FULLY3D'03TUAM2–31ExtendedParallelBackprojection(EPBP)forArbitraryConeAngleandArbitraryPitch3DandPhase–Correlated4DCTReconstructionMarcKachelrie,WilliKalenderAbstract—RecentdevelopmentsinmedicalCTaimatfasterrotationspeedsandahighernumberofsimultane-ouslyacquiredslices.
TheseeortsarepushedfurtherbycardiacCTwhichiscurrentlythemostprominentspecial–purposeapplicationinCT[1].
Today,16-slicescannersarestate–of–the–art.
ButCTmanufacturershavealreadyan-nouncedscannerswithfarmoreslicesandsomeevenpro-moteprototypeswithupto256slices.
MedicalCTmustsupportthecircularscantrajectory(se-quencescan)andthespiraltrajectory(spiralscan).
Arbi-trarypitchselectionisofhighimportance.
Inanycase,thecompleteareaofthedetectoristobeexposedandeachmeasuredrayshouldcontributetotheimage.
Onlythen,optimizeddoseusagecanbeachieved.
Theserequirementscannotbefullledbycurrentre-constructionapproaches.
Exactcone–beamreconstruction,thatiscapableofreconstructinglargecone–angledata,can-notcopewitharbitrarypitchandphase–correlateddataseg-ments.
Onlyapproximatereconstructionapproacheshavethepotentialtohandleallrequirements.
Currently,theonlyknownapproachthatcanhandlephase–correlatedtruecone–beamdataisanextensiontotheAdvancedSingle–SliceRebinning(ASSR)algorithm[2],[3].
However,thisgeneralizedapproachASSRCIislimitedto32slices.
WehavethereforedevelopedanewapproximateFeldkamp–typealgorithm,theextendedparallelbackpro-jection(EPBP)[4].
Itsmainfeaturesareaphase–weightedbackprojectionandavoxel–by–voxel180normalization.
Therstfeatureensures3Dand4Dcapabilitieswithoneandthesamealgorithm,thesecondensures100%detectorusage(eachraycounts!
).
Thealgorithmisevaluatedusingsimulateddataofathoraxphantomandacardiacmotionphantomforscannerswithupto256slices.
Thestandardreconstructions(EPBPStd)areofexcellentqualityevenforasmanyas256slices.
Thecardiacrecon-structions(EPBPCI)areofhighqualityaswellandshownosignicantdeteriorationofobjectsevenfarothecenterofrotation.
SinceEPBPCIusesthecardiointerpolation(CI)phaseweightingthetemporalresolutionisequivalenttothatofthewellknownsingle–sliceandmulti–slicecardiacapproaches180CI,180MCI,andASSRCI,respectively,andliesintheorderof50msto100msforrotationtimesbetween0.
4sand0.
5s.
I.
IntroductionMEDICALcomputedtomographyiscurrentlyevolv-ingfasterthanever.
Increasedspatialresolution,de-creasedscantime,increasedtemporalresolution,decreasedpatientdose,andincreasedvolumecoveragearesomeoftheimportanttrendstomention.
Aslittleasveyearsago,single–slicespiralCTwasthestate–of–the–art.
Then,therst4–slicescannersbecameavailable.
Alreadyin2001,16–slicescannersstartedtore-InstituteofMedicalPhysics,UniversityofErlangen–N¨urnberg,Krankenhausstr.
12,91054Erlangen.
Correspondingauthor:PDDr.
MarcKachelrie,E–mail:marc.
kachelriess@imp.
uni–erlangen.
deplacethe4–slicemachines.
Apparently,thenearfuturewillshiftthenumberofsimultaneouslyscannedslicesto32,64andevenmore.
Slicethickness,andtherebyspatialresolu-tion,willcontinuetodecreasetofurtherimprovediagnosticaccuracy.
Atthesametime,doseutilizationwillincreasetokeeptheeectivepatientdoseatanacceptablelevel:dosemodulationtechniques,automaticexposurecontrolandimproveddetectormaterialswillhelptodoso[1].
