Transfer of normal 99m Tc-ECD brain SPET databases between different gamma cameras

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

Abstract. A stereotactic, normal perfusion database isimperative for optimal clinical brain single-photon emis-sion tomography (SPET). However, interdepartmentaluse of normal data necessitates accurate transferability ofthese data sets. The aim of this study was to investigatetransfer of three normal perfusion databases obtained inthe same large population of healthy volunteers who un-derwent sequential scanning using multihead gammacameras with different resolution. Eighty-nine healthyadults (46 females, 43 males; aged 20–81 years) werethoroughly screened by history, biochemistry, physicaland full neurological examination, neuropsychologicaltesting and magnetic resonance imaging. After injectionof 925 MBq technetium-99m labelled ethyl cysteinatedimer (ECD) under standard conditions, 101 scans wereacquired from all subjects (12 repeat studies) on a triple-head Toshiba GCA-9300A (measured average FWHM8.1 mm). Ninety-one sequential scans were performedon a dual-head Elscint Helix camera (FWHM 9.6 mm)and 22 subjects also underwent imaging on a triple-headPrism 3000 (FWHM 9.6 mm). Images were transferredto the same processing platform and reconstructed by fil-tered back-projection with the same Butterworth filter(order 8, cut-off 0.9 cycles/cm) and uniform Sorensen at-tenuation correction (µ=0.09). After automated rigid in-trasubject registration, all subjects were automatically re-oriented to a stereotactic template by a nine-parameteraffine transformation. The databases were analysed us-ing 35 predefined volumes of interest (VOIs) with nor-malisation on total VOI counts. For comparison, thehigh-resolution data were smoothed with a 3D Gaussiankernel to achieve more similar spatial resolution. Hoff-man phantom measurements were conducted on all cam-

eras. Partial volume effects after smoothing varied be-tween –6.5% and 10%, depending on VOI size. Be-tween-camera reproducibility was 2.5% and 2.7% for theToshiba camera versus the Helix and the Prism database,respectively. The highest reduction in between-cameravariability was achieved by resolution adjustment incombination with linear washout correction and a Hoff-man phantom-based correction. In conclusion, transfer ofnormal perfusion data between multihead gamma cam-eras can be accurately achieved, thereby enabling wide-spread interdepartmental use, which is likely to have apositive impact on the diagnostic capabilities of clinicalbrain perfusion SPET.

Keywords: Regional cerebral blood flow – 99mTc-ECD –Normal database – Resolution

Eur J Nucl Med (2001) 28:435–449DOI 10.1007/s002590000461

Introduction

Instrumentation and methodological advances are ex-pected to result in more accurate and widespread diag-nostic use of brain perfusion single-photon emission to-mography (SPET). Improved resolution and sensitivitydue to the use of fan-beam collimation allows acquisitionof technetium-99m labelled radioligands with a resolu-tion of 7–8 mm full-width at half-maximum (FWHM),while correction algorithms for physical factors that de-grade image quality, such as scatter [1] and non-uniformattenuation [2], have become commercially available.Substantial progress has also been made towards the au-tomated analysis of functional imaging data involvingoperator-independent retrospective intramodality as wellas intermodality co-registration [3, 4].

Current (semi)quantitative image analysis methodsin functional brain imaging are based on the compari-

Koenraad Van Laere (✉ )Division of Nuclear Medicine, P7, Ghent University Hospital, De Pintelaan 185, 9000 Ghent, Belgiume-mail: koen.vanlaere@rug.ac.beTel.: +32-9-2403028, Fax: +32-9-2403807

Original article

Transfer of normal 99mTc-ECD brain SPET databases between different gamma camerasKoenraad Van Laere1, Michel Koole2, Jan Versijpt1, Stefaan Vandenberghe2, Boudewijn Brans1, Yves D’Asseler2, Olivier De Winter1, Alain Kalmar1, Rudi Dierckx1

1 Division of Nuclear Medicine, P7, Ghent University Hospital, De Pintelaan 185, 9000 Ghent, Belgium2 Medical Signal and Image Processing Department (MEDISIP), Faculty of Applied Sciences, Ghent University, Belgium

Received 20 September and in revised form 5 December 2000 / Published online: 17 February 2001© Springer-Verlag 2001

son of data from a single patient or patient group withdata from a control group. Therefore, such a controlgroup, preferably consisting of healthy volunteers,plays a most important, but also a sensitive and vulner-able role in the quantitation of functional neuroimagingdata [5].

To date, most studies have used their own age-matched, usually small normal database, acquired on thesame camera and preferably under the same conditions.Nevertheless, reference standards based on high-resolu-tion functional data could improve comparative interde-partmental studies as well as data exchange for largerstudies and could also improve the impact of clinicalbrain SPET by allowing quantitative analysis of singlepatient data with a controlled reference set. The avail-ability of such a large database will depend entirely onthe possibility of accurate transfer to other gamma cam-eras. Moreover, if transfer from high-resolution to low-resolution systems is feasible, it will enhance researchpossibilities by permitting the combination of data ob-tained over a number of years, during which time tech-nological advances have resulted in new scanners withimproved sensitivity and resolution time [6].

Only a small number of studies have addressed thisissue using actual patient or volunteer data with brainpositron emission tomography (PET) and SPET, andthese have been based on small samples or specific agegroups [6, 7]. Also in the domain of nuclear cardiology,few studies have been published on the comparison ofnormal SPET data obtained using different cameras or indifferent departments [8].

The GO AHEAD project (Ghent Optimized AbsoluteHigh-resolution ECD Adult Database) was initiated inour institution with the aim of constructing a large age-and gender-stratified normal 99mTc-ethyl cysteinate di-mer (ECD) perfusion and 3D-magnetic resonance imag-ing (MRI) database obtained in thoroughly screenedhealthy volunteers, with non-uniform attenuation- andscatter-corrected data (Van Laere et al. 2000, submittedfor publication). The core SPET database was acquiredon a high-resolution triple-headed gamma camerasystem (89 volunteers+12 repeat studies within 2weeks); most volunteers also underwent same-day sub-sequent scanning on a dual-head gamma camera (n=91),and in addition some underwent imaging on anothertype of high-resolution triple-headed gamma camerasystem (n=22).

The aim of this study was therefore threefold. Firstly,it was investigated whether the SPET resolution of thethree gamma cameras could be approximated as uniformand isotropic, so that three-dimensional Gaussian filtersfor conversion to lower resolution gamma camera datacould be used. Secondly, a high-count Hoffman phantomstudy was conducted in order to investigate such transferin the absence of (confounding) interindividual physio-logical variables among human subjects. These datawere examined semiquantitatively by means of an auto-

mated, anatomically standardised procedure based on awhole-brain volume of interest (VOI) analysis. Thirdly,the volunteer databases acquired on the three gammacamera systems were compared after adjustment for res-olution, adjustment for inhomogeneous washout of99mTc-ECD and application of a between-camera correc-tion factor, based on the previous high-count Hoffmanstudies. The relative impact of these correction factorswas investigated with respect to between-camera repro-ducibility of the data sets.

It is known that the accuracy of human (semi)quanti-tative SPET and PET studies depends on a large numberof factors. These include, firstly, the physical and intrin-sic tomograph characteristics, such as attenuation, scat-ter, resolution recovery, statistical uncertainties due tocamera sensitivity, head holder effects and filtering. Sec-ondly, several physiological parameters such as age, gen-der, injection conditions and inhomogeneous tracerwashout contribute to the intersubject variation. Thirdly,there are factors that influence the anatomical standardi-sation, such as minimisation algorithms with their asso-ciated parameters. Finally, the region of interest (ROI),VOI or voxelwise procedure used for (semi)quantifica-tion determines the actual outcome values. Therefore, inthis work, the same volunteers were used for the creationof the databases and most processing parameters werekept constant in order to obtain a comparison with mini-mal confounding factors.

