NeuroImage 223 (2020) 117362

Contents lists available at ScienceDirect

NeuroImage
journal homepage: www.elsevier.com/locate/neuroimage

The human corticocortical vestibular network

T.M. Raiser a,b,1, V.L. Flanagin a,b,1, M. Duering b,c,e, A. van Ombergen a,f, R.M. Ruehl a,d,
P. zu Eulenburg a,b,g,∗
a
German Center for Vertigo and Balance Disorders, University Hospital, Ludwig-Maximilians-University Munich, Germany
b
Graduate School of Systemic Neurosciences, Munich, Germany
c
Institute for Stroke and Dementia Research, University Hospital, Ludwig-Maximilians-University Munich, Germany
d
Department of Neurology, University Hospital, Ludwig-Maximilians-University Munich, Munich, Germany
e
Munich Cluster for Systems Neurology (SyNergy), Munich, Germany
f
Department of Translational Neurosciences, Faculty of Medicine and Health Sciences, University of Antwerp, Antwerp, Belgium
g
Institute for Neuroradiology, University Hospital, Ludwig-Maximilians-University Munich, Marchioninistrasse 15, 81377 Munich, Germany

a r t i c l e i n f o a b s t r a c t

Keywords: Background: Little is known about the cortical organization of human vestibular information processing. Instead
Functional network of a dedicated primary vestibular cortex, a distributed network of regions across the cortex respond to vestibular
Structural network input. The aim of this study is to characterize the human corticocortical vestibular network and compare it to
Vestibular system
established results in non-human primates.
Comparative connectomics
Methods: We collected high-resolution multi-shell diffusion-weighted (DWI) and state-of-the-art resting-state
functional MR images of 29 right-handed normal subjects. Ten cortical vestibular regions per hemisphere were
predefined from previous vestibular stimulation studies and applied as regions of interest. Four different structural
corticocortical vestibular networks accounting for relevant constraints were investigated. The analyses included
the investigation of common network measures and hemispheric differences for functional and structural con-
nectivity patterns alike. In addition, the results of the structural vestibular network were compared to findings
previously reported in non-human primates with respect to tracer injections (Guldin and Grusser, 1998).
Results: All structural networks independent of the applied constraints showed a recurring subdivision into iden-
tical three submodules. The structural human network was characterized by a predominantly intrahemispheric
connectivity, whereas the functional pattern highlighted a strong connectivity for all homotopic nodes. A signifi-
cant laterality preference towards the right hemisphere can be observed throughout the analyses: (1) with larger
nodes, (2) stronger connectivity values structurally and functionally, and (3) a higher functional relevance. Simi-
lar connectivity patterns to non-human primate data were found in sensory and higher association cortices rather
than premotor and motor areas.
Conclusion: Our analysis delineated a remarkably stable organization of cortical vestibular connectivity. Differ-
ences found between primate species may be attributed to phylogeny as well as methodological differences. With
our work we solidified evidence for lateralization within the corticocortical vestibular network. Our results might
explain why cortical lesions in humans do not lead to persistent vestibular symptoms. Redundant structural rout-
ing throughout the network and a high-degree functional connectivity may buffer the network and reestablish
network integrity quickly in case of injury.

1. Introduction the past years it has become a research target to build a detailed human
connectome for answering questions in neuroscience (Van Essen et al.,
With the evolution of methods like functional MRI and diffusion- 2013). Disease connectomics (Klauser et al., 2015) and comparative con-
weighted imaging, we are able to build a model of the connections nectomics (e.g. Goulas et al., 2014; van den Heuvel et al., 2016) have
within the human brain from a functional and structural point of view. gained attention for investigating differences and common substrates
The concept of a connectome was introduced by Sporns et al. (2005). In within and among species. In this study we aimed to comprehensively

∗
Corresponding author at: Institute for Neuroradiology, University Hospital, Ludwig-Maximilians-University Munich, Marchioninistrasse 15, 81377 Munich,
Germany.
E-mail address: [email protected] (P. zu Eulenburg).
1
These authors indicates equal contribution.

