Anda di halaman 1dari 8

~

IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 13, NO. 4, DECEMBER 1994 641

Triple Energy Window Scatter


Correction Technique in PET
Lhgxiong Shao, Richard Freifelder, and Joel S. Karp

Abstract-A practical triple energy window technique (TEW) technique [24], [25]. However, if the source distribution is
is proposed, which is based on using the information in two not uniform, this technique may not be accurate [20]. Another
lower energy windows and one single calibration, to estimate the
scatter within the photopeak window. The technique is basically
PET scatter assessment method, the multispectral technique
a conventional dual-window technique plus a modification factor, [21], [22], has also been reported to be useful for PET. In this
which can partially compensate object-distribution dependent work, we propose a triple energy window (TEW) technique
scatters. The modification factor is a function of two lower scatter which is aimed at taking the position and source dependence
windows of both the calibration phantom and the actual object. into account. This practical technique is based on using the
In order to evaluate the technique, a Monte Carlo simulation
program, which simulates the PENN-PET scanner geometry, was information from two energy windows below the photopeak,
used. Different phantom activity distributions and phantom sizes one narrow and one wide, and a prior phantom calibration, to
were tested to simulate brain studies, including uniform and estimate the scatter included in the photopeak window.
nonuniform distributions. The results indicate that the TEW The motivation for this method is as follows. It is known
technique works well for a wide range of activity distributions from our simulations, as well as those of Thompson [25],
and object sizes. The comparisons between the TEW and dual
window techniques show better quantitative accuracy for the that scattered and unscattered events exist within the energy
TEW, especially for different phantom sizes. The technique is window placed around the photopeak. By using a second
also applied to experimental data from a PENN-PET scanner to energy window below the photopeak, an estimate of the scatter
test its practicality. under that photopeak can be obtained. This is because, in a
narrow window just below the photopeak (in the range of
I. INTRODUCTION 350450 keV), most of the scattered events, like those in the
photopeak window itself, have only undergone single Compton
C OMPTON scattering is always a problem in quantitative
PET volumetric imaging studies. Most previous scatter
correction techniques in PET were based on convolution
scattering through small angles. This forms the basis of the
DEW scatter correction method [ 191. The method, however,
methods using measured scatter response functions [ 11-[ 121. has been shown to suffer from source distribution and object
With the use of NaI(T1) detectors and the improvement of size dependencies [20], thus limiting its applicability.
BGO detectors, energy spectrum-based scattering correction If the ratio of events in the lower, scatter window to the
techniques have become more promising. events in the photopeak window is compared for measured and
For the last decade, most research on energy-based correc- simulated data, using a similar source distribution, then indeed,
tion techniques were mainly in SPECT. They include the dual scatter under the measured photopeak can be compensated for
window [131, [141, dual photopeak window [15], [16], three quite well (see Section 11). However, if these distributions are
window [ 171, and multiple window techniques [ 181. In general, dissimilar, the compensation is inaccurate. It is, of course,
these techniques have been shown to provide a good estimate not possible to simulate data for the exact distribution of
of scatter for a source distribution that is approximately activity in a patient study, but it can only be approximated
uniform. Some of these techniques are simple to implement with a phantom. In that case, since the distribution of scattered
[13], [14], [17]. Some of these techniques can partially account events in both the lower and photopeak energy windows is not
for scatter dependencies in the shape, size, and density of the strictly the same, additional information is required to form
object [15], [16], [18]. Since the scatter energy distribution for a scatter correction method which is less sensitive to source
PET is not the same for SPECT 1201, 1241, 1251, most of the distributions and object size dependencies.
above techniques require modifications and reevaluations. A The additional information that we have chosen to use
PET dual energy window technique (DEW) has been recently is a third energy window (385450 keV) that is wholly
reported to show some promising results [19], and a notched contained within the larger “DEW-scattering” window (i.e.,
window has been suggested to improve the PET dual window 350450 keV). We will compare the ratio of the two
“scattering” windows, the one at 385450 keV and the other
Manuscript received May 19, 1993; revised December 20, 1993 and April at 350450 keV, to the ratio used in the DEW technique, the
13, 1994. This work was supported by grants awarded by the NIH, NS-14867,
R29-CA56627-01Al and the DOE, DE-FG02-88ER60642. The Associate 350450 keV scattering window and the photopeak window.
Editor responsiblefor coordinating the review of this paper and recommending Comparing these ratios to each other (and the necessary
its publication was C. J. Thompson. simulated numbers) allows us to calculate a compensation
The authors are with the Department of Radiology, University of Pennsyl-
vania, Philadelphia, PA 19 104 USA; e-mail: ling@confucius.pet.upenn.edu. factor that, at least to the first order, can correct for the above
IEEE Log Number 9406856. dependencies. Since this compensation factor is not exact,
0278-0062/94$04.00 0 1994 IEEE
642 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 13, NO. 4, DECEMBER 1994

