一种抗金属伪影干扰的双流自注意力分割网络
DOI:
作者:
作者单位:

1.天津大学;2.天津仁爱学院

作者简介:

通讯作者:

中图分类号:

TP391.4

基金项目:

国家自然科学基金项目(面上项目,重点项目,重大项目)


An Anti-Metal Artifact Interference Dual-Stream Self-Attention Segmentation Network
Author:
Affiliation:

1.Tianjin University;2.Tianjin Renai College

Fund Project:

The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    计算机断层扫描(computed tomography, CT)影像受金属伪影干扰时其质量会大幅降低。目前许多深度学习方法被用来去除金属伪影,但是校正后图像与正常CT存在数据分布差异。即使经过伪影去除,后续的CT影像分割任务也难以获得理想分割结果。本文提出了一种抗金属伪影干扰的分割网络,可以在伪影数据和无伪影数据上直接完成语义分割。此网络采用双流复合连接结构,对两条主干上的编码器进行交互信息传导,两条主干分别用于提取带金属伪影图像和无伪影图像的特征。在带伪影的主干中,提出基于Transformer的焦点自注意力机制模块,在多尺寸分辨率下对全局信息进行编码,并使用混合损失和辅助监督优化网络的训练过程。实验结果表明,此网络在金属伪影干扰下分割结果的平均Dice系数、MIoU和Recall分别为86.40%,93.11%和90.76%。该网络在面向CT影像语义分割时具有很好的抗金属伪影干扰效果,无需伪影校正的情况下可获得较高的分割精度。

    Abstract:

    The quality of computed tomography (CT) images is degraded when being disturbed by metal artifact. Many deep learning methods are proposed to remove metal artifact, but there are data distribution differences between the corrected and normal images. Even removing the artifact, it is difficult to achieve compatibility for downstream semantic segmentation task. To address this problem, this paper proposes a segmentation network that is resistant to metal artifact interference, and it can accomplish semantic segmentation directly on both artifact and artifact-free data. The network uses a dual-stream composite connection structure to interact, which are used to extract features of metal artifact images and features of metal-free images, respectively. We develop a focal self-attention Transformer block for extracting global encoding at multi-size resolutions. The mixture loss and auxiliary supervision are adopted to optimize training process. The experimental results show that this network on metal artifact data could reach 86.40%, 93.11% and 90.76% in average Dice coefficient, MIoU and Recall, respectively. This network has great anti-metal artifact interference effect in semantic segmentation for CT images, and it could reach high segmentation accuracy without artifact reduction.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-03-09
  • 最后修改日期:2023-04-26
  • 录用日期:2023-05-10
  • 在线发布日期:
  • 出版日期: