Content uploaded by Ruben C Gur
Author content
All content in this area was uploaded by Ruben C Gur on Jul 23, 2015
Content may be subject to copyright.
developed from sectioned brains (1), published atlases or
stereotactic definition of ROIs (Z3). More recently, it has
become feasible to perform routine anatomic imaging in
parallel with physiologic studies to define ROIs, which are
then overlaidand/orco-registeredwith the physiologic im
ages for quantification (4—6).
The use of an anatomicallybased system for ROl defi
mtion has a greatdeal of intrinsic appealand is essential in
the presence of gross structuralpathology. However, with
the currentspatialresolution of PET imaging,the benefits
obtained by individualizing template-based ROl definitions
to each subject's anatomy in the absence of gross pathol
ogy are unclear. There are little data comparing this labor
intensive approach to a simplermethod based on propor
tional adjustmentof standardtemplates directly upon the
physiologic images.
The present study was designed to examine the stability
of PET quantitation across a number of approaches to
image analysis. The following questions were addressed:
1. What is the interoperator reliability of a template
system for ROl definition and quantitation?
2. How doesanatomicallybasedROl localizationcorn
pare with the use of a standard template overlaid
directly on PET images?
MATERIAI@SAND METHODS
Sub@ts
The potentialof anatomic imagingto improvethe quantitative
accuracy of fun@onal brain ima@ngthrough refined regknal
definitionis widelyaccepted. However,there are littledata ad
dressingthe impactofapproach to regionallocalizationon quan
titationof metabolic images in the absence of gross structural
pathology.We compared MRI-baSedversus PET-based ap
proaches to the analysis of PET 18F-fluorodeoxyglucose(FDG)
images using a standard adjustable template based on simple
geomethc regions. Forthe MRI-baSedapproach, templates and
indMdualregions were adjusted to e@h Individual'sanatomy,
whereas the PET-based definitioninvolvedonlyglobal propor
tionaladjustmentofthe standard templates. Metabolicrates for
glucose and volume-to-wholebrain ratios were determined by
twooperatorsfor78 volumesofinterestin fivesubjects. Palrwise
correlationsindicatedhigh interoperatoragreement for each ap
proach and high intraoperatoragreement for MRI-baSedversus
PEr-based metaboliCvalues. The stabilityofthe metaboliCrates
and ratios among operators and analysis approaches was sup
portedbylowcoeffidentsofvarlationacross measurements and
small average differences in paired comparisons. Thus, witt@n
the currentspatialresolutionof PET imaging,quantitationof
metabolicimages is relativelyrobustto imageanalysisapproach
in the absence of gross structuralabnormality.To take advan
tags of the greater quantitativeaccuracy promised by high-res
olutionanatomiCand functionalimaging,more refineddelinea
tionof anatomicimageswillbe necessary.
J Nuci Med 1993; 34:2201-2207
ThesampleconsistedoffiveresearchsubjectsfromtheMental
HealthClinicalResearchCenter(MHCRC)on schizophrenia(7).
Three subjects (2 male, 1 female) were patients with schizophre
ma and two were normalvolunteers(1 male, 1 female).The
female control subject was subsequently excluded from further
MHCRCevaluationdue to informationprovidedat follow-upthat
indicateda pasthistoiyof substanceabuse.Meanage was26 ±
2.5 yr.
Measurements
MRL MRS!S@@flSwere obtained on a GE Signa 1.5 Tesla mag
neticresonancescanner.Followinga briefsagittallocalizingscan,
spin-echo imageswere acquired in the axial plane, with TR = 3000
and TE = 30/80,a slice thicknessof 5 mm and no gap between
slices. All scans received a clinical interpretation by experienced
neuroradiologists. The scans of the two controls and the female
te quantification of regional brain physiology
with PET ultimately depends upon accurate localizationof
regions of interest (ROIs). Methods for the definition of
PET ROIs fall broadly into two categories: physiology
based versus anatomy-based approaches. Early physio
logic schemes involved placement of ROIs on “hotspots―
or areas of increased radioactivity concentrationson PET
scans. Subsequent approaches were based on templates
ReceivedOct. 29. 1992;revIsionacceled Aug.5, 1993.