Eb Eb only concentrate on the application of the technique to the


volumetric imaging PENN-PET scanner which uses NaI(T1)
lower window, LW detectors with good energy resolution. The investigation is
mainly based on the Monte Carlo simulation and intends
to provide a feasibility study of the TEW technique for its
implementation in clinical scanners.
In Sections I1 and 111, we will study the DEW technique
and then the TEW technique. In Section IV, we will discuss
how to implement and evaluate the techniques. Finally, in our
results section (Section V), we will show the performance of
the techniques.

' I
pM' '
_I
1W -c I 11. DUALENERGYWINDOWTECHNIQUE
Ea
A. Monte Carlo Simulation

(a)
V scatter In order to develop and evaluate the technique, we used a
Monte Carlo simulation code [23], which simulates the PENN-
PET scanner geometry. The program can provide coincidence
energy spectra for both scatter and true (unscattered) events.
Specifically, in the program, the generated photon pairs are
tracked through a phantom object and allowed to scatter
within the phantom. After exiting the phantom, the photons

- true only
are tracked until they enter a detector, where they are allowed
to undergo scattering and photoelectric absorption. The energy
and position resolutions are simulated by applying Gaussian
smearing with appropriate widths to the recorded energy and
positions of the photons. In this study, we assume an energy
resolution of 10% (FWHM) and position resolution of 5 mm
in the transaxial direction and 6 mm in the axial direction to be
consistent with the performance of the large position sensitive
NaI(T1) detectors used in the PENN-PET.
I
Energy (keV)
B . Energy Spectrum
(b)
Since an event in PET results from a coincidence between
Fig. 1. (a) Illustration of a 2-D coincidence spectrum for PET photons and
the definitions of the photopeak window and lower windows used in the two y-rays, where either one or both y-rays may have scat-
studies. (b) A Monte Carlo simulated PET energy spectrum for scatters and tered, both y-ray energies are displayed as a 2-D spectrum
trues from one detector. in Fig. l(a). A photopeak event is one in which both y-
rays are detected in the photopeak energy range, namely p w ,
it is used in an error-feedback fashion to modify the DEW and fall into the 2-D photopeak window p w in Fig. l(a). A
technique. Even so, this compensation is only approximate scattered event in which only one y-ray scatters will likely
since the TEW technique, as well as the DEW technique, result in one detector measuring the energy with range lw
assumes that the scatter in the lower energy windows has while the other detector likely measures the energy with range
close distributions to the scatter in the photopeak windows. pw. These events fall mainly in the 2-D window LW in
Other choices of windows, with different locations and widths, Fig. l(a). It should be noted that the probability of an event
are possible. One example is the three window technique in in the comer region marked G in Fig. I(a), where both y-rays
SPECT [17]. However, due to the different shapes of the PET scatter, is negligible. In general, it contains less than 1% of
and SPECT spectra, we cannot predict if it compensates well the total events for the scatter windows that we used in this
for source dependencies. Using additional lower windows investigation.
below 350 keV, or alternatively the multiple window fitting Fig. l(b) shows the simulated energy spectra of unscattered
technique [18], does not necessarily improve the estimate of (true) and scattered events for one detector. Compton scat-
scatter under the photopeak since multiple and large angle tering in the detector results in true events of energy below
scattering complicate the situation. 511 keV, although very few fall within the narrow window,
The general concept of the TEW technique can be applied in lwl,from 385450 keV since this requires multiple scattering
principle to both NaI(T1) and BGO scanners, assuming that the within the detector before the y-ray escapes. A wider window,
energy resolution of the BGO system is adequate for the above 1202, is also set from 3 5 0 4 5 0 keV for comparisons and to
assumptions to still be true. However, in this investigation, we determine the modification factor in our studies.
SHAO ef al.: TRIPLE ENERGY WINDOW SCATTER CORRECTION TECHNIQUE 643

h
.- , ---- total =trues + scatters
-O- simulated scatters
Y

E
h
/----------., dualwindow(lw1)
I -C
Y
600 I !-
I
dual window (lw2)
P
v
I
I I

z 500 - I
I
I
I
Y
I I
d - I
C
400
I
I \
.-
Y 300 - I I

E
B
-J 200 -
100 -

10 50 90 130 170 210 250 0 1 ’ ~ ~ ~ ~ ” ’ ” ”


Position r (mm) 0 40 80 120 160 200 240
Position r (mm)
Fig. 2. Simulated scatter ratio (total counts in scatter window/scatter within

:\-.\.
photopeak) for several source distributions. The scatter ratio for the calibration (a)
phantom is defined as the calibration factor throughout the paper.

