fMRI resting state networks define distinct modes of long-distance
interactions in the human brain
M. De Luca,a,b,*C.F. Beckmann,aN. De Stefano,bP.M. Matthews,aand S.M. Smitha
aOxford Centre for Functional Magnetic Resonance Imaging of the Brain, UK
bInstitute of Neurological Science, University of Siena, Italy
Received 26 May 2005; revised 16 August 2005; accepted 25 August 2005
Available online 2 November 2005
Functional magnetic resonance imaging (fMRI) studies of the human
brain have suggested that low-frequency fluctuations in resting fMRI
data collected using blood oxygen level dependent (BOLD) contrast
correspond to functionally relevant resting state networks (RSNs).
Whether the fluctuations of resting fMRI signal in RSNs are a direct
consequence of neocortical neuronal activity or are low-frequency
artifacts due to other physiological processes (e.g., autonomically
driven fluctuations in cerebral blood flow) is uncertain. In order to
investigate further these fluctuations, we have characterized their
spatial and temporal properties using probabilistic independent
component analysis (PICA), a robust approach to RSN identification.
Here, we provide evidence that: i. RSNs are not caused by signal
artifacts due to low sampling rate (aliasing); ii. they are localized
primarily to the cerebral cortex; iii. similar RSNs also can be identified
in perfusion fMRI data; and iv. at least 5 distinct RSN patterns are
reproducible across different subjects. The RSNs appear to reflect
‘‘default’’ interactions related to functional networks related to those
recruited by specific types of cognitive processes. RSNs are a major
source of non-modeled signal in BOLD fMRI data, so a full under-
standing of their dynamics will improve the interpretation of functional
brain imaging studies more generally. Because RSNs reflect interac-
tions in cognitively relevant functional networks, they offer a new
approach to the characterization of state changes with pathology and
the effects of drugs.
D 2005 Elsevier Inc. All rights reserved.
Keywords: Functional MRI; Brain activation; Resting state; Independent
component analysis; Functional connectivity; Resting state networks;
PICA; Perfusion fMRI
The functioning of the human brain during rest can be
investigated using different functional imaging techniques (Biswal
et al., 1995; Shulman et al., 1997; Gusnard and Raichle, 2001).
While the resting state is an ill-defined condition, consistent
functional patterns across individuals should represent common
‘‘default’’ or ‘‘idling’’state activity. Long-range coherences in these
activities therefore could reflect strong functional connectivities.
fMRI images obtained using blood oxygen level dependent
(BOLD) contrast show signal fluctuations at rest. These fluctua-
tions occur at low frequencies (0.01–0.05 Hz) and have been
shown to be coherent across widely separated (although function-
ally related) brain regions (e.g., bihemispheric sensorimotor
cortices) (Biswal et al., 1995; Lowe et al., 1998; Cordes et al.,
2000). Regions showing coherent fluctuations therefore constitute
a ‘‘resting state network’’ (RSN). We and others have appreciated
that there is more than one spatially distinct RSN in a resting brain
image dataset, with each RSN having a distinct signal time-course
(De Luca et al., 2002; Greicius et al., 2003).
Whether the fluctuations of resting fMRI signal in RSNs are a
direct consequence of neuronal activity or whether they reflect
phenomena such as cardio-respiratory motion or vascular modu-
lation is uncertain. The normally low sampling rate of fMRI
images (Jezzard et al., 2002) causes temporal aliasing of variations
of the BOLD fMRI signal induced by cardiac and respiratory
cycles into a low-frequency range, similar to that of the RSN signal
fluctuations. Some low-frequency coherences in resting BOLD
fMRI data are clearly a consequence of this physiological noise
(Lowe et al., 1998; Xiong et al., 1999; Cordes et al., 2000).
However, studies conducted in ways that avoid aliasing of the
fMRI signal (using a fast image sampling rate) show that many
low-frequency coherences are still present, suggesting that RSNs
and (higher frequency) physiological noise are phenomenologi-
cally distinct processes (Biswal et al., 1995; Lowe et al., 1998).
Additional patterns related directly to vascular processes inde-
pendent of cortical neuronal function have been identified as low-
frequency fluctuations in resting fMRI data (Kiviniemi et al., 2000;
Wise et al., 2004). The most direct data relating some patterns of
1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved.
* Corresponding author. FMRIB Centre, University of Oxford, John
Radcliffe Hospital, Headley Way, Headington, Oxford OX3 9DU, UK. Fax:
+44 1865 222717.
E-mail address: firstname.lastname@example.org (M. De Luca).
Available online on ScienceDirect (www.sciencedirect.com).
