Reliability of task-specific neuronal
activation assessed with functional
PET, ASL and BOLD imaging
, Godber M Godbersen
, Verena Pichler
, Sebastian Klug
, Manfred Kl€
, Wolfgang Wadsak
, Marcus Hacker
, Rupert Lanzenberger
and Andreas Hahn
Mapping the neuronal response during cognitive processing is of crucial importance to gain new insights into human
brain function. BOLD imaging and ASL are established MRI methods in this endeavor. Recently, the novel approach of
functional PET (fPET) was introduced, enabling absolute quantification of glucose metabolism at rest and during task
execution in a single measurement. Here, we report test-retest reliability of fPET in direct comparison to BOLD imaging
and ASL. Twenty healthy subjects underwent two PET/MRI measurements, providing estimates of glucose metabolism,
cerebral blood flow (CBF) and blood oxygenation. A cognitive task was employed with different levels of difficulty
requiring visual-motor coordination. Task-specific neuronal activation was robustly detected with all three imaging
approaches. The highest reliability was obtained for glucose metabolism at rest. Although this dropped during task
performance it was still comparable to that of CBF. In contrast, BOLD imaging yielded high performance only for
qualitative spatial overlap of task effects but not for quantitative comparison. Hence, the combined assessment of fPET
and ASL offers reliable and simultaneous absolute quantification of glucose metabolism and CBF at rest and task.
fMRI, functional PET (fPET), PET/MRI, task-specific activation, test-retest reliability
Received 29 September 2020; Revised 22 April 2021; Accepted 3 May 2021
Human brain function has been subject to research for
centuries and is yet not fully understood due to its vast
complexity. Thus, characterizing the neuronal response
during cognitive processing is of pivotal importance
and can be achieved with various approaches.
The most common imaging method to investigate
task-induced changes in the brain is functional magnet-
ic resonance imaging (fMRI) based on the blood
oxygen level-dependent (BOLD) signal. Advantages
are high sensitivity to changes in blood oxygenation,
readily accessible MRI sequences, established tasks and
high temporal and spatial resolution. However, the
BOLD signal is a composite of changes in cerebral
blood flow (CBF), cerebral blood volume (CBV) and
and consequently, only an indirect,
non-specific proxy for neuronal activation.
Other drawbacks are the instability of the BOLD
signal for longer task durations, spurious effects such
as scanner drifts
heart rate and respiratory influences,
These are non-trivial issues and to a certain
Department of Psychiatry and Psychotherapy, Medical University of
Vienna, Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Division
of Nuclear Medicine, Medical University of Vienna, Vienna, Austria
Department of Pharmaceutical Sciences, Division of Pharmaceutical
Chemistry, University of Vienna, Vienna, Austria
Center for Biomarker Research in Medicine (CBmed), Graz, Austria
Andreas Hahn, Department of Psychiatry and Psychotherapy, Medical
University of Vienna, Waehringer Guertel 18-20, 1090 Vienna, Austria.
Journal of Cerebral Blood Flow &
!The Author(s) 2021
Article reuse guidelines:
extent reasons that absolute quantification of BOLD
signal changes require substantial effort.
Another fMRI-based approach is arterial spin label-
ing (ASL) which enables quantification of CBF. Here,
arterial blood is magnetically labeled to use it as endog-
enous contrast agent. Perfusion-weighted images are
then computed from a labeled and an unlabeled
image, enabling estimation of CBF.
quantification represents a major advantage of ASL,
facilitating a comparison between rest and task-
specific flow changes. However, the requirement to
acquire images in pairs reduces the temporal resolution
and also the spatial resolution is inferior compared to
Task-induced changes are also reflected in altered
glucose metabolism. Hence, positron emission tomog-
raphy (PET) using the glucose analogue [
F]FDG is a
suitable tool to map neuronal activation. Indeed, task-
specific changes in glucose metabolism were already
assessed before BOLD imaging and ASL were intro-
A crucial drawback of this method is that
measurements have to be conducted in at least two
separate sessions, mostly carried out even on separate
days, one at rest and another one during task execu-
tion. Changes in daily performance or resting activity
between these measurements will influence the results
and subjects are exposed to ionizing radiation twice.
Recently, these issues were resolved with the novel
approach of functional PET (fPET). The method ena-
bles the quantification of resting and task-specific glu-
cose metabolism within a single measurement.
protocol was further optimized by the administration
F]FDG as bolus plus constant infusion.
increases the signal-to-noise ratio and the amount of
freely available radiotracer to track even subtle task-
related changes with a temporal resolution of minutes
(instead of hours or days as previously required).
Disadvantages include the necessity for arterial cannu-
lation to enable absolute quantification and the radia-
tion exposure of participants.
In sum, task-specific neuronal activation is reflected
in subtle changes from the resting activity in glucose
metabolism, blood flow and oxygenation. The above-
mentioned technical challenges and individual physio-
logical effects might limit the detection of these
changes. Hence, a valuable imaging approach is not
only sensitive to task-induced changes but also has to
provide high reliability during rest and task conditions.
This is of pivotal importance for scientific and clinical
applications to ensure that subtle effects can be robust-
ly detected despite the variance inherent to repeated
measurements. As such applications commonly aim
to assess complex cognitive functions, it is essential to
know the reliability for correspondingly complex tasks,
since reliability of simpler tasks may not be extrapolat-
While ASL and BOLD imaging underwent optimi-
zation for decades already, fPET is still in its infancy
and the applicability of this novel imaging approach in
longitudinal studies was not yet assessed. Therefore, we
conducted a test-retest study and employed a challeng-
ing task with varying levels of cognitive load, given by
the video game TetrisV
R. The introduction of fully-
integrated PET/MRI scanners enabled the simulta-
neous acquisition of fPET, ASL and BOLD imaging.
Thus, the test-retest reliability between these imaging
modalities can be most directly compared in a single
scan session. We aimed to assess i) the capability of
fPET to track task-specific changes with a high reliabil-
ity between measurements and ii) its performance in
comparison to that of the well-established modalities
of ASL and BOLD imaging. For this purpose, we
investigated common parameters for each modality
(influx constant K
and cerebral metabolic rate of glu-
cose (CMRGlu) for fPET, CBF for ASL and parame-
ter estimates (i.e., beta values obtained from the general
linear model) for BOLD imaging) and assessed their
test-retest reliability with frequently used metrics
(intraclass correlation, coefficient of variation, Dice
Material and methods
For this study, 53 healthy subjects were initially
recruited, data from 40 were used and test-retest reli-
ability was assessed for 20. Among the 13 drop out
subjects, 7 discontinued voluntarily after the first mea-
surement, for 1 subject arterial blood sampling failed,
for 2 subjects ASL could not be acquired during the
first measurement because of temporary technical chal-
lenges with the PET/MRI scanner (only affecting
ASL), for 3 subjects only MRI was carried out due to
arterial puncture and/or radiotracer synthesis failure.
Cross-sectional data of a subsample was already
included in a previous analysis.
All subjects under-
went a routine medical investigation at a screening
visit including electrocardiography, blood tests, neuro-
logical and physiological tests, and a urine drug test.
Psychiatric disorders were ruled out with the Structural
Clinical Interview DSM-IV conducted by an experi-
enced psychiatrist. Female participants additionally
underwent a pregnancy test at the screening visit and
before each PET/MRI measurement. Participants had
to fast for at least four hours prior to the scan, includ-
ing no consumption of sweetened beverages and caf-
Exclusion criteria were weight above 100 kg for
reasons of radiation protection, current or previous
2Journal of Cerebral Blood Flow & Metabolism 0(0)
neurological, physiological or psychiatric disorders,
current breastfeeding or pregnancy, left-handedness,
substance abuse, MRI contraindications, participation
in a study including ionizing radiation exposure (past
10 years) and regularly playing TetrisV
Ror similar games
(including mobile phone games) within the last 3 years.
After detailed explanation of the study protocol, all
subjects gave written informed consent. All subjects
were insured and reimbursed for their participation.
The study was approved by the Ethics Committee of
the Medical University of Vienna (ethics number:
1479/2015) and all procedures were carried out in accor-
dance with the Declaration of Helsinki. The study was
registered at ClinicalTrials.gov (ID: NCT03485066).
Forty healthy subjects (20 male, mean age sd ¼
23.0 3.4 years, all right-handed) underwent a single
PET/MRI scan on a fully-integrated PET/MRI
system (Siemens Biograph mMR, Siemens
Healthineers, Germany). To assess the test-retest reli-
ability of functional imaging parameters, a subgroup of
20 subjects (10 male, 23.1 3.1 years) also underwent a
second measurement (4.2 0.7 weeks apart). The scans
of the remaining 20 subjects (10 male, 23.0 3.7 years)
solely served for an independent region of interest def-
inition. Measurements were initiated with the acquisi-
tion of a structural T1-weighted image followed by an
ASL sequence during rest. Subsequently, [
was administered as bolus followed by constant infu-
sion while a complex cognitive paradigm was
employed. After 8 minutes of rest, 4 task conditions
were carried out with varying difficulty (6 min each, 2
easy, 2 hard, pseudo-randomized order). Each task was
followed by a rest condition (5 min) where subjects
were instructed to look at a black crosshair on grey
background and to let their thoughts wander.
Simultaneously with the task blocks, ASL was acquired
during one easy and one hard condition. During the
remaining task blocks, BOLD imaging was acquired
for functional connectivity as described elsewhere.
Immediately after fPET and ASL acquisitions, the
same task was carried out in a conventional block
design and BOLD data was acquired (4 easy, 4 hard
and additionally, 4 control blocks, 30 s each, pseudo-
randomized order), again separated by rest blocks (10 s
each). Although the duration and number of task
blocks differed across imaging modalities, we aimed
to employ an optimal acquisition protocol for each
parameter. The entire study design is depicted in
An adapted version of the video game TetrisV
ried out, representing a complex visuo-spatial motor
task, which combines mental rotation, spatial planning
and hand-eye coordination, thus, targeting various
higher-order brain regions. The task comprised two
levels of difficulty (easy and hard) to induce different
cognitive loads (Figure 1(b)). The aim of the task was
to complete full horizontal lines by moving and rotat-
ing bricks falling from the top of the screen. Subjects
played only with their right hand on an MR-
compatible button box (index/small finger: move left/
right, middle finger: rotate, ring finger: move down).
The two levels varied regarding the speed at which the
bricks were falling and the initial number of incomplete
lines at the beginning (Figure 1(b)). During BOLD
acquisition, a control level was introduced where
bricks had to be lead through a chimney sufficiently
Figure 1. Study design and cognitive task. (a) Measurements were initiated with a structural T1-weighted image (grey, 8 mins) and
ASL at rest (green, 6 mins). Thereafter, fPETwas acquired (blue, 52 mins) and an adapted version of the video game TetrisV
four times with varying cognitive load (6 mins, 2 easy, 2 hard, pseudo-randomized order) separated by resting periods. Simultaneously,
ASL was acquired during one easy and one hard condition (green). BOLD data for functional connectivity was acquired during the
second easy and hard task blocks, but these were not used in the current study. Immediately after fPET and ASL, BOLD data was
acquired with the same task and an additional control condition (red, 12 task blocks, 30 s each, 10 s rest). (b) The task consisted of the
conditions easy, hard and control, whereas the latter was only carried out during BOLD imaging. Easy and hard conditions differed by
the speed of the falling bricks and the initial number of incomplete lines. In the control condition bricks had to be guided through a
chimney and were automatically removed at the bottom. This figure was adapted from Hahn et al. (2020) under CC-BY license.
Rischka et al. 3
wide that no rotation was necessary (Figure 1(b)). For
the control level, bricks were removed at the bottom.
Thus, no completion of lines was possible. In general,
subjects were instructed to gain as many points as pos-
sible and were explained that completion of several
lines at once scores more points. Right before the
start of the measurement, each condition was played
once (30 s each) in the scanner to familiarize the par-
ticipants with the procedure and the controls.
PET and MRI data acquisition
The radioactive glucose analogue [
F]FDG was fresh-
ly synthesized every morning using FASTlab FDG cas-
settes with phosphate buffer formulation
FASTlab platform (GE Healthcare). The substance
was administered via a cubital vein as bolus for
1 minute followed by constant infusion for 51 minutes
with an infusion pump (Syramed mSP6000, Arcomed,
Switzerland, dosage: 5.1 MBq/kg, bolus speed: 816 ml/
h, infusion speed: 42.8 ml/h, bolus-infusion ratio of
activity: 20:80%), which was placed in an MR-shield.
PET data was acquired in list-mode, enabling the ret-
rospective definition of frame lengths during
A T1-weighted structural image was acquired with a
magnetization prepared rapid gradient echo
(MPRAGE) sequence prior to radiotracer administra-
tion (TE/TR ¼4.21/2200 ms, voxel size ¼11
1.1 mm, matrix size ¼240 256, slices ¼160, flip
angle ¼9,TI¼900 ms, 7.72 min). The image was
used to rule out severe brain disorders, for attenuation
correction and normalization to MNI space.
A 2 D pseudo-continuous arterial spin labeling
(pcASL) sequence was recorded at rest prior to radio-
tracer administration and during task conditions simul-
taneously with PET acquisition (TE/TR ¼12/4060 ms,
post-labeling delay ¼1800 ms, labeling dura-
tion ¼1.5 sec, readout duration per slice ¼35 ms,
voxel size ¼3.44 3.44 5mmþ1 mm gap, matrix
size ¼64 64, slices ¼20, flip angle ¼90, 6 min). The
labeling plane was 9 cm inferior of the center of the
field of view, which was placed on the anterior-
posterior commissure line. No background suppression
or further optimizations were applied.
BOLD data was acquired with an echo-planar
imaging (EPI) sequence following PET acquisition
(TE/TR ¼30/2000 ms, voxel size ¼2.5 2.5 2.5 mm
þ0.825 mm gap, matrix size ¼80 80, slices ¼34, flip
angle ¼90, 8.17 min).
Prior to each PET/MRI measurement, the individual
fasting blood glucose level was measured (Glu
triplicate measurement). Arterial blood samples were
drawn from a radial artery throughout the radiotracer
administration (time points: 3, 4, 5, 14, 25, 36 and
47 min) and were timed not to interfere with task per-
formance and the MRI acquisition. Blood samples
were processed as previously described.
whole blood activity and plasma activity after centrifu-
gation were measured in a c-counter (Wizard
Perkin Elmer, USA). The whole blood curve was line-
arly interpolated and resampled to match the time
points of the reconstructed PET frames. The plasma-
to-whole-blood ratio was averaged across time points.
The whole blood curve was then multiplied with the
mean plasma-to-whole-blood ratio to obtain an arterial
input function for absolute quantification.
PET data preprocessing and quantification of glucose
Data was reconstructed with an ordinary Poisson -
ordered subset expectation maximization algorithm
(OP-OSEM, 3 iterations, 21 subsets, matrix size:
344 344, slices: 127, voxel size: 2.09 2.09 2.03 mm)
and binned into 104 frames of 30 s. In addition to stan-
dard corrections such as dead time and decay, attenu-
ation and scatter correction was performed with a
based on the structural MRI
acquired at the first measurement. Preprocessing and
quantification steps were similar to our previous
SPM12 was used for head movement
correction (quality ¼1, registration to mean image),
spatial normalization to MNI space and spatial
smoothing with an 8 mm Gaussian kernel. The spatial
normalization was estimated with the structural MRI.
The mean PET image was then coregistered to the
structural MRI and both transformations (coregistra-
tion, normalization) were applied to the dynamic PET
data. Images were masked so that only grey matter
voxels were present and a low pass filter with a cutoff
frequency of half the task duration was applied to the
time course of every voxel. A general linear model
(GLM) was employed to separate task-specific and
baseline metabolism including four regressors: baseline,
one for each task condition (easy/hard, linear ramp
function, slope ¼1 kBq/frame) and the first principal
component of the six movement regressors estimated
during the movement correction step. The baseline
term was defined with a multimodal approach (as
well as an independent approach, see statistical analy-
sis). The individual BOLD data was used to identify
voxels in MNI space that exhibit significant task effects
(see below and Figure 2) in the hard vs rest condition
(p <0.05 FWE corrected voxel level). These voxels
were then masked out in the spatially normalized
PET frames. The remaining grey matter voxels were
4Journal of Cerebral Blood Flow & Metabolism 0(0)
considered inactive during the task and were averaged
within each frame, yielding a task-free baseline time
course. This approach has been proven useful in our
and yields similar task effects
compared to a BOLD-independent baseline defini-
The Gjedde-Patlak plot was applied to obtain
the influx constant K
with linearity set to 15 min after
tracer application resulting in 3 K
maps: rest, easy vs
rest and hard vs rest. Finally, CMRGlu was quantified
as CMRGlu ¼K
/LC * 100, with LC being
the lumped constant ¼0.89.
ASL data preprocessing and cerebral blood flow
The 2 D pseudo-continuous arterial spin labeling data
was processed as described previously:
intensity below 80% of the mean of raw ASL data
(across all voxels within each frame) were set to 0 to
remove areas with insufficient signal. Movement cor-
rection was carried out with SPM12 (quality ¼1) fol-
lowed by calculation of the equilibrium magnetization
) map as the temporal average of all unlabeled
images. The brain was extracted from the M
with the brain extraction tool implemented in FSL
and the resulting mask was applied to the images of
the time series. Data was spatially normalized to MNI
space via the T1-weighted structural image as done for
the PET data (i.e., mean image coregistered to T1, both
transformations applied to all ASL data) and
smoothed with an 8 mm Gaussian kernel. CBF was
where kis the blood-tissue water partition coefficient
(¼0.9 ml/g), DM the difference between pairs of labeled
and unlabeled images, R1a the longitudinal relaxation
rate of blood (¼0.67 sec
), athe tagging efficiency
(¼0.8), xthe post-labeling delay time adapted for
slice timing (¼1800 ms at slice 1) and sthe labeling
pulse duration (¼1.5 sec). CBF was averaged across
the time series.
Since the resulting maps acquired during task per-
formance represent the sum of baseline and task
effects, pure task-specific CBF was calculated by sub-
tracting CBF at rest from CBF obtained during the
easy and hard condition, respectively.
BOLD-derived task changes
Data preprocessing was carried out in SPM12 as
BOLD data was slice timing
corrected to the middle slice, realigned to the mean
image (quality ¼1), spatially normalized to MNI
space and smoothed (8 mm Gaussian kernel). First
level analysis was performed as block design with one
regressor for each task condition (control, easy, hard)
in the GLM. Additionally, regressors for movement,
white matter and cerebrospinal fluid (CSF) were
included. The following contrasts of interest were esti-
mated from the GLM’s beta values: control vs rest,
easy vs rest, hard vs rest, easy vs control, hard vs
Region of interest definition
In order to obtain an unbiased comparison between the
different imaging parameters, we defined functional
regions of interest (ROIs) based on all three imaging
modalities similar to our previous investigation
Figure 2. Task-specific changes and functional ROIs. Task effects were obtained with functional PET (fPET), arterial spin labeling
(ASL) and blood oxygenation (BOLD). The presented maps depict group t-maps of the contrasts hard vs rest for fPET and ASL (a and
b) and hard vs control for BOLD (c), all p
<0.05 corrected cluster level, height threshold of p <0.001 uncorrected voxel level. For
a robust analysis of the test-retest reliability, ROIs were determined with a conjunction analysis across all three modalities (inter-
section, d), revealing overlapping task changes in the frontal eye field (blue), intraparietal sulcus (pink), occipital cortex (green) and
supplementary motor area (SMA, red). We focused on large, robust clusters (>500 voxels) therefore, analysis of the SMA was
omitted. The slices z ¼4 and z ¼56 are presented in MNI space.
Rischka et al. 5
described below. To avoid potential bias, the ROI def-
inition and the test-retest evaluation was carried out
with separate study cohorts (see study design).
A group-wise one-sample t-test was performed for
each modality (fPET: K
, hard vs rest; ASL: CBF,
hard vs rest; BOLD: GLM beta values, hard vs control,
<0.05 corrected cluster level, height threshold
of p <0.001 uncorrected voxel level). The BOLD con-
trast hard vs control was chosen for a similar extent of
task effects in the three modalities. Finally, a conjunc-
tion analysis (i.e., intersection) across the three FWE-
corrected and binarized t-maps was computed to
obtain task-specific ROIs. The left and corresponding
right side of each ROI were merged. We focused on the
largest clusters (>500 voxels) to provide a robust def-
inition of task-relevant changes (Figure 2). These
included the occipital cortex (OCC), intraparietal
sulcus (IPS) and the frontal eye field (FEF) (see
results), which were also identified in our previous
work with a partly overlapping sample.
mean value of each ROI was extracted for further
Statistical analyses were based on commonly reported
parameters to enable comparison with previous litera-
ture, namely K
and CMRGlu for fPET, CBF for ASL
and parameter estimates for BOLD imaging. Similarly,
frequently used metrics of test-retest reliability were
used to compare imaging parameters between the two
measurements. Data were visually inspected for normal
distribution (Figure 4 and 5).
Quantitative comparisons were assessed using the
group-wise median within-subject coefficient of varia-
tion (CoV [%] ¼SD/mean*100) and the intraclass cor-
, equation (2)) for each modality, each
ROI and each condition:
where MSBS is the mean square between subjects and
MSE the mean square error. MSBS and MSE were
calculated from an n k matrix with n ¼20 observa-
tions and k ¼2 measurements.
Spatial similarity between the task effects of the first
and second measurement was assessed with the
Sørensen-Dice similarity coefficient (DICE) at group
level. Group level was chosen because no individual
statistical maps could be computed for ASL with the
current study design. Second-level analyses were per-
formed with a one-sample t-test (n ¼20) for each
measurement and each task condition for all three
modalities. T-maps were thresholded at p <0.05 FWE
corrected cluster level. DICE was calculated as
where Xare all active voxels in the first measurement of
one modality and one condition and Yare all active
voxels in the corresponding second measurement.
Additional analyses were conducted to rule out spu-
rious findings and corresponding statistics were cor-
rected for multiple testing with the Bonferroni-Holm
procedure (multiple ROIs and conditions). The use of
BOLD data to identify task-specific voxels in the fPET
analysis may introduce dependencies between the two
modalities. Thus, we also computed task changes in
glucose metabolism independent from the BOLD
data by modeling the fPET baseline term with a
and then re-calculated reli-
Next, we investigated potential effects of task per-
formance on test-retest reliability. Individual task per-
formance was computed as points per minute for each
condition following the original NintendoV
system for TetrisV
R. Differences in performance between
the two measurements were assessed by paired t-tests
and associations with imaging parameters were calcu-
lated by Pearson’s correlation.
Finally, the effects of head motion were assessed.
Average framewise displacement was computed from
the realignment parameters for each subject and imag-
Differences in framewise displacement
between the two measurements were assessed by paired
t-tests and associations with coefficients of variation
were calculated by Pearson’s correlation.
The following paragraphs represent the key findings.
A comprehensive list of all results can be found in
Table 1. Figure 3 shows voxel-wise data from a
representative subject (i.e., with average coefficient of
variation). Furthermore, we provide scatter and Bland-
Altman plots to compare imaging parameters between
the two measurements at rest (Figure 4) and during
task performance (Figure 5). Reliability of CMRGlu
is presented in Supplementary Table 1.
Regions of interest
The conjunction analysis revealed task-induced
changes across all three imaging modalities in the
6Journal of Cerebral Blood Flow & Metabolism 0(0)
Table 1. Reliability of K
, CBF and BOLD changes.
CoV (IQR) [%] ICC
FEF IPS OCC FEF IPS OCC Whole brain
Rest 5.4 (4.2) 4.6 (2.6) 5.0 (4.0) 0.88 0.89 0.87 -
Easy vs Rest 26.1 (43.1) 27.9 (43.1) 29.9 (52.2) 0.50 0.52 0.33 0.53
Hard vs Rest 21.0 (34.8) 17.8 (16.2) 13.6 (21.3) 0.65 0.76 0.65 0.61
Rest 10.1 (10.4) 6.4 (10.5) 5.3 (7.8) 0.39 0.69 0.76 -
Easy vs Rest 38.7 (62.0) 32.4 (28.3) 24.9 (39.3) 0.51 0.43 0.67 0.68
Hard vs Rest 15.0 (31.8) 25.1 (43.2) 17.5 (24.5) 0.67 0.48 0.72 0.73
Easy vs Rest 31.7 (31.2) 20.5 (37.3) 28.9 (49.8) 0.40 0.45 0.23 0.81
Hard vs Rest 16.4 (19.5) 21.1 (24.2) 24.5 (20.4) 0.41 0.13 0.09 0.78
Control vs Rest 22.3 (40.9) 38.9 (49.1) 24.5 (51.4) 0.72 0.44 0.13 0.78
Easy vs Control 61.4 (80.0) 43.4 (54.8) 40.2 (141.9) –0.06 0.46 0.55 0.35
Hard vs Control 34.0 (63.7) 33.8 (104.4) 49.4 (51.7) 0.12 –0.08 0.12 0.70
Commonly used metrics for test-retest reliability were estimated, namely median within-subject coefficient of variation (CoV), and intraclass cor-
relation (ICC) for each ROI and each condition. Additionally, whole-brain DICE coefficient was calculated. At resting state, K
exhibited the highest
reliability. During task performance, reliability dropped but was comparable between K
and CBF. With BOLD imaging the best performance was
achieved with the DICE coefficient, i.e. the qualitative spatial overlap of task effects between measurements.
FEF: frontal eye field, IPS: intraparietal sulcus, OCC: occipital cortex.
Figure 3. Visual comparison between measurements 1 and 2 (M1 and M2) for a representative (i.e. average test-retest reliability)
subject. (a) At rest, fPET and ASL demonstrate highly similar patterns of K
and CBF between the two measurements. (b) Task-specific
changes in K
, CBF and BOLD signal show clear activations for the regions of interest frontal eye field (FEF), intraparietal sulcus (IPS)
and occipital cortex (OCC) in both measurements. Similar slices as in Figure 2 are presented (z ¼8 and z ¼56, MNI space).
Rischka et al. 7
frontal eye field (FEF), intraparietal sulcus (IPS) and
the occipital cortex (OCC), depicted in Figure 2.
Additionally, the precentral gyrus and supplementary
motor area were identified but analyses were omitted
because of the limited size of the clusters.
Coefficient of variation
The smallest CoV were found for K
at rest with a
median of 4.6–5.4% for the different ROIs. Slightly
higher CoV were obtained for CBF at rest, namely
5.3–10.1%. For the two task conditions (easy/hard vs
rest, respectively) the CoV of K
between 13.6 and 29.9%. Similarly, higher CoV was
observed for CBF during task performance (between
15.0% and 38.7%). The CoV of BOLD changes com-
pared to rest fluctuated between 16.4% and 38.9%.
Higher CoV were obtained for the contrast of task
conditions vs control (33.8–61.4%). CoV of CMRGlu
was similar to K
(rest: 7.0–7.9%, task: 17.4–29.3%).
Similar to CoV, excellent reliability was obtained for K
at rest (ICC ¼0.87–0.89 for the different ROIs).
Contrarily, the ICC of CBF was lower during rest com-
pared to that of K
(0.39–0.76). Again, the reliability
decreased during task performance, with higher ICC
during the hard vs rest condition for K
and CBF (0.48–0.72). During the easy vs rest condition,
ICC was lower for K
(0.33–0.52), whereas ICC for
CBF was similar to hard (0.43–0.67). For BOLD
signal changes relative to rest, the ICC varied substan-
tially: easy (0.23–0.45), hard (0.09–0.41) and control
(0.13–0.72). Comparisons of task conditions vs control
even reached negative values (0.06–0.55). For
CMRGlu, ICC was slightly lower compared to K
ranging from 0.68 to 0.74 at rest and 0.31 to 0.61
during task performance.
Different from the other metrics, whole-brain similarity
of the task changes between the first and second mea-
surement was highest for BOLD changes for easy vs
rest (DICE ¼0.81) and hard vs rest (0.78). When con-
trasting against the control condition, DICE slightly
decreased to 0.70 for hard but dropped to 0.35 for
easy. For the other imaging modalities, DICE coeffi-
cients were lower than BOLD (K
: easy/hard: 0.53/0.61,
CMRGlu: easy/hard: 0.52/0.61 and CBF: 0.68/0.73).
We have carried out a number of additional analyses to
further support our findings. First, we calculated task-
Figure 4. Test-retest variability at rest. Scatter and Bland-Altman plots of glucose metabolism (K
) and cerebral blood flow (CBF)
comparing measurements 1 and 2 (M1 and M2). Both imaging modalities demonstrate good overall agreement between the two
measurements with few outliers and no proportional bias. No significant differences were observed between the measurements (all
p>0.5 corrected), indicating no relevant systematic bias. See Table 1 for complementary results. Colors for ROIs match those in
Figure 2. FEF: frontal eye field, IPS: intraparietal sulcus, OCC: occipital cortex.
8Journal of Cerebral Blood Flow & Metabolism 0(0)
specific changes in glucose metabolism independent of
the BOLD data. In line with our previous results this
showed similar activation patterns
reliability (Supplementary table 2). We further investi-
gated the potential effect of changes in task perfor-
mance. Even though the subjects did not train
between the two measurements there was a slight but
significant increase in performance for the easy (points
per minute 427 vs 710) and the hard task levels (1176 vs
1621, both p <0.01 corrected). However, there were no
significant changes in any of the corresponding imag-
ing parameters (all p >0.5 corrected) or any significant
correlations with changes in task performance (all
p>0.5 corrected). Finally, head motion was not differ-
ent between the two measurements for any of the imag-
ing modalities (all p >0.1 corrected). Furthermore, no
association between individual framewise displacement
and coefficients of variation was observed (all p >0.5
corrected), indicating that motion did not drive the
In this study, the test-retest reliability of functional
imaging parameters reflecting neuronal activation
during a complex visuo-spatial task was assessed. The
multimodal data acquisition approach facilitated a
direct comparison of fPET test-retest performance to
the well-established methods of ASL and BOLD imag-
ing. We observed excellent reliability of glucose metab-
olism and CBF at rest. The reliability decreased during
task performance for both K
and CBF but was lowest
for BOLD signal changes.
Measurements at resting state reflect resting activity,
which plays a key role in research and clinical applica-
tions, for example, to investigate altered metabolism or
blood flow in patients. A high test-retest reliability of
glucose metabolism at rest was already demonstrated
decades ago with 2-[1-
within the same day
but also with [
intervals of one to twelve weeks between measure-
Similar reliability was achieved in this study
with excellent ICC and small CoV. In a comprehensive
evaluation of CBF, test-retest reliability was compared
within session and 1 hour and 1 week apart.
expected, the CoV increased with the time between
measurements, which could be based on technical chal-
lenges and varying resting activity. Although our meas-
urements were separated by 4 weeks, we could achieve
similar CoV. With the caveat of longer measurement
exhibited a higher reliability than CBF in
terms of CoV and ICC, suggesting fPET as a robust
quantitative tool to map resting activity. Other than K
and CBF, estimation of BOLD changes during rest is
theoretically possible by calibrating the signal
usually not performed due to technical challenges.
Consequently, a comparison to K
and CBF was omit-
ted in this work.
In addition to imaging parameters at rest, a compre-
hensive assessment of brain function requires analysis
of changes during stimulation and performance of
Figure 5. Task-specific test-retest variability. Scatter and Bland-
Altman plots of glucose metabolism (K
), cerebral blood flow
(CBF) and blood oxygen level dependent signal (BOLD, param-
eter estimates) comparing measurements 1 and 2 (M1 and M2).
As for the rest condition, the Bland-Altman plots indicate good
agreement without proportional bias or major outliers and
comparable limits of agreement between the easy (a) and the
hard task (b). No significant differences were observed between
the measurements (all p >0.5 corrected), indicating no relevant
systematic bias. See Table 1 for complementary results. Colors
for ROIs match those in Figure 2. FEF: frontal eye field, IPS:
intraparietal sulcus, OCC: occipital cortex.
Rischka et al. 9
specific tasks. Although a drop in reliability for glucose
metabolism and CBF in both task conditions was
observed, the rather novel fPET technique showed sim-
ilar task-specific reliability compared to the established
ASL approach, however, coming at the cost of longer
CMRGlu had slightly lower reliability than K
definition, these inconsistencies can be ascribed to
changes in the LC or plasma glucose level between
measurements. Since LC is assumed to be stable in
the differences are likely to occur
due to alterations in the plasma glucose levels. Thus,
acquiring a temporal profile of the latter might
improve the reliability. Regarding CBF, even lower
ICC values were previously reported for target regions,
although the measurement interval was only 1–2 days
compared to 4 weeks in our study. Also robust para-
digms such as finger tapping exhibited lower reliability
within a measurement interval of 1 to 4 weeks.
reliability of fMRI has been subject to long-lasting dis-
cussions as results vary widely. Recently, a meta-
analysis on task-specific BOLD changes, including
more than 50 studies, revealed a low test-retest reliabil-
ity (mean ICC ¼0.40) across various task durations,
test-retest intervals and task types.
A similar average
ICC was achieved with our task when compared to rest
(ICC ¼0.33), which was even lower for contrasts of
task vs control (ICC ¼0.17). For comparison, the aver-
age ICC of K
and CBF was markedly higher with
values of 0.52 and 0.58. Interestingly, the DICE coef-
ficient (i.e., a metric that quantifies the spatial extent of
overlapping activation, independent of its amplitude)
was highest for BOLD changes compared to CBF
in almost all conditions. Together, ICC and
DICE indicate a high spatial overlap of task effects
for BOLD, but with fluctuating effect sizes. This sug-
gests to use BOLD more selectively, such as in research
questions were simple identification of active regions is
of greater importance than reliable quantitative
In summary, the test-retest reliability during task
performance varied across the presented imaging
approaches. A possible explanation is that each modal-
ity reflects different factors that are coupled to neuro-
nal activation to varying degrees. While a close
relationship was observed between metabolic processes
and CBF, a mismatch occurs upon task performance
between oxygen supply and utilization.
task-specific neuronal activation, the glutamatergic
release is increased which triggers several neurovascu-
lar signal pathways, including vasodilating agents such
as nitric oxide or prostaglandins.
These factors yield a
higher CBF and blood oxygen level in capillaries and
arterioles. The net decrease in deoxygenated hemoglo-
bin is the underlying mechanism of the BOLD effect.
Hence, BOLD signal changes are mediated by gluta-
mate release and influenced by the abovementioned
factors. A more direct measure of metabolism is
given by the radioactive glucose analogue [
Astrocytes metabolize glucose to supply neurons with
energy in form of lactate.
Also, in neurons them-
selves, glucose is transformed into ATP for action
potentials and synaptic transmission.
is irreversibly bound in cells and thus, task-specific
increase in energy consumption is proportional to
Of note, also glucose metabolism
influences the vasodilation by certain neurovascular
signal pathways, including lactate.
This complex neu-
rophysiological interplay between the presented imag-
ing approaches suggests to consider them as
complementary and not as substitution of one another.
Another, more technical, reason for differences in
test-retest variability might arise from different mea-
surement durations between the three modalities.
Acquisition times for ASL and BOLD imaging were
shorter than for fPET. It is known that the temporal
signal-to-noise ratio (tSNR) increases with the square
root of the measurement time, which may also result in
higher reliability. Hence, extending the measurement
time of BOLD imaging and ASL to that of fPET
(52 min) would theoretically increase the tSNR by
approximately 150% and 70%, respectively (see
Supplementary figure S1). Of note, this does not
imply an increase in test-retest reliability by the same
amount, because external influences such as physiolog-
ical effects are not considered. Thus, it is likely that the
test-retest reliability of task-based fMRI reaches a pla-
teau similar to that of resting-state fMRI.
Currently, BOLD imaging is often the first choice to
map task changes on the whole-brain level, although
several pitfalls are known. As mentioned, the BOLD
effect is rather unspecific due to its signal complexity
which is often compensated for by employing a control
condition. However, finding a suitable control condi-
tion is challenging as it bears the risk to remove poten-
tial active regions after contrasting the conditions and
has the drawback of introducing additional variance as
also shown in this work.
Based on our results and aforementioned difficulties
with BOLD imaging, we propose to consider fPET and
ASL as a robust combined tool to tackle novel research
questions using fully-integrated PET/MRI systems.
This enables the simultaneous acquisition of two com-
plementary aspects of neuronal activation. Different to
BOLD imaging, fPET and ASL directly reflect glucose
metabolism and CBF, respectively, making them less
dependent on control conditions to achieve higher spe-
cificity. The advantage of absolute quantification fur-
ther enables longitudinal comparisons during task
performance and at rest. However, the acquisition
10 Journal of Cerebral Blood Flow & Metabolism 0(0)
requires an adaption of already established paradigms.
While fPET requires longer tasks, which have to be
stable for several minutes, BOLD imaging employs
shorter paradigms with several repetitions.
Furthermore, BOLD changes can resolve event-
related task effects due to the higher temporal resolu-
tion. Differently, fPET and ASL are currently limited
to block designs, although real-time and event-related
designs were proposed for ASL within a specific tech-
Given the current improvement of PET
scanners with higher sensitivity,
of up to 100 milliseconds and advanced reconstruction
event-related designs also seem to be
highly feasible with fPET in the near future.
In cases where the simultaneous acquisition of fPET
and BOLD is favorable, we suggest short task blocks
or a hierarchical design.
For the latter, a fast on/off
design is embedded in longer blocks, enabling a simul-
taneous acquisition of fPET and BOLD signal with
respect to the paradigm requirements. Utilizing a
dual-echo ASL sequence
would further allow the
acquisition of all three modalities at the same time
with such a design. Still, this would imply tradeoffs
mostly at the cost of the BOLD signal as ASL employs
longer TR. Although a hierarchical design enables a
direct comparison between the modalities, the short
rest periods within a block might influence the fPET
The assessment of task-specific glucose metabolism
may also prove feasible in clinical routine. The current
European Association of Nuclear Medicine (EANM)
procedure guidelines for [
F]FDG PET brain imaging
suggest the acquisition of a 15-30 min static image. This
may also include stimulation paradigms outside the
scanner to identify brain areas involved in specific
task performance e.g., before surgery.
omits metabolic dynamics and therefore potentially
misses important functional alterations connected to
a disease. With fPET, resting and task-specific glucose
metabolism can be robustly determined within a single
30 min measurement, enabling an optimized therapy
planning (e.g., in tumor patients), but also potentially
allowing for an enhanced monitoring of progression in
Although high test-retest reliability of fPET was
demonstrated, we have to note a few limitations.
fPET currently requires arterial cannulation for abso-
lute quantification of glucose metabolism but there is
great effort to substitute the AIF with an image-derived
input function which will broaden the applicability of
Another drawback is the radiation
exposure of participants (here approx. 6.4 mSv for a
subject with 75 kg body weight). However, the develop-
ment of PET systems with high sensitivity
allow further reduction of the radiation burden up to
an order of magnitude.
Reliability obtained with a
complex visuo-spatial motor task with different task
loads might be more susceptible to individual daily per-
formance than simple tasks such as checkerboard stim-
ulation or finger tapping. However, mapping the
neuronal response to complex behavioral processes
requires performance of similarly challenging tasks,
which makes knowledge about the corresponding reli-
ability an essential aspect. Another limitation is the
slight improvement in task performance between the
two measurements. This may be ascribed to an
improved control to align the bricks, which may have
resulted in altered task-specific activation in the second
measurement. To diminish this effect, the measure-
ments were separated by 4 weeks and the ICC
used describing consistency rather than absolute agree-
ment. Furthermore, no changes in imaging parameters
or correlations with task performance were observed.
This indicates that the reported test-retest reliability
was not affected by the change in task performance.
Lastly, task-specific test-retest metrics were directly
compared between modalities, although data acquisi-
tion varied across modalities in duration and the
number of task blocks. Strictly speaking, only K
CBF were acquired simultaneously and BOLD changes
immediately afterwards within the same session. These
differences emerge from the different requirements of
imaging modalities. fPET and ASL perform best with
relatively long task blocks, whereas the BOLD signal is
unstable for long continuous task durations. To meet
these requirements, we employed a study design that is
technically most feasible but also allows to optimize the
task-specific acquisition independently for each imag-
ing modality. Our results therefore provide test-retest
data similar to commonly employed designs and it is
likely that longer measurement times would increase
the test-retest reliability of BOLD imaging and ASL
because of a higher tSNR. However, this has to be
evaluated systematically with the acquisition of more
task blocks and longer measurement durations to iden-
tify a plateau of reliability.
In conclusion, matching task-specific neuronal acti-
vations were robustly detected with fPET, ASL and
BOLD imaging in both task conditions. Despite its
recent introduction, similar or higher reliability was
achieved for fPET in comparison to the optimized
and well-established modalities of ASL and BOLD
imaging, albeit with a longer measurement time.
Indeed, BOLD changes exhibited a substantially
lower reliability, especially when using a control condi-
tion. Nevertheless, BOLD effects showed the highest
qualitative overlap between the measurements, suggest-
ing to use BOLD imaging preferably for the spatial
assessment of task changes. On the other hand, the
combination of fPET and ASL offers a robust tool,
Rischka et al. 11
enabling the simultaneous absolute quantification of
glucose metabolism and blood flow during rest and
task performance. Considering that fPET will benefit
from numerous further advancements such as progres-
sive modeling strategies, more sensitive scanners and
increased temporal resolution, an even higher reliabil-
ity of this technique is to be expected in the near future.
This enables the application of fPET not only in
research, but also in clinical routine.
Data availability statement
Raw data will not be publicly available due to reasons of data
protection. Processed data can be obtained from the corre-
sponding author with a data-sharing agreement.
The author(s) disclosed receipt of the following financial sup-
port for the research, authorship, and/or publication of this
article: This research was funded in whole, or in part, by the
Austrian Science Fund (FWF) KLI 610, PI: Andreas Hahn.
For the purpose of open access, the author has applied a CC
BY public copyright license to any Author Accepted
Manuscript version arising from this submission.
L. Rischka and M. Kl€
obl are recipients of a DOC
Fellowship of the Austrian Academy of Sciences at the
Department of Psychiatry and Psychotherapy, Medical
University of Vienna. S. Klug is supported by the MDPhD
Excellence Program of the Medical University of Vienna. The
scientific project was performed with the support of the
Medical Imaging Cluster of the Medical University of
Vienna. The sponsors were not involved in any part of
The authors are especially grateful to Danny JJ Wang from
the Stevens Neuroimaging and Informatics Institute,
University of South Carolina and the Regents of the
University of California, for providing the pCASL sequence.
We also want to thank J. Raitanen, J. V€
olkle and A.
Pomberger for radioligand synthesis and K. Papageorgiou,
A. Basaran, L. Silberbauer and the diploma students of the
Neuroimaging Labs (NIL) for medical support. The authors
are further grateful to K. Einenkel and E. Sittenberger for
subject recruitment and A. Jelicic for task implementation.
Declaration of conflicting interests
The author(s) declared the following potential conflicts of
interest with respect to the research, authorship, and/or pub-
lication of this article: M. Hacker received consulting fees
and/or honoraria from Bayer Healthcare BMS, Eli Lilly,
EZAG, GE Healthcare, Ipsen, ITM, Janssen, Roche, and
Siemens Healthineers. S. Kasper received grants/research
support, consulting fees and/or honoraria within the last
three years from Angelini, AOP Orphan Pharmaceuticals
AG, Celgene GmbH, Eli Lilly, Janssen-Cilag Pharma
GmbH, KRKA-Pharma, Lundbeck A/S, Mundipharma,
Neuraxpharm, Pfizer, Sage, Sanofi, Schwabe, Servier, Shire,
Sumitomo Dainippon Pharma Co. Ltd., Sun Pharmaceutical
Industries Ltd. and Takeda. R. Lanzenberger received travel
grants and/or conference speaker honoraria within the last
three years from Bruker BioSpin MR, Heel, and support
from Siemens Healthcare regarding clinical research using
PET/MRI. He is a shareholder of BM Health GmbH since
2019. W. Wadsak declares to having received speaker hono-
raria from the GE Healthcare and research grants from Ipsen
Pharma, Eckert-Ziegler AG, Scintomics, and ITG; and work-
ing as a part time employee of CBmed Ltd. (Center for
Biomarker Research in Medicine, Graz, Austria). All other
authors report no conflict of interest in relation to this study.
LR, VP, WW, MH, SKa, RL and AH designed the study.
GMG, PM, VR recruited the subjects. VP and WW produced
the radiotracer. LR, AH, GMG, PM and SKl acquired the
data. LR and AH analyzed the data and were supported by
MK. MH, SKa, RL and AH supervised the study. LR and
AH wrote the manuscript. All authors reviewed and
approved the final version of the manuscript.
Supplemental material for this article is available online.
Lucas Rischka https://orcid.org/0000-0002-6766-857X
Sebastian Klug https://orcid.org/0000-0001-8714-6608
Andreas Hahn https://orcid.org/0000-0001-9727-7580
1. Logothetis NK and Pfeuffer J. On the nature of the
BOLD fMRI contrast mechanism. Magn Reson Imaging
2004; 22: 1517–1531.
2. Smith AM, Lewis BK, Ruttimann UE, et al.
Investigation of low frequency drift in fMRI signal.
Neuroimage 1999; 9: 526–533.
3. Bollmann S, Kasper L, Vannesjo SJ, et al. Analysis and
correction of field fluctuations in fMRI data using field
monitoring. Neuroimage 2017; 154: 92–105.
4. Glover GH, Li TQ and Ress D. Image-based method for
retrospective correction of physiological motion effects in
fMRI: RETROICOR. Magn Reson Med 2000; 44:
5. Turner R. How much cortex can a vein drain?
Downstream dilution of activation-related cerebral
blood oxygenation changes. Neuroimage 2002; 16:
6. Pike GB. Quantitative functional MRI: concepts, issues
and future challenges. Neuroimage 2012; 62: 1234–1240.
7. Williams DS, Detre JA, Leigh JS, et al. Magnetic reso-
nance imaging of perfusion using spin inversion of arte-
rial water. Proc Natl Acad Sci U S A 1992; 89: 212–216.
8. Haller S, Zaharchuk G, Thomas DL, et al. Arterial spin
labeling perfusion of the brain: emerging clinical applica-
tions. Radiology 2016; 281: 337–356.
12 Journal of Cerebral Blood Flow & Metabolism 0(0)
9. Borogovac A and Asllani I. Arterial spin labeling (ASL)
fMRI: advantages, theoretical constrains and experimen-
tal challenges in neurosciences. Int J Biomed Imaging
2012; 2012: 1–13.
10. Phelps ME, Kuhl DE and Mazziota JC. Metabolic map-
ping of the brain’s response to visual stimulation: studies
in humans. Science 1981; 211: 1445–1448.
11. Schaefer SM, Abercrombie HC, Lindgren KA, et al.
Six-month test-retest reliability of MRI-defined PET
measures of regional cerebral glucose metabolic rate in
selected subcortical structures. Hum Brain Mapp 2000;
12. Hahn A, Gryglewski G, Nics L, et al. Quantification of
task-specific glucose metabolism with constant infusion
of 18F-FDG. J Nucl Med 2016; 57: 1933–1940.
13. Villien M, Wey H-Y, Mandeville JB, et al. Dynamic func-
tional imaging of brain glucose utilization using fPET-
FDG. Neuroimage 2014; 100: 192–199.
14. Rischka L, Gryglewski G, Pfaff S, et al. Reduced task
durations in functional PET imaging with [18F]FDG
approaching that of functional MRI. Neuroimage 2018;
15. Elliott ML, Knodt AR, Ireland D, et al. What is the test-
retest reliability of common task-functional MRI meas-
ures? New empirical evidence and a meta-analysis.
Psychol Sci 2020; 31: 792–806.
16. Hahn A, Breakspear M, Rischka L, et al. Reconfiguration
of functional brain networks and metabolic cost converge
during task performance. Elife 2020; 9: 1–18.
17. Varrone A, Asenbaum S, Vander Borght T, et al. EANM
procedure guidelines for PET brain imaging using [18F]
FDG, version 2. Eur J Nucl Med Mol Imaging 2009; 36:
18. Long JZ, Jacobson MS and Hung JC. Comparison of
FASTlab 18F-FDG production using phosphate and cit-
rate buffer cassettes. J Nucl Med Technol 2013; 41: 32–34.
19. Burgos N, Cardoso MJ, Thielemans K, et al. Attenuation
correction synthesis for hybrid PET-MR scanners: appli-
cation to brain studies. IEEE Trans Med Imaging 2014;
20. Hahn A, Gryglewski G, Nics L, et al. Task-relevant brain
networks identified with simultaneous PET/MR imaging
of metabolism and connectivity. Brain Struct Funct 2018;
21. Wang J, Zhang Y, Wolf RL, et al. Amplitude-modulated
continuous arterial spin-labeling 3.0-T perfusion MR
imaging with a single coil: feasibility study. Radiology
2005; 235: 218–228.
22. Smith SM. Fast robust automated brain extraction. Hum
Brain Mapp 2002; 17: 143–155.
23. Bartlett EJ, Brodie JD, Wolf AP, et al. Reproducibility of
cerebral glucose metabolic measurements in resting human
subjects. J Cereb Blood Flow Metab 1988; 8: 502–512.
24. Maquet P, Dive D, Salmon E, et al. Reproducibility of
cerebral glucose utilization measured by PET and the
[18F]-2-fluoro-2-deoxy-d-glucose method in resting, healthy
human subjects. Eur J Nucl Med 1990; 16: 267–273.
25. Chen Y, Wang DJJ and Detre JA. Test-retest reliability
of arterial spin labeling with common labeling strategies.
J Magn Reson Imaging 2011; 33: 940–949.
26. Barrio JR, Huang S-C, Satyamurthy N, et al. Does 2-
FDG PET accurately reflect quantitative in vivo glucose
utilization? J Nucl Med 2020; 61: 931–937.
27. Yang FN, Xu S, Spaeth A, et al. Test-retest reliability of
cerebral blood flow for assessing brain function at rest
and during a vigilance task. Neuroimage 2019; 193:
28. Sari H, Erlandsson K, Law I, et al. Estimation of an
image derived input function with MR-defined carotid
arteries in FDG-PET human studies using a novel partial
volume correction method. J Cereb Blood Flow Metab
2017; 37: 1398–1409.
29. Gusnard DA, Raichle ME and Raichle ME. Searching
for a baseline: functional imaging and the resting human
brain. Nat Rev Neurosci 2001; 2: 685–694.
30. Fox PT, Raichle ME, Mintun MA, et al. Nonoxidative
glucose consumption during focal physiologic neural
activity. Science 1988; 241: 462–464.
31. Attwell D, Buchan AM, Charpak S, et al. Glial and neu-
ronal control of brain blood flow. Nature 2010; 468:
32. Attwell D and Iadecola C. The neural basis of functional
brain imaging signals. Trends Neurosci 2002; 25: 621–625.
33. Pellerin L and Magistretti PJ. Glutamate uptake into
astrocytes stimulates aerobic glycolysis: a mechanism
coupling neuronal activity to glucose utilization. Proc
Natl Acad Sci U S A 1994; 91: 10625–10629.
34. Harris JJ, Jolivet R and Attwell D. Synaptic energy use
and supply. Neuron 2012; 75: 762–777.
35. Zimmer ER, Parent MJ, Souza DG, et al. [18F]FDG
PET signal is driven by astroglial glutamate transport.
Nat Neurosci 2017; 20: 393–395.
36. Iadecola C. The neurovascular unit coming of age: a
journey through neurovascular coupling in health and
disease. Neuron 2017; 96: 17–42.
37. Birn RM, Molloy EK, Patriat R, et al. The effect of scan
length on the reliability of resting-state fMRI connectiv-
ity estimates. Neuroimage 2013; 83: 550–558.
38. Hernandez-Garcia L, Jahanian H, Greenwald MK, et al.
Real-time functional MRI using pseudo-continuous arte-
rial spin labeling. Magn Reson Med 2011; 65: 1570–1577.
39. Hernandez-Garcia L, Lee GR, Vazquez AL, et al. Fast,
pseudo-continuous arterial spin labeling for functional
imaging using a two-coil system. Magn Reson Med
2004; 51: 577–585.
40. van Sluis J, de Jong J, Schaar J, et al. Performance char-
acteristics of the digital biograph vision PET/CT system.
J Nucl Med 2019; 60: 1031–1036.
41. Zhang X, Cherry SR, Xie Z, et al. Subsecond total-
body imaging using ultrasensitive positron emission
tomography. Proc Natl Acad Sci U S A 2020; 117:
42. Jamadar SD, Ward PG, Li S, et al. Simultaneous task-
based BOLD-fMRI and [18-F] FDG functional PET for
measurement of neuronal metabolism in the human
visual cortex. Neuroimage 2019; 189: 258–266.
43. Ghariq E, Chappell MA, Schmid S, et al. Effects of back-
ground suppression on the sensitivity of dual-echo arte-
rial spin labeling MRI for BOLD and CBF signal
changes. Neuroimage 2014; 103: 316–322.
Rischka et al. 13
44. Sundar LK, Muzik O, Rischka L, et al. Towards quan-
titative [18F]FDG-PET/MRI of the brain: automated
MR-driven calculation of an image-derived input func-
tion for the non-invasive determination of cerebral glu-
cose metabolic rates. J Cereb Blood Flow Metab 2019; 39:
45. Cherry SR, Jones T, Karp JS, et al. Total-Body PET:
maximizing sensitivity to create new opportunities for
clinical research and patient care. J Nucl Med 2018;
14 Journal of Cerebral Blood Flow & Metabolism 0(0)