ciple, our method is similarto thatproposed by Koral et al.(11). Their method employs the acquisition of spatiallydependent energy spectra and estimates scattered photoncomponents for each region on the basis of the energyspectra acquired. Unlike their method, however, ourmethod does not requirethe acquisitionof fullenergy spectra, but only utilizes total values for three energy windowsof the spectra.

We performeda physical evaluationof the TEW methodusing phantoms and have also applied this method to apatients in a clinical trial.

MATERIALSAND METhODS

All images in this investigationwere acquired with a threeheaded, rotating camera dedicated SPEC!' system GCA9300A/HG(ToshibaAmericaMedicalSystems,Tustin, CA).

TEW Method for Compton Scatter CompensationFigure 1 shows the energy spectra of primaryand Compton

scatteredphotonswhichwere simulatedfora @“@Tclinesourceatthe center of a water-filledcylindricalvessel with a 20-cm diameter.Theenergyspectrumis estimatedat the centralaxisusingthe MonteCarlomethod(14). In the previouspaper(10), welocatedthe centerof a scatter rejectionwindowat each endof themain window. The scatter rejection windows and the main windowthusoverlapeachotherby a widthof Ws/2.Inthephantomand clinicalstudies, it is impossibleto acquireusingoverlappedwindows. For clinical implementationof the TEW method to agamma camera system, we removed the overlap by locating thesubwindowsadjacentto the mainwindow.We obtainedthe appropnate locationandwidthofthe scatterrejectionwindowbasedon the analysisof a numericalphantomgeneratedby the MonteCarlo method (15).

Imagequalitywas comparedusingthe root meansquare errorof SPECTvaluesbetweenthe imagereconstructedfromprojectionsof primaryphotonsandfromthe correctedprojectionsofscatteredphotonsusingthe TEWmethod.Basedon theseresults,thecombinationof a widthof 24%forthemainwindowanda 3keV scatterrejectionwindowseems appropriatefor @“@Tc(141keVgammaray).Thesewidthschangedependingon theenergyof the primaiy photon.

We obtainedthe countsfromthese threewindowsand estimatedthe amountof scatteredphotonsmixinginto the mainwindow.We employeda trapezoidalapproximationfor this esti

Thespatialdistributionofscatteredphotonsvanes dependingonmany factors suth as oL@ectsize and source distribution. Wepropose a th@-energy window (TEW) scatter compensationmethod for determining position-dependent Compton scatter.We estimatedthecountofprimaryphotonsateachpixelintheacquired image using the 24% main windowcentered at thephotopeak energy and 3 keVscatter rejectionwindowson bothsides of the main window.We conducted a physicalevaluationof thismethodusingphantomsandalso appliedthismethodtopatients in a dinical trial. The TEW method performed Comptonscatter compensationwithgood accuracy.

J NuciMed1993;34:2216-2221

everal factors adversely affect quantitative analysis ingamma camera-based SPECT'. Among these, the most significant are Compton-scattered photons mixing into thepreset energy window and the attenuation of photonswithin the human body. Accordingly, several techniqueshave been proposed for Compton scatter compensation(1—9).In some reports, a Monte Carlo modeling techniqueis employed to study the distributingproperties of scattered photons.

According to these reports, photons generated withinthe patient are scattered, which affects their direction andenergy. These scattered photons then mix into the mainenergy window. Most of the scattered photons in the mainwindow are single-scatter photons. Furthermore, singlescatter photons vary depending on scatterer distributionand the position ofthe radioactive source. Consequently, it isnecessary to estimate single scatter intensity for each pixel.

We conducted a simulationanalysis of scattered photoncomponents mixed into SPECT images by Monte Carlomodeling and proposed a triple-energy window (TEW)compensation method (10) for scattered photons. In pm

ReceivedFeb. 10, 1993;revIsionacceptedAug.5, 1993.Forcorrespondenceorreprintssor*artTsic@Iththara,ToshibaNaruWorks,

NudearMedicineSysterrsEnglneedngDept,1385Shimoishigoni,Otawera-shi,Tochigi-ken,324J@i.

2216 The Journalof NudearMedicine•Vol.34 •No. 12 •December1993

Compton Scatter Compensation Using theTriple-Energy Window Method for Single- andDual-Isotope SPECTTakashi Ichihara, Koichi Ogawa, Nobutoku Motomura, Atushi Kubo and Shozo Hashimoto

ToshibaNasu Workr,Nuclear Medicine Systems EngineenngDepartment, Shànoiihigam4Otawara-shi, Tochigi-ken@Japan; Department ofElectiical Engineethz@ College ofEngineering Hosei University, Koganei-sh4 Tokyo, Japan; andKeio University School ofMedicine, Tokyo, Japan

Eq.1

TEW scatter compensation can be acquired by any gamma camerasystemusingthethree-or two-windowacquisitionmode.

IfeachPMTandeachcameraareahasslightoffsets,countsinsubwindowsare unstableand noiseincreasesin estimatedscatterimages. To avoid these phenomenon,we developed a gammacamera stabilization unit OFFOTUNE'. Unifonnity and energyspectrum peak fluctuations were stable using the stabilization unitfortheTEWmethod.

Phantom S@dyEvaluation of TEW Method Using Hot Rods with a 10-mm

Diameter. To evaluate the proposed scatter-compensationmethod,we usedhotrods10mmindiameterinacylindricalwaterpoolphantomwitha20-cmdiameter.Tenhotrodswereifiledwithdifferentactivitiesof @°‘Tl@‘Tcor ‘DI.In case of single-isotopeSPECT,we usedsix rods to evaluateTEWscattercompensation.Theotherscontainedno isotope.To measuretheactivityof therods with accuracy, each rod could be taken out of the cylindricalpool phantom. Technetium-99m was to be used for single-isotopestudies.In thispaperwe evaluatetheTEWmethodusing @“@Tcwithsingle-isotopeSPECF.For @°‘Tland‘@Imyocardialstudies,the combinationof blood flow (2°―fl)and metabolism(‘DIBMIPP)or the combinationof bloodflow(@°‘Tl)and the statusofsympatheticinnervation(‘@I-MIBG)were considered for dualisotope SPEC!'. We tested the applicability of TEW method todual-isotopeSPECTusingtenrods.

WeusedInoue'sreconstructionmethod,includingattenuationcompensation (12). In the uniform attenuation phantom, the validity of reconstructed SPECF values was assessed. A 10-mmdiametercalibrationsourcewasusedtoacquireSPECFdatainairand its activitywas measured.The SPECTvalue was measuredwithina circularROl with a radiusmeasuring[rod radius+(FWT@Mof SPED.' image)12]. We then computed the cross-calibrationcoefficient ([mCi][secj/[SPECFvaluej).

Weusedtheattenuationcoefficientofwater(0.15cm' forthe140 keV gamma ray of@Tc; 0.146 cm' for the 160 keV gammarayof ‘DI;0.187cm@'forthe70keVx-rayof @°‘Tl).

WedefinedSPEC!'valuesas countsof a reconstructedimagepixel. The SPECF value is proportionate to the counts of projection images used for reconstruction.After scatter and attenuationcompensation, the SPECT value is proportionate to the true activity of the corresponding position of the radioactive source.

Using 2o@11,‘@Iand @Tc,cross-calibratedSPECT valuesversus activitywas plotted. Scatter compensationin SPECTcanbe verifiedby thegraphbelow.The radioactivityvaluesof the hotrods are plotted on the x-axis. Cross-calibrated SPECF valuesafteraccurateattenuationcompensationareplottedon the y-axis.InSPECTimages,if scattercompensationis perfect,thecounts(SPECF value) of a pixel, which has no activity, should be zero.After attenuation compensation, the counts (SPECT value) ofpixels with activity are proportionate to the true activity of thecorrespondingpositionof the radioactivesource. After crosscalibration is performed, the SPECT value should be equal to thetrue activity of the corresponding position of the radioactivesource. If scatter compensationwas not done for the SPECTimagesbut was done for the cross-calibrationsource images,theslope deviatedfrom 1 and the y-interceptwas not zero. On theotherhand,if scattercompensationwas done for the SPECFimagesand not donefor the cross-calibrationsource images,or ifa bad cross-calibrationcoefficient([mCi][secj/[SPECFvalue])was used, the slopedeviatedfrom1but the y-interceptwas zero.In thisprocess,accuracyof theslopevaluestronglydependson

