Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
DOI 10.1007/s00259-009-1156-3
ORIGINAL ARTICLE
Advancement in PET quantification using 3D-OP-OSEM
point spread function reconstruction with the HRRT
Andrea Varrone & Nils Sjöholm & Lars Eriksson &
Balazs Gulyás & Christer Halldin & Lars Farde
Received: 30 December 2008 / Accepted: 17 April 2009 / Published online: 13 May 2009
# Springer-Verlag 2009
Abstract
Purpose Image reconstruction including the modelling of
the point spread function (PSF) is an approach improving
the resolution of the PET images. This study assessed the
quantitative improvements provided by the implementation
of the PSF modelling in the reconstruction of the PET data
using the High Resolution Research Tomograph (HRRT).
Methods Measurements were performed on the NEMAIEC/2001 (Image Quality) phantom for image quality and
on an anthropomorphic brain phantom (STEPBRAIN). PSF
reconstruction was also applied to PET measurements in
two cynomolgus monkeys examined with [18F]FE-PE2I
(dopamine transporter) and with [11C]MNPA (D2 receptor),
and in one human subject examined with [11C]raclopride
(D2 receptor).
Results PSF reconstruction increased the recovery coefficient (RC) in the NEMA phantom by 11–40% and the grey
to white matter ratio in the STEPBRAIN phantom by 17%.
PSF reconstruction increased binding potential (BPND) in
the striatum and midbrain by 14 and 18% in the [18F]FEPE2I study, and striatal BPND by 6 and 10% in the [11C]
MNPA and [11C]raclopride studies.
A. Varrone (*) : N. Sjöholm : L. Eriksson : B. Gulyás :
C. Halldin : L. Farde
Karolinska Institutet, Department of Clinical Neuroscience,
Psychiatry Section and Stockholm Brain Institute,
R5:02, Karolinska Hospital,
17176 Stockholm, Sweden
e-mail: andrea.varrone@ki.se
L. Eriksson
Siemens Molecular Imaging,
Knoxville, TN, USA
L. Eriksson
Department of Physics, University of Stockholm,
Stockholm, Sweden
Conclusion PSF reconstruction improved quantification by
increasing the RC and thus reducing the partial volume effect.
This method provides improved conditions for PET quantification in clinical studies with the HRRT system, particularly
when targeting receptor populations in small brain structures.
Keywords Resolution . Modelling . Small brain nuclei .
Partial volume effect
Introduction
The performance of positron emission tomography (PET) has
improved considerably since the technique was introduced in
the 1970s. The implementation of detector crystals with high
stopping power, high light output and fast decay, together with
improved ring design and electronics has led to major
advancements. The High Resolution Research Tomograph
(HRRT) (Siemens Molecular Imaging) is a PET system with
high spatial resolution designed for imaging of the human
brain. The HRRT system has a spatial resolution of approximately 2.5 mm in the centre and 3.5 mm at 14 cm off-centre
full-width at half-maximum (FWHM) [1] and is primarily
used for studies of brain metabolism and radioligand binding
to neuroreceptors [2–5]. PET imaging with the HRRT would
provide better quantification of neuroreceptors in the human
brain, particularly for those monoamine transporter populations that are highly expressed in small brain nuclei, such as
the locus coeruleus or the substantia nigra.
Along with improvement of hardware, software development for more accurate image reconstruction and quantification has been of great importance in the field of molecular
imaging. Image reconstruction including modelling of the
scanners’ point spread function (PSF) in the system matrix is
an approach improving the spatial resolution of the PET
1640
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
images [6, 7]. This approach has been validated and
implemented in new whole-body PET/CT systems, such
as the Siemens Biograph (Hi-Rez) scanner, in which it has
been shown that PSF reconstruction improves the resolution and also provides a more uniform resolution across the
field of view [6].
Partial volume effect (PVE) is a main factor contributing to
reduced accuracy in the quantification of the PET signal [8–11]
in particular in small brain structures. The combination of
high-resolution imaging with the HRRT and the improved
spatial resolution using PSF reconstruction may provide a
means for improving the quantification and reducing the PVE
in neuroreceptor PET studies. Indeed, PSF reconstruction has
recently been implemented and validated in the HRRT system
and applied to human [11C]PE2I PET measurements, showing
approximately 20–25% increase in dopamine transporter
binding potential in the striatum as compared with the
conventional iterative reconstruction method [7].
The reconstruction algorithm that includes modelling of
the PSF [7] has been implemented in the HRRT system at
the Karolinska Institutet PET Centre. In this study we
examined the effects of the improved resolution of the
HRRT images obtained with PSF reconstruction on the
recovery coefficient (RC), the time-activity curves (TACs)
in target and background regions and the quantitative
outcome measure, the binding potential, of dopamine
transporter or dopamine D2 receptor density. The quantitative performance of PSF reconstruction was compared with
the one from the standard reconstruction algorithm. This
comparative analysis was based on phantom measurements
as well as on studies of dopaminergic markers in two nonhuman primates and in one human subject.
