Segmentation of Breast Cancer on DCE-MRI Images With MRF Energy and Fuzzy Speed Function
-
摘要: 乳腺癌灶的精确分割是乳腺癌计算机辅助诊断的重要前提. 在动态对比增强核磁共振成像(Dynamic contrast-enhanced magnetic resonance imaging, DCE-MRI)的图像中, 乳腺癌灶具有对比度低、边界模糊及亮度不均匀等特点, 传统的活动轮廓模型方法很难取得准确的分割结果. 本文提出一种结合马尔科夫随机场(Markov random field, MRF)能量和模糊速度函数的活动轮廓模型的半自动分割方法来完成乳腺癌灶的分割, 相对于专业医生的手动分割, 本文方法具有速度快、可重复性高和分割结果相对客观等优点. 首先, 计算乳腺DCE-MRI图像的MRF能量, 以增强目标区域与周围背景的差异. 其次, 在能量图中计算每个像素点的后验概率, 建立基于后验概率驱动的活动轮廓模型区域项. 最后, 结合Gabor纹理特征、DCE-MRI时域特征和灰度特征构建模糊速度函数, 将其引入到活动轮廓模型中作为边缘检测项. 在乳腺癌灶边界处, 该速度函数趋向于零, 活动轮廓曲线停止演变, 完成对乳腺癌灶的分割. 实验结果表明, 所提出的方法有助于乳腺癌灶在DCE-MRI图像中的准确分割.
-
关键词:
- 乳腺癌 /
- 动态对比增强核磁共振成像 /
- 马尔科夫随机场能量 /
- 活动轮廓模型 /
- 模糊聚类
Abstract: Accurate segmentation of breast cancer is an important step for computer-aided diagnosis. In images obtained from dynamic contrast-enhanced magnetic resonance imaging technique, the traditional active contour model method is difficult to obtain accurate segmentation results due to the low contrast, blurred boundary and intensity inhomogeneous of the breast cancer images. In this paper, a semi-automatic segmentation of active contour model combining Markov random field (MRF) energy and fuzzy velocity function is proposed to perform the segmentation of breast cancer in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) images. This method has the advantages of fast speed, objectivity and the ability to reproduce the segmentation result compared to the manual segmentation of a professional doctor. First, the MRF energy of DCE-MRI is calculated to enhance the difference between the target area and the background. Second, the posterior probability of each pixel is calculated in the energy map. Then, region term of active contour model based on the posterior probability is developed. Finally, a fuzzy speed function, which derived by combining the image intensity, time domain characteristics of DCE-MRI and the Gabor texture feature, is introduced into the active contour model as edge function. At the boundary of breast cancer, the edge function approaches zero and the evolution of the contour curve will stop. The experimental results showed that the proposed segmentation method can accurately segment breast cancer in the images of DCE-MRI. -
图 2 乳腺DCE-MRI图像的MRF能量和后验概率
Fig. 2 MRF energy and posterior probability of breast DCE-MRI image
图 6 肿块形乳腺癌灶分割结果((a)医生手动分割; (b)结合二值水平集和形态学的分割方法; (c)基于边界的ACM; (d)基于局部高斯能量的ACM; (e)基于区域的ACM; (f)基于MRF的区域ACM; (g) U-net; (h) V-net; (i)本文方法)
Fig. 6 Tumor-shaped breast lesion segmentation results ((a) Expert manual segmentation; (b) Segmentation method based on binary level set and morphological operation; (c) Edge-based ACM; (d) ACM based on local Gaussian energy; (e) Region-based ACM; (f) Regional ACM based on MRF; (g) U-net; (h) V-net; (i) The method in this paper)
图 7 非肿块形乳腺癌灶分割结果((a)医生手动分割; (b)结合二值水平集和形态学的分割方法; (c)基于边界的ACM; (d)基于局部高斯能量的ACM; (e)基于区域的ACM; (f)基于MRF的区域ACM; (g) U-net; (h) V-net; (i)本文方法)