Forcorrespondenceor reprintscor1@ SM. ReenlclçPhD,GRCINLA,Lato
ratoryofPersonaRtyand Cognition,Room2-C-14,4940EasternAve.,Baltimore,
MD21224.
MRIVersusPET-BasedFDGQuantitation•ResniCkatal. 2201
Comparison of Anatomically-Defined Versus
Physiologically-Based Regional Localization:
Effects on PET-FDG Quantitation
Susan M. Resnick, Joel S. Karp, Bruce Turetsky and Raquel E. Our
Laborato,y ofPe@onality and Cognition, GRC/NIA, Baltimo,@ Mwyland and Depamnents ofP@sychiatiy and Radiologj@,
Unive,city ofPennsylvania, Philade@ohia
patientwere readas normal.Onemale patienthad a minimal
increase in sulcal prominence for age, and the second male patient
hadmultiplesmallpunctateareasof abnormalsignal,increased
for age, in the white matterof the superiorposteriorparietallobe.
PET. PET studies ofregional cerebral glucose metabolism were
performed using ‘8F-fluorodeoxygluoose (FDG) and the PENN
PETscanner(8). Fluorine-18-FDGwas injectedas a bolus,and
arterial blood samples were obtained over the next 90 mm. A
styrofoamheadcastwas made for each subjectprior to radioiso
tope administrationto aid in immobilizationduring scanning.
Studieswereperformedunder“restingconditions―witheyesand
ears unocciuded. Followinga 40-mmuptake period, the subject
was positioned in the PENN-PET scanner and the distribution of
‘8Fwasimagedoverthenext50 mm.A laserbeammountedon
thescannerwas employedfor headpositioningto allowimaging
alongthecanthomeatalaxis.
Images were acquired in 2-mm slices, yieldinga total of 45
transverse slices. To improve statistics, successive 2-mm sections
weresummedinanoverlappingfashiontoyield43slicesof6-mm
thickness,with slicecentersseparatedby 2mm. The2-mmaxial
sampling of the PENN-PET scanner is unique and minimizes the
partialvolume effect in the axial direction, leadingto more accu
rate visualizationand quantitationof reslicedimages(9). This is
accomplished using large continuous detectors without plane-dc
finingsept@These studies were acquired using the prototype
PENN-PETscanner,whichhas anaxialfieldofviewof9 cm.The
relativelysmallfieldof view leadsto some error in reslicingand
regionalquantitationbecausetheentirebrainisnotimagedandall
slicesmaynotbe availablefor particularvolumes.The intrinsic
transverse and axial resolution of the PENN-PET scanner is 5-6
mm FWHM, although the average resolution with reconstructive
ifiteringistypically7—8mm.Thefinitespatialresolutionlimitsthe
quantitationof smallstructures, as the partialvolumeeffectwill
resultin an underestimationof counts in a region.
Local cerebral metabolic rate for glucose (CMRglc) was deter
mined as described previously (10) using experimentally deter
mined values of the lumped constant and rate constants (11).
Image Analysis. PET image analysis was accomplished using
thePETVIEWsoftwarepackage,developedbyourPETgroupin
collaboration with UGM Medical Systems. The software includes
modulesforimagedisplayin three orthogonalplanes(transverse,
sagittal, coronal), for reslicing in oblique planes and for RO!
definitionand quantification.
Template Development. We have developed a series of stan
dard templates, whichcan be placed on either the ME! or PET
images, using the PETVIEW software. Twenty-one templates
havebeendefinedin planesparallelto the anteriorcommissure
posteriorcommissure(AC-PC)line.The templatesare separated
by 4 mm along the z-axis and include 78 volumes, each volume
includingseveral ROIs from successive slices (see Table 2 for
definitionof volumes). ROl definitionwas guidedby brain atlas
sections and correspondingoverlays(12) and by the Talairach and
Tournoux brain atlas (13). Template regions were relatively sim
plcgeometricforms(Fig.1),viewedassamplesratherthaninclu
sive of activity in particularstructures.For each template, polyg
onalorellipticalregionsweredrawnon onehemisphereandthen
“mirrored―to the contralateralhemisphere,ensuringthat ROIs
were initiallyof identicalsize andorientationfor the two hemi
spheres.
