diaphragm attenuation) and the anterior wall (due to attenuation from breast tissue) (1,2); such artifacts can lead toincorrect interpretation in these regions, particularly sincethe degree and exact location of attenuation artifacts canvary between individuals.

Several methods have been proposed to correct for theseartifacts by measuring attenuation in the chest with a transmission source (3—7),but these methods require expensiveadditionalhardwareand can require significant additionalimaging time. This paper proposes a method to estimateattenuation in the chest which does not require additionalhardware and adds only minimal imaging time. The lungregions are defined with use of an injection of @°@Tc-MAAand body outlines are obtained from off-peak images.

METhODS

CardilacEmIssion DataAn elliptical phantom with lung and cardiac inserts (Data Spec

trum Corp., Chapel Hill, NC) was employed for the phantomexperiments. The lung inserts were filled with damp Styrofoambeads, the heart was filled with a solution containing 0.5 mCi

@Tcpertechnetate and the rest of the phantomwas filled withwater to represent soft tissue. To emphasize the effect of diaphragm attenuation, the cardiac insert was fastened approximately 1.5 cm below its standard position, placing the inferiormyocardial wall below the lower lung boundary. Imaging wasperformedin64positionsover 180°from45°right-anterior-obliqueto 45°left-posterior-obliqueusing a rotatinggamma camera(Multi-SPECr,SiemensMedicalSystems,HoffmanEstates, IL),with a total acquisition time of 10 mm. To simulate a typical 180°cardiac acquisition, the extra projection images from this tripleheaded gamma camera were not utilized in the reconstruction.The projections were reconstructed using rampfilteredbackprojection; high frequency noise was decreased with use of postreconstruction three-dimensional Wiener filtering (8).

Patientcardiacimageswereobtainedusingourstandardclinical protocols on the same rotating gamma camera using a 20%windowcentered about either 70keV (@‘Tl)or 140keV (@mTc).Patientswere injectedduringtreadmillstress or dipyridimoleinfusion with either 3.0 mCi @°‘11or 7.0 mCi @°‘Tcsestamil,i;images were acquired following stress and again after a 3-4 hrdelay.Patientswhowerestudiedusing @°“Fc-sestamibiwerereinjected with a 25 mCi dose before the delayed images wereobtained. Reconstructionswere performed as described aboveusing 180 degrees of pr@ection dat@@

Theaccuracyof SPECTcardiacperfusionima@ngis impairedbyartifactsinducedbynonunffom@gamma-rayattenuation.Thisstudyproposesa methodto estimateattenuationinthechestofpatientsw@outthe additionalhardwareandexpenseof transmissionimaging.Methods:Afterthe standard2@@11or @rcsestamibidelayedimageswereobtained, @°@Tcrnacroaggregated albumin(MM) was injectedand dual-energySPECTacquisitionwas performedwithwindowscenteredat 140keVand94 keV.Lungcontourswereobtainedby thresholdingtheon-peak(140keV)reconstructions.Outerbodycontoursweredefinedfrom imagesproducedby reconstructionof the lowerenergyscatterwindowobtained ssmuftaneousiyatthetime of thelung(MM) irn@ng.Followingassignmentof standardattenuatlonvaluesto the lungandnonlung(softtissue)regionsattenuationcorrectionwasachievedbymeansofa modifiediterativeChangalgorithm.The resultswerequantitativelyevaluatedbyimagingofa cardiacphantomfilledwithuniformactMtyplacedina chestphantom.Sensitivitytothechoiceof lungandsofttissueattenuationvalues,the choiceof the thresholdusedfor lungsegmentation,anderrorsin registrationof the attenuationmapwere assessed. Results: App1@ationofthls technique in a chestphantomand in patientsimagedwith both @°i1@sestamibiresultedin improvementin artifactuallydecreasedinferiorwallactivitywithoutadverselyafFectingtheOtherWaIISTheresuftswerereletivelyinsensftivetocholceofvaluesforlungandsoft-tissueattenuation,lungthresholding,andsmall( I .3cm)registrationerrors.Conclusion:Thissimplemethodcorrectsfornonuniformèttenuationin males; studies are underwayto adaptthe methodto determinebreastcontourin femalesandto detem*@ethevalueof the methodin dinicalpractice.

Key Words: SPECT;attenuationcorrection;cardiacperfusionimaging;coronaryarterydisease

J NucIMed1995;36:506-512

ardiacperfusion imagingusing SPECF is useful in thediagnosis ofcoronaiy artery disease. However, study quality is impaired by significant artifacts induced by nonuniformgamma-rayattenuationin the chest. This can result inapparently decreased uptake in the inferior wall (due to

ReceivedJai. 14,1994;revIsionacceptedAug.22,1994.Forcorrespondenceor repririsco.@ JeroldW.Walls,M.D.,Mdlnckrodt

Inelhtuteof R&Iiology.510SouÃKingshi@iway,St.Louis,MO63110.

506 TheJournalof NudearMedicine•Vol.36•No.3 •March1995

Attenuation Correction in Cardiac SPECTwithout a Transmission MeasurementJerold W. Wallis, Tom R. Miller and Paul Koppel

Mallinckrodt Institute ofRadiolo@', Was!thzgton University School ofMedicine, St. LouL@,Missouri

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

IA

B

T\IUD

toyield lungsofthe appropriatesizeby processingofa cylindricalphantom of known dimensions.

In patients studied with @Fc-sestami1,i,the heart was visibleon the @°@Tc-MAAimages.Inorderto removethemyocardiumfrom these images, the sestamibi images obtained before MAAinjection were subtracted from the @Tc-MAAimages beforedeterminationofthe lungboundaries.Theproportionof sestamibiimages subtracted was chosen visually to substantially eliminatecounts in the region of the myocardium on the resulting images.Processinglung data from the phantom study differedslightlyfrom processing of patient lung data in one respect—there wassignificantlygreater attenuation of the phantom's water-filledlungs during the lung emission imaging, a first-order Chang correction (10)was appliedto the reconstructedlung data. The alternativeof addingradioactivityto Styrofoambeadsin the lungcavity was not chosen because of the difficulty in achieving auniform distribution of tracer.

Outer Body Boundaries. Outer body contours were definedfromimagesproducedbyreconstructionof a lowerenergyscatterwindowobtainedsimultaneouslyat the time of the lung (MAA)imaging (Fig. 1C—E).A 30% window centered at 94 keV wasempiricallychosentoyieldthebestdefinitionofthebodycontour.Ramp-reconstructedimageswere smoothedusinga tbree-dimensional Butterworthifiter (order5, cutoff 0.225/pixel) with subsequentthree-dimensional5 x 5 x 5 medianifitering.Thesefilteringoperationswere acceleratedusing three-dimensionalversions ofthemethodsof McClellan(8,11)andHuang(12),respectively.Asmallbackgroundregionexteriortothebodyonallsliceswasthenchosen manuallyto serve as a count reference.A thresholdwasthenchosen interactivelyto differentiatethe degreeof activityexteriorto thebodyfromthatwithinthebody;avalueof 1.4—1.7x the mean value in this backgroundregion on each individualslice was found to optimally distinguishthe body boundary.Thebody region was then automatically defined by the computer on allslices.Thisuseof a referenceregionallowedtheactualthresholdvalue to vaiy across slices; such variationwas found to be necessary because the source distribution from which scatteredcounts were emanating was not constant. Other edge detectiontechniquescouldlikelybe employedwithsimilarresults.Validation of outer body detection was performedby processing of thechest phantomwith comparisonto the knownellipticaldimensions of the phantom.

Attenuation Coefficients. For technetium-energy myocardialstudies,the broad-beamattenuationcoefficientfor water (0.13cm@') was assigned to the regions designated to be soft tissue,representedbyallthevolumewithintheouterbodycontournotinthelungs.A valueof .047cm' was usedforthelungsbasedonpublishedestimates(413) anda valueof .07cm@was usedforthe dampStyrofoam.Whenthalliumwas used as the perfusionagent,thesevalueswereincreasedby 20%basedon theratioofthe narrow-beamattenuationcoefficientsinwaterforthe 68—83keVHgx-raysfrom @°‘11andthe140keVgamma-rayfrom @Tc(14).The finalattenuationmapis shownin FigureiF.

Reconstruction and ProcessingAttenuation correction was achieved by use of an iterative

reconstruction algorithm. A modified version of the iterativeChangalgorithm(10)was employedwhichallowedforvariableattenuation, with the attenuation map utilized in the projectionstep of the iterativereconstruction.The depth-dependentthreedimensionalpointspreadfunctionwas not modeledin the projectionstepbecauseofthe increasedcomputationalcomplexity.Five

shownfor a singleslicein the cardiacregionof the chestphantom.TheWeinerfifteredon-peaktranswdalslicesof the lungs(A)arethresholdedto forma binaryimage(B);a lineprofilethroughthelungsIsalsoshown.TheOff-peaktranSaxlalslicesarereconstructedusinga relativetysmoothButterworthfilter (C)and then a medianfilterIsapplied(D).ThisimageisthresholdedtoformabinaryImage(E),whichis combinedwiththe earlierlungoutlinesto yieldthefinalattenuationmap (F).

DetermInation of AttenuatIon MapThechestattenuationmapconsistsprimarilyof regionsof soft

tissue (water) density and regions of lung. The spine does notinterfere with imaging in a typical 180°cardiac acquisition, and theremainingbones of the thoraxare relativelythin;they are notdistinguishablefromwaterdensityon transmissionimagesobtamed using a gamma-camera flood source (4). Thus, an estimatedattenuationmapcanbecreatedfromknowledgeof theouterbodycontour, lung boundariesand approximateattenuationcoefficients for lung and water. The steps in the generation of theattenuationmapareshownin Figure1.

LungBounda,ies. The lung contours were Obtainedafter injection of @“Tc-MAA(Fig. 1A, B). To determinethe marginsof thelungsin the phantomstudy, the Styrofoamwas removedaftertheemissiondatawereobtained,andthe lungcavitieswerefilledwitha solutioncontaining2 mCiof @9'cpertechnetate.Thephantomwasthenreplacedunderthegammacamerainapositionidenticalto thatusedfortheemissionimaging.Lungboundariesinpatientstudieswere obtainedby injectionof 4.0 mCi @°Tc-MAAaftercompletionof thedelayed @°‘Tlor @°‘Tc-sestamibiimageS.Carewas takento avoidpatientmotionbetweenthe delayedcardiacimagesand the lung images.Projectiondata were obtainedover360°using 90°angles, with a total acquisition time of6 mm. A 20%windowcentered at 140keV was used for the lung acquisition;simultaneous acquisition of a lower energy window was also performedto obtain the outer body contour, as describedbelow.

Thefirststep in dataanalysiswas reconstructionof the transaxial lung emission images, using the same techniques as emplayed for the cardiac emission data. Segmentation of the resulting imagesto extract the lungboundarieswas performedusingathresholdof35%of peaklungactivityfollowedbya three-dimensional dilation/erosionoperation (9). After thresholdingto producea startingmask,a dilation(whichincreasesthe regionsizebyone pixelin alldirections)was performedto eliminateanysmallholesin thelungregions,followedby anerosion(decreasingtheregion size by one pixel) to preserve the original geometry. Arelativelylow thresholdwas chosento permitareasof hypoperfusedlung to be correctly classified. This operation was confirmed

AttenuationCorrection in Cardiac SPECT •WaOisat al. 507

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

FiGURE2. FoUo@ngprocessingof the bodyphantomas describedabove,theestimatedattenuationm@Isshownat multiplelevels(top);for comparison,PE@transmissionImagesare alsoshown(bottom)at approodmatalythesamelevels.Sightirreguishtiesintheautomaticallydetectedcontourduetothespineandintherightchestwall (viewei@sleftycouldbecorrectedeitherautomaticaltyorwfthoperatorintervention,butwouldnotbeslgnfficantforatypical180°acquisthon.

body contour, evident in Figure 2. This did not affect thereconstruction as it was not within the projection path ofthe 180°emission data; if 360°acquisition were being employed, a structureof bone density could be automaticallyplacedin the midlineposteriorbody to represent the spine.The automaticallydeterminedbody contours in the limitedpatient studies correspondedwell with expected body outlines and with visually evident boundaries on the scatterimages.

Cardiac Reconstniction. Figure 3 shows bullseye plotsof the cardiac phantomcontaininguniformactivity imagedin the chest phantom. The clinical reconstruction (withoutattenuation correction) and the attenuation corrected reconstruction are shown. The apparentlydecreased uptakein the inferior wall (due to diaphragmattenuation) is notpresent after attenuationcorrection. Quantitative analysisshowed that the ratioofthe mean counts in the anteriorandposterior regions of the bullseye image improved from 1.17without attenuation correction to 1.02 following correction The uniformity of the myocardium as measured bythe relative s.d. of the counts in the bullseye plot improvedfrom 13.7% without attenuation correction to 8.6% withcorrection, corresponding to the visually more uniformactivity evident in the corrected image in Figure 3.

SensitivityAnaa@ysis. Figure 4 shows the effect of increasing offsets of the attenuation map on the reconstructedheart. It can be seen that a misregistrationerrorof4 cm canresult in artifactualnonuniformityin the myocardium of adegree approaching that seen without attenuation correction. Visually and quantitativelythere was minimalchangein uniformityofthe myocardiumwhen a 1.3 cm registrationerrorwas present,whilea 23 cmregistrationerror resultedin mild but definitenonuniformityin the appearanceof thebullseye. These findings are in approximate agreementwith data from PET imaging, where a 2 cm misalignmentwas judged to result in significant cardiac artifacts (16).Note, however, that the chest phantomemployed had a 1.5cm region of soft-tissue density between the heart and thelungs, as might be seen in an obese person with pericardialor epicardialfat. The requirementsfor alignmentare likely

•0iterationswere foundto be sufficientfor attenuationcorrection.Followingreconstruction, standard SPECF cardiac processingwasperformed,leadingto imagesin theshort,verticallong-,andhorizontal long-axis planes and a bullseye display (15). When thesame study was processed under varyingconditionsof attenuationcorrection, identicalcardiacrotationangles andbullseye samplingwas performed.

Error SensitivityMisregistration Error. To determine the effect of misregistra

tion errors between the attenuationmap and the emission data,reconstructionsof the chest phantomcontaininga plasticheartfilledwith uniformactivitywere performedfollowingshiftingofthe attenuationmapby 2, 4 and6 pixels,correspondingto 1.34,2.68and 4.02cm, respectivelyin the x (lateral),y (antero-posterior)andz (cranio-caudal)directions.Inadditiontovisualassessmentof themis-registeredreconstructions,theuniformityof activity in the reconstructed myocardium was quantified by creationof abullseyeplotanddeterminationof therelatives.d. (s.d./meancounts) of the bullseye counts. For this quantitative analysis theentirebullseyewas analyzedwith the exceptionof the extremebasal portion of the bullseye (the edge of the myocardium). Theratio of activity in the anterior and inferior wall (omitting thepen-apicalregionofthe inferiorwall)was alsocomputed;thiswillbe referredto as the anterior/posteriorratio.

Lung ThirsholdingError Se@&citivity.The same chest and heartphantomwasusedtoassesstheeffectofdifferentthresholdvaluesfordeterminationof lungboundaries.Choiceof differentthresholdsresultedin slightlydifferentlungsizesin the attenuationmap.Reconstructions were performed with threshold values of 25%,35% and 45% of peak lung activity. Following reconstructionswith each of the attenuation maps, bullseye displays were generated and the uniformityof activityin the myocardiumwas assessed as describedabove.

Sensitivity to E,ror in Ass4gned Attenuation COeffiCients. Although the attenuation coefficient of the lungs will ‘@raryslightlyfromone individualto another(6), a singleestimatedvalue isemployed in this technique. To assess the sensitivity to choice ofattenuationvalue, the lung attenuationcoefficient was varied inreconstructionsofthe chestphantomoverarangefrom0.0to0.13cm@‘,with assessment of the effect on the reconstructed cardiacimagesas describedearlier.The effectof variationof the softtissueattenuationcoefficientwas alsoassessed,withreconstructionsemployingvaluesfrom0.09cm' to 0.17cm'.

RESULTSPhantomStud@s

Contour Validation. The attenuation map following cxtractionof the lung and outer body contours and combination of these regions is shown in Figure 2 (top) for thephantom study. The majorand minor axes of the effipticalouter body contour were computed from the segmentedimages to be 30.8 cm and 22.1 cm, respectively; thesecompared to measurements of 31.5 cm and 235 cm in theactual phantom. To evaluate the lung contours, the segmented imagewas comparedto a transmissionimageof thephantom (obtained from a separate PET scan), shown inFigure 2 (bottom). As can be seen in the figure, the lungoutlinesare similarto those derivedfromthe transmissionstudy. The increased attenuation from the plastic spineresulted in a posterior indentation in the estimated outer

508 TheJournalof NudearMedicine•Vol.36•No.3 •March1995

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

14

;:13@ 12No

correctionI

.2

ii.'-°—--°@;-;:;:-7@@‘1\

@. i:@ i 0.9 Corrected

.@ 8 Corrected@ @- 0.8A 0 0.05 0.1 0.15 B 0 0.05 0.1 0.15

i&.IuDI(1/cm) @t.lung(1/cm)

*

AA2ttenuatlon0

2 4Map Shift(cm)0

@ G@@\

0B8@2O24Attenuation Map Shift(cm)—9--.

x Shift—4---Y ShiftZ Shift —0-—No Conecti@

14@ 1.2.2No correction@

@. @1.15 Nocoriection@12

NNm 1L1‘a p1.05

@10

@ a0.95@8

@ . . . . 0.9

A 0.09 0.11 0.@130'IS 0.17 B 0.09 0.11 0.13 0.15 0.17i&.sofltissue (1/cm) I@.softhun. (1/cm)

FIGURE 5. Graphsdepictingthe effectof vat@lngthe lungattenuationcoeffidentuponthereconstructedmyocardiumintheuniformcardiacphantomImagedinthechestphantom.(A)and(B)areasinFigure4.

The results following utilizationof differentlung attenuation coefficients are shown in Figure 5. While use ofextreme values such as those corresponding to water or airresults in significantnonuniformityof the cardiacbullseye,this data suggests that variation within two s.d. of an average lung attenuation value (0.045cm' s.d. .013) (13)would still result in substantialimprovementin myocardialuniformity. The effect ofchoice of different lung thresholdswas assessed; the relatively sharp lung boundaries, as cvident in Figure 1A resulted in only minimalchange in lungboundaiy position. Utilizing a value of 25% or 45% for thethresholdratherthanour standardvalue of 35%resulted inonly a 3 mm shift in lung boundaiy positions, and had nosignificant effect on the bullseye uniformity or anterior!posterior ratio. The effect of variation in the soft-tissueattenuation coefficient is presented in Figure 6. Minorchanges in the effective soft-tissue attenuation might beencountered due to varying degrees of scatter in obesepatients. Errorsof 10%in the soft-tissue attenuationcoefficient had minimal effect on the myocardium, and errors of30% still resulted in more accurate reconstructions thanthose obtained without attenuation correction.

Patlent Stud@Figures 7 and 8 illustrate the estimated attenuation maps

and the reconstructed clinical images both with and without attenuationcorrection in two patients, one imagedwith

phantomimagedIn the chestphantom,withreconstructionbothudthotd@A4andwfth(B)attenuationcorrection.

to be slightly more demandingin a thin patient, in which a1.3 cm shift would move the heart substantially into thelung region.

It canbe seen in Figure 4Athat a —1.3-cmshift in theydirection actually improved cardiac uniformity slightly.While it is conceivable that this represents a registrationerror, resulting from repositioningof the phantom afterifiling the lungs with activity during our experiment, it ismore likely that a slight offset in one of the six directionshappenedby chance to slightly improve the nonuniformityarising from other sources (e.g., partial volume effect,depth-dependent resolution). It can be seen in Figure 4Bthat the anterior/posterior ratio is most accurate whenthere is no z-axis shift, appropriatelyplacing the inferiorwall in the abdomen and the anterior wall in the chestdensity portions of the phantom; significantmovement cither in the cranial or caudal directions places these twowalls in the same attenuation region, lessening the effect ofthe correction. Motion in the x and y directions slightlyincreased or decreased the effect of the correction by placing the heart closer or furtherfrom the chest wall.

