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.%A semiautomatic segmentation method based on level set is proposed to solve the weak boundaries problem in segmentation of myocardial DE-MRI (delayed enhancement magnetic resonance Image).This method uses multiphase level sets and the Chan-Vese model to simultaneously extract the boundaries of endocardium,epicardium and infarction.It relies on the circular shape of the cardiac left ventricle that a circular penalized term is introduced to keep the shape of the evolving contour close to a circle,and thus the difficulty caused by the weak boundaries between the endocardium and trabeculation is overcome.A dynamic balloon force is introduced according to the distance between the myocardial borders to preserve the distance between the two evolving contours within a given range,therefore the segmentation accuracy of the epicardium is improved.Since the gray level of the normal myocardium is significantly different from that of the infarcted myocardium,the Chan-Vese model is used to extract the infarction region of myocardium and to obtain smooth object boundaries.The proposed method is validated on the data set from STACOM 2012 challenge.The average Dice similarity coefficient of the method is 0.72,and is 0.07-0.12 bigger than those of other methods.Experimental results also show that the proposed method obtains better reproducibility,and is more robust than those compared methods.
展开▼