Template Implementation. MEl and PET images were recon
structed into volumes (a stack of transverse sections) and then
each dataset was resliced parallelto the AC-PC line, using the
Ia
,
F:,.,. .@1. MjustedtemplateregionsovedalduponMRI(top)
and correspordng PET (bottom) slices.
“oblique―softwaremodule.SincethePETimagematrixconsists
of 2-mm3 voxels, the interpolation errors are small and do not
significantly affect the quantitation. The voxel size in the MR.!
imageis typically0.78 x 0.78 x 5mm,which isfairlycoarse inthe
axial direction but did not lead to difficulty in visualizing anatomic
structures after reslicing.The AC-PC linewas definedfrom the
sagiualview and, following Talairach and Tournoux Methodology
(13),cutsthroughthesuperiorportionoftheanteriorcommissure
and the inferiorportionof the posteriorcommissure.On the PET
images,theAC-PCline was notclearlyvisible,butan approxi
marionwas providedby a linepassingthroughthemostanterior
part of the frontal lobe, the anterior corpus callosum and the
occipitalpole(14).AlternatePET sliceswereprocessed,yielding
quantificationof approximately21sliceswith4-mmspacing.
FortheanatomicallybasedRO!definition,thestandardtern
plateswere first overlaidon the reslicedMR.!images.The rnost
superiorslicecontainingthecaudatenucleusandoverallfitwere
used as guides for positioning in the z-axis. The templates were
customized to each individual's anatomy in three steps:
1. A global proportional adjustmentfor the whole template
series was performedthroughtranslation,rotationand re
sizing.
2. Regionswithineachslicewereadjustedproportionallyas a
group.
3. Individual ROIs were positioned over appropriate struc
tures by translation, rotation and resizing.
These MEl-based templates were then overlaid upon corre
spending PET slices, again using the most superior slice contain
ing caudate nucleus as a guide for positioning in the z-axis. Due to
scalingdifferencesbetweenour MR.!and PETimages,itwas also
necessaiy to globally resize each MM-adjusted template slice to
theresliced(AC-PCoriented)PETsectionsusingthe50%thresh
old of theedgeof the brainas a guide.Fromthe reslicedPET
images, count densities were then determined for each region and
regional cerebral glucose metabolism was quantified (10,11).
Area-weightedaverage metabolicrates for glucose metabolism
were computedacrossallslices toyieldmetabolicrates forthe 78
volumesofinterest.“Whole-brain―metabolismwasestimatedby
an area-weighted average of metabolic rates for all regions. Be
cause most of the volumes define gray matter structures, the
whole brainvalue primarilyreflectsgraymattermetabolism.
For the physiologically-basedRO!definitionand quantifies
tion, the originaltemplateswere placed directlyupon the PET
images.Theheadofthecaudatenucleusandoverallfitwereagain
used as guides for positioning in the z-axis. The global atlas- and
slice-based adjustments were performed using the 50% threshold
2202 The Journalof NudearMedicine•Vol.34 •No. 12 •December1993
OperatorI Operator21.
MRI 2. PET 3. MRI 4. PET 5. MAI-PET15
•Operator2'sMAt-basedtemplateuponthereslicedPETImageofOperator1.
Correlationsfor absolute CMR@Care above the diagonal and those for volume-to-wholebrain ratios are belowthe diagonal. For clarfty,
Interoperatorcorrelatlons uslngthe same approach are InItalics,Whilelntraoperatorcorrelatlons between MRt-basedversus PET-based appro@hes
areInbold.
Numberof paired observationsvaries between339 and 365. Allcorrelationsare slgnfficantat p = .0001.