FOV and 10 cm off-centre. List mode data of the point
source were reconstructed using the ordinary Poisson 3D
ordered subset expectation maximization (OP-3D-OSEM)
algorithm, with 6 iterations and 16 subsets. Data were
reconstructed with the OP-3D-OSEM and not with analytical reconstruction according to the National Electrical
Manufacturers Association (NEMA) standards.
More recently, image reconstruction has been performed on
a Quad Core PC running a fast reconstruction algorithm
developed by Hong et al. [12]. Based on the measurements
performed by Sureau et al. [7], PSF modelling has recently
been implemented within the fast reconstruction algorithm
(the software was kindly provided by Dr. Merence Sibomana,
Rigshopitalet, Copenhagen University Hospital, Denmark).
The PSF model used in the fast reconstruction algorithm has
been described in detail by Comtat et al. [13] in a multi-centre
evaluation of the PSF reconstruction which also included our
HRRT system. This PSF model differs from the model used
by Sureau et al. [7] with regard to the projector used. In the
regular HRRT reconstruction software for sinogram data the
projector is based on a 3-D implementation of the Joseph’s
algorithm [14] and has a Gaussian shape. In the present
approach an isotropic 3-D kernel was chosen for the OPOSEM reconstruction algorithm with resolution modelling as
described in Eq. 7 of Comtat et al. [13]. The PSF parameters
used in the reconstruction were the default values of
FWHM1 = 2.1 mm, FWHM2 = 5.9 mm, Gauss2/Gauss
1ratio=0.05. The PSF modelled in the reconstruction
was isotropic and stationary. To measure resolution using
PSF modelling, list mode data of the point source were
also reconstructed using the advanced algorithm which
includes resolution modelling of the PSF, with 10
iterations and 16 subsets (OP-3D-OSEM-PSF).
Material and methods
The NEMA IEC/2001 phantom
The HRRT system
This is a phantom primarily used for assessment of the image
quality of whole-body PET systems. The phantom consists
of a large background compartment of approximately 6 l and
6 spheres of different diameter. The spheres and background
can be filled with solutions having different radioactivity
concentration. The diameters of the spheres were: 10, 13, 17,
22, 28 and 37 mm. Different concentrations were used in the
spheres and different isotopes were used in the spheres and in
the background since we aimed to simulate TACs in regions
of different sizes, different radiotracer concentration and
with different kinetic behaviour. In our experimental setting,
the large sphere was filled with a solution of 18F at a
concentration of ~7 kBq/ml. The remaining spheres were
filled with a solution of 18F at a concentration of ~21 kBq/ml,
whereas the background was filled with a solution of 11C at a
concentration of ~5–6 kBq/ml. A solution of 11C was used
for the background to simulate a low activity region having
The HRRT system (Siemens Molecular Imaging) was
installed in 2007 at the PET Centre, Karolinska University
Hospital, Solna, Sweden. The system consists of eight
panel detectors with an octagonal configuration. Each panel
has 9*13 phoswich blocks viewed by 10*14 photomultiplier tubes (PMTs). The total number of crystals is 59,904×
2 = 119,808. Each block has two scintillator layers,
Lu2SiO5:Ce (LSO) and Lu0.6Y1.4SiO5:Ce (LYSO) with an
8*8 crystal configuration per layer. The crystal size is
2.15*2.15*10 mm3. The centre-to-centre distance between
crystals is 2.4 mm. The axial field of view (FOV) is
25.2 cm, corresponding to 207 planes in the reconstructed
images, with a pixel size of 1.218×1.218×1.218 mm3 [1].
The spatial resolution of the system was measured in air
with an 18F point source positioned in the centre of the
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
lower signal to noise ratio than that of the spheres with
different diameter. At the time of imaging the ratio of sphere
to background concentration was measured in a well counter.
The ratio for the 37-mm sphere was 1.4, whereas it was 4.3
for the remaining spheres.
The NEMA phantom was positioned in the HRRT and
list mode data were acquired for 120 min. Images were
reconstructed with 20 frames each having a duration of
6 min. A 6-min transmission scan with a rotating 137Cs
source was performed 12 h after the list mode acquisition to
allow for physical decay of the radioisotopes.
Two reconstruction algorithms were applied: (1) OP-3DOSEM with 6 iterations and 16 subsets and (2) OP-3DOSEM-PSF, with 10 iterations and 16 subsets. The number
of iterations was based on a previous observation that
inclusion of resolution modelling with OP-3D-OSEM-PSF
decreases the convergence rate of the algorithm [13].
However, to confirm the choice of the number of iterations
for each algorithm data were also reconstructed using 10
iterations for the OP-3D-OSEM and 16 iterations for the
OP-3D-OSEM-PSF. Images obtained with algorithm 1 will
be referred to as native, while those obtained with
algorithm 2 as PSF. Images were reconstructed without
decay correction. Native images were also smoothed with a
Gaussian filter of 2 mm FWHM for visual evaluation.
Volumes of interest (VOIs) were drawn on each sphere
according to the inner diameter and then eroded by one
pixel to reduce partial volume sampling. VOIs of the same
size were copied to the background according to the NEMA
standards [15]. The volumes of the VOIs for the series of
spheres were: 0.2, 0.5, 1.5, 2.9, 7.3 and 16.9 cm3.
