摘要
提出RDU-Net网络进行肺结节的分割工作。该网络以U-Net网络为基础,针对网络训练过程中经常出现的梯度消失现象,引入残差单元对基础网络进行改进,该操作很好地解决了网络模型训练过程中出现的梯度消失问题;为了提高网络的泛化能力,在网络中增加了Dropout层,以避免网络在训练过程中过拟合现象,进一步提高分割精度。该网络在LIDC-IDRI数据集上进行实验,其AUC和Dice分别达到了0.89和0.76,相较于基础网络其分割精度和分割效果都有一定的提高。
This paper proposes the RDU-Net network to segment lung nodules.The network is based on the U-Net network.Aimed at the phenomenon of gradient disappearance that often occurred during network training,the residual unit was introduced to improve the basic network.This operation solved the problem of gradient disappearance during the network model training process.In order to improve the generalization ability of the network,a Dropout layer was added to the network to avoid overfitting during the training process and further improve the segmentation accuracy.The network was tested on the LIDC-IDRI dataset,and its AUC and Dice reached 0.89 and 0.76,respectively.Compared with the basic network,its segmentation accuracy and segmentation effect were improved to a certain extent.
作者
陈亚浩
韩林
刘艳青
张悦
Chen Yahao;Han Lin;Liu Yanqing;Zhang Yue(School of Information Engineering,Zhengzhou University,Zhengzhou 450000,Henan,China;National Supercomputing Zhengzhou Center,Zhengzhou 450052,Henan,China)
出处
《计算机应用与软件》
北大核心
2024年第7期177-183,共7页
Computer Applications and Software
关键词
深度学习
U-Net
肺片分割
肺结节
Deep learning
U-Net
Lung slice segmentation
Lung nodules