https://doi.org/10.1016/j.neuroimage.2020.117362
Received 29 October 2019; Received in revised form 2 September 2020; Accepted 4 September 2020
Available online 9 September 2020
1053-8119/© 2020 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

characterize the vestibular corticocortical structural and functional net- 2. Materials and methods
work of the human brain.
Besides whole brain connectomes, researchers work on subnetworks We recruited 29 right-handed participants (age range: 20.5 - 36.7
of several species on various scales. On a cellular level (microscale nm to years, mean age: 27.4 ± 4.3 years, 16 female), from a larger pool of
μm; mesoscale μm to mm), retrograde tracers identify (long) interneu- participants (N~150) that did not have any history of a peripheral or
ronal connections, usually in rodents and non-human primates. These central neuro-otological disease and showed normal vestibular func-
invasive methods can also resolve directionality. On a macroscale level, tion during a clinical head-impulse-test, Romberg stance, Babinski step-
diffusion weighted imaging is used as an indirect and non-invasive mea- ping and Babinski pointing tasks. All participants also had to exhibit
surement of fiber orientations. This method is widely used for research- a detectable torsional nystagmus in the video-oculography surveillance
ing the human brain and cannot detect retrograde or anterograde di- during galvanic vestibular stimulation for the functional localizer ses-
rectionality. Mapping the entire human brain on a cellular level in the sion of the study. All participants gave their written informed con-
living organism is a major technological challenges that is not (yet) real- sent and were paid for participation. The local ethics committee of the
izable in comparison to other species (e.g. mouse connectome, Oh et al., Ludwig-Maximilians-Universität, Munich, approved the study in accor-
2014). dance with the latest revision of the Helsinki declaration from 2008.
Different sensory networks have been characterized over the past The imaging data were acquired on a clinical 3T MRI scanner
years. In non-human primate research visual areas have been defined, (SKYRA Siemens, Erlangen, Germany) with a 64-channel head/neck
connectomes and networks constructed and compared across species coil. During all scans, participants wore ear plugs and a dedicated head
(Orban et al., 2004; Rosa and Tweedale, 2005; Wang et al., 2012). Au- fixation device to limit head motion artifacts during structural and func-
ditory (Misic et al., 2018), pain (Kucyi and Davis, 2015), and olfactory tional imaging (®Pearltec Crania adult, Schlieren, Switzerland).
(Milardi et al., 2017) networks have also been established. A high-resolution T1-weighted image was acquired for each subject
Independent of the modality or methods used, it is essential to de- at the start of the study (TR= 2060 ms, TE= 2.17 ms, flip angle = 12 deg.,
fine the cortical areas of the specific (sensory) network or to choose FoV= 240mm, 320 × 320 matrix, 0.75mm isotropic voxel resolution,
an appropriate parcellation of the brain. In contrast to the auditory or A-P phase encoding, GRAPPA 2). Multi-shell diffusion weighted images
visual system, where unimodal stimulation is feasible, and the system were obtained for three shells of (1)b= 1000 s/mm2 , (2)b= 1750 s/mm2 ,
well characterized, manifold confounds have until recently hampered and (3)b= 2500 s/mm2 at a 1.75mm isotropic spatial resolution. Ten
the definition of the human cortical vestibular system, in particular the non-diffusion and 150 directions encoding scans were obtained with
multisensory nature of its inputs (Ruehl et al., 2017). Electrophysiolog- the following parameters: acquisition matrix 128 × 128, 72 slices, TR=
ical and retrograde tracer studies in the macaque have identified the 3800ms, TE= 104.8ms, FoV= 224 × 224 mm, slice thickness= 1.75 mm,
primary vestibular cortical regions and their connectivity (Guldin and multi-band factor 3 (Feinberg et al., 2010). Reverse phase encoding in
Grusser, 1998). Human homologue areas of these regions have also been a subsequent otherwise identical second run was applied to minimize
identified (Lopez et al., 2012; zu Eulenburg et al., 2012; zu Eulenburg susceptibility-induced distortions during post-processing.
