基于空间FCM与MRF方法的乳腺MRI序列三维病灶分割研究  

3D Segmentation of Breast DCE-MRI Sequence Using Spatial FCM-MRF Method

作　　者：张承杰[1] 厉力华[1] 

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,杭州310018

出　　处：《中国生物医学工程学报》2014年第2期202-211,共10页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(61271063);国家重点基础研究发展计划资助(2013CB329502);国家杰出青年科学基金(60788101)

摘　　要：针对乳腺DCE-MRI病灶分割,提出一种空间FCM聚类与MRF随机场相结合的三维分割方法。首先,对MRI图像进行空间FCM粗分割,提取病灶粗轮廓。然后,在其基础上进行MRF精分割,并结合病灶三维信息:用相邻切片分割结果对应标号矩阵初始化MRF精分割标号场,同时用该张切片粗分割所得隶属度矩阵对MRF精分割进行参数自适应调整。用该方法与空间FCM、水平集、模糊MRF方法对50例MRI数据进行分割对比实验,得到良、恶性病灶分割重叠率分别为76.4%、75.5%;相比于空间FCM的68.7%、69.5%,水平集的70.8%、72.6%以及模糊MRF的72.9%、73.6%有明显提升。对所有175例MRI数据分割结果进行非监督评价,得到良、恶性病灶区域均匀性均大于0.92;区域内差异性良性病灶92%小于150、恶性病灶98%小于150;区域间差异性良性病灶87%大于0.25、恶性病灶90%大于0.3。综上表明,该方法具有较高的分割精度。Breast MRI image segmentation is a challenging issue. This paper presents a 3D segmentation method, which is based on spatial FCM clustering and Markov random field. First, the MRI image was coarse segmented by spatial FCM to extract lesion contours. And then MRF segmentation was conducted to refine the result. We combined the 3D information of lesion in the MRF process by using segmentation result of contiguous slices to constraint the slice segmentation. At the same time, a membership matrix of FCM segmentation result is used for adaptive adjustment of Markov parameters in MRF segmentation process. The segmentation performance of this method was compared with that of spatial FCM, level set and fuzzy MRF on a database including 50 breast DCE-MRI examinations. Results show that average overlap rate for benign and malignant of our method is 76.4% and 75.5% respectively, compared with 68.7% and 69.5% for spatial FCM, 70.8% and 72.6% for level set method, and 72.9% and 73.6% for fuzzy MRF. It is demonstrated that our method has a better performance in accuracy. In addition, we used unsupervised evaluation method to evaluate the segmentation result of all the 175 breast DCE-MRI image sequences in the database, The uniformities of intra region （URs） for both benign and malignant lesion was more than 0.92. The differences within region （DRs） of 92% of the benign lesions and 98% of the malignant lesions were less than 150, the differences of the inter region （DIR） of 87% of benign lesion were more than 0. 25, while those of 90% of malignant lesion were over O. 3. The results demonstrated the proposed method has a good performance in segmentation accuracy.

关 键 词：乳腺DCE-MRI图像 病灶分割 FCM-MRF 三维分割 参数自适应 

分 类 号：R318[医药卫生—生物医学工程]