FiGURE4. Theeffectofmisregistrabonoftheattenuationmapinthex,yandzdirectionsupon(A@theuniformityofthereconstructedrnyocardlum as assessed by the standard deviation of the pbcelsInthe bullseye(seetext)and (B)theanterior: posteriorratio.Theuniformcardiacphantomwasimagedin thechestphantom.Forreference,thevaluescorrespondingto theuncorrectedmyocardialreconstru@onswe alsoshown.

FIGURE6@Theeffectof variationin thesoft-tissueattenuationcoefficientuponthereconstructedmyocardium.(A)and(B)areasInFigure4.

AttenuationCorrectionin CardiacSPECT•Walksat al. 509

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

DISCUSSIONA B

Importance of Attenuation CorrectionAttenuation effects hinder interpretationof both planar

and tomographic myocardial perfusion studies. Attenualion by the diaphragmresults in significant inferior wallartifacts in planar perfusion imaging, leading to the clinicalpractice of right lateral decubitus imaging for the steepoblique view (17,18). Attenuation due to the diaphragm isalso present in SPECF imaging performed in the supineposition (2), and can result in erroneous interpretationofdecreased activity in the inferior wall. The degree of attenuation effect can vaiy from one individual to another, making it difficult to ascribe a fixed inferior wall defect toattenuation or to infarction in a particular patient. Alternative approaches, such as prone imaging (19,20), have beenproposed to partially alleviate this problem, but such methoth can lead to anterior wall artifacts and have not becomepopular in clinical practice.

Similarly, attenuation from breast tissue is a problemboth in planar (21,22) and SPECF (2,23) imaging ofwomen. In planar imaging the breast contours are frequently visualized, allowingcorrection for the effect ofattenuation during image interpretation.While the abnormalities in SPECF reconstructions are predominantly inthe anterior region, it is often difficult to determine if aparticular abnormality is due to attenuation artifact or tocoronaly disease.

Methods of Attenuation CorrectionSeveral methods have been proposed for attenuation

correction in SPECF. These methods involve determination of the distribution of attenuation coefficientsin thechest (the attenuation map). The attenuation coefficientshave most frequently been determined by a transmissionmeasurement. Focused collimators with point or linessources, parallel collimators with flood or moving linesources and an uncollimated gamma camera with a pointsource have all been used (3—7).While these methods aregenerally effective, there are associated difficulties such as:the mechanical apparatus can be awkward and expensive;use of focused collimators may result in truncation artifacts; use of different energies for transmission and emission data can result in incorrect attenuation values; andscatter from simultaneous presence of both the emissionand transmission sources can degrade the reconstructions.

Other methods, including the one proposed here, estimate ratherthan measure the attenuationmap. Data fromcomputed tomography(CT) have been used for this purpose (24), with registration of the transmission data withthe SPECT emission study priorto attenuationcorrection.A recent paper (25) also proposed using lung perfusionimaging for definition of lung outlines, with use of a radioactive elastic bandage to mark the outer body contourOther methods to define a convex outer body contour fromprojection data have also been developed (26-28). Assignment of noise-free attenuation coefficients to segmentedtransmission images has been proposed in both SPECF

9)

E

FIGURE7. Imagesfroma cardiacstudyempb@ng@°iiii apatlentwltha largeanteroseptalandapicalInfarction.(A)attenuationmapusedinclinicalreconstruction(I.e.none).(BandC)Mid-cardIacshort-a,dssheaand bulseyeplotwffl@otdattenuationcorrection,showing inferior attenuation artIf@ as well as the apical Infarct (D)attenuationmapfroma mid-cardiactransadalslice,showingbodyoutlineandlungs;thedomeoftherightheml-dlaphragmisevidentatthislevel.(EandF) Mid-cardiacshort-adssliceandbullseyeplotfolbv@ngattenuationcorrection.

201T1(Fig. 7) and one with @“Tc-sestamibi(Fig. 8). Thefirst patient had a large anteroseptal and apical myocardialinfarction with extensive disease in the left anterior descendingcoronaiy arteiy. There was no evidenceby ECOor catheterization of coronaiy disease in the inferiorwall.The second patient underwent a submaximal treadmillstudy for evaluation of syncope. In this patient with a lowclinical suspicion of coronaiy disease, a mild reversibleseptal defect on the stress images was felt to be due toexercise-inducedleft bundle branch block, and the fixedinferior defect was attributed to diaphragm attenuation.Note the improved appearance of the inferior and inferoseptal walls following attenuation correction in both ofthese patients with no evidence of coronaiy arteiy diseasein this region. The areas of known abnormalitieswere notaffected by the attenuation correction.

FIGURE8. Imagesfroma cardiacstudyempIo@4ng @cseatsmibi,Ina patientwithlowclinicalsuspicionof coronaryarterydisease,noprevioushlsto@yof Infarction,anda mildfixedInferiorwaltdefectconsistentwithdiaphragmattenuation.Imageformatas InFigure7.

