European Journal of Radiology 82 (2013) 862869

Contents lists available at SciVerse ScienceDirect

European Journal of Radiology

journal homepage: www.elsevier.com/locate/ejrad

Clinical evaluation of PET image reconstruction using a spatial

resolution model
Flemming Littrup Andersen a , Thomas Levin Klausen a , Annika Loft a , Thomas Beyer a,b ,
Sren Holm a,
a
b

Department of Clinical Physiology, Nuclear Medicine and PET, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark
cmi-experts GmbH, Zurich, Switzerland

a r t i c l e

i n f o

Article history:
Received 4 September 2012
Received in revised form 7 November 2012
Accepted 8 November 2012
Keywords:
Positron emission tomography
PET/CT
Spatial resolution
Point spread function
Image reconstruction

a b s t r a c t
Purpose: PET image resolution is variable across the measured eld-of-view and described by the point
spread function (PSF). When accounting for the PSF during PET image reconstruction image resolution is
improved and partial volume effects are reduced. Here, we evaluate the effect of PSF-based reconstruction
on lesion quantication in routine clinical whole-body (WB) PET/CT imaging.
Materials and methods: 41 oncology patients were referred for a WB-PET/CT examination (Biograph 40
TruePoint). Emission data were acquired at 2.5 min/bed at 1 h pi of 400 MBq [18F]-FDG. Attenuationcorrected PET images were reconstructed on 336 336-matrices using: (R1) standard AW-OSEM (4 iter,
8 subsets, 4 mm Gaussian) and (R2) AW-OSEM with PSF (3 iter, 21 subsets, 2 mm). Blinded and randomised reading of R1- and R2-PET images was performed. Individual lesions were located and counted
independently on both sets of images. The relative change in PET quantication (SUVmax , SUVmean , volume) of lesions seen on R1 and R2 is reported as (R2 R1)/R1. Furthermore, SUVmax and SUVmean was
measured for a 3 cm spherical norm region in the right lobe of the healthy liver for R1 and R2.
Results: Clinical reading revealed 91 and 103 positive lesions for R1 and R2, respectively. For all lesions
SUVmax (R2) was higher than SUVmax (R1). Regression analysis indicated that the relative increase in
SUVmax (and SUVmean ) decreased with lesion size, whilst it increased with increasing radial distance from
the centre of the eld of view (FOV). There was no signicant difference in SUVmean in homogenous liver
tissue between R1 and R2.
Conclusion: In whole-body FDG-PET/CT using routine clinical protocols, PSF-based PET reconstruction
increases lesion detection and affects SUVmax measurements compared to standard AW-OSEM PET reconstruction.
2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction
Since the introduction of the rst integrated PET/CT system in
1998 [1], this combined modality has gained widespread use in
oncology for the diagnosis, staging and restaging of disease, using
primarily the glucose analogue [18F]-uoro-deoxy-glucose (FDG)
as a tracer of choice [2]. Dual-modality PET/CT is also used for the
denition of target lesions during the course of radiotherapy planning [3]. PET has also been suggested to be benecial for dose
painting [4], that is to dene volumes with high tracer uptake that
could gain from a specic localized dose escalation, or radiation
boost [5].

Corresponding author at: Department of Clinical Physiology, Nuclear Medicine

and PET, Rigshospitalet, Blegdamsvej 9, DK-2100 Copenhagen, Denmark.
Tel.: +45 3545 1681; fax: +45 3545 3898.
E-mail address: sholm@pet.rh.dk (S. Holm).
0720-048X/$ see front matter 2012 Elsevier Ireland Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ejrad.2012.11.015

PET images must be of high quality and reliable quantitative accuracy whenever employed for diagnostic and therapeutic
strategies. However, the spatial resolution of PET is signicantly
lower than that of images acquired with CT or MRI. This is caused
by several factors [6]. The positron range effects [7] and the acolinearity of emitted annihilation photons cause the registered
coincidence line to be offset from the actual point of decay. In
addition, the nite detector size of the PET tomograph as well
as photon-detector interactions and scatter further degrade the
spatial resolution. Finally, the actual image reconstruction and
post-reconstruction ltering limit the resolution of the PET images.
In general, the spatial resolution of a PET system can be
described by a point spread function (PSF). The description of
the PSF is simplied by reporting the full width at half maximum (FWHM) along three orthogonal axes (radial, tangential, and
axial) with reference to the cylindrical geometry of the PET detector arrangement. Of note, the PSF is spatially variable within the
measured eld-of-view (FOV). This is mainly related to depth-ofinteraction (DOI) effects within the detector elements. Photons

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

originating near the scanner centre or travelling along a coincidence line near the centre of the FOV impinge the crystal surface
at a right angle. Thus they experience a rst interaction within that
very crystal. In contrast, photons travelling along a line that is more
tangential to the FOV, enter the crystal surface of a given detector element at a smaller angle and therefore may penetrate into
neighbouring crystals before being stopped. As a result, the event
is assigned to an incorrect more centrally located coincidence line,
thus, broadening the PSF in the radial direction, while the tangential
PSF remains unaffected.
As a consequence of the limited spatial resolution, partial volume effects (PVE) are observed in PET imaging that lead to a
signicant underestimation of the activity concentration of lesions
smaller than two to three times the FWHM [8,9]. Given the spatial
variability of the PSF, the resulting PVE varies with the location of
the lesion with respect to the main tomographic axis.
Nonetheless, the variability of the PET image resolution can be
described by a model and introduced into the image reconstruction process [10]. Today, all major vendors of whole-body PET/CT
systems offer such algorithms for clinical use: TrueX (Siemens
Healthcare) [11,12], SharpIR (GE Healthcare) [13,14] and Astonish
TF (Philips Healthcare) with PSF model-based resolution recovery
capabilities.
The effect of PSF-reconstructions has been assessed in phantoms
[15,16] and in patients [1720]. When accounting for the PSF during
image reconstruction, spatial resolution is markedly improved and
may well approach isotropic values across the measured FOV [11].
In turn, partial volume effects in the reconstructed images can be
reduced, and lesion uptake values are expected to increase mainly
for small lesions.
Our study aims at assessing the impact of iterative reconstruction using a PSF model in clinical routine application of whole-body
(WB) PET/CT imaging. The impact of the spatial resolution recovery model is assessed by means of clinical lesion detectability and
measurements of the relative changes in standardized uptake values (SUV) with and without PSF-modelling. The impact and relative
changes are measured against the ndings in standard clinical
PET/CT protocols without PSF modelling as used in our daily routine.
2. Materials and methods
In this prospective study, 41 consecutive clinical patients were
included. All patients were referred for a clinical whole-body
PET/CT examination using FDG. Clinical indications for staging or
restaging included: gynaecological cancers (15), lymphomas (8),
mesotheliomas (3), lung cancer (2), melanomas (3), CUP (2), sarcomas (2) and others (6). Patients included 10 males (mean weight
80 kg (5495 kg), mean age 48 years (1568 years)) and 31 females
(mean weight 67 kg (4297 kg), mean age 60 years (2387 years)).
2.1. PET/CT imaging
All examinations were performed on a 4-ring LSO-based PET/CT
system, Biograph 40 TruePoint (Siemens Healthcare). Each detector ring consists of 48 detector blocks with 4 mm 4 mm 20 mm
crystals arranged in 13 13 panels. In total there are 32,448 crystals covering an axial eld-of-view of 21.8 cm, reconstructed into
109 transaxial image planes.
All patients underwent a standard whole-body PET/CT examination based on the EANM guidelines [21]. Patients were asked to fast
for 6 h prior to an activity injection of (400 40) MBq FDG. Patients
rested in a dimly lit room during the uptake time ((69 10) min;
range 5394 min). Prior to the PET/CT examination patients were
asked to void. Patients were then positioned on the PET/CT patient

863

handling system with their arms raised and supported above the
head.
The WB-PET/CT acquisition started with a topogram view,
followed by a spiral CT scan (120 kVp, automatic tube-current
modulation). PET emission data were acquired in 3D-mode at
2.5 min/bed position. CT images were reconstructed with a 70 cm
FOV and a slice thickness of 3 mm. Emission data were corrected
for scatter and attenuation using the acquired CT images [22] prior
to PET image reconstruction (AC-PET).
We have chosen WB-acquisition protocol and reconstruction
parameters as suggested by the vendor and following careful evaluation by our expert readers regarding signal and noise properties
in a clinical setting. Clearly, optimum choices of these reconstruction parameters are scanner model specic as they depend on
counting statistic and system geometry. Here, two reconstruction protocols were used, both based on attenuation weighted
ordered subset expectation maximization (AW-OSEM) [23]: (R1)
standard AW-OSEM (4 iterations, 8 subsets, 4 mm Gaussian postlter), and (R2) AW-OSEM with the PSF-model included in the
system matrix [19] (3 iterations, 21 subsets, 2 mm Gaussian). In
both cases (R1,R2), the images were reconstructed on 336 336
matrices with a pixel size of 2 mm 2 mm and a slice thickness
of 3 mm.

2.2. PET/CT image analysis

An experienced nuclear medicine physician with more than 8
years of clinical PET/CT experience reviewed the reconstructed ACPET images. The reader was blinded to the image reconstruction
protocol (R1 or R2). The PET images were presented in random order
with a minimum of 2 weeks between reconstructions (R1 or R2) of
the same patient.
Firstly, the reader was asked to grade the overall image quality of the whole-body AC-PET images on a scale of 5 (grade
1 = very poor, 2 = poor, 3 = acceptable, 4 = good, 5 = very good). Subsequently, the reader was asked to count the number of visible
lesions on the AC PET images with respect to the main anatomical regions: head/neck, axilla, mediastinum, lungs, liver, spleen,
abdomen, pelvic, inguinal, bones, other. Here, the term visible
refers to a PET-positive nding, as interpreted in a clinical setting,
whereby a PET-positive lesion is a clearly delineated hypermetabolic focus with a contrast sufciently high to differentiate the
lesion from background. In case of bulky and heterogeneous malignancies the lesion count was one. Clusters of lesions with more than
ve foci were localized and counted as ve.

2.3. PET quantication

Volume of interest (VOI) analysis was performed on a sample
of 58 lesions present in both image sets identied by the clinical
reader. Firstly, the radial distance (mm) from the centre of the lesion
to the centre of the transverse FOV was measured on the AC-PET
images from R1 and R2. In addition, the minimum diameter (mm) of
the hypermetabolic foci visible on PET (R1 and R2) as measured on
the transaxial PET image was recorded. The line measurement tool
in Osirix (version 4.1, 64 bit) was used for all linear measurements.
Secondly, a VOI analysis of all 58 lesions present on R1- and R2images was performed. Here, the orthogonal view with the most
prominent lesion activity on AC-PET was chosen (Siemens Multimodality Imaging Workstation, TrueD) and a 3D isocontour volume
was marked encompassing the hypermetabolic lesion. The isocontour threshold value was set to 40% of the SUVmax . In addition,
the SUVmean for that VOI and the corresponding VOI-volume per

864

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

lesion were recorded. All SUV measurements were normalized to

the weight of the patient and computed as:
SUV =

Tumour activity (Bq/ml) body weight (g)

injected dose (Bq)

(1)

Standardised uptake values (g/mL) and lesion volumes (mL)

were compared for the PET image reconstructions R1 and R2 in
order to assess the quantitative effect of PSF-modelling. The relative
change was calculated as:
R2 R1
(SUVmax (R2) SUVmax (R1)) 100
%=
R1
SUVmax (R1)

(2)

SUV, lesion volumes and relative changes are reported as functions of the distance from the centre of the FOV and the size of the
lesion.
Finally, as a reference value for homogenous tissue, a 3 cm diameter sphere was placed in a healthy region in the right lobe of the
liver (on both R1 and R2) and the SUVmean was reported. Graphs
include regression lines as determined from Excel (Microsoft Ofce,
2007). Furthermore, a simultaneous regression of the relative
changes in SUVmax , the lesion size and the radial position of the
lesion was performed. A Wilcoxon signed rank test (Sigmaplot 11.1,
Systat Software Inc, USA) was used for the paired data analysis of
SUVmax , SUVmean , and VOI-volume.

Patient #
41
39
37
35
33
31
29
27
25
23
21
19
17
15
13
11
9
7
5
3
1

PSF
AW-OSEM

Lesions

10

12

Fig. 1. Number of PET-positive lesions detected in all 41 patients following PET

reconstruction with R1 and R2; 91 and 103 hot-spots were located using R1 and R2,
respectively.

3. Results
3.1. PET/CT image analysis
The mean overall quality of the AC-PET images was 4.2 0.6
(R1) and 4.3 0.7 (R2) with the difference not being statistically
signicant (paired, 2-tailed t-test).
In total, 91 and 103 PET-positive lesions were counted following standard AW-OSEM reconstruction (R1) and AW-OSEM with
PSF model (R2), respectively. This difference was statistically signicant (p < 0.05, paired t-test). Fig. 1 shows the number of lesions
counted per patient for the two reconstruction methods. In 12/41
(29%) patients no lesion was detected on R1 and R2. In 17/41 (41%)
patients all lesions seen on R1 were seen also on R2. In 3/41 (7%)
patients the standard reconstruction (R1) led to 3 lesions not being

detected on R2. In 9/41 (22%) patients, image reconstruction with

PSF (R2) detected more lesions (15 in total) than on R1 (Fig. 2).
Fig. 3 shows illustrative clinical cases from the patient cohort
studied here, where PET image reconstructed with R2 resulted in
higher lesion counts than those reconstructed with R1.
3.2. PET quantication
Lesions were located within 5 mm (minimum) and 139 mm
(maximum) from the centre of the FOV. Fig. 4 shows the relative
change in SUVmax between R1- and R2-reconstruction for the 58
PET-positive lesions plotted over the distance from the centre of
the FOV. Linear regression analysis yielded a weak correlation with
r2 = 0.15.

Fig. 2. A 7-y/o child with a sarcoma referred for a WB FDG-PET/CT was scanned with the arms down. The injection site (aka port) is clearly seen on image (A) where the
injections site following standard PET reconstruction (R1) corresponds to a large hot spot. This uptake pattern is further resolved on the PSF-based reconstruction (R2) thus
tting individual tracer volumes to the ow structure of the injection system (B), further exemplied by the zoom-in fusion image in (C). PET images are scaled to SUV (g/mL)
and displayed between 0 and 1.4.