Materials and methods

Gamma cameras. SPET studies were performed on three gammacamera systems. The core data set was acquired on a ToshibaGCA-9300A triple-headed camera (Dutoit Medical, Wommelgem,Belgium), equipped with super-high-resolution lead fan-beam(SHRFB) collimators and transmission CT (TCT) sources (camera1). It was operated in a brain-dedicated mode with a head domeand at a fixed 132 mm radius of rotation. A fixed head rest, con-sisting of 3 mm carbon fibre and 5 mm mousse, was attached tothe bed. Volunteer data on camera 1 were acquired with gadolini-um-153 transmission sources placed at the focal lines of the fan-beam collimators, and scatter windows for triple-energy window(TEW) scatter correction. Camera 2 was an Elscint (General Elec-tric, Haifa, Israel) Helix dual-headed system with low-energyhigh-resolution (LEHR) collimators. The radius of rotation was setat the minimal circular rotation and with a mean of 135 (SD 4)mm for the volunteer data. A removable head rest was used, con-sisting of 2 mm carbon fibre and 2 mm rubber. Neither scatter nornon-uniform attenuation corrections were used for camera 2, sincefor this dual-headed camera, neither multiple window acquisitionsnor transmission imaging was available.

Camera 3 was a Marconi (Cleveland, Ohio, USA) Prism 3000triple-headed system with ultra-high-resolution fan-beam collima-tors and equipped with a single 153Gd line source for non-uniformattenuation correction. The radius of rotation was fixed at158 mm, which is required for the operation of the STEP attenua-tion correction. A head rest consisting of an 11-mm compositehead support and a 6-mm carbon fibre head holder with 6 mmfoam was utilised. Although transmission data were acquired,

436

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

non-uniform attenuation correction was not used for this camerasince the Prism-STEP attenuation correction protocol does not al-low a 99mTc downscatter correction to be made for the 153Gd win-dow, which is known to produce an incorrect central brain quanti-fication [9]. Because of the single-window acquisition on camera2, we preferred not to use scatter correction on this data set and tocompare only data sets uncorrected for scatter in this study. Thismakes the approach [one window and filtered back-projection(FBP) with uniform attenuation correction] usable on any currentgamma camera.

SPET resolution measurements. For all three gamma camerasused, attenuation- and scatter-free SPET resolution maps in airwere acquired by means of a matrix of point sources using 20370±4 kBq (10.0±0.1 µCi) point sources. Markers for the pointsources were placed at equidistant locations of 35 mm in the axialand 40 mm in the radial direction. The resulting point source di-ameter was consistently between 1.5 and 2 mm in all cases.

SPET measurements were acquired during 20 min with thesame angular sampling and radial distances as were used in thevolunteers’ study. For the Helix camera, measurements were car-ried out at the average camera rotation radius of 135 mm.

Reconstruction was performed with FBP. By means of thegaussfit function in the IDL programming language (RSI Systems,USA), Gaussian fits along the axial, tangential and radial direc-tions were calculated. The FWHM was determined from the calcu-lated standard deviation σ:

(1)

corrected for the source diameter modelled as a Gaussian distribu-tion with σ=0.5 mm. The resolution correction under a uniformand isotropic assumption was similarly achieved by smoothing the

data with a Gaussian kernel κ, where σκ,n for camera n (n=2,3),given by:

(2)

Hoffman phantom data. In order to determine systematic differ-ences between the gamma cameras without physiological intersub-ject confounds, a Hoffman phantom experiment was performed.The Hoffman phantom [10] was filled with 175 MBq 99mTcO4

–. Inall cases, the phantom was oriented with the cerebellum on thehead rest and the midsagittal plane perpendicular to the horizontalplane. SPET measurements of the same filled phantom were ac-quired during 40 min (triple-headed cameras) or 60 min (dual-headed cameras). The same angular sampling and radial cameradistances as were used in the volunteers’ study were taken forcameras 1 and 3, while for camera 2 the average volunteer studyrotation radius of 135 mm was taken. In total, 20–25 Mcountswere acquired. To investigate the reproducibility of the phantommeasurements due to phantom manipulations and registration, theHoffman phantom was also taken out of camera 1 and measuredagain in approximately the same position.

Healthy volunteer data. All volunteers were thoroughly screenedby history, biochemistry, physical and full neurological examina-tion (by a skilled board-certified neurologist) and neuropsycholog-ical testing (by a skilled board-certified psychiatrist), and all hadnormal MRI scans (T2 and MPRAGE). Handedness was evaluatedthrough a questionnaire according to Briggs and Nebes [11]. Be-tween February and November 1999, all 89 selected volunteerswere injected with 925 MBq 99mTc-ECD (Dupont PharmaceuticalsLtd., Brussels, Belgium) under resting circumstances (eyes closed,low ambient noise level). The selection criteria for normality aresummarised in Table 1. Normative uptake data, effects of age and

437

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

FWHM = 2 2ln 2σ

σ σ σκ ,n n= −212

Table 1. Exclusion criteria for the normal volunteers

Known disorders of the central nervous system (epilepsy, head trauma, known structural abnormality on CT or MR scan)

Previous unexplained unconsciousness

Known psychiatric disease or first- or second-degree relative with hospitalisation for psychiatric disease, especially mood disorders or schizophrenia, or known dementia of the Alzheimer or frontal lobe type in first- and second-degree relatives

Implanted electronic devices (pacemaker, neurostimulator, infusion pump, etc.)

Substance abuse (alcohol, drugs or medication) or existing dependency or previous (poli)clinical treatment for complications of substance abuse

Consumption of psychoactive medication during past 3 months (antidepressives, hypnotics, sedatives, etc.)

Uncontrolled hypertension with repeated diastolic pressure >95 mmHg

Diabetes mellitus

Auto-immune disease

Major internal heart, lung, liver or kidney disease

Mini-Mental State Examination< 27 (age over 60) or < 28 (age under 60)

Abnormal blood or urine investigation

Possible pregnancy

Medical investigation with radiation burden above 1 mSv during past 6 months

Abnormal neurological, neuropsychological or psychiatric testing by board-certified neurologist and psychiatrist

gender and voxelwise structural correlation of these confoundershave been described elsewhere (Van Laere et al. 2000, submittedfor publication). Of the 89 volunteers, 86 underwent subsequentscanning on camera 2, and 22 underwent scanning on camera 3. Intotal, 12 volunteers underwent repeated 99mTc-ECD SPET scanswithin 2 weeks of the first scan on camera 1, and five also under-went rescanning on camera 2. The extended set of 91 studies oncamera 2 was used to compare cameras 1 and 2. The common setof data for all three cameras, consisting of a subset of 22 overallyounger volunteers, was available for the transfer study betweenthe three cameras. Demographic parameters and scan data aresummarised in Table 2.

All volunteers gave written informed consent to participationin the study, which was authorised by the local Ethics Committeeof Ghent University Hospital.

Data acquisition and processing. To restrain head movement, thesubject’s head was strapped. All subjects co-operated optimallyduring the whole procedure. On cameras 1 and 3, the acquisitionwas performed in continuous mode with a total of 90 projectionsin a 128×128 matrix. Scan duration was 40 s per projection angle.For camera 2, 60 projections in a 128×128 matrix were acquired,each also 40 s in duration. In all cases, the main acquisition win-dow was a 20% window over the 140-keV peak.

Fan-beam projections were converted to 128×128 parallel ge-ometry data in 4° bins using the floating-point rebinning softwaresupplied by the manufacturers.

The raw projection data were transferred after Interfile 3.3conversion to a Hermes imaging platform (Nuclear Diagnostics,Hägersted, Sweden). On this system, images were reconstructedby FBP in a 128×128×128 matrix. The resulting cubic voxel sizewas 3.2 mm for camera 1, 3.4 mm for camera 2 and 2.08 mm forcamera 3. For all images, the same prefiltering was performedwith a Butterworth filter of order 8 and cut-off 0.90 cycles/cm.

A pre-reconstruction Sorenson uniform attenuation correctionwith an optimised effective attenuation coefficient of 0.09 cm–1

was applied [9, 12]. Automated elliptical attenuation contourswere drawn at a threshold of 5%, approximately corresponding tothe outer skull rim. Manual adjustment was necessary for the low-er cerebellar slices because of the low extracerebral uptake.

