| 戴垚均,闫士举,宋成利.基于密集网络改进的肺结节良恶性分类模型[J].中国医学影像技术,2018,34(7):1104~1109 |
| 基于密集网络改进的肺结节良恶性分类模型 |
| Benign or malignant lung nodules classification model based on modified DenseNet |
| 投稿时间：2017-10-09 修订日期：2018-04-05 |
| DOI：10.13929/j.1003-3289.201710013 |
| 中文关键词: 深度学习 肺结节 良恶性分类 肺部图像数据库联盟 |
| 英文关键词:Deep learning Lung nodules Benign or malignant classification Lung image database consortium |
| 基金项目: |
|
| 摘要点击次数: 2641 |
| 全文下载次数: 1533 |
| 中文摘要: |
| 目的 探讨改进后的卷积神经网络模型对肺结节进行良恶性分类的准确率。方法 以分类模型密集网络（DenseNet）为基础模型，采用中间密度投影方法将肺结节的三维信息输入卷积神经网络进行训练，并针对肺结节良恶性分类问题适应性改进神经网络结构，将传统损失函数Cross Entropy Loss替换为Focal Loss，使网络能着重学习难以分辨的肺结节。结果 改进后神经网络模型对良恶性肺结节分类的准确率为89.93%，曲线下面积为0.947。结论 适应性改进后的卷积神经网络模型判断良恶性肺结节准确率较高。 |
| 英文摘要: |
| Objective To investigate the accuracy of classification of benign and malignant lung nodules with computer-aided diagnosis (CAD) systems.Methods DenseNet was used as the basic model, and 3D information of lung nodules was incorporated into training phase of the convolutional neural network via performing median density projection. Then the network structure was adapted to the classification of benign and malignant lung nodules. Traditional Cross Entropy Loss was replaced with Focal Loss to enable the network to focus on learning difficult-to-resolve lung nodules.Results The accuracy of the modified network model for classification of benign and malignant lung nodules was 89.93%, and the area under the curve was 0.947.Conclusion The adapted convolutional neural network model has high accuracy in the classification of benign and malignant lung nodules. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