A B

*

EI

2

510 TheJournalof NudearMedians•Vol.36•No.3 •March1995

:.Q

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

Protocol1Protocol2Standard

InItialImagingStandard InitialImagIngln@on of lowdose°@‘Tc-MMD_ImagIngStendard

@ ImagIngStendwd deiayedImagIngInjectionof °9m•ro@MMInjectIon of higherdose@°“Tc-MMDua@energy

ImagIngDual-energy ImagIng

(29) and PET (30,31) to decrease noise propagation fromthe transmission to the emission reconstructions.

All of these methods employ an iterative reconstructionalgorithmto correct for the variableattenuation,since thecommonly used first-order Chang correction (10) is onlyapproximate, and its accuracy diminishes in the setting ofnonuniformattenuation. In our experience,both iterativelikelihood-based techniques employing the expectationmaximization algorithm (32,33) and techniques using iterative ifitered backprojection (34) result in adequate compensation for attenuation after 5—10iterations. While it ispossibleto modelthe point-spreadfunctionduringiterativereconstruction to improve image resolution, as has beendone both by us (34,35) and others (36—38),this was notdone in this investigation to decrease the computationalrequirements.

Comparison of ThIs Method to Use of aTransmIssion Source

The principal advantages of this method of attenuationcorrection lie in two areas: (a) simplicity and cost and (b)image quality. The method is indeed simple, requiringonly5—10min of additionalcamera time (withoutmanipulationof apparatus such as collimators or flood sources) andsemiautomatic additional image analysis. The cost of thisapproach is negligible, consisting only of the expense of a4 mCi dose of @“@Tc-MAAand camera time. These factorsare in contrast to the effort and expense associated withspecialized collimators and flood sources required fortransmission measurement. The added patient radiationdose is only 7%—11%of the @°‘Tlor Tc-sestamibi dose.

Issues of image quality are more complex. The methodpresented here could be criticized as being approximateandinexact, especially in comparisonto measurementwithtransmission sources. The tradeoffs between the two methods may not favor the transmission approach. While theaccuracy of the method described here may be less thanthat of transmissionmeasurement, our sensitivity analysisindicates that anticipatederrors in the value of the attenuation coefficients due to normalpatientvariationwill onlyweakly affect the results. In both this method and whenutilizing transmission measurements, care must be taken tokeep the emission data and attenuation map aligned. Thetransmission measurement may lead to loss of contrast inthe emission data due to scattered counts when the floodsource is imaged simultaneously with the emission data.Conversely, the emission gamma rays may interfere withthe transmissionmeasurement. The method proposed herehas no such contrast loss or transmission interferencebecause the @‘@Tc-MAAis injected only after the emissiondata are obtained. Furthermore,a noisy transmissionmeasurement will propagate noise into the reconstruction ofthe emission data. The method proposed here involves nonoise propagation, although there may be a slight offset orbias of counts from the truevalue. Thus, compared to themethod presented here, the transmissiontechnique mayyield slightlymore accurate attenuationcorrection,but in

TABLE IImagingProtocolsfor EstimatedAttenuationCorrection

exchange for reduced image contrast and increased noise.Further studies will be required to determine whichmethodyields better clinicalresults.

Imaging In Patients with Lung DiseaseGeneration of the body outlines by use of scatter from

the lungs following @Tc-MAAinjection was found to beaccurate in men where the scatter source is close to thechest wall. However, in women the breast marginmay besignificantly farther from the lungs than is the chest wall inmen and the body marginis not convex, thus making thismargin difficult to distinguish on scatter images. It may benecessary to utilize other methods to create an accurateattenuation map in women. A radioactive binder, such asthatproposed by Madsen (25), would reliablydelineate theanteriorchest wall, but still would not accurately portrayconcave portionsof the body outline. Alternatively, a scattering source outside the body, such as a string markerplacedalongthe sternum,wouldaugmentthe scatter fromthe lung activity. Furtherwork in this area is ongoing.

The method assumes that the distributionof lung perfusion adequately reflects the air-filledportions of the lung;this may not be the case in patients with chronic obstructive lung disease who may have nonperfused, air-filled lungregions. To minimize the effect of heterogeneous perfusion, a relatively low threshold was utilized for lung segmentation (so that areas of moderately hypoperfused lungwould still be segmented as lung), and a sequence of dilation/erosion steps was employed in the segmentation process to fill in any remaining small defects in the segmentedlung region.

Imaging ProtocoisWe envision two possible imaging protocols. Concern

over loss of image contrast from scattered lung counts hasled us to inject the @‘@‘Tc-MAAonly after all emission datawere obtained (Table 1, Protocol 1). There is no loss ofcontrast due to scattered counts from the @‘Tc-MAAbecause the @“Tc-MAAis injected only after all emissiondata have been collected. The disadvantage of this approach—that the first set of images are not corrected forattenuation—maynot be a critical issue. For example, if afixed defect is seen in the inferior wall that disappears afterattenuation correction of the second set of images, thedefect is likely a diaphragmartifact. Otherwise, the defectlikely is due to coronaiy disease.