5000

4000

600

500

400

300

200

100

60

FiGURE 1. PositIon-dependentCOmptOn-ScattercompensationmethodçFEWmethod)and a simulatedenergyspectrumof @Fcusing MonteCarlo modeling.This figureshows the total energyspectrum (rotal), the first to the fourth order scattering (1st Scatter, 4th Scatter), and the sum of allorders of scattering (ScatterSum).Mostof the scatteredphotonsinthe mainwindowarefirstorderscatteredphotons.To reducescatteredphotonsaccurately,itis necessaryto estimatethe scattersum Inthe mainwindow,whichis shownas a shaded thangleusingthe counts in narrowscatterrejection windows A and C, adjacent to the main window.

mate.For @Tc,Figure1 showsthe totalenergyspectrum(Total), the first to the fourthorder scattering (1st Scatter, . . . , 4thScatter),and the sumof all ordersof scattering(ScatterSum).Mostof the scatteredphotonsin themainwindowarefirst-orderscattered photons. To reduce scattered photons accurately,it isnecessary to estimate the scatter sum in the main window, which isshown as a shadedtriangleusingthe counts in narrowscatterrejectionwindowsA andC, adjacentto themainwindowinFigure1.

Thecountsandwindowwidthsaredefinedas follows:

cleft = Counts in scatter rejection window A.Cright Counts in scatter rejection window C.

W, = Widthof scatterrejectionwindow.Wm Width of main window.Clotal Counts in main window.

Cscat = Counts of scattered photons mixed in main window.

Cpnm Counts of non-scattered photons in main window.

Then

“CleftCright―ø@WmCscator(@.@. W,/ 2

Thus, the primaryphotons for each pixelcan be calculatedfromthe followingequation:

C@=c;o@-Cscat.

We definetheTEWmethodas Equations1 and2. TEWscattercompensationis postprocessing.This methodcan applyto notonly SPECT imaging but planar and dynamic imaging. Data for

Eq.2

2217Compton Scatter Compensation•ichiharaat al.

450———4nn___,,@u

@uu————@——n@I

ICieft 0 \24

FiGURE 2. Examp@ of scatter compensation for myocard@SPECT images.

determinationof the calibrationcoefficient.Basedon this, a y-intercept of roughlyzero indicatescorrect scatter correction.

P@rlimina@y ClinicalEvaluation of@'―Tc-Tetrofosmin. Follow

ing administrationof 370 MBq (10 mCi) of tetrosfosmin, imageswere acquiredin continuousrotatingmodeusinga high-resolution,parallel-holecollimatorwitha 128x 128matrix.Imageswereacquired at 30 sec/projection for a total imaging time of 15 mm.Prejection data were acquired for two windows. The main windowwas 24%;the lowerscatter rejectionwindowwas 3 keV and theupperscatterrejection(C@J countwas assumedto be zero.

ScattercomponentswereeliminatedforeachpixelaccordingtoEquations1and2. Thisprocesswasexemplifiedby the projectiondata near the LAO 60 position and was shown in Figure 2. Theprojectiondata acquiredby the mainwindowandscatterrejectionwindowswere subjectedto Butterworthfilterprocessingbeforereconstruction, as shown in Figure 2. The cutoff frequencies ofthe filter for these two images were 0.2 and 0.1 [cycles/pixel],respectively, with an order of 8. We applied the Butterworth filterto remove high-frequencycomponents. Frequency componentshigherthan the camera resolutionwere eliminatedas noise components. Different cutoff frequency values were applied to mainwindow images and scatter rejectionwindow images. The cutofffrequencies for scatter rejection window images were definedbased on the investigationof spatialdistributionof multiple-orderComptonscatter by Floyd et al. (14,16).

I.2

0C)w 2@> --

w

eFIGURE 3.Profilecurvesalonglinea-f.

2218 TheJournalof NudearMedicine•Vol.34 •No. 12 •December1993

O@@a@ o@

I I x@+Ot@*@)@I@ I@$ a@;.@1i(&o@)I @5IftIa1I(8,O.1O)

—

Acquisition was performed under the following conditions:Cri.ght 0, Wm 24% and W@ = 2.14%. Therefore, Equation 1 can

be written as follows:

Eq.1'

TheC@ imageswerecalculatedfromEquation1'. As showninEquation2, the scatteredphotonimagewas subtractedfromtheprojectionimageof the main window.The profilecurves alongline c-f are shown in Figure 3. Image reconstructionwas doneusingfilteredbackprojectionand attenuationcorrectionwas notperformed.

