4, AUGUST 1999
’i
1.2
C. Statement of the problem
1 .o
The fan-sum approximation is not valid when there are
0.8
large fluctuations in crystal efficiencies (Figure 2). This
0.6 situation occurs more frequently if narrow energy windows
0.4
are used.
;g
Y 0.2
If, for example, a crystal or a detector block belonging to a
crystal fan has an efficiency much higher (lower) than the
0.0
average, then the summation in (4) is also higher (lower) than
-0.2 . , , .I . . . , I . , . , i average and the efficiency given by (2) is overestimated
0 50 ioa 150
X ProfilesOFTOSS c e n t e r of cylinder (underestimated). These errors in the estimated efficiencies
directly propagate to all the sinograms corrected using these
Figure 1 - a) The effect of different offsets of a uniform cylinder efficiencies, and may also influence the performance of
from the center of the FOV, on the efficiencies calculated using the
scatter correction, generating artifacts and quantitation errors
fan-sum method, in an ECAT EXACT HR+. The correct efficiencies
are all equal to 1. In the offset correction method by Chatziiouannou, in the reconstructed images.
the obtained efficiencies are divided by the function corresponding to Methods have been proposed to avoid the use of
the offset of the phantom. b) Profiles across the center of a uniform approximation (4) [4-6,9-10]. However, only the method by
cylinder, reconstructed from data corrected with the fan-sum method Chesler et a1 can be directly applied with an uniform cylinder,
and no correction for offset. The artifacts increase with the offset the others requiring sinograms with equally illuminated
from the center. LORs. Moreover, the accuracy of this method is limited when
there is a strong asymmetry of efficiency in the ring.
11. THEORY
Let us consider a single detector ring containing N crystals
with efficiencies ci (i=O,l, ...,N-1). Each block contributes Coincidenceswith
The scale factor f is equal to the ratio between the sum of C. Correction for cylinder offset from the center
counts of the average fan of group g and the sum of counts We propose a correction for the phantom offset different
using only the LORs corresponding to opposed “good” from the methods described above [4,8], due to the
crystals. This factor is used to compensate for the eventual particularities of the discrimination algorithm. This correction
missing LORs due to opposed “bad” crystals. is based on the synthesis of a “perfect” sinogram, i.e., a
If crystal i is “bad”, the same expression can be applied. If sinogram corresponding to the actual phantom position, but
all crystals in the fan of crystal i are “good”, this expression is with constant crystal efficiencies and perfect statistics. The
equivalent to the fan-sum. attenuation and scatter component are also included, in order
to mimic as much as possible an ideal measurement of the Ai,
B. Method 2: Iterative Method values. The phantom position is obtained by detecting the
limits of the phantom in the measured sinogram, through
In this method, we consider that all crystals view the same
smoothing and thresholding, to determine the positions of the
total activity A, since the source is axially symmetric and we
center of the phantom for each projection. These positions are
assume that it is well centered in the FOV:
then fitted to a sinusoid.
A/,,;= CA,=A, i=O,l,N-l
je/m i
To correct the efficiencies calculated with the
discrimination method, we apply this method to the perfect
Using expression (1) to substitute A , , we can write: sinogram, forcing the algorithm to use the same discriminated
crystals. This way, we obtain separately the systematic errors
1065
111.METHODS
We compared the two new methods of efficiency
calculation with the fan-sum and the efficiencies calculated
with expression (5) ("HR+ method"), using simulations and
data acquired experimentally on this tomograph. We used a 0 200 400 600
discrimination threshold of 3 in the discrimination method and Crystal
250 iterations in the iterative method. Figure 4 - (a) Efficiencies used in the simulations with noise and
(b) efficiencies calculated from a sinogram with real data, acquired
A. Tomograph
We used the ECAT EXACT HR+ [ l l ] , i h i c h has
i using a narrow energy window.
l00.000€'" ' I ' " ' I " ' ' 1 ' ~ ' ' I ' " '3 "t j
i 6
5
iE io.000
0.0011 I . . . , . . , , I , , .. ' . . . . I . . , . I
0 50 100 150 200 250
Iteration number
D sc r mtnat cr
per crystal fan. The number of counts usually used in the
ECAT EXACT HR+ for crystal efficiency calculation is
approximately 500000 per sinogram.
