Abstract:

The left ventricular MR image segmentation algorithm based on the Distance Regular Level Set model（DRLSE） has a strong dependence on the gradient information, and is easy to fall into the local optimum in the weak edge region of the image, and is sensitive to the selection of the initial contour. In order to reduce the sensitivity of the algorithm to the initial contour and improve its ability to segment the weak edge of the left ventricle image, this paper proposes a left ventricular segmentation algorithm suitable for weak edge information. Based on DRLSE, the segmentation algorithm proposes a new local term based on the proposed coefficient of variation based segmentation model（PSM）. The algorithm relies on gradient and image local information to drive the curve evolution, which reduces the sensitivity of DRLSE to the initial contour. In order to overcome the situation where the DRLSE algorithm causes boundary leakage when dividing the weak boundary of the epicardium, shape constraint is introduced. In order to verify the accuracy of the algorithm segmentation proposed in this paper, based on the database provided by the Imaging Department of the Children’s Hospital of Toronto City, the endocardium is segmented by DRLSE, the Convexity Preserving Level Set model（CPLSE）, U-Net network and the endocardium algorithm of this paper; the epicardium is segmented by DRLSE, introduction of epicardium shape binding DRSLE（DRLSE-shape）, U-Net network and the epicardium algorithm of this paper. The experimental results show that the proposed algorithm is superior to the above algorithm for the left ventricular and adventitia, which can reduce the sensitivity of DRLSE to the initial contour and improve the accuracy of MR image segmentation of the left ventricular weak boundary.

Key words: left ventricle, image segmentation, level set, weak boundary

摘要：

基于距离正则水平集模型（DRLSE）的左心室MR图像分割算法对梯度信息有很强的依赖性，在图像弱边缘区域容易陷入局部最优，且对初始轮廓的选取敏感。为降低算法对初始轮廓的敏感程度，提高其在左心室图像弱边缘的分割能力，提出一种适用于弱边缘信息的左心室分割算法。在DRLSE的基础上，该分割算法提出运用拟合方法计算基于变异系数分割模型（PSM）的新局部项，算法依靠梯度与图像局部信息驱动曲线演化，降低了DRLSE对初始轮廓的敏感度；引入形状约束力，克服DRLSE算法在左心室外膜弱边界处出现边界泄露的情况。为验证所提算法分割的准确性，基于多伦多市患病儿童医院影像科提供的数据库，利用DRLSE、保持凸性水平集模型（CPLSE）模型、U-Net网络以及提出的内膜算法对心内膜进行分割；利用DRLSE、引入外膜形状约束力的DRLSE模型（DRLSE-shape）、U-Net网络以及提出的外膜算法对心外膜进行分割。实验结果表明，针对左心室内、外膜，所提算法优于上述算法，能降低DRLSE对初始轮廓的敏感程度，提升对左心室弱边界MR图像分割的精确度。

关键词: 左心室, 图像分割, 水平集, 弱边界