RESULTS

Evaluation of TEW MethodTo verify the proposed method of scatter compensation,

we performed SPEC!' acquisition using @°‘Tl,@@uTcand1@I and a 10-mm diameter hot rod phantom. After scatter

compensation, we used Inoue's method for attenuationcompensation. Figure 4 shows the SPECF results obtainedfor six hot rods of @“Tc.The numbers 1—6correspond tothe location of the hot rods. The relative SPEC! valueswere plotted on the y-axis, and the radioactivityvalues ofnucides sealed in the rod were plotted on the x-axis. TheSPECF values after scatter compensation are located veryneara straightline which passes roughlythroughthe originof the graph. Compared to the results obtained withoutcompensation, the degree of improvement is remarkable,andtheeffectofcompensationis evident.For @Tcscattercompensation, we assumed that C@,1= 0 in Equation 1.Based on these results, it has been shown that, two-window image acquisition is sufficientwith @“@Tc.

In Figures 5 to Figure 7, the cross-calibrated SPECTvalues were plotted on the y-axis and the radioactivityvalues of nucides sealed in the rod were plotted on thex-axis. Figure 5 shows the results of the proposed scattercompensation and attenuation compensation method usingfive hot rods of ‘@Iin a water-filled cylindrical phantom.The cross-calibrated SPECF data were fitted to a straightline using the least squares method so that:

—S.- Ctotal

—•— Cscat.

.-‘-- Cprim.

20 40 60 80 100 120f

PIXEL Cprim

@@3UUU@—@%nnnn@—225000.——@———

F—————52-

—-- 4@J—- ———- —@1—,J%Jvu

SOOn,——

I withcorrection

a withoutcorrectionsource

w

-I

>I-C.)@ w0@(1)@

>.

A, FIGURE 4. Technetium-99m-fihled10-mm hot rodSPECT usingan SHR parallalholecoffimatorand 10-mmhot rods and poolphantoms.

Eq. 3 compensation, it is evident that the degree of separation isimproved. In Figure 8, it can be seen that the SPECTvalues in the cardiac chambers are almost zero due toscatter compensation. It can safely be said that this is notcaused by overcompensation of scattered photons, because absence of overcompensation is verified by profflecurve analysis in the data processing of scattered photonsas shown in Figures2 and3, and Figure4, which shows theSPECF results obtained for six hot rods of @“Tc.If weknow the ratio between the concentration of @‘@‘Tc-tetrofosmin in the blood after intravenous injection and myocardialuptake, we can prove that the SPEC!' values in theleft ventricle become zero.

0 0.5 1.5 2 2.5 3 3.5 4x:ACTIVITY

y=1.087x —0.009.

Figure 6 shows the same evaluation using hot rods filledwith 20111 in the same phantom. The cross-calibratedSPEC!' values were fitted to a straightline:

y = 0.912 x + 0.011. Eq. 4

Iodine-123 has two photopeaks: 160 keY and 520 keV.Thallium-201has three main photopeaks: 70 keY, 135keYand 167 keY. Therefore, it is not possible to suppose that

@ =0.Figure7 shows the dual-isotope SPECF results obtained

for five hot rods of ‘@Iand five hot rods of @°‘Tlusing thesix-window method. The cross-calibrated SPECT valuesof scatter compensation using three windows for each isotope were fitted to a straightline:

y=0.908x —0.0038, Eq.5

for 201'fland,

for 1@I.

y= 1.01x —0.0208, Eq.6

Preliminary Clinical EvaluationScatter Compensation for Myocardial Imaging Using

@“Tc-Tetrofosmin.If the image with scatter compensationis subtracted from the image without compensation, it ispossible to observe the error caused by scattered photonsat the time of SPECF reconstruction. This should not beconsidered as the distributionof scattered photons, but aserrors in the reconstruction calculation. This effect increases fromthe posteriorwall to the liver (spleen), andthelevel of the effect is not constant as shown in Figure8. Thismeans that backgroundcancellation with a constant cutoffvalue is not effective in the display of SPECT images.Accordingly, the calculation error decreases in regionswhere the posterior wall is in contact with the liver due toscatter compensation. Compared with the data without

C-)Ew

>I-C-)w0@U)

YtDwI-

-J

C)C,)U)0a:C@)

0.2 0.3 0.4 0.5ACTIVITY(mCi) X

FiGURE 5. Iodine-123-fihled10-mm hot rod SPECT using anSHRparallel-holecollimator.