TABLE I
PairwiseCorrelationsAmong Measurements
OperatorI
I. MAt-based
2. PET-based
Operator2
3. MAt-based
4. PET-based
5. MAP-baSedPETI*
0.95
0.96 0.98
0.98
0.95
0.92
0.90
0.96
0.90
0.92
0.95
0.98 0.97
0.96
0.91
0.96
0.96
0.95
0.93
of theedgeof the brainforplacementof the outerregions,but
ROIswere not manipulatedindividually.Countdensitiesand
CMRg1c for glucose metabolism were determined as described
above.
Procedures
Five measurementsof metabolicactivitywere made for each
subject's PET image:
1. Operator 1, MEl-based.
2. Operator1, PET-based.
3. Operator2, MEl-based.
4. Operator 2, PET-based.
5. Operator 2, MEl-based using the resliced PET image of
Operator 1.
Thus,thefirstfourmeasurementsprovideassessmentsof in
traoperator agreementacross ME!-and PET-basedmethodsand
interoperatoragreementwithin and acrossmethods.The fifth
measurementwas performedto evaluatethe extentto which
interoperatorvariabilitymightbe due to differencesin reslicing
the PETimagesversusfittingtemplatesto the ME!.
Angle of reslicing, positioning of templates in the z direction
and template manipulationvaried between operators, allowing
assessment of inter-rater reliability for the entire procedure. Since
thepurposeof the intraoperatorcomparisonsof ME!-andPET
based quantitationwas to evaluate the variabifityin metabolic
rates as a function of template manipulation on anatomic versus
physiologicimages,positioningof the templatesin the z direction
andangleof reslicingwere fixedwithinoperatorsforthesecorn
parisons. These factors were also fixed across operators for the
fifth measurement and were identical to those for the first and
secondmeasurements.
Statistical Analysis
All analyses were performedfor absolute values of CMRglc
and volume-to-wholebrain ratios. Pairwisecorrelationsamong
measurements were calculated across subjects and volumes to
examine:interoperatoragreementfora givenapproach(e.g., 1
versus 3, 2 versus 4), intraoperatoragreementacross MR.!-and
PET-based approaches (e.g., 1 versus 2, 3 versus 4) and the
agreementbetweenoperatorswhen usingthe samereslicedPET
image (1 versus 5). Variability in these values across measure
mentswas examinedby analysisof coefficientsof variation.For
each subject,coefficientsof variationacross the five measure
ments were calculated for each volume. These coefficients of
variation were then averaged across subjects and across both
subjects and volumes to provide, respectively, mean coefficients
ofvariation for each volume and an overall mean value. Note that
these values do not includevariabilityin metabolismbetween
subjects,andvariabilitybetweensubjectsis reflectedonlyto the
extentthatagreementamongmeaswrment@differsbetweensub
jects. In addition, paired t-tests were performed to examine the
magnitude,directionandconsistencyofdifferencesbetweenmea
surements. T-tests were done for the following comparisons: in
teroperator for MEl-basedand PET-basedvalues (1 versus 3, 2
versus 4), intraoperatorfor MEl-basedversus PET-basedvalues
(1versus2, 3versus4)andbetweenoperatorsfortheMEl-based
approach using a fixed resliced PET image (1 versus 5).
RESULTS
Paii-wise correlations among the five measurements are
presented in Table 1. Correlationsfor absolute metabolic
rates are above the diagonal and correlations for volume
to-whole brain ratios are below the diagonal. All correla
tions indicate significant and high agreement among mea
sures. As expected, correlations for absolute values are
higher than those for volume-to-whole brain ratios, be
cause global metabolic differences between subjects con
tnl,ute to the former. Scatterplots of intraoperatorand
interoperatorcorrelationsforvolume-to-whole brainratios
are illustratedin Figure 2.