et al., 2013; zu Eulenburg et al., 2018b). Resting state fMRI images were obtained during which participants
The vestibular system supplies us with information about head trans- had their eyes open fixating a dot straight ahead. Here, eye movements
lation, rotation and orientation in a gravitational environment. Interest- were recorded (with the MRC camera built into MRI compatible binoc-
ingly, contrary to other sensory modalities, central vestibular dysfunc- ular goggles; NordicNeuroLab, Bergen, Norway). The images (589 vol-
tion is typically transient, and the respective brain regions, both cortical umes) with whole-brain coverage (2.5mm isotropic resolution) were ac-
and subcortical, are thought to be responsible for these compensation quired in an interleaved slice order with multiband acceleration factor
mechanisms (Brandt et al., 1997). Interestingly, disorders or lesions of 6 (TR = 700 ms, TE = 33 ms, flip angle = 45 °, FoV = 210 mm, matrix
these central vestibular regions can cause vestibular deficits in the acute size = 84 × 84, slice thickness 2.5mm, anterior-posterior phase encod-
phase, but vertigo symptoms never persist in these patients (Brandt and ing direction, prescan normalized). The resting-state scanning session
Dieterich, 2017). This suggests a highly robust and redundant vestibular contained 589 volumes at a TR of 0.7 s (total duration 6 minutes 52
cortical network organization. seconds).
Hemispheric lateralization has been observed in several networks For a functional localizer of the vestibular regions, an additional
like language (Knecht et al., 2000) and the auditory system (Misic et al., fMRI with bimastoidal galvanic vestibular stimulation (GVS) was con-
2018). Dieterich et al. (2003) also found a correlation with handedness ducted. GVS was performed after topical anesthetic (Emla ○ R creme, As-
and vestibular dominance of the ipsilateral hemisphere. Lesions in the traZeneca, Wedel, Germany; 5 g for each side, active substances: 25 mg
right hemisphere seem to cause vestibular symptoms more often than Lidocain and 25 mg Prilocain) was applied to the skin of the mastoid
lesions in the left hemisphere (Eguchi et al., 2019) and symptoms per- bone and surrounding areas and left to soak for at least 45 minutes. Con-
sist longer (e.g. Abe et al., 2012). Very recently it has been shown that trolled by custom programs written in Matlab a 2.5mA sinusoidal stim-
functional lateralization of certain tasks or states correlates negatively ulus with a frequency of 0.9Hz and duration of 10 seconds elicited natu-
with the corpus callosum probability of the connection (Karolis et al., ralistic vestibular sway sensations fifteen times during this session. The
2019). MR images (500 volumes) with whole-brain coverage (2.5mm isotropic
In our current study we aimed for a corticocortical vestibular net- resolution) were acquired in an interleaved slice order with multiband
work on a macroscale with non-invasive diffusion weighted imaging acceleration factor 6 (TR = 700 ms, TE = 33 ms, flip angle = 45 °,
(DWI) and functional connectivity from high-quality multiband resting- FoV = 210 mm, matrix size = 84 × 84, slice thickness 2.5mm, anterior-
state data. The human corticocortical vestibular network was to be com- posterior phase encoding direction).
pared to the tracer-based structural network in non-human primates Prior to data analysis, all data were entered into the MRI Quality
(Guldin and Grusser, 1998). We analyzed the network with respect to Control tool (MRIQC, Version 0.10.3) (Esteban et al., 2017) to exclude
its robustness, vulnerability and how this relates to the hubness of in- any data sets with low quality data. Average framewise displacement
dividual nodes as measured by degree and betweenness centrality. Last (FD) and standardized DVARS were used as the main quality control
but not least a goal of this study was to investigate if there is a lat- measures for the functional sessions. Participants were excluded if any
eralization of the structural and/or functional vestibular corticocorti- single average run FD value exceeded 0.6 mm, if more than 25 % of
cal network and the role of commissural connections via the corpus a participant’s time points from one run were above 0.2 mm FD, or if
callosum. the temporal derivative of the voxel time course variance (DVARS) re-
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

