一种心肌延迟强化磁共振图像的半自动分割算法

发布时间：2018-05-16 16:37

本文选题：心肌延迟强化磁共振 + 水平集算法　；参考：《西安交通大学学报》2017年08期

【摘要】：针对心肌延迟强化磁共振图像分割中边界模糊的问题,提出了一种基于水平集的半自动分割算法。该算法通过结合多相水平集算法和Chan-Vese模型,可以同时分割心肌内膜、外膜和心肌梗塞区域的边界。根据心脏左心室的圆形特征,通过增加圆形惩罚项,使得分割心肌内膜的曲线在迭代过程中逐渐接近于圆形,从而解决了心肌内膜边界模糊的问题。根据心肌内膜和外膜之间的距离关系,提出了动态气球力,通过将两条演化曲线之间的距离约束在一定范围内,使得曲线在真实边界附近收敛,从而提高了分割心肌外膜边界的精度。由于正常心肌与发生梗塞的心肌在灰度上存在着明显的差异,采用Chan-Vese模型代替阈值法分割心肌梗塞区域以获取光滑的目标边界。该算法已在STACOM 2012国际竞赛数据集上进行了验证测试,结果表明,算法的平均Dice相似性系数为0.72,较其他算法的平均Dice系数提高了0.07~0.12;算法具有较好的实验可重复性的同时也具有更好的稳定性。
[Abstract]:A semi-automatic segmentation algorithm based on level set is proposed to solve the problem of boundary blur in myocardial delayed enhanced magnetic resonance image segmentation. By combining the multiphase level set algorithm and the Chan-Vese model, the boundary of myocardial endocardium, adventitia and myocardial infarction region can be divided simultaneously. According to the circular characteristics of the left ventricle, the curve of myocardial endocardium segmentation is gradually close to the circle in the iterative process by increasing the circular penalty term, thus solving the problem of blurring the boundary of myocardial endocardium. According to the distance relationship between endomyocardium and adventitia, the dynamic balloon force is proposed. By limiting the distance between the two evolution curves in a certain range, the curve converges near the real boundary. Thus, the accuracy of dividing the boundary of myocardial adventitia is improved. Due to the obvious difference between the normal myocardium and the infarcted myocardium, the Chan-Vese model is used to segment the infarct area instead of the threshold method to obtain the smooth target boundary. The algorithm has been tested on the STACOM 2012 international competition data set. The results show that, The average Dice similarity coefficient of the algorithm is 0.72, which is 0.070.12 higher than the average Dice coefficient of other algorithms. The algorithm has better repeatability and stability.
【作者单位】：西安交通大学电子与信息工程学院;
【基金】：国家自然科学基金资助项目(61672423)
【分类号】：R542.22;TP391.41

【相似文献】