正电子发射断层成像统计迭代重建算法分析-analysis of statistical iterative reconstruction algorithm for positron emission tomography.docxVIP

InvestigationofPositronEmissionTomographyImageStatisticalReconstructionAbstractPET(PositronEmissionTomography)isoneofthemostadvancedtechniquesintheworld.Itjoinsthelatestadvantageofthenuclearphysics,computervision,molecularbiology,medicalimages,anditisoneofnoninvasivemedicalimagingtoolsobservingdynamicallyandquantitativelythephysiologicalandbiochemicalchangesinvivo.Mathematicalmethodsareperformedonthesinogramtorecovertheactivitymap,whichindicatesthemetabolicchangesoftheorgans.PETimagereconstructionalgorithmsavailableareintroducedinthefirstpartofthethesis.Currently,itcanbeclassifiedintoanalyticalanditerativemethod.Theanalyticalmethodsaresimplicityandfastness.However,itsreconstructedimagesareverynoisy.Thoughthefamousmaximum-likelihoodexpectation-maximization(MLEM)canbetterexpresssystemmodelsofphysicaleffectsandmodelingthestatisticalPoissoncharacterofthedata,thereconstructedimagesalwaysstartdeterioratingtoproduce“checkerboardeffect”astheiterationproceedsanditsuffersslowconvergence.ThepopularBayesianreconstructioncangreatlyimprovereconstructionbyincorporatingimagepriorinformation.Butitalwaysfindthat,heavilyreliedontheinformationwithinalimitedneighborhood,conventionalBayesianmethodscanonlycontributelimitlocalpriorinformationtoreconstruction,whichmayleadstounfavorableoversmoothingeffectorbringstaircaseedgeartifacttoreconstruction.Toimprovetheshortageoftheiterativemethod,wedothefollowingworkonPETreconstructionalgorithms:ThetraditionalLeast-Squaresreconstructionalgorithmsofpositronemissiontomographycannoteffectivelysuppressthenoiseandhaveslowconvergence.Inordertosolvetheproblem,weintroduceapenaltypriortermintotheleast-squaresalgorithm,andcombinewithOrderedSubsets(OS)method,formingthepenalizedleast-squaresalgorithm,whichcanreconstructahigherqualityimage,butalsoenhancetheconvergencepreferable.WiththeThinPlatepriorintheBayesianreconstruction,combiningtheanisotropicdiffusionfilterwithabetterdiffusivity,wehaveanewreconstructionalgorithmwhichcannotonlyprotecttheedgeoftheimageasthe