Temporal Image Fractionation: Rejection of Motion Artifacts in

Temporal Image Fractionation: Rejection of Motion Artifacts in Myocardial SPECT Guido Germano, Paul B. Kavanagh, Hosen Kiat, Kenneth Van Train and Daniel S. Berman Department ofMedical Physics and Imaging Division ofNuclear Medicine, Department oflmaging and the Division of Cardiolo@, Depamnent ofMedicine, Cedars-Sinai Research Institute, Cedars-Sinai Medical Center, and the Division of Nuclear Medicine and Biophysics, Department ofRadiological Sciences and the Department ofMedicine, UCLA School of Medicine, Los Angeler, California unrecognized, creating artifacts in the reconstructed im Methods: We have developed a protocol, termed “temporalages, adversely affecting their visual or quantitative inter image fractionation,― in which static myocardial perfusion pretation (1—3).Several methods of detection/correction of

SPECTstudiesareacquiredasthree-interval dynamicstudies motion have been reported in the literature, with all the (threetemporalframes,eachconsis@ng of a full projectionset), correction algorithmsrequiringsome form of postacquisi utilizing continuous altemathg detector rotabon and a mufti tion processing (4—10).Acquisition strategies to reduce detectorcamera.Theframesare indMduallyexaminedfor mo motion have essentially consisted of trying to keep the tionbycinedisplay,thensummedtogetherintoa staticSPECT filewhichis reconstructed withstandardprocedure. Thisap patient still on the bed. In this paperwe propose an acqui image fractionation,― proath offersthree potentialadvantagesin reducingor eliminat sition approach, termed “temporal which helps eliminate or reduce the effects of motion in ing imageartifactsresultingfrompatientor organmotion:(1)If myocardial perfusion SPECT. severemotionoccursin oneframe,onlythe remainingtwo are summedandreconstructed (motion-purging); (2)Aftemating de The introductionof@―Fc-teboroxime, a myocardialper tector rotationreducesartifactsfrom mono-directional, drifting fusion agent with rapid uptake and washout, has made motion dunng acquisition (i.e., upward creep of the heart); and (3) Generally, with multiple rotations, motion is spread over a

dynamic myocardial

SPED.'

a reality over the past few

years (11—15), and modern gamma cameras can support

largerangularrangeandthereforehasa lessereffectonthefinal dynamic SPEC!' acquisition protocols. It is therefore pos reconstructed images.Results:Theseadvantagesaredemon stratedand quantifiedin this paperusingdinicaldata (A)and sible to acquire a tomographic study employing several simulatedmotionon phantomdata (B and C). In the phantom alternating rotations, in otherwords, to break down a static experiments, fractionated images were found to be 48.9%, study into a dynamicstudy of the same global duration.We 35.8% and 35.9% “more similar― to the original images than

call this approach “temporalimage fractionation― because

nonfrac@onated imagesfor simulated1.67-cmupwardcreep, it fractionates the total acquisition time into several tern I .1-cmnonreturning axialmotionand I .65-cmlateralmotion, poral frames, each frame consisting of a complete set of respectively. Conclusion:Thisprotocolrequireslittleextrapro projection images.

cessing and no final extra data storage compared to standard acquisition, and it has nearly eliminated instances in which a

There are several advantages to image fractionation. First, the individual frames can be reviewed side-by-side studyhadto be repeateddueto patientmotion.Step-and-shoot by displaying their projections in cine mode before recon acquisffionis notrecommended inconjunctionwfththisprotocol,

asftwould lengthen thetirne necessaiyto obtaln thesamecount struction.This allowseliminationof framesin which obvi

ous motion occurred; sum the remaindertogether and they

statisticsas in nonfractionated acquisition. Key Words: myocardialSPECT;dynamicacquisition;fraction ation;motioncome@on

can be reconstructed into “motion-purged― tomographic

images (of course, there is an associated count statistics loss, dependent on the level of fractionationand the num J NuciMed 1994; 35:1193-1197 ber of frames contaminated by motion). Even if the pro jection images are not reviewed and the frames are simply summed together and reconstructed, image fractionation may still be less sensitive to motion than conventional, atient or organ motion is perhaps the most common static SPEC!' acquisition protocols. For example, acquir cause of technical inadequacy in myocardial perfusion ing adjacent frames using alternatingdetector rotation re SPEC!', resulting in the need for a repeat study, and, if duces artifacts from mono-directional, drifting heart mo tion during acquisition (i.e., upward creep of the heart Received Nov.29,1993;revIsion aocepted March1,1994. For cxxrespondence or reprints con@

Gu@o Germano, PhD, D@or,

following

exercise)

(16—18) by averaging

such motion

at

Nu

clearMedicine Physics, Cedars-Sinai Medical Cerderfr4@47 N,8700Beverly Blvd., each projection angle. Fractionationof static into dynamic LosAng&es,CA90048. SPEC!' studies would also alleviate artifactsfrom the rapid

Temporal Image Fractionation• Germano at al.

1193

variation

of a tracer's distribution

in the camera's

field of

view, as reported in pelvic SPEC!' studies (19—20) as well as in @Tc-teboroxime myocardial SPECT studies (21). Generally, most types of motion should result in lesser

artifactswhen imagefractionationis used, since motionis “spread― and averaged over a larger angular range than with conventional acquisition. For example, if severe mo tion of duration T affects T/15 projections out of 60 total projections (or T/5 degrees of the acquisition arc) in a nonfractionated, 180°,3°/projection,15 sec/projection ac quisition protocol,

the same amount ofmotion

would affect

n(T/15) projections out of nóOtotal (or nT/5 degrees of the

disease, and prompted him to move halfway through the first 5-mis acquisition frame. Four 5-mis frames were acquired, the last for control purposes. Frames 2 + 3 + 4 (control), as well as

Frames 1 + 2 + 3 (fractionatedacquisition,nomnotion-purged) and Frames 2 + 3 (fractionatedacquisition, motion-purged)were

reconstructedand reoriented. Upward Heart Creep To testwhetherimagefractionation mightreduceartifactsas sociatedwith upward heart creep, a chest phantom (Data Spec trum 2230 with cardiac insert 7070) was imaged over three 120-

projections,1-mmframes.The phantom'smyocardiumcontained 400 .tCiof @°@‘Fc, and the three frames were also summed to gether to represent standardor nonfractionatedacquisition. Up wardcreep of3 pixels (L65 cm) over the entirestudy durationwas

acquisition arc) in an n-frame, 180°,3°/projection,15/n sec/projection fractionated acquisition protocol. The pur simulated in the standard projection set by shifting projections pose of this paper is to visually and quantitativelydemon from45° to — 135° (45° RAOto LPO)by a linearlyinterpolated strate (A) the advantagesof image fractionation+ “motion distance uniformly varying from 0 to 3 pixels, respectively. In purging― in a clinical study with induced motion, and (B)

the motion-rejectingeffect of image fractionationin a phan tom model with simulated upward creep, lateral motion and axial motion of the heart. We use the specific fraction ation protocol routinely employed our clinic. METhODS Theimagefractionation protocolwe investigatedcalledforthe acquisitionof three frames,each of durationequalto one-thirdof the conventional“long― or nonfractionated frame(three5-mis frames versus one 15-mm frame for clinical 20111, three

