ORIGINAL RESEARCH ARTICLE

published: 07 November 2014

doi: 10.3389/fnhum.2014.00897

Dynamic connectivity states estimated from resting fMRI

Identify differences among Schizophrenia, bipolar disorder,
and healthy control subjects
Barnaly Rashid 1,2 , Eswar Damaraju1,2 , Godfrey D. Pearlson 3,4,5 and Vince D. Calhoun1,2,3,4 *
1
The Mind Research Network, Albuquerque, NM, USA
2
Department of Electrical and Computer Engineering, University of New Mexico, Albuquerque, NM, USA
3
Olin Neuropsychiatry Research Center – Institute of Living, Hartford, CT, USA
4
Departments of Psychiatry, Yale University School of Medicine, New Haven, CT, USA
5
Departments of Neurobiology, Yale University School of Medicine, New Haven, CT, USA

Edited by: Schizophrenia (SZ) and bipolar disorder (BP) share significant overlap in clinical symptoms,
Giacomo Koch, IRCCS Fondazione brain characteristics, and risk genes, and both are associated with dysconnectivity among
Santa Lucia, Italy
large-scale brain networks. Resting state functional magnetic resonance imaging (rsfMRI)
Reviewed by:
Vaibhav A. Diwadkar, Wayne State
data facilitates studying macroscopic connectivity among distant brain regions. Standard
University School of Medicine, USA approaches to identifying such connectivity include seed-based correlation and data-driven
Maide Bucolo, University of Catania, clustering methods such as independent component analysis (ICA) but typically focus
Italy on average connectivity. In this study, we utilize ICA on rsfMRI data to obtain intrinsic
*Correspondence: connectivity networks (ICNs) in cohorts of healthy controls (HCs) and age matched SZ and
Vince D. Calhoun, The Mind Research
Network, 1101 Yale Blvd NE,
BP patients. Subsequently, we investigated difference in functional network connectivity,
Albuquerque, NM 87106, USA defined as pairwise correlations among the timecourses of ICNs, between HCs and
e-mail: vcalhoun@mrn.org patients. We quantified differences in both static (average) and dynamic (windowed)
connectivity during the entire scan duration. Disease-specific differences were identified in
connectivity within different dynamic states. Notably, results suggest that patients make
fewer transitions to some states (states 1, 2, and 4) compared to HCs, with most such
differences confined to a single state. SZ patients showed more differences from healthy
subjects than did bipolars, including both hyper and hypo connectivity in one common
connectivity state (dynamic state 3). Also group differences between SZ and bipolar
patients were identified in patterns (states) of connectivity involving the frontal (dynamic
state 1) and frontal-parietal regions (dynamic state 3). Our results provide new information
about these illnesses and strongly suggest that state-based analyses are critical to avoid
averaging together important factors that can help distinguish these clinical groups.
Keywords: dynamic functional connectivity, intrinsic connectivity networks, independent component analysis,
schizophrenia, bipolar disorder

INTRODUCTION need to identify altered connectivity in relevant core brain

Schizophrenia (SZ) and bipolar disorder (BP) are two common networks.
psychiatric conditions characterized by gray and white matter Recently, FC has been used to examine the functional orga-
abnormalities and disrupted connectivity across large-scale brain nization of brain networks in various psychiatric illnesses,
networks (Mohamed et al., 1999; Kubicki et al., 2007). Such where FC is defined as the temporal covariance of neu-
dysconnectivity includes disruption of both structural (Kubicki ral signals between multiple spatially distinct brain regions
et al., 2007; Rotarska-Jagiela et al., 2008, 2009) and functional (Friston et al., 1993). Different analytic tools have been applied
connectivity (FC; Meyer-Lindenberg et al., 2001; Uhlhaas and to resting-state fMRI data to describe brain FC, including
Singer, 2006; Garrity et al., 2007; Calhoun et al., 2008a, 2011) seed-based analysis (Biswal et al., 1995; Greicius et al., 2003),
that may be related to clinical symptoms, including cogni- data-driven methods, such as independent component analy-
tive dysfunction. SZ is often referred to as a dysconnection sis (ICA; Hyvärinen and Oja, 2000; Calhoun et al., 2001b, 2009;
syndrome, where the term “dysconnection” refers to over- or Damoiseaux et al., 2006; Fox and Raichle, 2007; Calhoun and
under-connection of neural circuits with respect to a healthy Adali, 2012), clustering (Cordes et al., 2002), multivariate pat-
control group (Friston et al., 1993). Because changes in the tern analysis (MVPA; Norman et al., 2006; Zhu et al., 2008;
function of a single brain region cannot explain the range of Zeng et al., 2012), graph theory (Achard et al., 2006; Buckner
impairments observed in SZ or BP (Achim and Lepage, 2005; et al., 2009) and centrality (Lohmann et al., 2010). In seed-based
Van Snellenberg et al., 2006; Minzenberg et al., 2009; Ragland approach, the connectivity patterns are based on a selected
et al., 2009; Wang et al., 2009; Chepenik et al., 2010), researchers seed region of interest (ROI), while ICA-based methods do not

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 1

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

require prior knowledge of brain activity or seed ROI selection BP (Öngür et al., 2010; Calhoun et al., 2011). There are numer-
(Erhardt et al., 2011a). ous studies suggesting abnormal default network connectivity in
Until recently, most fMRI studies assumed that FC is station- SZ and BP (Zhou et al., 2007, 2008; Calhoun et al., 2008b, 2011),
ary throughout the entire scan period (or at least static during although both increased and decreased connectivity have been
a giving task or condition such as rest). This assumption of reported. Different analytical techniques could account for these
stationarity is likely an oversimplification since it is likely that inconsistent findings, as seed-based and data-driven analyses and
individuals are engaged in slightly different mental activities at varying preprocessing steps do not necessarily produce the same
different points in time. In addition, previous works showing results. Also each intrinsic brain network comprises a collection
evidence of fluctuation in FC (Arieli et al., 1996; Makeig et al., of multiple network components, only a few of which might be
2004; Onton and Makeig, 2006) are consistent with the idea affected throughout a specific period of illness.
that dynamic changes in FC occur during the course of the Prior studies have identified abnormal connectivity in other
experiment. Recent studies show that connectivity dynamics can intrinsic networks. For example, patients with persistent audi-
capture uncontrolled but reoccurring patterns of interactions tory verbal hallucinations may have increased connectivity in
among intrinsic networks during a task or at rest (Sakoǧlu et al., the cingulate cortex within the speech-related network (Wolf
2010; Allen et al., 2012; Hutchison et al., 2013; Calhoun et al., et al., 2011). In attention and executive control networks, patients
2014). These studies provide results that cannot be detected with demonstrated abnormal connectivity in precuneus and right lat-
static FC analyses. In a dynamic connectivity study using wavelet- eral pre-frontal areas. Few studies have examined both bipolar and
based time-frequency coherence analysis, significant results were SZ patients. A recent study of both disorders (Öngür et al., 2010;
observed for resting state connectivity variation between poste- Meda et al., 2012; Khadka et al., 2013) found subgenual and medial
rior cingulate cortex and an anti-correlated network (Chang and prefrontal anomalies in BP patients and dorsal medial prefrontal
Glover, 2010). Another approach for studying dynamic connectiv- anomalies in SZ patients, although considerabe overlap among
ity is the sliding-window correlation technique (Allen et al., 2012; groups.
Hutchison et al., 2013). In this paper, we implement a recently published approach to
Resting state BOLD studies have proven useful recently to inves- assess functional network connectivity (FNC) dynamics between
tigate abnormal FC, as the absence of a specific task complements healthy controls (HCs) and SZ and bipolar patients, which
task-specific study by measuring intrinsic functional brain orga- includes group spatial ICA, dynamic FNC via sliding time win-
nization without any differential behavioral performance and task dow correlation, and k-means clustering of windowed corre-
activity between diagnostic groups, and thus makes it easier for lation matrices (Allen et al., 2012). We hypothesized that dis-
cognitively compromised patients to participate in such studies. rupted functional integration in SZ and bipolar patients can
Resting-state fMRI connectivity has been used to identify dif- be found in several brain regions including temporal, frontal,
ferences in multiple patient groups including SZ (Calhoun et al., visual, and DMNs as suggested by previous studies. To test our
2009, 2011; Sakoǧlu et al., 2010; Damaraju et al., 2014), BP (Cal- hypothesis we conducted group difference analyses in connec-
houn et al., 2011), Alzheimer’s disease (Greicius et al., 2004; Sorg tivity using independent two sample t-tests. The results show
et al., 2007), autism (Starck et al., 2013), and others. However, to that dynamic FNC captured by sliding time window analysis
our knowledge, no study to date has evaluated changes in connec- can reveal significant differences between patients and con-
tivity patterns over time in fMRI in patient groups versus controls. trols that cannot be found using conventional stationary FNC
It is not yet known how spatial and temporal dynamics of rest- analysis.
ing state networks contribute to individual psychopathological
disorders. Both SZ and BP are diagnosed using cross-sectional MATERIALS AND METHODS
clinical symptoms along with longitudinal course and outcome PARTICIPANTS
measures. There are significant overlaps in symptoms and dis- We assessed 159 total subjects comprising 61 screened HCs [HC,
ease progression between these two disorders that can make it age 35.44 ± 11.57 (range), 28 females], 60 patients diagnosed
difficult to differentiate them without repeated clinical diagnos- with SZ or schizoaffective disorder (SZ, age 35.85 ± 12.01, 13
tic assessment (Keshavan et al., 2011). By determining a reliable females) and 38 bipolar subjects (BP, age 38.96 ± 10.90, 20
diagnostic indicator (‘biomarker’) based on biological features of females), matched for age with no significant differences among
these diseases, a baseline for developing more accurate and reliable three groups, where age: p = 0.303, F = 1.2031, DF = 2. Signifi-
differentiating tools for diagnosis, and ultimately treatment, can cant differences in sex among three groups were found, where sex:
in theory be provided (Keshavan et al., 2013). p = 0.002, X 2 = 11.81, DF = 2. Diagnoses were based on detailed
Previous studies show both similarities and differences in static medical and psychiatric history, chart reviews, and the Structured
FC between SZ and BP. Most prior studies focused on quantifying Clinical Interview for DSM Disorders (Gibbon et al., 1997). None
the underlying characteristics of sensory, auditory, cognitive con- were acutely ill at the time of scanning. Bipolar patients were a
trol (CC) and emotional processes of the brain. For example, the mixture of psychotic and non-psychotic by history.
default mode network (DMN), consists of a set of brain regions
known to be activated during internally focused tasks and may Data acquisition
be involved in processes such as attention to internal emotional Resting-state fMRI scans were acquired at the Institute of Living,
states, self-referential processing or task- independent thoughts Hartford, CT, USA on a 3T Siemens Allegra head-only scanner
(Buckner et al., 2008). DMN data may distinguish between SZ and with 40 mT/m gradients and a quadrature head coil. T2* -weighted

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 2

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

functional images were acquired using gradient echo planar imag- depicted peak cluster locations in gray matter with minimal
ing (EPI) method with repetition time (TR) = 1.5 s, echo time overlap with white matter, ventricles and edges of the brain
(TE) = 27 ms, field of view = 24 cm, acquisition matrix 64 × 64, and also exhibit higher low frequency temporal activity. Subject
flip angle = 700, voxel size = 3.75 mm × 3.75 mm × 4 mm, slice specific time courses and spatial maps were obtained via back
thickness = 4 mm, gap = 1 mm, number of slices = 29, 210 frames reconstruction.
and ascending acquisition. Subjects were instructed to keep their Additional post-processing steps including linear, quadratic
eyes open, look at a fixation cross on a monitor display and to rest and cubic detrending, multiple regression of the six realignment
quietly during the scan session. parameters and their temporal derivatives, removal of detected
outliers, and low-pass filtering with a high frequency cutoff of
DATA PRE-PROCESSING 0.15 Hz were applied to the component TCs in order to remove
Functional images were pre-processed using an automated trends associated with scanner drift and movement-related arti-
pipeline based around SPM 51 . Pre-processing included the facts. We have detected the outliers based on the median absolute
removal of the first four image volumes to avoid T1 equilibration deviation, as implemented in 3D DESPIKE (Cox, 1996). Outliers
effects, realignment using INRIalign (Freire et al., 2002), slice- were replaced with the best estimate using a third-order spline fit
timing correction using the middle slice as the reference frame, to the clean portions of the TCs.
spatial normalization into Montreal Neurological Institute (MNI)
space2 , reslicing to 3 mm × 3 mm × 3 mm voxels, and smoothing FC ESTIMATION
with a Gaussian kernel (FWHM = 5 mm). Voxel timeseries were The static FNC for each subject was estimated from the TC matrix,
z-scored to normalize variance across space, minimizing possi- as the C × C sample covariance matrix (see Figure 1A). In addition
ble bias in subsequent variance-based data reduction steps (Allen to the standard FNC analyses, we computed correlations between
et al., 2012). ICN TCs using a sliding temporal window [Tukey window (see
In order to limit the impact of motion we excluded from anal- Figure 1B)] having a width of 22 TRs = 33 s; sliding in steps
ysis subject data with a maximum translation of >2 mm or with of 1 TR), resulting in W = 180 windows to capture the vari-
SFNR <275 (Signal-to-fluctuation-noise ratio, where the signal is ability in connectivity. To characterize the full covariance matrix,
the average voxel intensity in all the ROIs defined in the object, we estimated covariance from the regularized precision matrix or
averaged across time, and the fluctuation noise is the temporal the inverse covariance matrix (Smith et al., 2011). Following the
standard deviation of the spatial mean in the same ROIs, after graphical LASSO method of (Friedman et al., 2008), we placed
removing the slow drift from the temporal series). Patient and a penalty on the L1 norm of the precision matrix to promote
control groups were age matched. Additional processing steps were sparsity. The regularization parameter lambda was optimized sep-
taken to mitigate against residual motion effects as described later. arately for each subject by evaluating the log-likelihood of unseen
data (windowed covariance matrices from the same subject) in a
GROUP ICA AND POST-PROCESSING cross-validation framework. Final dynamic FC estimates for each
Imaging data were decomposed into functional networks using window, were concatenated to form a C × C × W array represent-
a group-level spatial ICA (Calhoun et al., 2001a; Calhoun and ing the changes in covariance (correlation) between components
Adali, 2012). Group ICA was performed using the GIFT tool- as a function of time.
box (Calhoun, 2004). In order to obtain functional parcellation,
we used a high model order ICA (number of components, DYNAMIC STATES AND CLUSTERING
C = 100) to decompose the functionally homogeneous cortical From all of the dynamic windowed FNC matrices, we selected win-
and subcortical regions exhibiting temporally coherent activity dows of higher variability as subject exemplars and used K-means
(Kiviniemi et al., 2009; Smith et al., 2009; Abou-Elseoud et al., clustering to obtain group centrotypes. We repeated the clus-
2010). In the subject-specific data reduction principle com- tering method using different distance functions (correlation,
ponent analysis (PCA) step, 120 principal components were cosine, rather than the L1-norm) and also found very similar
retained (retaining >99% of the variance of the data). Group results. We determined the number of clusters to be five using the
data reduction retained C = 100 PCs using the expectation- elbow criterion of the cluster validity index, which is computed
maximization (EM) algorithm as implemented in the GIFT as the ratio between within-cluster distances to between-cluster
toolbox. The Infomax group ICA (Calhoun et al., 2001b) algo- distance. These centrotypes are then used as starting points to
rithm was repeated 20 times in ICASSO (Himberg and Hyvari- cluster all of the dynamic FNC data. Group specific centrotypes
nen, 2003) and the resulting components were clustered to were computed. Subject specific centrotypes were used to perform
estimate the reliability of the decomposition (Himberg et al., independent sample t-tests to probe for group differences.
2004). Subject-specific spatial maps (SMs) and time-courses
(TCs) were estimated using the GICA1 back-reconstruction RESULTS
method based on PCA compression and projection (Calhoun INTRINSIC CONNECTIVITY NETWORKS (ICNs)
et al., 2001b; Erhardt et al., 2011b). Out of the 100 compo- ICA was successfully used to identify the intrinsic connectivity
nents obtained, we characterized 49 components as ICNs that networks (ICNs) in HCs and patients with SZ and bipolar, and to
identify differences in FNC among these ICNs. The spatial maps of
1 http://www.fil.ion.ucl.ac.uk/spm/software/spm5 49 ICNs identified with group ICA are shown in Figure 2A. ICNs
2 http://www.mni.mcgill.ca/ are grouped by their anatomical and functional properties, which

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 3

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 1 | (A) An overview of the sliding window analysis. Group (TCs). (B) Stationary FC between components (left) is estimated as the
independent component analysis (ICA) is used to decomposed resting-state covariance of TCs. Dynamic FC (right) is estimated as the series of regularized
data from 159 subjects into 100 components, 49 of which are identified as covariance matrices from windowed portions of each subject’s component
intrinsic connectivity networks (ICNs). GICA1 back-reconstruction method is TCs and then the matrices are aggregated across subjects [Adapted from
used to estimate the subject specific spatial maps (SMs) and time courses (Allen et al., 2012)].

include the following: sub-cortical (SC), auditory (AUD), sensori- were initially ordered using algorithms in the brain-connectivity
motor (SM), visual (VIS), CC, default mode (DM), and cerebellar toolbox (Rubinov and Sporns, 2010) that maximize modularity
(CB) components. The observed ICN networks are very simi- of the connectivity matrix. These were manually partitioned into
lar to those found in previous studies with low model order ICA subgroups as in our earlier work (Allen et al., 2012). The aver-
(Calhoun et al., 2008a) as well as high model order ICA (Kiviniemi age connectivity matrix demonstrates strong positive connectivity
et al., 2009; Smith et al., 2009; Allen et al., 2011). within subcortical, VIS, SM, default-mode, and CB networks. A set
The SC networks are represented by four components (ICs 48, of CC regions also shows this positive connectivity among them-
91, 78, and 61) with activations focused in the amygdala, putamen selves and are also connected to certain VIS networks. These CC
head, putamen tail, and thalamus. The AUD network is repre- and VIS regions show anti correlation to default-mode regions.
sented by a single component (IC 36) with bilateral activation of Two sample t-tests did not reveal any group differences in static
the superior temporal gyrus (STG). The SM regions are captured or overall connectivity. Previous studies have found differences in
by nine components (ICs 1, 70, 15, 38, 9, 80, 14, 35, and 27). FNC in similar groups, but not with such a high model order that
The visual system (VIS) is represented by ten components (ICs produces more focus brain regions, but also more comparisons.
31, 10, 11, 16, 21, 29, 32, 33, 54, and 87), which matches with In our case, several FNC pairs showed a trend level of significance,
the functional and structural characterization of occipital cortex. but did not quite reach a corrected level of significance for the
The cognitive control network (CC) includes the ICN components static FNC analysis.
involved in directing and monitoring behavior, mediating mem- However, we also computed an analysis of FNC differences
ory and language functions (ICs 64, 66, 92, 42, 60, 63, 94, and 95). within groups of components (e.g., DMN components re-
The DMN is captured by eight components. Finally, we classify combined), called a network group (NG). To do this we computed,
the CB network with three components with activations in both for each NG, the average connectivity between it and all other
right and left cerebellum. NGs (Repovs et al., 2011). We then applied an FDR correction for
multiple comparisons of the between-NG connectivity. Several
STATIC FNC between-NG pairs showed significant group SZ/control differ-
Group mean FC or static FNC between ICN timecourses is shown ences, including sub-cortical and sensory-motor, sub-cortical and
in Figure 2B. The ICN components in the static FNC matrix CC, and DM and cerebellum. One pair, sub-cortical and CC,

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 4

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 2 | (A) Non-artifactual ICNs and (B) Group mean static FNC (L) or right (R). See table S1 for more detailed information on each
between ICN timecourses. ICNs are divided into groups and arranged intrinsic component. STG, superior temporal gyrus; SMA,
based on their anatomical and functional properties. FC was averaged supplementary motor area; MCC, middle cingulated cortex; Bi-FFG,
over all subjects and displayed as inverse Fisher-transformed. All of the bi-fusiform gyrus; MTG, middle temporal gyrus; IPL, inferior parietal
ICN labels in (B) indicate the brain region with peak amplitude and lobule; SPL, superior parietal lobule; dMPFC, dorso-medial prefrontal
should be considered as bilateral activation unless mentioned as left cortex; Cereb, cerebellum.

showed a significant difference between SZ and BP patients. No connectivity within a module was computed and stored as
between-NG connectivity difference was found between HCs and “component weight vector.” These positive or negative weights
bipolars. were then used to create weighted spatial map containing
all contributing components for a given dynamic state, and
DYNAMIC CONNECTIVITY STATES AND GROUP DIFFERENCES finally the weighted spatial maps were projected onto a 3-
We use k-means clustering method to identify re-occurring pat- dimensional MNI surface using the AFNI-SUMA (Saad et al.,
tern of FC states (Figure 3). Dynamic FNC analysis suggests that 2004).
patients make fewer transitions to some states (States 1, 2, and In state 1, two component pairs captured the differences
4) compared to HCs. Significant differences were found between between HCs and SZ patients as well as between the two patient
groups in dynamic FNC states 1, 2, 3, and 4, between healthy groups (Figures 4 and 5). Compared to HCs, SZ patients showed
control and patient groups as well as between SZ and bipolar greater connectivity between the component pair STG (C36) and
patients. left angular gyrus (C65), in the temporal-parietal region. Also
Figure 4 summarizes the difference between groups measured compared to BP, SZ patients showed greater connectivity between
by the connectivity between ICN component pairs. For better two frontal components: right motor (C15) and dorso-medial
visualization purpose, brain connectome for each of the sig- prefrontal cortex (DMPFC; C46).
nificant dynamic states is shown in Figure 5. Also, Figure 6 In dynamic connectivity state 2 (Figures 4 and 5), HCs showed
shows the rendering maps for main effects of dynamic con- greater connectivity between a sub-cortical component, putamen
nectivity for all the subjects. To create the rendering maps, tail (C78) and a frontal component, ventral motor (VM C1),
we first identified the modularity in the dynamic FC matrix compared to the patients with SZ.
for each state using the Brain Connectivity Toolbox (Rubi- In dynamic state 3 (Figures 4 and 5), most of the differ-
nov and Sporns, 2010). For each component, the average ences in connectivity were captured between HCs and SZ patients,

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 5

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 3 | Results from clustering approach for k = 5. Group specific centroids of the states (state 1 to state 5) are obtained from k-means clustering. The
total number and percentage of occurrences is listed above each centroid

and between the two patient groups. These connectivity differ- bipolar patients. We identified several ICNs that differentiate SZ
ences were found in frontal, parietal, occipital, temporal and and BP from HCs.
CB regions of the brain. SZ patients showed greater connectiv- Dynamic FC captures stable connectivity patterns that are not
ity between several temporal-parietal components, compared to observed in the stationary FC. FC of the brain is not stationary;
HCs. The connectivity between most of the frontal-parietal and rather it’s changing over time. Thus observing group-wise dif-
frontal-occipital components was greater in HCs compared to SZ ferences in connectivity across time as captured by the discrete
patients, whereas the connectivity between most of the parietal- dynamic states gives us more valuable information that cannot be
occipital components was greater in SZ patients. Also greater found within the stationary or mean FC.
connectivity in SZ was found between two frontal-parietal com- In Figure 3, each matrix represents the centroid of a cluster
ponent pairs, VM, and left supramarginal gyrus, and VM and right and signifies a connectivity state stably present within data. These
superior parietal lobule (SPL), compared to BP. This is the only dynamic connectivity states are fully reproducible and present in
dynamic state that captured differences between these two patient numerous subjects. Dynamic state 1 resembles the pattern of sta-
groups. tionary FC. FC patterns in state 2–5 represent connectivity show
Dynamic state 4 revealed differences between HCs and bipolar considerable deviation from the mean FC.
patients in temporal and parietal regions, where greater con- One of the notable features that differ between FC states is
nectivity in HCs was found between two parietal components, the connectivity within DMN regions and, between DMN and
paracentral and SPL, and greater connectivity in BP was found other functional networks. In state 3 and 5, the DMN regions
between a temporal component bilateral fusiform gyrus and a show strong synchronous activation with themselves, and mostly
parietal component left supramarginal gyrus. Dynamic state 5 asynchronous activation with other functional networks. Partic-
did not display any significant group differences in FC. Also, no ularly in state 3, the DMN regions show strong asynchrony with
significant correlation between symptoms and connectivity was most of the CC components. State 5 shows the similar nature
found. of connectivity between DMN and CC components, but with a
reduced number of CC components. Also in states 3 and 5, sev-
DISCUSSION eral sensori-motor components show negative correlations with
We explored dynamic FC patterns with ICA, sliding windows, and the DMN system, which is not visible in other states. In con-
clustering. Our analysis of connectivity dynamics in a relatively trast, state 1, 2, and 4 do not show similar FC patterns between
large sample (n = 159) provides, to our knowledge, the first whole- DMN and other ICN networks, where segregation of synchronized
brain characterization of regional differences in FC variability and activation between DMN and other ICN nodes can be
distinction of discrete FC states among healthy control, SZ, and observed.

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 6

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 4 | Difference in dynamic states. Group differences in dynamic schizophrenia and bipolar patient groups. The cells that have survived a
FC states are obtained using an independent two-sample t-test FDR threshold for multiple comparison correction are enclosed in black
between healthy control and patient groups, as well as between patch.

State 2 captures the FC differences between cortical and subcor- of DM regions and segregated synchronous activation between the
tical components, where strong asynchronous activation between nodes in other ICNs. Also, we can predict that hypersynchroniza-
subcortical regions (amygdala, putamen head, putamen tail, and tion between VIS areas will be accompanied by synchronization
thalamus) and sensori-motor, auditory, and VIS cortex were of DMN regions and strong asynchronous activation with other
found. Cerebellum also shows this asynchrony with these cortical functional networks.
regions. Also substantial reduction in connectivity between DMN Note that, state 4 is the only dynamic state where we have
regions can be observed in this state. As mentioned in several pre- found significant differences between healthy control and bipo-
vious studies, reduced thalamocortical connectivity (Spoormaker lar subjects. State 4 shows synchronous activity within most
et al., 2010), increased subcortical connectivity (Larson-Prior of the network nodes except few VIS and CC components,
et al., 2011) and a segregation of DMN connectivity (Spoormaker which show anti-correlation with themselves as well as with
et al., 2010; Larson-Prior et al., 2011) indicated a state of light sleep other ICN networks. The differences between HC and BP were
or drowsiness. Also similar dynamic state related to drowsiness was captured between a pair of SM component (paracentral gyrus)
found among healthy subjects in (Allen et al., 2012). and CC component(R SPL), and between a pair of SM com-
Hutchison et al. (2013), periods of hypersynchronization ponent (left supplementary motor area) and VIS component
were described where extremely high intra-network connectiv- (bi-fusiform gyrus).
ity between all nodes of oculomotor and motor networks were The differences between groups are not localized in a sin-
found in macaques and humans. This relates well to our observed gle dynamic state. Rather the group differences are distributed
discrete FC states where states 1,2, and 4 show time windows across four dynamic states (states 1, 2, 3, and 4). This dis-
with high correlations throughout the motor system (and some tributive nature of the group differences could be one rea-
motor components in state 5), while state 3 and 5 represent son they were not detected in the static FNC, since that
periods with synchronous activation between VIS areas. From metric only shows the average FNC for the run. Also the
our results, we can predict that periods of hypersynchronization dynamic states in Figure 4 show higher p-values for several
between motor nodes would also include synchronous activation t-tests between ICN components for different groups, which

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 7

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 5 | Brain connectome. A visual summary of significant activation unless mentioned as left (L) or right (R). STG, superior temporal
connectivity differences in dynamic states between different ICN gyrus; dMPFC, dorso-medial prefrontal cortex; AG, angular gyrus; VM,
components for control and patient groups. The gray cross mark indicates ventral motor; ITG, inferior temporal gyrus; CB, cerebellum; SMA,
that no component from that region showed any significant group supplementary motor area; IFG, inferior frontal gyrus; SMG, supramarginal
difference. Note that the colors of the connecting links and the domains gyrus; SPL, superior parietal lobule; LG, lingual gyrus; MOG, middle
are independent of each other. All of the component labels indicate the occipital gyrus; IOG, inferior occipital gyrus; Bi-FFG, bi-fusiform gyrus;
brain region with peak amplitude and should be considered as bilateral ParaCL, paracentral.

did not pass multiple comparison correction tests. With a showed that failure to deactivate default-mode regions corre-
larger sample size, more significant group difference could be sponded to gray matter losses in the dorsal ACC and medial
revealed. prefrontal cortex regions (Zhou et al., 2008; Pomarol-Clotet et al.,
Significant between-group differences in connectivity strength 2010; Skudlarski et al., 2010; Salgado-Pineda et al., 2011). How-
were found in several intrinsic networks including sub-cortical, ever, as mentioned earlier, both increased and decreased FC have
VIS, auditory, SM, CC, DM and cerebellum networks. Several com- been reported in the DMN in SZ. Medial prefrontal cortex is
ponents in the default-mode network (DMN) including DMPFC, a region known to be associated with information processing
right and left angular gyri (AG), and right and left precuneus when more than one course of action may be required, such
showed significant connectivity differences with the components as representing the thoughts, actions, and feelings of others
in VIS, CC, SM, auditory, and CB networks. Previous studies sug- across time (Gilbert et al., 2006). Several studies of both SZ and
gest that DMN may distinguish SZ and bipolar patients from BP (Öngür et al., 2010; Meda et al., 2012; Khadka et al., 2013)
HCs (Zhou et al., 2007, 2008; Calhoun et al., 2008b, 2011). The have reported subgenual and medial prefrontal abnormalities in
majority of previous studies report reduced task-related suppres- bipolar patients and dorsal medial prefrontal abnormalities in
sion in the DMN in SZ (Zhou et al., 2007, 2008; Jafri et al., 2008; SZ patients. (Huang et al., 2010) reported decreased amplitude
Bluhm et al., 2009; Jann et al., 2009; Kim et al., 2009; Park et al., of low frequency fluctuation (ALFF) in the medial prefrontal
2009; Pomarol-Clotet et al., 2010; Wang et al., 2011). Studies regions in never treated SZ patients, and found to become

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 8

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

FIGURE 6 | Main dynamic effects rendering maps. Using the Brain module was computed and stored as “component weight.” These weights
Connectivity Toolbox, modularity in the dynamic FC matrices was found. The were then used to create weighted spatial map containing all associated
FC matrix is divided into two modules with nodes of each module correlated components for a given dynamic state, and finally the weighted spatial maps
positively with each other in general and anticorrelated with nodes of other were projected onto a 3-dimensional MNI surface using the AFNI-SUMA
module. For each component the average positive connectivity within a (Saad et al., 2004).

normalized with antipsychotic therapy (Sambataro et al., 2009; Our study showed two other DMN components, left and right
Lui et al., 2010). precuneus, which are involved in a wide spectrum of highly inte-
Another DMN component found in our analysis is the angular grated tasks, including episodic memory (Cabeza and Nyberg,
gyrus (AG), which is known to be involved in language process- 2000; Rugg and Henson, 2002), mental imagery recall (Shallice
ing (Hall et al., 2005; Binder et al., 2009; Price, 2010; Clos et al., et al., 1994; Fletcher et al., 1996), and self-processing operations,
2014), as well as memory and social cognition. Therefore, AG such as first-person perspective taking (Cavanna and Trimble,
dysregulation can help differentiate SZ and bipolar patients from 2006). Garrity et al. (2007), higher positive symptoms were cor-
HCs. Our study showed greater connectivity in SZ between the related with increased deactivation in the medial frontal gyrus,
component pair STG and left AG. Notably, several studies also precuneus and the left middle temporal gyrus (MiTG). Compared
found FC abnormalities in STG, which is a major part of the to SZ patients, HCs showed greater connectivity between left cere-
dominant hemisphere language network. Also both structural and bellum and both left and right precuneus. The cerebellum may
functional abnormalities in the STG have been demonstrated in influence motor systems by estimating inconsistencies between
SZ patients in multiple studies as well as in psychotic BP and con- intention and action and by adjusting the motor operations appro-
stitute the best-replicated brain differences correlating with the priately (Kandel et al., 2000), as well playing a role in cognition and
severity of psychotic symptoms in SZ, most specifically auditory emotion (Schmahmann and Caplan, 2006). Prior studies reveal
hallucinations and formal thought disorder collectively; abnor- impaired functional integration of cerebellum in SZ (Honey et al.,
malities in these regions likely underpin psychotic phenomena 2005; Becerril et al., 2011). Collin et al. (2011) proclaimed the FC
(Aguayo, 1990; Swerdlow, 2010; Fusar-Poli et al., 2011). In our to other brain regions [left thalamus, middle cingulate gyrus, and
study, group variations in connectivity strength were observed supplementary motor area (SMA)] to be disconnected from the
in several temporal lobe components [STG, bi-fusiform gyrus cerebellum in SZ patients.
(FFG) and left inferior temporal gyrus (ITG)], known to pro- In our study, several SM components including SMA, right
cess auditory information (Kim et al., 2009; Sui et al., 2011). This and left motor, VM, supramarginal gyrus (SmG) and paracen-
reinforces the fact that aberrant temporal lobe coherence patterns tral showed between-group connectivity differences that were
may exhibit significant abnormality in both SZ, and to a lesser distributed across different dynamic states. (Jeong et al., 2009)
extent BP (Pearlson, 1997; Calhoun et al., 2008b). These findings reported decreased correlation of the left inferior frontal gyrus
may be useful in explaining the language and thought disruptions (IFG) with left middle temporal gyrus (MTG)/ left superior
in SZ. temporal sulcus, left SPL/supramarginal gyrus and other brain

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 9

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

regions. Our results showed connectivity differences between SmG We characterized FC as the covariance between ICN time-
and other brain components [with lingual gyrus (LG) in HC < SZ; courses. Characterization of FC matrices based on higher-order
with VM in SZ > BP; with bi-FFG in HC < BP]. Previous stud- statistics (e.g., mutual information) or lag-insensitive measures
ies found impaired FC between cerebellum and LG in SZ patients (e.g., cross-correlation) could efficiently recover the underlying
(Collin et al., 2011). biological structure of networks. Another limitation of the study
Other findings in our analysis include connectivity differences is that smaller acquisition parameters may not lead to optimum
in several CC components [left ITG, left middle frontal gyrus results by exploring all possible aspects of dynamic changes in FC.
(MiFG), MTG, left IFG, left superior medial gyrus (SMG), and Each subject in this study was scanned for only 5 min, which
SPL] with components from other brain networks. Abnormal FC is probably not optimal for considering the rate of change in
in left IFG, MiFG, and IFG was found in SZ patients (Jeong et al., dynamic states. A longer acquisition time (∼10 min) is recom-
2009; Müller et al., 2013). mended for a more accurate estimation of connectivity dynamics.
Another key finding in our study is greater connectivity in HCs To identify centroids of dynamic FC we used k-means clustering,
between putamen tail and VM regions, compared to SZ patients. which has several limitations, including difficulty in separating
The putamen may be involved in the generation of spontaneous clusters with different sizes and densities, and a high susceptibility
language, and linked to auditory/verbal hallucinations (Hoffman to outliers. Future work could include application of alternative
and Hampson, 2012). Several SZ studies showed FC anomalies in clustering models (fuzzy-clustering or density-based clustering
the putamen (Hoffman et al., 2011; Hoffman and Hampson, 2012; techniques) in the connectivity dynamics. Future work focusing on
Tu et al., 2012). an improved understanding of the association between disease and
connectivity dynamics could actually enrich our knowledge of the
LIMITATIONS AND FUTURE DIRECTIONS dynamic properties of the healthy functional brain. In addition,
Several experimental and methodological limitations must be con- recent work has shown that there are time-varying changes not
sidered while performing the sliding-window analysis method and only in the covariance but also in the associated spatial patterns
interpreting results. One limitation is that the non-stationary noise (Ma et al., 2014). Future studies to characterize both covariance
sources in fMRI time series can influence changes in FC over time. and spatial changes over time are warranted.
Synchronous global modulations of fMRI time series can be caused
by variations in respiratory and cardiac rates, as they predom- CONCLUSION
inantly occupy the low frequencies (<0.1 Hz; Wise et al., 2004; We have performed, to our knowledge, the first whole-brain char-
Chang and Glover, 2009). Also head motion could generate spa- acterization of intrinsic regional differences in FC variability and
tially structured artifacts in FC (Power et al., 2012; Yan et al., 2013). a comprehensive analysis of discrete FC states in SZ, BP and HCs.
Even though ICA reasonably separates the component sources for One key finding was the aberrant FC patterns found in several
sliding-window analysis, it may not have completely separated the default-mode components including DMPFC, bilateral angular
effects from other sources of interest. Therefore, to interpret the gyrus, and bilateral precuneus, in the patient groups. Other sig-
dynamic results, efficient denoising as well as recording of res- nificant findings include connectivity anomalies in VIS, SM and
piration and cardiac events should be considered. In the current cognitive control networks in both patient groups. These findings
study we performed careful quality control as well as incorporating could be used as distinctive characteristic markers in SZ and BP,
multiple motion regression steps to mitigate against the impact of and also could help diagnose the patients based on their biologi-
motion. cal features, rather than exclusively depending on cross-sectional
Another important issue for sliding-window analysis is the clinical symptoms and information on longitudinal course and
choice of window size. (Sakoǧlu et al., 2010) reported that the outcome.
ideal window size should be able to estimate FC variability and
capture the lowest frequencies of interest in the signal, as well as to ACKNOWLEDGMENT
detect interesting short-term effects. In this study, dynamics were This study was supported by NIH/NIBIB: 2R01 EB000840-06.
estimated using an empirically validated fixed sliding-window of
22TRs (33s) similar to that used in (Allen et al., 2012). Future work SUPPLEMENTARY MATERIAL
should evaluate changes across a variety of windows lengths that The Supplementary Material for this article can be found online at:
could be performed using separate windows (Cribben et al., 2012) http://www.frontiersin.org/journal/10.3389/fnhum.2014.00897/
or perhaps combined with multi-scale approaches such as wavelet abstract
transform (Chang and Glover, 2010).
Several recent studies on microstate-based EEG-fMRI resting- REFERENCES
Abou-Elseoud, A., Starck, T., Remes, J., Nikkinen, J., Tervonen, O., and Kiviniemi,
state datasets have showed that EEG microstates and some number V. (2010). The effect of model order selection in group PICA. Hum. Brain Mapp.
of fMRI-based ICNs show correspondence between themselves 31, 1207–1216. doi: 10.1002/hbm.20929
(Britz et al., 2010; Musso et al., 2010; Yuan et al., 2012). A brain Achard, S., Salvador, R., Whitcher, B., Suckling, J., and Bullmore, E. D.
microstate can be defined as a functional/physiological state dur- (2006). A resilient, low-frequency, small-world human brain functional net-
ing which specific neural processes occur (Musso et al., 2010). work with highly connected association cortical hubs. J. Neurosci. 26, 63–72.
doi: 10.1523/JNEUROSCI.3874-05.2006
Using concurrent EEG-fMRI data, the underlying physiologi- Achim, A. M., and Lepage, M. (2005). Episodic memory-related activa-
cal correlates of these dynamic states can be well assessed as tion in schizophrenia: meta-analysis. Br. J. Psychiatry 187, 500–509. doi:
demonstrated in (Allen et al., 2013). 10.1192/bjp.187.6.500

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 10

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

Aguayo, J. (1990). Auditory hallucinations and smaller superior temporal gyral Chang, C., and Glover, G. H. (2009). Relationship between respiration, end-tidal
volume in schizophrenia. Am. J. Psychiatry 147, 1457–1462. CO2 , and BOLD signals in resting-state fMRI. Neuroimage 47, 1381–1393. doi:
Allen, E., Eichele, T., Wu, L., and Calhoun, V. D. (2013). “EEG signatures of func- 10.1016/j.neuroimage.2009.04.048
tional connectivity states,” in Proceedings of the Human Brain Mapping, Seattle, Chang, C., and Glover, G. H. (2010). Time–frequency dynamics of resting-
WA. state brain connectivity measured with fMRI. Neuroimage 50, 81–98. doi:
Allen, E. A., Damaraju, E., Plis, S. M., Erhardt, E. B., Eichele, T., and Calhoun, V. D. 10.1016/j.neuroimage.2009.12.011
(2012). Tracking whole-brain connectivity dynamics in the resting state. Cereb. Chepenik, L. G., Raffo, M., Hampson, M., Lacadie, C., Wang, F., Jones, M. M., et al.
Cortex 24, 663–676. (2010). Functional connectivity between ventral prefrontal cortex and amygdala
Allen, E. A., Erhardt, E. B., Damaraju, E., Gruner, W., Segall, J. M., Silva, R. F., et al. at low frequency in the resting state in bipolar disorder. Psychiatry Res. 182,
(2011). A baseline for the multivariate comparison of resting-state networks. 207–210. doi: 10.1016/j.pscychresns.2010.04.002