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

865

Fig. 3. The effect of PET image reconstruction with an integrated model of the PSF: (A) 62-y/o female with ovarian cancer and a lesion to the upper liver that was originally
missed on R1 and (B) 74-y/o female with lymphoma and a lesion in the left neck noted only on R2. The relative change in SUVmax from (R1, top) to (R2, bottom) was (A) 29%
(2.73.8 g/mL) and (B) 34% (2.53.8 g/mL).

The smallest and largest lesions were 4 mm and 44 mm in diameter, respectively. Fig. 5 illustrates the relationship between the
relative increase in SUVmax as a function of the minimum diameter
of the lesions. The correlation coefcient of the linear regression
was r2 = 0.11.
Table 1 summarizes the results from the VOI analysis for
all 58 lesions described on both reconstructions R1 and R2.
The relative changes in SUVmax and SUVmean were similar.
The mean relative changes were (46 27)% and (45 27)%
for SUVmax and SUVmean , respectively. Fig. 6A illustrates the

y = 1.34x + 0.62, with r 2 = 0.87

(3)

Fig. 7 shows a box plot comparing SUVmax , SUVmean and volume

for R1 and R2. Difference values for the two reconstruction schemes
were statistically signicant (p < 0.001, Wilcoxon signed rank test).
Following R2, lesion volumes decreased by 2% (minimum) and
79% (maximum) with the exception of one lesion (lesion 38), which

% SUVbw

% SUVbw

120

120

y = -1.23x + 59.46
R = 0.11

100

y = 0.31x + 24.09
R = 0.15

100

linear regression of SUVmax for R2 on R1 with the regression line

being:

80

80
60

60
40

40

20

20

0
0

20

40

60

80

100

Distance from center (mm)

120

140

Fig. 4. Relative change (%) in SUVmax for inclusion of PSF model as a function of the
radial distance of the PET-positive lesion from the centre of the FOV.

10

20

30

40

50

Tumor size (mm)

Fig. 5. Change (%) in SUVmax for inclusion of PSF model as a function of the size
(minimum diameter) of the PET-positive lesion. Activity concentration in smaller
size lesions increases more when using R2 compared to the change in SUVmax for
larger size lesions.

866

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

Table 1
Results from the distance measure and VOI analysis of 58 PET-positive lesions detected on PET images following AW-OSEM (R1) and AW-OSEM/PSF (R2) reconstruction.
Lesion

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58

Location

Pelvis
Abdomen
Abdomen
Pelvis
Pelvis
Pelvis
Pelvis
Abdomen
Other
Abdomen
Abdomen
Abdomen
Abdomen
Lung
Lung
Lung
Lung
Lung
Abdomen
Mediastinum
Mediastinum
Mediastinum
Mediastinum
Mediastinum
Mediastinum
Mediastinum
Mediastinum
Axilla
Axilla
Axilla
Head/neck
Head/neck
Abdomen
Bones
Bones
Abdomen
Abdomen
Abdomen
Abdomen
Pelvis
Inguinas
Inguinas
Lung
Pelvis
Head/neck
Head/neck
Pelvis
Pelvis
Pelvis
Lung
Lung
Lung
Mediastinum
Liver
Head/neck
Head/neck
Lung
Head/neck

Distance (mm)

17
34
87
63
35
96
45
110
63
77
95
99
85
105
148
58
137
127
139
6
110
72
34
25
27
61
36
100
91
85
54
54
73
97
100
22
31
46
44
54
69
34
62
44
49
53
58
44
49
56
140
124
59
81
67
62
93
90

Size (mm)

15
7
13
19
15
12
15
9
6
7
17
6
6
11
8
19
7
5
6
5
8
12
10
10
14
13
11
5
5
5
7
7
9
9
7
11
9
4
44
9
6
9
31
16
4
6
7
18
25
8
8
6
6
11
7
6
16
27

AW-OSEM
SUVmax

SUVmean

14.2
6.7
7.9
10.6
10.7
6.4
12.5
6.1
2.7
5.6
7.4
5.5
6.8
9.2
5.2
10.4
4.9
4.8
4.0
9.2
5.8
11.3
4.7
5.4
7.4
4.5
5.6
2.1
1.3
1.4
3.4
2.8
4.9
9.7
4.4
8.0
7.1
4.9
8.5
4.7
2.4
8.8
20.0
11.4
2.5
3.9
6.9
8.7
4.2
2.5
6.9
2.7
6.1
3.9
6.1
6.1
4.0
2.3

8.2
4.1
4.7
6.4
6.1
3.7
6.9
3.6
1.6
3.2
4.4
3.1
3.7
5.3
2.9
6.4
2.8
2.7
2.3
5.1
3.2
6.4
2.7
3.0
4.2
2.6
3.2
1.2
0.9
0.9
2.0
1.7
3.2
5.9
2.7
4.5
4.0
2.5
4.6
2.6
1.4
5.2
11.6
6.6
1.6
2.1
4.2
5.4
2.1
1.5
4.1
1.5
3.1
2.8
3.2
3.7
2.3
1.5

