IEEE Transactions on Nuclear Science, Vol. NS-30, No. 1, February 1983
PERFORMANCE EVALUATION AND CALIBRATION OF THE NEURO-PET SCANNER
Victor J. Sank, Rodney A. Brooks, Walter S. Friauf, Stephen B. Leighton, Horace E. Cascio,and Giovanni Di Chiro
National Institutes of HealthBethesda, Maryland 20205
Summary
The Neuro-PET is a circular ring seven-slicepositron emission tomograph designed for imaging humanheads and small animals. The scanner uses 512 bismuthgermanate detectors 8.25 mm wide packed tightly togeth-er in four layers to achieve high spatial resolution(6-7 mm FWHM) without the use of beam blockers.Because of the small 38 cm ring diameter, the sensitiv-ity is also very high: 70,000 c/s per true slice withmedium energy threshold (375 keV) for a 20 cm diameterphantom containing 1 pCi/cc of positron-emittingactivity, according to a preliminarymeasurement. Thereare three switch-selectable thresholds, and the sens-itivity will be higher in the low threshold setting.
The Neuro-PET is calibrated with a round orelliptical phantom that approximates a patient's head;this method eliminates the effects of scatter andself-attenuation to first order. Further softwarecorrections for these artifacts are made in the recon-struction program, which reduce the measured scatterto zero, as determined with a 5 cm cold spot. With a1 cm cold spot, the apparent activity at the center ofthe cold spot is 18% of the surrounding activity, whichis clearly a consequence of the limits of spatialresolution, rather than scatter. The Neuro-PET hasbeen in clinical operation since June 1982, and approx-imately 30 patients have been scanned to date.
Descri ption
The Neuro-PET scanner1'2 began routine clinicaloperation in June of 1982; at the time of writing about30 patients have been scanned. This 4-ring 7-slicepositron emission tomograph was designed to fill theneed for a high resolution, high sensitivity scannerat the National Institutes of Health for human headand animal studies. At the present time the electron-ics for three rings is operational, permitting simul-taneous scanning of five slices; the remaining elec-tronics chassis containing discriminators for thefourth ring will be installed shortly.
To achieve the mutually desirable goals of highresolution and high sensitivity, we designed each ringto contain 128 tightly packed bismuth germanate cryst-als 8.25x20x35 mm around a 38 cm circle (inside diam-eter). The reason for the small ring diameter is toincrease the sensitivity, an extremely important cons-ideration when doing high resolution scanning. Inaddition, we didn't want the number of detectors per
ring to exceed 27.
The collimator assembly contains three washer-shaped disks of depleted uranium 3 mm thick, for reduc-ing the number of scattered and out-of-plane detectedgamma rays. This assembly reduces the patient apertureto 25 cm diameter, which so far has presented no diffi-culties in patient scanning, although the dental chairused for patient support has not been adequate for a
few patients who could not sit up or lie in a supineposition. One study done without the collimator hasproduced images of good quality, so that this modalitymay be used when a larger patient aperture is required.
Crystal face center-to-center spacing is 9.3 mm,so that a scanning motion is required to produce thesampling needed for maximum spatial resolution.
Details of the wobble motion used have been reported
previously. This motion, which is under the controlof a single SLOSYN stepper motor connected with a chaindrive to two of the three eccentric shafts, is softwarecontrollable and has performed flawlessly to date. Wecurrently use an 8 mm wobble diameter with 8 discretedata acquisition points spaced uniformly around thewobble circle.
The random coincidence count rate is a function ofthe coincidence resolving time T of the electronicdetectors, as well as of the single gamma count rateper detector. The Neuro-PET offers two switch-select-able resolving times, although all studies to date havebeen done with the longer T of 12 ns. For dynamicstudies with very high count rates the 5.5 ns T may beused to further reduce the random coincidence rate,with a sensitivity reduction of 39%.
Performance Evaluation
Spatial resolution
Spatial resolution of the Neuro-PET was measuredby scanning a line source placed in a 20 cm diametersolid plastic phantom. The source contains 68Ge en-closed in a 3 mm diameter steel jacket to suppress thepositron range; thus the results are fully representa-tive of imaging with the isotope 18F, which has a verysmall positron energy. All measurements were made onthe reconstructed image without any software smoothing.This is the same reconstruction method used for patientimages, although at the present time some patientimages are later smoothed to reduce statisticalnoise due to limited counts. The need for this smooth-ing is expected to disappear when all seven slices areoperational.
