左心室核磁共振图像的自动分割

Auto Segmentation of the Left Ventricle MR Images

目前左心室核磁共振图像的分割方法,大部分是半自动的,如Snake方法;为了能实现全自动分割,该文先采用SVM对图像进行左心室定位,然后用水平集(LevelSet)方法进行分割.针对水平集符号距离函数构造计算量大的问题,提出了一种新的符号距离函数(SDF)的生成方法———中线延拓方法.它只需对图像进行一次扫描就可以生成SDF,同时还可以记下每点对应的曲线上的最近邻点,为速度项中曲率的扩展提供条件.针对核磁共振图像成像特点,特别是对加标记线的左心室核磁共振图像,引入了块像素变差和灰度相似性的思想,对水平集方法的速度项进行了改进,提高了分割精度.该方法能全自动、快速、准确地实现左心室的分割.文中给出了合成图像和左心室核磁共振图像的分割结果.

Currently the segmentation algorithms of the left ventricle MR (Magnetic Resonance) images are mostly semi-auto, such as Snake method. To realize the auto segmentation, in this paper the authors adopt SVM (Support Vector Machine) to localize the left ventricle of images, and then carry out the segmentation with level set method. As to the problem that the construction of a signed distance function (SDF) is computationally expensive, a new generation method of the SDF, namely midline extending method is introduced. The SDF can be generated simply by scanning the image for a time. In the meantime it can take down a nearest point on an initial curve for each point, which supports for extending the curvature of the speed term. Focusing on the imaging characteristics of MR images, especially on that of the left ventricle MR images with tag lines, the authors introduce the method of block-pixel variation and intensity comparability to improve the speed term of level set, so as to increase the segmentation precision. This method can automatically, quickly and accurately achieve the goal of segmentation of the left ventricle. In this paper, the segmentation results of synthetic images and the left ventricle MR images are offered.