Mean coefficients of variation for the 78 volumes are
presented in Table 2. The average volume size was 3.47 ±
4.9cm3,containinganaverageof 216.9voxels(each2 x 2
x 4mm).ThevolumesizesrangedfromameanofO.07cm@
for the uncus to a mean of 26.8 cm@for the cerebellum. As
described above in the statistical analysis section, the mean
coefficients of variation reflect the average coefficients of
variation across the five measurements and the five sub
jects. The mean ±s.d. coefficient of variation across all
volumes was 6.1% ±3.8% (rangefrom 1.6%to 27.6%) for
absolute values of CMRg1cand 6.2% ±4.1% (range from
MRIVersusPET-BasedFDGQuantitation•Resnickatal. 2203
2.0
1.8
0
w 1.6
@ 1.4
@ 1.2
01 1.0
a:
0 0.8
I-
@ 0.6
w
@ 0.4
0.2
0.0
2.0 2.0
1.8 1.8
0 0
wl.6 wl.6
q2 Cl)
@1.4 @1.4
ti1.2 ti1.2
a. a-
@1.o @1.0
a: a:
@0.8 @0.8
@0.6 @0.6
w w
a-0.4 a-0.4
0 0
0.2 0.2
0.0 0.00.00.20.40.60.81.01.21.41.61.82.0
OPERATOR 2: MAt-BASED
2.0
1.8
0
w 1.6
@ 1.4
L@1.2
a-
i@1.0
a:
0 0.8
@ 0.6
w
a. 0.4
00.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
OPERATOR1: MAt-BASED 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
OPERATORI: PET-BASED
FiGURE 2. Scatterplots of
coffelabonsbatween measure
meritsforvolume-to-wholebrain 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
OPERATOR I: MAt-BASED
1.3%to 28.8%)forvolume-to-whole brainratios. Six of the
78 coefficients ofvariation for absolute values and 9 of the
ratiovalues aregreaterthan 10%.Withthe exception of the
medulla, which is subject to error due to reslicing with a
9-cmfield ofview acquisition, allvolumes have coefficients
of variation less than 15%. Note that smaller subcortical
and limbic structures, including the basal ganglia, hip
pocampus, and amygdala, show a similar degree of stabil
ity across measures as do the largercorticalvolumes.
As described above, differences in CMRglc and ratio
values between pairs ofmeasurements (1versus 3, 2versus
4, 1 versus 2, 3 versus 4, 1 versus 5) were examinedto
assess the consistency of the varied approaches to image
analysis. Average paired differences across the 78volumes
for CMRglc and ratio values are presented in Table 3.
There were small but consistent differences in absolute
values of CMRg1c across operators, reflecting slightly
higher metabolic rates for Operator 1 compared with Op
erator 2. The average differences in CMRg1cin Table 3
correspond to less than 2% of the mean CMRglc values.
Note that corrections for partialvolume effects, which we
are not currentlyperforming,would result in much greater
increases in metabolic rates, but would affect all measure
ments. For volume-to-whole brain ratios, there were no
significant differences between methods.
The t-test comparisons for individualvolumes were per
formed for exploratory purposes, using a 0.01 level of
significance due to the large number of tests and small
numberof observations. With thiscriterion,a maximumof
4 of the 78volumesreachedsignificancefor eachcompar
ison of two measurements, but no particular volume
emerged as more sensitive to the type of analysis.