The second protocol (Table 1, Protocol 2) yields atten

AttenuationCorrectionin CardiacSPECT•Wallisat al. 511

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

uation maps for both sets of perfusion images and shouldwork well when a one-day Tc-sestamibi study is performed. Followingthe first low dose of sestamibi (e.g. 7mCi), 1.5 mCi @‘Tc-MAAis injected and an attenuationmap is created. The second dose of sestainibiwifi typicallybe much larger(e.g. 25 mCi), thus overwhelming the scatter from the partly-decayed first MAA dose. Finally, asecond 2.5 mCi @9@c-MAAinjection is made to producethe second attenuationmap. Ourconcern with scatter in a201Tl study may be excessive (25); it is possible that Protocol 2 (Table 1)will yield acceptable resultswith @°‘Tl,andfurther study is warranted. Reduction of the first @9'c-MAA dose to 1—2mCi may yield acceptable lung boundaries while leading to a barely-detectable loss of contrast inthe delayed @‘°‘Tlemission images following a 3—4hr decayinterval.

cONCLUSIONS

In summaiy, a method is described that yields attenuation maps for use in correction of myocardial images. Thissimple, inexpensive method contrasts with most reportedtechniques that involve expensive and potentially cumbersome transmission measurements. While the anatomic accuracy of the method described here may be slightly inferior to that achieved with the transmission approach,results presented here show good correction for nonurnform attenuation and compare favorably with publishedreports employingthe more complex techniquewhile offeringpotentialadvantagesof superior imagecontrast andlower noise. Studies are underway to determine the valueof this method in clinical practice.

ACKNOWLEDGMENT

This work was supported in part by a grant from SiemensMedicalSystems, HoffmanEstates, IL.

REFERENCES1. ManglosSH,ThomasFD,GagneGM,HdllwigBJ.Phastomsmdyofbreast

tissueattenuationin myocardialimaging.INudMed 199334:992—996.2. DePueyEG,GarciaEV.Optimalspecificityofthallium-201SPECTthrough

reco@on of imagingartifacts.INuciMed 1989,30'.441-449.3. ManglosSH, BassanoDA, ThomasFD. Cone-beamtransmissioncorn

puted tomography for nonuniform attenuation compensation of SPEC'I'images.JNUCIMt'4199132:1813—1820.

4. Tsui BMW, Guilbeig GT, Edgerton ER, Ballard JG, Perry JR, McCartneyWH, BergJ. Correctionof nonuniformattenuationin cardiacSPECI@ünaging.JNUCZMe41989@,3O497-5O7.

5. Frey EC, Tsui BM, Peny JR. Simultaneousacquisitionof emissionandtransmissiondatafor improvedthalliurn-201cardiacSPECFimiigingusinga technctiurn-99rntransmissionsource.JNUC1Me4 199233:2238-2245.

6. Manglos SH, Bassano DA, Thomas FD, Grossman ZD. Imaging of thehumantorsousingcone-beamtransmissionCT implementedona rotatinggammacamera.JNUCIMe4199233:150-156.

7. TungCH, GullbergGT, ZengGL, OuistianPP.,DataFL, Morgan1fF.Nonuniformattenuationcorrectionusingsimultaneoustransmissionandemissionconvergingtornography.IEEETrai&sNuclSci199239:1134-1143.

8. Miller TR, WallisJW, WilsonAD. Interactivereconstructionin singlephotontornography.EurlNuclMed 1989;15:189-193.

9. Serra J. Image ana'ysis and mathematical moq@zoIog3@.New York: Acadernic Press; 1982.

10.changLT. A methodforattenuationcorrectioninradionuclidecomputedtomography.IEEETransNuci Sd 1978;NS-26:638-643.

11.McQellanJH,ChanDSK.A two-dimensinnalFIR filterstructurederivedfromtheChebyshevrecursion.IEEETransCircuitsSystems1977;7:372-378.

12.HuangIS, YangGJ,YangGY. A fast two-dimensionalmedianfilteringalgorithm.IEEET@w&cAcoustSpeechandSWPTOC1979ASSP-27:13-18.

13.RowellNP, GlahoimJ, FlowerMA, CroninB, McCreadyVR. Anatomicallyderivedattenuationcoefficientsfor usein quantitativesinglephotonemissiontornographystudiesofthe thorax.EuriNuclMed 1992;19:36-40.

14.SorensonJA, PhelpsME. Physicsin nuclearmedicine,secondedition.Orlando:GrimeandStratton,inc.; 1987:AppendixD.

15.Miller1k, StarrenJB,GrothePA. Three-dimensionaldisplayofpositronemissiontomographyof theheart.JNUC1Med1988;29:530—537.

16.McCOrdME, BacharachSL, BonowRO,DilsizianV, CuocoloA, Freedman N. Misalignment between PET transmission and emission scans: itseffect on myocardial imaging.INuci Med 199233:1209-1214.

17.JohnstoneDE,WackersFiT, BergerHi, Ctat.Effectofpatientpositioningon left lateralthallium-201myocardialscans.JNuclMed 197920:183.