For intrasubject data sets acquired on different cameras, aswell as the Hoffman phantom data, the intrasubject data were first-ly realigned by a six-parameter rigid registration, based on a countdifference cost function with an iterative downhill-simplex searchalgorithm (MultiModality, Nuclear Diagnostics, Sweden). These

realigned brain SPET data were then automatically registered toan anatomically standardised (stereotactic) template with a voxelsize and slice separation of 3.59 mm (64×64×64 matrix) (BRASS,Brain Registration and Automated SPET Semiquantification, Nu-clear Diagnostics, Sweden) [13, 14, 15]. For the latter, the samealgorithm was used but based on all nine linear degrees of free-dom (translation, rotation and scaling).

On the above-mentioned template, 35 VOIs were defined withinclusion of all grey matter for semiquantification. These VOIsconstitute a so-called region map. The size of these VOIs rangedfrom 4.2 to 37.5 ml. Phantom 99mTc and volunteer 99mTc-ECD up-take was adjusted to the whole brain VOI uptake (i.e. normalisationon total VOI counts). Such an automated, objective and operator-independent VOI approach is currently our routine procedure forsemiquantification of SPET studies which allows the automatedextraction of quantitative information on regional uptake values,reproducibility, asymmetry indices and anteroposterior gradients.

For the regional analysis of radial effects such as attenuation,the VOI set was also summarised in radially divided clusters: neo-cortex and cerebellum, subcortex (anterior and posterior gyrus cingulus, mesial temporal lobe) and central VOIs (striatum, thala-mus, pons). Right-left cerebral perfusion asymmetries were exam-ined with relative differences defined as the asymmetry indexAI=[(R–L)/(R+L)]×200, where R=99mTc-ECD uptake in a right-sided VOI and L=the uptake in the homologous left-sided region.

Anteroposterior gradients were defined by classifying theVOIs into three cluster regions: η=1: anterior, defined by all VOIswith a midpoint anterior to the coronal plane with Talairach coor-dinate y=10 mm (relative to anterior commissure; location approx-imately of the caudate head); η=2: central, defined between anteri-or and posterior; η=3: posterior, defined by all VOIs with a mid-point posterior to the coronal plane with Talairach coordinatey=–30 mm (posterior border of thalamus).

Correction for inhomogeneous washout. 99mTc-ECD is character-ised by an inhomogeneous washout in the brain occurring after in-jection [16]. Since the magnitude of this washout is not negligiblecompared to the achievable confidence intervals with currentSPET analysis techniques, a correction factor was applied. As afirst-order approximation, a linear correction was carried outbased on the measured time difference between the midpoint ofthe scans on the different cameras. In other words, a correctionbased on a position-dependent homogeneity gradient ∂H(x,y,z,t)/∂twas applied. This gradient was assumed to be constant over time,which holds true for the first 4 h after injection [16]. The values

438

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

Table 2. Demographic vari-ables and acquisition parame-ters

Camera type Camera 1 Camera 2 Camera 3Toshiba GCA9300A Elscint Helix Picker Prism 3000

Collimator SHRFB LEHR UHRFBNo. (# repeat studies) 101 (12) 91 (5) 22 (0)Average age (years) (SD) 45.0 (17.4) 44.4 (17.2) 25.2 (6.3)

Range 20–81 20–81 20–44Men/women 47/54 42/49 6/16Start time p.i. (min) (SD) 35 (8) 85 (20) 157 (17)

Range 25–67 57–144 111–189Time from scan 1 (min)a – 50 (18) 120 (20)

Range – 27–107 83–156Rotation radius (mm) (SD) 132 (fixed) 135 (4) 158 (fixed)

Range – 124–150 –

SHRFB, Super-high-resolutionfan-beam; LEHR, low-energy,high-resolution; UHFRB, ultra-high-resolution fan-beama Relative time difference be-tween start scan on camera 1and camera i (i=2, 3)

used for this correction were applied relative to the gradientsfound for neocortical values as published by Ichise et al. (Table 3)[16]. Since not all regions which were defined on our VOI mapwere reported in this reference, for other regions an estimate basedon these values was taken, as also indicated in Table 3. No age-re-lated differences in the ECD clearance were assumed since atpresent no data are available demonstrating age-related differencesin specific cytosol or membranous esterase activity responsible forthe intracellular entrapment of ECD in the brain.

Statistics. Descriptive statistics were calculated for uptake values,age groups, asymmetry indices and anteroposterior gradients. Pri-or to the parametric tests, each data set underwent Kolmogorov-Smirnov testing for normality and Levene’s test for equality ofvariances. For the comparison of the Hoffman phantom data, thebetween-camera reproducibility (σh,mn) was defined as one stan-dard deviation of the unweighted relative VOI activity differenceAm–An between cameras m and n (m,n=1..3):

(3)

Similarly, weighting by VOI size was conducted to exclude largecontributions of small VOI areas with higher measurement vari-ance. For the healthy volunteer data, multivariate analysis of vari-ance based on the general linear model with repeated measures(RMANOVA) was conducted in order to identify differences be-tween 99mTc-ECD uptake values. Bonferroni correction for multi-ple comparisons was applied. A Student’s t test was used to studydifferences between average asymmetry indices. Bivariate correla-tions were investigated with the parametric Pearson correlationtest. Non-parametric correlations (Spearman test) were investigat-ed between the anteroposterior region clusters and 99mTc-ECDVOI uptake value differences. Statistical significance is reported

as being at or above the 95% (P≤0.05) limit. All statistical analys-es were performed with SPSS (v9.0 for Windows, SPSS Inc., Heverlee, Belgium).

Results

SPET resolution

The FBP reconstructed slices of multiple point sourcesexhibited no significant star artefacts. In Fig. 1, projec-tion and 3D plots of the resolution maps obtained fromthe three cameras are shown for the central 3×5 pointmatrix which was in the common field of view area ofall three cameras. The resolution values along the radial,tangential and axial directions, are given in Table 4. Thespatial resolution was slightly anisotropic in the three or-thogonal planes for the two fan-beam cameras. Analysisof variance (Bonferroni corrected) showed that, for allcameras, no differences were present in radial, tangentialor axial FWHM along points in the axial direction. How-ever, with radial distance, a significantly different axialFWHM was present for camera 1 (P=0.007, decreasingwith radial distance). Also a significantly different (in-creasing) tangential FWHM was present for camera 1(P=0.014), camera 2 (P=0.001) and camera 3 (P=0.001).

The largest FWHM difference in the cylinder definedby this matrix (diameter 16 cm, approximately corre-sponding to the average anteroposterior diameter of thehuman cerebrum) was 0.9 mm. Because of these rela-

439

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

Table 3. Correction factor forinhomogeneous washout (basedon Ichise et al. [16])

Region Additional VOIsa Clearance Correction(%/h) (±SD) (%/h)b

Frontal 5.44 (1.11) 0.21Temporal Medial temporal 4.7 (1.04) –0.53Parietal 4.44 (1.14) –0.79Occipital 6.34 (1.21) 1.11Average neocortical 5.23 (0.57) –Cerebellum 4.23 (1.13) –1.00Basal ganglia 4.65 (1.03) –0.58Thalamus Pons 2.45 (0.94) –2.78White matter 2.39 (1.57) –2.84

a For regions not defined in[16]b Relative to the average neo-cortical value

σh mnn m

m n

A AA A, (%) = × −

+

standard deviation 200

Table 4. Spatial resolution for the gamma cameras used in this work, as a function of radial distance R from the central axis of rotation

Cameraa Average FWHM FWHM x FWHM y FWHM z

R (cm) 0 8 0 8 0 8

1 8.1 (0.7) 7.5 (0.7) 7.5 (0.6) 7.5 (0.2) 8.9 (0.8) 9.4 (0.3) 8.4 (0.4)2 9.6 (0.9) 10.2 (0.2) 10.3 (0.4) 9.5 (0.7) 9.8 (0.3) 8.6 (0.4) 9.5 (0.6)3 9.6 (0.9) 10.2 (0.4) 9.6 (0.1) 8.3 (0.3) 9.3 (0.2) 10.0 (0.6) 10.6 (0.3)

a Camera 1 = Toshiba CGA-9300A triple-headed, SHRFB, detector radius Rdet=13.2 cm; camera 2 = Elscint Helix dual-headed, LEHR,average Rdet=13.5 cm; camera 3 = Picker Prism 3000 triple-headed, UHRFB, Rdet=15.8 cm

440

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

tively small changes in FWHM over the radial distanceand the absence of axial effects, the effect of resolutiondegradation was modelled, within a first-order approxi-mation, as isotropic and uniform. The average FWHM,calculated as the mean of all values over the 3×5 matrix,is also given in Table 4. Since the measured resolution ofboth camera 2 and camera 3 turned out to be lower, thedata set from camera 1 was smoothed with a 3D isotro-pic Gaussian kernel to result in the same average FWHM(Eq. 2). These smoothed data will be referred to as set#1′.