NeuroImage 29 (2006) 1359 – 1367
low-frequency coherence in fMRI data to neuronal activity come
from evidence that the underlying fluctuations are correlated with
modulations of cortical electrical activity detected by EEG (Gold-
man et al., 2002; Leopold et al., 2003; Moosmann et al., 2003;
Laufs et al., 2003). The observation of changes in patterns with
neurological disease (e.g., Alzheimer’s disease; Greicius et al.,
2004) is consistent with this.
An important concern in studying RSNs is whether the method
used for their identification is appropriately sensitive, yet relatively
unbiased. Methods based on direct correlations with time-courses
of signal change identified from a ‘‘seed’’ voxel are limited to
applications to regions for which there is an a priori expectation of
a network pattern.
Here, we have applied probabilistic ICA (PICA) to the
characterization of RSNs in resting brain BOLD contrast datasets.
We have made a series of observations designed to test: i. the
independence of PICA-defined RSNs from artifacts related to
cardio-respiratory motion; ii. the localization of potential gener-
ators of RSNs; iii. the relation of BOLD RSNs to coherences
defined with perfusion imaging; iv. the reproducibility of RSNs
across subjects; and v. the specific patterns of coherent activity
across the brain.
All fMRI data were acquired from healthy volunteers (age
range 22–51 years). In all experiments, subjects were at rest; they
were instructed to relax with their eyes closed, without falling
asleep, as confirmed by the subjects after completion of the
experiment. MRI data were acquired on a 3 T Varian/Siemens MRI
system at the Oxford Centre for Functional Magnetic Imaging of
the Brain, except the data of experiment 1, which were collected on
a 1.5 T Philips Gyroscan MRI system at the NMR Centre of the
University of Siena. Temporal and spatial resolutions of fMRI data
varied across the experiments; they are detailed in the following
All data were first pre-processed using tools from the FMRIB
Software Library (FSL, http://www.fmrib.ox.ac.uk/fsl) (Smith et
al., 2004), applying the following procedures: motion correction
(Jenkinson et al., 2002), spatial smoothing using a Gaussian kernel
of FWHM 5 mm, mean-based intensity normalization of all
volumes by the same factor, and high-pass temporal filtering
(Gaussian-weighted least-squares straight line fitting, with high-
pass filter cut-off of 250s). Following the pre-processing, the data
were analyzed using MELODIC (Multivariate Exploratory Linear
Optimised Decomposition into Independent Components), an
implementation of probabilistic independent component analysis
(PICA) (Beckmann and Smith, 2004), also part of FSL.
Independent component analysis (ICA) is becoming a popular
exploratory method for analyzing complex data such as that from
fMRI experiments. ICA views the 4D data as a sum of a set of
spatiotemporal components, each of which consists of a spatial
map modulated in time by that component’s associated time-
course. It attempts to separate the different components by making
the assumption that the spatial maps are statistically independent of
each other, and, having different time-courses, they will ideally
each represent a different artefact or activation pattern. By using
the entire 4D dataset at once in this multivariate analysis, this kind
of approach does not need to be fed any temporal model. In
attempting to find RSNs in fMRI data, it is preferable to use a
methodology that does not require the additional experimental
sessions, extra analysis steps, and potential bias associated with
The application of ‘‘model-free’’ methods such as ICA,
however, has previously been restricted both by the view that
results can be hard to interpret, and by the lack of ability to
quantify statistical significance for estimated spatial maps. Beck-
mann and Smith (2004) proposed a probabilistic ICA (PICA)
model for fMRI which models the observations as mixtures of
spatially non-Gaussian signals and artefacts in the presence of
Gaussian noise. It was demonstrated in the same work that using
an objective estimation of the amount of Gaussian noise through
Bayesian analysis of the number of activation and (non-Gaussian)
noise sources, the problem of overfitting can be overcome. The
approach proposed for estimating a suitable model order (i.e., how
many ICA components to find) also allows for a unique
decomposition of the data and reduces problems of interpretation
as each final component is more likely to be due to only one
physical or physiological process.
The objective of this experiment was to assess if aliasing of
physiological processes (cardiac and respiratory cycles) is distin-
guishable from ‘‘true’’ RSNs observed in fMRI data. The cardiac
and respiratory cycles occur around 1 Hz and 0.3 Hz respectively.
Consequently, these can become aliased at typical TRs (2–3 s),
giving significant power at the frequencies typical of RSNs.
Because at low TRs (below 125 ms) such aliasing is avoided, in
this experiment, BOLD fMRI data were collected using a very
short TR (120 ms); see also Lowe et al. (1998). In addition, longer
TR (3 s) BOLD fMRI data were collected and compared to the
results from the low TR data. This was in order to test (via spatial
comparison of PICA components) whether RSN signal is distinct
(and distinguishable) from aliased signal changes related to the
cardiac or respiratory cycles.