Besidesimprovedspatialresolution,improvedcontrastresolutionandlowpatientdoseoneisfurtherinterestedinhighesttemporalresolutiontoallowimagingtheheart.
Thisisdoneusingshortrotationtimescombinedwithde-dicatedphase–correlatedreconstructionalgorithmsastheyareavailablesince1996/1997[5].
Therelevanceforimagereconstructionalgorithmsliesintheexpectedincreaseinconeangleandinthedemandforphase–correlatedreconstruction.
Neglectingthecone–angleofthescannerasdoneinall4–sliceandinallcom-mercialcardiac16–slicereconstructionalgorithmswillyieldunacceptableimageartifactsforfuturescanners.
Therearefastandecientcone–beamalgorithmswithacceptableimagequalityavailableforcone–beamspiralCT.
However,theydonotensure100%doseusageand/ortheydonotworkforarbitraryspiralpitch.
Further,noneofthemiscapableofcombiningthesedemandswithphase–correlatedimagingatwiderconeangles.
Tollthemissinggap,weproposeEPBP,anewapprox-imatecone–beamreconstructionthatallowsforarbitrarypitchandforphase–correlatedreconstructionandthaten-sures100%detectorusage.
DetailsofcardiacCTscanningandtherestrictionsonthemaximumpitchasafunctionofthepatient'sheartratecanbefoundinreferences[6],[7],[8].
Thephaseselectionmechanismsdescribedthere,namelythecardiointerpolation(CI)andthecardiodelta(CD)approachareusedforEPBPaswell.
EspeciallyEPBPCIthatusesCI,themostpromisingmulti–phaseweightingcurrentlyavailable,isevaluatedhere.
Inthispaper,wewilloutlinetheEPBPStdandtheEPBPCIalgorithmandgivesomedescriptiveexamples.
II.
SimulationsToevaluateournewapproachwehavesimulatedspiralcone–beamdatacorrespondingtothein–planegeometryofatypicalmedicalCTscanner(1160projectionsperro-tation,672detectorchannelsperdetectorrow,andafanangleΦ=52)usingadedicatedx–raysimulationtool(ImpactSim,VAMPGmbH,M¨ohrendorf,Germany).
Two2FULLY3D'03TUAM2–3phantomshavebeensimulated:thethoraxphantomde-scribedinthephantomdatabasehttp://www.
imp.
uni–erlangen.
de/forbildandthecardiacmotionphantomde-scribedin[7].
Thesimulatedscanprotocoluses0.
42srotationtime(143rpm),M*0.
75mmcollimationwithM=2msimul-taneouslyscannedsliceswherem=4,8,andapitchofp=0.
375.
Thetableincrementcanbefoundasd=MSp.
III.
GeometryA.
ScanGeometryThescangeometryassumedhereisafan–beamgeome-trywithcylindricaldetectorsandaspiralfocustrajectory.
TheEPBPapproachisbasedonarebinningtoparallelgeometry.
Othergeometries,suchasatdetectors,canbeeasilyincorporatedbymodifyingthecorrespondingrebin-ningandtransformequations.
Notethatinthelimitofp=0,thespiralreducestoacirculartrajectory.
EPBPcopeswithsequencedataaswellasitdoeswithspiraldata.
Thesourcetrajectoryisparameterizedbytheviewangleαass(α)=RFsinαcosα0+d001α2π.
(1)RFdenotestheradiusofthefocalspottrajectoryandddenotesthetableincrementperrotation.
Thecoordinatevectorofthedetectorelement(α,β,b)isgivenasr(α,β,b)=s(α)+RFDsin(α+β)cos(α+β)0+b001;βandbdenotethetransaxialandlongitudinaldetectorcomponents,respectively.
Torebinthetransaxialcomponentsofthecone–beamdatatoparallelgeometryweparameterizearaybyitsdis-tanceξtotheaxisofrotationandbyitsanglewithre-specttothenegativey–axis.