2219Compton Scatter Compensation •Ichihara at al.

O.q.

0.3-————_4,@ ‘2@I

7@-——8/

/46•lOdII/a•a°@::2____

J@!I@y@@ II;:IIII0.3——-@L@::@!

a 1-123withoutcorrectionC 1-123withcorrection

I—L@

(4)

WVTHOUTCORRECT)ON

WITH SCATTER CORRECTiON0.1 0.2 0.3 0.4

ACTIVITY (mCi) X0.5

FiGURE8. Result of scatter compensationfor myocard@SPECT images.

ues were fitted to a straightline. In the case of the singleisotope SPECT study, straight lines 3 and 4 are slightlyoffset from the origin. The accuracy of the y-intemceptvalues was within 2% of the maximum activity of the hotrods. If attenuation and scatter compensation were performedideally, the gradientof these straightlines would beunity. The gradients of straight lines 3 and 4 were 1.087 and0.912, respectively. At this time, net accuracyfor measurements of activity using SPEC!', including cross-calibration, was estimated within ±10%.These values stronglydepend on the cross-calibrationprocess.

For dual-isotope SPECT of 20―fland ‘@Iusing six windows, y-intercept values decreased 0.01 mCi. In spite ofthe cross-talk effect from @°@Tlphotons of 135keY and 167

FIGURE 6. Thallium-201-filted10-mm hot rod SPECT using anSHR parallel-hOlecollimator.

DISCUSSION

An experiment was carriedout on a hot rod phantomtoevaluate the compensation of scattered photons. When theshape of the phantom is known and distribution of attenuation coefficients is uniform. For such an ideal situation,Inoue's method of SPEC!' reconstruction is capable ofgiving a complete analytical solution, including attenuationcompensation. For this reason, his method was used forthis evaluation. After scatter compensation, attenuationcompensation was performed ideally. A graph was drawnby plotting the cross-calibrated SPECT values against theradioactivityof the rods. The cross-calibratedSPECT val

TL-201withoutcorrection

TI-201withcorrectionC) A@Ew-J

>F—C)w0@U)

V W..7 I—

a:

-I

C)(I)U)0a:C)

(10)(9)

y

0.2 0.3 0.4 0.5ACTIVITY(mCi) x

0.1ACTIVITY(mCi) x

FiGURE 7. Dual-@sotopeSPECT resufts for five rods of 1@ttand @°ii.

2220 TheJournalof NudearMedicine•Vol.34 •No. 12 •December1993

a 11-201withcorrection:=@ • without correctionC)

w:D-J>-I-C)w0@U)

y0wI-

a:

C)(0U)0@'a: 0C)

accurately compensates for Compton-scatter. The TEWmethod is easier to perform, contributingsignificantly toquantitative analysis in single- or dual-isotope studies. Inthe future, it will be possible to apply this method to dinical use for routine examinations.

REFERENCES1. Bloch P, Sanders 1. Reduction of effects of scattered photons on a sodium

iodineimagingsystem.I Nuci Med 1972;25:67-fl.2. Floyd CE, JaszczakRi, HarrisCC, ColemanRE. Energyand spatial

distributionof multipleorder Comptonscatter in SPECF: a Monte Carloinvestigation.Phys Med Bid 1984;29:1217-1230.

3. Bachrach 5L, Green MV, Bonow RO, Findly5L, Daube-Witherspoon ME,Larson SM. The effectof energy windowon cardiac ejection fraction.INuciMed 1988;29:385—391.

4. JaszczakRi, GreerKL,FloydCEJr,HarrisC, ColemanE. ImprovedSPECFquantificationusingcompensationfor scatteredphotons.I NuciMed 1984;25:893-900.