TACs were obtained for the different spheres and for the
background. The area under the curve (AUC) was
calculated (kBq · ml−1 · min) as an estimate of the
cumulative radioactivity concentration in each region.
For each of the spheres having the ratio to background
equal to 4.3 (10- to 28-mm diameter), the contrast recovery
coefficient CRChot was measured according to the following
equation [15]:
CRChot
Chot =Cbkg
¼
ahot =abkg
1
1
where Cbkg is the concentration in the background, Chot is
the concentration in the spheres, ahot is the actual activity in
spheres and abkg is the actual activity in the background.
The ratio ahot /abkg was measured in the well counter and
decay corrected to the time of imaging. For simplicity,
CRChot will be referred to as recovery coefficient (RC). For
the calculation of Chot/Cbkg TACs for each sphere and
background were decay corrected to the start time of
imaging and then the data of the first 75 min were averaged
1641
to remove possible bias due to low counting statistics in the
background region [16].
To examine the different noise characteristics using either
of the above approaches the standard deviation (SD) of the
average radioactivity concentration in each VOI was taken as
an estimate of the noise and plotted over time along with the
mean value on native images, images smoothed with a
Gaussian filter of 2 mm and on PSF images. To examine the
effect of the number of iterations on the noise of native and
PSF images, the coefficient of variance was calculated with
different iteration number for each VOI.
The STEPBRAIN phantom
This STEreolithographed Phantom of the Brain (STEPBRAIN ©) is anthropomorphic and contains different
fillable compartments for grey matter (GM) and white
matter (WM) [17] (for a detailed description see http://lab.
ibb.cnr.it/). The STEPBRAIN phantom was filled with
[18F]FDG in both compartments to obtain a GM to WM
ratio of 7.9. This ratio was targeted to simulate a PET image
of the benzodiazepine receptor ligand [11C]flumazenil [18].
List mode acquisition of the phantom was performed for
60 min followed by a 6-min transmission scan. Images
were reconstructed as described above for the NEMA
phantom using algorithms 1 and 2.
To calculate the average radioactivity concentration in
the GM and WM, binary masks of the two compartments
were obtained by the segmentation of a computed tomography (CT) scan. To obtain the masks, the phantom was
scanned in the 64 slice CT System Siemens Biograph
(Siemens Molecular Imaging). The voxel size of the CT
images was 0.39×0.39×5.0 mm. First, a CT scan of the
empty phantom (CT-air) was performed to visualize the
thickness of the walls. A new CT scan was performed when
the GM compartment was filled with diluted contrast
medium and the WM compartment with water, with
approximately a ratio of 10 in Hounsfield units between
GM and WM (CT). The CT was manually segmented in
order to generate GM and WM masks, removing the
contribution of the walls measured with the CT-air.
The GM and WM masks were used to measure the average
radioactivity concentration in each compartment. For this
analysis CT-air and CT were coregistered to PET images using
a normalized mutual information algorithm in the software
PMOD 2.85 (PMOD Group, Zurich, Switzerland). The
same coregistration parameters were applied to the GM and
WM masks. Figure 1 displays the CT-air, the CT and the
GM and WM masks that all were coregistered to the PET
images. In addition, VOIs were drawn in the centrum
semiovale of both hemispheres to estimate the radiotracer
concentration in a region where the PVE is considered to be
negligible (Fig. 1).
1642
Fig. 1 Transaxial PET (a)
and CT images (b, c) of the
STEPBRAIN phantom at
the level of the basal ganglia.
Corresponding masks (d, e) for
the grey matter (GM) and the
white matter (WM) and small
VOIs (f) at the level of the
centrum semiovale (CSO)
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
Cynomolgus monkeys
Two female cynomolgus monkeys (housed at the Astrid
Fagraeus Laboratory, SMI, Solna, Sweden) were included.
The study was approved by the Animal Ethics Committee of the
Swedish Animal Welfare Agency and was performed according
to the “Guide for the Care and Use of Laboratory Animals” [19].
PET measurements were performed in the HRRT system.
In the first monkey (6.5 kg) one PET measurement was
conducted after intravenous injection of the dopamine
transporter radioligand [18F]FE-PE2I (67 MBq) [20].
Anaesthesia was induced and maintained by repeated
intramuscular injections of a mixture of ketamine hydrochloride (3.75 mg/kg per h Ketalar®, Pfizer) and xylazine
hydrochloride (1.5 mg/kg per h Rompun® Vet., Bayer). List
mode data were acquired for 180 min and images were
reconstructed with a series of 28 frames of increasing
duration (60 s×5, 180 s×5, 360 s×5, 600 s×13).
In the second monkey (4.9 kg) one PET measurement was
conducted after intravenous injection of the dopamine D2
receptor radioligand [11C]MNPA (197 MBq) [21]. Anaesthesia was induced by intramuscular injection of ketamine
hydrochloride and maintained by a mixture of sevoflurane
(2–6%) and air after endotracheal intubation. List mode data
were acquired for 93 min. Images were reconstructed with a
series of 28 frames of increasing duration (30 s×4, 60 s×4,
180 s×11, 360 s×9). In all PET measurements the head was
immobilized with a fixation device [22]. Body temperature
was maintained by Bair Hugger Model 505 (Arizant
Healthcare Inc., Eden Prairie, MN, USA) and monitored by
an oral thermometer. ECG, heart rate and respiratory rate
were continuously monitored throughout the experiments.
