IEEE TRANSAmIONS ON NUCLEAR SCIENCE, VOL 37,NO.

2, APRIL 1990 783

Measuring PET Scanner Sensitivity: Relating Countrates to Image

Signal-to-Noise Ratios using Noise Equivalent Counts

S.C. Strother*', M.E. Casey' and E.J. Hoffman'

*Neurology Department, Memorial Sloan-Kettering Cancer Center, New York, New York 10021
'CTI PET Systems, Inc., 810 Innovation Drive, Knoxville, Tennessee 37932
'Laboratory of Nuclear Medicine (DOE), UCLA School of Medicine, Los Angeles, California 90024

Abstract
Sensitivityparameters derived from a plot of a scanner's as it eliminates the 25%-30% change in the minor to major
true coincidence count (TCC) rates as a function of activity axes of the elliptical cross section of the average human
in a 20cm cylindrical phantom have no direct link to image head. More experimental work is needed to address these
quality. Noise equivalent count (NEC) rate curves, which issues.
incorporate the noise effects of subtracting the randoms and However, even if a set of standard phantoms is available
scatter count components provide a direct link between the issue of what performance parameters to measure
image signal-to-noise ratios and the scatter, randoms and remains. We suggest, as have others [6-91 that for the
trues coincidence count rates. We measured TCC and NEC purposes of measuring PET scanner sensitivity,image quality
curves with a standardized 20cm diameter nylon cylinder for should be defined in terms of the reconstructed image signal-
five different CI'I/Siemens PET scanners with several to-noise ratio (SNR) in a uniform object which is directly
scanner-collimator combinations: (1) 831/08-12 with lmm related to a PET scanner's noise equivalent count perfor-
collimator septa; (2) 933/08-12 and 933/08-16 with 9 1 m m mance. The basic NEC expressions are obtained below from
tapered collimator septa; (3) 931/08-12 with P l m m tapered a simple tomographicS N R derivation and are then extended
and a lmm collimator septa and the 931/08-16 with 9 1 m m to include the additional noise effects of realistic scatter and
tapered collimator septa. In addition, we compared TCC randoms corrections.
and NEC curves on the 933/08-12 with those from an The concept of noise effective (or equivalent) counts
Alderson brain phantom. In general we found that the TCC derives originally from the work of Shaw [lo]. The idea has
curves indicated peak count rates and activity levels that been used in the development of a unified theory of radiation
were as much as 50% higher than the corresponding values imaging and as a measurement of noise equivalent quanta
from NEC curves. The primary factor causing this diff- (NEQ) has been applied to the assessment of imaging
erence is the noise effect of the randoms component. In efficiency in forming a figure of merit for S N R transport
addition, we demonstrated that compared to the Alderson through an imaging system. This figure of merit, called the
brain phantom the standard 20cm cylinder is a poor pred- detective quantum efficiency (DQE) has been applied to a
ictor of count rate performance for PET brain imaging. wide range of medical imaging devices [ll].

Introduction Theory
The 20cm diameter cylindrical phantom has become Consider the S N R of a reconstructed image element
the de facto object for measuring PET scanner sensitivity. following the approach of Brownell et al. [12].
The derived performance measurements range from the low
activity sensitivity specification - count rates in kcps/pCi/cc SNR = c te(vAFt,)-'/2
typically measured at about .1 pCi/cc - to the phantom
activity levels for (1) 50% deadtime, (2) saturation and (3) where c = a constant [13].
randoms equal to trues (plus scatter). For brain scanning te = true coincidence counts in a reconstructed
the use of the 20cm diameter cylinder has been questioned image element.
by Links and Wagner [1,2] in favor of a 15cm diameter and VAR, = L(weighted variance of the samples from
several authors have reported PET scanner sensitivities and each of the m projection angles contri-
scatter fractions for a range of diameters from 14cm-20cm buting to the image element).
[3,4]. None of these test cylinders bears a particularly close
resemblance to the elliptical dimensions of the average adult The coincidence components without attenuation and dead-
human head [5] - Males:19.6cmxl5.5cm; Females:18.5cmx time correction, in each projection sample are - trues(tp),
14.5cm. A test cylinder diameter of 7 inches or 17.8cm, scatter (s ) and randoms (rp). The variance of m projection
while better than 20cm may still not be a very good standard samples [t, t sp + rp)-(spt rp)], is combined to give