5. Jaszczak Ri, Floyd CE Jr., Coleman E. Scatter compensation techniquesfor SPECT. IEEE Trans Nuci Sci 198532:786-793.

6. AxelssonB, MasakiP, IsraeLssonA. Subtractionof Compton-scatteredphotons in single photon emission computerized tomography. I Nuci Med1984;25:490—494.

7. Hamill JJ, Dc Vito RP. Scatter reduction with energy-weighted acquisition.IEEETransNuciSd 1989;36:1334-1339.

8. De Vito RP, HamillJJ. Determinationof weightingfunctionsforener@weigjnedacquisition.I NuciMed 1991;32:343—349.

9. KingMA, HademenosGJ,GlickSJ A dual-photopeakwindowmethodforscatter compensation.I Nuci Med 199233:605—612.

10.OgawaK,HarataY,IchiharaT,KuboA,Hashimoto5. Apracticalmethodfor position-dependent Compton scatter compensation in single photonemissionCF. IEEE TM!1991;10:408-412.

11. I(oral KF, WangX, RogersWL, aonth@ene NH, WangX. SPECTCompton-scatter compensation by analysis of energy spectra. I NuciMed 1988;29195-202.

12. Inoue 1, Kose K, Hasegawa A. Image reconstruction algorithm for singlephoton emission computed tomography with uniform attenuation. PhysMedBid1989;34:299-304.

13. Kubo A, Nakamura K, Hashimoto J. Phase I clinical trial of a new myocardialimagingagent @“Tc-PPN1011.IpvzINuclMed 1992;29:1165—1176.

14.OgawaK, HarataY, Ichihara1, KuboA, HashimotoS. Estimationofscatter componentin SPECF planar imageusinga Monte Carlo method.J@mI Nuci Med 1990;27:467—475.

15. OgawaK, ChugoA, Ichiharal, KuboA, HashimotoS. Quantitativeimagereconstructionusingposition-dependentscatter correctionin singlephotonemissionct. ConfRec 1992IEEENuc Sd Sym 1992;2:1011-1013.

16.FloydCE,JaszczakRi, HarrisCC,ColemanRE. Energyandspatialdistributionof multiple-orderComptonscatter in SPEC!': a Monte Carloinvestigation.PhysMedBid 1984;29:1217-1230.

2221ComptonScatterCompensation•Ichiharaat al.

keY to the ‘@I160 keV main window, the gradients ofstraight lines 5 and 6 were almost the same, as in thesingle-isotope SPEC!' study. When using a nucide withmultiple energy peaks such as 1@I,the results for the rodphantomshowed that three windows were required.But inthe case of a nuclide with a single-energy peak of emittedphotons such as @‘@Tc,the scattered photons can be eliminatedby usingtwo windows: a mainwindow anda scatterrejection window to the left of the main window.

For a preliminaryclinical evaluation, the proposed scatter compensation method required a 24% main windowimage and one or two 3-keY scatter rejection window images which were acquiredwith one scan using a three- ortwo-window acquisition. Scatter compensation was performed at postprocessing. Projectiondatafroma 24%mainwindow and one or two 3-keY scatter rejection windowscan be reconstructed either as conventional SPECT imageswithout scatter compensation or as SPECF images withscatter compensation. These data are easily compared.Clinical evaluation for the proposed scatter compensationmethod can be done without affectingroutineSPEC!' studies. This is one of the advantages of the ‘FEWmethod. Inmyocardial imaging using @Tc-tetrofosmin,the fact thatSPECT values in the cardiac chambers become almostzero is an expected result following the report concerningthe dynamics after the administrationof @‘@‘Fc-tetrofosmin(13). Within 30 min to 1 hr after intravenous administration, the myocardial uptake is 1.8% (injected dose) andblood concentration is 0.2% (injected dose/liter). If thefluctuation in cardiac chamber volume is considered, thisresult appears reasonable. In this way, SPEC!' values become zero in regions where there is no distribution ofradiopharmaceuticals. This fact contributes significantly toreproducibility and quantitative accuracy in the identification of regions of distributionand regions of ischemia onthe data display when clinical images are evaluated.

CONCLUSIONS

We have determined that the TEW method using onemain energy window and two scatter rejection windows