The discrimination and iterative methods were applied to
all sinograms. The simulated sinogram was used as the
0 200 400 600
perfect sinogram in the off-center correction. The fan-sum Crystal
and HR+ methods were applied only to the sinograms of the
Figure 6 - a) Efficiencies calculated with the fan-sum and HR+
centered phantom.
methods, compared with the simulated efficiencies. No random
variations were simulated in this case, to simplify the interpretation.
E. Data acquisition An offset was added to each curve to facilitate the visualization. The
The measurements were performed without septa in the curves obtained with the discrimination and iterative methods were
ECAT EXACT HR+, using a 20 cm diameter uniform virtually superimposable on the simulated curve. Figures b) and c)
cylinder containing 25 MBq of %e. The standard energy show the corresponding efficiency errors, showp as a percentage of
the correct efficiencies: b) HR+ and fan-sum methods; c)
window (350-650 keV) and several narrow energy windows
discrimination and iterative methods. Note the different scale in (b)
were used. Each LOR was registered separately in the and (c).
sinogram. Only the sinograms containing LORs that join
crystals in the same ring were used for the efficiency
calculation. IV. RESULTS
F. Data analysis
A. Convergence of the iterative algorithm
In the simulations, we computed the absolute error in
Figure 5 shows the evolution of the average and maximum
efficiency for each crystal, as a percentage of the true
efficiency, 1 0 0 ~ ~ e , , , ~ -E,~, ,~I
I E ~ ~and
~ used
, ~ ,their mean and
efficiency errors with iteration number, corresponding to the
results presented in Figures 6, 7 and 8. The 250 iterations took
maximum values to compare the various algorithms and to
50 seconds on a Sun UltraSPARC 1. After 10 iterations the
assess the convergence of the iterative algorithm.
average error was less than 1.3% and after 180 iterations it
To evaluate the magnitude and the distribution of the was less than 0.01 %. The errors decreased exponentially
quantification errors in the sinogram, we calculated the ratio between these iteration numbers. The same general behavior
between the efficiency correction factors obtained by each was observed for the other situations simulated.
algorithm and the ideal factors. In the reconstructed images,
the difference between the images corrected for efficiency and B. Simulation of the effect of a defective block
the real simulated image is presented.
Figure 6 shows the efficiencies and corresponding relative
The measured sinograms of the uniform cylinder were errors calculated by each of the 4 methods tested. The method
corrected for geometrical factors, before using each efficiency used in the HR+ presents errors of larger magnitude, 6.9% on
calculation method. We evaluated the uniformity of the average and reaching a maximum of 27.9% in the simulated
corresponding sinograms corrected for efficiency by situation. The fan-sum method has smaller errors, but still
computing the mean and variance, both of the total counts per reaching an average of 2.3% and a maximum of 10.2%. The
angle of projection and of the number of counts at each radial errors observed with the discrimination (iterative) methods
position.
1067
tended to zero: the mean error was 0.1%(0.0%) and the the sinograms are shown in Figure 7, as a percentage of the
maximum was 0.2%(0.0%). correct factors, for each method. The differences between the
The shape of the errors in the HR+ and fan-sum methods corresponding reconstructed images and the image
follows the shape of the activity seen by the crystals, with reconstructed with the ideal factors are shown in Figure 8.
maximum amplitude for the crystals directly opposing the
defective block. The HR+ method also presents errors in the C. Simulation of experimental situations
zone of the defective block. The results are shown on Table 1, for different statistic
The errors in the efficiency correction factors applied to levels. The HR+ method has the highest errors, which increase
only slightly with the decrease of the number of counts in the
HR+ Method sinogram. The iterative method has an error of 0.0 with no
156 33 30
noise, but the errors increase faster with the decrease of the
216 21 69 number of counts. These errors are slightly smaller than the
-;144 IOOY
standard deviations of the counts in the fan, as given by the
Poisson distribution: these are 1.7%, 2.4%, 3.8% and 5.4%,
1 52
72
for the sinograms with 1000000,500000,200000 and 100000
0 L ] - 1 3 I2 counts, respectively. The errors were larger than these values
0 72 li4 213 246
Rod 01 tn for all the other methods.
In the simulations made with different phantom offsets, the
average and maximum errors of the discrimination and
iterative methods remained the same, for all noise levels.
V. DISCUSSION
The iterative method is the most accurate of the methods
compared, tending to the exact solution in the simulations
made without noise. It is also fast and simple to implement.
