摘要
Thispaperaimstomeettherequirementsofreducingthescanningtimeofmagneticresonanceimaging(MRI),acceleratingMRIandreconstructingahighqualityimagefromlessacquisitiondataasmuchaspossible.MRImethodbasedoncompressedsensing(CS)withmultipleregularizations(tworegularizationsincludingtotalvariation(TV)normandL1normorthreeregularizationsconsistingoftotalvariation,L1normandwavelettreestructure)isproposedinthispaper,whichisimplementedbyapplyingsplitaugmentedlagrangianshrinkagealgorithm(SALSA).TosolvemagneticresonanceimagereconstructionproblemswithlinearcombinationsoftotalvariationandL1norm,weutilizedcompositesplitdenoising(CSD)tosplittheoriginalcomplexproblemintoTVnormandL1normregularizationsubproblemswhichweresimpleandeasytobesolvedrespectivelyinthispaper.Thereconstructedimagewasobtainedfromtheweightedaverageofsolutionsfromtwosubproblemsinaniterativeframework.BecauseeachofthesplittedsubproblemscanberegardedasMRImodelbasedonCSwithsingleregularization,andforsolvingthekindofmodel,splitaugmentedlagrangealgorithmhasadvantageoverexistingfastalgorithmsuchasfastiterativeshrinkagethresholding(FIST)andtwostepiterativeshrinkagethresholding(TwIST)inconvergencespeed.Therefore,weproposedtoadoptSALSAtosolvethesubproblems.Moreover,inordertosolvemagneticresonanceimagereconstructionproblemswithlinearcombinationsoftotalvariation,L1normandwavelettreestructure,wecansplittheoriginalproblemintothreesubproblemsinthesamemanner,whichcanbeprocessedbyexistingiterationscheme.Agreatdealofexperimentalresultsshowthattheproposedmethodscaneffectivelyreconstructtheoriginalimage.ComparedwithexistingalgorithmssuchasTVCMRI,RecPF,CSA,FCSAandWaTMRI,theproposedmethodshavegreatlyimprovedthequalityofthereconstructedimagesandhavebettervisualeffect.
出版日期
2014年03月13日(中国Betway体育网页登陆平台首次上网日期,不代表论文的发表时间)