@

1-mm frames

versus

@Tc-sestamibi or

one 3-mis frame for

@°‘Tc phan

other words, dependence from a multidetectorcamerawas effec tively eliminated by considering only projections corresponding to the 180° reconstructionarc, assuming that projections along that arc had been acquired sequentially. In the fractionated sets, pro jections from 45°to 135°in Frame 1, Frame 2 and Frame 3 were

similarly shifted by distances uniformly varying from 0 to 1, 2 to 1 and 2 to 3 pixels, respectively, to account for the alternating detector rotation. Frame 1, Frame 2 and Frame 3 of the fraction

ated study were summed together and reconstructed simulta neously with the nonfractionated frame, to ensure perfect corre

lation of the two image datasets. To quantify improvements effected by fractionation, normalized maximal count circumferen

tial proffleswere calculated for the unmoved (C0_J, moved frac

tornexperiments).Dynamic(fractionated)SPEC!'studieswere tionated (CfTaCt) and moved nonfractionated images. The acquiredwith continuous,alternatingCCWand CWrotationand point-by-point sum of the absolute differences between a pair of 30

projections

using

the

standard

dynamic

SPEC.!'

acquisition

60-point proffles was taken as a measure of their disagreement,

protocolavailableon our camera.A macrowas writtento display whichin turn led to the followingformulato calculatethe percent theprojectionsintheindividualframesincisc modeimmediately improvement derived by using fractionation over standard acqui after acquisition; if no major motion is seen in any frame, all

sition.

frames are automaticallysummedtogether projection-by-projec

@

tion. The entire process adds less than 1 mm to processing time. If severe motion is seen in one frame, that frameis discardedand

Percent improvement over standard acquisition =

@: -[@k@I

the remainingtwo aresummed(thisapproachis stillvalidwhen the affectedframeis the middleone, as longas motionis of the returning type). If two frames (adjacent or not) are contaminated by motion, they are both discarded, and the limitation of the

1—i@1

*100.

Eq.1

methodis the statisticalqualityof the remainingframe. Oncethe i@ I[C@k@ - [@k@I summedifieis generated,processingis resumedas with standard staticSPED' studies.The projectiondataareprefilteredwith a two-dimensional Butterworthifiter(order= 2.5 andCriticalfre It should be noted that in patient studies the position of the heart quency= 0.333cycles/pixelfor @°‘Tc-sestamibi, order= 5 and with respect to surrounding attenuators is also changing as the critical frequency = 0.25 cycles/pixel for @°@Tl, pixel size = 0.53 heartcreeps. Thus, a more rigoroussimulationof the upward cm for our acquisitions), reconstructed over 180° (45°RAO to heart creep phenomenonwould keep into account the changing 120) with a ramp filter and filtered backprojectionand reori attenuationpatterns of the fractionatedprojectiondataset. ented.No attenuationcorrectionwasused.A triple-detector cam era (Prism3000, Picker)with LEHR collimatorswas used for all Nonretumlng Motion To test whetherimagefractionationmightgenerallyimprove patient studies and phantom experiments. reconstructedimagequalityof studiesaffectedby motion,a non ClInical Patients returningmovementof 2 pixels(1.1cm) in the y direction(along Over the past 15mo, more than 1000patients have had myo thepatient'saxis)at 45 sec fromacquisitionstartwas also simu cardial perfusion @Tc-sestamibi and @°@11 SPECF studies per latedin thedatasetsacquiredin thepreviousexperiment.Projec to 0° in the nonfractionated StUdyandprojections formedwiththe fractionatedprotocolandthe separatedual-iso tionsfrom45° to —90° inFrame1of thefractionatedstudywereshifted topetechnique(22).To demonstratetheeffectof motion-purging from45° in conjunction with the fractionation protocol, we selected a by 2 pixelsin y. Again,Frame1, Frame2 andFrame3 of the @Tc-sestamibi patient with low likelihood of coronaiy artety

1194

fractionated

study were summed

together

and reconstructed

si

The Journalof NuclearMedicine• Vol.35 • No. 7 • July1994

multaneouslywith the nonfractionatedframe, andEquation1was used to quantify the percent improvementderived by using frac tionationover standardacquisition. Finally, nonreturningmotion

of 3 pixels (1.65 cm) in the x direction(perpendicular to the patient's axis) at 45 sec from acquisition start was simulated as

describedabove.Ofnote, such lateralmotioncannotbe corrected by standardmethods. C)

r)