The discrimination method is more robust than the fan-
C 32 61 56 126 0 32 61 96 128
sum, yielding better results when the efficiency is very
variable and tending to the fan-sum as the efficiency becomes
Figure 8 - Difference between the image obtained from the
reconstruction of the sinograms corrected with each method, and the more homogeneous and less crystals are discriminated.
true simulated image. The true image is a uniform cylinder with a value Although not as exact as the iterative method, it has the
of 1 in the interior and 0 in the exterior. Note the different limits of the advantage of being less sensitive to errors in the determination
gray scale used. of the displacement of the phantom from the center.
1068
Table 1 Average and maximum absolute errors, in percentage of the true efficiencies, for
~
the various efficiency calculation methods, as a function of the number of counts in the
sinogram. The efficiencies simulated are represented in Figure 4.
Average (maximum) absolute error, in percentage of the true efficiencies
Method Total counts in sinogram
Infinite 1000000 500000 200000 100000
HR+ 10.0 (27.8) 10.1 (32.7) 10.1 (30.1) 10.4 (37.3) 10.9 (47.2)
Fan-sum 4.2 (9.2) 4.3 (13.2) 4.4 (14.5) 4.9 (15.8) 5.8 (22.2)
Discrimination 3.6 (9.4) 3.7 (11.7) 3.9 (16.1) 4.5 (16.3) 5.5 (20.5)
Iterative 0.0 (0.0) 1.6 (8.0) 2.2 (8.3) 3.3 (14.7) 4.6 (21.0)
Although the implciiicntations prcscntctl lierc only [41 D.A. Chcslcr, C.W. Stearns, "Calihration of detector
consider the case of ii ring of crystals, thcy can in principle bc sensitivity in positron cameras", IEEE Ti.ans Nucl Sei,
applied to ohliquc sinograins (Ixtwccn any 2 different rings), vol. 37, no. 2, pp, 768-772, 1990
as long 21s the apPr('xiiiiatioii of the synnnctry of the SOLI~CC 151 12, ~ermallseli, T,J, spil1ks, l , , ~ ~ Calliici,
,, *,A,
reniains valid. We intend to cxtcnd these mcthotls to full 311 Iaiinicrtsma, "Calculation of dctcctor clficicncics and
acquisitions, which would rctlucc the acquisition tinic o f the cxtension of the norinalization sinogram in 1'
uniform cylinder data by a factor of 32, in the casc of the Mcd p,iol, 42, pp,l 143-1 154, 1997
E X 4 1 EXACT HR+.
161 1C.D. Badawi, M A . Lodgc, P.K. Marsdcn, "Algorithnis for
calculating detector cflicicncy normalization coellicicnts
VI. CONCLUS~ONS for triic coincidcnccs in 3D 1'1 , Phys Mcd B i d , 43, pp.
The new incthocls dcscribctl arc practical ways o l 189-205, 1998
calculating crystal efficiencies lrom unilorm cylinder data, 17 I M.E. Cascy, H. Gadagkar, 11. Newport, "A coinponent
with increased accuracy iii ii larger range ol' acquisition I~ascdmcttiod for norinalizatioii in voluinc l'I<T'', Proc.
conditions than the 1':ui-surri mcthotl and tlic tncthod cnrrcntly 3rd Intcrnational Meeting on Fully llircc-l)iiiicnsional
iisctl in the I l C h l ' ICXACT HR+. Image Reconstruction in Radiology and Nuclear
Medicine, Aix-lcs-Bains, pp. 67-7 1, 1995
VII. ACKNOWLEIXMENTS 1x1 A.-X. Cliatziioannou, "Mcasnrcnienls and calculations
We thank Mike Cascy and C~harlcsWatson for access to towuds quantitative whole body 1'El' imaging", PhD
the ECAT EXACT HR+ norinalimtion so~ircccode and for Thesis, University of California, L o s Angclos, 1996
hclpliil discussions. 191 M.17,. Cascy, E.J. Hoflinan, "Quaiititation i n Positron
This work was supported by kkindaqiio para a Citncia c ii Emission Coniputcd l'omograpliy: 7. A tcchniquc to
'l'ccnologia (Prograina Praxis XXI, Portugal) and by AKS- rcdnce noise in mklcntal coincidence niciisnrcincnts and
Ccntro (l'ortllgal). coincidence efficiency calibration", Journal of Coniputcd
Let1 l'oinograpliy, vol. 1 0 , no. 5, pIi. 845-850, 1986