Fig. 7 Non tumor-shaped breast lesion segmentation results ((a) Expert manual segmentation; (b) Segmentation method based on binary level set and morphological operation; (c) Edge-based ACM; (d) ACM based on local gaussian energy; (e) Region-based ACM; (f) Regional ACM based on MRF; (g) U-net; (h) V-net; (i) The method in this paper)
表 1 V-net和U-net训练集结果
Table 1 Results of V-net and U-net in training sets
分割
算法$R_{\rm{TP}}$
(均值±标准差)$R_{\rm{FP}}$
(均值±标准差)$D_{\rm{S}}$
(均值±标准差)U-net 0.9706±0.1189 0.6673±0.6662 0.6252±0.1456 V-net 0.9408±0.0637 0.0732±0.1287 0.8836±0.0917 
下载: 导出CSV
表 2 50个临床样本分割结果
Table 2 Segmentation results of 50 clinical samples
分割算法 $R_{\rm{TP}}$ (均值±标准差)$R_{\rm{FP}}$ (均值±标准差)$D_{\rm{S}}$ (均值±标准差)结合二值水平集和形态学的分割方法 0.9759±0.0276 0.9100±1.6139 0.6910±0.2537 基于边界的ACM 0.9676±0.0551 0.6270±0.4238 0.5146±0.2272 基于局部高斯能量的ACM 0.9548±0.0818 0.8368±1.0217 0.6147±0.2040 基于区域的ACM 0.9892±0.0135 1.0604±1.0193 0.5857±0.2357 基于MRF的区域ACM 0.8709±0.1178 0.2018±0.3271 0.7524±0.1438 U-net 0.9389±0.1156 0.5988±0.3772 0.6143±0.1456 V-net 0.8762±0.1408 0.1131±0.0961 0.7920±0.1402 本文方法 0.9297±0.0413 0.0337±0.0222 0.8996±0.0410
下载: 导出CSV
表 3 算法的时间复杂度
Table 3 Time complexity of the algorithm
分割算法 时间复杂度 平均执行时间 (s) 结合二值水平集和形态学的分割方法 ${\rm O}(n)$ 3.3500 基于边界的ACM ${\rm O}(n)$ 3.2550 基于局部高斯能量的ACM ${\rm O}(n)$ 3.6790 基于区域的ACM ${\rm O}(n)$ 1.1188 基于MRF的区域ACM ${\rm O}(n^2)$ 2.7925 U-net ${\rm O}(n)$ 1.6102 V-net ${\rm O}(n)$ 3.3669 本文方法 ${\rm O}(n^2)$ 4.6575
下载: 导出CSV
-
[1] Fitzmaurice C, Akinyemiju T F, Al Lami F H, Alam T, Alizadeh-Navaei R, Allen C, et al. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life-years for 29 cancer groups, 1990 to 2016: A systematic analysis for the global burden of disease study. JAMA Oncology, 2018, 4(11): 1553−1568 doi: 10.1001/jamaoncol.2018.2706 [2] Uematsu T, Kasami M, Watanabe J. Is evaluation of the presence of prepectoral edema on T2-weighted with fat-suppression 3 T breast MRI a simple and readily available noninvasive technique for estimation of prognosis in patients with breast cancer? Breast Cancer, 2014, 21(6): 684−692 [3] Cheon H, Kim H J, Lee S M, Cho S H, Shin K M, Kim G C, et al. Preoperative MRI features associated with lymphovascular invasion in node-negative invasive breast cancer: A propensity-matched analysis. Journal of Magnetic Resonance Imaging, 2017, 46(4): 1037−1044 doi: 10.1002/jmri.25710 [4] Fasihi M S, Mikhael W B. Overview of current biomedical image segmentation methods. In: Proceedings of the 2016 International Conference on Computational Science and Computational Intelligence. Las Vegas, USA: IEEE, 2016. 803−808 [5] van Dijk N P, Maute K, Langelaar M, van Keulen F. Level-set methods for structural topology optimization: A review. Structural and Multidisciplinary Optimization, 2013, 48(3): 437−472 doi: 10.1007/s00158-013-0912-y [6] 马超, 刘亚淑, 骆功宁, 王宽全. 基于级联随机森林与活动轮廓的3D MR图像分割. 自动化学报, 2019, 45(5): 1004−1014Ma Chao, Liu Ya-Shu, Luo Gong-Ning, Wang Kuan-Quan. Combining concatenated random forests and active contour for the 3D MR images segmentation. Acta Automatica Sinica, 2019, 45(5): 1004−1014 [7] 孙文燕, 董恩清, 曹祝楼, 郑强. 