(a

RESULTS In the 100-patient subsample for which records were available, the need for motion-purging(the occurrence of motion in one or more of the fractionatedframes)was 14% (28/200), with no significant difference between

@‘@‘Tc-ses

tamibi (15/100) and @°@Tl (13/100). The majority of cases involved one frame (25/28 or 89.3%): the first (11/25 or 44%) and the last frame (10/25 or 40%) were most fre

FiGURE 1. ReconStrUCted short-a,ds,midveritilcularslices of a @“Tc-sestamibi patientwith a low likelihoodof coronaryarterydis

quently affected, unlike the second frame(4/25or 16%,two

easeand severemotionin Frame1 of fractionatedacquisition.

201Tland two

Summingand processingof Frame2 and Frame3 yiskl a normal perfusionpattern(center),identicaltOthatinthecontroldataset(left). Summingand processingFrame1, Frame2 and Frame3 resultsin

@Tc-sestamibi).This would be expected, as

a patient is more likely to move at the beginning of the acquisition (when he/she may still be trying to find a corn fortable position), or at the end of the acquisition (when

thedistortion oftheLVshapeandthedevelopment ofaninferiorwall defectinthereconstructed image(right).

muscle strain sets in, especially if his/her arms are out

stretched). A breakdown of the four cases with motion in the middle frame showed motion as likely to be of the (right). The same amount of motion with fractionated ac nonreturning(2/4, one @°‘Tl and one @“@Tc-sestamibi) as of quisition does not affect the LV shape and does not change the returningtype (2/4, one 201'fland one @Tc-sesthmibi). its perfusion pattern (center), resulting in images much In the first instance, Frame 1 and Frame 3 were summed more similarto those producedfromthe originaldata (left). together, in the second the study was repeated (@°‘Tl) or Figure 3 shows how this similarity was quantified. First, only Frame 1 was used (@Tc-sestamibi). Motion in two maximal count circumferential profiles were calculated for out of the three frames was rare (3/28or 10.7%).In partic the images in Figure 2 using appropriate software tools. ular, motion in the first two frames occurred once (@Tc Then, absolute difference profiles were computed for the sestamibi), motion in the first and third frame twice (one original/fractionated and the original/nonfractionated pro 201Tl and one @Tc-sestamibi). Again, while the @Tc

count statistics allowed the processing of only one frame, the @°“fl study had to be repeated. In conclusion, 26 (92.9%)of the 28 studies affected by motion were salvaged by fractionated acquisition, and only 2 (7.1%) had to be repeated. Figure 1 shows reconstructed short-axis, midventricular slices of the @°‘Tc-sestamibi patient with a low likelihood of coronary artery disease (23—24) and induced motion in Frame 1. As expected, cine displayof the projectionsin the fractionated acquisition frames clearly demonstrated se vere motion in Frame 1. Summingand processing Frame 2 and Frame 3 only yielded images of good counting statis tics and a normalperfusion pattern(center) identicalto the pattern shown in the control set (Frames 2 + 3 + 4, left). In contrast, summing and processing Frame 1, Frame 2 and Frame 3 resulted in the artifactualdevelopment of a mild anterior and a moderate inferior wall defect in the reconstructed

images (right).

Figure 2 shows reconstructed short-axis, midventricular slices of the phantom with simulated upward heart creep. Motion in the projection dataset representingthe nonfrac tionated protocol results in distortion of the left ventricle (LV) shape and development of artifactual defects in the anteroseptal and inferior wall of the reconstructed image

Temporal Image FractiOnatiOn• Germano at al.

file pair, and their sums ratioed as in Equation 1. Based on

F@1.