Phantom measurements

The schematic procedure for subsequent phantom andhuman measurements and analysis is summarised inFig. 2.

For the estimation of within-camera reproducibility,the same phantom was measured twice in camera 1 withthe same filling but after replacement and subsequentsix-parameter rigid co-registration. The within-cameraphantom reproducibility, calculated as defined in Eq. 1but with repeated measurements (i=j=1), resulted in avalue of σh,11=1.1% (difference range –2.1%–2.0%).

For the Hoffman data, the anterior temporal VOI wasexcluded from the subsequent analysis since this region isincompletely incorporated in the phantom. A specificsmaller cerebellar VOI was constructed because of similarincomplete inclusion in the standard Hoffman phantom.

Differences between data set 1 and its smoothed 1′counterpart ranged from –6.5% to +10.1%, with a stan-dard deviation of 3.8% (correlation Rh,11′=0.952,P<0.001) (Fig. 3A, B). Highest effects, due to partial vol-ume effects from (the absence of) uptake in extracerebraltissue, white matter and ventricular space, were observedin the orbitofrontal cortex (∆Ah,1′l=Ah,1–Ah,1′=–10.5%)and pons (∆Ah,1′l=–6.7%). The superior parietal regionshowed the highest increase in relative uptake. The abso-lute value of these differences was not correlated to VOIsize (P=0.9), and differences were also not significantlyrelated to the radially grouped clusters (P=0.79). Howev-er, significant differences were found as a function of an-teroposterior gradient, where the relative uptake in theposterior regions increased compared to the anterior up-take values (ANOVA P=0.019, Bonferroni corrected).

Concerning the camera comparison, there were nosignificant differences between the smoothed data set 1′

Fig. 1. Point spread function measurements in the central field ofview in the three gamma cameras used for this work at the (aver-age) radius of rotation for the healthy volunteer studies. A 3×5point grid was measured. y, Axial direction; x, radial direction. Forcomparison, all pixel sizes were converted to 2 mm. The values ofthe z-axis are relative values. The radial and axial profiles are pro-jections from the central (third) radial row and central (rightward)axial column, respectively

441

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

and the other Hoffman data sets from cameras 2 and 3(paired t test, P>0.50). Highly significant correlationswere found between these data, as can be seen in Fig. 3Cand D. The Pearson correlation coefficients wereRh,1′2=0.989 and Rh,1′3=0.984 (P<0.0001). The correla-tion coefficient was lower for the comparison to the un-smoothed data: Rh1,2=0.952 and Rh,13=0.970 (P<0.0001).

The unweighted between-camera reproducibility be-tween data sets 1′ and 2 (σh,1′2) was 1.7% (mean uptakedifference <∆Ah,1′2>=<Ah,2–Ah,1′>=–0.2%, range –4.4%to +2.5%). For the Hoffman data sets 1′ and 3, the repro-ducibility σh,1′3 was 2.0% (<∆Ah,1′3>=–0.2%, range–5.1% to +6.6%). In all cases, the highest uptake differ-ences were found in small VOIs such as the orbitofrontalcortex and the basal ganglia. When weighting accordingto VOI size was applied, the reproducibility improved toσh,1′2=1.4% and σh,1′3=1.7%.

Boxplots of the data set comparisons as a function ofanteroposterior and radial gradients are shown in Fig. 3Eand F. ANOVA between the difference of the relative up-take values and radial grouping revealed no significantdifferences between neocortical, subcortical or centraluptake values for any of the camera pairs (P>0.15). Sig-nificant differences were present in the anteroposteriorgradient for the uptake difference ∆Ah,1′3 (P=0.022, Bon-ferroni corrected), but not for the other comparisons. Thenon-parametric correlation between the anteroposteriorcoefficient η and ∆Αh,1′2 did not reach significance(P=0.9) but was highly significant for ∆Ah,1′3 (Spearmanρ=–0.54, P=0.001). The absolute magnitude of the rela-tive 99mTc uptake differences was not correlated to VOIsize (P=0.5).

Healthy volunteer data

Figure 4A shows the effect of smoothing and thus partialvolume effects on the semiquantification of the extendeddata set 1 (n=91) versus data set 1′. A similar pattern asfor the Hoffman data is observed. RMANOVA showedsignificant differences between the two data sets 1 and 1′(P=0.024). There was a significant region-by-smoothinginteraction effect in the bilateral orbitofrontal and anteriortemporal cortex and the cerebellum (all decreases) as wellas the occipital and superior parietal cortex (all increases).

Regression analysis between smoothed and originalhuman data for this camera resulted in a regression coef-

Fig. 2A–D. Schematic procedure for the transfer and analysis ofthe volunteer data. Data from the reference camera with the high-est resolution and point spread function PSF1 (A) were correctedby a 3D Gaussian smoothing kernel to the mean value of the othertwo cameras. Data from the cameras with the lower resolutionwere obtained later and were thus corrected for inhomogeneouswashout (B). After rigid intrasubject registration, all subject datawere registered by a nine-parameter affine transformation to a ste-reotactic template (count difference minimisation) (C). On thistemplate, a predefined VOI set had been constructed, allowing anautomated VOI extraction with normalisation on the total numberof counts (D)

442

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

ficient Rv,1′1 =0.945 (P<0.0001) (Fig. 4B). The unweight-ed between-camera reproducibility for the volunteer dataset was σv,1′l=3.6% (<∆Av,1′l>=–0.4%, range –9.7% to+6.1%). With VOI weighting, this resulted in σv,1′l=3.1%.

Concerning the database comparison between cam-eras, Fig. 5 shows box plots of the semiquantitative99mTc-ECD uptake values for the overall comparison forthe common data set (n=22), as well as the comparisonbetween cameras 1 and 2 for the extended group of vol-unteers (n=91). The correlation coefficients between thewashout-corrected VOI data sets were Rv,1′2=0.977

(n=91) and Rv,1′3=0.962 (n=22) (all P<0.0001). Thesevalues were not altered by more than 0.01 by caseweighting according to VOI size. The correlation wasbetter than between the original data set from camera 1:Rv,12=0.914 and Rv,13=0.919.

