基于超声的腹部软组织漂移的校正系统的实现

上海交通大学硕士学位论文基于超声的腹部软组织漂移的校正系统的实现姓名：赵晶申请学位级别：硕士专业：软件工程指导教师：顾力栩20080101ICT(computertomography)CTBCTCTCTIIOpen-by-reconstructionCTCT(FRE)(SSD,SAD)NCCABSTRACTIIIIMPLEMENTATIONOFULTRASOUND-BASEDCORRECTIONSYSTEMFORABDOMENSOFTTISSUESHIFTABSTRACTTraditionalimage-guidednavigationsystemsestablish3Dmodelofpatient’sanatomicalstructureusing3Dreconstructionandvisualizationofpreoperative2Dimages.Onedrawbackofthisapproachisthemodalwillnotreflectthechangeofunderlyinganatomicalstructurecorrectlyiforgansdeformduetorespirationandbodymovement,whichwillresultininaccuratenavigation.Duetoitsreal-time,non-invasive,andhighlyflexibleadvantages,ultrasoundimaginghascomprehensiveclinicalapplications.However,itsdrawbackisthatimagesusuallyexhibitspeckleandhavelimitedregionofscanning.Bycombiningthepowerofbothpre-operativeCTandintra-operativeultrasound,it’spossibletotrackdeformedorgansandobserveinternalstructureclearly.Atpresent,3Dultrasoundguidednavigationsarewidelystudiedandappliedabroad.However,theyarenotyetpopularizedindomesticclinicalapplications.Undersuchcircumstance,weimplementacorrectionsystemforabdomensofttissueshiftbasedonintra-operative2Dultrasound.Throughregistrationandfusionofpre-operativeCTandintra-operativereal-timeultrasound,wecorrectthedeformedabdomenorgansinCT.ABSTRACTIVInthispaper,wefocusonalgorithmswhichareadoptedinseveralmodulesofcorrectionsystem,aswellastheirimplementations.Thesemodulesincludeultrasoundsimulation,ultrasoundfiltering,fusionofpre-operativeCTandintra-operativeultrasound.Inultrasoundsimulationmodule,weanalysisandimplementanalgorithmwhichhasahighdegreeofsimulation.Ittakesintoaccountboththeultrasoundacquisitionmodelandthespeckleformationmodel.Stagesincludingimagesampling,specklesimulation,andimageinterpolationareintroduced.Whenvalidatingtheeffectsofultrasoundfilteringalgorithmsandmultimodalregistrationalgorithms,theuseofcomputersyntheticimagesmakesitpossibleforustoapplysomeobjectiveevaluationcriteria.Inultrasoundfilteringmodule,weanalysisandimplementmathematicalmorphologybasedmethodsandmedianfilterbasedmethodsrespectively.Thebasicrequirementofultrasoundfilteringistopreservedetailedinformationusefultolaterdiagnosisasmuchaspossiblewhilesuppressingspeckleeffectively.WeobtainalgorithmOpen-by-reconstructionbycombiningbasicoperatorslikeerosionanddilationinmorphology,whichcaneliminatelightpointsinimageswhilepreservingimage’sgreyscalevaluesonageneralbasis.Inthemeantime,weusedecompositionofcommonstructuringelementstoacceleratefiltering.Median-typefiltershavegoodabilityofdetailpreservation,thusveryappropriatetoultrasoundfiltering.Weintroducemulti-stagemedianfilterandadaptiveweightedmedianfilterbasedonstandardmedian,applythemtoultrasoundimagesandreceivegoodresults.InCT-USfusionmodule,weapplyiterativeclosestpointalgorithmABSTRACTVandthin-platesplinesalgorithmtoglobalrigidtransformationandlocalnon-rigidtransformationrespectively.Meanwhile,similaritymeasuresincludingfiducialregistrationerror(FRE),averageintensitydifference(SSD,SAD),normalizedcrosscorrelation(NCC)areadoptedtoverifytheregistrationresults.Finally,fusionofregisteredimagesisperformed,andinformationfrombothmodalsaresynthesizedanddisplayedbyadjustmentofcontrastandtransparency.Keywords:image-guidednavigation,softtissueshift,ultrasoundsimulation,specklesuppression,imagefusion20081920081920081911.1[1-4]imageguidedsurgeryIGSCT[5][6]regionofinterestROICTCT[7][8][9][6][10][6]CT[10]CT[11]2CTCTCT1.21.2.11-11-1Fig.1-1Image-guidednavigationprocedureCTMRI[12][13]MRI/CT3[14][15]operatingroomOR1.2.21.31.3.1B[16]speckle[17]41-21-2BFig.1-2LiverB-modescan[18]1.3.2[19][20]free-hand1-351-3Fig.1-3Free-handultrasoundimagingsystem[17]1-41-4Fig.1-4Flatmagnetictransmitter5freedomofdegreeFOD3266CTCTCT1.4CTCTCTCT1.3.2CTCTCTCTCT71.5Open-by-reconstruction——CTICPTPSCT82.1Goodman[21]Burckhardt[22]BamberPizurica[23][24]2.2echo9resolutiontemporalresolutionspatialresolutionaxiallateral2-1Fig.2-1Sectoralscanultrasonogram[25]resolutioncell2-12.3[26]B2.3.1B2-2102-32-2BFig.2-2SectoralscanofB-modeultrasound2-3Fig.2-3Samplinggridandalgorithmparameters7nmΘwy0dmindmaxnmΘwy0w/2dmindmax2.1[26]I0IgIg(ρ,θ)[(3π-Θ)/2,(3π+Θ)/2]n11Θ/(n-1)[dmin,dmax]m(dmax-dmin)/(m-1)xyI0IgSampleI0nmΘy0dmindmaxn:mΘy0dmindmax1copyI0intoIg2foreachpixelIgdo3Ig←unassigned4forθ←(3π-Θ)/2to(3π+Θ)/2bynsteps5forρ←dmintodmaxbymsteps6x←ρ×cosθ+w/27y←-ρ×sinθ+y08Ig(x,y)←I0(x,y)9returnIg2.1Algorithm2.1Samplingalgorithm0-10-1VTKvtkImageDataSetScalarComponentFromDouble()IgIg255×255Lena2-4(a)2-4(b)2-12-1Θ12abLena2-4LenaFig.2-4SamplingonLena2-1LenaTable2-1SamplinggridparametersofLena2.3.2GoodmanBurckhardt[26]nmΘwy0dmindmax5050°55255306725513coherentu(x,y,z,t)=A(x,y,z)ei2πυt2-1A(x,y,z)υtA(x,y,z)=|A(x,y,z)|eiθ(x,y,z)2-2|A(x,y,z)|θ(x,y,z)I=f(|A(x,y,z)|2)2-3Charles[26]I0IgI=a|A(x,y,z)|2+b2-4ab10I=|A(x,y,z)|22-5φi(x,y,z)∑==NiixzyxA1z)y,,(),,(φ2-6NIg2-6vcMjIMNiCAAxxzyxAji+=+=∑∑=−=11z)y,,(z)y,,(),,(φφ2-7AcAvMAcIg2-5Av[0,2π]θ~U(0,2π)2-8[21]2222221v),(σσπσvueug+−=2-9uvσ140σ|A|~Rayleigh(0,σ)2-10e-iθ=cosθ-isinθ2-11eiθ=cosθ+isinθ2-2uvu=|A(x,y,z)|cosθ2-12v=|