1
Present Address: PET Imaging Section (llp), VA Medical Center, 1 Veterans Dr., Minneapolis, MN 55417

0018-9499/90/0400-0783$01.000 1990 IEEE

784

total counts within the complete field-of-view (fov). When

VAR, = &,.,wm(tp+sp+rp) calculating NEC curves particular care must be taken to
reduce random rates for the fov to those that apply to the
= &,.,wmtp(l+asp
+arp) (2) appropriate
__ cylinder diameter.
where asp = s /tP and arp = rp/tp are respectively the
& for the expression in equation 2, equation 6 requires
subtraction of noiseless estimates of the image scatter and
scatter and ranioms fractions for a projection sample at a
randoms coincidence signals. This is approximately true for
particular angle. Note that this variance expression assumes
randoms when they are calculated from directly measured
that sp and rp are accurately known without any significant
singles count rates or when the smoothing technique
measurement noise. Consider the S N R in the center of a
described by Casey and Hoffman [14] is used with an offset
reconstructed image of a uniform cylinder of activity, with
coincidence window. However, when scatter is corrected as
diameter D and image elements (dxd), then by symmetry
a deconvolution procedure or the offset coincidence window
technique is used to estimate randoms, these measurements
VAR, = w(Dt,ac/d)( 1+ asp+arp) (3) are inherently noisy and the additional noise components
must be added into VAR, in equation 2.
since for all m projection angles tp = (DtJdmaJ with an
For randoms estimated with an offset window the mean
attenuation and deadtime correction factor for the center of
randoms component (rp) has a variance of r . The noise
the cylinder, ac. For T, the total true image counts without
deadtime and attenuation correction
variance term in equation 2 therefore becomes gP. Similarly,
the process of deconvolving the scatter component from each
projection may be approximated as subtracting a constant
t, = avg(aJT/(KD2/4d2) (4) fraction of the (tp+sp) signal that remains after randoms
subtraction. This fraction may be written as a (t +s )
where avg(aJ is the average attenuation and deadtime
where a = s /(tp + sp) =asp/(l + asp). How:$erp e!lt
correction factor for a projection. Therefore, from equations .cP
(tP+ sp> signal gas a variance of (tP+ sR + rp>or (tp+ sp + 2rp>
1,3 and 4 with w-l12 absorbed into c
depending on the randoms correction procedure used.
SNR = m(d/D)3/2[avg(aJ/aJ 1/2 Therefore, the variance of the noise component added as a
result of deconvolving the scatter is given by
x [~/(1+asp+arp)l''~ (9 (aCd2(t +s + r ) or (ac )2(t + s +2rp). This leads to the
P P P
NEC = T/(l+asp+arp) (6)
following expression for h8in Perms of single slice count
rates for true (T), random (Ri) and scattered (Si)
Equation 5 is in the same form as the expressions for coincidences with acp, aspand arpapproximated by ad, ad
tomographic reconstruction noise given by Budinger et al. and arirespectively.
[l3,14] with the addition of the object dependent attenuation
and deadtime correction factor and the NEC expression for NEC = +/[(1+aEi)(T+Si+2Ri)]
image counts. The noise equivalent counts expression of
= T/[(1 +aEi)(l+ ad + 2clri)] (7)
equation 6 is in the general form given by Derenzo [6],and
by Lupton and Keller [SI where it is referred to as the Methods
quality factor (a). NEC can be seen to be the reduced T
Data Collection
that without randoms and scatter components, i.e.,
asp= arp=0, produces the same image S N R at the center of
"F, l l C and 13N were used with different scanners to
create uniform flood activity sources in standardized 2Ocm
the cyhder as the true coincidence rate obtained by
diameter by u)cm long nylon phantoms. The total activity
subtracting randoms and scatter components from the total
loaded into the phantom was measured in a well counter and
coincidence count rate.
divided by the phantom's volume to obtain the initial activity
Since we typically do not know the random and scatter
concentration. The activity in the phantom was allowed to
fractions in a projection element, equation 6 must be
decay over a wide activity range while collecting multiple
rewritten in terms of the total image random and scatter
lu)-300sec image frames. Total random and true+scatter
fractions, ariand ad respectively. To the extent that arpand
coincidences per frame were converted to count rates by
asp are constant across a projection they may be
dividing by the frame time. The llC and 13N data sets were
approximated by ariand ad respectively. This substitution
adjusted for decay during the frame. In all cases the offset
was investigated by Lupton and Keller [SI for the Therascan
3128 PET scanner and was found to be a good coincidence window was used to measure random
coincidences. All data sets were obtained without a headrest
approximation provided the random coincidence component
was not substantially greater than the trues component. to eliminate additional attenuation effects.
Note that ariand a refer only to the scatter and ;andom Measuring Scatter Fractions
d.
counts within the cyhder boundaries, as distinct from the Scatter fractions (scatter/trues) were measured for most
 785