PSF
Volume (ml)
4.2
2.6
2.1
22.4
11.5
3.2
4.7
3.1
1.0
3.3
8.7
3.2
4.6
1.8
2.8
3.6
2.9
1.8
0.7
1.7
1.6
1.1
3.6
4.6
6.5
6.3
6.4
1.3
1.6
4.0
2.0
2.1
4.2
2.2
1.7
2.1
2.3
1.2
211.0
2.2
2.0
3.8
11.7
7.5
1.1
2.1
0.8
8.3
35.0
0.9
0.9
1.0
4.5
6.4
2.3
2.3
18.0
5.5

SUVmax

SUVmean

Volume (ml)

SUVmax (%)

SUVmean (%)

volume (%)

20.9
10.2
11.2
11.9
12.7
8.7
16.8
10.8
4.7
8.1
9.2
8.3
9.9
14.0
6.5
14.0
6.3
7.3
8.1
9.8
9.6
20.5
6.2
7.1
10.6
5.5
5.9
3.2
1.7
1.9
4.3
3.7
5.5
19.4
9.4
11.9
9.1
4.8
8.6
5.8
4.2
13.9
26.0
14.0
3.8
5.6
13.8
12.6
5.9
4.8
13.3
4.8
9.4
4.8
8.9
10.2
7.2
2.9

12.4
6.1
6.6
7.4
7.0
5.3
9.1
5.9
2.6
4.6
5.5
4.4
5.7
7.9
3.6
8.4
3.4
3.9
4.8
5.5
5.5
11.8
3.5
3.9
5.8
3.1
3.5
1.9
1.2
1.2
2.6
2.1
3.5
11.4
5.4
7.0
5.2
2.6
4.7
3.4
2.4
8.3
15.0
8.1
2.1
3.3
8.6
7.4
2.8
2.8
7.5
2.7
5.1
2.8
5.0
6.3
3.7
1.9

2.6
1.4
1.2
18.8
8.8
1.8
2.4
1.5
0.4
1.6
6.3
1.8
1.3
1.0
1.8
2.0
2.1
1.5
0.3
1.7
0.8
0.5
2.4
3.0
3.1
5.1
5.3
0.6
1.0
2.3
1.4
1.5
3.1
1.0
0.6
0.8
1.3
2.3
206.0
1.8
0.6
1.3
8.7
4.3
0.9
0.6
0.3
5.2
16.5
0.4
0.5
0.4
1.0
4.7
1.0
0.9
5.7
4.1

47
53
43
12
18
35
34
77
71
46
25
50
46
52
26
35
30
50
105
7
65
82
32
32
44
22
6
54
33
28
29
34
12
101
111
49
28
1
1
23
72
57
30
23
51
45
101
45
41
91
91
79
53
23
47
66
79
27

52
49
42
14
15
44
31
64
68
44
25
41
55
51
26
31
21
44
110
7
71
85
30
31
38
18
9
57
32
41
29
24
10
92
99
54
31
5
2
29
79
59
29
22
29
55
106
36
29
87
83
78
64
1
56
71
63
26

39
46
42
79
24
44
49
53
61
53
27
44
71
42
35
46
27
14
63
2
51
54
33
34
53
19
17
53
37
43
31
29
26
57
67
60
43
5
2
20
68
65
26
42
21
71
63
38
53
58
50
58
79
27
58
59
68
26

The table includes the anatomical location of the lesion, the minimum diameter (mm), the distance from the centre of the FOV (mm), as well as SUVmax , SUVmean and lesion
volume for R1 and R2. SUVmean and volume were determined from VOI analysis based on 40% iso-contour. The table also lists the relative changes in SUVmax , SUVmean and
lesion volume.

increased in volume by 5%. Fig. 6B shows the SUVmean in the norm

region of the liver for both PET reconstructions R1 and R2. The linear
regression line was found to be:
y = 1.01x + 0.05 with r 2 = 0.97

(4)

The tracer concentration in the liver norm region varied by

(3 3)% (1.7% 13.5%) between R1 and R2.
A regression analysis of the relative change in SUVmax as a function of both the lesion diameter and the distance from the centre

was also performed. An independent regression analysis of the relative changes in SUVmax as a function of lesion size (mm) (Eq. (5))
and distance (mm) from the centre (Eq. (6)) showed:
R2 R1
% = 59.5 1.2 size
R1

(5)

R2 R1
% = 24.1 + 0.31 distance
R1

(6)

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

867

PSF (SUVbw)
(g/ml)
30

25
20
15
y = 1,34x + 0,62
R = 0,87

10
5
0

Fig. 7. Box plots of SUVmax , SUVmean and lesion volume for AW-OSEM (R1) and
AW-OSEM + PSF (R2) reconstruction. In all cases the change proved highly statistical
signicant (P < 0.001) using Wilcoxon signed rank test.