RMANOVA for the common cases showed no overallsignificant differences between the VOI uptake values,corrected for multiple comparisons (P=0.07–0.78). Asignificant camera-VOI interaction was present in mostVOIs. The most significant camera-VOI interactionswere present in the prefrontal, parietal, occipital and pos-terior cingulate cortex. The same analysis for the extend-

Fig. 3A–F. Hoffman phantomcomparative measurements. AEffect of resolution degradationon uptake for the various (bilat-eral) VOIs from camera 1. Dataset 1′ represents the smoothedstudies. Error bars represent 2standard deviations. OFR, Or-bitofrontal; PRFR, prefrontal;LAFR, lateral frontal; SUFR,superior frontal; SMC, sensori-motor; TSUP, temporal superi-or; TMI, temporal medial/infe-rior; TMES, mesial temporal;PAI, parietal inferior; PAS, pari-etal superior; OCC, occipital;CBL, cerebellum; PGP, puta-men and globus pallidus; CNC,caput nucleus caudatus; THA,thalamus; GCA, cingulate gy-rus, anterior; GCP, cingulategyrus, posterior. B Scatter plotof the Hoffman reproducibility(open triangles) and effect ofresolution degradation (filledcircles). C, D Reproducibilityscatter plots for 99mTc Hoffmanphantom uptake between cam-eras 2 and 3, respectively, ver-sus the uptake of smoothed da-ta from camera 1. E, F Uptakedifference between the variouscameras (open boxplots, 1′ vs2; horizontal lines, 1′ vs 3;hatched, 2 vs. 3), as a functionof radial clusters (neocortex,subcortex and central VOIs)(E) and anteroposterior VOIposition (anterior, central andposterior) (F)

443

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

ed comparison between data sets 2 and 1′ showed themost significant camera-VOI interactions in the orbito-frontal, prefrontal, anterior cingulate and parietal cortex.

For all cameras, 99mTc-ECD uptake coefficients ofvariation were similar for all regions, ranging from 2.4%up to 8.8% in the small orbitofrontal and anterior tempo-ral pole region (Table 5). The standard deviation was sig-nificantly higher compared to the same common data setfor cameras 1′ (SD=4.0%) and 2 (SD=3.4%) (paired ttest, P=0.001).

The asymmetry index was evaluated for the bilateralVOIs, providing a measure of both magnitude and direc-tion of asymmetry. Figure 6 shows the results for thecommon (Fig. 6A) and extended set (Fig. 6B). There wasan average rightward asymmetry for most cerebral re-gions (t test, all P<0.04) (Table 5). Among the threecameras, there were no statistical differences in asymme-try for the regions considered (paired t test, P=0.4–0.9).

ANOVA showed significant differences (P<0.001)between 99mTc-ECD uptake differences and anteroposte-rior index for the comparison between cameras 1 and 2and that between cameras 1 and 3. A significant correla-tion was present between the anteroposterior index andthe difference between cameras 1′ and 2 (Spearmanρ=–0.62, P<0.001), as well as that between cameras 1′and 3 (Spearman ρ=–0.44, P<0.001). There was no sig-nificant correlation between the difference in uptake andthe radially defined clusters (neocortical, subcortical,central).

Correction of volunteer databases by Hoffman phantommeasurements

In order to study the possibility of improved databasetransfer by means of an intermediary brain phantom cor-

Fig. 4. A Effect of resolution degradation on 99mTc-ECD uptakefor the bilateral VOIs for camera 1, shown as boxplots for all heal-thy volunteers. Data set 1′ represents the smoothed studies. TANT,Temporal anterior; otherwise the abbreviations are as in Fig. 3A.B Scatterplot of the resolution degradation with linear regressionanalysis. Error bars denote 1 standard deviation from the individ-ual VOI measurements

Fig. 5. A Relative uptake of 99mTc-ECD normalised to the totalVOI counts (%), for the common data set obtained on all threecameras (n=22). Open boxplots, camera 1′ (smoothed); hatched,camera 2; horizontal lines, camera 3. B Relative uptake of 99mTc-ECD normalised to the total VOI counts (%) for the extended dataset obtained on cameras 1 and 2 (n=91). Abbreviations as previ-ously defined

444

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

Table 5. 99mTc-ECD regional uptake values, standard deviation and asymmetry index (AI) for the three gamma cameras with uniformresolution and linear washout correction

VOI Cam1 Cam1′ Cam2 Cam3FWHM adjusted Washout corrected Washout corrected

Uptake (SD) AI Uptake (SD) AI Uptake (SD) AI Uptake (SD) AI(n=91) (n=91) (n=91) (n=22)

Orbitofrontal Left 97.4 (5.4) 0.2 88.5 (6.5) –1.1 92.0 (5.3) 0.0 90.3 (8.2) 0.3Right 97.6 (5.0) 87.5 (6.5) 92.0 (5.4) 90.5 (8.7)

Prefrontal Left 94.6 (3.2) 4.4 94.3 (3.0) 4.1 97.7 (3.1) 3.80 100.7 (4.8) 2.8Right 98.9 (2.9) 98.3 (2.6) 101.5 (3.0) 103.5 (4.3)

Lateral frontal Left 97.7 (2.9) 2.0 98.7 (2.9) 2.02 101.0 (2.8) 3.16 102.4 (2.7) 2.8Right 99.7 (3.2) 100.7 (3.0) 104.3 (2.7) 105.3 (3.4)

Superior frontal Left 99.9 (4.0) 1.9 102.6 (3.7) 1.7 103.0 (3.5) 0.3 106.2 (4.4) –0.2Right 101.8 (3.4) 104.4 (3.4) 103.2 (3.2) 105.9 (4.2)

Sensorimotor cortex Left 101.9 (3.3) 0.1 104.7 (3.3) 0.5 101.7 (2.9) 2.1 102.6 (3.7) 1.1Right 102.1 (3.0) 104.8 (3.2) 103.8 (2.4) 103.7 (4.4)

Anterior temporal Left 79.4 (4.5) 2.0 75.3 (6.5) 0.3 77.1 (4.4) –3.5 74.2 (8.8) –1.0Right 81.0 (4.3) 75.5 (6.4) 74.4 (5.1) 75.5 (5.7)

Superior temporal Left 108.1 (3.8) –3.9 107.7 (3.3) –2.6 104.1 (3.0) –1.1 100.5 (4.6) 1.0Right 104.0 (3.8) 105.0 (3.7) 103.0 (2.7) 101.5 (2.5)

Inferior and Left 96.6 (3.1) 1.0 95.7 (3.1) 1.1 92.8 (2.9) 3.3 91.1 (3.7) 4.3medial temporal Right 97.6 (3.1) 96.7 (3.2) 96.0 (3.0) 95.2 (2.6)

Mesial temporal Left 75.0 (3.9) 3.8 74.4 (4.8) 4.1 77.3 (4.2) –0.9 75.8 (4.2) 2.0Right 77.9 (4.1) 77.5 (4.8) 76.6 (3.8) 77.3 (5.2)

Inferior parietal Left 106.4 (3.4) –0.4 106.5 (3.4) 0.8 101.9 (3.4) 1.3 103.2 (4.8) 2.8Right 106.0 (3.5) 107.4 (3.7) 103.3 (3.2) 106.1 (4.2)

Superior parietal Left 96.4 (5.8) 1.0 102.5 (5.5) –0.2 99.6 (4.8) –0.1 103.1 (5.4) 0.6Right 97.5 (5.1) 102.2 (4.7) 99.5 (4.8) 103.7 (6.7)

Occipital Left 110.2 (3.6) –1.2 113.9 (3.5) –2.4 110.0 (3.6) 1.4 105.5 (3.9) 1.6Right 108.9 (3.8) 111.2 (3.7) 111.5 (3.9) 107.2 (4.8)

Cerebellum Left 103.3 (5.2) 0.5 98.1 (5.6) 0.5 97.4 (6.6) 1.2 95.7 (6.0) 2.6Right 103.8 (4.5) 98.6 (5.1) 98.6 (6.2) 98.2 (5.7)

Nucleus lentiformis Left 107.3 (4.9) 3.9 110.8 (4.5) 2.5 107.6 (4.6) 3.1 111.3 (6.6) 0.3Right 111.7 (5.0) 113.6 (4.6) 111.0 (4.8) 111.7 (5.8)

Head of caudate Left 99.0 (7.0) 5.7 101.2 (5.9) 4.0 99.5 (6.6) 3.8 107.6 (8.2) –2.2Right 104.8 (6.0) 105.4 (5.5) 103.4 (5.5) 105.2 (5.2)

Thalamus Left 92.2 (4.3) 3.5 93.3 (4.0) 2.3 93.0 (4.6) 1.5 91.5 (6.5) 2.3Right 95.5 (4.7) 95.5 (4.4) 94.4 (4.5) 93.7 (5.3)

Pons 80.0 (4.4) 78.9 (4.6) 78.3 (4.5) 79.4 (3.5)

Anterior cingulate 101.0 (4.9) 98.9 (5.5) 101.7 (5.0) 101.5 (4.0)

Posterior cingulate 106.1 (5.6) 104.4 (6.0) 103.1 (5.0) 96.1 (4.2)

Unweighted average 98.3 (4.2) 1.5 98.2 (4.3) 1.1 98.3 (4.0) 1.2 98.4 (5.0) 1.4

Weighted average 100.0 (3.8) 1.1 100 (3.9) 1.0 100 (3.8) 1.6 100 (4.5) 2.0

rection set, volunteer data were adjusted by the ratio be-tween the regional VOI uptake in the Hoffman phantomfor camera 1′ (smoothed) and the regional VOI uptakefor Hoffman phantom data in the other cameras.

