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均值方法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.
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.
作者
王喆
赵世凤
田沄
王学松
周明全
WANG Zhe;ZHAO Shifeng;TIAN Yun;WANG Xuesong;ZHOU Mingquan(College of Information Science and Technology,Beijing Normal University,Beijing 100875,China;Beijing Key Laboratory of Digital Preservation and Virtual Reality for Cultural Heritage,Beijing 100875,China)
出处
《郑州大学学报（工学版）》
CAS
北大核心
2019年第1期18-23,31,共7页
Journal of Zhengzhou University（Engineering Science）
基金
国家自然科学基金资助项目(61672103
61472042)
北京市自然科学基金资助项目(4174094)
青岛市科技计划项目(2017-4-3-2-xcl)
关键词
脑血管分割
有限混合模型
K均值
cerebral vessel segmentation
finite mixture model
K-means