Blood pressure was monitored every 15 min.
Transmission scans were acquired for 6 min using a single
137
Cs source immediately before radioligand injection.
Images were reconstructed using algorithms 1 and 2
described above. In monkey 1, VOIs were drawn for the
caudate, putamen, midbrain and cerebellum on average
images from 11 to 180 min using PMOD 2.85. In monkey
2, VOIs were drawn on the caudate, putamen and cerebellum
for [11C]MNPA. Delineation of VOIs was guided by an atlas
of a cryosected cynomolgus monkey head [22].
Radioligand uptake was expressed as percent standardized uptake value (%SUV) and calculated as follows:
Radioactivity concentration (kBq/cm3) ÷ [injected activity
(MBq) / body weight (kg)] ·100. TACs were obtained for
each region and the AUC was calculated for each TAC as
an estimate of the cumulative radioactivity concentration in
each region. The binding potential (BPND) was estimated
with the simplified reference tissue model (SRTM) [23]
using the cerebellum as reference region. In the [18F]FEPE2I study, BPND was also estimated from native images
reconstructed with 10 iterations and from PSF images
1643
reconstructed with 16 iterations. We selected [18F]FE-PE2I
for this purpose since this radioligand provides a quantifiable signal from brain regions with different transporter
density, such as caudate, putamen and midbrain.
Human subject
One male control subject, 25 years old, underwent one PET
measurement in the HRRT system with [11C]raclopride
(432 MBq), a reference radioligand for the D2 dopamine
receptor [24]. Before PET, a plaster helmet was made and
used with a head fixation system during PET data
acquisition as previously described [25]. The radioligand
was administered as a bolus over 10 s and the IV line was
immediately flushed with saline. A transmission scan was
acquired for 6 min using a single 137Cs source immediately
before radioligand injection. List mode data were acquired
for 63 min. Images were reconstructed with algorithms 1
and 2 with a series of 26 frames of increasing duration
(15 s×4, 30 s×4, 60 s×6, 180 s×6, 360 s×6). PET images
were coregistered with 3-D T1-weighted MRI data. ROIs
were drawn on the caudate, putamen and cerebellum, using
the Human Brain Atlas (HBA) software [26]. BPND was
estimated with the SRTM using the cerebellum as reference
region [23] from native images reconstructed with 6 and 10
iterations and from PSF images reconstructed with 10 and
16 iterations.
Results
The HRRT system
The in plane resolution with OP-3D-OSEM reconstruction
was 2.3 mm FWHM in the centre of the FOV and 3.1 mm
at 10 cm off-centre directions. The corresponding in plane
resolution with OP-3D-OSEM-PSF was 1.5 mm FWHM in
the centre of the FOV and 2.4 mm at 10 cm off-centre
directions.
The NEMA IEC/2001 phantom
PSF reconstruction improved the visualization of the
smallest spheres of the NEMA phantom as compared
with native images (Fig. 2). TACs obtained from the
spheres having a diameter between 10 and 28 mm showed
that with PSF reconstruction the activity in all spheres was
increased and the activity in the smallest spheres came
closer to the spheres of large diameter (Fig. 2). The
increase in AUC of the spheres with a diameter between
10 and 28 mm was inversely related to the diameter itself,
ranging from 24.7% (10-mm diameter) to 4.6% (28-mm
diameter). Even in the case of the largest sphere with 37-mm
1644
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
Fig. 2 a Transaxial PET images
of the NEMA 2001 phantom. b
The line profile drawn at the
level of the two smallest spheres.
c Corresponding plots of the line
profile obtained from the reconstruction without (native) and
with modelling of the point
spread function (PSF). d TACs
obtained from background and
spheres of different diameter
(10–37 mm) of the NEMA
phantom. Note that the 37-mm
sphere was filled with lower
radioactivity
d
25
37 mm - PSF
37 mm - Native
28 mm - PSF
28 mm - Native
17 mm - PSF
17 mm - Native
10 mm - PSF
10 mm - Native
Background - PSF
Background - Native
Radioactivity [kBq/cm3]
20
15
10
5
0
0
20
40
60
80
100
120
Time [min]
diameter the AUC increased by 4.2% using PSF reconstruction. Only a slight decrease (1.8%) in activity was
observed in the background region with PSF reconstruction
(Fig. 2).
The RC for the smallest sphere (10 mm) was 0.81 as
compared to 0.58 for native images, while for the spheres
having a diameter between 13 and 28 mm it was between
Table 1 Comparison of RC measured in spheres having a diameter
from 10 to 28 mm using the two reconstruction algorithms, each with
two different numbers of iterations
RC
10mm
13mm
17mm
22mm
28mm
Native, 6 iterations
Native, 10 iterations
PSF, 10 iterations
PSF, 16 iterations
0.58
0.56
0.81
0.83
0.72
0.71
0.91
0.91
0.76
0.75
0.89
0.88
0.87
0.86
0.97
0.96
0.92
0.92
1.01
1.00
0.91 and 1.01 as compared to between 0.72 and 0.92 for
native images (Table 1). The RC changed by 2% or less
when the number of iterations was increased from 6 to 10
for the native images and from 10 to 16 for the PSF images
(Table 1). The SD measured in each of the VOIs using PSF
reconstruction was lower than in native images and came
closer to the one measured in images smoothed with a
Gaussian filter of 2 mm (Fig. 3a–c). The COV% for the
VOIs of the background and of the spheres was higher in
native images than in PSF images and increased with
increasing number of iterations (Fig. 3d, e).
