{"title":"Deformable medical image registration based on unsupervised generative adversarial network integrating dual attention mechanisms","authors":"Meng Li, Yuwen Wang, Fuchun Zhang, Guoqiang Li, Shunbo Hu, Liang Wu","doi":"10.1109/CISP-BMEI53629.2021.9624229","DOIUrl":null,"url":null,"abstract":"Registration is a basic subject of medical image research, and has been a research hotspot for decades. In the process of optimizing each pair of images, the traditional registration requires a lot of calculation, which is very time-consuming for a large amount of data. In recent years, the existing deep learning network framework, especially the model based on U-Net structure, has not only improved the computing speed, but also greatly improved the registration performance. However, the feature loss occurs in the UpSampling process of this structure. Hence, We propose a generative adversarial network using a dual attention mechanisms without any supervised information. In UpSampling process of the registration network, the dual attention mechanism is introduced to improve feature recovery ability. The dual attention mechanism consists of channel attention mechanism and location attention mechanism. For the registration network, local crosscorrelation loss functions are proposed to improve image similarity. Experiments show that our method has achieved perfect registration effect, especially in the edge region.","PeriodicalId":131256,"journal":{"name":"2021 14th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"94 27","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 14th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/CISP-BMEI53629.2021.9624229","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 2
Abstract
Registration is a basic subject of medical image research, and has been a research hotspot for decades. In the process of optimizing each pair of images, the traditional registration requires a lot of calculation, which is very time-consuming for a large amount of data. In recent years, the existing deep learning network framework, especially the model based on U-Net structure, has not only improved the computing speed, but also greatly improved the registration performance. However, the feature loss occurs in the UpSampling process of this structure. Hence, We propose a generative adversarial network using a dual attention mechanisms without any supervised information. In UpSampling process of the registration network, the dual attention mechanism is introduced to improve feature recovery ability. The dual attention mechanism consists of channel attention mechanism and location attention mechanism. For the registration network, local crosscorrelation loss functions are proposed to improve image similarity. Experiments show that our method has achieved perfect registration effect, especially in the edge region.