Thenormalformoftheray'sx–y–componentsisgivenasxcos+ysinξ=0.
Thisdenitionwaschosentohavethecentralraysforfan–beam(β=0)andforparallelbeam(ξ=0)coincidingforα=.
Therelationbetweenarayinfan–beamcoordinates(α,β)andaparallel–beamray(,ξ)isthefamiliartransform=α+βξ=RFsinβandα=+arcsinξ/RFβ=arcsinξ/RF.
(2)B.
PointProjectionForbackprojection,weneedtoknowthedetectorcoor-dinates(ξ,b)thatresultfromprojectingthepoint(x,y,z)froms(α)ontothecylindricaldetector.
Theradialcoordi-nateisgivenasξ=xcos+ysin.
Thelongitudinaldetectorcoordinatecanbecomputedus-ingtheintersectiontheorem.
ThetransaxialdistanceoftherespectivevoxeltothesourceisgivenasD2=(RFsinαx)2+(RFcosα+y)2=R2F2RFrsin(α)+r2with(x,y)=(rcos,rsin)or,equivalently,D=RFcosβ+η=R2Fξ2+ηwithη=ycosxsin.
Now,ndbbyscalingtheaxialdistancezdα2πfromDtoRFD:b=RFDD(zdα2π).
(3a)Andwendanotherrepresentationofξ:ξ=RFsinβ=RFxcosα+ysinαD.
(3b)IV.
ReconstructionTheextendedparallelbackprojectionalgorithmconsistsofthefollowingvesteps:-azimuthalrebinning:p(α,β,b)→p(,β,b),-longitudinalrebinning:p(,β,b)→p(,β,l),-radialrebinning:p(,β,l)→p(,ξ,l),-convolution:p(,ξ,l)→p(,ξ,l),-weightingandbackprojection:p(,ξ,l)→f(x,y,z).
A.
AzimuthalRebinningTheoriginalprojectiondatap(α,β,b)areconvertedfromfan–beamtofan–parallelgeometryusing(2)asfollows:p(,β,b)=p(α,β,b)withα=β.
B.
LongitudinalRebinningConvolvingspiraldatainthedetectorrowdirection(con-stantb)yieldsseverecone–beamartifacts.
AsindicatedbyASSR[2],SMPR[9],andexactcone–beamreconstruc-tion[10]theoptimaldirectionofconvolutionisthetangentds(α)/dα.
Toalignthefan–paralleldetectorrowswiththeoptimaldirectionofconvolutionalongitudinalrebinningisrequired.
Therefore,weareinterestedintherelationshipbetweenbandξwhenmovingalongds.
Using(3)onendsdbdξ=d(bD)d(ξD)=RFDdzRF(dxcosα+dysinα)=RFDd2πR2F;inthelaststep(1)wasusedtoinsertthecomponentsofds.
Now,wedeneanewlongitudinalvariablelasb=l+λξwithλ=dbdξ=dRFD2πR2Fsuchthatdl/dξ=0inthedirectionofds.
Then,dothelongitudinalrebinningp(,β,l)=p(,β,b)withb=l+λξ=lλRFsinβtoswitchtolasthenewindependentvariable.
FULLY3D'03TUAM2–33Wheneverbexceedsthedetectorlimitsbminandbmaxforsomeβweextrapolatebyrepeatingtheoutermostde-tectorrow.
Thevaluesthataremadeupbyextrapolationarerequiredduringconvolutionwhichoperatesoncom-pletedetectorrows,always.
Duringbackprojectiontheseextrapolatedpointsarenotaccessed;backprojectionratherrespectsthephysicaldetectorarea!
C.
RadialRebinningTheradialrebinningconvertstoequidistantparallelco-ordinates.
Weuse(2)tondp(,ξ,l)=p(,β,l)withβ=arcsinξ/RF.
D.
ConvolutionNow,convolutionofthedetectorrowsisperformedusingastandardconvolutionkernelk(ξ),as,forexample,theShepp–Logankernel.
p(,ξ,l)=p(,ξ,l)k(ξ)yieldstheconvolveddatapneededforbackprojection.