一种基于模糊主动轮廓的鲁棒局部分割方法. 自动化学报, 2017, 43(4): 611−621Sun Wen-Yan, Dong En-Qing, Cao Zhu-Lou, Zheng Qiang. A robust local segmentation method based on fuzzy-energy based active contour. Acta Automatica Sinica, 2017, 43(4): 611−621 [8] Zhang K H, Zhang L, Lam K M, Zhang D. A level set approach to image segmentation with intensity inhomogeneity. IEEE Transactions on Cybernetics, 2016, 46(2): 546−557 doi: 10.1109/TCYB.2015.2409119 [9] Rouhi R, Jafari M. Classification of benign and malignant breast tumors based on hybrid level set segmentation. Expert Systems with Applications, 2016, 46: 45−59 doi: 10.1016/j.eswa.2015.10.011 [10] 张旭梅, 范虹, 乔柱. 融合全局和局部信息的水平集乳腺MR图像分割. 计算机应用研究, 2015, 32(1): 307−311 doi: 10.3969/j.issn.1001-3695.2015.01.072Zhang Xu-Mei, Fan Hong, Qiao Zhu. Level set method combined global and local information and its application to breast MRI. Application Research of Computers, 2015, 32(1): 307−311 doi: 10.3969/j.issn.1001-3695.2015.01.072 [11] 虞红伟, 厉力华, 徐伟栋, 刘伟, 张娟, 邵国良. 一种基于水平集的多尺度乳腺肿块分割方法. 仪器仪表学报, 2010, 31(6): 1418−1423Yu Hong-Wei, Li Li-Hua, Xu Wei-Dong, Liu Wei, Zhang Juan, Shao Guo-Liang. Multi-scale approach for breast mass segmentation using level set method. Chinese Journal of Scientific Instrument, 2010, 31(6): 1418−1423 [12] 范虹, 朱艳春, 王芳梅, 张旭梅. 多分辨率水平集算法的乳腺MR图像分割. 物理学报, 2014, 63(11): Article No.118701Fan Hong, Zhu Yan-Chun, Wang Fang-Mei, Zhang Xu-Mei. Segmentation of breast MR images based on multiresolution level set algorithm. Acta Physica Sinica, 2014, 63(11): Article No.118701 [13] Kuo H C, Giger M L, Reiser I, Boone J M, Lindfors K K, Yang K, et al. Level set segmentation of breast masses in contrast-enhanced dedicated breast CT and evaluation of stopping criteria. Journal of Digital Imaging, 2014, 27(2): 237−247 doi: 10.1007/s10278-013-9652-1 [14] Rodtook A, Makhanov S S. Multi-feature gradient vector flow snakes for adaptive segmentation of the ultrasound images of breast cancer. Journal of Visual Communication and Image Representation, 2013, 24(8): 1414−1430 doi: 10.1016/j.jvcir.2013.09.009 [15] Xie Q Y, Pan X, Zhu H Q. A multiphase level set clustering approach using MRF-based Student's-t mixture model. In: Proceedings of the 8th International Conference on Wireless Communications and Signal Processing. Yangzhou, China: IEEE, 2016. 1−5 [16] Yang X, Gao X B, Tao D C, Li X L, Li J. An efficient MRF embedded level set method for image segmentation. IEEE Transactions on Image Processing, 2015, 24(1): 9−21 doi: 10.1109/TIP.2014.2372615 [17] Kallenberg M, Petersen K, Nielsen M, Ng A Y, Diao P F, Igel C, et al. Unsupervised deep learning applied to breast density segmentation and mammographic risk scoring. IEEE Transactions on Medical Imaging, 2016, 35(5): 1322−1331 doi: 10.1109/TMI.2016.2532122 [18] Rouhi R, Jafari M, Kasaei S, Keshavarzian P. Benign and malignant breast tumors classification based on region growing and CNN segmentation. Expert Systems with Applications, 2015, 42(3): 990−1002 doi: 10.1016/j.eswa.2014.09.020 [19] Ashraf A B, Gavenonis S, Daye D, Mies C, Feldman M, Rosen M, et al. A multichannel Markov random field approach for automated segmentation of breast cancer tumor in DCE-MRI data using kinetic observation model. In: Proceedings of the 14th International Conference on Medical Image Computing and Computer-Assisted Intervention. Toronto, Canada: Springer, 2011. 546−553 [20] Achuthan A, Rajeswari M, Ramachandram D, Aziz M E, Shuaib I L. Wavelet energy-guided level set-based active contour: A segmentation method to segment highly similar regions. Computers in Biology and Medicine, 2010, 40(7): 608−620 doi: 10.1016/j.compbiomed.2010.04.005 [21] Shahvaran Z, Kazemi K, Helfroush M S, Jafarian N, Noorizadeh N. Variational level set combined with Markov random field modeling for simultaneous intensity non-uniformity correction and segmentation of MR images. Journal of Neuroscience Methods, 2012, 209(2): 280−289 doi: 10.1016/j.jneumeth.2012.06.012 [22] 宋艳涛, 纪则轩, 孙权森. 基于图像片马尔科夫随机场的脑MR图像分割算法. 自动化学报, 2014, 40(8): 1754−1763Song Yan-Tao, Ji Ze-Xuan, Sun Quan-Sen. Brain MR image segmentation algorithm based on Markov random field with image patch. Acta Automatica Sinica, 2014, 40(8): 1754−1763 [23] Li B, Chen K, Peng G M, Guo Y X, Tian L F, Ou S X, et al. Segmentation of ground glass opacity pulmonary nodules using an integrated active contour model with wavelet energy-based adaptive local energy and posterior probability-based speed function. Materials Express, 2016, 6(4): 317−327 doi: 10.1166/mex.2016.1311 [24] 张喆, 韩德强, 杨艺. 基于证据马尔可夫随机场模型的图像分割. 控制与决策, 2017, 32(9): 1607−1613Zhang Zhe, Han De-Qiang, Yang Yi. Image segmentation based on evidential Markov random field model. Control and Decision, 2017, 32(9): 1607−1613 [25] Li C M, Kao C Y, Gore J C, Ding Z H. Minimization of region-scalable fitting energy for image segmentation. IEEE Transactions on Image Processing, 2008, 17(10): 1940−1949 doi: 10.1109/TIP.2008.2002304 [26] Khan S, Hussain M, Aboalsamh H, Mathkour H, Bebis G, Zakariah M. Optimized Gabor features for mass classification in mammography. Applied Soft Computing, 2016, 44: 267−280 doi: 10.1016/j.asoc.2016.04.012 [27] Kwedlo W. A new random approach for initialization of the multiple restart EM algorithm for Gaussian model-based clustering. Pattern Analysis and Applications, 2015, 18(4): 757−770 doi: 10.1007/s10044-014-0441-3 [28] 郑强, 董恩清. 一种新的基于二值水平集和形态学的局部分割方法. 电子与信息学报, 2012, 34(2): 375−381Zheng Qiang, Dong En-Qing. A new local segmentation method based on binary level set and morphological operation. Journal of Electronics and Information Technology, 2012, 34(2): 375−381 [29] Xu J, Monaco J P, Madabhushi A. Markov random field driven region-based active contour model (MaRACel): Application to medical image segmentation. In: Proceedings of the 13th International Conference on Medical Image Computing and Computer-Assisted Intervention. Beijing, China: Springer, 2010. 197−204 -
图(7) / 表(3)
计量
- 文章访问数: 1498
- HTML全文浏览量: 282
- PDF下载量: 176
- 被引次数: 0
