摘要
由于CT图像是三维图像,在原始的V-Net模型分割中,易出现结节漏检和边界分割不清晰,以及损失函数Dice训练时不稳定等问题。根据这些问题,提出3D多尺度SE V-Net,简称MSEV-Net网络,同时通过联合损失函数来提高训练的稳定性。该网络模型在V-Net网络的基础上,使用多尺度卷积模块来替换原有的5×5×5卷积,同时在残差连接后加入SE通道注意力模块,通过不同尺度的特征融合和学习不同通道之间的关系,解决肺结节小不易分割的问题。同时在V-Net网络残差连接基础上加一条短跳跃连接,使得整个网络更好利用全局特征。联合损失函数选择Dice和交叉熵损失函数进行融合,可以很好地解决训练不稳定问题。提出的MSEV-Net网络模型和联合损失函数在平均分割准确率PA达到0.998,DSC达到0.837。实验结果表明,该方法在提高肺结节分割精度方面具有一定的效果。
Since CT images are three-dimensional images,in the original V-Net model segmentation,it is prone to the problems of nodal omission and unclear boundary segmentation,as well as the instability during the training of the loss function Dice.According to these problems,we propose 3D multi-scale SE V-Net referred to as MSEV-Net network,while improving the stability of training by joint loss function.This network model is based on the V-Net network,using the multi-scale convolution module to replace the original 5×5×5 convolution,while adding the SE channel attention module after the residual connection to solve the problem of small lung nodules that are not easy to segment by fusing features of different scales and learning the relationship between different channels.At the same time,a short jump connection is added on top of the residual connection of the V-Net network,which makes the whole network better utilize the global features.The joint loss function selects Dice and cross-entropy loss function for fusion,which can well solve the problem of training instability.The MSEV-Net network model and joint loss function proposed reach 0.998 in the average segmentation accuracy PA and 0.837 in the DSC.The experimental results show that the proposed method is effective in improving the segmentation accuracy of lung nodules.
作者
李丽
林晓明
彭丰平
潘家辉
LI Li;LIN Xiao-ming;PENG Feng-ping;PAN Jia-hui(School of Software,South China Normal University,Foshan 528225,China;Guangdong Foshan Lianchuang Engineering Graduate School,Foshan 528311,China)
出处
《计算机技术与发展》
2024年第4期82-88,共7页
Computer Technology and Development
基金
广东省普通高校特色创新项目(2022KTSCX035)
国家自然科学基金面上项目(62076103)。
