北京生物医学工程

基于全卷积神经网络的直肠癌肿瘤磁共振影像自动分割方法

Automatic segmentation method based on full convolution neural network for rectal cancer tumors in magnetic resonance image

作者：冉昭简俊明王蒙蒙赵星羽高欣

单位：中国科学技术大学（合肥 230026）中国科学院苏州生物医学工程技术研究所（苏州215163）

关键词：直肠肿瘤分割；神经网络；多边输出；磁共振影像

分类号：R318.04

出版年·卷·期（页码）：2019·38·5（465-471）

摘要：

目的直肠肿瘤（rectum cancer, RC）的图像精确分割是直肠癌诊断和治疗的基础和关键。目前，直肠肿瘤的分割通常是由放射科医生逐切片进行，这种方式主观性强，工作量大。为此，本文提出了一种直肠肿瘤磁共振影像全自动分割网络，在有效减少放射科医生负担的同时提高了肿瘤分割结果的可重复性。方法首先采用一个预训练的ResNet50提取特征，并在网络隐藏层添加三个边输出模块，实现图像数据的多尺度特征提取，最后融合三个边输出模块获得最终的分割结果。将所提网络架构的分割结果与基于U-net网络架构的分割结果进行比较，并分析不同损失函数和感兴趣区域（region of interest, ROI）尺寸对所提网络分割性能的影响。结果本研究使用中山大学附属第六医院512例患者的影像数据对模型进行训练及测试，其中随机选取的461名患者的T2加权磁共振影像用于网络训练，剩下51名患者的T2加权磁共振影像用于网络测试。结果表明，所提网络分割结果的平均Dice相似性系数（Dice similarity coefficient, DSC）、平均敏感度（sensitivity）、平均特异度（specificity）及平均豪斯多夫距离（Hausdorff distance, HD）分别达到了83.61%、89.10%、96.36%和8.49，均优于基于U-net的分割方法。对于包含了肿瘤组织的ROI，尺寸越小，分割效果越好。对于给定尺寸的ROI，几种损失函数并无太大差异。结论该算法能够准确地勾画肿瘤边界，将有助于提升医生工作效率。

Objective Accurate segmentation of rectal tumors form images is a basic and crucial task for diagnosis and treatment of rectal cancer. Currently, the segmentation of rectal tumors is usually performed by radiologists on a slice-by-slice basis, which is highly subjective and requires a large amount of work. Therefore, this paper proposes an automatic segmentation network for magnetic resonance image of rectal tumors, which can not only effectively reduce the burden of radiologists, but also improved the repeatability of tumor segmentation results. Methods A pre-trained Resnet50 model was introduced for feature extraction, and three side-output modules were added to the hidden layer of Resnet50 to guide multi-scale feature learning. The final boundaries of tumor were determined by the fusion of the predictions from side-output modules. The proposed model was compared with a U-net based model, and the impacts of different region of interest (ROI) sizes and loss functions were also evaluated. Results We trained and evaluated the models on the data of 512 patients from Sixth Affiliated Hospital of Sun Yat-sen University (Guangzhou, China), in which the T2-weighted magnetic resonance images (T2W-MRIs) of 461 patients were randomly selected for model training, while the T2W-MRIs of the remain 51 patients were used for performance evaluation. The proposed model was superior to the U-net based model and achieved an average Dice similarity coefficient of 83.61%, an average sensitivity of 89.10%, an average specificity of 96.36%, and an average Hausdorff distance of 8.49. In addition, when the ROI contained rectal tumor tissue, the smaller the ROI size was, the higher the segmentation accuracy would be. For a certain ROI size, there were no significant differences in segmentation results among these loss functions. Conclusion The proposed network can accurately delineate the tumor boundaries and could help improve physician productivity.

参考文献：

[1] Parkin DM, Pisani P, Ferlay J. Global cancer statistics [J]. CA: A Cancer Journal for Clinicians, 1999, 49(1): 33-64.

[2] 陈万青, 张思维, 曾红梅, 等. 中国2010年恶性肿瘤发病与死亡 [J]. 中国肿瘤, 2014, 23(1): 1-10.

Chen WQ, Zhang SW, Zeng HM, et al. Report of cancer incidence and mortality in China, 2010 [J]. China Cancer, 2014, 23(1): 1-10.

[3] 李国东, 董新舒, 刘明. 直肠癌外科治疗进展 [J]. 中华肿瘤防治杂志, 2015, 22(5): 402-406.

Li GD, Dong XS, Liu M. Progress of surgical treatment in rectal carcinoma [J]. Chinese Journal of Cancer Prevention and Treatment, 2015, 22(5): 402-406.