The full width at half maximum (FWHM) of thereconstructed line source image was found to be 6 mmat the center of the field of view, and 7 mm if thesource is placed 9 cm off center. These values applyto both true and cross slices. In the near future weplan to increase the number of wobble points from 8 to9, and to improve the method of interpolation froma linear approach to a multi-point polynomial approach.We believe that these improvements will enable us toachieve a resolution approaching 5 mm at the center ofthe field of view. In addition, we plan to provideoptional beam blockers which will reduce the size ofthe detector apertures and thereby offer a resolut-ion capability of 3-4 mm, although at the expense ofreduced sensitivity. The beam blockers will be design-ed primarily for use with animal studies.
SensitivityOur sensitivity measurements were made using the
conventional 20 cm diameter phantom containing a unif-orm distribution of activity. The phantom had a thinstainless steel wall to reduce the effect of wall atten-uation, and the count rate was kept low enough that theeffects of random coincidences and dead time were
negligible. When the count rate was divided by theactivity concentration, we obtained a sensitivity of70,000 (c/s)/(pCi/cc) for a single true slice, usingthe medium energy threshold. This value is higher thanwe had anticipated, based on measurements reported withprototype components.2 Part of the reason for the
U.S. Government work not protected by U.S. copyright.
636
discrepancy is that the earlier phantom had 1 cm thickplastic walls, and part may be due to the method ofcalibrating the activity. Until we investigate thematter further, the values reported herein must beregarded as tentative.
The sensitivity measured for a single cross-sliceis 80,000 c/s, providing a total sensitivity for allseven slices of 520,000 (c/s)/(pCi/cc) at mediumthreshold. The sensitivity at low and high energythresholds has not yet been determined, but ourpresent feeling based on patient studies done to dateis that the improvement in sensitivity offered by thelow energy threshold will be of significant help inreducing statistical noise, and that this modalitywith itsincreased sensitivity may become more widelyused.
Scatter
The effect of scattered coincidences was evalua-ted using cylindrical "cold spots" placed in a 20 cmdiameter phantom filled with a uniform distributionof activity. One cold spot was a 5 cm diameter plastictube filled with water; the other was a 1 cm diameterplastic rod. After scanning the phantom, the imagewas reconstructed the same way as patient images, andthe apparent activity in the center of the cold spotwas measured. The 5 cm cold spot was scanned in thecenter of the phantom, and the 1 cm cold spot wasscanned approximately 7 cm from the center.
The results showed an apparent activity of zerofor the 5 cm cold spot, not only at the center of thespot, but throughout the entire area. For the 1 cmcold spot the apparent activity at the center was 18%of the surrounding activity. This 18% figure obviouslyis due to the tails of the line spread function, ratherthan to scatter, since the 5 mm radius of the rod isless than the FWHM of the line spread function. Theseresults show that the scatter correction program inthe reconstruction routine is operating properly.
To evaluate the scatter without software correc-tion, reconstructions were made in which the correctionwas omitted. In this case the apparent activitieswere 19% for the 5 cm cold spot and 33% for the 1 cmcold spot. We therefore conclude that, at the mediumenergy threshold, the scatter contribution in theNeuro-PET scanner is 15-20% before software correctionand zero afterward. However, we must emphasize thatthe present scatter correction program is not verysophisticated and will not perform as well when thedistribution of activity is more complicated than asingle cold spot. A more accurate program is underdevelopment.
Patient images
Approximately thirty patients have been scannedto date with 18F-deoxyglucose (FDG), a radiopharmaceu-tical which is taken up in proportion to the glucoseutilization rate of the tissue. These images thusprovide a metabolic map of the brain. The Neuro-PEThas produced images with excellent delineation of thecerebral cortex and separation of the basal ganglia,and has proven particularly invaluable in revealingthe very thin rim of malignant tissue that sometimesis found in gliomas that possess a large necroticcenter.
Figs. 1-5 illustrate typical results obtainedfor both normal and pathological structures. Becauseof limited counts, these pictures have been smoothedslightly, using the SMOOTH option of the VIEW program.The resultant spatial resolution has thus been degraded
637approximately 1 mm beyond the values quoted earlier.We expect that the need for smoothing will disappearwhen all seven slices are operational, so that longerscans can be made. The scan time for most of thesestudies waslO minutes. We should also mention thatquantitative measurements made with the Region of Int-erest program are based on unsmoothed image numbers,and so are not degraded by the smoothing which isdone solely for visual appearance.
