《计算机应用研究》|Application Research of Computers

基于残差双注意力U-Net模型的CT图像囊肿肾脏自动分割

Automated segmentation of cystic kidney in CT images using residual double attention motivated U-Net model

免费全文下载 (已被下载 次)  
获取PDF全文
作者 徐宏伟,闫培新,吴敏,徐振宇,孙玉宝
机构 1.南京信息工程大学 自动化学院 江苏省大气环境与装备技术协同创新中心,南京 210044;2.中国人民解放军 63936部队,北京 102202;3.东部战区总医院 a.医学工程科;b.泌尿外科,南京 210044
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)07-066-2237-04
DOI 10.19734/j.issn.1001-3695.2019.03.0092
摘要 人体肾脏存在形状的多样性和解剖学的复杂性,囊肿病变也会导致肾脏形状发生大幅变化。为应对CT图像囊肿肾脏自动分割存在的诸多挑战,提出一种新型深度分割网络模型。该模型设计有带残差连接的双注意力模块,在残差结构的基础上,联合空间注意力和通道注意力机制自适应学习更加有效的特征表达。依据U-Net架构,以残差双注意力模块为基础模块构建编码器和解码器,设置层级间的跳跃连接,使网络能够更加关注肾脏区域特征,有效应对肾脏的形状变化。为了验证所提模型的有效性,从医院共采集79位肾囊肿患者的CT图像进行训练和测试,实验结果表明该模型能够准确分割CT图像切片中的肾脏区域,且各项分割指标优于多个经典分割网络模型。
关键词 CT图像; 囊肿肾脏分割; 深度网络分割模型; 注意力机制
基金项目 国家自然科学基金资助项目(61672292)
江苏省高等学校自然科学研究重大资助项目(18KJA52007)
江苏省“六大人才高峰”资助项目(DZXX-037)
本文URL http://www.arocmag.com/article/01-2020-07-066.html
英文标题 Automated segmentation of cystic kidney in CT images using residual double attention motivated U-Net model
作者英文名 Xu Hongwei, Yan Peixin, Wu Min, Xu Zhenyu, Sun Yubao
机构英文名 1.Jiangsu Collaborative Innovation Center on Atmospheric Environment & Equipment Technology,School of Automation,Nanjing University of Information Science & Technology,Nanjing 210044,China;2.No. 63936 Unit,People's Liberation Army,Beijing 102202,China;3.a.Dept. of Medical Engineering,b.Dept. of Urinary Surgery,General Hospital of Eastern Theater Command,PLA,Nanjing 210044,China
英文摘要 Human kidneys have the variety of shapes and the complexity of anatomy. Cyst lesions can also cause large changes in kidney shape. This paper proposed a new deep network segmentation model to cope with the many challenges of automatic segmentation of CT image cysts. The proposed model deployed a dual attention module with residual connection. Based on the residual structure, it adopted the joint spatial attention and channel attention mechanism to learn more effective feature expression. According to the U-Net architecture, it built the encoder and decoder with the residual dual attention module as the building block, and also set the jump connections between the layers, so that the network could pay more attention to the characteristics of the kidney region and cope well with the changes in kidney shape. In order to verify the validity of the proposed model, it collected CT images of 79 patients with renal cysts from the hospital for training and testing. The experimental results show that the model can accurately segment the kidney regions in CT image slices, and the segmentation indicators are better than some classic segmentation network models.
英文关键词 CT image; cyst kidney segmentation; deep segmentation network; attention mechanism
参考文献 查看稿件参考文献
 
收稿日期 2019/3/15
修回日期 2019/4/19
页码 2237-2240
中图分类号 TP393.04
文献标志码 A