北京生物医学工程

基于生成对抗网络的肝脏CT图像分割

Liver CT image segmentation based on generative adversarial network

作者：邓鸿邓雅心丁廷波严中红王富平陈忠敏

单位：重庆理工大学药学与生物工程学院(重庆 400054) <p>通信作者:陈忠敏，教授。E-mail:chenzhongmin@cqut.edu.cn</p> <p> </p>

分类号：R318.04

出版年·卷·期（页码）：2021·40·4（367-376）

摘要：

目的从腹部计算机断层扫描(computed tomography, CT)图像中分割出肝脏区域对于肝脏疾病早期诊断、肝脏大小估计以及3D重建十分重要，精准快速地分割出肝脏边缘成为研究要点。方法采用公开发表的肝脏肿瘤数据集为研究对象，融合生成对抗网络和UNET网络对CT图像实现肝脏的自动分割。首先将腹部CT图像输入到UNET网络进行分割预测，然后通过生成对抗网络(generative adversarial networks, GAN)进行对抗训练，使得预测结果更加接近于真实结果，同时在进行对抗训练的过程中探索了不同的距离约束函数对于分割结果的影响；预测的分割结果通过Dice分数(dice similarity coefficient，Dice)、IoU分数(intersection over union, IoU)、像素精确度(pixel accuracy，PA)、相对体积误差(relative volume difference，RVD)以及相对表面积误差(relative surface area error，RSSD)在CT-核磁健康腹部器官分割挑战数据集[Combined (CT-MR) Healthy Abdominal Organ Segmentation Challenge Data, CHAOS]数据集上进行评价。结果 L2距离约束的GAN-UNET网络可以很好的对肝脏进行分割，其Dice、IoU和PA分别达到了94.9%、91.3%、99.4%，相比于UNET的Dice、IoU和PA为92.3%、86.7%、95.8%有明确的提升。在三维指标中，本文的方法在RVD、RSSD为0.026、0.079，相比于UNET的0.042、0.191有明显下降。结论通过对UNET网络进行生产对抗训练以及在训练过程中引入距离约束函数可以提高肝脏分割的性能，肝脏分割结果可以应用于计算机辅助诊断系统中。

Objective Segmenting the liver area from computed tomography (CT) images of the abdomen is the first step in early diagnosis of liver disease, liver size estimation and 3D reconstruction, and it is also a very important step. How to accurately and quickly segment the edge of the liver is a problem worth studying. Methods Using publicly published liver tumor datasets as the research object, this paper merges the generative adversarial network and the UNET network to realize the automatic segmentation of the liver on the CT image. First, the abdominal computer tomography CT image is input into the UNET network for segmentation prediction, and then generative adversarial networks conducts adversarial training to make the prediction result closer to the real result. At the same time, the impact of different distance constraint functions on the segmentation result is explored during the adversarial training process; the predicted segmentation result passes the Dice score, IoU score, Pixel Accuracy (PA), relative volume error (RVD), and relative surface area error (RSSD) are evaluated on CHAOS (Combined (CT-MR) Healthy Abdominal Organ Segmentation Challenge Data) data set. Results The L2 distance constrained GAN-UNET network can segment the liver well, and its Dice, IoU and PA reached 94.9%, 91.3%, and 99.4% respectively, compared with UNET's Dice, IoU and PA of 92.3%, There was a certain increase in 86.7% and 95.8%. Among the three-dimensional indicators, the RVD and RSSD of 0.026 and 0.079 in the method of this paper are significantly lower than UNET's 0.042 and 0.191. Conclusions The performance of liver segmentation can be improved by performing production confrontation training on UNET network and introducing distance constraint function in the training process, and the segmentation can be used in computer-aided diagnosis systems.

参考文献：

[1] ?杨昌俊,杨新.基于图割与快速水平集的腹部 CT 图像分割[J].CT 理论与应用研究, 2011, 20(3):291-300.

Yang CJ, Yang X. Abdominal CT image segmen-tation based on graph cuts and fast level set[J].CT Theory and Applications,2011,20(3)：291-300.

[2] Nakayama Y, Li Q, Katsuragawa S, et al. Automated hepatic volumetry for living related liver transplantation at multisection CT[J]. Radiology,2006, 240(3):743–748.

[3] Kallini JR , Gabr A , Salem R , et al. Transarterial radioembolization with yttrium-90 for the? treatment of hepatocellular carcinoma[J]. Advances in Therapy, 2016, 33(5):699-714.

