北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2022, Vol. 45 ›› Issue (4): 28-34.

• 论文 • 上一篇    下一篇

基于数据增强及注意力机制的肺结节检测系统

李阳,高轼奇   

  1. 长春工业大学
  • 收稿日期:2021-09-01 修回日期:2021-12-16 出版日期:2022-08-28 发布日期:2022-06-26
  • 通讯作者: 高轼奇 E-mail:gaozed@163.com
  • 基金资助:
    国家自然科学基金项目;吉林省教育厅项目

A lung CAD system based on Data Augmentation and CA-YOLO-V4 detection framework

1, 2   

  1. 1.
    2. Changchun University of Technology
  • Received:2021-09-01 Revised:2021-12-16 Online:2022-08-28 Published:2022-06-26

摘要: 为了解决带标注的医学影像数据过少导致模型学习能力有限以及深度检测过程中下采样带来的微小结节特征容易丢失的问题,设计了一种基于CT-GAN的数据增强及改进YOLO-V4检测框架的肺CAD系统. 第一部分,在结节生成算法CT-GAN中引入DropBlock正则化方法,实现带标注医学影像的数据增强,以提升肺结节的生成质量. 第二部分,在YOLO-V4中引入注意力机制,构造CA-YOLO-V4框架,以捕捉肺结节的位置感知、方向感知和跨通道的信息,帮助模型更加精确地检测肺结节感兴趣区域. 实验结果表明,所提肺CAD系统,在LUNA16数据集上,数据增强及结节检测的性能指标均优于对比算法,能够有效扩充数据集,提升结节检测性能.

关键词: 肺计算机辅助检测系统, 数据增强, 肺结节检测, CA-YOLO-V4, 改进CT-GAN

Abstract: For solving the problem of limited model learning ability caused by too little labeled medical image data and easy loss of tiny nodule features caused by sub-sampling in the process of deep detection, a lung CAD system based on CT-GAN data augmentation and improved YOLO-V4 detection framework was designed. In the first part, the regularization method DropBlock was introduced into the nodule generation algorithm CT-GAN to augment the data of annotated medical images, so as to improve the generation quality of pulmonary nodules.In the second part, attention model was introduced in YOLO-V4, and CA-YOLO-V4 framework was constructed to capture the position perception, direction perception and cross-channel information of pulmonary nodules, helped the model to detect the region of interest of pulmonary nodules more accurately. The experimental results show that the performance indexes of data augmentation and nodule detection in LUNA16 data set of the proposed lung CAD system are superior to the comparison algorithm, which can effectively expand the data set and improve the performance of nodule detection.

Key words: Lung Computed Aided Detection, Data Augmentation, Pulmonary nodule detection, CA-YOLO-V4, The improved CT-GAN

中图分类号: