量函数进行约束,以改进在模糊区域水平集的分割效果,从而有效地实现对任意侧断面中下颌管的分割。通过利用分割精度对感兴趣目标的分割结果进行度量,得出形状导向水平集算法的平均分割精度为4.19%,而本算法的平均分割精度为1.82%。该方法能有效对患者的下颌管进行精确的分割,为临床下颌管的定位提供了一种行之有效的新方法。In CT images,the mandibular canal is difficult to distinguish from other surrounding tissues due to its tubular structure.This paper proposed a shape-driven level-set algorithm restrained by local information to segment the mandibular canal with high accuracy.Firstly,according to the location of the distribution of the mandibular canal,we reconstructed many cross sectional images of corresponding parts of the mandibular canal,and extracted the mandibular canal from the cross sectional images,and then applied principal component analysis（PCA） to carry out the shape priori statistics of the mandibular canal.Finally,based on shape-driven level-set algorithm,restrain the evolution of the level-set energy function to improve the segment result in the fuzzy region by introducing the local information of the mandibular canal＇s surrounding tissue.By applying the criterion of segmentation accuracy in the region of interest,the segmentation accuracy of shape-driven level-set algorithm was 4.19%,while the segmentation accuracy of our method is 1.82%.Experimental results showed that this method could effectively segment patients＇ mandibular canal and provide an effective method for locating the position of the mandibular canal.

关 键 词：CT图像 下颌管 先验形状 形状导向 局部信息 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]