形状统计Mumford-Shah模型的MR图像左心室外轮廓分割

Shape Statistics Mumford-Shah Model for Outer Contour Segmentation of the Left Ventricle MR Image

左心室的分割是左心室运动重建的前提,分割的精度直接影响重建的真实性.由于左心室外轮廓存在弱边界,甚至边界的断裂,使得精确分割左心室外轮廓变得相当困难.文章在详细分析左心室外轮廓的基础上,采用了形状统计Mumford-Shah模型的分割方法,同时对原模型做了如下两点改造:(1)用期望最大(EM)算法求得图像中每点属于心肌的后验概率,通过此后验概率构造“伪灰度”图像来替代原灰度图像,以达到目标与背景灰度呈阶跃型分布的目的;(2)用边缘图替代Mumfor-Shah模型中的梯度项,以增强弱边缘,提高分割精度.实验结果证明,此方法可以有效地改进分割精度,适合整个心动周期.

The segmentation of left ventricle is the premise of its motion reconstruction, and the segmentation precision will influence the reconstruction facility. Due to the weak boundary, even the break boundary, on the outer contour of the left ventricle, it is very difficulty to segment the outer contour precisely. Based on the detailed analysis of the outer contour, shape statistics Mumford-Shah model is adopted. At the same time, the authors reconstruct the original model in the following two parts： （1）Using expectation maximum （EM） algorithm to resolve the posterior probability of each pixel belonging to the myocardium, and using it to construct ＂artificial intensity＂ image to substitute original intensity image, in order to make the object and background intensity distributed abruptly; （2） Using edge image to substitute the grads term of the Mumford- Shah model, in order to enhance weak boundary and improve segmentation precision. The experi ment results prove that this method can effectively improve the segmentation precision and is appropriate for the whole heart beating cycle.