3D-OSEM and FP-CIT SPECT quantification: benefit for
studies with a high radius of rotation?
Walter Koch, Peter Bartenstein and Christian la Fouge
`
re
Objectives Dopamine transporter imaging with single-
photon emission computed tomography (SPECT) is a
valuable tool for both clinical routine and research studies.
Recently, it was found that the image quality could be
improved by introduction of the three-dimensional
ordered subset expectation maximization (3D-OSEM)
reconstruction algorithm, which provides resolution
recovery. The aim of this study was to systematically
evaluate the potential benefits of 3D-OSEM in comparison
with 2D-OSE M under critical imaging conditions, for
example, scans with a high radius of rotation.
Materials and methods Monte Carlo simulation scans of
a digital brain phantom with various disease states and
different radii of rotation ranging from 13 to 30 cm were
reconstructed with both 2D-OSEM and 3D -OSE M
algorithms. Specific striatal binding and putamen-to-
caudate ratios were determined and compared with true
values in the phantom.
Results The percentage recovery of true striatal binding
was similar between both reconstruction algorithms at
the minimum rotational radius; however, at the maximum
rotational radius, it decreased from 53 to 43% for
3D-OSEM and from 52 to 26% for 2D-OSEM. 3D-OSEM
matched the true putamen-to-caudate ratios more closely
than did 2D-OSEM in scans with high SPECT rotation radii.
Conclusion 3D-OSEM offers a promising image quality
gain. It outperforms 2D -OSEM, particularly in studies with
limited resolutions (such as sc ans acquired with a high
radius of rotation) but does not improve the accuracy of the
putamen-to-caudate ratios. Whether the benefits of better
recovery in studies with higher radii of rotation could
potentially increase the diagnostic power of dopamine
transporter SPECT in patients with borderline striatal
radiotracer binding, however, needs to be further
examined. Nucl Med Commun 34:971–977
c
2013 Wolters
Kluwer Health | Lippincott Williams & Wilkins.
Nuclear Medicine Communications 2013, 34:971–977
Keywords: 3D-OSEM iterative reconstruction, dopamine transporter, FP-CIT,
Monte Carlo simulation, radius of rotation
Department of Nuclear Medicine, University of Munich, Munich, Germany
Correspondence to Walter Koch, MD, Department of Nuclear Medicine,
University of Munich, Marchioninistr 15, 81377 Munich, Germany
Tel: + 49 89 7095 4646; fax: + 49 89 7095 7646;
e-mail: walter.koch@med.uni-muenchen.de
Received 17 March 2013 Revised 21 May 2013 Accepted 28 June 2013
Introduction
Imaging of the presynaptic dopamine transporter (DAT)
has evolved to be an important diagnosti c tool in patients
with Parkinsonian syndromes [1]. DAT single-photon
emission computed tomography (SPECT) scans are used
to confirm or exclude a neurodegenerative Parkinsonian
syndrome [2] and, in combination with semiquantifica-
tion [3,4], can detect subtle changes in DAT binding in
striatal subregions and allow monitoring of disease
progression [5,6].
Recently, the enhanced image reconstruction algorithm
three-dimensional ordered subset expectation maximiza-
tion (3D-OSEM) has become available for DAT imaging.
The algorithm takes the depth response of the collima-
tors into account and has been shown to provide a
superior image quality in comparison with 2D-OSEM [7].
Its value for clinical routine use has been demon-
strated [8]. A superiority in low-count images enables
reduction of the injected radiotracer dose or the imaging
time [9–11].
The potential benefits of resolution recovery (as im-
plemented in 3D-OSEM) for semiquantitative analyses,
however, have not been examined yet. An increased
resolution could particularly aid in critical imaging
conditions, such as performing a DAT SPECT scan with
a high rotational radius, as is sometimes required because
of anatomical reasons or claustrophobia. Recently, the
rotation radius dependence of I-123-FP- CIT quantifica-
tion was shown for 2D-OSEM reconstructions [12].
The aim of this study was to systematically evaluate the
potential benefits of 3D-OSEM in comparison with
2D-OSEM for semiquantitative analyses and disease
detection based on the Monte Carlo simulation of studies
with various radii of rotation and different extents of
Parkinson’s disease.