Front. Syst. Neurosci. 5:2. doi: 10.3389/fnsys.2011.00002 Clos, M., Langner, R., Meyer, M., Oechslin, M. S., Zilles, K., and Eickhoff, S.
Arieli, A., Sterkin, A., Grinvald, A., and Aertsen, A. (1996). Dynamics of ongoing B. (2014). Effects of prior information on decoding degraded speech: an fMRI
activity: explanation of the large variability in evoked cortical responses. Science study. Hum. Brain Mapp. 35, 61–74. doi: 10.1002/hbm.22151
273, 1868–1871. doi: 10.1126/science.273.5283.1868 Collin, G., Pol, H. E. H., Haijma, S. V., Cahn, W., Kahn, R. S., and Van Den Heuvel, M.
Becerril, K. E., Repovs, G., and Barch, D. M. (2011). Error process- P. (2011). Impaired cerebellar functional connectivity in schizophrenia patients
ing network dynamics in schizophrenia. Neuroimage 54, 1495–1505. doi: and their healthy siblings. Front. Psychiatry 2:73.
10.1016/j.neuroimage.2010.09.046 Cordes, D., Haughton, V., Carew, J. D., Arfanakis, K., and Maravilla, K. (2002).
Binder, J. R., Desai, R. H., Graves, W. W., and Conant, L. L. (2009). Where Hierarchical clustering to measure connectivity in fMRI resting-state data. Magn.
is the semantic system? A critical review and meta-analysis of 120 functional Reson. Imaging 20, 305–317. doi: 10.1016/S0730-725X(02)00503-9
neuroimaging studies. Cereb. Cortex 19, 2767–2796. doi: 10.1093/cercor/bhp055 Cox, R. W. (1996). AFNI: software for analysis and visualization of functional
Biswal, B., Zerrin Yetkin, F., Haughton, V. M., and Hyde, J. S. (1995). Functional magnetic resonance neuroimages. Comput. Biomed. Res. 29, 162–173. doi:
connectivity in the motor cortex of resting human brain using echo-planar mri. 10.1006/cbmr.1996.0014
Magn. Reson. Med. 34, 537–541. doi: 10.1002/mrm.1910340409 Cribben, I., Haraldsdottir, R., Atlas, L. Y., Wager, T. D., and Lindquist, M. A. (2012).
Bluhm, R. L., Miller, J., Lanius, R. A., Osuch, E. A., Boksman, K., Neufeld, R. W. J., Dynamic connectivity regression: determining state-related changes in brain
et al. (2009). Retrosplenial cortex connectivity in schizophrenia. Psychiatry Res. connectivity. Neuroimage 61, 907–920. doi: 10.1016/j.neuroimage.2012.03.070
174, 17–23. doi: 10.1016/j.pscychresns.2009.03.010 Damaraju, E., Allen, E. A., Belger, A., Ford, J. M., McEwen, S., Mathalon, D.
Britz, J., Van De Ville, D., and Michel, C. M. (2010). BOLD correlates of EEG H., et al. (2014). Dynamic functional connectivity analysis reveals transient
topography reveal rapid resting-state network dynamics. Neuroimage 52, 1162– states of dysconnectivity in schizophrenia. Neuroimage Clin. 5, 298–308. doi:
1170. doi: 10.1016/j.neuroimage.2010.02.052 10.1016/j.nicl.2014.07.003
Buckner, R. L., Andrews-Hanna, J. R., and Schacter, D. L. (2008). The brain’s default Damoiseaux, J. S., Rombouts, S., Barkhof, F., Scheltens, P., Stam, C. J., Smith, S. M.,
network. Ann. N. Y. Acad. Sci. 1124, 1–38. doi: 10.1196/annals.1440.011 et al. (2006). Consistent resting-state networks across healthy subjects. Proc. Natl.
Buckner, R. L., Sepulcre, J., Talukdar, T., Krienen, F. M., Liu, H., Hedden, T., et al. Acad. Sci. U.S.A. 103, 13848–13853. doi: 10.1073/pnas.0601417103
(2009). Cortical hubs revealed by intrinsic functional connectivity: mapping, Erhardt, E. B., Allen, E. A., Damaraju, E., and Calhoun, V. D. (2011a). On network
assessment of stability, and relation to Alzheimer’s disease. J. Neurosci. 29, 1860– derivation, classification, and visualization: a response to Habeck and Moeller.
1873. doi: 10.1523/JNEUROSCI.5062-08.2009 Brain Connect. 1, 105–110. doi: 10.1089/brain.2011.0022
Cabeza, R., and Nyberg, L. (2000). Imaging cognition II: an empirical Erhardt, E. B., Rachakonda, S., Bedrick, E. J., Allen, E. A., Adali, T., and Calhoun, V.
review of 275 PET and fMRI studies. J. Cogn. Neurosci. 12, 1–47. doi: D. (2011b). Comparison of multi-subject ICA methods for analysis of fMRI data.
10.1162/08989290051137585 Hum. Brain Mapp. 32, 2075–2095. doi: 10.1002/hbm.21170
Calhoun, V. D. (2004). Group ICA of fMRI Toolbox (GIFT). Available at: http://icatb. Fletcher, P. C., Shallice, T., Frith, C. D., Frackowiak, R. S. J., and Dolan, R. J. (1996).
sourceforge.net Brain activity during memory retrieval The influence of imagery and semantic
Calhoun, V. D., and Adali, T. (2012). Multisubject independent component analysis cueing. Brain 119, 1587–1596. doi: 10.1093/brain/119.5.1587
of fMRI: a decade of intrinsic networks, default mode, and neurodiagnostic Fox, M. D., and Raichle, M. E. (2007). Spontaneous fluctuations in brain activity
discovery. Biomed. Eng. IEEE Rev. 5, 60–73. doi: 10.1109/RBME.2012.2211076 observed with functional magnetic resonance imaging. Nat. Rev. Neurosci. 8,
Calhoun, V. D., Adali, T., Pearlson, G. D., and Pekar, J. (2001a). “Group ICA of 700–711. doi: 10.1038/nrn2201
functional MRI data: separability, stationarity, and inference”, in Proceedings of Freire, L., Roche, A., and Mangin, J. F. (2002). What is the best similarity measure for
the International Conference on ICA and BSS, San Diego, CA. motion correction in fMRI time series? Med. Imaging IEEE Trans. 21, 470–484.
Calhoun, V. D., Adali, T., Pearlson, G. D., and Pekar, J. J. (2001b). A method for mak- doi: 10.1109/TMI.2002.1009383
ing group inferences from functional MRI data using independent component Friedman, J., Hastie, T., and Tibshirani, R. (2008). Sparse inverse covariance
analysis. Hum. Brain Mapp. 14, 140–151. doi: 10.1002/hbm.1048 estimation with the graphical lasso. Biostatistics 9, 432–441. doi: 10.1093/bio-
Calhoun, V. D., Kiehl, K. A., and Pearlson, G. D. (2008a). Modulation of temporally statistics/kxm045
coherent brain networks estimated using ICA at rest and during cognitive tasks. Friston, K. J., Frith, C. D., Liddle, P. F., and Frackowiak, R. S. J. (1993). Functional
Hum. Brain Mapp. 29, 828–838. doi: 10.1002/hbm.20581 connectivity: the principal-component analysis of large (PET) data sets. J. Cereb.
Calhoun, V. D., Maciejewski, P. K., Pearlson, G. D., and Kiehl, K. A. (2008b). Blood Flow Metab. 13, 5–5. doi: 10.1038/jcbfm.1993.4
Temporal lobe and “default” hemodynamic brain modes discriminate between Fusar-Poli, P., Borgwardt, S., Crescini, A., Deste, G., Kempton, M. J., Lawrie, S., et al.
schizophrenia and bipolar disorder. Hum. Brain Mapp. 29, 1265–1275. doi: (2011). Neuroanatomy of vulnerability to psychosis: a voxel-based meta-analysis.
10.1002/hbm.20463 Neurosci. Biobehav. Rev. 35, 1175–1185. doi: 10.1016/j.neubiorev.2010.12.005
Calhoun, V. D., Liu, J., and Adali, T. (2009). A review of group ICA for fMRI data Garrity, A., Pearlson, G., Mckiernan, K., Lloyd, D., Kiehl, K., and Calhoun, V.
and ICA for joint inference of imaging, genetic, and ERP data. Neuroimage 45, (2007). Aberrant “default mode” functional connectivity in schizophrenia. Am. J.
S163–S172. doi: 10.1016/j.neuroimage.2008.10.057 Psychiatry 164, 450–457. doi: 10.1176/appi.ajp.164.3.450
Calhoun, V. D., Miller, R., Pearlson, G., and Adalı, T. (2014). The chronnectome: Gibbon, M., Spitzer, R. L., and First, M. B. (1997). User’s Guide for the Structured
time-varying connectivity networks as the next frontier in fMRI data discovery. Clinical Interview for DSM-IV Axis II Personality Disorders: SCID-II (Arlington,
Neuron 84, 262–274. doi: 10.1016/j.neuron.2014.10.015 VA: American Psychiatric Publishing).
Calhoun, V. D., Sui, J., Kiehl, K., Turner, J., Allen, E., and Pearlson, G. Gilbert, S. J., Spengler, S., Simons, J. S., Steele, J. D., Lawrie, S. M., Frith, C. D.,
(2011). Exploring the psychosis functional connectome: aberrant intrinsic et al. (2006). Functional specialization within rostral prefrontal cortex (area 10):
networks in schizophrenia and bipolar disorder. Front. Psychiatry 2:75. doi: a meta-analysis. J. Cogn. Neurosci. 18, 932–948. doi: 10.1162/jocn.2006.18.6.932
10.3389/fpsyt.2011.00075 Greicius, M. D., Krasnow, B., Reiss, A. L., and Menon, V. (2003). Functional
Cavanna, A. E., and Trimble, M. R. (2006). The precuneus: a review of its connectivity in the resting brain: a network analysis of the default mode hypoth-
functional anatomy and behavioural correlates. Brain 129, 564–583. doi: esis. Proc. Natl. Acad. Sci. U.S.A. 100, 253–258. doi: 10.1073/pnas.01350
10.1093/brain/awl004 58100

