Statistical parametric maps of 18F-FDG PET and 3-D autoradiography in the rat brain: a cross-validation study

Abstract

PurposeAlthough specific positron emission tomography (PET) scanners have been developed for small animals, spatial resolution remains
one of the most critical technical limitations, particularly in the evaluation of the rodent brain. The purpose of the present
study was to examine the reliability of voxel-based statistical analysis (Statistical Parametric Mapping, SPM) applied to
18F-fluorodeoxyglucose (FDG) PET images of the rat brain, acquired on a small animal PET not specifically designed for rodents.
The gold standard for the validation of the PET results was the autoradiography of the same animals acquired under the same
physiological conditions, reconstructed as a 3-D volume and analysed using SPM.

MethodsEleven rats were studied under two different conditions: conscious or under inhalatory anaesthesia during 18F-FDG uptake. All animals were studied in vivo under both conditions in a dedicated small animal Philips MOSAIC PET scanner
and magnetic resonance images were obtained for subsequent spatial processing. Then, rats were randomly assigned to a conscious
or anaesthetized group for postmortem autoradiography, and slices from each animal were aligned and stacked to create a 3-D
autoradiographic volume. Finally, differences in 18F-FDG uptake between conscious and anaesthetized states were assessed from PET and autoradiography data by SPM analysis and
results were compared.

ResultsSPM results of PET and 3-D autoradiography are in good agreement and led to the detection of consistent cortical differences
between the conscious and anaesthetized groups, particularly in the bilateral somatosensory cortices. However, SPM analysis
of 3-D autoradiography also highlighted differences in the thalamus that were not detected with PET.

ConclusionThis study demonstrates that any difference detected with SPM analysis of MOSAIC PET images of rat brain is detected also
by the gold standard autoradiographic technique, confirming that this methodology provides reliable results, although partial
volume effects might make it difficult to detect slight differences in small regions.

KeywordsRat brain–Positron emission tomography–Three-dimensional autoradiography–SPM

Full text

Page 1

Statistical parametric maps of 18F-FDG PET and 3D autoradiography in the
rat brain: a cross-validation study.

Prieto Elena1, Collantes María2, Delgado Mercedes3, Juri Carlos4,5,6, García-
García Luis3, Molinet Francisco2, Fernández-Valle María Encarnación7, Pozo
Miguel Ángel3, Gago Belén4, Martí-Climent Josep1, Obeso José A. 4,5, Peñuelas
Iván1,2

1 Nuclear Medicine Department, Clínica Universidad de Navarra, Pamplona,
Spain

2 Small Animal Imaging Research Unit, Center for Applied Medical Research
(CIMA) and Clínica Universidad de Navarra, Pamplona, Spain

3 Brain Mapping Unit, Universidad Complutense de Madrid, Madrid, Spain

4 Movement Disorders Group, Neurosciences Division, Center for Applied
Medical Research (CIMA), and CIBERNED, University of Navarra, Pamplona,
Spain

5 Department of Neurology and Neurosurgery, Clínica Universidad de Navarra
and Medical School, Pamplona, Spain

6 Department of Neurology. Pontificia Universidad Católica de Chile. Santiago,
Chile.

7 MRI Research Center, Universidad Complutense de Madrid, Madrid, Spain

Corresponding author:
Iván Peñuelas
Nuclear Medicine Department
Clínica Universidad de Navarra
Av. Pío XII 36, 31008, Pamplona, Spain
Tel: +34 948 255400
Fax: +34 948 296500
E-mail: ipenuelas@unav.es


KEYWORDS
Rat brain, positron emission tomography, 3D-autoradiography, SPM

1

Page 2

ABSTRACT
Although specific PET scanners have been developed for small animals, spatial
resolution remains one of the most critical technical limitations, particularly in
the evaluation of the rodent brain. The purpose of the present study was to
examine the reliability of voxel-based statistical analysis (SPM) applied to 18F-
fluorodeoxyglucose (18F-FDG) PET images of the rat brain, acquired on a small
animal PET not specifically designed for rodents. The gold standard for the
validation of the PET results was the autoradiography of the same animals
acquired under the same physiologic conditions, reconstructed as a 3D volume
and analyzed using SPM. Methods: Eleven rats were studied under two
different conditions: conscious or under inhalatory anesthesia during 18F-FDG
uptake. All animals were studied in vivo under both conditions in a dedicated
small animal Philips MOSAIC PET scanner and magnetic resonance images
were obtained for subsequent spatial processing. Then, rats were randomly
assigned to a conscious or anesthetized group for post-mortem
autoradiography and slices from each animal were aligned and stacked to
create a 3D autoradiographic volume. Finally, differences in 18F-FDG uptake
between conscious and anesthetized states were assessed from PET and
autoradiography data by SPM analysis and results were compared. Results:
SPM results of PET and 3D autoradiography are in good agreement and led to
the detection of consistent cortical differences between the conscious and
anesthetized groups, particularly in the bilateral somatosensory cortices.
However, SPM analysis of 3D autoradiography also highlighted differences in
the thalamus that were not detected with PET. Conclusion: This study
demonstrates that any difference detected with SPM analysis of MOSAIC PET