C. Theory 1
-,‘’ --b
J‘

The TEW technique is basically an extension of the DEW 3501 --..-,


--’ \,----
simulated scatters
iotal= trues + scatters
technique, to which a phantom-size and activity-distribution 300 ; dual window (Iwl)
- dual window (lw2)
dependent compensation factor is introduced. Hence, we start
with the theory for the DEW technique. While the DEW and
TEW techniques can be applied to volume imaging (3-D), we
will, for simplicity, derive the formula for 2-D projection data
P ( 0 ,T ) , where 0 is the projection angle and r is the position of
the projection data. We also assume that the scattering medium
has a uniform attenuation coefficient (a reasonable assumption
for brain studies).
For a DEW technique, the true projection data under the
photopeak, Pp-tr(O, T ) , can be expressed by
0 40 80 120 160 200 240
pp-tr(0, r ) = p p - t o t ( 0 , r ) - Pp-sc(0,r )
Position r (mm)
= Pp-tot (0, r ) - Ps-tot(0, r ) / C F u n i f o r m ( o , r )
(b)
(1) Fig. 3. Estimated scatters with the DEW technique for simulated data
(calibration phantom size is 18 x 10 cm): (a) a small cylinder (14 x 10
where Pp-tot(O,T ) and P p - s c (r~),are the total counts and the cm) and (b) a large cylinder phantom (22 x 10 cm).
scatter counts in the photopeak window P W , respectively.
Ps-tot(O, r ) are the total counts in the lower window LW1. phantom (22 x 10 cm). It can be observed that the scatter ratios
C F u n i f o r m is the scatter ratio, which is the scatters in the lower and calibration factors are source-distribution dependent and
window divided by the scatters in the photopeak window. phantom-size dependent. If we only use the DEW technique
Since CFuniform is determined from a uniform calibration with one calibration factor, as indicated by the solid line in
phantom, it is also called calibration factor. That is, Fig. 2, the mismatch of the estimated scatters will be very
appreciable.
CFuniform(0, r ) = Ps-tot(@,~)/Pp_sc(0,
r). (2)
Usually, Pp-,,(O,r ) for a uniform phantom is unknown. It can D.Results
be obtained from Monte Carlo simulation data or experimen- As an example, Fig. 3 shows the performance of the DEW
tally from a phantom that is a reasonable approximation for technique applied to a simulated small uniform cylinder (14 x
the objects being imaged (e.g., heads). In our investigation, we 10 cm) and a simulated large uniform cylinder (22 x 10 cm)
used several line sources placed in a uniform water phantom by using the above calibration phantom. Two different lower
to reconstruct the calibration factor of a uniform distribution. windows were tested for both cases; lwl and lw2. The results
The detailed design will be given in Section IV-B. indicate that the technique is very sensitive to the phantom
The simulated calibration factor (2) is shown in Fig. 2 for sizes and the width of the lower window. For example, given
the calibration phantom which is an 18-cm diameter by 10-cm a scatter window (lwl), we underestimate scatter in one case
length cylinder (all phantom dimensions will be quoted as a (small phantom shown in Fig. 3(a)), while we overestimate
diameter times a length throughout the paper). Also shown the scatter in other case (large phantom shown in Fig. 3(b)).
are the scatter ratios for rod sources (4 x 10 cm) in cold and Hence, we conclude the DEW technique may not be robust
warm cylinders (1 8 x 10 cm) and a large uniform cylinder enough for a change in the object size.
644 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 13, NO. 4, DECEMBER 1994

111. TRIPLEENERGYW m o w TECHNIQUE


4000
A . Triple Window Technique
The triple energy window technique can be expressed as
1000
Pp-tr(0, r ) = Pp-tot(0, r ) - M ( 0 ,r )
n
* Ps-tot(0r r)/CFuniform(er r )
I .