scanner-collimator combinations as the ratio of the counts elliptical cross-section of 19cmxl5cm, we choose a, = 0.10
under the exponential tails to the central peak counts, for a based on the 18cm cylinder value for the 831 PET scanner
scatter proftle obtained from a rod source in the center of a [l5],and f=0.36 (for an 18cm diameter).
2Ocm diameter water filled phantom [SI. The closest values Axial Inter a,=
of a,, (2Ocm cylinder) available for most scanners integrated Scanner Ring Crystal Septa Septa Septa (Scatter/
the scatter tails out to l5cm from the center of the profile. Model Diam. Width Depth Width Gap' TNS)
For two data sets (see Table 1)it was necessary to estimate 931/06-l2 l a m 12Anm 18Scm 3-lmm 105mm 0.12
the scatter fraction from the values measured for other 93l/oEl2 " " lmm l25mm 0.14'
scanner-collimator combinations using the approximations 931/06-16 " 15.4mm " 3-1mm UOmm 0.15
[6,8], a, a (septa depth).' and asa (inter-septa gap).
?33/06-l2 76cm 12%m l25cm 3-1mm 105mm 0.18
Calculating NEC Rates 933/06-16 " 15Amm m r n " UOmm 025
All NEC curves were calculated using equation 7 with
aci= 0, asi=agfrom Table 1and R, = fR. The randoms rate 831/CXLl2 64cm 12Anrn 145cm lmm l25mm 0.18'
'Exparcd crystal width
(R) was reduced by the fraction of a projection subtendedby
the 2Ocm cylinder (f):(l) 831, f=0.5; (2) 933, f=0.4; (3) 931, Table 1: Relevan! physical parameters of the PET
f=0.33 . For the Alderson brain phantom with a maximum scanners tested. Calculated values, see text.

Results
Figure 1: The mean true and random coincidence count curves for cross and direct slices, and the derived noise equivalent
count (NEC) curves (see text) for the 831/08-12, 933/08-12 and 933/08-16 scanning a 2Ocm diameter phantom.
831/08-12, lmm Septa
12500 12500

loo00 10000

7500 7500
Counts/ Counts/
Second Moo
Second 5000

2500 2500

0 0
(a) 0 0.5 1 1.5 2 2.5 3 3.5 (b) 0 0.5 1 1.5 2 2.5 3 3.5
microCuries/cc microCuries/cc

933/08- 12, 3mm Tapered Septa 933/08- 12. 3mm Tapered Septa
12500 12500

10000 low0

7500 7500
Counts/ Counts/
Second 5wo Second 5ooo

2500 2500

0 0
(c) 0 0.5 1 1.5 2 2.5 3 3.5 (a 0 0.5 1 1.5 2 2.5 3 3.5
microCuries/cc microCuries/cc

933/08- 16, 3mm Tapered Septa

12500

10000

7500
Counts/ Counts/
Second Second 5000

2500

0
0 0.5 1 2 2.5 3 3.5 0 05 1 1.5 2 2.5 3 3.5
(e) 1.5
(0
microCuries/cc microCuries/cc
786

931/08-12, 3mm Tapered Septa 931/08-12. 3mm Tapered Septa

17500 17500
lM00 1KY)o

12500 12500

Counts/ Counts/ loaM

Second 7500 Second 7500
5ooo 5oOo
2500 2500

0 0
(a) 0 1 2 3 4 5 6 7 (b) 0 1 2 3 4 5 6 7
microCuries/cc rnicroCuries/cc

931/08-12. lmrn Septa

17500 17500
15oM) 1KY)o

12500 12500

Counts/ loo00 Counts/ loaM

Second 7500 Second 7500
m m
2500 2500
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
(c) (dl
microCuries/cc microCuries/cc

931/08-16. 3mm TaDered SeDta 931/08- 16, 3mm Tapered Septa

17500 17500
lKKl0 15ooo

12500 12500

Counts/ lWa0 Counts/ lOoa0

Second 7500 Second 7500

5ooo 5ooo
2500 ?500

0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
(e)
rnicroCuries/cc
(0 microCuries/cc

Figure 2:The mean true and random coincidence count curves for cross and direct slices, and the derived noise equivalent
count (NEC) curves (see text) for the 931/08-12and 931/08-16scanning a 2Ocm diameter phantom.

