«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

32

期数:

2022年06期

页码:

162-166

栏目:

应用前沿与综合

出版日期:

2022-06-10

文章信息/Info

Title:

An Improved Double Level Set Algorithm for Left Ventricular Segmentation of Cardiac MRI Images

文章编号:

1673-629X(2022)06-0162-05

作者:

赵昊宸; 苑金辉; 朱恩嵘; 乔 艳; 胡晓飞

1. 南京邮电大学 通信与信息工程学院,江苏 南京 210003;

2. 南京邮电大学 地理与生物信息学院,江苏 南京 210003

Author(s):

ZHAO Hao-chen1 ; YUAN Jin-hui1 ; ZHU En-rong2 ; QIAO Yan2 ; HU Xiao-fei2

1. School of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;

2. School of Geography and Bioinformatics,Nanjing University of Posts and Telecommunications,Nanjing 210003,China

关键词:

水平集方法; 距离规则化函数; 后处理算法; 左心室图像分割; 深度学习

Keywords:

level set method; distance regularization function; post-process algorithm; left ventricular image segmentation; deep learning

分类号:

TP301

DOI:

10. 3969 / j. issn. 1673-629X. 2022. 06. 027

摘要:

针对基于深度学习的心脏 MRI 图像左心室分割网络仅使用简单的阈值法对输出的概率图进行二值化,使得分割的结果边缘模糊的问题,提出了一种改进的基于双水平集的后处理算法。 在使用传统水平集方法对左心室内外膜进行分割时,存在水平集函数演化不稳定、边缘分割精度低和分割效率低等问题。 提出新的距离规则化函数对水平集的能量函数进行改进,新的距离规则化函数可以更好地约束边缘正则性,能够很好地弥补深度网络输出结果当中的缺陷,并使用双水平集,即 0 水平集函数和 k 水平集函数分别向左心室内外膜演化。 利用多伦多市儿童病医院影像科提供的数据集,用上述算法对基于深度网络的心脏 MRI 图像左心室分割结果进行后处理,实验结果表明,左心室内膜和外膜分割的 Dice 相似系数分别为 0. 933 8 和 0. 958 7。 对比其他分割模型,分割精度获得了明显提高。

Abstract:

Aiming at the problem that the edge of the ventricular segmentation results are fuzzy only using a simple threshold method tobinarize the output probability map of the left ventricular segmentation network of cardiac MRI images based on deep learning, animproved post-processing algorithm based on two-level sets is proposed. When the traditional level set method is used to segment theleft ventricle endocardium and epicardium,there are problems such as unstable evolution of the level set function,low accuracy of edgesegmentation,and low segmentation efficiency. A new distance regularization function is proposed to improve the energy function of thelevel set. The new distance regularization function can better constrain the edge regularity,and can make up for defects in the output ofthe deep learning network. Simultaneously,we use the two - level set algorithm,that is,the 0 level set function and the k level set,toevolve to the left ventricle endocardium and epicardium respectively. The data set provided by the imaging department of the TorontoChildren’s Hospital is applied in this thesis. The aforementioned algorithm is used to post-process the left ventricular segmentation resultsof cardiac MRI images based on the deep network. The experimental results show that the Dice similarity coefficients of the leftventricular endocardium and epicardium segmentation are 0. 933 8 and 0. 958 7,respectively. Compared with other segmentation models,the segmentation accuracy of this thesis has been significantly improved.

常用功能

更新日期/Last Update: 2022-06-10