10
15
20
AW-OSEM (SUVbw) (g/ml)

25

compares favourably with the independent regression analyses in

Eqs. (5) and (6).

PSF (SUVbw)
(g/mL)
4

4. Discussion

2
y = 1.01x + 0.05
R = 0.97

2
3
AW-OSEM (SUVbw) (g/mL)

Fig. 6. (A) PET-positive lesions: SUVmax (R2) (g/mL) plotted over SUVmax (R1). (B)
Liver: SUVmean (g/mL) for R2 plotted over SUVmax (g/mL) for R1.

Assuming a simplistic linear relationship of the two variables

size (mm) and distance (mm) the regression can be described as:
R2 R1
% = 37.7 1.0 size + 0.27 distance
R1

(7)

where (R2 R1)/R1% is the relative change (%) in SUVmax , size is the
minimum diameter of the lesion and distance is the distance of that
lesion from the centre of the FOV. While the regression coefcient
r2 of the dual-parameter t was somewhat higher (r2 = 0.22) than
that obtained from independent 2D-ts in Figs. 3 and 4, Eq. (7)

This study aimed at comparing two independently adopted WBPET/CT protocols from our clinical routine workow. We compared
the lesion detectability and the relative differences in lesion volume, size and uptake for two routinely used image reconstruction
protocols. We used a consecutive patient cohort from our clinical
PET/CT programme.
This study indicated a signicant effect of iterative image reconstruction employing a spatial image resolution model (PSF) on
clinical PET image quality and quantication. We observed an
increase in PET-positive lesions, as well as a signicant change
in the measured SUV and volumes of these lesions. We compared
two reconstruction protocols that are standard in our clinical routine. The reconstruction parameters for R1 and R2 were selected
independently for each protocol by our clinical readers. The settings were in line with previous reports on reconstruction protocols
for performance evaluations of the TruePoint series of WB-PET/CT
tomographs (Siemens Healthcare) for optimal signal-to-noise ratio
[12].
It is known that PSF-based reconstruction (R2) can improve
resolution compared to standard AW-OSEM without PSF modelling [11,14] and, therefore, we reduced the Gaussian postreconstruction lter applied in the image-domain from 4 mm to
2 mm to take advantage of this improvement.
In our patient cohort, the number of PET-positive lesions
reported was increased from 91 in R1 by 13% to 103 in R2, that is
when using iterative image reconstruction together with a model
of the spatial resolution function of the PET system (Fig. 1). This
increase in detected PET-positive lesions was statistically signicant (p < 0.05, paired t-test). Further, visual comparisons of
standard (R1) and PSF-based (R2) reconstruction (Fig. 3) supported
an increased reader condence for reporting a positive nding.
Fig. 2 shows a visually convincing case, albeit of little clinical signicance, with prominent changes in the activity pattern following
(R2)-reconstruction.
It is well known that the use of a PSF model causes overshoot
along the edges (Gibbs artefacts) [10]. Whilst this observation is
not widely published it is under investigation by several imaging
groups. First reports suggest that the overshoot effect is rather small
at clinical noise levels [13]. Recent studies by Lasnon et al, indicate a