The slopes of the early portions of the TCC curves in figures Scanner Septa scadtivity-Trues. pc9AEtivity
1 and 2 are given in Table 2. This trues sensitivity at low Model Width Cross Direct Total Trurs@pCi/cc NE@pCi/cc
activity increases by a factor of 2.0with increasing inter-septa pn/aslz 3.lmm 881r 4- 1Wr 1wr 6.0 WL 4.0
gaps and decreasing ring diameter, a trend which is slightly 9 3lm2 lmm 1 U
l 5Jk 119L 158ir 4.0 9sL 27
decreased if the trues sensitivities are converted to NEC !l3lml6 3 . l m m U S L 7% 1ssL 1% 35 105L 2.2
rates using the asfractions of Table 1. The maximum factor
?
33m
l23.lmm
of 1.8 change in the NEC sensitivity over all scanners at low
2oereglindtr UBr 75k 1ssL m 2 5 75kl.6
activity, corresponds to a 41.8 or 36% SNR advantage based
Ht.apbanhm loslr 6.Ok WL 164k 4.1 118k 3.1
on equation 5. Comparing cross versus direct slices
between scanners, the maximum sensitivity change betweem mm16 s l m m 182k loak 2U7k l2tk 1.7 67k 1.1
the direct slices of the 931/08-12and the cross slices of the 8
31m
l2 lmm l4& 95k 178k 107k l.7 62k 1.1
933/08-16is 3.7 for trues, and 3.3 for NEC’s, a 82% S N R ‘cps/&i/cc @ O.lpCj/cc
advantage. However, if sensitivities are expressed per mm
of axial FWHM resolution [6]the advantage of the 933/08- Table 2 For all PET scanners tested the measured values
16 is reduced by approximately 7/9=0.8 [16]. Expressing for:(l) low activity sensitivity (trues for cross, direct and
sensitivity per mm of axial FWHM resolution makes the total);(2) saturation counting levels of true and NEC count
831/08-12the most sensitive scanner at low activities. rates @ phantom activity levels.
 787

The low activity sensitivity for the head phantom compared direct slices is 4.9-7.lkcps and 3.3-4.lkcps respectively,
to the 20cm cylinder in Table 2 is decreased by 20% because corresponding to SWR advantages of 20% and 11%.
of the smaller volume of activity and increased attenuation The efficiency with which increased activity is used at mid
due to the skull. to high activity levels may be estimated by comparing the
Figures 1 and 2 demonstrate the much lower peak average NEC rates for a given scanner’s cross or direct slices
counting (saturation) rates and phantom activity levels at at the peak counting rate activity and at half that activity.
which these rates occur for the NEC curves compared to the The ratio for any pair of cross slice or direct slice rates at
TCC curves. These results are summarized for total trues these two activities, ranges from 1.19-1.23 giving a
and NEC counting rates in Table 2. In contrast to the low consistently small S N R advantage of about 10% for a
activity sensitivitiesthe peak rates and corresponding activity doubling of the activity concentration. Finally consider the
levels tend to decrease with increasing inter-septa gaps and relative cross/direct slice efficiency advantage for each
decreasing ring diameter. individual scanner tested. For the 4 PET scanners with 3mm
In particular consider the change of collimator on the tapered septa the cross/direct slice ratio is between 1.7 and
931/08-12. When moving from the 3mm tapered septa to 1.8 for the activity range up to the peak counting rate for
the lmm septa there is an average 18% increase in low both TCC and NEC curves. For the 931/08-12 and 831/08-
activity, trues sensitivitywhich correspondsto an average 8% 12 with lmm septa the cross/direct ratio is 1.65 and 1.5
SNR advantage based on NEC rates. The randoms rate respectively. These NEC ratios imply S N R advantages for
rises much more sharply in Figure 2(c) for the lmm septa the cross slice over the direct slice of 34% to 22%.
than in Figure 2(a) for the 3mm tapered septa. This causes Figure 3 demonstrates the effect of reducing the
the NEC curves of Figure 2(d) to turn over at a lower count randoms rate when testing with a real, brain sized object as
rate and activity level than in Figure 2(b) so that by opposed to the 20cm diameter cylinder. Compare Figure 3,
approximatelylpCi/cc the lmm septa collimator has lost its (a) and (b) to Figure 1, (c) and (d). The small SNR
8% S N R advantage and the 3mm tapered septa starts to advantage of about 8% for the 20cm cylinder, based on low
perform at a higher NEC rate. activity NEC rates is lost by OSpCi/cc in favor of the brain
Similarly for the 933/08-12 in Figure l(c) and (d) phantom. At almost twice the activity concentration of the
compared to the 933/08-16 in Figure l(e) and (f) the 20cm cylinder the brain phantom has a peak NEC counting
increased inter-septa gap results in a large increase in the advantage of about 1.53 giving a SNR advantage of 24%.
randoms rate at a given activity level. Based on the NEC
counting rates in Figure l(d) and l(f) the average 12% S N R 933/08-12, 3mm Tapered Septa
17500
advantage of the 933/08-16 at low activities has been lost by 15000
lpCi/cc. In addition, if the rates are expressed per mm of 12500
FWHM axial resolution the 933/08-16 has no image S N R
Counts/ 10~00
advantage over the 933/08-12 at any activity level. This
Second 7500
result is in contrast to the 34% sensitivity advantage of the
933/08-16 based on the standard, low activity sensitivityrates 5000