Two BOLD echo planar imaging fMRI datasets were collected
from a single subject at 3 T, with the following parameters. In the
first dataset (long TR), three axial slices covering the motor cortex
(in-plane resolution 3.75 ? 3.75 mm, slice thickness 7 mm, no gap,
TR = 3000 ms, TE = 30 ms, 200 volumes) were collected during
bilateral finger tapping (30s ON–OFF paradigm). In the second
dataset (short TR), one single slice covering the motor cortex (in-
plane resolution 3.75 ? 3.75 mm, slice thickness 7 mm, TR = 120
ms, TE = 30 ms, 2200 volumes) was collected during rest.
The comparison of the maps obtained from the two experiments
was performed by means of spatial correlation coefficients. In
Fig. 1. Resting state networks and the aliasing problem. (A and B) BOLD fMRI data collected with high temporal resolution (TR = 120 ms). Spatial
distribution of probabilistic independent components and their relative power spectra respectively for (A) a resting state network in the motor cortex and (B)
cardiac fluctuation (with harmonics). PICA separates the physiological noise induced by the cardiac cycle from the RSN. (C and D) BOLD fMRI data collected
with low (more typical) temporal resolution (TR = 3000 ms). Spatial distribution of probabilistic independent components and their relative power spectra for
resting state networks (C) and aliased cardiac-related artifact (D). PICA can separate the aliased physiological noise induced by the cardiac cycle from the RSN.
All the maps are thresholded with alternative hypothesis, P > 0.5.
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
addition, an estimation of the aliased frequency, at different TRs, of
the fundamental frequency, was also carried out for the cardiac and
The objective of this experiment was to address the issue of
spatial localization of RSNs, specifically their localization with
respect to gray matter localization. BOLD fMRI images with
relatively high spatial resolution, compared to a typical fMRI
resolution (such as 4 ? 4 ? 7 mm), were collected to
investigate whether RSNs are localized within gray matter.
Two BOLD fMRI datasets were collected from two subjects
at 3 T during rest. In the first subject, the following parameters
were employed: 12 axial slices, in-plane resolution 2 ? 2 mm,
slice thickness 6 mm, no gap, TR = 3000 ms, TE = 40 ms,
300 volumes. In the second subject, the following parameters
were employed: 30 axial slices, in-plane resolution 2 ? 1.5
mm, slice thickness 1.75 mm, no gap, TR = 10 s, TE = 40 ms,
200 volumes. The parameters were optimized to achieve high
spatial resolution and are different in the two subjects as we
varied the balance between resolution and signal-to-noise in the
More than one mechanism could contribute to the origin of
these signals. If low-frequency coherences localized to gray matter
arise from neuronal activity, then we hypothesize that they should
also be reflected as local increases in blood flow. To test for such
coherences specifically in cerebral blood flow changes across the
brain, arterial spin-labeling perfusion imaging was used to acquire
serial images of the resting brain.
For this purpose, we acquired resting ASL (arterial spin
labeling) perfusion fMRI data (Kwong et al., 1992; Biswal et
al., 1997). The ASL contrast mechanism is purely sensitive to
blood flow, as opposed to BOLD fMRI, which is also sensitive
to local oxygenation. Three perfusion fMRI datasets were
collected from one single subject at rest, with the following
parameters: 5 axial slices (totaling 15 slices covering whole
brain), in-plane resolution 4 ? 4 mm, slice thickness 6 mm, no
gap, TR = 2000 ms, TE = 20 ms, TI = 1400 ms, 200 volumes.
Previously, an RSN in the motor cortex was found in ASL data
using a seeding approach (Biswal et al., 1997). As with the
BOLD contrast, we applied PICA to the resting ASL data to
define spatiotemporal networks, enabling us to look for multiple
independent RSNs (if present) without the need for prespecifi-
cation of the number of RSNs expected or selection of a seed
The objective of this experiment was to investigate the spatial
reproducibility of the RSNs across different subjects. Spatial
reproducibility was assessed through spatial correlation of the
RSN maps. Whole-brain BOLD fMRI datasets were collected from
10 subjects at 3 T, during rest, using the following parameters: 45
axial slices, in-plane resolution 3 ? 3 mm, slice thickness 3 mm, no
gap, TR = 3400 ms, TE = 40 ms, 200 volumes.
After the separate single-subject PICA analyses, in order to
combine the results from different subjects, RSN maps were first
aligned to the subjects’ structural images and then into a standard
(MNI152) space. They were then smoothed using a 5 mm FWHM
Gaussian kernel. This was carried out to reduce the effect of
structural differences between subjects (i.e. equivalent to the
smoothing often applied in standard multi-subject fMRI experi-
ments) (Jezzard et al., 2002, Chapter 14).