000 FiGURE 2. RecOnStrUCted short-ads,midventticularslicesof the phantomwith simulatedupwardheart creep.The nonfractionated protocolresultsin distortionof the LV shape and developmentof artifactualdefectsin the anteroseptaland inferoseptalwall of the

reconstructed image(right).FraCtiOnated acquisition doesnotafFect the LVshapeor its perfusionpattern(center),resultingin images more“similar―to thoseproducedfromthe originaldata(left),as also demonstratedby the circumferentialprofilesin Figure3.

1195

i@

0 @t I@' 0. 200—@---—100@——o_-@-

4@:

a C

a a E U IC)

000

-.—*-—Non-fr@tk [email protected]

0

10

20

30

40

50

60

Sector (clockwise from 9 o'clock) FiGURE 3. Ma,Gmalcountdrcumferenbalprofilesfor the images in Figure 2. Quantitative analysis ofthese profiles based on Equation

FiGURE 5. ReCOnStructed short-was,midventilcularslicesof the phantomwith simulatednonretuminglateralmotion.The nonfrac tionatedprotocolresultsin distortionof the LV shapeand develop

mentofartifactual defectsintheanteriorandinferolateral wallofthe 1 resuftedin the fractionatedimagebeingjudged48.9%“more reconstructedimage(right).Fractionatedacquisitiondoesnot affect similar― to the original than the nonfractionated image. the LV shapenordoesft producedefects(center),matchingthe this criterion, the fractionatedimage was 489% more sim ilar to the original than the nonfractionatedimage. Figure 4 shows reconstructed

short-axis,

normalperfusionpatternofthe originaldata (left).Quantitativeanal ysisofthe mwdmalcountcircumferentialprofilesbasedon Equation 1 resufted in the fractIOnated image being judged 35.9% “more

similar― to theoriginalthanthenonfractionated image.

midventricular

slices of the phantom with simulated nonreturningaxial motion. Motion in the projection dataset for the nonfrac tionated protocol results in distortionof the LV shape and development of an artifactualperfusion defect in the ante nor wall of the reconstructed image (right). The same amount of motion with fractionated acquisition does not affect the LV shape nor does it produce defects (center), better matchingthe normalperfusionpatternof the original data (left). Quantitative maximal circumferential proffle

analysis based on Equation 1 resulted in the fractionated

image beingjudged 35.8%more similarto the originalthan the nonfractionated image. Figure 5 shows reconstructed

short-axis, midventricular slices of the phantom with sim ulated nonreturning lateral motion. Motion in the projec

tion dataset for the nonfractionatedprotocol results in dis tortion of the LV shape and development of artifactual perfusion defects in the anterior and inferolateral wall of the reconstructed image (right). In contrast, fractionated acquisition

does not affect the LV shape nor does it pro

duce defects (center), matching (with only a slight decrease in the wall-to-LV cavity count ratio) the normal perfusion

pattern of the original data (left). Quantitative maximal circumferential profile analysis based on Equation 1 re

000@

sulted in the fractionated image being judged 35.9% more similar to the original than the nonfractionated image.

DISCUSSION

In this paper, we have presented a new acquisition/ processing protocol to reduce or eliminate artifacts and inaccuracies derived from patient or organ motion in myo cardial SPEC!' studies. The technique uses (but does not necessarily require) a multidetector camera in conjunction with continuous, alternatingrotation to fractionate the ac FiGURE 4. ReCOnStrUcted short-wds,m@ventricuisr slicesof the phantomwith simulatednonretumingaxialmotion.The nonfraction quisition frame into three sub-frames, and proved visually atedprotocolresuftsindistortion oftheLVshapeanddevelopment and quantitatively effective in correcting motion-corrupted of an artifactualdefect in the anteriorand inferolateralwall of the clinical and phantom data. The appeal of this technique lies reconstructedimage(right).FractiOnated acquisitiondoesnotaffect in the fact that improvements in image quality and artifact the LV shapenordoesft producedefects(center),matchingthe reduction are achieved with little or no extra processing, normalperfusionpatternofthe onginaldata (left).Quantitativeanal ysisofthe ma,dmalcountcircumferentialprofilesbasedon Equation and ultimately no additional data storage requirements. Fractionation of static into dynamic SPECT studies also 1 resuftedin the fractiOnatedimage being judged 35.8% “more minimizes artifacts from the rapid variation of a tracer's similar― to theoriginalthanthenonfractionated image.