Materials and methods
Phantom
The Zubal digital brain phantom (http://noodle.med.yale.edu/
zubal/, G. Zubal , Yale University, New Haven, Connecti-
cut, USA; [13]) was modified to simulate the typical
profiles of the normal radiotracer binding status as well as
neurodegeneration in Parkinsonian syndromes (loss of
DAT binding [2]). On the basis of previous measure-
ments with a physical phantom [14] and patient
Original article
0143-3636
c
2013 Wolters Kluwer Health | Lippincott Williams & Wilkins DOI: 10.1097/MN M.0b013e328364a9fd
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
scans [15], the activity distribution within the digital
phantom was chosen to reflect a realistic situation found
in healthy controls and patients. For simulation of normal
DAT binding, the activity concentrations of I-123 ratios
between the striatal structures of each hemisphere and
the remaining brain were 6 to 1 [15]. To simulate
neurodegeneration, an exponential loss of DAT binding
was modeled separately for the caudate and the putamen,
based on t values previously published in a long-term
follow-up study on patients w ith idiopathic Parkinsonian
syndromes [16] according to the formula: C
s
¼C
0
exp
T
t
;
where C
s
is the activity concentration of the respective
striatal region, C
0
equals the striatal concentration in a
healthy state, T is the years of disease, and the t values
(reflecting the rate of disease progression) being derived
from the study by Schwarz et al.[16].
Monte Carlo simulation
The SIMIND Monte Carlo code [17] was used to
calculate projection data based on the digital brain
phantom (256 256 matrix with 128 slices, 1.1 1.1
1.4 mm pixel size). A dual-headed MiE ECAM variable
SPECTcamera (MiE, Seth, Germany) equipped with low-
energy, high-resolution parallel hole collimators (parallel
hexagonal holes with cells of 1.11 mm diameter, 2.405 cm
height, and 0.16 mm septal thickness) was entirely
modeled in the software. The comparability of the
simulated data of this camera type with real equipment
has been confirmed elsewhere [18]. The acquisition
parameters were based on recommendations outlined in
the procedure guidelines for neurotransmission SPECT
with DAT ligands published by the European Association
of Nuclear Medicine [3] and were applied to the Monte
Carlo simulation. All acquisitions were optimized not only
to obtain a high spatial resolution but also to reflect the
clinical use of the SPECT systems. A total of 120
projections were obtained for each simulation, with the
detector heads following a 3601 circular orbit in a
128 128 matrix with a main energy window from 143.1
to 174.9 keV. In addition, a lower (131.9–143.0 keV) and an
upper (175.0–186.1 keV) scatter window adjacent to the
main window were acquired. The pixel size was
3.0 3.0 mm. Physical effects, such as photon attenuation
and scatter in the phantom and the crystal, degradation
due to collimator resolution, septal penetration, photon
interaction in the collimator [19], and backscatter from
the detector cover material were included in the
simulations. The full energy spectrum of I-123 was
simulated. Ten million counts in the main energy window
were simulated for each acquired projection to obtain low
noise simulation data. Study counts of the main window
were then scaled to obtain total counts of 2.5 million per
acquisition, as typically acquired in true patient scans.
The scatter windows were consecutively scaled with
identical factors. Finally, Poisson-distributed noise was
added to the projection data. Apart from the original main
energy window acquisitions, scatter-corrected data were
calculated based on the triple energy window correction
method [20,21].
Simulated disease states and radii of rotation
A healthy state (T = 0 years) as well as disease states 2, 4,
6, 8, and 10 years after disease onset were simulated. The
simulations therefore covered a wide range, from entirely
normal to preclinical as well as far-progressed disease
states. Each disease state was imaged with 13, 14, 15, 16,
17, 18, 19, 20, and 30 cm radii of SPECT rotation.