Spatial consistency between different RSN maps was quantified
by finding the (spatial) normalized correlation coefficient of each
map from one subject with each map of another subject. The
correlations were thresholded at 0.15, corresponding to a proba-
bility level P < 0.00015.
RSN maps that were spatially consistent across all subjects
were detected by looking for consistent sets of pair-wise
correlations between all subjects (in all the directions). In other
words, let us suppose we had only three subjects. If map j of group
1 (where group indicates the set of maps obtained from a PICA
decomposition of one subject’s data) was correlated with map i of
group 2 and with map k of group 3, in order to declare these maps
consistent, we had to verify that map k was also correlated with
map j. This was not always true since we were dealing with
thresholded correlation coefficients.
After identifying spatial maps that are consistent across
subjects, we then created group maps using a fixed-effects analysis.
For inference, we then ran mixture-modeling (alternative hypoth-
esis testing, thresholded at P > 0.5 for ‘‘activation’’ vs. null
(Beckmann and Smith, 2004) to create the thresholded results for
each group-level RSN.
Finally, we tested whether diffusion-derived anatomical (white
matter) connectivity supports the found networks. We used a
probabilistic representation of thalamic nuclei derived from
diffusion tensor data (Johansen-Berg et al., 2005, http://
www.fmrib.ox.ac.uk/connect) to test whether RSN peaks lying in
the thalamus were both functionally connected to particular cortical
areas in the RSN maps (i.e., part of the same RSN) and
anatomically connected in the diffusion atlas to these same cortical
Characterization of spatiotemporally distinct patterns of coherent
signals in BOLD and ASL images from the unstimulated brain:
resting state networks
PICA applied to a time series of echo planar brain images
acquired from a subject at rest using either typical (3 s) or short
(0.12 s) TR generates a series of spatiotemporally distinct patterns
of coherent signal changes defined by BOLD fMRI (Fig. 1).
Coherent RSN patterns can be identified having most power at
very low frequencies (0.01–0.05 Hz, Figs. 1A, C). These are
spatially very similar at both long and typical TR (compare also
with Figs. 3–5) and are also temporally very similar and having
the characteristic power spectrum of RSNs. A cardiac-related
component can be clearly seen in Fig. 1B, where the (temporal)
power spectrum peaks at the expected frequency of approximately
1 Hz. The associated spatial map has very strong similarity to Fig.
1D—suggesting that PICA has successfully identified the cardiac
component even in the typical TR (where the time-course is
aliased, as can be seen in the power spectrum of Fig. 1D) and
successfully separated the RSN from the cardiac component. In
these datasets, the only robust physiological components found
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
were the RSNs and the cardiac pulsation—that is, we did not find
strong signal relating to respiration.
The apparent anatomical co-localization of low-frequency RSN
coherences with gray matter were confirmed using higher-
resolution fMRI. In both higher-resolution datasets, patterns were
seen which corresponded (spatially) very well with the more
typical resolutions acquired. For example, Fig. 2 shows an RSN
spatial component resulting from PICA applied to the 2 ? 2 mm
in-plane resolution data. This has clear spatial similarity to the
maps shown in Figs. 4 and 5, and it can be seen that the voxels
involved in the RSN do indeed lie within gray matter. The 2 ?
1.5 ? 1.75 mm data gave a similar general spatial pattern, but the
greatly reduced voxel size in this dataset resulted in much noisier
and less interpretable results.
The ASL perfusion data showed low-frequency coherences in a
pattern similar to that found with BOLD contrast (Fig. 3). PICA
performed on perfusion fMRI resting data disclosed five inde-
pendent component (ICs) whose spatial and temporal character-
istics strongly matched the RSNs observed in BOLD fMRI resting
data (compare with Figs. 4 and 5).
Resting state networks are found consistently across subjects and
define functional–anatomically related regions in the brain
If RSNs define ‘‘default’’ states of coherent activity across the
brain, then they should be reproducible between healthy, alert
individuals. The number of components extracted by ICA from
each subject varied from 42 to 67. This included scanner-related
artefacts such as EPI ghosting and physiological artefacts such as
cardiac pulsation. Spatial cross-correlation showed that a consistent
set of five spatiotemporally distinct patterns was identified for all
10 subjects studied. Fig. 4 illustrates the five RSNs found with a
typical single subject’s dataset.
To understand the anatomical relations of these resting fluctua-
tions, the 5 spatiotemporally distinct RSNs from the 10 different
using PICA. The resulting group average RSN maps confirmed
of the group average networks together emphasizes the comple-
mentary patterns of activation.
The coordinates of maxima in each activation cluster defined in
the group maps were used to localize functional–anatomical
regions attributed to the RSN. Individual RSNs then were
classified spatially both on the basis of coordinates in standard
space (Table 1) and by regional anatomy:
1. RSN1: a posterior network characterized by involvement predo-
minantly of occipital cortex, as well as temporal–parietal regions;
2. RSN2: a posterior–lateral and midline network involving
primarily the precuneus and anterior pole of the prefrontal
lobe, as well as parietal regions.
3. RSN3: a lateral and midline network including the pre- and
post-central gyri, as well as midline regions including the
thalamus and hippocampus.
Fig. 2. Cortical localization of resting state networks. In relatively high-resolution BOLD fMRI datasets (in-plane resolution of 2 ? 2 mm), RSN components
are localized in cortical gray matter. The green lines represent the gray–white matter border in the selected area, after manually segmenting the image.
Fig. 3. Resting state networks are identifiable in perfusion fMRI data. PICA performed on perfusion fMRI resting data disclosed five independent component
(ICs) whose spatial and temporal characteristics strongly matched the RSNs observed in BOLD fMRI resting data (compare with Fig. 5).
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
4. RSN4: a network involving dorsal parietal and predominantly
lateral prefrontal cortex.
5. RSN5: a ventral network dominated by coherences between the
inferior occipital parietal, temporal, and inferior prefrontal cortices.
Using the coordinates in Table 1, we were able to test more
specifically the relationship between the anatomy of signal
correlations defined in the RSNs and anatomical connectivity based
on our prior definition of thalamo-cortical pathways using diffusion
tensor imaging (Johansen-Berg et al., 2005). Using a probabilistic
representation of thenormal humanthalamus definedon thebasis of
white matter connectivity to cortical regions, the anatomical
relations between the thalamic activation cluster in RSNs 2 and 3
(Fig.6) andthecortical regionswereexplored. Thethalamic peakof
coherence from the group RSN2 map corresponds to a region in the
probabilistic thalamic atlas (Johansen-Berg et al., 2005, http://
www.fmrib.ox.ac.uk/connect) with strongest connectivity to the
in this RSN. By contrast, the anatomical localization of the thalamic
cluster in RSN3 (Fig. 6) corresponds to a region that connects most
anatomically strongly with motor cortex. Together, these results are
consistent with a correspondence between regions showing RSN
coherence and those with strong anatomical connectivities.
Several previous reports have described specific patterns of
low-frequency coherent signal in time series of gradient echo MRI
from unstimulated brain (the brain ‘‘at rest’’). The most commonly
recognized pattern includes particularly the sensorimotor cortex
bilaterally (corresponding to RSN 3 defined here) (Biswal et al.,
Fig. 4. Consistently identified resting state networks. Five RSNs from a
single subject illustrating those found consistently from all subjects are
shown. Maps are thresholded at P > 0.5 (alternative hypothesis threshold,
for activation versus null). Each row represents the three most interesting
slices of one distinct RSN. The RSNs are shown on the corresponding
structural image transformed into standard space.
Fig. 5. Group resting state network maps. From top to bottom: (1) RSN 1
including visual cortical areas. The RSN reported here includes the main
visual functional network. (2) RSN 2 including visuospatial and executive
system. The RSN reported here includes the emotion/visuospatial
processing functional network. (3) RSN 3 including sensory and auditory
system. (4) RSN 4 including the dorsal pathway. (5) RSN 5 including
ventral pathway. The crosses indicate the positions of the centers of the
major clusters, and the corresponding coordinates are reported in Table 1
for each corresponding map (map obtained with alternative hypothesis
threshold P > 0.5).
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
1995; Lowe et al., 1998; Xiong et al., 1999). Other work
characterized a predominantly occipital network (corresponding
to RSN1 here) (Goldman and Cohen, 2003; Moosmann et al.,
2003). Our study extends this description, using a relatively
unbiased approach to analysis based on probabilistic ICA
(Beckmann and Smith, 2004). We have identified 5 spatiotempor-
ally distinct patterns of low-frequency coherences across the brain.
The PICA method clearly distinguishes these patterns of activity
from those associated with cardio-respiratory motion of the brain,
even without sampling that is rapid with respect to the primary
frequencies of these processes. We additionally have provided
evidence for cortical localization of these coherences and for
similar patterns associated with changes in local blood flow,
consistent with the neuronal origin of the signals.
Our analysis using PICA found multiple patterns of coherence
involving distinct functional–anatomical networks across the
brain. While the coordinated neuronal activity that we infer is
reflected in these hemodynamic changes may have a specific
processing function, at this point, any such functions are unclear.
Instead, therefore, we interpret the coherences more generally as
indicative of ‘‘default’’ or ‘‘idling’’ mode of interactions between
functionally integrated regions. As such, the coherences provide
insight into the dynamic functional architecture of the brain in the
absence of activity coordinated for a specific task.
Previous work has provided evidence that some RSNs are
correlated with slow modulations of EEG-measured neuronal
activity in the alpha band (Goldman and Cohen, 2003; Goldman
et al., 2002; Laufs et al., 2003) and mu band (Moosmann et al.,
2003). Changes in the strengths of some coherences have been
reported with neurological diseases (Greicius et al., 2004). Note
that, while Goldman showed alpha-related changes in (our) RSNs
1 and 3, Laufs’ results appear spatially more similar to (our) RSNs
4 and 5. Therefore, though we can conclude that there does seem to
be a strong correlation between the RSN time-courses and
modulation of the alpha EEG component, much remains to be
understood as to the exact nature of this link.
The PICA approach offers specific advantages relative to
correlation-based analysis with ‘‘seeding’’ of a region identified by
a prior stimulus activation study. The latter limits analysis to
be ideally sensitive to all such longer-range coherences, our results
emphasize the potential richness of the time series data and the
substantial extent of long-range coherences in fMRI datasets. A
coherences are related directly to regions identified by arbitrarily
be localized to the specific regions of functional cortex probed. For
example, the low-frequency coherences between regions of sensor-
imotor cortex identified here were not maximal in the motor cortex
methodological differences potentially account for the more
extensive patterns of activation implicated in RSNs reported here,
as well as for the increased number of distinct patterns of coherent
activity that we have identified. Reassuringly, where comparable
regions are explored, results from correlation in PICA (or other ICA
methods) are generally consistent (Greicius et al., 2004; De Luca et
al., in press).
The reproducibility of patterns of coherence across the brain was
explored directly in our study. Reproducibility of patterns between
the individuals was good. There is some evidence that prior
experience (e.g., training on a specific task) may modulate the
relative strengths of coherences (Waites et al., 2005), though it is
Coordinates of the major clusters of the RSNs, as shown in Fig. 5 (cross
points) in stereotactic space of Talairach and Tournoux (1988)
Brain regions are identified by putative Brodmann area (BA).
Fig. 6. RSN and thalamic connectivity. Top: the cross indicates the voxel
in RSN 2 (x = 6, y = ?19, z = 6 in MNI space) used for seeding an
anatomical connectivity investigation, using a standard-space anatomical
thalamus connectivity atlas. Bottom: the cross indicates the seeding voxel
in RSN 3 (x = 12, y = ?17, z = 0 in MNI space). A distinct pattern of
anatomical connectivity is associated with the different resting state
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
clear from our data that, while such modulation can occur, the RSNs
are quite robustly found across subjects. It remains to be seen
absence of an experimental task.
The ability to differentiate physiological noise variations from
different sources is inherent to multivariate decomposition techni-
ques such as PICA. In addition to signal changes that are potentially
neuronally mediated and those related to cardio-respiratory cycles,
coherences can be found that appear more specifically localized to
regions with large draining veins and may represent changes
associated with either cerebral blood volume modulation or gross
vessel movements (Kiviniemi et al., 2000). Instrument-related
artefacts also can be identified (Beckmann and Smith, 2004).
The group-level RSNs identified relate to functional–anatom-
ically distinct systems. RSN1 includes the striate and extra-striate
cortex, regions involved in visual processing (Haxby et al., 1994).
The relationship to slow modulation of alpha wave activity
previously described suggests that this RSN could be modulated
by levels of alertness, although this has not yet been demonstrated.
RSN2, which involves the precuneus, anterior pole, and midline
structures including the thalamus and hypothalamus, as well as
medial parietal cortex, is closely related to patterns described as
‘‘deactivated’’ during active tasks in PET cerebral blood flow
studies (Shulman et al., 1997; Mazoyer et al., 2001). These
regions have been suggested to be associated in a functional
network related to internal monitoring and states of consciousness
(Gusnard and Raichle, 2001). RSN3 involves the post-central
gyrus, insula, and midline cingulate in superior frontal gyrus.
These regions are involved in motor control and somatosensation
(Hsiech et al., 1999), suggesting that the network reflects
functional and anatomical interactions relevant to the control of
action. RSN4 includes occipital, dorsal, parietal, and prefrontal
regions. Parietal and prefrontal regions are closely functionally
integrated in a wide range of cognitive processes. The pattern here
may recall more specifically the network of brain regions implicated
in visual perception for action, the so-called ‘‘where’’ pathway
(Ungerleider and Haxby, 1994). RSN5, by contrast, involves
predominantly more inferior regions of occipital, parietal, and
prefrontal cortex in a pattern recalling the complementary visual
perceptual ‘‘what’’ pathway (Ungerleider and Haxby, 1994).
The agreement of RSN spatial localization and anatomical
connectivity suggests that RSNs follow similar spatial organization
to that inferred from DTI anatomical connectivity.
In summary, our analysis with PICA has identified several
independently varying patterns of signal coherence across the brain
in resting state BOLD fMRI. Similar patterns were shown to be able
to be found with the more specific measure of hemodynamic
response provided by perfusion imaging. The cortical localization
of the generators and the similarity of the patterns to known
functional–anatomical networks suggest that these arise with long-
range coherences in neuronal activity. Although the interactions
may have independent functional roles, these are not yet apparent.
Their association with the unstimulated or ‘‘resting’’ brain suggests
that they arise from ‘‘default’’ or ‘‘idling’’ state of these functional
networks. They could thus simply represent a form of ‘‘noise’’
distributed across the networks as a consequence of their functional
connectivity. Even in this instance, however, they potentially
provide information on functional systems and the dynamics of
interactions within them. They also may prove to be a useful probe
for functional alterations in the brain as a consequence of changes in
brain state, disease, or pharmacological interventions.
We are grateful to Dr. Giandomenico Iannetti and Dr. Joe
Devlin for helpful discussion on RSN neuroscience meaning. We
are grateful for financial support from the European PhD program,
the UK Medical Research Council, the UK Engineering and
Physical Sciences Research Council, and GlaxoSmithKline. We are
grateful to S. Clare for the high spatial resolution data, M. Lowe for
the low TR data, and to P. Chiarelli for the ASL data.
Beckmann, C., Smith, S.M., 2004. Probabilistic independent component
analysis for functional magnetic resonance imaging. IEEE Trans. Med.
Imag. 23, 137–152.
Biswal, B., Yetkin, F.Z., Haughton, V.M., Hyde, J.S., 1995. Functional
connectivity in the motor cortex of resting human brain using echo-
planar MRI. J. Magn. Reson. Med. 34, 537–541.
Biswal, B., Kylen, J.V., Hyde, J.S., 1997. Simultaneous assessment of flow
and bold signals in resting-state functional connectivity maps. NMR
Biomed. 10, 165–170.
Cordes, D., Haughton, V.M., Arfanakis, K., Wendt, G.J., Tursky, P.A.,
Moritzk, C.H., Quigley, M.A., Meyerand, M.E., 2000. Mapping
functionally related regions of brain with functional connectivity MR
imaging. AJNR Am. J. Neuroradiol. 21, 1636–1644.
De Luca, M., Beckmann, C.F., Behrens, T., Clare, S., Matthews, P.,
De Stefano, N., Woolrich, M., Smith, S.M., 2002. Low frequency
signals in fMRI—‘‘resting state networks’’ and the ‘‘intensity
normalisation problem’’. Proc. Int. Soc. of Magnetic Resonance
De Luca, M., Smith, S., De Stefano, N., Federico, A., Matthews, P.M., in
press. BOLD contrast resting state networks are relevant to functional
activity in the neocortical sensorimotor system. Exp. Brain Res.
Goldman, R.I., Cohen, M., 2003. Tomographic distribution of resting alpha
rythm sources revealed by independent component analysis. Ninth Int.
Conf. on Functional Mapping of the Human Brain.
Goldman, R.I., Stern, J.M., Engel Jr., J., Cohen, M.S., 2002. Simultaneous
EEG and fMRI of the alpha rhythm. NeuroReport 13 (18), 2487–2492
(Edition, 13 (18)).
Greicius, M.D., Krasnow, B., Reiss, A.L., Menon, V., 2003. Functional
connectivity in the resting brain: a network analysis of the default mode
hypothesis. Proc. Natl. Acad. Sci. U. S. A., 253–258.
Greicius, M.D., Srivastava, G., Reiss, A.L., Menon, V., 2004. Default-
mode network activity distinguishes Alzheimer_s disease from
healthy aging: evidence from functional MRI. Proc. Natl. Acad. Sci.
U. S. A., 4637–4642.
Gusnard, D.A., Raichle, M.E., 2001. Searching for a baseline: functional
imaging and the resting human brain. Nat. Rev., Neurosci. 2, 685–692.
Haxby, J.V., Horwitz, B., Ungerleider, L.G., Maisog, J.M., Pietrini, P.,
Grady, C.L., 1994. The functional organization of human extrastriate
cortex: a pet–rcbf study of selective attention to faces and locations.
J. Neurosci. 14, 6336–6353.
Hsiech, J.C., Meyerson, B.A., Ingvar, M., 1999. Pet study on central
processing of pain in trigeminal neuropathy. Eur. J. Pain 3 (1), 51–65.