The STEPBRAIN phantom
In the STEPBRAIN phantom, with PSF reconstruction the
average GM radioactivity concentration increased by 4.4%,
whereas the average WM radioactivity concentration
decreased by 10.5% as compared with native images. This
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
1645
d
a
Native
Background
700
37 mm
28 mm
10 mm
Background
35
30
25
20
15
Native 10 iterations
PSF 16 iterations
Native 6 iterations
PSF 10 iterations
600
500
COV%
Radioactivity [kBq/cm3]
40
400
300
200
10
100
5
0
0
0
20
40
60
80
100
0
120
20
40
Time [min]
b
e
2 mm smoothing
37 mm Native 10 ite
37 mm Native 6 ite
10 mm Native 10 ite
10 mm Native 6 ite
35
37 mm
28 mm
10 mm
Background
30
25
20
15
10
200
COV%
Radioactivity [kBq/cm3]
80
100
120
100
120
Spheres
250
40
37 mm PSF 16 ite
37 mm PSF 10 ite
10 mm PSF 16 ite
10 mm PSF 10 ite
150
100
50
5
0
0
20
40
60
80
100
120
0
0
20
Time [min]
c
40
60
80
Time [min]
PSF
40
Radioactivity [kBq/cm3]
60
Time [min]
37 mm
28 mm
10 mm
Background
35
30
25
20
15
10
5
0
0
20
40
60
80
100
120
Time [min]
Fig. 3 TACs obtained from three representative spheres and from the
background of the NEMA phantom showing the average radioactivity
concentration plus one SD of each VOI as an indicator of the relative
noise in the native image (a), the image smoothed with a Gaussian
filter of 2 mm (b) and in the PSF image (c). Note that the 37-mm
sphere was filled with lower radioactivity. The COV% in the VOIs of
the background (d) and of two representative spheres (e) was plotted
over time and as a function of the number of iterations
difference corresponds to a 17% increase of the GM to WM
ratio with PSF reconstruction as compared with native
images (percent of true ratio: 71 vs 61%, respectively).
With PSF reconstruction the radioactivity concentration in
the centrum semiovale decreased by 3.2%. When the
centrum semiovale was used as WM region, 97% of the
true GM to WM ratio was achieved with PSF reconstruction as compared with 90% of native images.
Cynomolgus monkeys
Visual inspection of [18F]FE-PE2I images reconstructed
with PSF showed better separation of caudate from putamen and better visualization of midbrain/substantia nigra as
compared with native images (Fig. 4a). In PSF images the
AUC of the TACs from the striatum and the midbrain were
16 and 14% higher than those measured in native images,
1646
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
Fig. 4 a Transaxial (first and
third columns) and coronal
(second and fourth columns)
HRRT PET images of [18F]FEPE2I binding in the same
cynomolgus monkey obtained
without and with PSF reconstruction. b TACs of [18F]FEPE2I binding in different brain
regions obtained from native
and PSF images
1200
Putamen PSF
Caudate PSF
Putamen Native
1000
Caudate Native
Midbrain PSF
Midbrain Native
Cerebellum PSF
800
%SUV
Cerebellum Native
600
400
200
0
0
30
60
90
120
150
180
Time [min]
while the AUC of the cerebellum TAC was only 3% higher
(Fig. 4b). PSF reconstruction increased BPND in the caudate
and the putamen by 14% and in the midbrain by 18%
(Table 2). In the native images the difference in BPND
between 6 and 10 iterations was less than 2%, whereas in
the PSF images the difference in BPND between 10 and 16
iterations was less than 5% (Table 2).
Image quality of the striatum was also improved by
visual inspection of the PSF images of [11C]MNPA (Fig. 5).
The AUC of the TACs of [11C]MNPA obtained from PSF
reconstructed images were 9 and 8% higher than the native
images in the caudate and the putamen and 6% higher in
the cerebellum (Fig. 5). In the [11C]MNPA study, using PSF
reconstruction BPND increased by 6% in the caudate (from
1.54 to 1.63) and the putamen (from 1.47 and 1.55).
Human subject
Improved visualization of human striatum was also shown
in [11C]raclopride images obtained with PSF reconstruction
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
1647
Table 2 Comparison of regional binding potential values for [18F]FEPE2I in a cynomolgus monkey using the two reconstruction
algorithms, each with two different numbers of iterations
Using PSF reconstruction, BPND increased by 12% in the
caudate and by 8% in the putamen (Table 3). In the native
images the difference in BPND between 6 and 10 iterations
was less than 2%, whereas in the PSF images the difference
in BPND between 10 and 16 iterations was less than 3%
(Table 3).
Binding potential
Native, 6 iterations
Native, 10 iterations
PSF, 10 iterations
PSF, 16 iterations
Caudate
Putamen
Midbrain
5.52
5.50
6.31
6.32
6.03
6.02
6.87
6.90
1.01
1.03
1.19
1.25
Discussion
This study was intended to further examine the implementation of PSF reconstruction in studies performed with the
HRRT system. We applied the same approach of image
reconstruction as previously reported by Sureau et al. [7]
and used different experimental conditions to evaluate the
effect of PSF reconstruction on quantitative parameters
and the AUC of the TACs from the caudate and the
putamen were 9 and 6% higher than those obtained from
the native images (Fig. 6). No difference was found in the
cerebellum TAC when comparing native and PSF images.
Fig. 5 PET images of [11C]
MNPA binding obtained in the
same monkey without (a) and
with (b) PSF reconstruction. c
Corresponding regional TACs
c
600
Caudate PSF
Putamen PSF
Caudate Native
Putamen Native
Cerebellum PSF
Cerebellum Native
500
%SUV
400
300
200
100
0
0
10
20
30
40
50
Time [min]
60
70
80
90
1648
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
Fig. 6 PET images of [11C]
raclopride binding in a human
subject obtained without (a) and
with (b) PSF reconstruction. c
Corresponding regional TACs
c
600
Putamen PSF
Putamen Native
Caudate PSF
Caudate Native
Cerebellum PSF
Cerebellum Native
500
% SUV
400
300
200
100
0
0
10
20
30
40
50
60
Time [min]
using different radioligands for the dopamine system. The
study demonstrates that reconstruction using modelling of
the PSF provides sizeable improvements of performance.
The basis of such improvement relies on the improved
Table 3 Comparison of regional binding potential values for [11C]
raclopride in the human subject using the two reconstruction
algorithms, each with two different numbers of iterations
Binding potential
Native, 6 iterations
Native, 10 iterations
PSF, 10 iterations
PSF, 16 iterations
Caudate
Putamen
3.08
3.14
3.45
3.54
4.16
4.23
4.48
4.57
image resolution obtained by compensation for the system’s
response function through modelling of the PSF in the
reconstruction algorithm. The improved resolution translates into improved recovery and reduced PVE.
Improved recovery (RC) of the radioactivity concentration was demonstrated in the NEMA IEC/2001 phantom, in
which PSF reconstruction yielded an RC of 0.81 for the
smallest sphere of 10-mm diameter as opposed to an RC of
0.59 for native images. It has to be noted that in this study
the purpose of the NEMA phantom was to simulate
structures of different size and different kinetics between
target and background. Therefore, it might be possible that
the RC was slightly overestimated due to different physical
properties (energy and positron range) of 18F and 11C. This
difference would however not influence the results of the
comparison between native and PSF images.
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
The visual examination of the line profiles across the
different spheres and the TACs for the different regions
examined showed that PSF reconstruction had larger impact
on the radioactivity in the spheres than the background.
With regard to the more uniform background radioactivity,
this observation was in agreement with a previous report
[7], also showing that PSF reconstruction decreases the
variance among neighbouring voxels. In addition, the noise
level of the VOIs for different spheres was lower in the PSF
images than in the native images and came closer to the
images smoothed with a Gaussian filter of 2 mm. Therefore, it appears that PSF reconstruction leads to an
improvement in image resolution reducing also the noise
in the image, as compared with the standard reconstruction
using the OP-3D-OSEM.
The results in the STEPBRAIN phantom showed that PSF
reconstruction reduces PVE and improves quantification of
HRRT images. The line profiles across different sections of
the phantom showed that the “spill-in” from the GM to WM
was reduced in PSF images and the GM to WM ratio was
increased. When the GM to WM ratio was calculated by
using a small VOI at the level of the centrum semiovale,
where PVE is negligible, the ratio was close to 100% of the
true ratio, further confirming that PSF reconstruction
improves the accuracy of the quantification.
The results in the cynomolgus monkeys and in the
human subject showed that PSF reconstruction increases
the BPND values mainly by an effect on the measured
radioactivity concentration in the target region, whereas the
effect was smaller in the reference region. This differential
effect appears to be related to the behaviour of the
radioligand used. In the case of [18F]FE-PE2I, the increase
of BPND was 14% in the striatum and 18% in the midbrain,
while for [11C]MNPA the increase of BPND was approximately 6%. In the cynomolgus monkey [18F]FE-PE2I has a
higher target to background ratio than [11C]MNPA. A
possible explanation is that the spill-over effect from ROIs
in high contrast images may be more pronounced than in
low contrast images resulting in lower recovery. When
reference tissue models are applied the improvement in
quantification using PSF reconstruction might thus be
higher for those radioligands with high specific to nondisplaceable binding ratio (i.e. [11C]PE2I).
With regard to the human PET measurement with [11C]
raclopride, the increase in BPND using PSF reconstruction
was approximately 10%. This increase was lower than the
improvement previously reported for [11C]PE2I [7], which
has a higher target to background ratio than [11C]raclopride,
suggesting again that the improvement in quantification
using PSF reconstruction is directly related to better
contrast recovery in the images.
The results from the NEMA IEC/2001 phantom as well
as the [18F]FE-PE2I and the [11C]raclopride studies suggest
1649
that only a marginal improvement in the outcome measures
is achieved by increasing the number of iterations. In
addition, a sizeable increase in the noise level of the images
was evident in native and PSF images reconstructed with
10 or 16 iterations, respectively. Therefore, we considered
that for PSF reconstruction 10 iterations should be
sufficient to reach convergence in different regions at
various experimental conditions.
Several methods based on segmentation of MR images
are currently available for PVE correction [8, 10, 11] and
implemented in applied studies with different PET radioligands [27, 28]. However, PVE correction methods based
on MR segmentation assume homogeneous uptake in each
of the segmented regions, which might not be appropriate
for each biological condition. In addition, brain structures
such as the midbrain or the brainstem, containing small
regions or nuclei such as the substantia nigra or the locus
coeruleus, cannot be segmented by standard MRI sequences
and acceptable post hoc methods for PVE correction are not
available. Therefore, PSF reconstruction appears to reduce
the need for sizeable PVE corrections in such regions. Along
with PVE, head motion is another important factor limiting
image resolution, particularly for the HRRT system. In this
study, the head fixation systems used for non-human
primates and the human subject were designed to minimize
head motion in the PET system. Methods of correction for
head motion are currently under investigation and might
become more widely available in the future. The contribution of head motion to the spatial resolution of the images
and the accuracy of quantification requires further evaluation, which was beyond the scope of this study.
Conclusion
This study provides further support for the suggestion that
PSF reconstruction of PET images improves neuroreceptor
quantification by improving the image resolution and
reducing PVE. In combination with the high resolution of
the HRRT system, this approach provides improved conditions for applied studies of psychiatric and neurodegenerative disorders, in particular when receptor populations in
small brain structures are examined.
Acknowledgments The authors would like to thank members of the
Karolinska Institutet PET Centre for assistance in the PET experiments. The authors also thank Drs. Inki Hong and Merence Sibomana
for implementation of the PSF modelling in the fast reconstruction
software, and Dr. Bruno Alfano for providing the STEPBRAIN
phantom.
This study was presented in abstract form at the Neuroreceptor
Mapping 2008 Meeting, Pittsburgh, PA, USA, and at the European
Association of Nuclear Medicine 2008 Congress, Munich, Germany.
This study was funded in part by the EC - FP6-project DiMI, LSHBCT-2005-512146 and by VR Swedish Science Council 48105.
1650
References
1. de Jong HW, van Velden FH, Kloet RW, Buijs FL, Boellaard R,
Lammertsma AA. Performance evaluation of the ECAT HRRT: an
LSO-LYSO double layer high resolution, high sensitivity scanner.
Phys Med Biol 2007;52:1505–26. doi:10.1088/0031-9155/52/5/019.
2. Heiss WD, Habedank B, Klein JC, Herholz K, Wienhard K,
Lenox M, et al. Metabolic rates in small brain nuclei determined
by high-resolution PET. J Nucl Med 2004;45:1811–5.
3. Horti AG, Fan H, Kuwabara H, Hilton J, Ravert HT, Holt DP, et
al. 11C-JHU75528: a radiotracer for PET imaging of CB1
cannabinoid receptors. J Nucl Med 2006;47:1689–96.
4. Leroy C, Comtat C, Trébossen R, Syrota A, Martinot JL, Ribeiro
MJ. Assessment of 11C-PE2I binding to the neuronal dopamine
transporter in humans with the high-spatial-resolution PET
scanner HRRT. J Nucl Med 2007;48:538–46. doi:10.2967/
jnumed.106.037283.
5. Hirvonen J, Johansson J, Teräs M, Oikonen V, Lumme V, Virsu P, et
al. Measurement of striatal and extrastriatal dopamine transporter
binding with high-resolution PET and [(11) C]PE2I: quantitative
modeling and test-retest reproducibility. J Cereb Blood Flow Metab
2008;28:1059–69. doi:10.1038/sj.jcbfm.9600607.
6. Panin VY, Kehren F, Michel C, Casey M. Fully 3-D PET
reconstruction with system matrix derived from point source
measurements. IEEE Trans Med Imaging 2006;25:907–21.
doi:10.1109/TMI.2006.876171.
7. Sureau FC, Reader AJ, Comtat C, Leroy C, Ribeiro MJ, Buvat I,
et al. Impact of image-space resolution modeling for studies with
the high-resolution research tomograph. J Nucl Med
2008;49:1000–8. doi:10.2967/jnumed.107.045351.
8. Quarantelli M, Berkouk K, Prinster A, Landeau B, Svarer C,
Balkay L, et al. Integrated software for the analysis of brain PET/
SPECT studies with partial-volume-effect correction. J Nucl Med
2004;45:192–201.
9. Rousset OG, Ma Y, Evans AC. Correction for partial volume effects
in PET: principle and validation. J Nucl Med 1998;39:904–11.
10. Rousset OG, Deep P, Kuwabara H, Evans AC, Gjedde AH,
Cumming P. Effect of partial volume correction on estimates of
the influx and cerebral metabolism of 6-[(18) F]fluoro-L-dopa
studied with PET in normal control and Parkinson’s disease
subjects. Synapse 2000;37:81–9. doi:10.1002/1098-2396(200008)
37:2<81::AID-SYN1>3.0.CO;2-#.
11. Müller-Gärtner HW, Links JM, Prince JL, Bryan RN, McVeigh E,
Leal JP, et al. Measurement of radiotracer concentration in brain
gray matter using positron emission tomography: MRI-based
correction for partial volume effects. J Cereb Blood Flow Metab
1992;12:571–83.
12. Hong IK, Chung ST, Kim HK, Kim YB, Son YD, Cho ZH. Ultra
fast symmetry and SIMD-based projection-backprojection (SSP)
algorithm for 3-D PET image reconstruction. IEEE Trans Med
Imaging 2007;26:789–803. doi:10.1109/TMI.2007.892644.
13. Comtat C, Sureau FC, Sibomana M, Hong IK, Sjöholm N,
Trébossen R. Image based resolution modeling for the HRRT
OSEM reconstructions software. Paper presented at: IEEE Nuclear
Science Symposium Conference Record, 2008; Dresden, Germany.
Eur J Nucl Med Mol Imaging (2009) 36:1639–1650
14. Joseph PM. An improved algorithm for reprojecting rays through
pixel images. IEEE Trans Med Imaging 1982;1:192–6.
doi:10.1109/TMI.1982.4307572.
15. Daube-Witherspoon ME, Karp JS, Casey ME, DiFilippo FP, Hines
H, Muehllehner G, et al. PET performance measurements using the
NEMA NU 2–2001 standard. J Nucl Med 2002;43:1398–409.
16. Reilhac A, Tomeï S, Buvat I, Michel C, Keheren F, Costes N.
Simulation-based evaluation of OSEM iterative reconstruction
methods in dynamic brain PET studies. Neuroimage 2008;39:359–
68. doi:10.1016/j.neuroimage.2007.07.038.
17. Alfano B, Prinster A, Quarantelli M, Brunetti A, Salvatore M.
STEPBRAIN: A stereolitographed phantom of the brain for
nuclear medicine, computed tomography, and magnetic resonance
applications. Paper presented at: RSNA, 2003.
18. Odano I, Halldin C, Karlsson P, Varrone A, Airaksinen AJ,
Krasikova RN, et al. [(18)F]Flumazenil binding to central
benzodiazepine receptor studies by PET—quantitative analysis
and comparisons with [(11)C]flumazenil. Neuroimage 2009;
45:891–902.
19. Clark JD, Gebhart GF, Gonder JC, Keeling ME, Kohn DF. Special
Report: The 1996 Guide for the Care and Use of Laboratory
Animals. ILAR J 1997;38:41–8.
20. Varrone A, Steiger C, Schou M, Takano A, Finnema SJ,
Guilloteau D, et al. In vitro autoradiography and in vivo
evaluation in cynomolgus monkey of [18F]FE-PE2I, a new
dopamine transporter PET radioligand. Synapse. 2009; In press.
21. Finnema SJ, Seneca N, Farde L, Shchukin E, Sóvágó J, Gulyás B,
et al. A preliminary PET evaluation of the new dopamine D2
receptor agonist [11C]MNPA in cynomolgus monkey. Nucl Med
Biol 2005;32:353–60. doi:10.1016/j.nucmedbio.2005.01.007.
22. Karlsson P, Farde L, Halldin C, Swahn CG, Sedvall G, Foged C,
et al. PET examination of [11C]NNC 687 and [11C]NNC 756 as
new radioligands for the D1-dopamine receptor. Psychopharmacology (Berl) 1993;113:149–56. doi:10.1007/BF02245691.
23. Lammertsma AA, Hume SP. Simplified reference tissue model for
PET receptor studies. Neuroimage 1996;4:153–8. doi:10.1006/
nimg.1996.0066.
24. Farde L, Hall H, Ehrin E, Sedvall G. Quantitative analysis of D2
dopamine receptor binding in the living human brain by PET.
Science 1986;231:258–61. doi:10.1126/science.2867601.
25. Bergström M, Boëthius J, Eriksson L, Greitz T, Ribbe T, Widén L.
Head fixation device for reproducible position alignment in
transmission CT and positron emission tomography. J Comput Assist
Tomogr 1981;5:136–41. doi:10.1097/00004728-198102000-00027.
26. Roland PE, Graufelds CJ, Wahlin J, Ingelman L, Andersson M,
Ledberg A, et al. Human brain atlas: for high-resolution functional
and anatomical mapping. Hum Brain Mapp 1994;1:173–84.
doi:10.1002/hbm.460010303.
27. Giovacchini G, Toczek MT, Bonwetsch R, Bagic A, Lang L,
Fraser C, et al. 5-HT 1A receptors are reduced in temporal lobe
epilepsy after partial-volume correction. J Nucl Med 2005;
46:1128–35.
28. Hasselbalch SG, Madsen K, Svarer C, Pinborg LH, Holm S, Paulson
OB, et al. Reduced 5-HT2A receptor binding in patients with mild
cognitive impairment. Neurobiol Aging 2008;29:1830–8.
doi:10.1016/j.neurobiolaging.2007.04.011.