SPECT processing
Simulated SPECT acquisition data were transferred to a
real MiE ECAM variable camera acquisition workstation
and reconstructed with a 2D-OSEM algorithm (OSEM
implementation based on the algorithm of Richard Larkin
from Macquarie University [22]) and with a 3D-OSEM
algorithm (depth response OSEM) using the MiE
Scintron software (MiE Medical Imaging Electronics,
Seth, Germany). For both 2D-OSEM and 3D-OSEM
reconstructions, projection data were smoothed using a
two-dimensional Gaussian filter with a full-width at half-
maximum of 5.65 mm. Smoothing was performed by
convolution of the projection with a filter mask in each
direction. The length of the filter mask is 3 with the
weighting [1, 2, 1]/4. Thus, each pixel in a projection is
first smoothed in the x-direction by a weighted sum
including twice itself and its left and right neighbor
divided by 4 in order to be count preserving. Thereafter,
the procedure is repeated in the y-direction using the
upper and lower neighbors of each pixel. Four iterations
with 16 subsets were used to reconstruct data, and the
reconstructions were corrected for attenuation (m = 0.11 /
cm, automated contour finding with separate contours for
each slice), following Chang’s method [3] for 2D-OSEM.
For 3D-OSEM, attenuation correction was integrated
into the reconstruction algorithm.
Automated VOI evaluation
Using shift transforms only, the digital phantom was
coregistered to the reconstructed transverse slices of the
13 cm rotational radius acquisition of the healthy state.
Three-dimensional volumes of interest (VOIs) for the
striatal regions were defined based on digital phantom
morphology (caudate or putamen). A large occipital
background region was added (8624 voxels), which served
as a reference for all semiquantitative analyses. The VOI
sizes are given in Table 1.
Semiquantitative evaluation
Specific binding within the striatum, caudate, and
putamen were calculated from the mean counts per
voxel, with the occipital cortex serving as a reference
[specific binding
striatum
= (striatum – occipital reference)/
occipital reference]. Because the underlying disease in
patients with Parkinsonian syndromes often affects the
caudate nucleus and putamen with a varying severity,
972 Nuclear Medicine Communications 2013, Vol 34 No 10
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
the putamen-to-caudate ratios (P-to-C ratios = ratio
between specific putaminal and specific caudate binding)
were also calculated.
Statistical analyses
Linear regression analyses were used to describe the
relationship between specific binding ratios and radii of
rotation. The slopes and SE of slopes were calculated.
To detect differences in the slopes of the linear
regression curves, analysis of covariance was applied,
investigating the significance of the interaction between
the classification effect (such as the reconstruction
method) and the covariate (the specific binding ratio).
To determine the influence of rotational radii on the
measured binding values from a statistical point of view,
multivariate general linear regression was used: true
striatal binding in the phantom (as the dependent
variable) was predicted on the basis of the covariates
measured binding and radius of rotation. A high slope/
correlation coefficient of the rotational radius therefore
indicates a strong impact of this facto r, whereas a low
correlation coefficient indicates a weak influence.
All statistical analyses were performed using SPSS
Software version 13 (SPSS Inc., Chicago, Illinois, USA).
For automation of digital phantom ‘filling’, multithreaded
Monte Carlo simulation, file format conversions, calcula-
tion of noise with Poisson distribution, DICOM packa-
ging, data exchange with a real MiE SPECT camera, and
automation of VOI quantification, an in-house software
written in VB.NET 2010 (Visual Studio 2010; Microsoft
Corp., Redmond, Washington, USA) was used. VOI
quantification was performed using the MIPAV software
(Center for Information Technology, National Institutes
of Health, Bethesda, Maryland, USA). Reconstruction of
the raw acquisition data sets on Sc intron software was
controlled by an in-house automation software based on
the AutoIt scripting technology (AutoIt Consulting Ltd,
Birmingham, UK, http://www.autoitscript.com).
The results, images, and statistics presented in text,
tables, and figures are based on acquisition data without
scatter correction, if not explicitly mentioned.
Results
Recovery
The measured specific striatal binding was co mpared
with the true specific binding ratios in the phantom for
both 2D-OSEM-reconstructed and 3D -OSEM-recon-
structed imag es in the healthy state. Independent
of the radius of rotation, the measured striatal binding
ratios were slightly higher for 3D-OSEM images than for
2D-OSEM images. At the minimum radius of rotation
(13 cm), the percentage recovery was 53% for 3D-OSEM
and 52% for 2D-OSEM. At the maximum rotational
radius (30 cm), the recovery decreased to 43% for
3D-OSEM and to 26% for 2D-OSEM. Addition of
scatter correction increased the recovery at 13 cm to 63
and 60% and at 30 cm to 51 and 42% for 3D-OSEM and
2D-OSEM, respectively.
Radius dependency of the measured striatal binding
The correlations between specific binding ratios and radii
of rotation in the healthy state for 2D-OSEM and 3D-
OSEM reconstructions are shown in Fig. 1 (uncorrected
data Fig. 1a, scatter-corrected data Fig. 1b). A linear loss
of striatal binding with an increasing radius of rotation
could be observed for both methods of reconstruction,
the observation being independent of whether uncor-
rected or scatter-corrected data were used. For 3D-
OSEM, the decrease, however, was less steep compared
with that fo r 2D-OSEM (significantly different slopes: F-
test P < 0.05), which is in line with the higher recovery
values observed for 3D-OSEM in scans with larger
rotational radii. Table 2 shows the results of the multi-
variate linear regression analyses. The results clearly
demonstrate a higher influence (correlation coefficient)
of the rotational radius on striatal binding in 2D-OSEM
images than in 3D-OSEM images.
Figure 2 exemplarily shows images of the healthy state
reconstructed with both 2D-OSEM and 3D-OSEM with
different radii of rotation. 3D-OSEM images showed a
better delineation of the caudate and the putamen and
a smoother background (nonspecific binding) compared
with 2D-OSEM images. With a higher radius, the
resolution decreases, the activity spreads beyond the
true borders of the striatum, and the partial-volume
effects increase in all three dimensions, resulting in a loss
of recovery, which is more pronounced in 2D-OSEM
images. Although a horizontal, 15-mm-wide line profile
through the striatal area in an acquisition with a low
rotational radius shows a steep count increase from
unspecific binding to striatal binding, the increase
becomes shallower with an increase in the radius of
rotation, with this effect being shown equally in both
methods of reconstruction.
Putamen-to-caudate ratios
To estimate the potential beneficial effects of 3D-OSEM
in comparison with 2D-OSEM in a clinical routine
setting, we directly co mpared the P-to-C ratios between
both methods of reconstruction as an objective paramet er
for determining the predominant putaminal binding loss
typically observed in Parkinson’s disease. Because low
Table 1 Volumes of interest used for SPECT quantification
Method of analysis Region Size (ml)
Morphological VOI Striatum 6.40
Caudate 2.39
Putamen 4.01
Occipital reference VOI Occipital 132.87
SPECT, single-photon emission computed tomography; VOI, volume of interest.
3D-OSEM vs. 2D-OSEM in FP-CIT SPECT Koch et al.973
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
striatal binding values result in statistical noise, these
analyses were confined to a disease progression of up to a
maximum of 10 years. Figure 3 exemplarily shows the
correlations between the measured and true P-to-C ratios
for 13 cm of rotation and 30 cm of rotation for both
methods of recon struction.
At 30 cm of rotation, the 3D -OSEM seems to outperform
2D-OSEM, with better reproduction of the true P-to-C
ratios. On defining an arbitrary threshold for disease
detection at a P-to-C ratio of 0.7 (30% relative loss of
putaminal binding), at 13 cm of rotation, di sease could be
detected 3.5 years after onset using 2D-OSEM and 3D-
OSEM; however, at 30 cm of rotation, the disease would
have had to have progressed 5.3 years in order to be
detected on 3D-OSEM images, but would have had to
have progressed to at least 8.0 years for detection with
2D-OSEM reconstructions.
This presumed difference between the reconstruction
methods, however, could only be observed on comparing
the minimum and maximum rotational radii. Across all
radii of rotation, no significant benefit of 3D-OSEM in
terms of P-to-C ratio reproduction could be shown (F-test
P = 0.782).
Discussion
Imaging of the presynaptic DAT has evolved into an
important diagnostic tool for patie nts with Parkinsonian
syndromes [1,23–25], and thus has become a routine
clinical procedure. Visual assessment of DAT SPECT
studies in many cases enables clinicians to decide
whether neurodegeneration of presynaptic neurons has
occurred and to confirm or exclude a neurodegenerative
Parkinsonian syndrome [2]. In particular, for an early
diagnosis, that is, the detection of subtle changes in DAT
binding in striatal subregions, and for monitoring disease
progression [5,6,26] or the beneficial effects of putative
neuroprotective drugs [5,6,16,27,28], additional semi-
quantitative measurements are essential [3].
3D-OSEM is assumed to have the potential to outper-
form filtered backprojection and 2D-OSEM in terms of
image quality [9–11].
Fig. 1
2.8
(a) (b)
3D-OSEM
2D-OSEM
Fit line for 3D-OSEM
Fit line for 2D-OSEM
Reconstruction method
3D-OSEM scatter corrected
2D-OSEM scatter corrected
Fit line for 3D-OSEM scatter corrected
Fit line for 2D-OSEM scatter corrected
Reconstruction method
2.6
2.4
Specific striatal binding
2.2
2.0
1.8
10 15 20 25 30
Radius of rotation
10
2.0
2.2
2.4
2.6
2.8
3.0
3.2
15 20 25 30
Radius of rotation
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
∗
Correlations between the specific binding ratios and radii of rotation in the healthy state in 2D-OSEM-reconstructed and 3D-OSEM-reconstructed
images for (a) uncorrected simulation data and (b) scatter-corrected data. OSEM, ordered subset expectation maximization.
Table 2 Relation between the radius of rotation and the measured striatal binding for 2D-OSEM and 3D-OSEM: results of the multivariate
linear regression analyses
Regression coefficient
Method of reconstruction Measured striatal binding±SE Radius of rotation±SE Constant±SE
2D-OSEM 2.146±0.026 0.320±0.004 – 0.643±0.76
3D-OSEM 2.027±0.016 0.180±0.002 – 0.331±0.48
OSEM, ordered subset expectation maximization.
97 4 Nuclear Medicine Communications 2013, Vol 34 No 10
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
FP-CIT semiquantification shows an acquisition radius
dependence that is attributed to partial-volume effects
due to a drop in the resolution with increasing radius, as
was recently shown by Larsson et al. [12]. With an
increase in the full-width at half-maximum of the spatial
resolution in reconstructed images, activity from within
the stri atum spreads over an increasing area as shown
in Fig. 2, resulting in a loss of counts within the
quantification VOIs (spill-out).
Radius dependence was less prominent when a three-
dimensional cone beam model was incorporated in
the iterative recon struction algorithm (3D-OSEM). The
resulting enhanced resolution recovery led to higher
binding ratios in scans with high radii of rotation.
The difference in recovery when comparing measured
and true specific striatal binding using both reconstruc-
tion methods in scans with minimal rotational radii was
low (1.9%); the annual loss of DAT binding in patients
with idiopathic Parkinsonian syndromes is B5.2% per
year [29]. It could also be attributed to the slight
differences in attenuation correction or filtering. The
more prominent differences in scans with high radii
of rotation (up to 19.4% at 30 cm) will most likely be
related to the higher spatial resolution in 3D-OSEM
images. 3D-OSEM therefore seems more robust to the
influence of the radius of rotation. The overall measured
binding values based on uncorrected data were in a
typical range of about half the true activity ratios in the
phantom, which can be attributed mainly to scatter and
partial-volume effects. Scatter correction led to a general
increase in recovery, in studies with both low and high
radii of rotation, but did not have an effect on the radius
dependence of the binding values. Scatter correction,
however, is typically not applied in a routine setting. If
further reduction of partial-volume effects is required,
the ‘Southampton’ semiquantification method proposed
by Fleming and colleagues [30,31] would provide a
promising method to overcome the influence of rotational
radius effects.
We would have expected that an increased spatial
resolution also has beneficial effects on the accuracy of
the P-to-C ratios. Parkinson’s disease typically affects the
putamen earlier, given the more prominent nigrostriatal
degeneration in that region [32,33] and the particular
degeneration of the ventrolateral substantia nigra pars
compacta [34], which innervates the posterior puta-
men [35]. P-to-C ratios therefore offer diagnostic informa-
tion, independent of total specific striatal binding [36].
Fig. 2
(c)(b)(a)
2D-
OSEM
2D-
OSEM
3D-
OSEM
3D-
OSEM
Examples for the effects of rotational radii on the image quality. Below each example, a 15 mm horizontal line profile centered on the striatal region is
shown. 2D-OSEM (first row), 3D-OSEM (second row), (a) a 13 cm radius of rotation, (b) a 20 cm radius of rotation, and (c) a 30 cm radius of
rotation. OSEM, ordered subset expectation maximization.
3D -OSEM vs. 2D-OSEM in FP -CIT SPECT Koch et al.975
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.