Fig. 1. Vestibular cortical regions. Bilaterally: Area 2v, Area 3av, Premotor, CSv, VIP, OP2, Area 7, VPS, SMA and hMST (in dorsal, right medial and frontal view).

flected more than 5% BOLD signal change (equivalent to a DVARS_nstd 2.2. Structural connectivity analyses
from MRIQC of over 50). None of the participants had to be excluded
(see Supplemental 4 for the motion metrics from the resting-state fMRI 2.2.1. Diffusion weighted data preprocessing and analysis
data). The DW images were visually inspected in all three orthogonal
views and no severe artifacts were revealed, i.e. no signal dropout
2.1. Vestibular Node definition and no gross geometric distortions could be seen. Preprocessing was
performed using MRtrix3 (http://www.mrtrix.org/; version: 3.0_rc3;
Defining robust, biologically motivated regions of interest is essen- Brain Research Institute, Melbourne, Australia), the FMRIB Software
tial for an accurate connectome or network. In this study we targeted Library (FSL; Jenkinson et al., 2012), and Advanced normalization
the vestibular cortical network. Ten cortical regions known in primates tools (ANTs; Avants et al., 2009). Preprocessing steps involved de-
for vestibular processing were motivated from a large group study us- noising (MRtrix ’dwidenoise’), Gibbs artefact removal (MRtrix ’mrdeg-
ing whole-nerve galvanic vestibular stimulation for each hemisphere ibbs’), correcting for eddy currents, motion and susceptibility-induced
(zu Eulenburg et al., 2018a; zu Eulenburg et al., 2018b) (online at Sci- distortions (FSL topup and eddy) and intensity bias corrections (ANTs
enceOpen.com DOI: 10.14293/S2199-1006.1.SOR-.PPCSDUO.v1) and ’N4biascorrection’).
used as regions-of-interest consisting of Area 2v, Area 3av, premotor MRtrix3 was further used for constrained spherical-deconvolution
area, cingulate sulcus visual (CSv), ventral intraparietal area (VIP), OP2, (CSD) and anatomically-constrained whole brain probabilistic tractogra-
Area 7, visual posterior sylvian area (VPS), supplementary motor area phy (Tournier et al., 2007, 2012; Tournier et al., 2019). T1 images were
(SMA) and human medial superior temporal area (hMST) (a detailed registered to the DW images using ANTs (Avants et al., 2009). Segmen-
list of the nodes with size and coordinates can be found in Supplement tation was obtained through generating five-tissue-type (5TT) images
1). We chose this vestibular atlas, instead of a whole-brain parcellation (consisting of: grey matter, white matter, CSF, subcortical grey matter,
such as the Schaefer atlas, because the dedicated vestibular regions often lesions). In none of the participants lesions were detected through the
cover multiple parts of individual parcels, and at the same time belong segmentation. The anatomically informed response function was then
to different networks, such as the salience network, the dorsal attention estimated from the data using the dhollander option (Dhollander et al.,
network and the somatomotor network (Supplemental 1,(Schaefer et al., 2016).
2018)). Fiber orientation distributions (FOD) were estimated from the data
For each subject we took the T1 to DWI co-registered image and with a spherical deconvolution followed by a multi-tissue informed
created a deformation field (T1 to Montreal Neurological Institute (MNI) log-domain intensity normalization. With the anatomically-constrained
space) using the CAT12 toolbox (version: r1432; http://www.neuro.uni- tractography (ACT; Smith et al., 2012) ten million streamlines (=107
jena.de/cat/index.html) in SPM12 (version: 7487). An inverse of the streamlines with a maximum length of 250 mm and a FOD amplitude
spatial deformation field was applied onto the atlas to generate an atlas cutoff of 0.06) were generated. The 5TT anatomical information in this
on single subject level. We then created an intercept between a binarized step results in biologically greater accuracy (Smith et al., 2012). For
grey matter mask (from the 5TT image; grey matter >0.1) and the atlas visual inspection a subset of 105 streamlines was generated for each
on single subject level to guarantee anatomical correctness with respect subject and overlaid on the 5TT images to check for correctness.
to tissue boundaries. For the tractography we used gray and white matter interface
The vestibular atlas in MNI space is visualized in Fig. 1. The volume cropped streamlines, the backtrack option and determined the seed
of the nodes varied across the established vestibular atlas. On average points dynamically with the help of the white matter norm mask. ACT
the right vestibular nodes were 23 % larger than the respective left nodes was used in combination with spherical deconvolution informed filter-
(t(9)=2.27, p< 0.05). (Fig. 2). ing of tractograms (SIFT2) to further improve the biological accuracy of
We compared this atlas with the subject’s and the cohort’s cortical streamlines (Smith et al., 2012, 2013, 2015).
response to galvanic vestibular stimulation (GVS). We calculated a sim-
ilarity map for the group and for each single-subject response between 2.2.2. Establishment of a structural corticocortical vestibular network
the GVS activation pattern and the vestibular atlas and calculated the (sCVN)
respective DICE overlap. The results can be found in Supplement 2. We By means of combining the delineated 10 million streamlines and the
also used the individual ROIs to calculate both the structural and func- above described vestibular atlas, we constructed corticocortical vestibu-
tional vestibular network connectivity, and modularity (see Supplement lar networks for each subject. Using the calculated weightings after
3), and found similar results to the atlas-based approach. SIFT2 filtering, we created symmetric, undirected structural connectiv-
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