(3) 0 40 EO 120 160 200 240


Position r (mm)
where M ( 0 , r ) is the modification factor which depends on
the object distribution and object size. There is no analytical large cold rod

solution for M ( 0 ,r ) since Compton scattering depends on the


object medium and scanner geometry. Most reported studies
depend on empirical fits to estimate the scatters under the
photopeak [15], [16], [18]. We also used a similar approach lo00

to obtain the modification factor. 0


0 40 EO 120 160 200 240
Position r (mm)
B. Procedure
Two steps are involved in the scatter correction. They are (a)
the following:
1.4-
1) Calibration:
h

a) Collect counts in two lower windows, Psl-tot(O,r) I. 1.2-


and Psz-tot (8,r ) from a uniform calibration phantom. a

b) Find the ratio between these two lower windows 5


s

1-,
I.
0.8-

-
0
Rcaiib(O,r ) = Psz-tot(O, r)/Psi-tot(6',r ) . e
8 0.6 cold rod in a warm background

; -
-t-

r ) as in the DEW technique


c) Calculate CFuniform(O, rod source in warm background
9
-
0.4 ..-C rod source in cold background
, , , ,
smallunifom
, , , ,
CFuniform(6,r ) = Psl-tot(6, r)/Ppsc(@,r ) . 0.2 largeuniform
calibration phantom
2 ) During the scattering correction for each object: 0.0
a) Use the same lower windows and find their ratio for 0 SO 100 150 200 250

the object, Position r (mm)


(b)
Robject(0,r ) = Psz-tot(0, T)/J'si-tot(0, r ) .
Fig. 4. (a) Simulated scatter profiles under the photopeak and in the two
lower windows for both the calibration phantom and the 6-cm diameter cold
b) Calculate M ( 0 , r ) in the following formula and use rod in a 16 x 10-cm warm cylinder. (b) Modification factors used in the
it in (3): TEW technique for the above rod source phantom and several other phantom
distributions and for the case of b = 0.5 in (4).
M ( 0 ,r ) = [Robject(o,r)/Rcalib(e, . ) Ib (4)
where the subscripts 1 , 2 mean lower window Number in (3) (e.g., the Ps-tot(O, r)/CFuniform(O,r ) term) since the
1 and 2, respectively, and b is a relaxation factor. narrow window contains mostly scattering from the medium
In the above procedure, several points should be mentioned
[25] and provides a better scatter estimation compared to the
here: 1) Since the TEW technique is an error-feedback ap- wide window setting (see also the energy spectra in Fig. l(b)).
proach and the modification factor is an empirical function, 3) As we mentioned earlier, the rationale for the modification
the amount of the feedback is determined through studies factor is that by normalizing Robject(O,T) with the ratio
(see Section V). This is why we introduce the relaxation from the uniform calibration phantom, it provides a partial
factor b, which can control the amount of the feedback. It correction or compensation for the source distribution and
must be optimized for a given class of studies. For example, object-size dependencies. Since the TEW technique is based
if b = 0, the feedback is zero and the TEW technique on an empirical formulation, many window arrangements,
becomes the DEW technique. This approach is similar to the including two separate lower windows, can be valid. This
dual photopeak window technique proposed by King [ 151. assumes that the events in the lower windows are mainly single
However, in our approach, we calculate the modification scattering events at relatively small angles. Initially, we use
factor for each object. 2 ) For the calibration phantom in the two overlapping lower windows lwl and Zwz to calculate M
simulation study, as we mentioned in the previous section, for the reason that we can easily control the compensation by
we will use an 18 x 10-cm cylinder. The widths of the two changing their window width ratio. Hence, we will concentrate
lower windows are defined in Fig. l(b) (350450 keV and on the overlapping window setting throughout the paper.
385450 keV). We chose the relaxation factor b = 0.5, for Fig. 4(a) shows the count profiles in the two lower
reasons which will be shown in the results section. We also windows and the scatters under the photopeak for both the
use the narrow lower window LWl in our first step calculation calibration phantom and a 16 x 10-cm cylinder with a large
SHAO ef al.: TRIPLE ENERGY WINDOW SCAlTER CORRECTON TECHNIQUE 645

TABLE I
CHI-SQUARED ‘ ‘ G ~ D N E SOF
S FIT” OF THE PROJECTION DATAFOR DIFFERENT “IQUES AS SHOWN IN FIG.5

I wide
dual window
I 1995.4 I 568.1 I 12821.0 1 1221.4 I 217.0 I
narrow
dual window 139.7 749.3 2816.3 4399.6 173.1

