基于新型深度全卷积网络的肝脏CT影像三维区域自动分割

doi:10.3969/j.issn.0258-8021.2018.04.001

摘要
图/表
参考文献(21)
相关文章 (15)

全文: PDF (14147 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要肝脏分割对于肝肿瘤肝段切除及肝移植体积测量具有重要的临床价值。由于在CT影像中肝脏与邻近脏器的灰度值相似性很高,因此对肝脏区域的三维自动分割是一项具有挑战性的难题。为解决精准肝脏分割的问题,提出一种新型的深度全卷积网络结构3DUnet-C2。该结构充分利用肝脏CT图像的三维空间信息,并有效结合肝脏区域的浅层特征和深层特征。特别地,还提出一种新的3DUnet-C2网络训练策略,通过选取清晰图像,并从图像中截取肝脏区域作为样本进行训练的方式,得到初步3DUnet-C2模型权重,并使用该权重来初始化3DUnet-C2的网络参数,从而使网络达到收敛。最后,针对3DUnet-C2网络分割肝脏边界不精准的问题,在原有3DUnet-C2网络模型的基础上,运用三维条件随机场构建3DUnet-C2-CRF模型来优化肝脏分割边界。为了验证所提出三维分割模型的性能,从ISBI2017 Liver Tumor Segmentation Challenge的数据集中选取100张CT图像用于训练、验证和测试,3DUnet-C2-CRF模型在随机选取的20张测试集上的分割准确率的Dice系数为96.9%,高于3DUnet和Vnet模型的Dice系数。实验结果表明,3DUnet-C2-CRF模型具有更好的特征表达能力以及更强的泛化性能,从而可提升模型的分割准确率。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	孙明建
	徐军
	马伟
	张玉东

关键词 ：肝脏分割, 深度全卷积网络, 条件随机场

Abstract：Liver segmentation has important clinical value in liver tumor resection and liver transplantation volume measurement. Because the intensity value of liver and adjacent organs is very close in CT images, the three-dimensional (3D) automated segmentation of the liver region is a challenged task. In order to make the accurate segmentation of liver region, a new deep fully convolutional network (FCN) structure 3DUnet-C2 was proposed. This network made full use of the three-dimensional spatial information of CT image, and combined well the characteristics of shallow and deep layers. In particular, a new network training strategy was proposed. The primary model was obtained by selecting the clear image and intercepting the liver region as a sample. Then the model was leveraged to initialize the network parameters so that the network can converge. Finally, on the basis of the original model, the 3DUnet-C2-CRF model was constructed by using the three-dimensional conditional random field to optimize the liver segmentation boundary. In order to verify the performance of the proposed 3DUnet-C2-CRF on 3D segmentation of liver regions, 100 CT images were chosen from the data set of the ISBI2017 Liver Tumor Segmentation Challenge. The Dice coefficient of the segmentation accuracy of the 3DUnet-C2-CRF model on 20 test images reached 96.9%, which is higher than the Dice coefficient of 3DUnet and Vnet models. Experimental results showed that the 3DUnet-C2-CRF model had better feature expression capability and more generalization performance, which improved the segmentation accuracy of the model.

Key words： liver segmentation deep fully convolutional networks conditional random field

收稿日期: 2017-09-05

PACS:

R318

基金资助:国家自然科学基金(61771249);江苏省“六大人才高峰”高层次人才项目(2013-XXRJ-019);江苏省自然科学基金(BK20141482)

通讯作者: E-mail: xujung@gmail.com

引用本文:

孙明建, 徐军, 马伟, 张玉东. 基于新型深度全卷积网络的肝脏CT影像三维区域自动分割[J]. 中国生物医学工程学报, 2018, 37(4): 385-393.
Sun Mingjian, Xu Jun, Ma Wei, Zhang Yudong. A New Fully Convolutional Network for 3D Liver Region Segmentation on CT Images. Chinese Journal of Biomedical Engineering, 2018, 37(4): 385-393.