Fig. 2. Visualization of parts of the pipeline

analyses. The tractogram as well as the adja-
cency matrices of the network were generated
with MRtrix3 (detailed pipeline commands can
be found in the supplement). Visualized is data
of one representative subject. The functional
data were generated with the CONN toolbox.
The brain connectivity toolbox was used for
generating network measures. AUC = Area un-
der the curve.

ity matrices of all cerebral nodes with a diagonal set to zero. When not zero (Rubinov and Sporns, 2010, 2011). No further thresholds were ap-
correcting for the distance between nodes, short-range connections are plied to the functional network.
typically overestimated and local and global network statistics are not
preserved (Betzel et al., 2019). We therefore scaled the weights by ei- 2.4. Brain connectivity analyses
ther inverse volume and/or inverse streamline length ((Hagmann et al.,
2008)). This resulted in an unscaled version of the network as well as 2.4.1. Network modularity
a streamline length scaled, an inverse node volume scaled and a scaled The weighted undirected networks were all ordered by applying the
combined network. The focus of our analyses was the structural scaled community Louvain algorithm (Blondel et al., 2008) of the brain con-
combined network. The three additional structural networks were es- nectivity toolbox (BCT, version 2017-15-01; Rubinov and Sporns, 2010)
tablished to see whether the modularity of the network is stable across iteratively onto the mean matrices of the four different scaled versions
the differing constraints. of the structural network as well as the functional network. With looping
over several gamma values (ranging from 0 to 2.5) we detected a stable
number of modules. The networks on single subject level were ordered
2.3. Establishment of a functional corticocortical vestibular network accordingly. As we are not comparing two groups, we normalized the
(fCVN) data for each subject to make the networks comparable to each other.
The presence of a normal distribution of certain network measures was
A functional connectivity analyses was performed with the rsfMRI tested with the Anderson Darling test and then pairwise t-tests were per-
data using CONN toolbox (version 18b; Whitfield-Gabrieli and Nieto- formed to precisely test for significant differences between the right and
Castanon, 2012; http://www.nitrc.org/projects/conn). Movement pa- the left hemispheres.
rameters, quality control timeseries and scrubbing regressors were em-
ployed by the CONN Toolbox as first level covariates. For denoising we 2.4.2. Network measures
used the following nuisance regressors’ time series to lessen their im- In addition to the modularity, we were interested in network mea-
pact: white matter (WM) and cerebrospinal fluid (CSF) confounds were sures that reflect the resilience of nodes in the network, and how this
considered with their first five principal components each. Furthermore, relates to the hubness of the individual nodes. This is of particular im-
six principal temporal components of the movement parameters (three portance for understanding the central vestibular compensation mecha-
translation and three rotation parameters) as well as their six deriva- nisms after vestibular loss or dysfunction (Brandt and Dieterich, 2017).
tives were employed. Images were denoised with a temporal band-pass The following measures were only calculated for the structural network
filter (0.008-0.09 Hz). Scrubbing was selected as an option with default with stream length and inverse node volume scaled values, as well as
parameters to remove any images with higher motion, but no volumes the functional network: vulnerability, degree centrality and between-
in any participant were selected for scrubbing. The ROI-to-ROI analysis ness centrality.
resulted in the functional weighted connectivity matrices with Fisher- To investigate the resilience of the network communication, we mea-
transformed correlation coefficients. These results were thresholded at sured the vulnerability as described in Iturria-Medina et al. (2008). The
p<0.05 FDR corrected (two-sided) together with 1000 permutations for vulnerability Vi was calculated for each node i by subtracting the global
non-parametric thresholding in the CONN toolbox. This fCVN was used efficiency without node i from the global efficiency, and then dividing
for detecting a stable number of modules within the network. For the it by the global efficiency. The global efficiency was calculated for each
subsequent network analyses, we set any negative correlation values to node using the BCT. As vulnerability values can be both negative and
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

Fig. 3. Mean normalized structural matrices in community Louvain order: the unscaled, the inverse node volume scaled, and the stream length scaled network are
listed on the left. The combined scaled network is shown bigger on the right. Higher values depict greater connectivity between regions. R corresponds to nodes in
right hemisphere, L to the ones in the left hemisphere. Order of the node labels is identical for all four networks. White frames indicate stable modules assigned by
the community Louvain algorithm.

positive and can change sign when applying different thresholds, we nections. We considered connections with values between 0.6 and 1 as
calculated the vulnerability for the adjacency matrices without thresh- densely connected and values ranging from 0.2 to 0.6 as moderately
olding them. Positive Vi values give rise to how important node i is for dense connected. As we cannot detect directionality with our measures,
the global efficiency and removing node i harms the network efficiency. we also extracted the non-human primate data without directionality
Negative values on the other hand indicate that the node is reducing the from the literature. In non-human primates the connections were only
global efficiency. Values around 0 denote that the global efficiency does reported as groups of either densely or moderately dense connected
not change when node i is damaged or removed from the network. nodes (Guldin and Grusser, 1998). For the densely connected nodes, we
To understand how the connectedness of individual nodes related to chose a value of 0.8 and for the moderately dense connections a value
its vulnerability, we also calculated the degree centrality (DC) and be- of 0.4 for visualization purposes. Within the scope of the non-human
tweenness centrality (BC), and can be combined with vulnerability. For primate tracer studies the majority but not all known cortical vestibu-
every node we calculated the DC using the BCT after applying different lar nodes have been investigated to date (e.g. periarcuate cortex, VIP)
thresholds onto the network, from 0 to 1 in 0.1 steps (corresponding to leaving us with eight out of ten nodes and its connections to compare
0 to 0.76 correlation coefficients in the functional data). Putting thresh- descriptively. Area 3av was separated into Area 3aHv and Area 3aNV in
olds on the matrices lowers the likelihood of false positive and false the primate study but was combined to one region in our analysis.
negative connections (de Reus and van den Heuvel, 2013). The area un-
derneath the curve (AUC) of the degree centrality was calculated for 3. Results
each node with jackknife resampling of all subjects. To further investi-
gate the homotopic organization of the brain, we performed a pair-wise 3.1. Characteristics of the structural CVN
t-test of the AUC values for left and right brain regions. For each node
we further calculated the mean BC. For BC we calculated the AUC as A stable subdivision into three modules were generated consistently
described for DC. across all structural corticocortical vestibular networks (sCVN): (1) the
left ROIs: Area 2v, 3av, IPS3, OP2, 7, VPS, hMST, (2) the right ROIs:
2.5. Comparison with primate data Area 2v, 3av, IPS3, OP2, 7, VPS, hMST, and (3) bilaterally the premotor
cortex, CSV and SMA (for the following gamma values: unscaled 0.59 ≤
We compared our human structural scaled combined network to the gamma ≤ 1.1; inverse node volume scaled 0.31≤ gamma ≤ 0.9; stream
connections reported in non-human primates of the homotopic vestibu- length scaled 0.88 ≤ gamma ≤ 1.05; combined 0.59 ≤ gamma ≤ 1.1;
lar ROIs. We took the mean values of the homotopic human nodes that Fig. 3). In all four networks the second module with right hemispheric
led to a mean 10 × 10 structural network without hemispheric infor- nodes reveals higher intensities than the first module with the respec-
mation. The connections between the analogous ROIs in primates, as tive left hemispheric nodes. Hemispheric differences are addressed in a
defined in (Smith et al., 2017), were all generated via tracer projec- separate section below. Volume correction is obvious when comparing
tions (Guldin and Grusser, 1998), whereas our results originate from the inverse volume scaled network to the unscaled version. Connections
noninvasive imaging methods. Hence, we applied a filtering normalized between Area 2v and 3av gain weight as the nodes are relatively small
threshold of 0.2 on our structural network to eliminate spurious con- in size and we corrected for the inverse of the node volume. Relatively
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

Fig. 4. Connectograms: (left) mean values across hemispheres of the normalized and weighted structural stream length and inverse node volume scaled network of
humans with a threshold at 0.2, grey color indicates that a homologous region/connection was not taken into account in primates; human brain template in MNI
space 3Drendered with MRIcroGL; (right) primate dense (thicker lines) and moderate dense (thinner lines) projections (right) between the respective homologous
nodes reported by Guldin and Grusser (1998); image of the macaque brain was acquired from the INIA19 Primate Brain Atlas; brain pictures are not scaled according
to their relative size.

