[1]王喆,赵世凤,田沄,等.基于自适应聚类中心的脑血管分割方法[J].郑州大学学报(工学版),2019,40(01):18-23.[doi:10.13705/j.issn.1671-6833.2019.01.004]
 Wang Zhe,Zhao Shifeng,Tian Yun,et al.Cerebral Vessel Segmentation Based on Adaptive Clustering Centers[J].Journal of Zhengzhou University (Engineering Science),2019,40(01):18-23.[doi:10.13705/j.issn.1671-6833.2019.01.004]
点击复制

基于自适应聚类中心的脑血管分割方法()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
40卷
期数:
2019年01期
页码:
18-23
栏目:
出版日期:
2019-01-10

文章信息/Info

Title:
Cerebral Vessel Segmentation Based on Adaptive Clustering Centers
作者:
王喆赵世凤田沄王学松周明全
1. 北京师范大学信息科学与技术学院;2. 文化遗产数字化保护与虚拟现实北京市重点实验室
Author(s):
Wang Zhe12Zhao Shifeng12Tian Yun 12Wang Xuesong12Zhou Mingquan12
1. School of Information Science and Technology, Beijing Normal University ;2. Beijing Key Laboratory of Cultural Heritage Digital Protection and Virtual Reality
关键词:
脑血管分割有限混合模型K均值
Keywords:
Cerebrovascular segmentationfinite mixture modelsK-means
DOI:
10.13705/j.issn.1671-6833.2019.01.004
文献标志码:
A
摘要:
脑血管分割是血管病变可视化、诊断和定量分析的关键步骤。但由于脑血管几何结构复杂,所占空间面积小,因此低对比度区域的血管分割依然是难点。在传统的基于密度的统计方法基础上,进一步采用基于梯度的自适应聚类中心的K均值进行血管提取。首先,根据磁共振血管成像(MRA)图像密度特征用一个瑞利分布和两个高斯分布函数,分别对背景区域、中间组织区域以及血管区域进行建模,采用期望最大的方法进行参数估计,利用后验概率获取血管的主体部分;之后根据剩余体素中包含血管的部分多为低密度区的细小血管以及血管边界的特点,对剩余体素进行梯度化处理,并提出改进的自适应聚类中心的K均值对剩余体素的梯度数据进行血管的聚类操作。实验结果表明,对剩余数据梯度化的聚类方法优于传统的仅基于密度的统计方法,且能更好地获取血管的细小分支及血管的边缘区域。
Abstract:
Cerebral blood vessel segmentation was a key step in three-dimensional (3D) reconstruction, computer aided diagnosis and quantitative analysis. Due to complex geometric structure, small area percentage, low contrast vessel segmentation was still a challenging problem. Based on traditional statistical method with intensity, an improved K-means algorithm based on self-adapting clustering centers with gradient of remaining voxels preserved from previous step was used for further extraction of thin vessels. Firstly, one Rayleigh distribution and two Gaussian distributions were adapted to model background, tissues and vessel areas, respectively. And EM algorithm was used to estimate parameters for Gaussian distributions. Then posterior probability is used to extract the main body of blood vessels. Secondly, the remaining part containing the low contrast vessel areas and vessel edges was computed for gradient. An improved K-means method with self-adapting clustering centers was proposed to detect those areas. Experiment result demonstrated that our method was better than traditional statistical methods, especially at low contrast branches and vessel edges.
更新日期/Last Update: 2019-02-28