E.
WeightingandBackprojectionInthisstep,weregardthebackprojectionofaxedvoxel,sayonelocatedatr=(x,y,z).
LetVdenotethesetofviewanglesunderwhichrismeasured.
AssumeatemporalwindowTthatcomprisesallthatcorrespondtoalloweddata.
Forthestandardreconstruc-tionEPBPStd,alldataacquiredarealloweddataandthereforeT=R.
Forthereconstructionofcardiacdata,TcanbedenedbyspecifyingacardiacmotionphasecRthatcountsrelativetosomesynchronizationpeaksandaphasewidth0EPBPCIchoosescassmallasallowedbythecompletenesscondition(seebelow).
Otherdenitionsmayincludeabsolutetiminginformationortherestrictiontoonlyonetemporalintervaloflengthπ(single–phasere-construction),ortwointervals(bi–phasereconstruction).
RegardlessofwhatconventionisusedtodeneT,theintersectionI=V∩T,thatcomprisesallviewstobeused,mustbe180–complete:k(I+kπ)=R.
Now,assumeaweightingfunctionw()whosesupportequalsI,i.
e.
w(R\I)={0},andkw(+kπ)=0.
ThelastconditioncaneasilybeachievedbyusingpositiveweightsonIonly.
ForEPBPCIweuseamulti–triangularweightfunction:trianglefunctionslocatedoneachofI'sdisjunctintervals.
Bynormalizingwasw()=w()kw(+kπ)weachievekw(+kπ)=1anddw()=π.
Backprojectionf(x,y,z)=dp(,ξ,l)w()withξ=ξ(x,y,)=xcos+ysinα=β=+arcsinξ/RFl=b(x,y,z,α)λξthenyieldsthedesiredvoxelvalueat(x,y,z).
V.
ResultsFigure1showsthatimagequalityofthethoraxphantomisexcellentwithEPBP,evenforasmanyas256slices.
Asindicatedbytheribs,ASSR(whichisinfactdesignedforuptoaboutonly60slices[2])cannotcopewiththislargecone–angle;thesameappliestothehighlyrelatedAMPRalgorithmdenedinreference[11].
Fig.
1.
Thorax,scannedwith256*0.
75mmcollimationandd=72mmtableincrement.
Aheartrateof120min1wassimulatedforEPBPCI.
(0/500)ConsideringthatEPBPCIusesonlyafractionofthedataavailable(here,roughly25%),dependingonthelocalheartrateandonthereconstructionposition,theimagesarealmostasgoodastheEPBPStdreconstructions,apartfromtheincreasedimagenoise.
Theonlyexceptionisaslightvariationinthereconstructeddensityclosetothevertebrae.
Figure2showsreconstructionsofthecardiacmotionphantomfora16–sliceanda256–slicescanner.
Sincethe4FULLY3D'03TUAM2–3(a)16*0.
75mmcollimation(b)256*0.
75mmcollimationFig.
2.
TransaxialslicesandMPRsofthecardiacmotionphantomreconstructedwithvariousalgorithmsusingastandard16–slicescannerandawidecone–angle256–sliceCT.
ThesagittalMPRs(bottom)showadditionalδ–objectsusedtomeasureresolution.
(0/500)eldofviewshowsonlythecentralpartsofthepatient,thein–planeimagesoftheASSRapproachareacceptableevenforthe256–slicescanner.
However,with256–sliceASSRthemultiplanarreformations(MPRs)tendtobeblurredinthez–directionandfullwidthathalfmaximumFWHMzoftheslicesensitivityproleisincreasedsignicantlywhereasthein–planeresolutionFWHMxyisthesameasforthe16–slicecase.
EPBP,incontrast,behavesverywellforallsimulatedscanners(16,32,64,128and256slices).
Spatialresolu-tionisslightlyhigherthanforthesingle–slicerebinningalgorithms.
ImagenoiseincreasesforEPBPCIduetothephase–weighting.
Thisobservationisvalidforalltheothersimulatedgeometriesandheartrates(wehavelookedintofH=40min1140min1).
EPBPgenerallybehavesequaltoorbetterthanASSR.
Finally,gure3givesanexampleofreconstructedpa-tientdata.
Thedatashownarecorrelatedtothepatientmotionfunction,theso–calledkymogram,whichcandi-rectlybederivedfromtheacquiredrawdata[12].
ThestandardreconstructionsofASSRStdandEPBPStdarecomparableduetothelownumberofslices;thephase–correlatedEPBPCIimagesareofhighimagequalityandcorrespondtothegold–standardincardiacCTimaging.
VI.
DiscussionTheextendedparallelbackprojectionappearstobeade-quateformedicalCTimagereconstructioninallrespects.
EPBPimagequalityisequivalenttoexisting4–or16–slicestandardandcardiacalgorithmsforawiderangeofFULLY3D'03TUAM2–35Fig.
3.
ASSRandEPBP,12*0.
75mmcollimation,3.
375mmtableincrement.
Thephase–correlatedimagesarereconstructedrelativetothekymogrampeaksat0%and50%ofK-K,respectively.
(0/500)simultaneouslyscannedslices.
EvendatawithM=256slicesyieldsexcellentimagequality.
Forstandardrecon-structionsthisisnotsurprisingsinceEPBPStdissimilartootherFeldkampalgorithms(aslongastheseperformconvolutionalongthetangentdirection).
Forwideconeanglecardiacdata,wherenootherphase–correlatedcone–beamalgorithmisavailableyet,EPBPCIperformsverywellevenforobjectsfarotheisocenter(ribsingure1).
Feldkamp–typealgorithmsaresuperiortoASSR[2],AMPR[11],orSMPR[9]forlargeM.
Wideconeanglecar-diacCTiscurrentlyonlypossiblewithEPBP.
Itsuniqueweightingstrategythatassignsindividualdatarangestoeachvoxel,ensures100%datausageandthusthemax-imumdoseutilizationpossible.
ThefutureofmedicalsequentialandspiralCTwillcertainlyincludetheideaofphase–correlated/phase–weighted3DbackprojectionofEPBP–type,ormodicationsthereof.
References[1]W.
A.
Kalender,ComputedTomography.
Wiley&Sons,2003.
[2]M.
Kachelrie,S.
Schaller,andW.
A.
Kalender,"Advancedsingle–slicerebinningincone–beamspiralCT,"Med.
Phys.
,vol.
27,pp.
754–772,Apr.
2000.
[3]M.
Kachelrie,T.
Fuchs,R.
Lapp,D.
-A.
Sennst,S.
Schaller,andW.
Kalender,"ImagetovolumeweightinggeneralizedASSRforarbitrarypitch3Dandphase–correlated4Dspiralcone–beamCTreconstruction,"Proc.
ofthe6thInt.
MeetingonFully3DImageReconstruction,pp.
179–182,Nov.
2001.
[4]M.
KachelrieandW.
A.
Kalender,"Extendedparallelbackpro-jectionforcardiaccone–beamCTforupto128slices,"Radiol-ogy,vol.
225(P),p.
310,Nov.
2002.
[5]M.
KachelrieandW.
A.
Kalender,"ECG–basedphase–orientedimagereconstructionfromsubsecondspiralCTscansoftheheart,"Radiology,vol.
205(P),p.
215,Nov.
1997.
[6]M.
KachelrieandW.
A.
Kalender,"Electrocardiogram–correlatedimagereconstructionfromsubsecondspiralcomputedtomographyscansoftheheart,"Med.
Phys.
,vol.
25,pp.
2417–2431,Dec.
1998.
[7]M.
Kachelrie,S.
Ulzheimer,andW.
A.
Kalender,"ECG–correlatedimagereconstructionfromsubsecondmulti–slicespi-ralCTscansoftheheart,"Med.
Phys.
,vol.
27,pp.