DISCUSSION
Despite the increasing use of anatomicallybased mea
surements for localization of ROIs on functional images,
there have been few attempts to directly evaluate the im
pact of differentapproaches to regionaldefinitionon quan
titative results. The findings of the present investigation
suggest that in the absence ofgross pathology, image quan
titationis relativelyrobustto the methodof image analysis
within the current resolution of PET. We observed high
interoperatorandintraoperatoragreementusingastandard
template approach for MEl-based and PET-based ap
proaches to image analysis. Correlationsamong measure
2204 The Journal of Nudear Medicine•Vol.34 •No. 12 •December 1993
Volume-to-WholebrainAbsolute
CMA9icratioRight
LeftAightLeft
TABLE 2
Mean Coeffidentsof Variationfor 78 Volumes Acrossthe Five Measurements and Five Subjects
1 AnterIorcingulate3.33.33.94.33
SuperIorfrontalgyrus3.94.03.03.05
Middle frontalgyrus5.83.14.73.27
Sensorlmotorcortex8.82.07.72.29
Superiortemporalgyrus7.02.15.72.111
Supramarglnalgyrust14.86.114.25.413
Corpus caltosumanterior6.67.26.56.615
Caudatenucleus4.13.94.54.617
Insuta1.52.63.04.019
Thaiamus3.32.84.03.421
Lateraloccipitalcortex6.65.35.64.823
Med@Ioccipitalcortex5.75.45.14.725
PosterIorcingulategyrus5.54.86.45.627
Corpus caJiosumposterior10.210.010.310.529
Middletemporalgyms7.13.35.62.831
InferIorfrontalgyrus8.63.98.03.833
Gtobuspallidus*1.32.22.51.635
Putamen2.21.82.62.437
Parahippocampalgyrus4.14.25.14.639
Hippocampus6.82.96.33.141
Cerebellum4.83.54.43.443
Midbrain3.03.82.73.945
OrbItalcortex10.08.910.28.947
Gyrusrectus6.55.77.36.449
Amygdala4.83.04.84.151
Ungualgyrus3.64.82.65.153
Lateraloccipltal-temporalgyms8.36.36.95.757
Hypothalamus6.28.76.28.059
Uncus6.712.27.411.861
Pons8.45.97.45.763
Medulta*28.819.627.618.165
Cingutategyrus(anterior+posterior)12.54.612.04.267
Angutargyrus6.36.55.16.169
Precuneus@6.43.47.32.871
ParacentralIObUIe@6.87.57.76.673
SuperIorfrontal/Mkldlefr@+@t5.84.15.94.675
MedIal superIor/Inferiorparietal9.7@7.39.4@8.377
Lateralsuperior/Inferiorparletal'8.59.78.310.179
Interiortemporalgyrus8.35.27.35.0
*MAt-t@@volumesforOperator1 onlybecause Operator2 was notconfidentof @suaiIzationofthisvolumeon MAt.
tAOtimagedInfoursubjects.
@R0timagedInthreesubjects.
‘AOlImagedintwosubjects.
ments were high for both absolute metabolic rates and
volume-to-whole brain ratios. The stabilityof image quan
titation across measurements was confirmed by coeffi
cients of variation of less than 10%for most volumes and
average differences in paired comparisons of less than 2%
for absolute metabolic rates and less than 1%for the vol
ume-to-whole brainratios. A variety of errors contributes
to variability in quantitation, includingaccuracy of reslic
ing reconstructed images, positioning of templates in the
axial direction and template manipulation. Despite these
errors, our results suggest that individual differences in
metabolism and in the regional distributionof metabolism
can be measured reliably using a variety of relatively sim
pie image analysis approaches.
While all correlations among methods were very high,
the correlations for volume-to-whole brain ratios between
MEl and PET-based measurements for each operator
(0.95)tendedto behigherthanthosebetweenoperatorsfor
a single method (0.92). This indicates that there is not a
clear advantage with the MEl-based method compared to
the PET-based method in the presence of other errors and
the variability among trained operators. An unexpected
findingwas that relatively smaller subcortical and limbic
structures, such as the basal ganglia, amygdala and hip
MRIVersusPET-BasedFDGQuantitation•ReSnICkatal. 2205
AbsoluteCMR9ic(mgflOO
9-Volume-to-Wholemiir1)brain
ratiosMean
(s.d.)Mean (s.d.)
greaterdifferencesbetween anatomicallybased and phys
iologically based approaches to image analysis. While this
is a future goal, the use of simple geometric regions and
transverse images for regional localization is typical of the
approachcurrentlyemployed in many PET facilities. Our
results offer an index of the potential error in applying
standardized templates for PET quantitation in the absence
of structuralpathology.
Although the robustness of the quantitation of PET met
abolic images is encouraging, we do not advocate aban
donment of anatomically based approaches to ROI local
ization. The limits of tolerance to pathology of proportional
adjustmentsof standardtemplates or stereotactic atlases
are unclear. MRI-basedRO! definitionwill be essential in
the presence of gross structural abnormalityand may be
critical in identifying subtle abnormalities or changes in
brainfunction across repeated studies. Difficulties in local
izing subtle changes between repeated studies highlight the
importance of anatomic measurement (16). Moreover, seg
mented MRIs can be used to estimate @SFand gray and
white mattervolumes for partialvolume correction (17,18).