Figure 7 summarises the changes in transfer reproduc-ibility σij between cameras. From this figure, it can beseen that the correction by Hoffman phantom ratios fur-ther improves the between-camera reproducibility, but

that the effect is lower than that obtained from resolutioncorrection.

Phantom correction changed the anteroposterior gradi-ent between data sets 1′ and 3 to a non-significant valueof ρ=–0.20 (P=0.16). The anteroposterior gradients be-tween 1′ and 2 improved modestly (ρ=–0.49, P=0.002).

445

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

Discussion

The number of confounding parameters in the compari-son of human SPET data between different cameras andacquisitions is large and these parameters are not inde-pendent. In this work, as many parameters as possiblewere kept constant between the different camera datasets. Scatter contributions were assumed to be compara-ble, while the same pre-reconstruction attenuation correc-tion was followed by utilising the same software algo-rithm independent of camera manufacturer, with an atten-uation contour drawn automatically around the cerebrumbased on a count threshold value. It is likely that non-uni-form attenuation correction would enhance the agreementbetween the databases since, among other things, headholder effects would be incorporated more accurately.

Filtering was kept constant although the camera sen-sitivity and the total number of counts per study may

have differed slightly. The number of projections in ac-quiring SPET studies may inversely influence resolution[17], and was lower for camera 2 than for the other cam-eras. However, these settings reflect those used in olderroutine SPET measurements and was preserved to allowretrospective study analysis with the current normal data.

The issue of processing variability resulting fromstandard software on different gamma camera supplierswas addressed recently in a nationwide audit by Jarritt etal. [18]. The authors found that substantial differences inreconstructed data could occur even as a result of soft-ware version changes by the same supplier. These possi-ble processing confounders were circumvented in thisstudy by using the exact same processing method afterInterfile transfer of the parallel projection data to a singlesoftware system.

Also, the anatomical standardisation and quantifica-tion procedure followed in this work was based on repro-ducible and automated methods, in order to reduce theinfluence of these factors in the database comparison [4,19]. The co-registration accuracy for a nine-parameterlinear transformations is of the order of 1–3 mm, whichis relatively small in comparison to the VOI sizes used.In a previous study, we showed that an intrasubject six-parameter rigid transformation led to a maximal semi-quantitative error of the order of 2% for 99mTc-ECD incomparable data and with the same analysis technique[20].

The global correction for the volunteer databases asused in this study can be summarised as a three-step pro-cedure, with subsequent resolution adjustment, washout

Fig. 6. A Boxplots of the asymmetry index (relative right-to-leftdifference, %) as a function of bilateral VOI for the common dataset obtained on all three cameras (n=22). Open boxplots, camera 1′(smoothed); hatched, camera 2; horizontal lines, camera 3.B Asymmetry index (relative right-to-left difference, %) for theextended data set obtained on cameras 1 and 2 (n=91). Abbrevia-tions as previously defined

Fig. 7. Effect of the processing corrections on the between-camerareproducibility σ (%) of the healthy volunteer data sets for uni-form FWHM correction, washout correction and Hoffman phan-tom correction ratios. Closed squares, camera 2 vs camera 1(n=91); open squares, camera 2 vs camera 1 (n=22, common dataset for all three cameras); closed triangles, camera 3 vs camera 1;open circles, camera 2 vs camera 3

correction and phantom data correction, as is outlined inthe Appendix.

Regarding the resolution adjustment, SPET measure-ments were performed to support the assumption that auniform and isotropic resolution correction is a reason-able starting estimate for a camera comparison. With thisuniform and isotropic resolution correction, it was dem-onstrated that accurate transfer of Hoffman data can beachieved and only small differences between the semi-quantitative uptake measurements were found, basedmainly on differences in anteroposterior gradients. Amore detailed analysis using an inhomogeneous resolu-tion correction of the data by applying the camera andcollimator response function with iterative reconstruc-tion techniques is currently under investigation and mayimprove the agreement between the databases.

The measured SPET resolution values for camera 3were lower than would have been expected in view ofthe use of ultra-high-resolution fan-beam collimators.This may have been partly due to the relatively large ro-tation radius necessary for the operation of the linesource transmission.

For the volunteer studies, resolution adjustment result-ed in significant VOI uptake differences, especially insmall VOIs at the cerebral borders, such as the orbitofron-tal and anterior temporal cortex. However, the cerebellum,merely surrounded cranially by the occipital cortex, alsoshowed a substantially lower uptake after resolution deg-radation. This is of clinical importance, since a semiquan-tification based on cerebellar values would therefore bemore prone to variability due to differences in resolution.Increases in activity in other regions, such as the superiorparietal cortex, most likely reflect an influx of counts fromthe cuneus/occipital cortex, which is known to be one ofthe regions with the highest 99mTc-ECD uptake.

Concerning washout effects, to our knowledge onlytwo literature sources have explicitly addressed these ef-fects for 99mTc-ECD. The correction for the relative99mTc-ECD clearance described by Ichise et al. [16] wascalculated in a sample of 20 healthy volunteers with amean age of 35±10 years, employing anatomically stan-dardised subject data, an ROI analysis and curve fittingtechniques. On the other hand, Flores et al. investigatedthis matter in seven volunteers and seven patients, withmanual redrawing of the ROIs on each image [21]. Over-all, significantly lower washout values were obtained,while the relative thalamic uptake was even reported toincrease with time. For this work, the washout correctionproposed in the former study was used, since that studyincluded more subjects and employed a more rigorousquantification procedure.

The washout effect may have been responsible for apotential camera order effect in the study design, giventhat the order of scanning was not randomised to camera.This limitation was incorporated since the first data set isto be used for clinical purposes with strictly similar tim-ing in order to minimise this confounder.

For the mesial temporal cortex and pons region,which were not specifically included in the data pub-lished by Ichise et al., it was assumed that temporal neo-cortical and thalamic washout values, respectively, couldbe taken. Indirectly, this assumption was justified by thegood agreement between the databases in these regions.The overall impact of the washout correction on the be-tween-camera reproducibility was modest, as can be seenfrom Fig. 7. This is readily explained by the fact that thiseffect is largest in a few small regions, such as the thala-mus (Table 3), and smaller in regions such as the neocor-tex, which have a greater weight factor on the overall re-producibility parameter.

Since no additional physiological variability waspresent in the Hoffman phantom studies, the between-camera reproducibility was significantly better than forthe human data. However, this additional between-sub-ject variability reflects the true clinical situation, and forthe transferability of human databases it is a prerequisitethat the accompanying uncertainties are lower than orcomparable to the overall intersubject variability of per-fusion SPET measurements. The overall uncertainty inintersubject VOI uptake is of the order of 2%–8% [de-pending on the VOI size (Van Laere et al. 2000, submit-ted for publication)], and averaged 4%–5% for all threedata sets in the present study. The transfer would intro-duce an additional variability of approximately 3%,thereby resulting in an overall average uncertainty of5%–6%; this would still represent a highly accuratesemiquantitative analysis. It should be noted, however,that the aforementioned additional variability should beinterpreted as a rather conservative estimate of the errorsinvolved in the transfer of rCBF values between camerasystems, since these include uncertainties in the washoutcorrection factor due to the sequential scanning proce-dure.