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 11

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

Greicius, M. D., Srivastava, G., Reiss, A. L., and Menon, V. (2004). Default- Larson-Prior, L. J., Power, J. D., Vincent, J. L., Nolan, T. S., Coalson, R. S., Zempel,
mode network activity distinguishes Alzheimers disease from healthy aging: J., et al. (2011). Modulation of the brain’s functional network architecture in the
evidence from functional MRI. Proc. Natl. Acad. Sci. U.S.A. 101, 4637–4642. transition from wake to sleep. Prog. Brain Res. 193, 227–294. doi: 10.1016/B978-
doi: 10.1073/pnas.0308627101 0-444-53839-0.00018-1
Hall, D. A., Fussell, C., and Summerfield, A. Q. (2005). Reading fluent speech from Lohmann, G., Margulies, D. S., Horstmann, A., Pleger, B., Lepsien, J., Goldhahn,
talking faces: typical brain networks and individual differences. J. Cogn. Neurosci. D., et al. (2010). Eigenvector centrality mapping for analyzing connectiv-
17, 939–953. doi: 10.1162/0898929054021175 ity patterns in fMRI data of the human brain. PLoS ONE 5:e10232. doi:
Himberg, J., and Hyvarinen, A. (2003). “Icasso: software for investigating the reli- 10.1371/journal.pone.0010232
ability of ICA estimates by clustering and visualization,” in Proceedings of the Lui, S., Li, T., Deng, W., Jiang, L., Wu, Q., Tang, H., et al. (2010). Short-term
13th Workshop on Neural Networks for Signal Processing, NNSP’03 (Espoo: IEEE), effects of antipsychotic treatment on cerebral function in drug-naive first-episode
259–268. schizophrenia revealed by “resting state” functional magnetic resonance imaging.
Himberg, J., Hyvärinen, A., and Esposito, F. (2004). Validating the independent com- Arch. Gen. Psychiatry 67, 783–792. doi: 10.1001/archgenpsychiatry.2010.84
ponents of neuroimaging time series via clustering and visualization. Neuroimage Ma, S., Calhoun, V. D., Phlypo, R., and Adali, T. (2014). Dynamic changes of
22, 1214–1222. doi: 10.1016/j.neuroimage.2004.03.027 spatial functional network connectivity in healthy individuals and schizophre-
Hoffman, R. E., Fernandez, T., Pittman, B., and Hampson, M. (2011). Elevated nia patients using independent vector analysis. Neuroimage 90, 196–206. doi:
functional connectivity along a corticostriatal loop and the mechanism of audi- 10.1016/j.neuroimage.2013.12.063
tory/verbal hallucinations in patients with schizophrenia. Biol. Psychiatry 69, Makeig, S., Debener, S., Onton, J., and Delorme, A. (2004). Mining event-
407–414. doi: 10.1016/j.biopsych.2010.09.050 related brain dynamics. Trends Cogn. Sci. (Regul. Ed.) 8, 204–210. doi:
Hoffman, R. E., and Hampson, M. (2012). Functional connectivity studies of 10.1016/j.tics.2004.03.008
patients with auditory verbal hallucinations. Front. Hum. Neurosci. 6:6. doi: Meda, S. A., Gill, A., Stevens, M. C., Lorenzoni, R. P., Glahn, D. C., Calhoun, V. D.,
10.3389/fnhum.2012.00006 et al. (2012). Differences in resting-state functional magnetic resonance imaging
Honey, G. D., Pomarol-Clotet, E., Corlett, P. R., Honey, R. A. E., Mckenna, P. functional network connectivity between schizophrenia and psychotic bipolar
J., Bullmore, E. T., et al. (2005). Functional dysconnectivity in schizophrenia probands and their unaffected first-degree relatives. Biol. Psychiatry 71, 881–889.
associated with attentional modulation of motor function. Brain 128, 2597–2611. doi: 10.1016/j.biopsych.2012.01.025
doi: 10.1093/brain/awh632 Meyer-Lindenberg, A., Poline, J.-B., Kohn, P. D., Holt, J. L., Egan, M. F., Weinberger,
Huang, X.-Q., Lui, S., Deng, W., Chan, R. C. K., Wu, Q.-Z., Jiang, L.-J., et al. D. R., et al. (2001). Evidence for abnormal cortical functional connectivity dur-
(2010). Localization of cerebral functional deficits in treatment-naive, first- ing working memory in schizophrenia. Am. J. Psychiatry 158, 1809–1817. doi:
episode schizophrenia using resting-state fMRI. Neuroimage 49, 2901–2906. doi: 10.1176/appi.ajp.158.11.1809
10.1016/j.neuroimage.2009.11.072 Minzenberg, M. J., Laird, A. R., Thelen, S., Carter, C. S., and Glahn, D. C. (2009).
Hutchison, R. M., Womelsdorf, T., Gati, J. S., Everling, S., and Menon, R. S. Meta-analysis of 41 functional neuroimaging studies of executive function in
(2013). Resting-state networks show dynamic functional connectivity in awake schizophrenia. Arch. Gen. Psychiatry 66, 811–822. doi: 10.1001/archgenpsychia-
humans and anesthetized macaques. Hum. Brain Mapp. 34, 2154–2177. doi: try.2009.91
10.1002/hbm.22058 Mohamed, S., Paulsen, J. S., O’Leary, D., Arndt, S., and Andreasen, N. (1999).
Hyvärinen, A., and Oja, E. (2000). Independent component analysis: algorithms and Generalized cognitive deficits in schizophrenia: a study of first-episode patients.
applications. Neural Netw. 13, 411–430. doi: 10.1016/S0893-6080(00)00026-5 Arch. Gen. Psychiatry 56, 749–754. doi: 10.1001/archpsyc.56.8.749
Jafri, M. J., Pearlson, G. D., Stevens, M., and Calhoun, V. D. (2008). A Müller, V. I., Cieslik, E. C., Laird, A. R., Fox, P. T., and Eickhoff, S. B. (2013).
method for functional network connectivity among spatially independent Dysregulated left inferior parietal activity in schizophrenia and depression: func-
resting-state components in schizophrenia. Neuroimage 39, 1666–1681. doi: tional connectivity and characterization. Front. Hum. Neurosci. 7:268. doi:
10.1016/j.neuroimage.2007.11.001 10.3389/fnhum.2013.00268
Jann, K., Dierks, T., Boesch, C., Kottlow, M., Strik, W., and Koenig, T. (2009). Musso, F., Brinkmeyer, J., Mobascher, A., Warbrick, T., and Winterer, G. (2010).
BOLD correlates of EEG alpha phase-locking and the fMRI default mode network. Spontaneous brain activity and EEG microstates. A novel EEG/fMRI analysis
Neuroimage 45, 903–916. doi: 10.1016/j.neuroimage.2009.01.001 approach to explore resting-state networks. Neuroimage 52, 1149–1161. doi:
Jeong, B., Wible, C. G., Hashimoto, R. I., and Kubicki, M. (2009). Func- 10.1016/j.neuroimage.2010.01.093
tional and anatomical connectivity abnormalities in left inferior frontal gyrus Norman, K. A., Polyn, S. M., Detre, G. J., and Haxby, J. V. (2006). Beyond mind-
in schizophrenia. Hum. Brain Mapp. 30, 4138–4151. doi: 10.1002/hbm.20835 reading: multi-voxel pattern analysis of fMRI data. Trends Cogn. Sci. (Regul. Ed.)
Kandel, E. R., Schwartz, J. H., and Jessell, T. M. (2000). Principles of Neural Science. 10, 424–430.
New York: McGraw-Hill . Öngür, D., Lundy, M., Greenhouse, I., Shinn, A. K., Menon, V., Cohen, B.
Keshavan, M. S., Clementz, B. A., Pearlson, G. D., Sweeney, J. A., and Tamminga, C. M., et al. (2010). Default mode network abnormalities in bipolar disorder and
A. (2013). Reimagining psychoses: an agnostic approach to diagnosis. Schizophr. schizophrenia. Psychiatry Res. 183, 59–68. doi: 10.1016/j.pscychresns.2010.04.008
Res. 146, 10–16. doi: 10.1016/j.schres.2013.02.022 Onton, J., and Makeig, S. (2006). Information-based modeling of event-related
Keshavan, M. S., Morris, D. W., Sweeney, J. A., Pearlson, G., Thaker, G., Seidman, brain dynamics. Prog. Brain Res. 159, 99–120.
L. J., et al. (2011). A dimensional approach to the psychosis spectrum between Park, I. H., Kim, J.-J., Chun, J., Jung, Y. C., Seok, J. H., Park, H.-
bipolar disorder and schizophrenia: the Schizo-Bipolar Scale. Schizophr. Res. 133, J., et al. (2009). Medial prefrontal default-mode hypoactivity affecting trait
250–254. doi: 10.1016/j.schres.2011.09.005 physical anhedonia in schizophrenia. Psychiatry Res. 171, 155–165. doi:
Khadka, S., Meda, S. A., Stevens, M. C., Glahn, D. C., Calhoun, V. D., Sweeney, J. 10.1016/j.pscychresns.2008.03.010
A., et al. (2013). Is aberrant functional connectivity a psychosis endophenotype? Pearlson, G. D. (1997). Superior temporal gyrus and planum temporale in
A resting state functional magnetic resonance imaging study. Biol. Psychiatry 74, schizophrenia: a selective review. Prog. Neuro Psychopharmacol. Biol. Psychiatry
458–466. doi: 10.1016/j.biopsych.2013.04.024 21, 1203–1229. doi: 10.1016/S0278-5846(97)00159-0
Kim, D. I., Manoach, D. S., Mathalon, D. H., Turner, J. A., Mannell, M., Brown, G. Pomarol-Clotet, E., Canales-Rodriguez, E. J., Salvador, R., Sarró, S., Gomar, J. J., Vila,
G., et al. (2009). Dysregulation of working memory and default-mode networks F., et al. (2010). Medial prefrontal cortex pathology in schizophrenia as revealed
in schizophrenia using independent component analysis, an fBIRN and MCIC by convergent findings from multimodal imaging. Mol. Psychiatry 15, 823–830.
study. Hum. Brain Mapp. 30, 3795–3811. doi: 10.1002/hbm.20807 doi: 10.1038/mp.2009.146
Kiviniemi, V., Starck, T., Remes, J., Long, X., Nikkinen, J., Haapea, M., et al. (2009). Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., and Petersen,
Functional segmentation of the brain cortex using high model order group PICA. S. E. (2012). Spurious but systematic correlations in functional connectivity
Hum. Brain Mapp. 30, 3865–3886. doi: 10.1002/hbm.20813 MRI networks arise from subject motion. Neuroimage 59, 2142–2154. doi:
Kubicki, M., Mccarley, R., Westin, C.-F., Park, H.-J., Maier, S., Kikinis, R., et al. 10.1016/j.neuroimage.2011.10.018
(2007). A review of diffusion tensor imaging studies in schizophrenia. J. Psychiatr. Price, C. J. (2010). The anatomy of language: a review of 100 fMRI studies published
Res. 41, 15–30. doi: 10.1016/j.jpsychires.2005.05.005 in 2009. Ann. N. Y. Acad. Sci. 1191, 62–88. doi: 10.1111/j.1749-6632.2010.05444.x

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 12

Rashid et al. Dynamic connectivity in schizophrenia and bipolar

Ragland, J., Laird, A., Ranganath, C., Blumenfeld, R., Gonzales, S., and Glahn, D. Swerdlow, N. R. (2010). “Integrative circuit models and their implications for the
(2009). Prefrontal activation deficits during episodic memory in schizophrenia. pathophysiologies and treatments of the schizophrenias,” in Behavioral Neurobi-
Am. J. Psychiatry 166, 863–874. doi: 10.1176/appi.ajp.2009.08091307 ology of Schizophrenia and Its Treatment, ed. N. R. Swerdlow (Berlin: Springer),
Repovs, G., Csernansky, J. G., and Barch, D. M. (2011). Brain network connectivity 555–583.
in individuals with schizophrenia and their siblings. Biol. Psychiatry 69, 967–973. Tu, P.-C., Hsieh, J.-C., Li, C.-T., Bai, Y.-M., and Su, T.-P. (2012). Cortico-striatal
doi: 10.1016/j.biopsych.2010.11.009 disconnection within the cingulo-opercular network in schizophrenia revealed
Rotarska-Jagiela, A., Oertel-Knoechel, V., Demartino, F., Van De Ven, V., Formisano, by intrinsic functional connectivity analysis: a resting fMRI study. Neuroimage
E., Roebroeck, A., et al. (2009). Anatomical brain connectivity and positive symp- 59, 238–247. doi: 10.1016/j.neuroimage.2011.07.086
toms of schizophrenia: a diffusion tensor imaging study. Psychiatry Res. 174, 9–16. Uhlhaas, P. J., and Singer, W. (2006). Neural synchrony in brain disorders: rele-
doi: 10.1016/j.biopsych.2010.11.009 vance for cognitive dysfunctions and pathophysiology. Neuron 52, 155–168. doi:
Rotarska-Jagiela, A., Schönmeyer, R., Oertel, V., Haenschel, C., Vogeley, K., and 10.1016/j.neuron.2006.09.020
Linden, D. E. J. (2008). The corpus callosum in schizophrenia-volume and Van Snellenberg, J. X., Torres, I. J., and Thornton, A. E. (2006). Functional
connectivity changes affect specific regions. Neuroimage 39, 1522–1532. doi: neuroimaging of working memory in schizophrenia: task performance as a
10.1016/j.pscychresns.2009.03.002 moderating variable. Neuropsychology 20, 497. doi: 10.1037/0894-4105.20.5.497
Rubinov, M., and Sporns, O. (2010). Complex network measures of brain Wang, F., Kalmar, J. H., He, Y., Jackowski, M., Chepenik, L. G., Edmiston, E. E., et al.
connectivity: uses and interpretations. Neuroimage 52, 1059–1069. doi: (2009). Functional and structural connectivity between the perigenual anterior
10.1016/j.neuroimage.2007.10.063 cingulate and amygdala in bipolar disorder. Biol. Psychiatry 66, 516–521. doi:
Rugg, M. D., and Henson, R. N. A. (2002). “Episodic memory retrieval: an (event-
10.1016/j.biopsych.2009.03.023
related) functional neuroimaging perspective,” in The Cognitive Neuroscience of
Wang, L., Metzak, P. D., and Woodward, T. S. (2011). Aberrant connectivity during
Memory: Encoding and Retrieval, eds A. Parker, E. Wilding, and T. Bussey (Hove:
self–other source monitoring in schizophrenia. Schizophr. Res. 125, 136–142. doi:
Psychology Press), 3–37.
10.1016/j.schres.2010.11.012
Saad, Z. S., Reynolds, R. C., Argall, B., Japee, S., and Cox, R. W. (2004). “SUMA:
an interface for surface-based intra-and inter-subject analysis with AFNI,” in Wise, R. G., Ide, K., Poulin, M. J., and Tracey, I. (2004). Resting fluctuations in arte-
Proceedings of the IEEE International Symposium on Biomedical Imaging: Nano to rial carbon dioxide induce significant low frequency variations in BOLD signal.
Macro, 2004 (National Institute of Health, Bethesda, MD: IEEE), 1510–1513. Neuroimage 21, 1652–1664. doi: 10.1016/j.neuroimage.2003.11.025
Sakoǧlu, Ü., Pearlson, G. D., Kiehl, K. A., Wang, Y. M., Michael, A. M., and Calhoun, Wolf, N. D., Sambataro, F., Vasic, N., Frasch, K., Schmid, M., Schönfeldt-Lecuona, C.,
V. D. (2010). A method for evaluating dynamic functional network connectivity et al. (2011). Dysconnectivity of multiple resting-state networks in patients with
and task-modulation: application to schizophrenia. MAGMA 23, 351–366. doi: schizophrenia who have persistent auditory verbal hallucinations. J. Psychiatry
10.1007/s10334-010-0197-8 Neurosci. 36, 366. doi: 10.1503/jpn.110008
Salgado-Pineda, P., Fakra, E., Delaveau, P., Mckenna, P. J., Pomarol-Clotet, E., and Yan, C. -G., Cheung, B., Kelly, C., Colcombe, S., Craddock, R. C., Di Martino, A.,
Blin, O. (2011). Correlated structural and functional brain abnormalities in the et al. (2013). A comprehensive assessment of regional variation in the impact of
default mode network in schizophrenia patients. Schizophr. Res. 125, 101–109. head micromovements on functional connectomics. Neuroimage 76, 183–201.
doi: 10.1016/j.schres.2010.10.027 doi: 10.1016/j.neuroimage.2013.03.004
Sambataro, F., Blasi, G., Fazio, L., Caforio, G., Taurisano, P., Romano, R., et al. (2009). Yuan, H., Zotev, V., Phillips, R., Drevets, W. C., and Bodurka, J. (2012).
Treatment with olanzapine is associated with modulation of the default mode Spatiotemporal dynamics of the brain at rest—exploring EEG microstates as
network in patients with Schizophrenia. Neuropsychopharmacology 35, 904–912. electrophysiological signatures of BOLD resting state networks. Neuroimage 60,
doi: 10.1038/npp.2009.192 2062–2072. doi: 10.1016/j.neuroimage.2012.02.031
Schmahmann, J. D., and Caplan, D. (2006). Cognition, emotion and the cerebellum. Zeng, L.-L., Shen, H., Liu, L., Wang, L., Li, B., Fang, P., et al. (2012). Identify-
Brain 129, 290–292. doi: 10.1093/brain/awh729 ing major depression using whole-brain functional connectivity: a multivariate
Shallice, T., Fletcher, P., Frith, C. D., Grasby, P., Frackowiak, R. S. J., and Dolan, R. J. pattern analysis. Brain 135, 1498–1507. doi: 10.1093/brain/aws059
(1994). Brain regions associated with acquisition and retrieval of verbal episodic Zhou, Y., Liang, M., Tian, L., Wang, K., Hao, Y., Liu, H., et al. (2007). Functional
memory. Nature 368, 633–635. doi: 10.1038/368633a0 disintegration in paranoid schizophrenia using resting-state fMRI. Schizophr. Res.
Skudlarski, P., Jagannathan, K., Anderson, K., Stevens, M. C., Calhoun, V. D., 97, 194–205. doi: 10.1016/j.schres.2007.05.029
Skudlarska, B. A., et al. (2010). Brain connectivity is not only lower but different in Zhou, Y., Shu, N., Liu, Y., Song, M., Hao, Y., Liu, H., et al. (2008). Altered resting-
schizophrenia: a combined anatomical and functional approach. Biol. Psychiatry state functional connectivity and anatomical connectivity of hippocampus in
68, 61–69. doi: 10.1016/j.biopsych.2010.03.035 schizophrenia. Schizophr. Res. 100, 120–132. doi: 10.1016/j.schres.2007.11.039
Smith, S. M., Fox, P. T., Miller, K. L., Glahn, D. C., Fox, P. M., Mackay, C. Zhu, F., Kapitan, J., Tranter, G. E., Pudney, P. D. A., Isaacs, N. W., Hecht, L., et al.
E., et al. (2009). Correspondence of the brain’s functional architecture dur- (2008). Residual structure in disordered peptides and unfolded proteins from
ing activation and rest. Proc. Natl. Acad. Sci. U.S.A. 106, 13040–13045. doi: multivariate analysis and ab initio simulation of Raman optical activity data.
10.1073/pnas.0905267106 Proteins 70, 823–833. doi: 10.1002/prot.21593
Smith, S. M., Miller, K. L., Salimi-Khorshidi, G., Webster, M., Beckmann, C. F.,
Nichols, T. E., et al. (2011). Network modelling methods for FMRI. Neuroimage Conflict of Interest Statement: The authors declare that the research was conducted
54, 875–891. doi: 10.1016/j.neuroimage.2010.08.063 in the absence of any commercial or financial relationships that could be construed
Sorg, C., Riedl, V., Mühlau, M., Calhoun, V. D., Eichele, T., Läer, L., et al. as a potential conflict of interest.
(2007). Selective changes of resting-state networks in individuals at risk for
Alzheimer’s disease. Proc. Natl. Acad. Sci. U.S.A. 104, 18760–18765. doi: Received: 31 May 2014; accepted: 20 October 2014; published online: 07 November
10.1073/pnas.0708803104 2014.
Spoormaker, V. I., Schröter, M. S., Gleiser, P. M., Andrade, K. C., Dresler, M., Citation: Rashid B, Damaraju E, Pearlson GD and Calhoun VD (2014) Dynamic
Wehrle, R., et al. (2010). Development of a large-scale functional brain network connectivity states estimated from resting fMRI identify differences among Schizophre-
during human non-rapid eye movement sleep. J. Neurosci. 30, 11379–11387. doi: nia, bipolar disorder, and healthy control subjects. Front. Hum. Neurosci. 8:897. doi:
10.1523/JNEUROSCI.2015-10.2010 10.3389/fnhum.2014.00897
Starck, T., Nikkinen, J., Rahko, J., Remes, J., Hurtig, T., Haapsamo, H., et al. (2013). This article was submitted to the journal Frontiers in Human Neuroscience.
Resting state fMRI reveals a default mode dissociation between retrosplenial and Copyright © 2014 Rashid, Damaraju, Pearlson and Calhoun. This is an open-access
medial prefrontal subnetworks in ASD despite motion scrubbing. Front. Hum. article distributed under the terms of the Creative Commons Attribution License
Neurosci. 7:802. doi: 10.3389/fnhum.2013.00802 (CC BY). The use, distribution or reproduction in other forums is permitted,
Sui, J., Pearlson, G., Caprihan, A., Adali, T., Kiehl, K. A., Liu, J., et al. (2011). provided the original author(s) or licensor are credited and that the original pub-
Discriminating schizophrenia and bipolar disorder by fusing fMRI and DTI lication in this journal is cited, in accordance with accepted academic practice. No
in a multimodal CCA+ joint ICA model. Neuroimage 57, 839–855. doi: use, distribution or reproduction is permitted which does not comply with these
10.1016/j.neuroimage.2011.05.055 terms.

Frontiers in Human Neuroscience www.frontiersin.org November 2014 | Volume 8 | Article 897 | 13