返回
cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network

cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network

ID：40370016

大小：1.18 MB

页数：10页

时间：2019-08-01

cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network_第1页
cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network_第2页
cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network_第3页
cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network_第4页
cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network_第5页
资源描述：

《cvpr18-Translating and Segmenting Multimodal Medical Volumes With Cycle- and Shape-Consistency Generative Adversarial Network》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、TranslatingandSegmentingMultimodalMedicalVolumeswithCycle-andShape-ConsistencyGenerativeAdversarialNetworkZizhaoZhang+∗,LinYang+,YefengZheng∗+UniversityofFlorida∗MedicalImagingTechnologies,SiemensHealthcareAbstractSynthesizedmedicalimageshaveseveralim

2、portantap-CTMRIplications,e.g.,asanintermediumincross-modalityimageregistrationandassupplementarytrainingsamplestoboostthegeneralizationcapabilityofaclassifier.Especially,syn-thesizedcomputedtomography(CT)datacanprovideX-rayattenuationmapforradiationth

3、erapyplanning.InMRICTthiswork,weproposeagenericcross-modalitysynthesisapproachwiththefollowingtargets:1)synthesizingreal-isticlooking3Dimagesusingunpairedtrainingdata,2)ensuringconsistentanatomicalstructures,whichcouldbechangedbygeometricdistortioninc

4、ross-modalitysynthesisFigure1:Ourmethodlearnstwoparallelsetsofgenera-and3)improvingvolumesegmentationbyusingsynthetictorsGA/BandsegmentorsSA/BfortwomodalitiesAanddataformodalitieswithlimitedtrainingsamples.WeshowBtotranslateandsegmentholistic3Dvolumes

5、.Herewethatthesegoalscanbeachievedwithanend-to-end3Dcon-illustrateusingCTandMRIcardiovascular3Dimages.volutionalneuralnetwork(CNN)composedofmutually-beneficialgeneratorsandsegmentorsforimagesynthesisandsegmentationtasks.Thegeneratorsaretrainedwithanpla

6、nning[4].adversarialloss,acycle-consistencyloss,andalsoashape-Machinelearning(ML)basedmethodshavebeenwidelyconsistencyloss,whichissupervisedbysegmentors,tore-usedformedicalimageanalysis[41,40],includingdetec-ducethegeometricdistortion.Fromthesegmentat

7、ionview,tion,segmentation,andtrackingofananatomicalstructure.thesegmentorsareboostedbysyntheticdatafromgener-Suchmethodsareoftengenericandcanbeextendedfromatorsinanonlinemanner.Generatorsandsegmentorsoneimagingmodalitytotheotherbyre-trainingonthetar-p

8、rompteachotheralternativelyinanend-to-endtraininggetimagingmodality.However,asufficientnumberofrep-fashion.Withextensiveexperimentsonadatasetincludingresentativetrainingimagesarerequiredtoachieveenoughatotalof4,496CTandmagneticresonanceimaging(

当前文档最多预览五页，下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页，下载文档查看全文
温馨提示：
1. 部分包含数学公式或PPT动画的文件，查看预览时可能会显示错乱或异常，文件下载后无此问题，请放心下载。
2. 本文档由用户上传，版权归属用户，天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容，确认文档内容符合您的需求后进行下载，若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误，付费完成后未能成功下载的用户请联系客服处理。