These enhancementswillbe particularlyimportantfor con
ditions associated with brain atrophy, such as dementia
and normal aging, where PET-based analysis is likely to
underestimate specific regional metabolic rates due to con
foundingeffects of tissue loss. To fully realize the greater
quantitative accuracy promised by high-resolution ana
tomic imaging, refined methods for segmentation of tissue
types and delineation of ROIs on structural images must be
concurrentwith improvements in the spatial resolution of
PET imaging.
ACKNOWLEDGMENTS
This research was supported by NIMH grants MH 43740,
MHCRC 43880, MH 48539, NIH grant NS 14867 and a Ben
FranklinPartnershipFoundationGrant. We thank Gerd Much
llehner, Shenji Guan and Manuel Angel for their efforts in the
developmentof PETVIEWandthestaffsof thePETandcyclo
tronfacilitiesfor theircontributionsin performingthe PETscans.
*p< 0.05;tp < 0.01; @p< 0.001.
‘Operator2's MAt-based template upon the resliced PET image of
Operator 1. Numberofvolumes ineach comparisonranges from336 to
358.
pocampus, showed stability across operators and ap
proaches comparable to that for the larger cortical vol
umes. The fact that ouranalyses were volume-based rather
than slice-based was a likely contributor to this stability.
Although this study was based on a particulartemplate
system, the consistency of findings across measurements
suggests these results may be general to similar systems
implemented in many other PET facilities. The relative
insensitivity of metabolic measures to variability in image
analysis approach suggests that stereotactic methods
should provide adequate estimation of region localization
in normal individuals. However, the extent to which the
findings can be generalized beyond FDG images is unclear,
because proportionaladjustmentof the template slices on
physiologic images requires a clearly defined brain con
tour.
The potentialof anatomically-basedROl localization to
improve quantitative accuracy was not fully tested in this
study due to limitations in regional definition on MR.!and
the use of relatively simple geometric regions. The MR.!
based measurements were limited by the ability to define
boundaries of specific cortical gyri and limbic structures,
e.g., hippocampus, from transverse sections, because this
is notoriously difficulteven for trained operators. In addi
tion, simple geometric regions do not accurately describe
convoluted structures such as cortical gyri and tap heter
ogeneous tissue composed of cerebrospinal fluid and white
matterin addition to gray matter. Schmidt et al. (15) have
recently described the errors introducedby tissue hetero
geneities in determinationof glucose metabolism using the
currentkinetic models of the FDO method. Thus, we be
lieve that more refined regional localization on MRI, in
cluding use of lateral surface views of three-dimensional
reconstructionsforgyral definition,considerationof tissue
heterogeneities and partial volume correction, will improve
accuracy in quantifying functional images and lead to
TABLE 3
Average Differencesin MetabolicRates and Ratios Across
Measures and Volumes
Operator 1 vs. Operator2MAt-basedO.07(O.39)t—0.006(0.10)PET-based0.04
(0.36)*—0.004(0.09)MAt
based on Operator 10.06(0.26)@-0.006(0.07)resliced
PET'MAt-based
vs.PET-basedOperator
10.02(0.26)0.007(0.07)Operator
2—0.02 (0.29)0.006(0.08)
REFERENCES
1. DanaR, Muehllehner0, RosenquistA. Computer-aideddataanalysisof
ECFdata[Abstract].JNucIMed1983;24:82.
2. Bohm C, Greitz T, Kingsley D, Berggren B, Olison L Adjustable comput
erized stereotaxicbrain atlas for transmissionand emission tomography.
AJNR 1983;4:731-733.
3. Fox PT, PerimutterJS, RaichieME. A stereotacticmethodof anatomical
localizationforpositronemissiontomography.I ComputAssistTomogr
1985;9:141—153.
4. EvansAC,BailC,MarreftS. ThompsonCJ,HakimA.Anatomicalfunc
tionalcorrelationusrngan adjustableMR.!basedatlaswithPET.I Ce@b
BloodFlowMetab1988;8:813—830.