[4] Dou Q, Yu L, Chen H, et al. 3D deeply supervised network for automated segmentation of volumetric medical images [J]. Medical Image Analysis, 2017, 41：40-54.

[5] Soomro MH, Giunta G, Laghi A, et al. Segmenting MR Images by Level-Set Algorithms for Perspective Colorectal Cancer Diagnosis [C]//Proceedings of the European Congress on Computational Methods in Applied Sciences and Engineering. Berlin ：Springer, 2017: 396-406.

[6] van Heeswijk MM, Lambregts DM, van Griethuysen JJ, et al. Automated and semiautomated segmentation of rectal tumor volumes on diffusion-weighted MRI: can it replace manual volumetry? [J]. International Journal of Radiation Oncology Biology Physics, 2016, 94(4): 824-831.

[7] Irving B, Cifor A, Papie? BW, et al. Automated colorectal tumour segmentation in DCE-MRI using supervoxel neighbourhood contrast characteristics [C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin：Springer, 2014: 609-616.

[8] Trebeschi S, van Griethuysen JJ, Lambregts DM, et al. Deep learning for fully-automated localization and segmentation of rectal cancer on multiparametric MR [J]. Scientific reports, 2017, 7(1): 5301.

[9] Huang YJ, Dou Q, Wang ZX, et al. HL-FCN: Hybrid loss guided FCN for colorectal cancer segmentation [C]//Proceedings of the Biomedical Imaging (ISBI 2018). Wahington DC：2018 IEEE 15th International Symposium on, IEEE, 2018: 195-198.

[10] Wang J, Lu J, Qin G, et al. Technical Note: A deep learning-based autosegmentation of rectal tumors in MR images [J]. Medical physics, 2018, 45(6): 2560-2564.

[11] Jian J, Xiong F, Xia W, et al. Fully convolutional networks (FCNs)-based segmentation method for colorectal tumors on T2-weighted magnetic resonance images [J]. Australasian Physical & Engineering Sciences in Medicine, 2018, 41(2): 393-401.

[12] Cheng R, Lay N, Mertan F, et al. Deep learning with orthogonal volumetric HED segmentation and 3D surface reconstruction model of prostate MRI [C]//Proceedings of the Biomedical Imaging (ISBI 2017). Wahington DC：2017 IEEE 14th International Symposium on, IEEE, 2017: 749-753.

[13] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition [C]//Proceedings of the Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 770-778..

[14] Du X, El-Khamy M, Lee J, et al. Fused DNN: A deep neural network fusion approach to fast and robust pedestrian detection [C]//Proceedings of the Applications of Computer Vision (WACV), 2017 IEEE Winter Conference on, IEEE, 2017: 953-961

[15] Peng C, Zhang X, Yu G, et al. Large kernel matters—improve semantic segmentation by global convolutional network [C]//Proceedings of the Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on, IEEE, 2017: 1743-1751.

[16] Seferbekov S, Iglovikov V, Buslaev A, et al. Feature pyramid network for multi-class land segmentation [C]//Proceedings of the The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018.

[17] Zhou Y, Zhu Y, Ye Q, et al. Weakly supervised instance segmentation using class peak response[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 3791-3800.

[18] Huang L, Xia W, Zhang B, et al. MSFCN-multiple supervised fully convolutional networks for the osteosarcoma segmentation of CT images [J]. Computer Methods and Programs in Biomedicine, 2017, 143：67-74.

[19] Xie S, Tu Z. Holistically-nested edge detection [C]//Proceedings of the Proceedings of the IEEE International Conference on Computer Vision, 2015: 1395-1403.

[20] Milletari F, Navab N, Ahmadi SA. V-net: Fully convolutional neural networks for volumetric medical image segmentation [C]//Proceedings of the 3D Vision (3DV), 2016 Fourth International Conference on, IEEE, 2016: 565-571.

[21] Chollet F. Keras[EB/OL]. 2015-06-27. https://keras.io.

[22] Kingma DP, Ba J. Adam: a method for stochastic optimization [C] San Diego：the 3rd International Conference for Learning Representations, 2015：1-5.

[23] Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks [C]//Proceedings of the Proceedings of the thirteenth international conference on artificial intelligence and statistics, 2010: 249-256.

[24] Taha A A, Hanbury A. Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool[J]. BMC medical imaging, 2015, 15(1): 29.

[25] Cardenes R, de Luis-Garcia R, Bach-Cuadra M. A multidimensional segmentation evaluation for medical image data [J]. Computer Methods and Programs in Biomedicine, 2009, 96(2): 108-124.

[26] Veropoulos K, Campbell C, Cristianini N. Controlling the sensitivity of support vector machines [C]//Proceedings of the Proceedings of the International Joint Conference on AI, 1999: 60.

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