[4] 李成刚.三维重建与虚拟现实技术在肝脏外科的应用[J].世界华人消化杂志,2020,28(13):515-518.

??? Li CG. Application of three-dimensional reconstruction and virtual reality technology in liver surgery[J].World Chinese Journal of Digestology,2020,28(13):515-518.

[5] Berzigotti A, Abraldes J G, Tandon P, et al. Ultrasonographic evaluation of liver surface and transient elastography in clinically doubtful cirrhosis[J]. Journal of Hepatology, 2010, 52(6): 846-853.

[6] 赵于前，闫桂霞，徐效文,等.基于先验信息水平集方法的肝脏CT序列图像自动分割[J].中南大学学报（自然科学版）, 2015, 46(4): 1310-1317.

Zhao YQ, Yan GX, Xu XW, et al. Automatic segmentation of livers from CT series based on level set method with prior knowledge[J]. Journal of Central South University(Science and Technology), 2015, 46(4): 1310-1317.

[7] ?吴健, 崔志明, 叶峰, 等. 基于轮廓形状的CT断层图像插值[J]. 计算机应用与软件, 2008, 25(11): 63-65.

?Wu J, Cui ZM, Ye F, et al.Interpolation based on contour shape for CT faulted images[J].Computer Applications and Software,2008, 25(11):63-65.

[8] ?Lu XQ, Wu JS, Ren XY, et al. The study and application of the improved region growing algorithm for liver segmentation[J]. Optik, 2014, 125(9): 2142-2147.

[9] Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651.

[10] Badrinarayanan V, Kendall A, Cipolla R. Segnet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495.

[11]Badrinarayanan V, Kendall A, Cipolla R. Segnet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495.

[12] Li XM, Chen H, Qi XJ, et al. H-DenseUNet: hybrid densely connected unet for liver and tumor segmentation from CT volumes[J]. IEEE Transactions on Medical Imaging, 2018, 37(12): 2663-2674.

[13] Goodfellow IJ, Pouget-Abadie J, Mirza M , et al. Generative adversarial networks[J]. Advances in Neural Information Processing Systems, 2014, 3: 2672-2680.

[14] Kim HJ, Lee D. Image denoising with conditional generative adversarial networks (CGAN) in low dose chest images[J]. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 2020, 954: 161914.

[15] Zhao D, Zhu D, Lu J, et al. Synthetic medical images using F&BGAN for improved lung nodules classification by multi-scale VGG16[J]. Symmetry, 2018, 10(10): 519.

[16] Zhou JY, Wong DW, Ding F, et al. Liver tumour segmentation using contrast-enhanced multi-detector CT data: performance benchmarking of three semiautomated methods[J]. European Radiology, 2010, 20(7): 1738-1748.

[17] Li Q, Song H, Chen L, et al. An overview of abdominal multi-organ segmentation[J]. Current Bioinformatics, 2020, 15(8): 866-877.

[18] Schindl MJ, Redhead DN, Fearon KC, et al. The value of residual liver volume as a predictor of hepatic dysfunction and infection after liver resection[J]. Gut, 2005, 54(2): 289-296.

[19] 刘博. 基于深度卷积神经网络的肝图像分割方法研究[D].哈尔滨: 哈尔滨工业大学,2019.

??? Liu B. Research on liver image segmentation based on deep convolutional neural network[D]. Harbin: Harbin Institute of Technology, 2019.

[20]李庆勃. 基于深度学习的肝脏肿瘤CT影像分割方法研究[D].西安：长安大学,2019.

Li QB. Research on CT image segmentation of liver tumor based on deep learning[D].Xi’an: Chang’an University, 2019.

[21] 黄泳嘉,史再峰,王仲琦,等.基于混合损失函数的改进型U-Net肝部医学影像分割方法[J].激光与光电子学进展,2020,57(22): 221003.

??? Huang YJ, Shi ZF, Wang ZQ, et al.Improved U-Net based on mixed loss functionfor liver medical image segmentation[J].Laser and Optoelectronics Progress,2020,57(22): 221003.

[22] 邓青松,吴传新,龚建平.3D打印技术在临床教学中的应用和效果[J].医学教育管理,2019,5(6):531-535.

Deng QS, Wu CX, Gong JP. Application and effect of 3D printing technology in clinical teaching[J].Medical Education Management,2019,5(6):531-535.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录