[1]杨靖祎,谢 洋,周晓叶,等.基于 3D CNN 的肺结节假阳性筛查模型[J].计算机技术与发展,2022,32(02):196-201.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 032]
　YANG Jing-yi,XIE Yang,ZHOU Xiao-ye,et al.False Positive Screening of Pulmonary Nodules with 3D CNN[J].,2022,32(02):196-201.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 032]
点击复制

基于 3D CNN 的肺结节假阳性筛查模型()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年02期
页码:
196-201
栏目:
应用前沿与综合
出版日期:
2022-02-10

文章信息/Info

Title:
False Positive Screening of Pulmonary Nodules with 3D CNN
文章编号:
1673-629X(2022)02-0196-06
作者:
杨靖祎1 ; 谢 洋2 ; 周晓叶2 ; 陈隆鑫2 ; 底 涛1*
1. 河北医科大学第二医院 数据中心,河北 石家庄 050051;
2. 河北医科大学第二医院 信息中心,河北 石家庄 050051
Author(s):
YANG Jing-yi1 ; XIE Yang2 ; ZHOU Xiao-ye2 ; CHEN Long-xin2 ;  DI Tao1*
1. Data Center,Second Hospital of Hebei Medical University,Shijiazhuang 050051,China;
2. Information Center,Second Hospital of Hebei Medical University,Shijiazhuang 050051,China
关键词:
肺结节; 假阳性筛查; 密集神经网络; 三维卷积神经网络; 深度学习
Keywords:
pulmonary nodule; false positive screen; DenseNet; 3D CNN; deep learning
分类号:
TP311
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 02. 032
摘要:
通过肺部 CT 影像进行肺结节检测是肺癌早期筛查的重要手段,而候选结节的假阳性筛查是结节检测的关键部分。 传统的结节检测方法严重依赖先验知识,流程繁琐,性能并不理想。 在深度学习中,卷积神经网络可以在通用的学习过程中提取图像的特征。 该文以密集神经网络为基础设计了一个三维结节假阳性筛查模型—三维卷积神经网络模型(TDN-CNN)。 首先利用 U-Net 提取 CT 图像的肺实质再截取以结节为中心的 VOI,通过平移和翻转扩充正样本数据;在 3维假阳性筛查网络中,通过稠密连接强化特征利用、扩大特征空间,采用瓶颈层降低参数冗余,训练中优化参数,最终获取最优模型。 与 2D CNN 相比,该模型充分利用了肺结节的三维空间特征。 该 3D CNN 在公开的 LIDC 数据集上的 CPM 得分达到 0. 840,显著高于其他几种 3D 模型。 实验结果证明了该模型的有效性,其适用于肺结节的假阳性筛查。
Abstract:
Pulmonary nodule detection through computer tomography images is the most primary way of early detection for lung cancer,and false positive screen is one of the most vital steps in automatic pulmonary nodule detection. Traditional nodule detection methods relyheavily on prior knowledge, cumbersome process and unsatisfactory performance. In deep learning, convolutional neural network canacquire image features in general learning process. In this paper,a 3D convolutional neural networks ( TDN-CNN) for reducing falsepositives of pulmonary nodules based on DenseNet is proposed. Firstly,the U-Net is used to segment the lung area,and then intercept theVOI centered on the pulmonary nodules and expand them by translation and flip. In the 3D CNN,the dense connection is used to achievefeature reuse and features reduction through the transfer layer,tune and optimized parameters in the training,and finally obtain the optimalmodel. Compared with 2D CNNs,it makes full use of the three - dimensional spatial features of pulmonary nodules. On the publiclyavailable LIDC dataset,the 3D CNN achieves CPM of 0. 840,which is significantly higher than that of other 3D CNNs. The experimentalresults demonstrate the effectiveness of the model,which is suitable for reducing false positives in pulmonary nodule detection systems.

相似文献/References:

[1]朱诗生,王慧娟,李淳鑫.基于深度学习和模型集成的肺结节分割方法[J].计算机技术与发展,2023,33(02):208.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 031]
　ZHU Shi-sheng,WANG Hui-juan,LI Chun-xin.Pulmonary Nodule Segmentation Method Based on Deep Learning and Model Integration[J].,2023,33(02):208.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 031]
[2]杨靖祎,张翠肖 *,戴 健,等.基于密集神经网络的肺结节假阳性筛查模型[J].计算机技术与发展,2021,31(04):147.[doi:10. 3969 / j. issn. 1673-629X. 2021. 04. 025]
　YANG Jing-yi,ZHANG Cui-xiao *,DAI Jian,et al.False Positive Screening of Pulmonary Nodules with DenseNet[J].,2021,31(02):147.[doi:10. 3969 / j. issn. 1673-629X. 2021. 04. 025]

更新日期/Last Update: 2022-02-10