您的位置:首页  > 论文页面

基于统计形态学和Hessian矩阵的血管树分割算法

发表时间:2018-05-31  浏览量:688  下载量:246
全部作者: 刘勇清,房斌,王翊,郑申海
作者单位: 重庆大学计算机学院
摘 要: 针对门静脉期计算机断层扫描(computed tomography,CT)图像中存在的背景不均匀和噪声多等问题,提出一种基于统计形态学和Hessian矩阵的肝脏内血管树分割算法。首先,对原始CT图像使用改进的三维中值滤波进行去噪和平滑;其次,为获得更好的Hessian增强效果,针对背景像素值的灰度不均问题,提出一种基于邻域像素直方图的顶帽操作方法;然后,为增强断续状血管,对图像进行灰度变换,并进行多尺度Hessian增强;最后,对Hessian增强后的图像使用各向异性扩散滤波平滑,并利用一种两阶段的区域生长算法获得最终的分割结果。在4例数据集上的实验结果表明,该分割算法相比区域生长和全卷积神经网络(fully convolutional network,FCN)这两种图像分割算法,能够更加有效地分割出丰满而平滑的血管树,其最高的敏感度和Jaccard系数分别为84.9%和70.8%.
关 键 词: 人工智能;血管树分割;形态学操作;Hessian矩阵;各向异性扩散滤波;区域生长
Title: Vessel tree segmentation algorithm based on statistical morphology and Hessian matrix
Author: LIU Yongqing, FANG Bin, WANG Yi, ZHENG Shenhai
Organization: College of Computer Science, Chongqing University
Abstract: In view of the problems of background heterogeneity and noise in the portal venous phase computed tomography (CT) images, a liver vessel tree segmentation algorithm based on statistical morphology and Hessian matrix is proposed. Firstly, the improved three-dimensional median filtering is used to denoise and smooth the original CT images. Secondly, in order to obtain better Hessian enhancement effect, an improved top-hat operation based on neighborhood pixel histogram is proposed to solve the grayscale unevenness problem of background pixel. Then, to enhance the intermittent blood vessels, the grayscale transformation and a multi-scale Hessian enhancement are performed. Finally, the Hessian enhanced image is smoothed by using anisotropic diffusion filtering, and the final segmentation results are obtained by a two-stage region growing algorithm. The experimental results on four datasets show that the proposed segmentation algorithm can segment the plump and smooth vascular tree more effectively than the region growing and the fully convolutional network (FCN) algorithms, and its highest sensitivity and Jaccard index are 84.9% and 70.8%, respectively.
Key words: artificial intelligence; vessel tree segmentation; morphological operation; Hessian matrix; anisotropic diffusion filtering; regional growth
发表期数: 2018年5月第10期
引用格式: 刘勇清,房斌,王翊,等. 基于统计形态学和Hessian矩阵的血管树分割算法[J]. 中国科技论文在线精品论文,2018,11(10):973-982.
 
0 评论数 0
暂无评论
友情链接