作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (2): 307-313. doi: 10.19678/j.issn.1000-3428.0057019

• 图形图像处理 • 上一篇    下一篇

基于双注意力3D-UNet的肺结节分割网络模型

王磐1, 强彦1, 杨晓棠2, 侯腾璇1   

  1. 1. 太原理工大学 信息与计算机学院, 山西 晋中 030600;
    2. 山西省肿瘤医院 放射科, 太原 030000
  • 收稿日期:2019-12-25 修回日期:2020-02-16 出版日期:2021-02-15 发布日期:2020-02-12
  • 作者简介:王磐(1994-),男,硕士研究生,主研方向为医学图像处理、深度学习;强彦,教授、博士;杨晓棠,主任医师、博士;侯腾璇,硕士研究生。
  • 基金资助:
    国家自然科学基金(61872261);山西省自然科学基金(201801D121139);虚拟现实技术与系统国家重点实验室开放基金(VRLAB2018A08)。

Network Model for Lung Nodule Segmentation Based on Double Attention 3D-UNet

WANG Pan1, QIANG Yan1, YANG Xiaotang2, HOU Tengxuan1   

  1. 1. School of Information and Computer, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China;
    2. Department of Radiology, Shanxi Province Cancer Hospital, Tainyuan 030000, China
  • Received:2019-12-25 Revised:2020-02-16 Online:2021-02-15 Published:2020-02-12

摘要: 为提高计算机辅助诊断系统对大尺寸肺结节分割的完整度以及小尺寸肺结节的分割精度,构建双注意力3D-UNet肺结节分割网络模型。将传统3D-UNet网络中的上采样操作替换为DUpsampling结构,通过最小化特征图的像素点与被压缩标签图像之间的损失,得到更具表达能力的特征图,进而提高网络收敛速度。在此基础上,融入空间注意力模块和通道注意力模块,使单通道与多通道中相似的特征彼此相关,增加特征图的全局相关性以提高分割结果的精度。实验结果表明,与3D-UNet等方法相比,该模型有效提高了肺结节分割的准确率,在公共数据集LIDC-IDRI上的MIoU分数达到89.4%。

关键词: 肺结节分割, 空洞卷积, 注意力模块, 3D-UNet网络, DUpsampling结构

Abstract: In order to improve the segmentation integrity of large pulmonary nodules and the segmentation accuracy of small pulmonary nodules in computer-aided diagnosis systems,this paper constructs a Double Attention 3D-UNet(DA 3D-UNet) for lung nodule segmentation.The traditional upsampling operation in the 3D-UNet network is replaced with the DUpsampling structure.By minimizing the loss between the pixels of the feature map and the compressed label image,a more expressive feature map is obtained,thereby improving the network convergence speed.On this basis,the spatial attention module and the channel attention module are integrated to make the similar features in the single channel and the multi-channel correlated with each other,increase the global correlation of the feature map,and improve the accuracy of the segmentation result.Experimental results show that compared with 3D-UNet and other methods,this model effectively improves the accuracy of lung nodule segmentation,and its MIoU score on the public data set LIDC-IDRI reaches 89.4%.

Key words: lung nodule segmentation, atrous convolution, attention module, 3D-UNet network, DUpsampling structure

中图分类号: