基于混合注意力机制的肺结节假阳性降低

唐秉航; 王艳芳; 马力; 陈庆武; 邵立伟; 黄德皇

doi:10.15953/j.ctta.2021.002

基于混合注意力机制的肺结节假阳性降低

1.
中山市人民医院影像科，广东中山528404
2.
中山仰视科技有限公司，广东中山528400
3.
中山市北京理工大学研究院，广东中山528405

基金项目: 中山市2019年高端科研机构创新专项（第一批）（基于人工智能CT时序列的肺癌早期预测及其应用）

详细信息

作者简介:
唐秉航: 男，中山市人民医院主任医师，硕士生导师，主要从事影像放射诊断，E-mail：zstangbh@sina.com

王艳芳: 女，中山仰视科技有限公司CEO，主要从事人工智能深度学习技术在医学影像上的开发与应用系列研究，E-mail：yfwang6@sina.cn

中图分类号: R 814
计量
- 文章访问数: 291
- HTML全文浏览量: 157
- PDF下载量: 41
出版历程
- 收稿日期: 2021-09-23
- 录用日期: 2021-11-23
- 网络出版日期: 2021-12-01
- 刊出日期: 2022-01-31

False Positive Reduction of Pulmonary Nodules Based on Mixed Attentional Mechanism

1.
Zhongshan City People’s Hospital, Zhongshan 528404, China
2.
Zhongshan Yangshi Technology Co., Ltd, Zhongshan 528400, China
3.
Zhongshan Research Institute, Beijing Institute of Technology, Zhongshan 528405, China

摘要

摘要:
为了解决肺结节CAD系统候选结节检测阶段高假阳性问题，本文提出一种基于混合注意力机制的肺结节假阳性降低方法。该方法可作为目前假阳性降低阶段最常用的3D CNN分类模型的替代方案，能有效回避3D CNN模型参数量及计算量大的问题。该方法将三维候选结节切片数据看作切片序列，使用时序分割模型，结合改进的包含混合注意力模块的2D Resnet-18骨干网络，在使用2D CNN的基础上，有效学习三维切片数据的时空特征。相对于3D CNN结构的肺结节分类模型，本文提出的方法在降低模型参数量和推理时间的基础上，提高了结节分类的准确率。
- 时序分割模型 /
- 混合注意力 /
- 肺结节
Abstract:
In order to solve the problem of high false positives in the candidate detection stage of pulmonary nodules CAD system, this paper proposes a method to reduce false positives of pulmonary nodules based on mixed attention mechanism. The method can be used as an alternative to the most commonly used 3D CNN classification model at the stage of false positive reduction. It can effectively avoid the problems of large number of parameters and computation in 3D CNN model. In this method, the 3D candidate nodule data is viewed as a slice sequence, and the temporal segment networks model is used in combination with the improved 2D ResNet-18 backbone network which contains mixed attention modules. On the basis of using 2D CNN, the spatial and temporal characteristics of the 3D slice data are effectively studied. Compared with the 3D CNN structure model for pulmonary nodules classification, the method proposed in this paper not only improves the accuracy of nodules classification but also reduces the number of model parameters and the inference time.
- temporal segment networks /
- mixed attention /
- pulmonary nodules

HTML全文

下载: 全尺寸图片幻灯片

图 1 网络模型整体结构

Figure 1. The overall structure of the network model

下载: 全尺寸图片幻灯片

图 2 SE模块

Figure 2. SE module

下载: 全尺寸图片幻灯片

图 3 ME模块

Figure 3. ME module

下载: 全尺寸图片幻灯片

图 4 CA模块

Figure 4. CA module

下载: 全尺寸图片幻灯片

图 5 真阳性结节切片序列

Figure 5. Slice sequence of true positive nodules

下载: 全尺寸图片幻灯片

图 6 假阳性结节切片序列

Figure 6. Slice sequence of false positive nodules

下载: 全尺寸图片幻灯片

1 注意力消融实验

模型	参数量/M	召回率/％	精确率/％
Baseline	11.17	97.52	98.24
Ours w/o ME	11.33	98.01	98.88
Ours w/o CA	11.30	98.05	98.93
Ours w/o SE	11.33	97.97	98.82
Ours（SE+ME+CA）	11.39	98.23	99.18

下载: 导出CSV

表 1 注意力消融实验

Table 1 Attention ablation experiment

模型	参数量/M	召回率/％	精确率/％
Baseline	11.17	97.52	98.24
Ours w/o ME	11.33	98.01	98.88
Ours w/o CA	11.30	98.05	98.93
Ours w/o SE	11.33	97.97	98.82
Ours（SE+ME+CA）	11.39	98.23	99.18