I triple window 1 608.5 1 265.8 1 119.0 I 431.5 I 62.6 I


cold rod (6-cm diameter x 10 cm long) located at the center. in Fig. l(a). Especially, since the lower window range for the
Fig. 4(b) shows the modification factors M(I9,r) with b = DEW technique with the narrow window is 385-450 keV,
0.5 for this phantom and several other phantom distributions these window settings are similar to the Notched Window
based on the calibration curve in Fig. 2. Since these phantoms area suggested in [241, [251.
are circularly symmetric, there is no &dependence for these For all of the above techniques, in order to reduce the
cases. However, for an arbitrary object, M(I9,r) will depend statistical noise, we processed the data using the following
on 0. In the figure, the other window settings are the same procedures:
as defined above. For the different source distributions given 1) We took enough counts for the calibration curve and
in the figure, the modification factors can range from 0.6 smoothed it using a three-point average where each
to 1.4, depending on source positions. point represents 8 mm in the projection data. Then, the
calibration curve can be interpolated back to the actual
IV. IMPLEMENTATION
pixel size, depending on the size of projection data.
A . Practical Implementation 2) The lower window projection data are smoothed by
a Gaussian kemel (FWHM = 12 mm) before it is
Although more studies will be done before the TEW tech-
substitutedinto (3). This step is necessary and reasonable
nique is implemented on our scanner, we propose a simple
since the lower window has much less counts than the
and practical method for separating the scatters and trues
photopeak window and the scatter profile is always much
under the photopeak without using simulated data, in order
smoother than the true projection data.
to obtain the calibration factor CFun;form.Since a uniform
distribution can be considered to consist of a superposition For the evaluation, besides the visual inspection of perfor-
of line sources distributed uniformly throughout the phantom, mance of the techniques, we used a numerical comparison
we can approximate this distribution with three line sources between the true and estimated scatters by employing a chi-
located at the center, 40 mm and 80 mm off the center in squared “goodness-of-fit’’ test, X 2 , [15], which is defined by
a water-filled cylinder. We partitioned the cylinder (20 x x 2= C [ ( S t ( i )- Se(2))2/St(2)] (5)
19 cm) into three annular components. We used the line source
inside each annulus as an approximation for any position in where & ( i ) is the true scatter projection data and S,(i) is
this annulus, and the relative areas of annuli are used to the estimated projection data. For the simulated data, X 2
weight the scatter fraction determined for each line source can be easily calculated. For the actual scanner data, St(2) is
measurement. In this way, for each individual line source, we unknown. Therefore, we cannot use (5) for the experimental
can easily obtain the scatter events under the photopeak and data. In order to study the feasibility of the implementation of
the scatter window separately, except in the area where the the TEW technique in the PENN-PET scanner [26], we apply it
line source actually is. However, we can remove that part of to the experimental data from a test phantom specified in [27].
data and extrapolate the scatters of this region by using scatter Specifically, the phantom is 20 cm in diameter by 19 cm long
information from adjacent positions. with a 5-cm diameter rod located 6 cm off center. Then, we
During data acquisition, the PENN-PET scanners provide calculated the contrast for the cold rod. The contrast is defined
for recording data in list mode format which allows flexibility as ( B- H ) / B , where B is the counts in the warm background
in choosing the energy window. However, for scanners which and H is the counts in the cold region. A more complete evalu-
do not provide list mode data, separate scans can be taken ation of optimal values for the energy windows and relaxation
with different window settings. parameter, in terms of contrast and the signal-to-noiseratio, is
necessary but beyond the scope of this feasibility study.
B. Evaluation
In order to illustrate the performance of the TEW technique, v . RESULTS AND DISCUSSION
we made comparisons between the DEW technique with We applied the three windowing techniques described in the
the wide lower window (LWz), with the narrow window previous section to different source distributions and different
(LW1) and the TEW technique with both LW1 and LW2 phantom sizes. They were a hot rod source (4 x 10 cm) in a
windows. Here, capital LW indicates the 2-D window region cold background (18 x 10 cm), a hot rod source in a warm
646 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 13, NO. 4, DECEMBER 1994

simulated scattRs
totaktrues + s c m
dual window (Iwl)
dual window ( I d ) 1 "'1
120
4
/I
I I
.
-0- simulatedscatters
total =trues +scatters
e dualwindow(Iw1)
- dual window (lw2)
+ triplewindow
Jo-
2
U
L '
0 40 80 120160200240
23-

10 -

0 ~ ' ~ ' ~ ' " " " ' '


0 40 80 120 160 200 240 0 40 80 120 160 200 240
Position r (mm) Position r (mm)

800 - 400 1

3
3
s
700

600-
-
' I
I
\
I
___
I -
-
-
simulatedscatters
total =mes +scatters
dualwindowOw1)
dualwindow(Iw2)

-
I
500 - I I
I
triple window
I I
i I triple window

"1 I I
I
200

150

200 100

100 50

".
0 40 80 120 160 200 240 0 40 80 120 160 ZOO 240
Position r (mm) Position r (mm)
(C) (4

