基于Transformer的肺肿瘤三维CT图像分割  

Transformer-based segmentation of CT images of patients with lung tumors

作　　者：王伟桐 玄萍[1,2] WANG Weitong;XUAN Ping(College of Computer Science and Technology,Heilongjiang University,Harbin 150080,China;College of Engineering,Shantou University,Shantou 515063,Guangdong,China)

机构地区：[1]黑龙江大学计算机科学技术学院,哈尔滨150080 [2]汕头大学工学院,广东汕头515063

出　　处：《智能计算机与应用》2024年第3期76-80,共5页Intelligent Computer and Applications

基　　金：国家自然科学基金(61972135);黑龙江省自然科学基金项目(LH2019F049);中国博士后科学基金(2019M650069);黑龙江省博士后科研启动基金(BHLQ18104)。

摘　　要：基于信息学技术自动分割病人的肺部CT图像,有助于医生对于肺癌患者的早期诊断,提取和整合图像区域间的空间关联,对于提升肺肿瘤分割性能是十分重要的。本文提出了一个新的基于Transformer的分割模型,用于肺肿瘤三维CT图像分割、学习和整合此类关联。本文分别设计了带有混合多头图像区域节点注意力的Transformer模块和类别注意力模块,学习并融合了肺部CT图像的空间层面和通道层面的信息。将新的基于Transformer的分割模型同其他较为先进的模型进行了对比实验,实验结果表明新的模型在骰子系数、交并比和豪斯多夫距离等方面优于其他模型。Automatic segmentation of CT images of patients with lung tumors based on informatics technology is helpful for promoting the early diagnosis of lung cancer patients.Extracting and integrating the spatial correlation among image regions is very important for improving the segmentation performance of lung tumors.The paper designs a Transformer module and a category attention module with a hybrid multi-head image region node attention,respectively,which learned and fused the spatial and channel-level information of the lung CT image.The paper designs the image region node attention mechanism and the category attention mechanism to learn and fuse the information of spatial level and that of channel level.The proposed method is compared with several segmentation methods for segmenting lung tumors.The experimental results show that the proposed segmentation model is superior to the compared segmentation models in Dice,IoU and HD distance.

关 键 词：肺部CT图像 图像区域节点注意力 TRANSFORMER 类别注意力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]