Overall, the asymmetry index was most reproduciblebetween the various data sets considered. This is also ofclinical importance since many of the current clinical de-cision-making strategies are based on comparison of nor-mal versus abnormal regional differences, and smallchanges in asymmetry may have clinical significance.For example, this is the case in early dementia, wheresmall asymmetries may be present at the initial examina-tion [22]. In accordance with literature data [23, 24], aconsistent rightward asymmetry was found in normalhealthy brains.

Whereas these comparisons were done with normalvolunteers, in clinical circumstances data blurring mayintroduce false-negative results owing to increased par-tial volume effects. Although not directly the issue ofthis study, this should be kept in mind as a potential limi-tation when transferring high-resolution data to lowerresolution or analysing them at lower resolution.

In the literature, only a few studies have specificallyaddressed the transfer of data sets to other cameras usingeither the same subjects or a matched group of different

446

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

subjects. In a single-centre trial with rescanning of thesame subjects, Lobaugh et al. investigated 28 healthy el-derly volunteers (70±6 years), some on single- versusdual-headed and some on dual- versus triple-headed sys-tems [7]. Employing a procedure based on interactiveROIs obtained from one subject’s MRI scan, they pri-marily found differences between single-headed and du-al-headed camera data, a finding which may have beenconfounded by data processing procedures such as dif-ferent attenuation correction. However, for their dual-versus triple-headed comparison, they also found that noVOI reached statistical significance in the comparison ofthe rCBF ratios. The present transferability study differsin several respects. First of all, a much larger and thor-oughly screened healthy volunteer sample was used inthis study. For the whole data set, the same individualswere rescanned on the same day at carefully documentedtimes post injection. The present study also used on av-erage younger volunteers, for whom a smaller intersub-ject reproducibility has been demonstrated, both structur-ally [25] and functionally (Van Laere et al., submitted forpublication, 2000). As mentioned above, the fact that ex-actly the same reconstruction procedure was followed ona common platform reduced possible software-specificconfounding factors in reconstruction. Furthermore, datawere analysed by an automated, subject-specific and op-erator-independent analysis, resulting in highly repro-ducible semiquantitative measurements and eliminatingintra- and interobserver contributions to the overall un-certainty. In a small multicentre trial employing a semi-automated ROI analysis in different subjects [26], it wasdemonstrated that a major part of the intersubject and in-terdepartmental variability may have been contributed bythe interactive ROI definition, largely confounding actu-al camera-based differences.

In the PET literature, only a recent study by Ribeiroet al. [6] explicitly addressed this camera transfer issue.In a limited sample of 14 volunteers, they also showedthat spatial resolution matching enabled the similar cal-culation of kinetic parameters for 18F-L-dopa data ob-tained with two different types of PET scanner.

Since for 99mTc-HMPAO, inhomogeneous washout ismuch less marked [27], it may have been a relative dis-advantage to choose 99mTc-ECD for this comparativestudy in healthy volunteers. The prolonged intrinsic invitro stability, high cerebral retention, lower radiationburden, rapid blood clearance of metabolites and rapidelimination from extracerebral tissues have made 99mTc-ECD of particular interest, although these clear advanta-ges have been partially countered by a stabilised form of99mTc-HMPAO with similar in vitro stability [28]. Fur-thermore, very few large normative data sets encompass-ing both young and old subjects and using 99mTc-ECDhave been published [29, 30]. The largest data set basedon healthy volunteers to be published previously wasthat reported by Tanaka et al., whose study group com-prised 48 volunteers aged between 22 and 95 years [31].

Although the correction based on high-count Hoffmanphantom data did not enhance the between-camera re-producibility to the same extent as the resolution adjust-ment (Fig. 7), some important bias could be correctedbetween the databases. This was especially true for thethick head holder used on camera 3, where the signifi-cant anteroposterior gradient in uptake difference wasrectified after the phantom VOI correction. The fact thatthis gradient remained present between cameras 1 and 2,with very similar head support thickness and composi-tion, may indicate that other physical or physiologicalcauses (e.g. incorrect assumptions regarding regionalwashout adjustment) may also contribute to the presenceof this observed gradient.

Overall, this work has several clinical and researchimplications. From these results, it follows that the coredata set from camera 1 can be transferred as a templateto other equal- or lower-resolution gamma camera sys-tems, with or without non-uniform attenuation and/orscatter correction. This would allow its multicentre use,provided the assumptions of homogeneous resolution ad-justment and linear position-dependent washout can bejustified. The former could be measured by similar SPETresolution measurements as described here in both cen-tres taking part in an exchange of normal data, while inroutine clinical settings washout effects could be min-imised by adhering to the recommended acquisition timeof 30 min post injection [32]. On this basis it would bepossible to create a larger statistical sample of strictlycontrolled normals than could be obtained in most cen-tres locally. Such an approach would also render morecomparable studies that have been performed over a period of several years, during which time old tomo-graphs have been replaced by new machines with bettersensitivity and resolution [6]. Moreover, matching(semi)quantitative results obtained with different tomo-graphs through an automated analysis procedure wouldbe an effective way to improve interdepartmental collab-oration and would enable a better appreciation of resultspublished by different groups. It would be advantageousto compare existing normal databases from larger groupsof patients such as this one and that described by Tanakaet al. [31] in order to evaluate such interdepartmentaldatabase comparisons in more detail.

Intrinsically, the same inference could be made for re-ceptor studies, although the actual outcome measures(e.g. regional versus non-specific uptake) may have to beverified both by phantom data and on the specific equip-ment used.

Conclusion

It was shown that functional perfusion SPET data ob-tained with a high-resolution gamma camera can betransferred to lower resolution gamma cameras withoutsignificant information loss and that this can be done ac-

447

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

curately in terms of the semiquantitative outcome mea-sures. This was achieved by a first-order isotropic anduniform resolution correction. Good agreement wasfound both for phantom data and for a large series of hu-man rCBF data acquired subsequently on three differentgamma cameras. For the latter, a correction for inhomo-geneous cerebral 99mTc-ECD washout is of importancewhen comparing individual studies or templates. A prac-tical scheme for the interdepartmental transfer of normal99mTc templates could be based on equivalent 3D SPETresolution measurements followed by an appropriate iso-tropic adjustment and a Hoffman phantom between-cam-era rCBF calibration to assess the accuracy of the dat-abase transfer. Such sharing of databases or pooling ofdata would have a positive benefit on standardisation ofperfusion SPET and both its clinical and research appli-cations.

Acknowledgements. The sole financial support for this work wasprovided by a Special Research Grant of the Ghent University andthe Flemish Government (BOZF 01104699). The authors grateful-ly acknowledge the logistical computing support obtained fromNuclear Diagnostics Ltd., Sweden and Sun Microsystems, Bel-gium. The authors are grateful to Dutoit Medical and Toshiba Eu-rope, as well as to Mr. W. Hawkins from Marconi International forhis help with the Prism 3000 fan-beam-to-parallel rebinning soft-ware.

Appendix

For camera i (i=1..3), the estimate the of regional cerebral perfu-sion (rCBFij) obtained by semiquantification for VOI number j,can be written as

(4)

where Aij is the VOI count for camera i, Airef is the reference counts

(total count normalisation was used); cljpυ is the partial volume cor-

rection factor obtained for VOI j on camera 1 smoothed to the res-olution of camera i; and cij

ω is the washout correction for VOI jgiven by:

(5)

for a scan time-difference ∆tij. The washout gradient is as-

sumed to be constant for the first hours postinjection. chij is the ex-

tra correction factor obtained from the Hoffman phantom mea-surements, and is given by the ratio of the normalised VOI activi-ties Ah

lj from camera 1 and camera i:

(6)

The linearisation correction for rCBF, clj, as originally proposed by

Lassen et al. [33], was not applied here as it was assumed to beconstant between scans.

References

1. Ichihara T, Motomura N, Ogawa K, Hasegawa H, HashimotoJ, Kubo A. Evaluation of SPET quantification of simultaneousemission and transmission imaging of the brain using a multi-detector SPET system with the TEW scatter compensationmethod and fan-beam collimation. Eur J Nucl Med 1996; 23:1292–1299.