* simulatedscatters
90 ] --- + scatters
total =trues
dual window (lwl)
dual window Ow2)
130.0{, II I
I
I I

lo,! . , . , . , . I . I . I

0 40 BO 120 160 200 240

Position r (mm)
(e)
Fig. 5. Estimated scatters with three different correction techniques from simulation data for: (a) a rod source in a cold background; (b) a rod source in a
warm background (5 : 1 ratio); (c) a small cylinder with uniform activity; (d) a large cylinder with uniform activity; (e) an elliptical phantom in the long axial
direction (90'). For three of the figures, the small plots in the right comer indicate the relative scale between the total counts and the scatters.

background (activity ratio 5 : 1), a small cylinder (14 x 10 cm) squared "goodness-of-fit" for these cases. Both Table I and
with uniform activity, a large cylinder (22 x 10 cm) with Fig. 5(a)-(e) show that the TEW technique provides better
uniform activity and an ellipse phantom (16 x 19 x 10 cm) scatter estimations for these phantoms. It should be noted
with uniform activity. The performance of the techniques that the DEW technique with a narrow lower window is
are shown in Fig. 5(a)-(e). In the figures, the statistics of better than the DEW technique with a wide lower window
the projection data are very typical, which gives about one when the source has high nonunifonnity, such as rod source
million counts per image. In addition, Table I shows the chi- in a cold background (Fig. 5(a)). This is expected since the
SHAO er al.: TRIPLE ENERGY WINDOW SCATTER CORRECTION TECHNIQUE

Cylindricalphantom
/-\

1
1 :
I -o- simulatedscatters
*
c
L

U
4nnn
- large uniform
- small uniform
-0-

-C
rod in warm background
rod in cold background
/I
0
M

F
i i I
1
I
; 2000
w
'?
.-
c
U

n 40 80 120 160 200 240 0.0 0.2 0.4 0.6 0.8 1.0
Position r (mm) Relaxation factor b

Fig. 7. Chi-squared "goodness of fit" for different phantoms versus the


relaxation factor b for the phantoms in Fig. 5
8000 - I
I
'. \
90degree ----- total=uues +scatters
1 1 z-1
:- dual window(1wl)

In
II
I
,
,,

,I 't
I
I

-1
,
triple rnnnn
0"W" ,
$ 1 I ,

s '
I

I '\ I I

\---.
I ,

II
4000-
; I
I
,
I
I# I
I

II
2000

phantom
-innno ' .
o so izn

-
20 40 60 100
Position r ( 2 mm)
simulated scatters
6OOo- ----- Fig. 8. Comparisons between the TEW technique and the DEW technique

i -
-t
total=trues +scatters
dual window (lwl)
triple
for the experimental data obtained from a PENN-PET scanner. The phantom
is a 20 x 19 cm cylinder with a cold rod (5 cm diameter) located 6 cm off the
center. The profiles are obtained from the reconstructed images. Since there
is no obvious difference between the images from the TEW and the DEW,
4000-
only the TEW image is shown (inset).

zoo0 -
The activity ratio between the rod and the ellipse is 1 : 5.
The estimated scatters with different techniques are shown in
Fig. 6. Since the ellipse phantom is asymmetric, the profiles
at 0" and 90" are both shown in the figure. For all three cases,
o 40 80 120 160 200 240
the TEW technique performs better than the DEW technique.
Position (mm) As mentioned above, the relaxation factor, b = 0.5, in
(C) M ( 0 ,r ) was used for the above results. However, the optimal
Fig. 6. Comparison between the TEW and DEW techniques for a cold rod b is different for different phantoms. Fig. 7 shows the chi-
(6 cm diameter) centered at a 16 x 10 cm hot cylinder (a), and a low activity squared "goodness-of-fit'' for different phantoms versus b. It
rod source (12 cm diameter) asymmetrically located in an elliptical phantom
(16 x 19 x 10 cm) at 90° (b) and 0' (c). The activity ratio between the is obvious that the choice of b = 0.5 is a good compromise
rod and the ellipse is 1 :5. between the different cases.
Finally, Fig. 8 shows an experimental comparison between
narrow window contains mainly the scatter information from the TEW technique and the DEW technique. The profiles are
scattering which occurs in the medium [24] and supports the obtained from a cross-section of the reconstructed image of a
assumption that the scattering window should be reasonably test phantom of [27] with the TEW technique. Table I1 shows
close to the photopeak window to provide for as accurate as the corresponding contrast for different regions of interests in
possible scatter compensation to the photopeak events. the cold rod. They are 2, 3, and 5 cm in diameters. The small
Similarly, we applied the techniques to a number of cold cold region of interest gives the best contrast since it has the
rod sources. The first one is the large cold rod phantom as least resolution blurring effect. In general, the TEW technique
shown in Fig. 4(a). The second one is a 12-cm diameter rod gives better contrasts than the DEW, although the results are
asymmetrically located in a 16 x 19 x 10 warm ellipse. very close, as shown in Table I1 and there is no noticeable
648 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 13, NO. 4, DECEMBER 1994