18.GordonDO, PfistererM, WilliamsR, CtaL Theeffectof diaphragmaticattenuationon @“11images.ClinNUCIMed1979;4:150.

19.Kiat H, VanTK, FriedmanID, CtaL QUantitatiVestress-redistributionthallium-201SPECFusingproneimaging:methodologlcdevelopmentandvalidation.JNUCJMed199233:1509-1515.

20. SegallGM, DavisMJ.ProneversussupinethalliummyocardialSPECF:amethod to decreaseartifactual inferior wall defects. I Nuci Med 198930:548—555.

21. GoodgoldHM, RehderjG,SamuelsLD,ChaitmanBR.Improvedinterpretation of exercise20!fl@ p@f@ @j@gy in women: characterization of breast attenuation artifacts Radiology 1987;165:361-366.

22. StolzenbergJ, KaminskyJ. Overlyingbreastasa causeof false-positivethallium scans.Clin NUCIMed 1978;3:229.

23. Garver PR, Wasnich 1W, Shibuya AM, Ych F. Appearance of breastattenuationartifactswith thaffiummyocardialSPECFimaging.ClinNuciMed1985;1O:694—696.

24. FlerningiS.A techniquefor usingCFimagesin attenuationcorrectionandquantificationin SPECF.NUC1MedCommun1989;1O:83—97.

25. MadsenMT, KirchnerFF,EdlinJ,NathanM, KahnD. An emissionbasedtechniquefor obtainingattenuationcorrectiondatafor myocardialSPECFstudies.NUCJMe4Commun1993;14:689—695.

26. YounesRB, MasM, BidetR. A fully automatedcontourdetectionalgorithm: the preliminaiy step for scatter and attenuation compensation inSPECF. EurlNuclMed 1988;14:586-589.

27. Hosoba M, wani H, Toyama H, Murata H, Tanaka E. Automated bodycontourdetectionin SPECF:effectson quantitativestudies.I Nuci Med198627:1184—1191.

28. Gullberg GT, Malko JA, Eisner RL Boundary determination methods ofattenuationcorrectionin 5PECF.In: EaserPD, ed.Emissioncomputedtomogrupljy:cwwztt7rndc. NY: SocietyofNuclear Medicine; 1983:33-53.

29. GaIt JR, cullorn SJ, Garcia EV. SPECTquantification: a simplified methodof attenuationand scattercorrectionfor cardiac imaging.I New! Med199233:2232-2237.

30. Meikle SR, Dahibom M, Cheriy SR. Attenuation correction using countlimited transmission data in positron emission tomography. I New! Med199334:143—150.

31. XuEZ, MullaniNA, GouldKL, AndersonVIL. A segmentedattenuationcorrectionfor PEF.JNuclMed 199132:161-165.

32. LangeK,CarsonR.EMreconstructionalgorithmsforemissionandtrans.mission tomography.I CompAssirt Tonaogr1984;8:306-316.

33. MIllerMI, SnyderDL, MillerTR. Maximum-likelihoodreconstructionforsingle-photonemissioncomputedtomography.IEEETransNuciSa 1985;NS-32:769-77&

34. Walls 3W, Miller TR. Rapidly converging iterative reconstruction algorithms in single.@on emissioncomputed tomography.JNucIMed 1993;34:1793-1800.

35. chornoboyES,ChenCJ,MillerMI, MillerTR,SnyderDL Anevaluationofmaximumlikelihoodreconstructionfor SPECF.IEEETransMedImag

@ 19909:99—110.

36. Zeng GL, Gullberg GT, Tsui BMW, Teriy JA. Three-dimensional iterativereconstructionalgorithmswith attenuationandgeometricpoint responsecorrections.IEEETransNud Sd 199138:693-702.

37. PenneyBC,KingMA.A pr@ector,back-pr@ectorpairwhichaccountsforthe two-dimensionaldepthanddistancedependentblurringin SPECF.IEEETransMw! Sd 1990@,37:681-686.

38. LiangZ, TurkingtonTO,GillandDR,JaszczakRi, ColemanRE.Simulta.neouscompensationfor attenuation,scatterand detectorresponseforSPECFreccmstnictionin threedimeasions.PI@uMedBIOI199237587-603.

512 TheJournalof NudearMedians•Vol.36•No.3 •March1995

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from

1995;36:506-512.J Nucl Med.   Jerold W. Wallis, Tom R. Miller and Paul Koppel  Attenuation Correction in Cardiac SPECT without a Transmission Measurement

http://jnm.snmjournals.org/content/36/3/506This article and updated information are available at:

  http://jnm.snmjournals.org/site/subscriptions/online.xhtml

Information about subscriptions to JNM can be found at:  

http://jnm.snmjournals.org/site/misc/permission.xhtmlInformation about reproducing figures, tables, or other portions of this article can be found online at:

(Print ISSN: 0161-5505, Online ISSN: 2159-662X)1850 Samuel Morse Drive, Reston, VA 20190.SNMMI | Society of Nuclear Medicine and Molecular Imaging

is published monthly.The Journal of Nuclear Medicine

© Copyright 1995 SNMMI; all rights reserved.

by on March 24, 2018. For personal use only. jnm.snmjournals.org Downloaded from