1196

TheJournalof NuclearMedicine• Vol.35 • No.7 • July 1994

distribution during acquisition (19—20),and is by no means

limited to myocardial applications. In fact, it is applicable to all organs (e.g., brain SPEC!', bone SPECT, gallium SPECT, and even static non-SPECT imaging) which re quire long acquisitions. It should be noted that if motion is substantial, images reconstructed

from fractionated

projection

data may still

contain artifactual defects. Such is the case in Figure 1 (right),where motion in Frame 1 is so severe as to cause an artifactual

inferior wall defect even when fractionation

is

used. Thus, we recommend cine review of the projection data and (if needed) motion-purging in all patients studied

2. CooperJ, NeumannP, McCandless B. Effectof patientmotionon tomo graphicmyocardialperfusionimaging.INuciMed 1992;33:1566-1571. 3. EisnerR. Sensitivityof SPECFthailium-201 myocardialperfusionimaging to patientmotion.JNuclMed 1992;33(8):1571—1573. 4. CooperJ, NeumannP, McCandless B. Detectionof patientmotionduring tomographic myocardial 34:1341-1348.

perfusion

imaging.

I

Nuci

Med

1993;

5. DcAgostiniA,MorettiR,BeilettiS,etal.Amotioncorrectionalgorithmfor an image realignment programme useful for sequential radionucide renog raphy. EurlNuciMed 1992;19:476-483.

6. EisnerR, NoeverT, NowakD, et al. Useof cross-correlation functionto detect patient motion during SPEC!' imaging. I Nuci Med 1987; 28:97—101. 7. FriedmanJ, Bennan D, Van Train K, et al. Patientmotionin thallium-201 myocardial SPECT imaging: an easily identified frequent source of artifac

tualdefect.JNuclMed 1988;13:321—324.

with the fractionated protocol. In cases where motion is found not to be confined to one frame, or is of the non returningtype and occurs in the middle frame, our experi

8. Germano G, Chua T, Kavanagh P. Kiat H, Berman D. Detection and correction ofpatient motion in dynamic and static myocardial SPEC!' using a multi-detector camera. INuciMed 1993;34:1349—1355.

ence shows that it is possible to obtain adequate data for

9. ParkerA. Effectof motionon cardiacSPEC!'imaging[Editorial].I Nuci

interpretation

from one frame only (5 min) in

@°@Tc-sesta

mibi studies, an option not available with regular acquisi

tion. However, 5-mis 201'flimages usually suffer from low count statistics. In these instances it is better to repeat the 20111 acquisition,

although

possible

alternatives

include

ac

quiring a fourth 5-mis frame, using a smoother prerecon struction filter or applying a more conventional motion

correction algorithmto the data (4—10). Finally, although not indispensable, continuous or pseudo-continuous (mod ified step-and-shoot) acquisition is highly desirable with fractionation (25). Assuming a “dead time―(time neces sary to step between adjacent projections) of 5 sec/projec tion for a single-detector camera operated in the step-and shoot, 180°,3°per projection acquisition mode (15 sec/ projection for a standard nonfractionated clinical acquisition), fractionation would cause a 50% increase in study time, from 20 to 30 min.

Additionalwork will be needed to establish optimalfrac tionation protocols. Also, conclusive evidence as to whether fractionationwithout motion-purgingis preferable to standard acquisition would require a more extensive investigation of motion patterns and myocardial activity distributions than attempted in this paper. Similarly, the use of fractionation in combination with other acquisition, pre-processing

and analysis techniques

should be carefully

investigated. We have not performed a quantitative or qualitative assessment of the changes in sensitivity and specificity for @Tc-sestamibi or 201'flmyocardial SPECF studies following the applicationof our fractionatedacqui

Med 1993;34:1355—1356. 10. Yang C, OrphanoudakisS, Strohbehn J, et al. A simulation study of motion artifacts in computed tomography. Phys Med Bid 1982;27:51-61.