下载: 导出CSV

2 本文模型与3D CNN基准比较

模型	参数量/M	召回率/％	精确率/％	推理时间/（s/单结节）
3D Resnet-18	33.16	97.95	98.81	0.011
Ours（SE+ME+CA）	11.39	98.23	99.18	0.005

下载: 导出CSV

表 2 本文模型与3D CNN基准比较

Table 2 Comparison between the model in this paper and the 3D CNN benchmark

模型	参数量/M	召回率/％	精确率/％	推理时间/（s/单结节）
3D Resnet-18	33.16	97.95	98.81	0.011
Ours（SE+ME+CA）	11.39	98.23	99.18	0.005

下载: 导出CSV

参考文献(15)

[1]	HAN F, WANG H, ZHANG G, et al. Texture feature analysis for computer-aided diagnosis on pulmonary nodules[J]. Journal of Digital Imaging, 2015, 28(1): 99−115. doi: 10.1007/s10278-014-9718-8
[2]	张婧, 李彬, 田联房, 等. 结合规则和SVM方法的肺结节识别[J]. 华南理工大学学报(自然科学版), 2011,39(2): 125−129, 147. ZHANG J, LI B, TIAN L F, et al. Lung nodule recognition combining rule-based method and SVM[J]. Journal of South China University of Technology (Natural Science Edition), 2011, 39(2): 125−129, 147. (in Chinese).
[3]	SETIO A, CIOMPI F, LITJENS G, et al. Pulmonary nodule detection in CT images: False positive reduction using multi-view convolutional networks[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1160−1169. doi: 10.1109/TMI.2016.2536809
[4]	ARMATO S G, ROBERTS R Y, MCNITT-GRAY M F, et al. The lung image database consortium (LIDC) and image database resource initiative (IDRI): A completed reference database of lung nodules on CT scans[J]. Academic Radiology, 2007, 14(12): 1455−1463. doi: 10.1016/j.acra.2007.08.006
[5]	高慧明, 赵涓涓, 刘继华, 等. 多尺度卷积神经网络用于肺结节假阳性降低[J]. 计算机工程与设计, 2019,40(9): 2718−2724. GAO H M, ZHAO J J, LIU J H, et al. Multi-scale convolutional neural network for pulmonary nodule false positive reduction[J]. Computer Engineering and Design, 2019, 40(9): 2718−2724. (in Chinese).
[6]	SETIO A A A, TRAVERSO A, de BEL T, et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The LUNA16 challenge[J]. Medical Image Analysis, 2017, 42: 1−13. doi: 10.1016/j.media.2017.06.015
[7]	王尚丽, 金戈辉, 徐亮, 等. 基于三维密集网络的肺结节检测方法[J]. 中国生物医学工程学报, 2020,39(1): 8−18. doi: 10.3969/j.issn.0258-8021.2020.01.02 WANG S L, JING G H, XU L, et al. Method for detecting pulmonary nodules based on three-dimensional dense network[J]. Chinese Journal of Biomedical Engineering, 2020, 39(1): 8−18. (in Chinese). doi: 10.3969/j.issn.0258-8021.2020.01.02
[8]	HUANG G, LIU Z, LAURENS V, et al. Densely connected convolutional networks[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 2261-2269.
[9]	WANG L, XIONG Y, ZHE W, et al. Temporal segment networks for action recognition in videos[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(11): 2740−2755. doi: 10.1109/TPAMI.2018.2868668
[10]	LIN J, GAN C, HAN S. TSM: Temporal shift module for efficient video understanding[C]// IEEE/CVF International Conference on Computer Vision (ICCV), 2019: 7082-7092.
[11]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770-778.
[12]	HU J, SHEN L, ALBANIE S, et al. Squeeze and excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011−2023. doi: 10.1109/TPAMI.2019.2913372
[13]	LI Y, JI B, SHI X, et al. TEA: Temporal excitation and aggregation for action recognition[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 906-915.
[14]	WANG Z, SHE Q, SMOLIC A. ACTION-Net: Multipath excitation for action recognition[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021: 13214-13223.
[15]	HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design[C]// Computer Vision and Pattern Recognition (CVPR), 2021: 13713-13722.

施引文献(3)

期刊类型引用(2)

1.	马力，黄德皇，王艳芳. 融合形状变换及纹理学习的肺结节生长预测. CT理论与应用研究. 2024(03): 317-324 . 本站查看
2.	朱玉婷，袁晓. 基于改进TransUNet模型的脑肿瘤图像分割方法研究. 计算技术与自动化. 2024(02): 98-104 . 百度学术