Fig. 1. Tomogram through the convexity of the brainshowing good delineation of the peripheral and mid-line cerebral cortex. Note in particular the inter-hemispheric fissure and the clear separation betweenwhite matter (dark areas) and gray matter (bright).
Fig. 2. Tomogram at the level of the basal ganglia.Note also the separation of the right and left post-erior cortex.
Fig. 3. Scan showing the temporal lobes and thecerebellum. Note the distinction between whitematter + temporal horns (darker areas) and cortexin the depth of the temporal lobes.
638
Fig. 4. Malignai- ylioma (bright spot) in theanterior left hemisphere. The reduced intensityin the greater part of this hemisphere is dueto edema associated with the tumor.
Fig. 5. A glioma in the left hemisphere exhibit-ing a large necrotic center (dark area) outlinedposteriorly by a thin bright rim of viable tumoraltissue.
CALIBRATION
The Neuro-PET detectors are calibrated in a uniquemanner, as compared with other PET scanners. Mostother scanners use an annular phantom, or a sheet phan-tom, in order to apply a known amount of radioactivity
between each pair of detectors.4'5 Subsequent softwarecorrections are then made during the reconstruction ofpatient scans for scatter, self-attenuation, and randomcoincidences.
The approach used with the Neuro-PET is similar tothat used in some CAT scanners, in that a cylindricalor elliptical phantom approximating the shape of a
patient's head is used for calibration. CAT scannershave a problem with beam-hardening artifact which isanalogous to self-attenuation in PET in that both, ifuncorrected, produce a "cupping" artifact wherein theintensity at the center of the image is reduced in in-tensity. In addition, scattered photons are a problemwith both modalities, adding a background haze to theimage. Several CAT manufacturers have found that by
calibrating the detectors with a round phantom of suit-able diameter, these problems can be eliminated tofirst order. Software corrections can then be made forthe residual error, which arises from the deviationsbetween the actual patient geometry and that of thecalibration phantom.
Up to the present time we have been using a 20 cmdiameter cylindrical phantom, although the softwareprograms allow for an elliptical phantom as well. Thecalibration procedure is as follows: The phantom isplaced in the gantry and centered with respect to thedetector ring, which is held stationary (no wobble)during the calibration scan. A typical calibrationmay last 30 minutes or more, to insure that thestatistical inaccuracy in the calibration coefficientsis substantially less than statistical fluctuations inpatient scan data. Coincidence counts acquired duringthe calibration are stored in extended memory of theData General Eclipse S/250 computer, using the samedata acquisition system as for patient scanning. Thecalibration program then performs the following calcu-lations for each detector pair (see also Fig. 6):
Random coincidences. Random coincidences cannot beincluded in the calibration coefficients, as is donefor scattered coincidences, because the random rateis proportional to the square of the true countrate. The total random coincidences for each detec-tor pair are calculated from the single count ratesof the individual detectors and subtracted from theaccumulated counts.
Decay correction. The number of counts is decaycorrected to the time at which the activity concen-tration was measured. This decay correction takesinto account the exponential nature of the decreaseduring the calibration scan. This is important be-
cause 68Ga, with a half-life of only 69 minutes, isoften used for calibration.
Path integral. The activity concentration (nCi/cc)is measured in a well counter and entered into thecomputer. The computer then calculates the pathlength through the phantom of the ray connecting thetwo detectors, using the proper geometrical formulafor an ellipse. The product of the activity concen-tration and the path length gives the ideal pathintegral (nCi/cm2) for that ray.
Calibration coefficient. The ratio of the correctedcounts per second to the ideal path integral is thencalculated and stored in a disk file called CAL.
The above calculation is done for 28,672 detectorpairs (7 slices x 128 views per slice x 32 detectorpairs per view). The cross slice coefficients actuallycontain contributions from two detector pairs, e.g.,IA-64B and 1B-64A, where A and B refer to two adjacentrings; however the calibration coefficient is calcula-ted for the combined count rate, not for the individualpairs. In addition, the Neuro-PET offers 3 switch-selectable energy thresholds and 2 coincidence resolv-ing times, making a total of six possible combinations,and calibration coefficients must be obtained for eachsetting. All of our present patient scans have beendone with medium threshold and the long coincidenceresolving time.