for total trues reported in Table 2. 2500

The 50% deadtime, and randoms equals trues activity (a) 0

0 05 1 1.5 2 25 3 35 4 45 5
levels provide little information on the form of the NEC
curves and hence on relative image SNR’s. The 50% microCuries/cc
deadtime activity level may be estimated by extrapolating the
12500
low activity sensitivity rates in Table 2 and comparing these
with the trues’ curves in Figures 1 and 2. For all scanners 10000
the randoms equals trues activitylevel is above the saturation
7500
activity of the TCC curves and well above that of the NEC Counts/
curves. Second 5000
The range of peak NEC rates from all scanners for cross
2500
and direct slice averages is 4.9-10.8kcps and 3.3-6.lkcps
respectively, corresponding to S N R advantages of 48% and 0
36%. However, this occurs over a range of activities from 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
l.l-4.0pCi/cc. If activities are restricted to specific values microCuries/cc
the S N R advantages are much smaller. At OSpCi/cc the
range of NEC rates for cross and direct slice averages over Figure 3: The mean true and random coincidence count
all scanners is 3.5-4.7kcps and 1.9-2.7kcps respectively, curves for cross and direct slices, and the derived NEC
corresponding to S N R advantages of 16% and 19%. curves for the 933/08-12 scanning the Alderson brain
Similarly at lpCi/cc the range of NEC rates for cross and phantom.
788

