Reduction of physiological noise with independent component analysis improves the detection of nociceptive responses with fMRI of the human spinal cord

Article (PDF Available)inNeuroImage 63(1):245-52 · July 2012with22 Reads
DOI: 10.1016/j.neuroimage.2012.06.057 · Source: PubMed
Abstract
The evaluation of spinal cord neuronal activity in humans with functional magnetic resonance imaging (fMRI) is technically challenging. Major difficulties arise from cardiac and respiratory movement artifacts that constitute significant sources of noise. In this paper we assessed the Correction of Structured noise using spatial Independent Component Analysis (CORSICA). FMRI data of the cervical spinal cord were acquired in 14 healthy subjects using gradient-echo EPI. Nociceptive electrical stimuli were applied to the thumb. Additional data with short TR (250 ms, to prevent aliasing) were acquired to generate a spatial map of physiological noise derived from Independent Component Analysis (ICA). Physiological noise was subsequently removed from the long-TR data after selecting independent components based on the generated noise map. Stimulus-evoked responses were analyzed using the general linear model, with and without CORSICA and with a regressor generated from the cerebrospinal fluid region. Results showed higher sensitivity to detect stimulus-related activation in the targeted dorsal segment of the cord after CORSICA. Furthermore, fewer voxels showed stimulus-related signal changes in the CSF and outside the spinal region, suggesting an increase in specificity. ICA can be used to effectively reduce physiological noise in spinal cord fMRI time series.
Technical Note
Reduction of physiological noise with independent component analysis improves the
detection of nociceptive responses with fMRI of the human spinal cord
G. Xie
a,b,1
, M. Piché
a,b,2
, M. Khoshnejad
a,b
, V. Perlbarg
c
, J.-I. Chen
b,e
, R.D. Hoge
b
, H. Benali
b,c
,
S. Rossignol
a
, P. Rainville
a,b,e
, J. Cohen-Adad
a,b,c,d,f,
a
GRSNC, Faculty of Medicine, Université de Montréal, Montreal, QC, Canada
b
Unité de Neuroimagerie Fonctionnelle, CRIUGM, Université de Montréal, QC, Canada
c
INSERMUniversité Pierre et Marie Curie Paris 06, UMiR-S 678, Laboratoire d'Imagerie Fonctionnelle, 75634 Cedex 13 Paris, France
d
A.A. Martinos Center for Biomedical Imaging, MGH, Harvard Medical School, Charlestown, MA, USA
e
Department of Stomatology, Université de Montréal, QC, Canada
f
Department of Electrical Engineering, Ecole Polytechnique de Montreal, QC, Canada
abstractarticle info
Article history:
Accepted 23 June 2012
Available online 6 July 2012
Keywords:
Spinal cord
fMRI
Physiological noise
ICA
Pain
The evaluation of spinal cord neuronal activity in humans with functional magnetic resonance imaging (fMRI) is
technically challenging. Major difculties arise from cardiac and respiratory movement artifacts that constitute
signicant sources of noise. In this paper we assessed the Correction of Structured noise using spatial Indepen-
dent Component Analysis (CORSICA). FMRI data of the cervical spinal cord were acquired in 14 healthy subjects
using gradient-echo EPI. Nociceptive electrical stimuli were applied to the thumb. Additional data with short TR
(250 ms, to prevent aliasing) were acquired to generate a spatial map of physiological noise derived from Inde-
pendent Component Analysis (ICA). Physiological noise was subsequently removed from the long-TR data after
selecting independent components based on the generated noise map. Stimulus-evoked responses were ana-
lyzed using the general linear model, with and without CORSICA and with a regressor generated from the cere-
brospinal uid region. Results showed higher sensitivity to detect stimulus-related activation in the targeted
dorsal segment of the cord after CORSICA. Furthermore, fewer voxels showed stimulus-related signal changes
in the CSF and outside the spinal region, suggesting an increase in specicity. ICA can be used to effectively reduce
physiological noise in spinal cord fMRI time series.
© 2012 Elsevier Inc. All rights reserved.
Introduction
Following the success of functional magnetic resonance imaging
(fMRI) in the investigation of brain function, fMRI of the spinal cord
was shown to be technically feasible in both humans (Backes et al.,
2001; Bouwman et al., 2008; Brooks et al., 2008; Cohen-Adad et al.,
2010; Eippert et al., 2009; Giulietti et al., 2008; Govers et al., 2007;
Madi et al., 2001; Maieron et al., 2007; Majcher et al., 2007; Stracke et
al., 2005; Stroman et al., 2004; Summers et al., 2010; Valsasina et al.,
2008; Yoshizawa et al., 1996) and animals (Cohen-Adad et al., 2009;
Endo et al., 2008; Lawrence et al., 2007; Lilja et al., 2006; Majcher et
al., 2007; Zhao et al., 2008). However, results remain controversial due
to the low reproducibility of spinal cord fMRI (Bouwman et al., 2008;
Giove et al., 2004). Because of the relatively small cross-sectional size
of the spinal cord grey matter, even very little motion can contaminate
the signal. Spinal cord motion and in-ow effects from the surrounding
cerebrospinal uid (CSF) along with cardiac and respiratory cycles
greatly degrade the quality of fMRI data by adding unwanted variance
in the time series (Figley and Stroman, 2007; Giove et al., 2004; Kong
et al., 2012; Stroman, 2006; Stroman et al., 2005). Non-invasive acquisi-
tion of functional images of the spinal cord is very challenging but re-
mains much needed even though there is currently no methodological
consensus on the adequate means to address this challenge.
Several methods exist to minimize physiological-related noise.
Respiratory-gated acquisition and breath-hold have been employed to
reduce the effect of respiratory activity (Stroman and Ryner, 2001;
Stroman et al., 1999), although with moderate success (Stroman,
2005). Cardiac gating has also been used in the brain (Guimaraes et
al., 1998) and spinal cord (Backes et al., 2001) but this requires substan-
tially longer acquisition time. Moreover, variable TR can introduce addi-
tional signal variance due to T
1
-effects, which are difcult to correct in
the spinal cord due to the necessity of acquiring a robust T
1
map. Addi-
tionally, heart rate can correlate with the experimental paradigm
especially for painful stimuli, therefore spurious activations can appear
NeuroImage 63 (2012) 245252
Corresponding author at: Department of Electrical Engineering, Ecole Polytechnique
de Montreal, 2500, Chemin de Polytechnique, Montreal, QC, Canada H3T 1J4. Fax: +1 514
340 4611.
E-mail address: jcohen@polymtl.ca (J. Cohen-Adad).
1
Current address: Department of Anesthesiology, University of S askatchewan,
Saskatoon, SK, Canada.
2
Current address: Department of Chiropractic, Université du Québec à Trois-Rivières,
QC, Canada.
1053-8119/$ see front matter © 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.neuroimage.2012.06.057
Contents lists available at SciVerse ScienceDirect
NeuroImage
journal homepage: www.elsevier.com/locate/ynimg
in the statistical map as TR would also correlate with heart rate
(Tousignant-Laamme et al., 2005).
Post-hoc correction can be used to reduce physiological noise in fMRI
time series, as shown in the brain (Behzadi et al., 2007; Deckers et al.,
2006; Glover et al., 2000; Hu et al., 2005; Lund et al., 2006; Thomas et
al., 2002; Tohka et al., 2008) and spinal cord (Brooks et al., 2008;
Figley and Stroman, 2007; Kong et al., 2012; Stroman, 2006). One meth-
od based on the RETROspective Image CORrection (RETROICOR) algo-
rithm (Glover et al., 2000) consists in estimating a set of regressors
based on external physiological recordings (pulse oxymeter and respira-
tion trace) to be included in the general linear model (GLM). Although it
has shown great success in capturing the variance of physiological noise
in several fMRI studies, this approach might inappropriately model the
shape of physiological-related signal by a combination of sine and cosine
functions. Alternatively, data-driven methods aim at extracting the
physiological noise part of the fMRI data from the data itself. One such
method performs CORrection of structured noise using Spatial Indepen-
dent Component Analysis (CORSICA) (Perlbarg et al., 2007). The princi-
ple of CORSICA is to estimate a noise map via the spatial independent
component analysis (ICA) and to remove the corresponding compo-
nents from the functional data before testing for the effect of interest.
The noise components can be identied using anatomical priors, as in
(Perlbarg et al., 2007), or based on separate data acquisition using a
short TR to assess the spatial distribution of noise at rest. Short-TR data
are used to avoid aliasing, particularly for cardiac signal , whose main
spectrum typically ranges from 0.8 to 1.4 Hz. Theref ore a sampling fre-
quency of at least 2 × 1.4 Hz (or TR b 350 ms) is required to satisfy the
Nyquist condition. CORSICA proved to be useful in reducing physiological
noise in brain fMRI time series (Schrouff et al., 2011; Vanhaudenhuyse et
al., 2010). The main assumption of CORSICA is that physiological noise is
spatially structured. Our recent investigation on the spatial distribution of
physiological noise in spinal cord fMRI demonstrated that cardiac-related
noise one major source of signal variance in spinal cord fMRI is spa-
tially structured and stable within each individual (Piché et al., 2009).
Brooks et al. perfor med ICA on spinal cord fMRI data acquired with
short TR (200 ms) to allow exploration of the physiological noise spectra
(Brooks et al., 2008). They showed that cardiac and respiratory effects are
easily extracted by the ICA decomposition. The cardiac-related signal was
mostly located in the CSF region and in the carotid and vertebral vessels,
while the respiratory-related signal usually appeared at the interface be-
tween connective tissues (neck, muscles). The interaction between cardi-
ac and respiratory signal (amplitude modulation) also appeared in the
ICA. In addition, a low-frequency component (b 0.1 Hz) was robustly
identied in multiple subjects. All these observations suggest that a
map of physiologic al noise can be computed for each subject and subse-
quently used within the same subject in various experiments.
In this paper we assessed whether CORSICA can improve the sen-
sitivity and specicity of BOLD responses to nociceptive stimuli in the
cervical spinal cord.
Materials and methods
Acquisitions
All experimental procedures conformed to the standards set by the
latest revision of the Declaration of Helsinki and were approved by
the Research Ethics Board of our institution (Comité mixte d'éthique
de la recherche du Regroupement Neuroimagerie Québec; CMER-
RNQ). All participants gave written informed consent, acknowledging
their right to withdraw from the experiment without prejudice, and re-
ceived compensation for their travel expenses, time and commitment.
FMRI acquisitions were carried out in 14 healthy volunteers with a 3
Tesla Siemens Trio system (Siemens Healthcare, Germany). RF recep-
tion was achieved by combining the product neck coil (4-channel)
with the spine matrix (the 6 most rostral elements were used). Coils
were sensitive from about C1 to T5. Subjects were between 20 and
49 years old (mean: 34), and had no history of neurological conditions,
including degenerative disorder or spinal trauma.
Subjects were carefully positioned to limit head movement and neck
lordosis and were requested not to move. An anatomical scan was
performed using a T
1
-weighted sequence (3D MPRAGE, sagittal orienta-
tion, 208 slices, TR=2250 ms; TE= 3.4 ms; ip angle= 9°; eld of
view=256 mm, 1 mm isotropic). Three functional scans were
performed in each subject using a T
2
*-weighted gradient-echo
echo-planar imaging (EPI) sequence. Sagittal slices were positioned to
cover the vertebral body of C3 to T1, with the middle slice centered
on the mid-sagittal plane. Parameters were: TR=2 s, TE=20 ms, ip
angle=82°, 9 sagittal slices, thickness=3 mm, no gap, in-plane resolu-
tion=1.6×1.6 mm
2
, 315 volumes. In addition, one short-TR run of
120 s was performed without stimulation (TR=250 ms, TE=20 ms,
ip angle=40°, 5 sagittal slices, thickness=3 mm, gap=1.5 mm,
in-plane resolution=1.6×1.6 mm
2
, 480 volumes). The short-TR data
were used to generate a map of physiological noise for each subject.
Functional paradigm
Nociceptive transcutaneous electrical stimuli (trains of 1-ms pulses
at 5 Hz) were delivered to the subject's right thumb by a pair of
custom-made surface electrodes (1 cm
2
; 2 cm inter-electrode distance)
and using a custom-made optically isolated constant current stimulator
triggered by a Grass S48 train generator (Grass Medical Instruments,
Quincy, MA, USA). Stimuli were synchronized with fMRI acquisitions
using a stimulus presentation program (E-Prime2, Psychology Software
Tools, Sharpsburg, PA, USA). Before scanning, subjects were familiarized
with the electrical stimuli and pain threshold was determined individu-
ally using the ascending method of limit. Series of supra-threshold stim-
uli were then administered to determine the stimulus intensity required
to produce moderate to strong painful sensations while insuring that the
subject could tolerate it for 60 s (i.e. rated 5070 on a 0100 numerical
rating scale with 0 dened as no pain and 100 as worst pain imaginable).
The functional paradigm consisted of alternating blocks of stimuli (S)
andrest(R)asfollows:R|S|R|S|R|S|R|S|R|S|R.Eachblock
was 60 s long (except for the rst block, which lasted 30 s). Total acqui-
sition time for each run was 10 min 30 s. Each subject had three func-
tional runs. BOLD responses to nociceptive stimuli of the thumb were
expected to occur mostly in the spinal cord at around C5 vertebral
level, which roughly corresponds to spinal level C6.
Physiological monitoring
A pulse oximeter probe was attached to the subject's left index n-
ger. A respiratory belt was attached around the subject's chest. Physio-
logical data were sampled at 1000 Hz and recorded on a MP150 system
(Biopac Systems, Inc., Goleta, CA, USA).
Data analysis
Data analysis was performed with Matlab® (The Mathworks Inc.,
Natick, MA, USA) using the sICA toolbo (2007 Inserm U678 V.
Perlbarg), the CORSICA toolbox (Perlbarg et al., 2007) and SPM2
(www.l.ion.ucl.ac.uk/spm/spm2.html). EPI data were corrected for
slice timing differences and motion-corrected using a rigid body align-
ment. The mean of the whole functional data set was calculated to ob-
tain a target volume for registration to the individual's T
1
-weighted
anatomical volume. No smoothing was applied to the data.
Generation of noise maps from short-TR data
Spatial ICA was performed on short-TR data to retrieve
physiological-related signal. A principal component analysis was
rst performed to whiten the data without reducing the data dimen-
sion, then spatial ICA was run (INFOMAX algorithm). Number of inde-
pendent components was set to 30, which is enough to explain most
246 G. Xie et al. / NeuroImage 63 (2012) 245252
physiological noise variance (Brooks et al., 2008; Piché et al., 2009).
The power spectrum of each component was examined to identify
three components showing the highest coherence with the power
spectra of cardio-respiratory data recorded in the corresponding
run. Namely, we identied one component having the highest coher-
ence with the cardiac trace, one component having the highest coher-
ence with the respiratory trace, and one component with the highest
coherence with both respiratory and cardiac spectra. Mean peak fre-
quencies of cardiac and respiratory uctuations were 1.2 Hz and
0.33 Hz, respectively (Fig. 1a).
The time course of the three selected components was regressed out
from the short-TR fMRI time-series on a voxel-by-voxel basis to obtain a
spatial distribution of physiological noise components. The resulting
T-maps indicated that most of the physiological noise variance was lo-
cated in the cerebrospinal uid (CSF) and along the spinal cord/CSF in-
terface, overlapping partly with spinal tissue (Fig. 2). T-maps were then
converted into a binary mask using an upper threshold of T =3.12
(P-uncorrected=0.001). Masks were not blurred.
Removal of physiological-related components from long-TR data
Spatial ICA was conducted on the long-TR data to generate 60 inde-
pendent components. The noisy components were then identied on
the basis of the similarity between the spatial maps of each component
and the noise map. To prevent us from selecting components related to
functional activity, only the components that shared the highest simi-
larity with the noise map were selected. Then, the components selec-
tion was achieved by a stepwise regression procedure. This procedure
iteratively selects a subset of components that explain each noise char-
acteristic signal. These noise characteristic signals were extracted from
the noise map by using a k-means clustering. Then, the frequency of se-
lection of each component (Fq) was calculated from the stepwise re-
gression procedure (one selection by noise characteristic signal). Fq
represents the spatial similarity between each component and the
noise map (Fig. 1b). The full procedure is detailed in (Perlbarg et al.,
2007). Fq was then thresholded based on the Otsu's approach (Otsu,
1979). The Otsu's approach consists in nding an adaptive threshold
of the histogram of Fq scores. This algorithm supposes that the histo-
gram is bimodal (i.e. there are two classes of components, the noise
components and the other ones) and nds the optimal threshold by
minimizes the intraclass variance.
Finally, data were reconstructed by combining all ICA components
without the noisy components, i.e., components with Fq score higher
than the threshold.
Analysis of stimulus-evoked responses
Statistical analysis was conducted on the long-TR data, using the
standard GLM implemented in SPM2, with high-pass ltering at 128 s
cut-off and motion regressors. Blocks of nociceptive stimuli were con-
volved with the default canonical hemodynamic response function.
For each subject, data from all three runs were combined using a
xed-effect analysis (i.e., intra-subject analysis). To assess the benet
of CORSICA, the GLM was conducted with and without noise correction.
We also compared CORSICA with the GLM that includes a CSF regressor
(
Kong et al., 2012). We used the CSF masks (see next section) to recon-
struct the averaged time course of the CSF signal for each run and each
subjects, and subsequently applied it as a regressor in the GLM for fur-
ther comparison with the CORSICA method.
Regions of interest (ROI)
Masks were created from the mean functional mask to avoid
mismatch between the T
1
-weighted anatomical image and the EPI
due to susceptibility distortions. A mask of the spinal canal (CSF+spinal
cord) and spinal cord was manually drawn on the mean functional
image (Fig. 3). The spinal cord mask was then split to a ventral and a
Fig. 1. a: The power spectrum of one component that exhibits both cardiac and respiratory signals. This is the rst of the 30 independent components extracted by sICA from a raw
time-series acquired using a short-TR. The cardiac and respiratory components were identied based on the coherence with the cardiac and respiratory activity in the power spec-
trum. b: Selection of the noise-related components in long-TR. The score Fq is plotted against all 60 components of long-TR data in one subject. An adaptive threshold of Fq was
automatically determined by CORSICA to select noise-related components (horizontal red line).
247G. Xie et al. / NeuroImage 63 (2012) 245252
dorsal mask for quantifying BOLD responses in the ventral and dorsal
segments. Masks were then divided into six parts corresponding to ver-
tebral levels, i.e., from C3 to T1. A mask of the CSF was created by
subtracting the spinal cord mask from the spinal canal mask.
The mean T-scores and the total voxel counts above threshold were
extracted within each ROI at the following thresholds: P=0.01, P=
0.001 (uncorrected for multiple comparisons) and P=0.05 (corrected
for multiple comparisons using Bonferroni). Results were compared
across methods using two-sample Student's t-test paired for subject,
with a threshold of P= 0.01 (corrected for comparisons between the
three conditions, i.e., no correction, CSF regressor and CORSICA).
Results
Fig. 4 shows T-map s of responses to nociceptive stimuli in four repre-
sentative subjects, without and with physiological noise correction using
CORSICA. Overall, higher peak T-score (subjects #1, #3, #4) and higher
voxel count (subjects #3, #4) of stimulus-related responses was
detected in the cord with CORSICA. Fewer stimulus-relate d responses
were observed outside the spinal cord with CORSICA, suggesting an in-
crease in spatial specicity. Similar results were obtained in the 10
other subjects. The time course of data after CORSICA exhibited lower
high frequency variations, as illustrated by Fig. 5.
Mean T-score and voxel count of signicant (Pb 0.01) responses to
nociceptive stimuli in all 14 subjects across vertebral levels in the spinal
cord mask were compared without correction, with the CSF regressor
and with CORSICA (Fig. 6). The voxel count and mean T-scores were sig-
nicantly increased with CORSICA at levels C4 and C6 (for voxel count)
and at C4 and C7 (for the mean T-score). Note that the T-maps shown in
Fig. 4 suggest a different interpretation on the distribution of positive
responses across vertebral levels, compared to the one described in
Fig. 6. However, the four subjects in Fig. 4 were presented only for illus-
trating the effect of T-score distribution before and after CORSICA. Fig. 6
should be considered the most reliable measure of the distribution of
responses across vertebral levels, as it includes all 14 subjects.
Fig. 7 shows signicant voxel counts in the ventral spinal cord, dor-
sal spinal cord and CSF. After CORSICA, the number of active voxels
was signicantly higher in the ventral and dorsal spinal cord
(Pb 0.05). No signicant change was detected in the CSF region.
Discussion
This study assessed the efciency of the CORSICA method to correct
physiological noise uctuation in spinal cord fMRI time series. This
method utilizes ICA from short-TR data to derive subject-dependent
spatial map of physiological noise and subsequently used to remove
physiological-noise-related components from long-TR fMRI data based
on the similarity criterion. CORSICA showed an increase in sensitivity
and specicity for the detection of BOLD responses to noxious stimuli
in the spinal cord.
Fig. 2. Four representative T-maps showing a mid-sagittal view of noise components extracted from the short-TR data (at rest). Each T-map was then converted into a binary mask with a
cutoff T-value of 3.12 (P-uncorrected= 0.001) and used as an input into CORSICA to identify spatially-matching components in the long-TR data. Image orientation: left=ventral, top=
rostral. Data were smoothed with a 3×3 ×6mm
2
kernel for clarity purpose.
Fig. 3. Example of masks of the spinal canal (CSF+spinal cord) and spinal cord used for ROI-based analysis of BOLD responses to nociceptive stimuli. Image orientation: left=ventral,
top=rostral.
248 G. Xie et al. / NeuroImage 63 (2012) 245252
Efciency of physiological noise correction
Although our main focus was the application of the CORSICA meth-
od, we also evaluated the use of a CSF regressor for modeling major
cardiac-related uctuations, as reported elsewhere to be one of the
main source of signal variance in the cord (Kong et al., 2012). We
were not able to compare our method with the RETROICOR method be-
cause some of the physiological traces recorded for the long-TR data
contained artifacts. Evaluation criteria were based on the spatial speci-
city of the responses by comparing number of activated voxels
(i.e., passing a certain T-threshold) and mean T-score at different levels
of the spinal cord and in the CSF area. Our results showed that the CSF
regressors and the CORSICA produced higher T-score, indicating that
the residual variance is decreased when modeling the physiological
noise in the fMRI time series, as conrmed by the time course (Fig. 5).
Testretest repeatability will be performed in future studies to assess
the robustness of the detected activations across sessions.
Spatial distribution of physiological noise
Here we utilized the information from short-TR data to identify the
spatial distribution of physiological noise. The main assumption of
CORSICA is that physiological noise is spatially structured, i.e.,some
areas should exhibit spatially organized patterns of physiological
Fig. 4. T-statistic maps of stimulus-evoked responses without (top) and with (bottom) physiological noise correction using CORSICA in four representative subjects. Statistical maps
are overlaid on the mid-sagittal slice of the individual anatomical image. Overall, higher number of stimulus-related responses is detected in the cord with CORSICA (e.g. subjects
#3, #4) with fewer false-positive responses outside spinal cord (e.g. subjects #1, #2, #4). Data were smoothed with a 3×3×6mm
2
kernel for clarity purpose.
Fig. 5. Time course of fMRI data before (blue) and after CORSICA (red) in one subject (single run). Motion-corrected and detrended data were selected out of a cluster at C6 of 13
voxels passing the P=0.01 threshold (regressor of interest from the GLM statistics), then averaged. The location of this cluster was chosen at the C6 vertebral level. The stimulation
paradigm is displayed in gray.
249G. Xie et al. / NeuroImage 63 (2012) 245252
noise. Our previous investigation on the spatial distribution of physio-
logical noise in spinal cord fMRI demonstrated that cardiac-related
noise one major source of signal variance in spinal cord fMRI is spa-
tially structured and stable across successive runs within individual
(Piché et al., 2009). Most cardiac-related variance was detected in the
CSF area as well as within large vessels. In addition, a recent study dem-
onstrated that the physiological noise was higher in the CSF and along
the spinal cord/CSF interface (Figley and Stroman, 2009). In the present
study, noise masks derived from short-TR data also showed that
physiological-related noise was mainly present within the CSF and
along the spinal cord/CSF interface (see Fig. 2).
One limitation is that components selected as noisy in the long-TR
data might have included some variance from the BOLD time course,
due to possible source mixing during ICA decomposition. This is a com-
mon problem with ICA (McKeown and Sejnowski, 1998). To improve
the separation between physiological noise and the signal of interest
(BOLD response), prior information could be added such as incorporat-
ing paradigm information into the ICA analysis (Calhoun et al., 2005).
Inter-subject variability
Our previous investigations showed that although spatial distribu-
tion of cardiac noise is stable within the same subject (testretest re-
peatability), it can vary quite substantially across subjects (Piché et al.,
2009). In the present study, noise maps were qualitatively different be-
tween individuals, as clearly evidenced in Fig. 2.Thissuggeststhata
template of noise map for the spinal cord is not advisable yet, and that
noise map should be estimated for each individual for maximum accu-
racy in the removal of physiological noise-related components.
External recording of physiological signals
Cardiac and respiratory traces were recorded using external probes
to identify three independent components from the short-TR data. Al-
though CORSICA has originally been introduced as a method that does
not necessarily require external physiological recording as opposed
to RETROICOR-based methods physiological recording has been
conducted for the sole purpose of validating the method for spinal
cord fMRI. In the future it is conceivable to use an algorithm that
would identify cardiac and respiratory frequency peaks based on a tem-
plate of the expected physiological spectra computed from a population.
Habituation effects
Previous studies have reported a habituation effect during repeated
nociceptive stimulations, characterized by a decrease of the BOLD re-
sponses (Becerra et al., 1999). To address the potential presence of ha-
bituation effect in our data, we inspected the time courses of fMRI
signals within signicant voxels in several subjects (see Fig. 5). We
could notice a decrease in the signal amplitude after a certain time
(~30 s), which could correspond to a decrease of the BOLD response as-
sociated with a habituation effect. Although this habituation effect may
have somewhat decreased the sensitivity to detect BOLD responses in
the cord, the main purpose of this study was to assess the efciency of
CORSICA to model physiological noise in fMRI time series.
Sagittal acquisition
Here we chose to acquire the long-TR and short-TR data in sagittal
orientation at relatively large thickness (3 mm). This resulted in partial
volume effect in the lateral direction. To extract cardiac-related variance
on the short-TR data, this scheme of acquisition appears appropriate,
given that most of the cardiac-related noise occurs in the large blood ves-
sel distributed at the periphery of the spinal cord (in the spinal venous
plexus). We have shown that the distribution of cardiac-related noise
is relatively smooth in space (Piché et al., 2009), therefore partial volume
effect is likely of minimum concern for extracting cardiac-related noise.
However, given that the spinal cord grey matter anatomy is arranged
in laminae in the cross-sectional plane, axial acquisition is often pre-
ferred for extracting the neuronal responses to a given stimulus with
high spatial accuracy (Backes et al., 2001; Brooks et al., 2008;
Cohen-Adad et al., 2010; Giulietti et al., 2008; Maieron et al., 2007;
Stroman and Ryner, 2001; Summers et al., 2010). Here, the reason for ac-
quiring the short-TR and long-TR data in sagittal orientation was to
match the FOV between the two acquisitions, in order to apply the esti-
mated noise mask without further registration, which could have
resulted into mis-registration and re-interpolation errors. However,
overcoming the latter issues wouldmakeitpossibletoacquirethe
Fig. 6. Mean T-score and voxel count for voxels passing the P = 0.01 threshold. Metrics
were averaged in the spinal cord at each vertebral level, from C3 to T1. Error bars show
standard error (SE) across subjects. Two sample Student'st-test paired for subject was
performed to test differences without correction, with CSF regressor and with CORSICA.
*: Pb 0.01; **: Pb 0.001 (corrected for comparisons across conditions and vertebral
levels).
Fig. 7. Signicant voxel counts (mean ± SE) in the ventral spinal cord, dorsal spinal cord and in the CSF surrounding the spinal cord. Voxel count was computed for three statistical
thresholds: P =0.01 (uncorrected), P = 0.001 (uncorrected) and P=0.05 (: corrected with Bonferroni). Two-sample Student'st-test (paired for subject) tested differences between
the three conditions. *: Pb 0.01; **: P b 0.001 (corrected for comparisons across conditions).
250 G. Xie et al. / NeuroImage 63 (2012) 245252
short-TR data in sagittal orientation, and the long-TR data in axial
orientation.
Localization of BOLD signal in response to nociceptive stimuli
BOLD responses were stronger in the dorsal versus in the ventral as-
pect of the spinal cord (see Fig. 7), which is consistent with the termina-
tion of nociceptive peripheral afferent bers in the ipsilateral dorsal
(posterior) horn of the spinal cord (Porro et al., 1991). Regarding the
laterality of the responses, we attempted to separate right and left
sides of the cord in order to assess stronger BOLD responses ipsilateral-
ly. However, the rightleft curvature of the cord associated with rela-
tively large slice thickness which (3 mm) resulted in un-balanced
partial volume effect between the right and the left side, prevented us
to pursue such analysis in a robust manner. According to previous stud-
ies, BOLD signal responses can be present ipsi- and contralaterally dur-
ing sensori and painful stimulations (Giove et al., 2004; Summers et al.,
2010). Contralateral activations could be the result of activity in reex
arcs, intraspinal and projection systems (Coghill et al., 1991)orcould
be false positives.
BOLD responses were mostly detected at C3C4 vertebral levels,
which is consistent with two previous studies employing nociceptive
stimuli of the thumb (Stracke et al., 2005; Stroman, 2006). Stracke et
al. found the strongest activation at C3C4 levels and Stroman et al.
found strongest activation at C2C3 levels. BOLD responses to nocicep-
tive stimuli of the thumb were expected to occur mostly in the spinal
cord at around C5 vertebral level, which roughly corresponds to the po-
sition of spinal level C6. Possible factors contributing to such discrepan-
cy are listed hereafter. Firstly, BOLD signals could have been spread out
across segments since axons in the lateral division of the dorsal-root -
bers split to form Lissauer's tract. These axons may ascend or descend in
Lissauer's tract for at least two spinal segments before entering into the
lateral portion of the dorsal horn to synapse on cells in the dorsal horn
(Murray, 2003). This implies that activation of nociceptive bers from
C6 dermatome may lead to activation of dorsal horn neurons distribut-
ed in spinal segments C4 to C8, extending roughly from vertebral levels
C3 to C7. This argument however does not by itself account for the fact
that there is such a discrepancy in the activated levels between (C3C4
instead of ~C5). Secondly, the stimulus at dermatome C6 can yield syn-
aptic relays via interneurons, thereby spreading BOLD responses across
levels (Stracke et al., 2005) (as clearly seen in Fig. 4). Moreover, the ex-
tension of the activation depends on the intensity of the nociceptive
stimulus (Coghill et al., 1991). Thirdly, the identication of spinal levels
based on vertebral levels may not be accurate (Stroman et al., in press).
Fourthly, the signal-to-noise ratio (SNR) is likely variable across verte-
bral levels, due to the uneven distribution of coil arrays elements. As a
result, the sensitivity to detect BOLD responses in lower segments can
be decreased. The use of highly sensitive coils for fMRI of the spinal
cord can alleviate this limitation (Cohen-Adad et al., 2011). This argu-
ment also raises concerns about the physiological noise to thermal
image noise ratio, which directly relates to the sensitivity to detect
BOLD responses (Triantafyllou et al., 2011). Decreasing the thermal
noise via more sensitive coils would in turn increase the contribution
of physiological noise, in which case post-processing methods such as
CORSICA may be even more useful in increasing the sensitivity to detect
fMRI activations.
Conclusion
One of the most challenging aspects of spinal cord fMRI is the effect
of physiological noise on the detection of the BOLD signal. As the fMRI
community has moved to higher
eld strengths, physiological noise
has become an increasingly important confound limiting the sensitivity
and the specicity of fMRI studies (Liu et al., 2006). It is therefore even
more important to limit this confound via post-processing techniques
such as the one presented here. While several approaches are being
proposed to correct physiological noise, we recommend that these de-
velopments be based on a careful assessment of the spatio-temporal dy-
namics of the noise (Piché et al., 2009), an explicit model of the
physiological signal to be removed (e.g. here based on spectral coher-
ence), a conrmation of the assumptions of the applied correction
model (e.g. here CORSICA assumes spatial stability of physiological
noise, a condition that has been veried by Piché et al.), and a convinc-
ing demonstration that the proposed method improves sensitivity,
specicity, or both. Further developments and in depth validation of
this correction method appear indispensable before spinal fMRI can be
applied more generally as a clinical assessment tool.
Acknowledgments
We thank C. Hurst and A. Cyr for assistance with MRI acquisitions.
We thank the reviewers for their very insightful comments. This work
was supported by the Canadian Institutes of Health Research and the
Quebec Pain Research Network of the Fonds de recherche en santé
du Québec (P.R.), the National MS Society [FG1892A1/1] (J.C.A), the
Canada Research Chair on the Spinal Cord and the SensoriMotor Reha-
bilitation Research Team (SMRRT) of the Canadian Institute of Health
Research (S.R.). G. Xie had a fellowship from the Groupe de Recherche
sur le Système Nerveux Central (GRSNC, Université de Montréal).
References
Backes, W.H., Mess, W.H., Wilmink, J.T., 2001. Functional MR imaging of the cervical
spinal cord by use of median nerve stimulation and st clenching. AJNR Am.
J. Neuroradiol. 22, 18541859.
Becerra, L.R., Breiter, H.C., Stojanovic, M., Fishman, S., Edwards, A., Comite, A.R.,
Gonzalez, R.G., Borsook, D., 1999. Human brain activation under controlled thermal
stimulation and habituation to noxious heat: an fMRI study. Magn. Reson. Med. 41,
10441057.
Behzadi, Y., Restom, K., Liau, J., Liu, T.T., 2007. A component based noise correction
method (CompCor) for BOLD and perfusion based fMRI. Neuroimage 37, 90101.
Bouwman, C.J., Wilmink, J.T., Mess, W.H., Backes, W.H., 2008. Spinal cord functional MRI at 3
T: gradient echo echo-planar imaging versus turbo spin echo. Neuroimage 43, 28829 6.
Brooks, J.C., Beckmann, C.F., Miller, K.L., Wise, R.G., Porro, C.A., Tracey, I., Jenkinson, M.,
2008. Physiological noise modelling for spinal functional magnetic resonance im-
aging studies. Neuroimage 39, 680692.
Calhoun, V.D., Adali, T., Stevens, M.C., Kiehl, K.A., Pekar, J.J., 2005. Semi-blind ICA of
fMRI: a method for utilizing hypothesis-derived time courses in a spatial ICA anal-
ysis. Neuroimage 25, 527538.
Coghill, R.C., Price, D.D., Hayes, R.L., Mayer, D.J., 1991. Spatial distribution of nociceptive
processing in the rat spinal cord. J. Neurophysiol. 65, 133140.
Cohen-Adad, J., Hoge, R.D., Leblond, H., Xie, G., Beaudoin, G., Song, A.W., Krueger, G.,
Doyon, J., Benali, H., Rossignol, S., 2009. Investigations on spinal cord fMRI of cats
under ketamine. Neuroimage 44, 328339.
Cohen-Adad, J., Gauthier, C.J., Brooks, J.C.W., Slessarev, M., Han, J., Fisher, J.A., Rossignol,
S., Hoge, R.D., 2010. BOLD signal responses to controlled hypercapnia in human
spinal cord. Neuroimage 50, 10741084.
Cohen-Adad, J., Mareyam, A., Keil, B., Polimeni, J.R., Wald, L.L., 2011. 32-channel RF coil op-
timized for brain and cervical spinal cord at 3 T. Magn. Reson. Med. 66, 11981208.
Deckers, R.H., van Gelderen, P., Ries, M., Barret, O., Duyn, J.H., Ikonomidou, V.N., Fukunaga,
M., Glover, G.H., de Zwart, J.A., 2006. An adaptive lter for suppression of cardiac and
respiratory noise in MRI time series data. Neuroimage 33, 10721081.
Eippert, F., Finsterbusch, J., Bingel, U., Büchel, C., 2009. Direct evidence for spinal cord
involvement in placebo analgesia. Science 326, 404.
Endo, T., Spenger, C., Westman, E., Tominaga, T., Olson, L., 2008. Reorganization of sen-
sory processing below the level of spinal cord injury as revealed by fMRI. Exp.
Neurol. 209, 155160.
Figley, C.R., Stroman, P.W., 2007. Investigation of human cervical and upper thoracic
spinal cord motion: implications for imaging spinal cord structure and function.
Magn. Reson. Med. 58, 185189.
Figley, C.R., Stroman, P.W., 2009. Development and validation of retrospective spinal
cord motion time-course estimates (RESPITE) for spin-echo spinal fMRI: improved
sensitivity and specicity by means of a motion-compensating general linear
model analysis. Neuroimage 44, 421427.
Giove, F., Garreffa, G., Giulietti, G., Mangia, S., Colonnese, C., Maraviglia, B., 2004. Issues
about the fMRI of the human spinal cord. Magn. Reson. Imaging 22, 15051516.
Giulietti, G., Giove, F., Garreffa, G., Colonnese, C., Mangia, S., Maraviglia, B., 2008. Char-
acterization of the functional response in the human spinal cord: impulse-
response function and linearity. Neuroimage 42, 626634.
Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction of
physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 44, 162167.
Govers, N., Beghin, J., Van Goethem, J.W., Michiels, J., van den Hauwe, L., Vandervliet, E.,
Parizel, P.M., 2007. Functional MRI of the cervical spinal cord on 1.5 T with
ngertapping: to what extent is it feasible? Neuroradiology 49, 7381.
251G. Xie et al. / NeuroImage 63 (2012) 245252
Guimaraes, A.R., Melcher, J.R., Talavage, T.M., Baker, J.R., Ledden, P., Rosen, B.R., Kiang,
N.Y.S., Fullerton, B.C., Weisskoff, R.M., 1998. Imaging subcortical auditory activity
in humans. Hum. Brain Mapp. 6, 33 41.
Hu, D., Yan, L., Liu, Y., Zhou, Z., Friston, K.J., Tan, C., Wu, D., 2005. Unied SPM-ICA for
fMRI analysis. Neuroimage 25, 746755.
Kong, Y., Jenkinson, M., Andersson, J., Tracey, I., Brooks, J.C.W., 2012. Assessment of
physiological noise modelling methods for functional imaging of the spinal cord.
Neuroimage 60, 15381549.
Lawrence, J., Stroman, P.W., Malisza, K.L., 2007. Comparison of functional activity in the
rat cervical spinal cord during alpha-chloralose and halothane anesthesia.
Neuroimage 34, 16651672.
Lilja, J., Endo, T., Hofstetter, C., Westman, E., Young, J., Olson, L., Spenger, C., 2006. Blood
oxygenation level-dependent visualization of synaptic relay stations of sensory
pathways along the neuroaxis in response to graded sensory stimulation of a
limb. J. Neurosci. 26, 63306336.
Liu, C.S., Miki, A., Hulvershorn, J., Bloy, L., Gualtieri, E.E., Liu, G.T., Leigh, J.S., Haselgrove,
J.C., Elliott, M.A., 2006. Spatial and temporal characteristics of physiological noise in
fMRI at 3T. Acad. Radiol. 13, 313323.
Lund, T., Madsen, K., Sidaros, K., Luo, W.-L., Nichols, T., 2006. Non-white noise in fMRI:
does modelling have an impact? Neuroimage 29, 5466.
Madi, S., Flanders, A.E., Vinitski, S., Herbison, G.J., Nissanov, J., 2001. Functional MR im-
aging of the human cervical spinal cord. AJNR Am. J. Neuroradiol. 22, 17681774.
Maieron, M., Iannetti, G.D., Bodurka, J., Tracey, I., Bandettini, P.A., Porro, C.A., 2007.
Functional responses in the human spinal cord during willed motor actions: evi-
dence for side- and rate-dependent activity. J. Neurosci. 27, 4182 4190.
Majcher, K., Tomanek, B., Tuor, U.I., Jasinski, A., Foniok, T., Rushforth, D., Hess, G., 2007.
Functional magnetic resonance imaging within the rat spinal cord following pe-
ripheral nerve injury. Neuroimage 38, 669676.
McKeown, M.J., Sejnowski, T.J., 1998. Independent component analysis of fMRI data:
examining the assumptions. Hum. Brain Mapp. 6, 368372.
Murray, M., 2003. Organization of the spinal cord, In: Conn, P.M. (Ed.), Neuroscience in
Medicine, 2e ed. Human Press Inc., Totowa, NJ, pp. 197211.
Otsu, N., 1979. A threshold selection method from gray level histograms. IEEE Trans.
Syst. Man Cybern. 9, 62 66.
Perlbarg, V., Bellec, P., Anton, J.L., Pelegrini-Issac, M., Doyon, J., Benali, H., 2007.
CORSICA: correction of structured noise in fMRI by automatic identication of
ICA components. Magn. Reson. Imaging 25, 3546.
Piché, M., Cohen-Adad, J., Nejad, M.K., Perlbarg, V., Xie, G., Beaudoin, G., Benali, H.,
Rainville, P., 2009. Characterization of cardiac-related noise in fMRI of the cervical
spinal cord. Magn. Reson. Imaging 27, 300 310.
Porro, C.A., Cavazzuti, M., Galetti, A., Sassatelli, L., Barbieri, G.C., 1991. Functional activ-
ity mapping of the rat spinal cord during formalin-induced noxious stimulation.
Neuroscience 41, 655665.
Schrouff, J., Perlbarg, V., Boly, M., Marrelec, G., Boveroux, P., Vanhaudenhuyse, A.,
Bruno, M.A., Laureys, S., Phillips, C., Pelegrini-Issac, M., Maquet, P., Benali, H.,
2011. Brain functional integration decreases during propofol-induced loss of
consciousness. Neuroimage 57, 198205.
Stracke, C.P., Pettersson, L.G., Schoth, F., Moller-Hartmann, W., Krings, T., 2005. Inter-
neuronal systems of the cervical spinal cord assessed with BOLD imaging at 1.5
T. Neuroradiology 47, 127133.
Stroman, P.W., 2005. Magnetic resonance imaging of neuronal function in the spinal
cord: spinal fMRI. Clin. Med. Res. 3, 146156.
Stroman, P.W., 2006. Discrimination of errors from neuronal activity in functional MRI
of the human spinal cord by means of general linear model analysis. Magn. Reson.
Med. 56, 452456.
Stroman, P.W., Ryner, L.N., 2001. Functional MRI of motor and sensory activation in the
human spinal cord. Magn. Reson. Imaging 19, 2732.
Stroman, P.W., Nance, P.W., Ryner, L.N., 1999. BOLD MRI of the human cervical spinal
cord at 3 tesla. Magn. Reson. Med. 42, 571576.
Stroman, P.W., Kornelsen, J., Bergman, A., Krause, V., Ethans, K., Malisza, K.L., Tomanek,
B., 2004. Noninvasive assessment of the injured human spinal cord by means of
functional magnetic resonance imaging. Spinal Cord 42, 5966.
Stroman, P.W., Kornelsen, J., Lawrence, J., 2005. An improved method for spinal func-
tional MRI with large vol ume coverage of the spinal cord. J. Magn. Reson. Imaging
21, 520526.
Stroman, P.W., Bosma, R.L., Tsyben, A., in press. Somatotopic arrangement of thermal
sensory regions in the healthy human spinal cord determined by means of spinal
cord functional MRI. Magn. Reson. Med http://dx/doi.org/10.1002/mrm.23292.
Summers, P.E., Ferraro, D., Duzzi, D., Lui, F., Iannetti, G.D., Porro, C.A., 2010. A quantita-
tive comparison of BOLD fMRI responses to noxious and innocuous stimuli in the
human spinal cord. Neuroimage 50, 14081415.
Thomas, C.G., Harshman, R.A., Menon, R.S., 2002. Noise reduction in BOLD-based fMRI
using component analysis. Neuroimage 17, 15211537.
Tohka, J., Foerde, K., Aron, A.R., Tom, S.M., Toga, A.W., Poldrack, R.A., 2008. Automatic inde-
pendent component labeling for artifact removal in fMRI. Neuroimage 39, 12271245.
Tousignant-Laamme, Y., Rainville, P., Marchand, S., 2005. Establishing a link between
heart rate and pain in healthy subjects: a gender effect. J. Pain 6, 341347.
Triantafyllou, C., Polimeni, J.R., Wald, L.L., 2011. Physiological noise and signal-to-noise
ratio in fMRI with multi-channel array coils. Neuroimage 55, 597606.
Valsasina, P., Agosta, F., Caputo, D., Stroman, P.W., Filippi, M., 2008. Spinal fMRI during
proprioceptive and tactile tasks in healthy subjects: activity detected using
cross-correlation, general linear model and independent component analysis. Neu-
roradiology 50 (10), 895902.
Vanhaudenhuyse, A., Noirhomme, Q., Tshibanda, L.J., Bruno, M.A., Boveroux, P.,
Schnakers, C., Soddu, A., Perlbarg, V., Ledoux, D., Brichant, J.F., Moonen, G.,
Maquet, P., Greicius, M.D., Laureys, S., Boly, M., 2010. Default network connectivity
reects the level of consciousness in non-communicative brain-damaged patients.
Brain 133, 161171.
Yoshizawa, T., Nose, T., Moore, G.J., Sillerud, L.O., 1996. Functional magnetic resonance im-
aging of motor activation in the human cervical spinal cord. Neuroimage 4, 174182.
Zhao, F., Williams, M., Meng, X., Welsh, D.C., Coimbra, A., Crown, E.D., Cook, J.J., Urban,
M.O., Hargreaves, R., Williams, D.S., 2008. BOLD and blood volume-weighted fMRI
of rat lumbar spinal cord during non-noxious and noxious electrical hindpaw stim-
ulation. Neuroimage 40, 133147.
252 G. Xie et al. / NeuroImage 63 (2012) 245252
    • "Because of the subject-specific prescription of axial slices and slice gap-size adjustments, the cervical EPI slices were already aligned along the rostrocaudal axis across subjects. As physiological noise has a major impact in the detection of spinal cord BOLD signal [26] , we then used independent component analysis to identify and account for noise components at the cervical cord level [56]. Thirty components were extracted for each subject, which accounted for about 95% of the BOLD signal variability. "
    [Show abstract] [Hide abstract] ABSTRACT: The spinal cord participates in the execution of skilled movements by translating high-level cerebral motor representations into musculotopic commands. Yet, the extent to which motor skill acquisition relies on intrinsic spinal cord processes remains unknown. To date, attempts to address this question were limited by difficulties in separating spinal local effects from supraspinal influences through traditional electrophysiological and neuroimaging methods. Here, for the first time, we provide evidence for local learning-induced plasticity in intact human spinal cord through simultaneous functional magnetic resonance imaging of the brain and spinal cord during motor sequence learning. Specifically, we show learning-related modulation of activity in the C6-C8 spinal region, which is independent from that of related supraspinal sensorimotor structures. Moreover, a brain-spinal cord functional connectivity analysis demonstrates that the initial linear relationship between the spinal cord and sensorimotor cortex gradually fades away over the course of motor sequence learning, while the connectivity between spinal activity and cerebellum gains strength. These data suggest that the spinal cord not only constitutes an active functional component of the human motor learning network but also contributes distinctively from the brain to the learning process. The present findings open new avenues for rehabilitation of patients with spinal cord injuries, as they demonstrate that this part of the central nervous system is much more plastic than assumed before. Yet, the neurophysiological mechanisms underlying this intrinsic functional plasticity in the spinal cord warrant further investigations.
    Full-text · Article · Jun 2015
    • "In each subject, temporal signal-to-noise ratio (TSNR) was measured in spinal gray matter upon completion of the functional-to-anatomical affine registration (step #9) as well as after the application of CSF and white matter 'regressors of no interest' (steps #11 and #12). Across all 22 subjects, we observed a 30% increase in median TSNR (from 29.3 to 38.1) after the application of these few regressors, demonstrating the importance of characterizing and removing structured noise sources (Xie et al., 2012). After band-pass filtering to isolate the frequency range of interest (0.01–0.08 Hz), single-subject analyses show that statistically significant correlations are measurable between selected regions and are reproducible across subjects. "
    [Show abstract] [Hide abstract] ABSTRACT: Functional magnetic resonance imaging using blood oxygenation level dependent (BOLD) contrast is well established as one of the most powerful methods for mapping human brain function. Numerous studies have measured how low-frequency BOLD signal fluctuations from the brain are correlated between voxels in a resting state, and have exploited these signals to infer functional connectivity within specific neural circuits. However, to date there have been no previous substantiated reports of resting state correlations in the spinal cord. In a cohort of healthy volunteers, we observed robust functional connectivity between left and right ventral (motor) horns, and between left and right dorsal (sensory) horns. Our results demonstrate that low-frequency BOLD fluctuations are inherent in the spinal cord as well as the brain, and by analogy to cortical circuits, we hypothesize that these correlations may offer insight into the execution and maintenance of sensory and motor functions both locally and within the cerebrum.
    Full-text · Article · Aug 2014
    • "ICA approaches offer the possibility to automatically identify sources on physiological noise, however, their usefulness must be judged against the possibility of true signal being identified as a noise source and removed from the data. One approach to mitigate this outcome is to acquire resting and task fMRI data, and use the resting data to define physiological noise components, for subsequent removal from the task data (e.g., Xie et al., 2012). Concerning the model-based techniques, such as RETROICOR (Hu et al., 1995; Josephs et al., 1997; Glover et al., 2000) and PNM (Brooks et al., 2008; Harvey et al., 2008) these will always be subject to the limitations of the basis functions used to describe the physiological processes they are attempting to model. "
    [Show abstract] [Hide abstract] ABSTRACT: The brainstem is directly involved in controlling blood pressure, respiration, sleep/wake cycles, pain modulation, motor, and cardiac output. As such it is of significant basic science and clinical interest. However, the brainstem's location close to major arteries and adjacent pulsatile cerebrospinal fluid filled spaces, means that it is difficult to reliably record functional magnetic resonance imaging (fMRI) data from. These physiological sources of noise generate time varying signals in fMRI data, which if left uncorrected can obscure signals of interest. In this Methods Article we will provide a practical introduction to the techniques used to correct for the presence of physiological noise in time series fMRI data. Techniques based on independent measurement of the cardiac and respiratory cycles, such as retrospective image correction (RETROICOR, Glover et al., 2000), will be described and their application and limitations discussed. The impact of a physiological noise model, implemented in the framework of the general linear model, on resting fMRI data acquired at 3 and 7 T is presented. Data driven approaches based such as independent component analysis (ICA) are described. MR acquisition strategies that attempt to either minimize the influence of physiological fluctuations on recorded fMRI data, or provide additional information to correct for their presence, will be mentioned. General advice on modeling noise sources, and its effect on statistical inference via loss of degrees of freedom, and non-orthogonality of regressors, is given. Lastly, different strategies for assessing the benefit of different approaches to physiological noise modeling are presented.
    Full-text · Article · Oct 2013
Show more