计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (24): 171-177.DOI: 10.3778/j.issn.1002-8331.1809-0190

• 图形图像处理 • 上一篇    下一篇

融合多维度卷积神经网络的肺结节分类方法

吴保荣,强彦,王三虎,唐笑先,刘希靖   

  1. 1.太原理工大学 信息与计算机学院,山西 晋中 030600
    2.吕梁学院 计算机科学与技术系,山西 吕梁 033000
    3.山西省人民医院 PET/CT中心,太原 030024
    4.山西农业大学 软件学院,山西 晋中 030600
  • 出版日期:2019-12-15 发布日期:2019-12-11

Fusing Multi-Dimensional Convolution Neural Network for Lung Nodules Classification

WU Baorong, QIANG Yan, WANG Sanhu, TANG Xiaoxian, LIU Xijing   

  1. 1.College of Information and Computer, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China
    2.College of Computer Science and Technology, Lvliang University, Lvliang, Shanxi 033000, China
    3.Department of PET/CT Center, Shanxi Provincial People’s Hospital, Taiyuan 030024, China
    4.College of Software, Shanxi Agricultural University, Jinzhong, Shanxi 030600, China
  • Online:2019-12-15 Published:2019-12-11

摘要: 针对CT图像肺结节分类任务中分类精度低,假阳性高的问题,提出了一种加权融合多维度卷积神经网络的肺结节分类模型,该模型包含两个子模型:基于二维图像的多尺度密集卷积网络模型,以捕获更宽泛的结节变化特征并促进特征重用;基于三维图像的三维卷积神经网络模型,以充分利用结节空间上下文信息。使用二维和三维CT图像训练子模型,根据子模型分类误差计算其权重,对子模型分类结果进行加权融合,得到最终分类结果。该模型在公共数据集LIDC-IDRI上分类准确率达到94.25%,AUC值达到98%。实验结果表明,加权融合多维度模型可以有效地提升肺结节分类性能。

关键词: 肺结节分类, 卷积神经网络, 深度学习, 多维度, 加权融合, CT图像

Abstract: In order to solve the problem of low classification precision and high false positive in the classification task of lung nodules in CT image, a benign and malignant classification model of lung nodules based on weighted fusion multi-dimensional convolution neural network is proposed. The model contains two sub-models:a multi-scale dense convolutional network model based on two-dimensional images to capture more extensive nodule variation features and promote feature reuse, and the three-dimensional convolutional neural network model based on three-dimensional images to make full use of spatial context information of nodules. 2D and 3D CT images are used to train the sub-models. The weights of the sub-models are calculated according to the classification errors, and then the weights are used to fuse the sub-models classification results. The more accurate classification results are obtained. The classification accuracy of the model is 94.25% and the AUC value is 98% on the public dataset LIDC-IDRI. The experimental results show that the weighted fusion multi-dimensional model can effectively improve the classification performance of lung nodules.

Key words: lung nodule classification, convolutional neural network, deep learning, multi-dimensional, weighted fusion, CT image