5. EvansAC,MarrcttS,TorrescorzoJ,KuS,CollinsL MRI-PETcorrelation
in three dimensionsusinga volume-of-interest(VOL)atlas.I CerebBlood
Flow Metal, 1991;11:A69—A78.
6. PelizarriCA, Oen GTY, SpelbringDR, WeichselbaumRB.,ChenC-T.
Accuratethree-dimemionalregistrationofCT, PET,andMRimagesofthe
brain.JComputAs@lot Tomogr 1989;13:20-26.
7. Gur RE, Mozley PD, Resnick SM, Ct at. Magnetic resonance imaging in
schiwphrenia.I.Volumetricanalysisofbrainandcerebrospinalflald.An@h
Ge,tPs@h@tiy1991;48:407-412.
8. K&pJS,Muehllehner0, MankoffDA,etal.Continuous-slicePENN-PET:
2206 TheJournalof NudearMedicine•Vol.34 •No. 12•December1993
14. FristonIU, PassinghamRE, Nuft JO, Heather ID, SawleGV, Frackowiak
RSJ. Localisationin PEr images:direct fittingof the intrecommissural
(AC-PC)line.I CerebBloodFlowMetab19899:698-695.
15.SchmidtK, Ludgnani0, MorescoRM,etal.Errorsintroducedbytissue
heterogeneityinestimationoflocal cerebralglucoseutilizationwith current
kineticmodelsofthe [‘8Fjfluomdeoxyglucosemethod.J CerebBloodFlow
Metab 1992;12:823—834.
16. Drevets WC, Videen TO, MacLeod AK, Hailer JW, Raichle ME. PET
imagesofblood flowchangesduringanxiety:correction Science19fl;256:
1696.
17. Meltzer CC, I@ealJP, Mayberg HS, Wagner MN Jr, Frost JJ. Correction of
PET dataforpartialvolumeeffectsin humancerebral cortexby MR imag
lag.I ConiputAssirtTomogr1990;14:561-570.
18.Muller-GartnerHW,links JM,PrinceJL, et al. Measurementof ra
diotracerconcentrationinbraingraymattarusiig positronemissiontomog
raphy: MRI-based correctionforpartialvolume effects.JCerebBloodFlow
Metab 1992;12:571—583.
apositrontomographwithvolumeimagingcapability.JNuclMed199031:
617—627.
9. KarpiS, Daube-WitherspoonME,Muehllehner0. Factorsaffectingaccu
racy and precision in PET volume imaging.I Ceirb Blood Flow Metab
1991;11A38-A44.
10.ReivichM, KuhiD, Wolf AP, GreenbergJ, Ct al. The @F.fluomde@
oxyglucosemethodforthe measurementoflocalcarebralglucoseutilization
in man. Cur Res 1979;44:127-137.
11.ReivichM,AlaviA, WoIfAP,CtaLGlucosemetabolicratekineticmodel
parameter determinationin man: the lumpedconstantsand rate constants
for1@F-fluorodeoxyglucoseand‘1C-deoxyglucose.I CerebBloodFlow
Metab 19855:179-lfl.
12.GeeJC,ReivichM, BajcsyR. ElasticallydefOrming3D atlasto match
anatomicalbrainimages.I ComputAssirt Tomogr1993;17:225—236.
13.TalairachJ,TournouxP.Co-planarstereota@catlasofthehumanbraA
3.dime?uionoJpm@onal3ystenL- an ajpivach to cerebwJimagvtg. New
York Theme Medical Publishers, Inc.; 1988.
NOTE
All 1991issues (Volume32)of the Journal ofNuclear Medicineare availablefor back order throughMarch 1994.
Thereafter, 1991 issues will be available only on microfilm. If you are missing issues from Volume 32, please take
advantage of this opportunity to order them before March 31, 1994. Orders can be made through Bookmasters,
Inc., 800-247-6553.
2207
MRIVersusPEr-BasedFDGQuantitatlon•Resrickatal.