Discussion and the saturation counting activity level. If possible this

Our results demonstrate that relatively large increases should be done for both a brain and a body phantom.
in true coincidence counting rates may represent only
moderate to no image S N R improvement, particularly at Acknowledgments
activity levels where there is a significant randoms fraction.
The present practice of quoting true coincidence counting This work was supported by National Institutes of Health
rates at low activities in a 2Ocm diameter cylinder provides grants NS25563 and NS23473.
a means of comparing relative image SNR's (at least for
20cm cylinders at low activity) because ar=Oand scatter is References
not an important factor for the scanners tested.
Changes in the scatter fraction and the additional noise 1. J.M. Links, H.N. Wagner. Specification of performance of positron
emission tomography scanners (Letter). J Nucl Med, 23:82, 1982
contributed by the scatter deconvolution process have a small
effect on image SNR's. For example, changing as from 0.2 2. J.M. Links, H.N. Wagner. Reply to comments on specifiying the
performance of a positron tomograph by RA.Brooks and VJ.Sank.
to 0.1 causes only a 4% change in the resulting relative J Nucl Med,23.942-943.1982
image SNR's and even less with a significant randoms
component. Furthermore, with as<0.25, ( l + a 3< 1.04 (see 3. El.Hoffman, M.E Phelps, S C . Huang. Performance evaluation of
a positron tomograph designed for brain imaging. J Nucl Med,
equation 7) allowing us to set a@ and not to worry about 24:245-257,1983
obtaining more accurate values for asthan those in Table 1.
4. I. Kanno, S. Miura, S. Yamamoto, H. Iida, M. Murakami, K
However, sensitivity values obtained at low activities Takahashi, K Uemura. Design and evaluation of a positron emission
should only be compared with several caveats in mind. tomograph HEADTOME 111. J Comput Assist Tomogr, 9931-939,
Relative image SNR's are dependent on the square root of 1985
the ratio of the NEC rates, not the direct ratio of the NEC 5. N. Diffrient, A.R Tilley, J.C. Bardagjy. Humanscale 1/2/3, M.I.T.
or TCC rates. Using total scanner sensitivities instead of Press, 1974
individual ditect and cross slice averages may mask 6. S.E Dennzo. Method for optimising side shielding in positron
significant performance differences. A comparison of low emission tomographs and for comparing detector materials. J Nucl
activity, total NEC rates from Table 2 indicates a 36% S N R Med, 21971-977,1980
advantage between the least and most sensitive scanners 7. LR Carroll, G.O. Hendry, J.D. Cumn.Design criteria for multi-slice
tested. However, this becomes an 82% S N R advantage positron emission-computed tomography detector systems. IEEE
when the least sensitive direct slices are compared with the Trans Nucl Sci, NS-273485-488,1980
most sensitive cross slices. 8. LR Lupton, NA. Keller. Performance study of single-slice positron
The maximum useful activity concentration, which is emission tomography scanners by Monte Carlo techniques. IEEE
Trans Med Img, M1-2154-168,1983
object dependent, occurs at the saturation or peak counting
rate. However, we found that the last 50% of the Zero to 9. K Ueda, E Tanaka, K Takami, F. Kawaguchi, K Okajima, K
saturation activity range was used very inefficiently resulting Ishimatsu. Evaluation of slice shield collimators for multi-layer
positron emission tomographs. IEEE Trans Nucl Sci, NS-29563-566,
in only about a 10% improvement in image SNR's. As a 1982
result a practical upper activity bound could probably be
10. J.C. Dainty, R Shaw. Image science:Principles, analysis and
defied well below the activity level for counting rate evaluation of photographic type imaging processes. Academic
saturation. Prrss(London), 1974
As demonstrated by Figures 1and 3 the singles rate and 11. RF. Wagner. Low contrast sensitivity of radiologic, (TINuclear
the resulting randoms rate and deadtime for a PET scanner Medicine and Ultrasound medical imaging systems. IEEETrans Med
are highly dependent on object size. For the 933/08-12 the Img, MI-2105-121, 1983
smaller volume of activity in a brain phantom compared to 12. G.L. Brownell, JA. Comia, RG. Zamenhof. Positron
a 20cm diameter cylinder results in a better image SNR Instrumentation. In: J.H. Lawrence, T.F. Budinger, &Recent
performance above about OSpCi/cc. This cross-over point Advances in Nuclear Medicine, Grune and Stratton (New York), 1,
1979
could probably be moved to higher activities by lengthening
the relatively short septa on the 933/08-12 (Table 1). The 13. T.F. Budinger, S.E. Derenzo, W.L. Greenberg,G.T. Gullberg, R H .
Huesman. Quantitative potentials of dynamic emission computed
standard u)cm cylinder is a poor predictor of countrate tomography. J Nucl Med, 19309-315, 1978
performancefor PET brain imaging, particularly above about
0.5-1.0pCi/cc. This problem of performance prediction for 14. M.E Casey, E.J. Hoffman. Quantitation in positron emission
computed tomography: 7A technique to reduce noise in accidental
brain imaging may be overcome by the adoption of an 18cm coincidence measurements and coincidence efficiency calibration. J
diameter cylinder although the problem of the mismatch with Comput Assist Tomogr, 10845850,1986
the elliptical dimensions of the average adult head remains. 15. E.J. Hoffman, W.M. Digby, G. Germano, J.C. Mazziotta, SC Huang,
Based on OUT results we suggest that when specifying M.E Phelps. Performance of a neuropet system employing 2D
PET scanner sensitivityasiand aribe provided to allow NEC modular detectors. J Nucl Med, 29983-984, 1988
rates to be calculated at at least 0.1, 0.5, 1.0 and 2.OpCilcc 16. Computer Technologyand Imaging, Inc., Knoxville,Tennessee,ECAT
Scanner equipment specifications.