11. BermanD, KiatH, MaddahiJ. Thenew @‘Tc myocardialperfusionimag ing agents: @“Tc-sestamibi and @“Tc-teboroxime. Circulation1991; 84:17—121. 12. Leppo J, DePuey E, Johnson L. A review ofcardiac imagingwith sestamibi and teboroxime. INuciMed 1991;32:2012-2022. 13. Li 0, Solot G, Frank T, Wagner H, Becker L. Tomographic myocardial

perfusionimagingwithtechnetium-99m-teboroxime at restandafterdypir idamole. INuciMed

1991;32:2000—2008. 15. Nakajima K, Taki 3, Bunko H, et al. Dynamic acquisition with a three

headedSPEC1@system: applicationtotechnetium-99m-SQ30217 myocardial imaging. INuciMed 1991;32:1273—1277. 16. Friedman J, Van Train K, Maddahi J, et al. “Upward creep―of the heart:

a frequentsourceof false-positivereversibledefectsduringthailium-201 stress-redistribution SPEC1@. INuciMed 1989;30:1718-1722. 17. Mester J, Weller R, aausen M, et al. Upward creep of the heart in exercise thallium 201 single photon emission tomography: clinical relevance and a simple correction method. EurlNuciMed 1991;18:184-190.

18. He Z, DarcourtJ, BenolielJ,et al. Majorupwardcreepof theheartduring exercise thallium-201 myocardial SPECF in a patient with chronic obstruc tive pulmonary disease. INuciMed 1992;33:1846-1847. 19. Bok B, Bice A, Qa@n M, Wong D, Wabner H. Artifacts in camera based single photon emission tomography due to time activity variation. Eur I NuciMed 1987;13:439—442. 20. O'Connor M, Kelly B. Evaluation of techniques for the elimination of

“hot― bladderartifactsin SPED' of the pelvis. I Nuci Med 1990; 31:1872—1875. 21. O'Connor M, Cho D. Rapid radiotracerwashout from the heart: effect on image quality in SPECF performed with a single-headed gamma camera system. INuciMed 1992;33:1146—1151. 22. Berman D, Kiat H, Friedman J, et al. Separate acquisition rest thallium

sition protocol. Future extensions of this work should in

dude validation of the technique in a prospective patient population using quantitativeanalysis to clearly assess the clinical significance of this novel motion rejection approach.

REFERENCES 1. BotvinickE, ThuY, O'ConnellW, DaeM. A quantitativeassessment of patientmotionandits effecton myocardialperfusionSPECFimages.INucl Med 1993;34:303—310.

Temporal Image Fractionation• Germano et al.

1991;32:1%8—1976.

14. StewartR, HeylB, O'RourkeR, BlumhardtR, MillerD. Demonstration of differentialpost-stenoticmyocardialtechnetium-99m-teboroxime clearance kinetics after experimentalischemia and hyperemic stress. I Nuci Med

201/stress technetium-99msestamil,i dual isotope myocardial perfusion SPECT:a clinicalvalidationstudy.JAm CoilCardiol 1993;22:1455-1464 23. Rozanski A, Diamond 0, Berman D, et al. The declining specificity of exercise radionudide ventriculography. NEngilMed 1983;309:518—522. 24. Rozanski A, Diamond 0, Forrester J, et al. Alternative referent standards

forcardiacnormality.Implicationsfordiagnostictesting.AnnInternMed 1984;101:164—171. 25. sionwith SPECF: current issues and future trends. In: Zaret B, Beller G, eds. Nuclearcwdiology: thestateoftheartandfuzw@direc&@is. St Louis: Mosby

YearBoolq1992:77-88.

1197