基于改进Swin⁃Unet腹部多器官图像分割方法研究  

Research on abdominal multi organ image segmentation based on improved Swin⁃Unet

作　　者：王雅丽 Wang Yali(School of Electronic Information,Southwest Minzu University,Chengdu 610225)

机构地区：[1]西南民族大学电子信息学院,成都610225

出　　处：《现代计算机》2023年第3期81-84,共4页Modern Computer

基　　金：西南民族大学中央高校基本科研业务费专项资金项目(2021NYYXS74)。

摘　　要：腹部多器官计算机断层扫描(CT)图像自动化分割可以辅助医生进行诊断,提高诊断效率。以往的分割方法由于腹部易发生形变、体积较小且器官密集导致分割效果相对较差。为此,提出基于改进Swin⁃Unet腹部多器官图像分割方法,使得网络在下采样过程中更加关注器官细节特征。实验表明,提出的方法相对于现有分割方法具有更高的准确性。Automatic segmentation of abdominal multi organ computed tomography(CT)images can assist doctors in diagno⁃sis and improve diagnostic efficiency.The previous segmentation methods are relatively poor for organs that are prone to deforma⁃tion,small and dense.Therefore,an improved Swin⁃Unet based abdominal multi organ image segmentation method is proposed,which makes the network pay more attention to organ details in the down sampling process.Experimental results show that the pro⁃posed method has higher accuracy than existing segmentation methods.

关 键 词：腹部多器官分割 语义分割 Swin⁃Unet TRANSFORMER 

分 类 号：R816.5[医药卫生—放射医学] TP391.41[医药卫生—临床医学]