TABLE II [6] M. Ljungberg, P. Msaki, and %-E. Strand, “Comparison of dual window
COMPARISON FOR EXPERIMENTALDATAIN FIG.
OF THE CONTRAST 8 and convolution scattering correction technique using the Monte Carlo
method,” Phys. Med. B i d , vol. 35, no. 8, pp. 1099-1 110, 1990.
[7] M. Ljungberg and %-E. Strand, “Scatter and attenuation correction
cold rod contrast in SPECT using density maps and Monte Carlo simulated scatter
functions,”J. Nucl. Med., vol. 31, no. 9, pp. 1560-1567, 1990.
[8] P. Msaki, B. Axelsson, and S. A. Larrson, “Some physical factors
influencing the accuracy of convolution scatter correction in SPECT,”
Phys. Med. B i d , vol. 34, no. 3, pp. 283-298, 1989.
[9] J. C. Yanch, M. A. Flower, and S. Webb, “A comparison of deconvo-
Triple window 0.90 0.93 0.97 lution and windowed subtraction techniques for scatter compensation in
SPECT,” IEEE Trans. Med. Imag., vol. MI-7, no. 1, pp. 13-20, 1988.
Dual window I 0.89 I 0.91 I 0.93 [lo] C. E. Frey and B. M. W. Tusi, “A practical projector-backprojector
modeling attenuation, detector response, and scatter for attenuation
scatter compensation in SPECT,” in Conf. Rec. I991 IEEE Nucl. Sei.
Sympos. Med. Imag. Conf., Santa Fe, NM,pp. 1777-1781, 1991.
[ 111 -, “Spatial properties of the scatter response function in SPECT,”
difference between the images. It should be emphasized that IEEE Trans. Nucl. Sci., vol. 38, no. 2, pp. 789-794, 1991.
[12] B. R. Zeeberg, A. N. Bice, and H. N. Wagner, Jr., A Theoretically-
this experimental data only serves as a demonstration and more Correct Algorithm to Compensate for a Three-Dimensional Spatially-
work is involved in the actual implementation. Variant Point Spread Function in SPECT Image in Information Pro-
In summary, it has been demonstrated, using simulated data, cessing in Medical Imaging. New York: Plenum Press, 1987, pp.
245-254.
that the calibration factor not only depends on the source distri- [13] R. J. Jaszczak, K. L. Greer, C. E. Floyd, Jr., C. C. Harris, and R.
butions and object sizes but also on the width of lower window, E. Coleman, “Improved SPECT quantification using compensation for
as shown in Fig. 2. By using the TEW technique, we can pro- scattered photons,” J. Nucl. Med., vol. 25, pp. 893-900, 1984.
141 M. C. Gilardi, V. Bettinardi, A. Todd-Pokropek, L. Milanesi, and
vide more accurate scatter estimation than by using DEW tech- F. Fazio, “Assessment and comparison of three scatter correction
niques, since the TEW technique partially takes the source dis- techniques in single photon emission computed tomography,” J . Nucl.
tribution and object size into account. The overall performance Med., vol. 29, pp. 1971-1979, 1988.
151 M. A. King, G. J. Hademenos, and S. J. Glick, “A dual window method
of the TEW technique is better than the DEW technique. In for scatter correction,”J. Nucl. Med., vol. 33, no. 5, pp. 605412, 1992.
fact, the TEW is the same as the DEW technique in the case of 161 G. J. Hademenos, M. A. King, and M. Ljungberg, “The influence of
phantom size, shape, and density, and collimators selection on the dual
b = 0.0. It is also much simpler than the multiple window tech- photopeak window scatter correction method,” in I992 IEEE Nucl. Sci.
nique [18], [21], [22].However, further studies should be con- Sympos. and Med. Imag. Conf., Orlando, FL, pp. 84&846, Oct. 1992.
ducted to optimize the modification factor for the experimental 171 K. Ogama, Y Harata, T Ichihara, A. Kubo, and S . Hashimoto, “A
practical method for position-dependent Compton-scatter correction in
data and evaluate the technique as applied to patient data. In single photon emission CT,” IEEE Trans. Med. Imag., vol. 10, pp.
addition, the optimal combination of two lower scatter win- 408412, 1991.
dows also needs to be studied. It is also possible that the model 181 K. F. Koral, X. Q. Wang, W. L. Rogers, N. H. Clinthome, and X. H.
Wang, “SPECT Compton-scattering correction by analysis of energy
for the modification factor (4) is not the optimal one. Another spectra,” J. Nucl. Med., vol. 29, pp. 195-202, 1988.
disadvantage of the TEW technique is that if the statistics 191 S. Grootoonk, T. J. Spinks, T. Jones, C. Michel, and A. Bol, “Correction
of the lower window are too low, fluctuations in the lower for scatter using a dual energy window technique with a tomograph
operated without septa,” in Conf. Rec. I991 IEEE Nucl. Sei. Sympos.
window could be amplified since usually CFuniformis less and Med. Imag. Conf., Santa Fe, NM, pp. 1569-1573, Nov. 1991.
than one, a shortcoming which the DEW technique also has. [20] R. L. Harrison, D. R. Haynor, and T. K. Lewellen, “Dual energy window
Overall, we conclude that the TEW technique provides a scatter corrections for Positron Emission Tomography,” in Conf. Rec.
1991 IEEE Nucl. Sci. Sympos. and Med. Zmag. Conf., Santa Fe, NM, pp.
practical, fast, and reasonably good scatter estimation in PET 17W1705, Nov. 1991.
for a wide variety of activity distributions and phantoms of [21] R. Lecomte, M. Bentourkia, P. Msaki, J. Cadorette, and M. Heon, “Po-
tentials of multispectral acquisition in Position Emission Tomography,”
different sizes. It only requires a single uniform phantom in I992 IEEE Nucl. Sci. Sympos. and Med. Imag. Conf., Orlando, FL,
calibration and it has the potential to become clinically useful. pp. 856858, Oct. 1992.
[22] P. Msaki, M. Bentourkia, J. Cadorette, M. Heon, and R. Lecomte,
“A normalization technique for multispectral acquisition in positron
REFERENCES emission tomography,” in 1992 IEEE -Nucl. Sci. Sympos. Med. Imag.
Conf.,Orlando, FL, pp. 976-968, Oct. 1992.
[l] M. Bergstrom, L. Eriksson, C. Bohm, G. Blomqvist, and J. Litton, [23] R. Freifelder, et al., “Increasing the axial extent in volume imaging PET
“Correction for scattered radiation in a ring detector positron camera by scanner: Is there a limit?” in Conf. Rec. I992 IEEE Nucl. Sci. Sympos.
integral transformation of the projections,” J. Comput. Assisted Tomog., and Med. Imag. Conf.. Orlando, FL, Oct. 1992.
vol. 7, no. 1, pp. 42-50, 1983. [24] C. J. Thompson and Y. Picard, “Two new strategies to increase the
[2] L. Shao and J. S. Karp, “Cross plane scattering correction-Point source signal noise ratio in positron volume imaging,” IEEE Trans. Nucl. Sci.,
deconvolution in PET,” IEEE Trans. Med. Imag., vol. 10, no. 3, pp. vol. 40, no. 4, pp. 956961, 1993.
234-239, 1991. [25] C. J. Thompson, “The problem of scatter correction in positron volume
131 J. S. Barney, J. G. Rogers, R. Harrop, and H. Hoverath, “Object shape imaging,” IEEE Trans. Med. Imag.,vol. 12, no. 1, pp. 124-132, 1993.
dependent scatter simulation for PET,” IEEE Trans. Nucl. Sci., vol. 38, [26] J. S. Karp, G. Muehllehner, D. A. Mankoff, C. E. Ordonez, J. M.
no. 2, pp. 719-725, 1991. Ollinger, M. E. Daube-Witherspoon, A. T. Haigh, and D. J. Beerbohn,
[4] C. E. Floyd, R. J. Jaszczak, K. L. Greer, and R. E. Coleman, “De- “Continuous-slice PENN-PET A position tomography with volume
convolution of Compton scatter in SPECT,” J . Nucl. Med., no. 26, pp. imaging capacity,” J . Nucl. Med., vol. 31, no. 5, pp. 617-627, 1990.
403408, 1985. [27] J. S. Karp, M. E. Daube-Witherspoon, E. J. Hoffman, T. K. Lewellen,
[5] P. Msaki, B. Axelsson, C. M. Dahl, and S. A. Larrson, “Generalized J. M. Links, W. H. Wong, R. D. Hichwa, M. E. Casey, J. G. Colsher,
scatter correction method in SPECT using point scatter distribution and R. E. Hitchens, “Performance standards in Positron Emission
functions,” J. Nucl. Med., no. 28, pp. 1861-1869, 1987. Tomography,” J . Nucl. Med., vol. 32, no. 12, pp. 2342-2350, 1991.

Anda mungkin juga menyukai