1881–1902,Aug.
2000.
[8]M.
Kachelrie,S.
Ulzheimer,andW.
A.
Kalender,"ECG–correlatedimagingoftheheartwithsubsecondmultisliceCT,"IEEETransactionsonMedicalImaging,vol.
19,pp.
888–901,Sept.
2000.
[9]K.
Stierstorfer,T.
Flohr,andH.
Bruder,"Segmentedmultipleplanereconstruction–anovelapproximatereconstructionformulti–slicespiralCT,"Phys.
Med.
Biol.
,vol.
47,pp.
2571–2851,2002.
[10]K.
Sourbelle,PerformanceEvaluationofExactandApprox-imateCone–BeamAlgorithmsinSpiralComputedTomogra-phy.
PhDThesis,Friedrich–Alexander–Universit¨atErlangen–N¨urnberg,2002.
[11]S.
Schaller,K.
Stierstorfer,H.
Bruder,M.
Kachelrie,andT.
Flohr,"Novelapproximateapproachforhigh-qualityimagereconstructioninhelicalconebeamCTatarbitrarypitch,"SPIEMedicalImagingConferenceProc.
,vol.
4322,pp.
113–127,2001.
[12]M.
Kachelrie,D.
-A.
Sennst,W.
Maxlmoser,andW.
A.
Kalen-der,"Kymogramdetectionandkymogram–correlatedimagere-constructionfromsubsecondspiralcomputedtomographyscansoftheheart,"Med.
Phys.
,vol.
29,pp.
1489–1503,July2002.
hosthatch在做美国独立日促销,可能你会说这操作是不是晚了一个月?对,为了准备资源等,他们拖延到现在才有空,这次是针对自己全球14个数据中心的VPS。提前示警:各个数据中心的网络没有一个是针对中国直连的,都会绕道而且ping值比较高,想买的考虑清楚再说!官方网站:https://hosthatch.com所有VPS都基于KVM虚拟,支持PayPal在内的多种付款方式!芝加哥(大硬盘)VPS5...
今天看到一个网友从原来虚拟主机准备转移至服务器管理自己的业务。这里问到虚拟主机和服务器到底有什么不同,需要用到哪些工具软件。那准备在下班之间稍微摸鱼一下整理我们服务器安装环境和运维管理中常见需要用到的软件工具推荐。第一、系统镜像软件一般来说,我们云服务器或者独立服务器都是有自带镜像的。我们只需要选择镜像安装就可以,比如有 Windows和Linux。但是有些时候我们可能需要自定义镜像的高级玩法,这...
v5.net一直做独立服务器这块儿的,自从推出云服务器(VPS)以来站长一直还没有关注过,在网友的提醒下弄了个6G内存、2核、100G SSD的美国云服务器来写测评,主机测评给大家趟雷,让你知道v5.net的美国云服务器效果怎么样。本次测评数据仅供参考,有兴趣的还是亲自测试吧! 官方网站:https://v5.net/cloud.html 从显示来看CPU是e5-2660(2.2GHz主频),...
小米3手机真实图片为你推荐
浏览器哪个好哪个浏览器好用?集成显卡和独立显卡哪个好集成显卡和独立显卡的区别?哪个更好?985和211哪个好985大学好 还是211是什么意思小说软件哪个好用免费什么软件看小说全免费炒股软件哪个好用用手机股票软件哪个好手机杀毒哪个好手机杀毒软件哪个最好用红茶和绿茶哪个好红茶和绿茶哪个更好?美国国际东西方大学你知道美国哪些大学有预科课程吗qq空间登录不上qq空间登不进去 怎么办qq空间登录网址开通QZONe
备案未注册域名 出租服务器 美国翻墙 gateone 火车票抢票攻略 好玩的桌面 网站挂马检测工具 北京双线机房 域名和空间 vip购优惠 吉林铁通 免费php空间 97rb zcloud winserver2008 websitepanel 百度新闻源申请 美国西雅图独立 免费服务器 ssd 更多