链接本文:

http://cjbme.csbme.org/CN/10.3969/j.issn.0258-8021.2018.04.001 或 http://cjbme.csbme.org/CN/Y2018/V37/I4/385

[1] Alshaikhli SDS, Yang MY, Rosenhahn B, et al. Automatic 3D liver segmentation using sparse representation of global and local image information via level set formulation[EB/OL]. https://arxiv.org/abs/1508.01521, 2015-10-04/2017-09-05.
[2] Heimann T, Meinzer H, Wolf I, et al. A Statistical deformable model for the segmentation of liver ct volumes using extended training data[C]// 3D Segmentation in the Clinic: A Grand Challenge. Canberra: CSIRO, 2007: 161-166.
[3] Kainmüller D, Lange T, Lamecker H. Shape constrained automatic segmentation of the liver based on a heuristic intensity model[C]// 3D Segmentation in the Clinic: A Grand Challenge. Canberra: CSIRO, 2007: 109-116.
[4] Long J, Shelhamer E, Darrell T, et al. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 39(4):640-651.
[5] Li Guodong, Chen Xinjian, Shi Fei, et al. Automatic liver segmentation based on shape constraints and deformable graph cut in CT images[J]. IEEE Transactions on Image Processing, 2015, 24(12): 5315-5329.
[6] Li Changyang, Wang Xiuying, Eberl S, et al. A likelihood and local constraint level set model for liver tumor segmentation from ct volumes[J]. IEEE Transactions on Biomedical Engineering, 2013, 60(10): 2967-2977.
[7] Linguraru MG, Richbourg WJ, Liu J, et al. Tumor burden analysis on computed tomography by automated liver and tumor segmentation[J]. IEEE Transactions on Medical Imaging, 2012, 31(10): 1965-1976.
[8] Heimann T, Van GB, Styner MA, et al. Comparison and evaluation of methods for liver segmentation from ct datasets[J]. IEEE Transactions on Medical Imaging, 2009, 28(8):1251-1265.
[9] Christ PF, Elshaer MEA, Ettlinger F, et al. Automatic liver and lesion segmentation in ct using cascaded fully convolutional neural networks and 3d conditional random fields[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2016: 415-423.
[10] Dou Qi, Chen Hao, Jin Yueming, et al. 3D deeply supervised network for automatic liver segmentation from ct volumes[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2016: 149-157.
[11] Krähenbühl P, Koltun V. Efficient inference in fully connected crfs with gaussian edge potentials[C]//Advances in Neural Information Processing Systems. New York: NIPS, 2011: 109-117.
[12] Glorot X, Bordes A, Bengio Y, et al. Deep sparse rectifier neural networks[J]. Journal of Machine Learning Research, 2011, 15: 315-323.
[13] Zeiler M D. ADADELTA: an adaptive learning rate method[EB/OL]. https://arxiv.org/abs/1212.5701, 2012-12-22/2017-09-05.
[14] Heimann T, Van GB, Styner MA, et al. Comparison and evaluation of methods for liver segmentation from ct datasets[J]. IEEE Transactions on Medical Imaging, 2009, 28(8):1251-1265.
[15] ÇiçekÖ, Abdulkadir A, Lienkamp SS, et al. 3D U-Net: learning dense volumetric segmentation from sparse annotation[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2016: 424-432.
[16] Ioffe S, Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//International Conference on International Conference on Machine Learning. Cham: Springer, 2015:448-456.
[17] Fedorov A, Beichel R, Kalpathy-Cramer J, et al. 3D slicer as an image computing platform for the quantitative imaging network[J]. Magnetic Resonance Imaging, 2012, 30(9):1323-1341.
[18] Ronneberger O, Fischer P, Brox T, et al. U-net: convolutional networks for biomedical image segmentation[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2015:234-241.
[19] Kleesiek J, Urban G, Hubert A, et al. Deep mri brain extraction: a 3d convolutional neural network for skull stripping[J]. NeuroImage, 2016, 129: 460-469.
[20] Milletari F, Ahmadi SA, Kroll C, et al. Hough-cnn: deep learning for segmentation of deep brain regions in MRI and ultrasound[J]. Computer Vision and Image Understanding, 2017, 164: 92-102.
[21] Milletari F, Navab N, Ahmadi SA, et al. V-net: fully convolutional neural networks for volumetric medical image segmentation[C]// 2016 Fourth International Conference on 3D Vision. California: IEEE, 2016: 565-571.