基于卷积神经网络和Transformer的肝脏CT图像分割方法  

Liver CT image segmentation method based on CNN and Transformer

作　　者：胡晓阳 李哲 HU Xiaoyang;LI Zhe(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110000,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,辽宁沈阳110000

出　　处：《中国医学物理学杂志》2023年第4期423-428,共6页Chinese Journal of Medical Physics

基　　金：辽宁省高校创新人才项目(LR2019058);辽宁省重点科技创新基地联合开放基金(2021-KF-12-05)。

摘　　要：针对现有的卷积神经网络在肝脏图像分割上精度较低的问题,提出了一种以U-Net网络模型为基础的分割算法。将多头自注意力机制引入到U-Net网络的跳跃连接中,在编码器部分使用空洞卷积,采用混合损失函数从而提高分割精度。在LITS数据集上通过实验结果表明,利用本文方法进行肝脏分割与传统U-Net方法相比Dice系数提升3.3%,平均交并比提升了2.4%,平均像素准确率提升了3.66%。Aiming at the problem of low accuracy of the existing convolutional neural network in liver image segmentation,a segmentation algorithm based on U-Net model is presented.The segmentation accuracy is improved using multi-head selfattention mechanism which was introduced into the skip connection of U-Net,atrous convolution in the encoder,and mixed loss function.The experimental results on LITS data set show that the Dice coefficient,mean itersection over union and mean pixel accuracy of liver segmentation using the proposed method are improved by 3.3%,2.4%and 3.66%as compared with traditional U-Net method.

关 键 词：卷积神经网络 肝脏 图像分割 多头自注意力机制 空洞卷积 

分 类 号：R318[医药卫生—生物医学工程] TP391[医药卫生—基础医学]