2. Bailey DL. Transmission scanning in emission tomography.Eur J Nucl Med 1998; 25:774–787.

3. Acton PD. Image analysis in brain SPECT and PET. In: Mur-ray IPC, Ell PJ, eds. Nuclear medicine in clinical diagnosis andtreatment. Edinburgh: Churchill Livingstone; 1998:613–627.

4. Friston KJ, Ashburner J, Frith CD, Poline JB, Heather JD,Frackowiak RSJ. Spatial registration and normalisation of im-ages. Human Brain Map 1995; 2:165–189.

5. Otte A. The importance of the control group in functionalbrain imaging. Eur J Nucl Med 2000; 27:1420.

6. Ribeiro MJ, Remy P, Bendriem B, et al. Comparison of clini-cal data sets acquired on different tomographs using 6-18F-L-dopa. Eur J Nucl Med 2000; 27: 707–712.

7. Lobaugh NJ, Caldwell CB, Black SE, Leibovitch FS, SwartzRH. Three brain SPECT region-of-interest templates in elderlypeople: normative values, hemispheric asymmetries, and acomparison of single- and multihead cameras. J Nucl Med2000; 41:45–56.

8. Naruse H, Daher E, Sinusas A, et al. Quantitative comparisonof planar and SPECT normal data files of thallium-201, tech-netium-99m-sestamibi, technetium-99m-tetrofosmin and tech-netium-99m-furifosmin. J Nucl Med 1996; 37:1783–1788.

9. Van Laere K, Koole M, Kaupinnen T, Monsieurs M, BouwensL, Dierckx RA. Non-uniform transmission in brain SPECT us-ing 201Tl, 153Gd and 99mTc static line sources: anthropomorphicdosimetry studies and differential influence on brain quantifi-cation. J Nucl Med 2000; 41:2051–2062.

10. Hoffman EJ, Cutler PD, Diby WM, Mazziota JC. Three di-mensional phantom to simulate cerebral blood flow and meta-bolic images for PET. IEEE Trans Nucl Sci 1990; 37:616–620.

11. Briggs GG, Nebes RD. Patterns of hand preference in a stu-dent population. Cortex 1975; 11:230–238.

12. Stodilka RZ, Kemp BJ, Prato FS, Nicholson RL. Importanceof bone attenuation in brain SPECT quantification. J NuclMed 1998; 39:190–197.

13. Talairach J, Tournoux P. Co-planar stereotactic atlas of thehuman brain. Stuttgart: Thieme Medical, 1988.

14. Slomka P, Stephenson J, Reid R, Hurwitz GA. Automatedtemplate-based quantification of brain SPECT. In: De DeynPP, Dierckx RA, Alavi A, Pickut BA, eds. SPECT in neurolo-gy and psychiatry. London: John Libbey; 1997:507–519.

15. Radau PE, Linke R, Slomka PJ, Tatsch K. Optimisation of au-tomated quantification of 123I-IBZM uptake in the striatum ap-plied to parkinsonism. J Nucl Med 2000; 41:220–227.

16. Ichise M, Golan H, Ballinger JR, Vines D, Blackman A,Moldofsky H. Regional differences in technetium-99m-ECDclearance on brain SPECT in healthy subjects. J Nucl Med1997; 38:1253–1260.

17. Kouris K, Clarke GA, Jarritt PH, Townsend CE, Thomas SN.Physical performance evaluation of the Toshiba GCA-9300Atriple-headed system. J Nucl Med 1993; 34:1778–1789.

18. Jarritt PH, Whalley DR, Cosgriff PS, Fleming JS, HoustonAS, Skrypniuk JV. IPEM/BNMS UK audit of quantitative pa-rameters obtained from nuclear medicine SPET software [ab-stract]. Eur J Nucl Med 2000; 27:942–942.

448

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001

rCBFijij

iref j

lljp

ij ijh

A

Ac c c c= ⋅ ⋅ ⋅ ⋅υ ω

cHt

tijj

ijω =

∂∂ ⋅∆

cA A

A Aijh lj

hih ref

lh ref

ijh

=⋅

⋅

,

, .

∂∂Ht

i

19. Fox PT, Mintun M, Reiman EM, Raichle ME. Enhanced de-tection of focal brain responses using intersubject averagingand change-distribution analysis of subtracted PET images. JCereb Blood Flow Metab 1988; 8:642–653.

20. Van Laere K, Koole M, D’Asseler Y, et al. Automated stereo-tactic standardization of brain SPECT receptor data using sin-gle-photon transmission images. J Nucl Med 2001; 42:inpress.

21. Flores LG, Jinnouchi S, Nagamachi S, et al. Retention of Tc-99m ECD in delayed SPECT of the brain. Ann Nucl Med1999; 13:1–4.

22. Celsis P, Agniel A, Cardebat D, Demonet JF, Ousset PJ, PuelM. Age related cognitive decline: a clinical entity? A longitu-dinal study of cerebral blood flow and memory performance. JNeurol Neurosurg Psychiatry 1997; 62:601–608.

23. Markus HS, Ring H, Kouris K, Costa DC. Alterations in re-gional cerebral blood flow, with increased temporal interhemi-spheric asymmetries, in the normal elderly: an HMPAOSPECT study. Nucl Med Commun 1993; 14:628–633.

24. Krausz Y, Bonne O, Gorfine M, Karger H, Lerer B, Chisin R.Age-related changes in brain perfusion of normal subjects de-tected by 99mTc-HMPAO SPECT. Neuroradiology 1998; 40:428–434.

25. Jernigan TL, Archibald SL, Berhow MT, Sowell ER, FosterDS, Hesselink JR. Cerebral structure on MRI. Part I. Localiza-tion of age-related changes. Biol Psychiatry 1991; 29:55–67.

26. Van Laere K, De Sadeleer C, Dobbeleir A, Bossuyt A, DeDeyn PP, Dierckx RA. Quantification of Tc-99m HMPAObrain SPET in two series of healthy volunteers using different

3-headed SPET configurations: normal databases and method-ological considerations. Nucl Med Commun 1999; 20:1031–1040.

27. Asenbaum S, Brücke T, Pirker W, Pietrzyk U, Podreka I. Im-aging of cerebral blood flow with technetium-99m-HMPAOand technetium-99m-ECD: a comparison. J Nucl Med 1998;39:613–618.

28. Barthel H, Kampfer I, Seese A, et al. Improvement of brainSPECT by stabilization of Tc-99m-HMPAO with methyleneblue or cobalt chloride. Comparison with Tc-99m-ECD. Nu-klearmedizin 1999; 38:80–84.

29. Kuji I, Sumiya H, Niida Y, et al. Age-related changes in thecerebral distribution of 99mTc-ECD from infancy to adulthood.J Nucl Med 1999; 40:1818–1823.

30. Barthel H, Wiener M, Dannenberg C, Bettin S, Sattler B,Knapp WH. Age-specific cerebral perfusion in 4- to 15-year-old children: a high-resolution SPET study using 99mTc-ECD.Eur J Nucl Med 1997; 24:1245–1252.

31. Tanaka F, Vines D, Tsuchida T, Freedman M, Ichise M. Nor-mal patterns on Tc-99m-ECD brain SPECT scans in adults. JNucl Med 2000; 41:1456–1464.

32. Juni JE, Waxman AD, Devous MD Sr, et al. Procedure guide-lines for brain perfusion SPECT using technetium-99m radio-pharmaceuticals. Society of Nuclear Medicine. J Nucl Med1998; 39:923–926.

33. Lassen NA, Andersen AR, Friberg L, Paulson OB. The reten-tion of [99mTc]-D,L-HMPAO in the human brain after intraca-rotid bolus injection: a kinetic analysis. J Cereb Blood FlowMetab 2000; 8 Suppl 1:S44–S51.

449

European Journal of Nuclear Medicine Vol. 28, No. 4, April 2001