868

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

slightly increased but statistically not signicant false positive rate

from these overshoot effects [25].
In our study, AW-OSEM (R1) outperformed AW-OSEM with PSF
(R2) in 3 patients by the number of lesions detected. In patient 7
(breast cancer) heterogeneous uptake in the left breast was read as
positive on R1 and negative on R2. In patients 31 and 37 one more
abdominal lesion was detected on the R1-images when compared
to R2. Upon retrospective comparison with both image volumes
displayed side-by-side the lesions were visible on either image set.
We attribute the disconcordance of these 3 readings to the intraobserver variability during the blinded reading of the randomly
presented PET image volumes.
All lesions showed a higher SUVmax when using PSFreconstruction (R2) compared to the standard AW-OSEM (R1),
except for abdominal lesion number 38, without any change in
SUVmax (Table 1). The change in SUV was analysed as a function of the radial distance from FOV (Fig. 4) as well as a function
of lesion size (Fig. 5). Our data (Fig. 4) conrm that image resolution improves for lesions further away from the centre when
PSF-modelling is included in the reconstruction [11]. Nonetheless,
this correlation was weak, since the relative change also depends on
the size of the lesion (Fig. 5). Neither Fig. 4 nor Fig. 5 can represent
the full scope of this complex relationship.
While this planar regression is not the most suitable model, it
does demonstrate a basic correlation of gain in SUVmax and both
lesion size and distance from the centre. The improvement in resolution is larger in the radial and axial directions, and the PSF-gain
for a non-spherical lesion is likely to depend on the orientation of
the reported minimum axis. We did not attempt to include this
variable. This may partly explain the low regression coefcient.
In addition to the change in SUVmax and similar changes in
SUVmean (Table 1) we observed a change in lesion volume. When
applying a 40% isocontour level the average reduction in lesion volume from R1 to R2 was 44.8% (range 1.878.6%). Fig. 7 presents
box plots for the changes in lesion volume as well as SUVmax and
SUVmean from introducing the PSF model in the standard reconstruction.
Interestingly, the observed relative change in SUVmax and
SUVmean following PSF-based reconstruction (R2) is of similar magnitude to the 25% threshold in classifying a change in therapy
response based on the EORTC criteria [24]. Therefore, PET and
PET/CT data of an individual patient undergoing multiple longitudinal examinations must be reconstructed using the same protocol
[21] in order to avoid unsubstantiated changes in patient management. Similarly, standardised protocols employing a xed set
of acquisition and reconstruction parameters should be applied
in study settings and multi-centre studies, in order to minimize
systematic quantication bias [26].
In addition to EORTC response criteria, total lesion glycolysis (TLG) has been suggested as a valid quantitative measure to
describe disease burden and changes thereof under therapy [27].
TLG is the product of the SUVmean and the lesion volume. Our results
in Table 1 indicate a rather stable TLG across the two image reconstruction methods R1 and R2 (21.5% 15.2% indicating that the
TLG may be considered an alternative to reporting tumour activity
and possibly therapy response when the reconstruction parameters cannot be fully standardized, such as in cases where external
PET or PET/CT readings are received prior to the requested in-house
PET examination.
Recently, Wahl et al. introduced a new therapy response assessment algorithm, referred to as PERCIST criteria [28], that is based on
norm reference uptake values in the liver or mediastinum and the
SUV normalized to the lean body mass (SUVlbm). We have assessed
the quantication in a norm region in the liver (Fig. 6B). Our study
shows the tracer concentration in the liver remains unchanged
(regression slope of 1.01) with an absolute variation of only 3%

when changing the reconstruction paradigm from R1 to R2. Therefore, PERCIST-based response assessment is unlikely to be affected
by the PSF-based reconstruction, as long as pre- and post-therapy
PET data are reconstructed using the same algorithms. Likewise,
changes in tumour-to-background (TBR) ratios, when based on the
liver, will follow the changes in the lesions.
However, considering the spatial dependence of the change in
SUV as a function of the distance of the lesion from the centre
(Fig. 3), care should be taken in repositioning the patient for the
follow-up exam in order to avoid signicantly different locations
of the known lesions inside the FOV.
Our study has several limitations. We reported and counted
PET-positive lesions without a histological or clinical reference.
Therefore, the effect of image reconstruction with a resolution
model (PSF, R2) can be described only as a relative change with
respect to the standard implementation of the existing reference
method (AW-OSEM, R1) as currently implemented on our PET/CT
systems. Finally, our study group of 41 patients with 58 lesions
assessed quantitatively, was comparatively small.
Nonetheless, our results indicate a signicant effect of PSF-based
reconstruction using routine protocols on lesion appearance and
quantication. The implications of this may vary with the choice
of the therapy response assessment. In any case, these observations warrant again the awareness for and the need to adopt
standardised acquisition and data processing protocols in single
and multi-centre settings.
5. Conclusions
Our study indicates increased lesion detection rates as well
as noticeable effects on lesion quantication for PSF-based
reconstruction compared to the standard AW-OSEM image reconstruction in FDG-PET. These quantitative changes may impact
diagnosis and staging as well as therapy response assessments
depending on the algorithm chosen. Therefore, standardised imaging protocols are encouraged for intra- and inter-patient analysis.
Conict of interest
Thomas Beyer is founder and president of cmi-experts GmbH
and reports no conict with this submission.
Acknowledgements
This work was supported by the John and Birthe Meyer Foundation.
We thank Marianne Federspiel and Elisabeth Abrahamson for
their help with the data acquisition, reconstruction and analysis.
References
[1] Beyer T, Townsend DW, Brun T, et al. A combined PET/CT tomograph for clinical
oncology. Journal of Nuclear Medicine 2000;41(8):136979.
[2] Czernin J, Allen-Auerbach M, Schelbert H. Improvements in cancer staging with
PET/CT: literature-based evidence as of September 2006. Journal of Nuclear
Medicine 2007;48(Suppl. 1):78S88S.
[3] Zhu A, Marcus D, Shu H, Shim H. Application of metabolic PET imaging in
radiation oncology. Radiation Research 2012;177(4):43648.
[4] Bentzen S. Theragnostic imaging for radiation oncology: dose-painting by numbers. Lancet Oncology 2005;6(2):1127.
[5] Elmpt Wv, Ruysscher DD, Salm Avd, et al. The PET-boost randomised phase II
dose-escalation trial in non-small cell lung cancer. Radiotherapy and Oncology
2012;104(1):6771.
[6] Bendriem B, Townsend D. The theory and practice of 3D PET. Dordrecht: Kluwer
Academic Publishers; 1998.
[7] Jdal L, Loirec CL, Champion C. Positron range in PET imaging: an alternative
approach for assessing and correcting the blurring. Physics in Medicine and
Biology 2012;57(12):393143.
[8] Soret M, Bacharach S, Buvat I. Partial-volume effect in PET tumor imaging.
Journal of Nuclear Medicine 2007;48(6):93245.

F.L. Andersen et al. / European Journal of Radiology 82 (2013) 862869

[9] Hoetjes NJ, Velden FHP, Hoekstra OS, et al. Partial volume correction strategies
for quantitative FDG PET in oncology. European Journal of Nuclear Medicine
and Molecular Imaging 2010;37(9):167987.
[10] Tong S, Alessio A, Kinahan P. Image reconstruction for PET/CT scanners: past
achievements and future challenges. Imaging in Medicine 2010;2(5):52945.
[11] Panin V, Kehren F, Michel C, Casey M. Fully 3-D PET reconstruction with system
matrix derived from point source measurements. IEEE Transactions on Medical
Imaging 2006;25(7):90721.
[12] Jakoby B, Bercier Y, Watson C, Bendriem B, Townsend D. Performance characteristics of a new LSO PET/CT scanner with extended axial eld-of-view and
PSF reconstruction. IEEE Transactions on Nuclear Science 2009;56(3):6339.
[13] Alessio A, Stearns C, Tong S, et al. Application and evaluation of a measured
spatially variant system model for PET image reconstruction. IEEE Transactions
on Medical Imaging 2010;29(3):93849.
[14] Bettinardi V, Presotto L, Rapisarda E, Picchio M, Gianolli L, Gilardi M. Physical performance of the new hybrid PET/CT Discovery-690. Medical Physics
2012;38(10):5394411.
[15] Alessio AM, Kinahan PE. Application of a spatially variant system model for 3D whole-body PET image reconstruction. Proceedings of the IEEE International
Symposium on Biomedical Imaging 2008;2008(May):13158.
[16] Kadrmas D, Casey M, Black N, Hamill J, Panin V, Conti M. Experimental comparison of lesion detectability for four fully-3D PET reconstruction schemes. IEEE
Transactions on Medical Imaging 2009;28(4):52334.
[17] Varrone A, Sjholm N, Eriksson L, Gulys B, Halldin C, Farde L. Advancement
in PET quantication using 3D-OP-OSEM point spread function reconstruction
with the HRRT. European Journal of Nuclear Medicine and Molecular Imaging
2009;36(10):163950.
[18] Mourik J, Lubberink M, Velden Fv, et al. In vivo validation of reconstructionbased resolution recovery for human brain studies. Journal of Cerebral Blood
Flow and Metabolism 2010;30(2):38190.

869

[19] Doll J, Henze M, Bublitz O, et al. High resolution reconstruction of PET images
using the iterative OSEM algorithm. Nuklear Medizin 2004;43(3):728.
[20] Rapisarda E, Bettinardi V, Thielemans K, Gilardi MC. Image-based point spread
function implementation in a fully 3D OSEM reconstruction algorithm for PET.
Physics in Medicine and Biology 2010;55(14):413151.
[21] Boellaard R, ODoherty M, Weber W, et al. FDG PET and PET/CT: EANM procedure guidelines for tumour PET imaging: version 1.0. Europena Journal of
Nuclear Medicine and Molecular Imaging 2010;37(1):181200.
[22] Kinahan PE, Townsend DW, Beyer T, Sashin D. Attenuation correction for a
combined 3D PET/CT scanner. Medical Physics 1998;25(10):204653.
[23] Hudson HM, Larkin RS. Accelerated image reconstruction using ordered subsets of projection data. IEEE Transactions on Medical Imaging 1994;13(4):
6019.
[24] Young H, Baum R, Cremerius U, et al. Measurement of clinical and subclinical tumour response using [18 F]-uorodeoxyglucose and positron emission
tomography: review and 1999 EORTC recommendations. European Organization for Research and Treatment of Cancer (EORTC) PET Study Group. European
Journal of Cancer 1999;35(13):177382.
[25] Lasnon C, Hicks RJ, Beauregard JM, et al. Impact of point spread function reconstruction on thoracic lymph node staging with 18F-FDG PET/CT in non-small
cell lung cancer. Clinical Nuclear Medicine 2012;53(37):9716.
[26] Boellaard R. Mutatis mutandis: harmonize the standard! Journal of Nuclear
Medicine 2012;53(1):13.
[27] Larson SM, Erdi Y, Akhurst T, et al. Tumor treatment response based on visual
and quantitative changes in global tumor glycolysis using PET-FDG imaging:
the visual response score and the change in total lesion glycolysis. Clinical
Positron Imaging 1999;2(3):15971.
[28] Wahl R, Jacene H, Kasamon Y, Lodge M. From RECIST to PERCIST: evolving
considerations for PET response criteria in solid tumors. Journal of Nuclear
Medicine 2009;50(Suppl. 1):122S50S.