The advantage of this method of calibration isthat the coefficients automatically include the effectsof scatter and self-attenuation for a uniform distribu-tion of activity, as well as the geometrical and physi-cal sensitivity factors for each detector. Our philo-sophy is to include as much as possible into the cali-bration so that the amount of correction that must be
7s/,it
l2S vMet*
-F, ,i.
:EL- , t F t r r .9X +, '. 2+' rIr;ucc'r_ct; L*: ti 'r A. _ ,..Y . .D/I C2.
+E T e En~S ci;, ( t-r twl;EA -:, si +u H ; 'u; CX S : . ,t _; , H) e' ** T -_-AEr-'W:Ti$.J
0 4, 7 T 7-C,
-~~ ~ ~ ~~TCF:f -~-}.t~@ r9 - i , i t t+ , ,*``S
wm~~~~~-_.,t,-;;LL D ,it .'-~tI
hr -
D D 4 ' W A C i.+.J " # '._r' = Si'k ~
F~~i i?&-
_.f i+-
-S _ .i l S+ C20t tj- S- Sr 'S E.
.7 ei2L,iarfC,
2 "C.r' 4_c- ..,
_~~~ ~ ~ ~ ~ ~ ~ _f _i ht' rz r.p; (ID..t.-R ,,I 1'ID. LE Ci- ,-;""EI )O t~~~~~~~~~~~~~~~~~~~~~~~~~~~' Ft'Ef fi,,_;4nL c, ri L.tUA
,~ ~ ~~ ~ ~ ~ ~~~~~~.ID.JS_,_ ,. rSr" C, s-ft ,' cr' tw'e E ,f*t.
jT-R*.D = J t , i, i r-i i UE> A / ;
CF~~~LLrI 1E6 bF1~ irj CA-L a t'e ICItic(J
7 tjY7'
J.qE E + *I
Fig. 6 Portion of calibration program in which coefficients CAL(ID,J) are calculated forthe IDth detector pair in the Jth view. The raw counts are stored in four pagesof extended memory and accessed through an array WINDOW(ID,J1), where ID is thenumber of the detector pair and J1 is the view number within the page.
applied later during reconstruction is held to a mini-mum. By including first-order scatter and attenuationin the calibration coefficients, we do not ignore theseproblems -- we take them into account in the simplestmost direct manner.
Peripheral calibration
The disadvantage of the above calibration methodis that it applies only to those detector pairs whichtransect the phantom. If the phantom is at least asbig as the largest patient head, this is not a problem,as long as the patient is properly centered. Howeverit is clearly desirable to have a proper calibrationfor the entire 25 cm aperture, i.e., for all 32 detec-tor pairs per view. At the time of writing, this hasnot been accomplished, but we have the followingapproaches in mind for extending the calibration. Oneidea is to use a separate annular phantom, as othershave done. Another idea is to use the scatter skirtsof the round phantom for calibrating the peripheraldetectors. In either case, the problem will be todevelop a smooth transition between the two areas ofcalibration, since the inner detector pairs includethe scatter in the coefficients, and the outer oneswill not.
References
1. RA Brooks, VJ Sank, G Di Chiro, WS Friauf, SB Leigh-ton, "Design of a high resolution positron emissiontomograph: The Neuro-PET". J Comput Assist Tomog 4,5-13 (1980)
2. RA Brooks, VJ Sank, WS Friauf, SB Leighton, HECascio, G DI Chiro, "Design considerations forpositron emission tomogrdphy". IEEE Trans BiomedEng BME-28, 158-177 (1981)
3. RA Brooks, VJ Sank, AJ Talbert, G Di Chiro, "Samplingrequirements and detector motion for positron emiss-ion tomography". IEEE Trans Nucl Sci NS-26 (1979)
4. M Bergstrom, L Eriksson, C Bohm, G Blomqvist, "Aprocedure for calibrating and correcting data toachieve accurate quantitative values in positronemission tomography". IEEE Trans NuCd Sci NS-29,555-557 (1982)
5. DC Ficke, DE Beecher, GR Hoffman, JT Hood, J Markham,N Mullani, MM Ter-Pogossian, "Engineering aspects ofPETT VI". IEEE Trans Nucl Sci NS-29, 474-478 (1982)
639