2

Page 3

images of rat brain is detected also by the gold standard autoradiographic
technique, confirming that this methodology provides reliable results, although
partial volume effects might make it difficult to detect slight differences in small
regions.

3

Page 4

1. INTRODUCTION

Positron emission tomography (PET) for in vivo imaging of metabolic processes
is a powerful tool to study the etiopathogenesis and progression of neurological
diseases and to test the efficacy of new therapeutic agents [1]. Recently,
specific PET scanners have been developed for small animals which provide
smaller fields of view and higher spatial resolution than clinical tomographs.
This allows longitudinal functional neuroimaging studies to be undertaken in
animal models of various neurological conditions, including Parkinson’s disease
[2, 3], Alzheimer’s disease [4], and epilepsy [5]. Moreover, due to the fact that
the same imaging technique can be used for both animal and human studies,
methodologies and results obtained from animal models may be easily
translated to the clinical setting.

In PET studies of the human brain, voxel-based analysis is the most extensively
used methodology for quantification. Statistical parametric mapping (SPM) is
the most popular software for this purpose [6]. This software applies a general
linear model to each voxel to perform statistical comparisons between images
of different subjects. The result is a statistical image that can be used to
spatially localize significant information. It should be noticed that use of SPM
requires that images from different individuals be spatially normalized, that is,
mapped into a common stereotaxic space by using a template image as
reference. In comparison to traditional analysis using regions of interest (ROIs),

4

Page 5

SPM is faster and includes the whole brain in the analysis, without any
preconception about the structures involved in the investigated pathology.
Recently, SPM analysis has also been applied to the study of animal models of
different neurological diseases [7, 8] and various templates have been
developed for spatial normalization of rodent brains [9, 10].

Although the resolution of small animal-dedicated tomographs has significantly
improved, high-resolution autoradiography still represents a key reference tool
for functional neuroimaging in small animal research. Autoradiography is a
widely- available low-cost technique with a spatial resolution approximately one
order of magnitude better than even the best small animal PET devices (100-
200 micrometers
vs 1-2 millimeters respectively) [11]. However,
autoradiography requires the sacrifice of the animal, thus negating its use in
longitudinal studies. These differences between in vivo small animal PET and
autoradiography are widely known, and several studies have compared the
results of these two techniques in detail [12-16]. Autoradiography is inherently a
two-dimensional (2D) sampling method in which slices are obtained
independently without correspondence between neighboring slices. Therefore,
quantification is classically obtained by the labor-intensive procedure of
manually drawing ROIs plane by plane. In order to avoid this drawback, several
authors have described different methodologies for voxel-based analysis, all of
them based on the creation of a three-dimensional (3D) volume stack of aligned
autoradiographic slices [17-19]. Moreover, these methodologies have been
successfully applied in the study of different animal models of neurological
pathologies [20, 21].

5

Page 6

In this study, the primary objective was to validate statistical voxel-based
analysis of PET rat brain studies performed in our Philips MOSAIC small animal
PET tomograph, using high-resolution 3D autoradiography as a reference. To
this end, an automated procedure for voxel-based analysis of PET and
autoradiography was designed and applied for comparison of
18F-
fluorodeoxyglucose (18F-FDG) metabolic studies of the rat brain. For
comparative purposes, animals were studied under two different conditions
during tracer uptake: conscious and under inhalatory anesthesia (isoflurane),
protocols that have been demonstrated to result in clear differences [22]. To the
best of our knowledge this is the first study to compare SPM results between 3D
autoradiographic volumes and PET images of the rat brain.

2. MATERIALS AND METHODS

2.1. Animals:
Eleven male healthy Sprague–Dawley rats (Harlan Ibérica, Spain) weighing 253
±14.5 g were individually housed under light-controlled (12:12 light/dark cycle,
with lights on at 08:00 h) and temperature-controlled (22 ± 1 ºC) conditions.
Animals had ad libitum access to food and tap water. Rats were deprived of
food for 8 h before both PET and autoradiography studies. The research
protocol was approved by the local Animal Ethics Committee (Universidad de
Navarra Institutional Committee on Care and Use of Laboratory Animals) and
was designed according to European Ethics Committee guidelines (decree

6

Page 7

86/609/EEC). All animals were studied with PET, magnetic resonance imaging
(MRI) and autoradiography techniques.

2.2. PET Studies
18F-FDG PET imaging was performed at Clínica Universidad de Navarra in a
dedicated small animal Philips MOSAIC tomograph (Cleveland, Ohio, USA).
This system is based on 14456 GSO crystals with dimensions of 2 x 2 x 10 mm,
arranged in 52 rings of 278 crystals each [23]. The crystals are read out by an
hexagonal array of 288 photomultiplier tubes. This gantry design leads to a
transverse field of view (FOV) 12.8 cm in diameter and with an axial extent of
12.0 cm, designed to image rodents or larger animals such as primates [24].
The scanner operates exclusively in 3D mode. A detailed description of this
system can be found elsewhere [25]. The radial spatial resolution of this system
is 2.7 mm full width at half-maximum (FWHM) in the radial direction and 3.4 mm
FWHM in the axial direction at the center of the FOV [23].

PET studies were conducted for two conditions in all rats (n=11): conscious or
under inhalatory anesthesia (2% isoflurane in 100% O2 gas) during the 18F-FDG
uptake period. For each animal, PET in conscious state was performed first and
all the studies were completed in a one week interval. In both conditions,
animals were anesthetized using isoflurane at the moment of the tracer injection
and during the PET studies. Thirty minutes after radiotracer administration via
the tail vein (20.2 ± 0.3 MBq in 0.2 ml of saline solution), animals were placed
prone on the PET scanner bed with their head positioned in the center of the
FOV to perform a static acquisition of 15 minutes. Images were subsequently

7

Page 8

reconstructed using an iterative 3D maximum likelihood algorithm (3D Ramla),
with 2 iterations and a relaxation parameter of 0.024. Corrections for dead time,
decay, random coincidences and scattering were applied. Images were
reconstructed on a 128 x 128 x 120 matrix, where the voxel size equals 1 x 1 x
1 mm. PET images were normalized for the injected dose of 18F-FDG and the
animal's body weight, resulting in parametric images representing SUV
(Standard Uptake Value).

2.3. MRI Acquisition
MRI studies of the same population of rats were performed at Complutense
University of Madrid (UCM) fifteen days after PET images. MRI scans were
acquired in a BIOSPEC BMT 47/40 (Bruker, Ettlingen, Germany) scanner
operating at 4.7 Teslas, and equipped with a 12 cm actively shielded gradient
system.

Rats were anesthetized with the same mixture of oxygen and isoflurane, after
which they were placed in a prone position inside a 7 cm birdcage
radiofrequency probehead with their head maintained in a fixed position. A
respiration sensor was used to control the animals during the scanning
procedure. Global shimming was performed, followed by three fast spin-echo
scout images in the axial, sagittal and coronal directions to localize the brain.
The acquisition parameters for these acquisitions were: Repetition Time (TR) =
2100 ms, effective Echo Time (TE) = 60 ms, FOV = 3.6 x 3.6 cm2, and matrix
size = 256 x 128. The acquired data were zero-filled to obtain images of 256 x
256 pixels. The acquisition time for each experiment was 33 s. A 3D fast spin-

8

Page 9

echo study was then performed using the following acquisition parameters: TR
= 1600 ms, effective TE = 80 ms, FOV = 3.5 x 3.5 x 3.5 cm3, and matrix size =
128 x 128 x 64. The reconstructed matrix size was 128 x 128 x 128 and the
pixel size was 0.27 x 0.27 x 0.27. The total acquisition time for this experiment
was 14 minutes.

2.4. Autoradiography
The regional glucose metabolism of the same animals was evaluated ex vivo by
autoradiography at the Brain Mapping Unit of Complutense University of
Madrid. Autoradiographic measurements were performed after MRI and were
completed in a one week interval. 18F-FDG was injected via the tail vein (54.5 
2.9 MBq in 0.2 ml of saline solution). The animals were then divided into two
groups based on whether they were conscious (n=5) or under isoflurane
anesthesia (2% in 100% O2; n=6) during the 30 min tracer uptake period. At the
end of the uptake period the animals were sacrificed and their brains were
quickly and carefully removed and frozen by immersion in cold isopentane in
dry ice. Serial coronal brain sections (40 µm thickness) from the olfactory bulb
to the level of the cerebellum were cut at -20ºC using a cryostat (Leica CM1850,
Germany). One out of every five sections was collected, yielding approximately
100 sections per animal, at a separation of 200 µm. The slices were thaw-
mounted onto Superfrost Plus® microscope slides (Menzel-Gläser,
Braunschweig, Germany), rapidly heat-dried, and exposed to an
autoradiographic film (Agfa Curix RP2 Plus, Mertsel, Belgium) for one hour. The
films were then manually developed and left to dry in a warm air stream. All the

9

Page 10

autoradiographically labeled brain slices were digitally captured at a resolution
of 1300 x 1030 pixels (Leica DC300F, Germany).

2.5. PET analysis
For whole brain analysis, images from the different rats were spatially
normalized into a standard stereotaxic space using the Paxinos and Watson
[26] coordinate system (Figure 1). Although a T2-weighted MRI template has
previously been published [10], neither PET nor autoradiography templates
were available. Hence, the normalization procedure required an initial step
registering each image to the corresponding MRI scan in order to use the MRI
rat template.

Taking into account the spatial resolution of our PET images, automatic
registration between the PET and MRI scans was not feasible. Therefore, a
manual registration process employing PMOD software (version 3.0; PMOD
Technologies Ltd., Adliswil, Switzerland) was used, based on MRI including the
skull and scalp and using anatomical structures such as the Harderian glands
as reference points. Each MRI scan was normalized to Paxinos space (SPM,
Wellcome Department of Cognitive Neurology, Institute of Neurology, London,
UK) and the transformation matrix was applied to each PET image and saved
for future application. Prior to statistical analysis, a brain mask obtained from
each MRI procedure was applied over the PET scan to exclude extra-cerebral
areas. Each of the scans was also individually smoothed with a Gaussian kernel
to reduce the impact of misregistration into template space and to improve the
signal to noise ratio. The smoothing was performed with a 0.6 mm FWHM filter.

10

Page 11

Differences in 18F-FDG uptake between the conscious and anesthetized state
were assessed using SPM8 software, including the 11 animals in the two
acquisition conditions. A two-sample paired t-test was used to account for the
fact that the both conditions were obtained from the same animal. To ensure
that the analysis only included voxels mapping cerebral tissue, a default
threshold of 80% of the mean uptake inside the brain was selected. Global
uptake differences between brain scans were adjusted using the “proportional
scaling” SPM option.
A significance level threshold of 0.005 (uncorrected for multiple comparisons)
and a minimum cluster size of 200 voxels were selected. The size of the
clusters exceeding the threshold and their corrected significance were
evaluated and anatomically located using a ROI template [27] in Paxinos
stereotaxic space.

2.6. Creation of a 3D Volume of Autoradiography and SPM analysis
The autoradiographic brain images were pre-processed with ImageJ v 1.37
software (Wayne Rasband, National Institutes of Health, Bethesda, MD, USA).
First, a rectangular bounding box (700 x 500 pixels) was defined around each
section (Figure 2.A). The image gray scale was then inverted to conform to the
conventions of human functional-imaging data (Figure 2.B) and the background
was subtracted using a rolling ball algorithm (Figure 2.C). A volume of interest
was also drawn as an isocontour over the autoradiographic images to generate
a mask of cerebral areas and to remove the possible contribution of adjacent
sections (Figure 2. D)

11

Page 12

In order to create a 3D image, all slices from each animal were imported as an
image stack using PMOD (Figure 2.E). Adjacent sections were aligned using
the PMOD Image Fusion tool. Due to the fact that the central slices provide the
best quality, an appropriately positioned artifact-free central slice was selected
as reference. The alignment of the whole volume was then performed
sequentially between adjacent slices. To preserve the shape of the sections, a
rigid transformation was chosen and the squared difference sum was used as
the dissimilarity metric for optimization. After the automatic registration and
visual inspection, a slight manual alignment was performed on some slices if
necessary. Thus, an autoradiographic 3D volume was obtained for each animal
(Figure 2.F). The fidelity of the alignment and 3D reconstruction process was
assessed by visual inspection of internal structures viewed in different
orthogonal sections, as well as by inspection of the smoothness of the cortex
after rendering the surface.

The resultant 3D volume then had to be spatially normalized to a standard
space (Figure 1). As a specific template of autoradiography does not exist, an
MRI template was used. For each animal, a mask was applied over its
corresponding MRI scan to remove the skull and scalp. Registration between
the skull-stripped MRI scan and the 3D autoradiography image was performed
automatically, followed by visual inspection and manual adjustment only if
necessary. The previously calculated transformation matrix of MRI
normalization was then applied to the autoradiographic 3D volume of the same
animal (SPM8).

12

Page 13

As previously done with the PET data, 3D autoradiographic volumes were
smoothed with a Gaussian filter with a FWHM of 1 mm. As each animal is only
acquired in one condition, a two sample t-test was performed in this case.
Model parameters included global normalization, with proportional scaling and
relative threshold masking at 80% of mean voxel value. Parameters were
estimated and contrasts were generated to create statistical parametric maps of
t-values. The statistical threshold was set at p<0.005 with an extent threshold of
100 contiguous voxels. Activated clusters were evaluated and anatomically
located using a ROI template [27].

3. RESULTS

3.1. PET studies and 3D Autoradiography
PET scans from a representative rat under conscious and anesthetized
conditions are shown in Figure 3.B and 3.C respectively, together with an MRI
image (Figure 3.A). Widespread glucose hypometabolism can be seen when
the animal was anesthetized versus conscious during the period of 18F-FDG
uptake.

Autoradiographic images were acquired, processed and successfully 3D
reconstructed using the described procedure. Visual inspection of axial slices
showed continuity of the borders of internal structures, such as the thalamus
and basal ganglia. Each 3D-rendered surface also exhibited a consistent global

13

Page 14

shape of the whole brain volume (Figure 4). The procedure was therefore well
designed and produced geometrically consistent volumes.

Figure 5 illustrates representative autoradiographic coronal brain slices of
conscious (Figure 5.B) and isoflurane-anesthetized animals (Figure 5.C). Under
both conditions, glucose uptake was heterogeneously distributed according to
regional differences in metabolic rate. As also noted with PET, a widespread
decrease in glucose metabolism during anesthesia was clearly observed in the
cortex, with hypometabolism also being apparent in subcortical structures.

3.2. Statistical analysis of PET studies
The areas displaying significantly decreased metabolism in the anesthetized
versus the conscious condition are shown in Figure 6. A statistical t map was
overlaid on the canonical MRI [10] coronal slices. Volume of interest contours of
some cortical areas and the thalamus were superimposed on these images to
show the location of the activation, resulting in the illustrated 3D volume-
rendered image of the statistical t map.

An extensive cortical area showed significant differences with SPM analysis
between the two acquisition conditions. SPM results are summarized in Table 1.
One widespread cluster was detected, covering bilateral somatosensory cortex
and other functional areas such as the auditory and visual cortices.

3.3. Statistical analysis of 3D Autoradiography

14

Page 15

The results of the SPM analysis of the generated 3D autoradiography are
shown in Figure 7. For visualization of the t-score map, significant voxels were
projected onto the 3D-rendered high-resolution brain MRI scan, thus allowing
anatomic identification. Based on a height threshold of p<0.005, there was
significant deactivation in extensive cortical areas and the thalamus of
anesthetized rats. Table 1 lists the activated clusters where significant
differences were observed, together with the coordinates and significance level
(Z and t scores and FWE-corrected p) of the maximum peak and the areas
covered by each cluster.

4. DISCUSSION
In the present study, we explored the feasibility of using SPM for the analysis of
18F-FDG PET of the rat brain acquired in our MOSAIC Philips small animal PET.
Voxel-based analysis is a widely applied technique in human studies, and holds
considerable promise for research using rodent models. Although several
studies have performed SPM analysis in rats [28, 29], none of them has
questioned the validity of the results taking into account the limitation in spatial
resolution.

As our goal was to undertake a feasibility study, we chose the most widely-
used radiotracer,
18F-FDG, which measures glucose metabolism. The
comparison selected was PET studies conducted in animals that were either
conscious or maintained under isoflurane anaesthesia during the tracer uptake
period, reflecting commonly encountered experimental protocols that have been
reported to produce significantly different results. This issue has been

15