Jenkinson, M., Bannister, P.R., Brady, J.M., Smith, S.M., 2002. Improved
optimisation for the robust and accurate linear registration and motion
correction of brain images. NeuroImage 17 (2), 825–841.
Jezzard, P., Matthews, P.M., Smith, S.M., 2002. Functional Magnetic
Resonance Imaging: an Introduction to Methods, J. Cogn. Neurosci.,
vol. 9. Oxford Univ. Press (Chap. 14–15).
Johansen-Berg, H., Behrens, T.E.J., Sillery, E.L., Ciccarelli, O., Wheeler-
Kingshott, C.A.M., Thompson, A.J., Smith, S.M., Matthews, P.M.,
2005. Functional–anatomical validation and individual variation of
diffusion tractography-based segmentation of the human thalamus.
Cereb. Cortex 15, 31–39.
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367
Kiviniemi, V., Jauhiainen, J., Tervonen, O., Paakko, E., Oikarinen, J., Download full-text
Vainionpaa, V., Rantala, H., Biswal, B., 2000. Slow vasomotor
fluctuation in fMRI in anesthesized child brain. J. Magn. Reson. Med.
Kwong, K.K., Belliveau, J.W., Chesler, D.A., Goldberg, I.E., Weisskoff,
R.M., Poncelet, B.P., Kennedy, D.N., Hoppel, B.E., Cohen, M.S.,
Turner, R., Cheng, H.M., Brady, T.J., Rosen, B.R., 1992. Dynamic
magnetic resonance imaging of human brain activity during primary
sensory stimulation. Proc. Natl. Acad. Sci. U. S. A. 89, 5675–5679.
Laufs, H., Kleinschmidt, A., Beyerle, A., Eger, E., Salek-Haddadi, A.,
Preibisch, C., Kraskow, K., 2003. EEG correlated fMRI of human alpha
activity. NeuroImage 19, 1463–1476.
Leopold, D.A., Murayama, Y., Logothesis, N.K., 2003. Very slow activity
fluctuations in monkey visual cortex: implications for functional brain
imaging. Cereb. Cortex 13, 422–433.
Lowe, M.J., Mock, B.J., Sorenson, J.A., 1998. Functional connectivity in
single and multislice echoplanar imaging using resting-state fluctua-
tions. NeuroImage 7, 119–132.
Mazoyer, B., Zago, L., Mellet, E., Bricogne, S., Etard, O., Houde, O.,
Crivello, F., Joliot, M., Petit, L., Tzourio-Mazoyer, N., 2001. Cortical
networks for working memory and executive functions sustain the
conscious resting state in man. Brain Res. Bull. 54 (3), 287–298.
Moosmann, M., Ritter, P., Krastel, I., Brink, A., Thees, S., Blakenburg, F.,
Taskin, B., Obrig, H., Villringer, A., 2003. Correlates of alpha rhythm in
functional magnetic resonance imaging and near infrared spectroscopy.
NeuroImage 20, 145–158.
Shulman, G.L., Fiez, J.A., Corbetta, M., Buckner, R.L., Miezin, F.M.,
Raichle, M.E., Petersen, S.E., 1997. Common blood flow changes
across visual task: II. Decreases in cerebral cortex. J. Cogn. Neurosci.
Smith, S.M., Jenkinson, M., Woolrich, M., Beckmann, C.F., Behrens,
T.E.J., Johansen-Berg, H., Bannister, P.R., De Luca, M., Drobnjak, I.,
Flitney, D.E., Niazy, R., Saunders, J., Vickers, J., Zhang, Y., De Stefano,
N., Brady, J.M., Matthews, P.M., 2004. Advances in functional and
structural MR image analysis and implementation as FSL. NeuroImage
23 (S1), 208–219.
Talairach, J., Tournoux, P., 2002. Co-planar Stereotaxic Atlas of the Human
Brain. Thieme Medical Publishers, New York.
Ungerleider, L.G., Haxby, J.V., 1994. FWhat_ and Fwhere_ in the human
brain. Curr. Opin. Neurobiol. 4, 157–165.
Waites, A.B., Stanislavsky, A., Abbott, D.F., Jackson, G.D., 2005.
Effect of prior cognitive state on resting state networks measured
with functional connectivity. Hum. Brain Mapp. 24 (1), 59–68.
Wise, R.G., Ide, K., Poulin, M.J., Tracey, I., 2004. Resting fluctuations in
arterial carbon dioxide induce significant low frequency variations in
BOLD signal. NeuroImage 21, 1652–1664.
Xiong, J., Parsons, L.M., Gao, J.H., Fox, P.T., 1999. Interregional
connectivity to primary motor cortex revealed using MRI resting state
images. Hum. Brain Mapp. 8, 151–156.
M. De Luca et al. / NeuroImage 29 (2006) 1359–1367