北京生物医学工程

基于改进Attention U-Net的胆囊自动分割模型研究

Research on gallbladder automatic segmentation model based on improved Attention U-Net

作者：尹梓名孙大运任泰周雷李永盛王广义王传磊曹宏刘颖斌束翌俊

单位：上海理工大学医疗器械与食品学院(上海200093) ,上海交通大学医学院附属新华医院普外科(上海200092),上海交通大学医学院附属仁济医院胆胰外科(上海200127),上海市胆道疾病研究重点实验室(上海200092) ,癌基因及相关基因国家重点实验室(上海200127),吉林大学白求恩第一医院肝胆胰外一科(长春130021) ,吉林大学中日联谊医院普外科(长春130033)  通讯作者：刘颖斌，E-mail: laoniulyb@163.com，束翌俊，E-mail: shuyjun19881125@163.com

关键词：深度学习; 胆囊; 图像分割; U-NET; 注意力机制

分类号：R318.04;TP391.5

出版年·卷·期（页码）：2021·40·4（346-353）

摘要：

目的基于多尺度注意力融合机制，提出改进Attention U-Net的胆囊自动分割模型，提高胆囊自动分割模型的性能，以辅助医生进行临床诊断。方法首先选取2017年1月-2019年12月上海交通大学医学院附属新华医院普外科、吉林大学第一医院肝胆胰外一科和吉林大学中日联谊医院普外科收治的88例病理诊断明确的胆囊癌，28例慢性胆囊炎胆囊结石患者和29例正常胆囊患者，构建胆囊分割数据集，然后通过对医学常用深度学习图像分割方法U-Net和Attention U-Net进行分析，提出基于多尺度融合注意力机制改进的Attention U-Net方法，并设计实验对三种方法进行对比评估。结果提出的改进Attention U-Net方法在验证集上的交并比阈值（IoU）分数，Dice系数，检测精度（Precision）和召回率（Recall）分别为0.72，0.84，0.92，0.79，全部优于传统U-Net和Attention U-Net方法。结论本文提出了基于多尺度融合注意力机制改进的Attention U-Net模型，其性能优于U-Net和Attention U-Net，证明了本方法中改进的注意力机制可以很好地改善U-Net模型在胆囊影像上的分割结果。

Objective According to multi-scale attention fusion mechanism, an improved gallbladder automatic segmentation model based on attention u-net is proposed to assist doctors in clinical diagnosis. Methods 88 cases of gallbladder cancer, 28 cases of chronic cholecystitis with gallbladder stones and 29 cases of normal gallbladder were selected from general surgery department of Xinhua Hospital Affiliated to Medical College of Shanghai Jiaotong University, first department of hepatobiliary and pancreatic surgery of first hospital of Jilin University, general surgery department of China-Japan Friendship Hospital, Jilin University from January 2017 to December 2019. Through the analysis of the commonly used deep learning image segmentation methods U-Net and Attention U-Net, we proposed an improved Attention U-Net method based on multi-scale fusion attention mechanism, and designed experiments to verify and evaluate the three methods. Results the results showed that the IOU score, Dice coefficient, precision and recall rate of the improved Attention U-Net method were 0.72, 0.84, 0.92 and 0.79 respectively, which were all better than those of the traditional U-Net and Attention U-Net methods. Conclusions the performance of our improved Attention U-Net model is better than that of U-Net and Attention U-Net, which proves that the improved attention mechanism in this method can improve the segmentation result of U-Net model in gallbladder image.

参考文献：

[1]孙晓敏, 徐萍, 马志红,等. 上海松江地区胆囊良性疾病的流行病学调查30901例[J]. 世界华人消化杂志, 2011, 19(27):2881-2885.

??? Sun XM, Xu P, Ma ZH, et al. An epidemiological survey of benign gallbladder disease in Songjiang District of Shanghai, China[J]. World Chinese Journal of Digestology, 2011, 019(027):2881-2885.

[2] 朱忠伟,李福军,胡柯铭,等.宁波镇海地区胆囊疾病的流行病学调查[J].现代实用医学,2012,24(11):1241-1242.

[3] 赵登秋, 张建淮, 王卫东. 江苏淮安地区胆囊疾病的流行病学调查[J]. 消化外科, 2004, 3(1):39-40.

??? Zhao DQ, Zhang JH, Wang WD, et al. An epidemic survey of gallbladder disease in Huaian district of Jiangshu Province[J]. Journal of Digestive Surgery, 2004, 3(1):39-40.

[4] ?Li ML, Zhang Z, Li XG, et al. Whole-exome and targeted gene sequencing of gallbladder carcinoma identifies recurrent mutations in the ErbB pathway[J]. Nature Genetics, 2014, 46(8):872-876.

[5] Naser MA, Deen MJ. Brain tumor segmentation and grading of lower-grade glioma using deep learning in MRI images[J]. Computers in Biology and Medicine, 2020, 121:103758.

[6] Hu HF, Pan N, Wang JY, et al. Automatic segmentation of left ventricle from cardiac MRI via deep learning and region constrained dynamic programming[J]. Neurocomputing, 2019, 347: 139-148.

[7] Almotairi S, Kareem G, Aouf M, et al. Liver tumor segmentation in CT scans using modified SegNet[J]. Sensors, 2020, 20(5):1516.

[8] 谭丹滟. 腹部CT影像中胆囊分割系统的设计与实现[D].武汉:华中科技大学, 2017.

??? Tan DY. Design and implementation of gallbladder segmentation system in abdominal CT images[D]. Wuhan: Huazhong University of Science and Technology, 2017.

[9] 段雪琦. 腹部CT影像中的胆管分割方法研究[D].武汉:华中科技大学, 2017.

??? Duan XQ. A research on bile duct segmentation in abdominal CT images[D]. Wuhan: Huazhong University of Science and Technology, 2017.

[10] Huang YF, Xu ZZ, Yan CR. Automatic gallbladder location and segmentation based on anatomical knowledge[J]. Applied Mechanics & Materials, 2014, 556-562:4697-4700.

[11] Lian J, Ma Y, Ma Y, et al. Automatic gallbladder and gallstone regions segmentation in ultrasound image[J]. International Journal of Computer Assisted Radiology and Surgery, 2017, 12(4):553-568.

[12] Li C, Tan Y, Chen W, et al. ANU-Net: attention-based nested U-Net to exploit full resolution features for medical image segmentation[J]. Computers & Graphics, 2020, 90: 11-20.

[13] Alom MZ, Yakopcic C, Hasan M, et al. Recurrent residual U-Net for medical image segmentation[J]. Journal of Medical Imaging, 2019, 6(1): 014006.

[14] Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[M]// International Conference on Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015. Switzerland: ?Springer, Cham, 2015: 234-241.

[15] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE Press, 2016: 770-778.

[16] 刘杨,朱斌, 何健,等. 意外胆囊癌的影像学特征及误漏诊分析[J]. 中国医药导报，2020，17(1)：161-164.

Liu Y, Zhu B, He J, et al. Analysis of imaging features of unsuspected gallbladder cancer and causes of misdiagnosis or missed diagnosis[J]. China Medical Herald，2020，17(1)：161-164.

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