large nodes’ (e.g. premotor) connectivity values do not change dras- OP2 (except a moderately dense connection to Area 3av (0.21) and to
tically with inverse node volume correction. Further, the influence of VPS (0.32)).
correcting for the stream length is evident when comparing all four dif-
ferently constrained networks. For example, Area VPS and Area 7 are in 3.3. Hemispheric differences
very close proximity in the brain and receive a highly dense connection
attributed in those networks, which do not correct for stream length We proceeded by comparing the left and right hemispheric node
(i.e. unscaled and inverse node volume scaled). The same holds true for values of the human streamline strength and inverse node volume
Area 2v and Area 3av. Stream length correction did not affect longer scaled network. The sum of the unthresholded intrahemispheric con-
connections (e.g. Premotor – hMST) as much as shorter ones. nections for each node was calculated for each hemisphere across sub-
The premotor regions are overall densely connected within the ac- jects. The connectivity density values followed a normal distribution and
cording hemisphere and are highly connected with each other, as well were significantly larger for the right hemisphere compared to the left
as with the SMA of both hemispheres. Right and left SMA are densely (F(1,28) = 9.467 p = 0.0022).
interconnected and after correcting for stream length, SMA is ranked Repeated measures analyses of AUC DC and AUC BC for the in-
the most densely connected node of the network. verse node volume and stream length scaled network showed signifi-
cantly higher values for the right hemisphere (F(1,28) = 3649.7, p ~ 0,
F(1,28) = 194.2, p ~ 0, respectively). The structural vulnerability was
3.2. Homotopic organization of the network
also higher for the right hemisphere (F(1,28) = 11.069, p=0.000937).

Structurally we compared our results of the streamline strength and 3.4. Network measures
inverse node volume scaled network to connections reported in pri-
mates (Guldin and Grusser, 1998). We identified overlap and differ- Premotor and SMA areas are characterized by relatively high vulner-
ences in connections between the vestibular nodes in human and non- ability values for the network (between 0.13 and 0.2) in comparison to
human primates (Fig. 4). Unfortunately, we could not compare all con- the vulnerability values of the rest of the nodes that accumulate around
nections of the vestibular nodes as two of the vestibular defined nodes 0 (Fig. 5). Even though the SMA (node 17 and 18) displays a lower
(IPS3/VIP and periarcuate/SMA) were not reported in the tracer study AUC BC than the premotor areas (node 5 and 6), the vulnerability is
of Guldin and Grusser (1998). On the one hand, we found some similar- comparably high (between 0.1 and 0.2). Premotor and SMA areas are
ities between the two species in the vestibular structural networks. Two characterized by a relatively high AUC DC in both hemispheres. The
connections with very high density described in the human sCVN are premotor areas are additionally characterized by a high AUC BC.
also reported to be densely connected in primates: Area 2v – Area 3av, The remaining nodes cluster with relatively low centrality (AUC DC
and VPS – Area 7. A rather moderately dense connection in humans 0.75-1.5, AUC BC 1.5-6) and low vulnerability values around 0 and none
of VPS and OP2 (0.33) is reported as a dense connection in primates of the nodes show lower values than -0.05 for the vulnerability.
(VPS-PIVC).
On the other hand, there is quite some difference in the reported 3.5. Characteristics of the functional CVN
connections. A moderate connection between hMST and the human pre-
motor cortex (0.45), as well as hMST and Area 7 (0.23) are not reported The functional network was divided into seven stable modules for
in primates. A dense connection of MST and VPS in primates is not ob- 1.41 ≤ gamma ≤ 1.68 with the iteratively performed community Lou-
servable in the thresholded human connectogram. Further the densely vain algorithm. Modules consist of bilateral (1) Area2v and Area 3aV,
connected PIVC in primates with several regions (VC, Area 6, Area 3av, (2) Premotor Area, (3) CSv and VIP, (4) OP2, (5) Area 7, (6) VPS, and
Area 2v, VPS and Area 7) are mostly absent in the human equivalent (7) SMA and hMST (Fig. 6).
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

No significant differences in the correlation coefficient between

hemispheres was found. Investigating centrality measures a clustering of
nodes is visible in the upper right corner of the graph (AUC DC between
3.9 and 5.9; AUC BC between 7.1 and 14.3). Both SMA regions (17 & 18)
are spatially separated from this cluster with lower AUC DC and AUC BC
values (see Fig. 6). For the functional network AUC DC and AUC BC did
not show any hemispheric differences, however in general the higher
the AUC DC and the higher the AUC BC, the higher the vulnerability
of the nodes. Functional vulnerability was not significantly different be-
tween the two hemispheres. Right hemispheric areas VIP and VPS show
the highest vulnerability in the functional network.

4. Discussion

The current results provide novel insights into the corticocortical

organization of vestibular information processing in the human brain.
We investigated the structural corticocortical vestibular network in hu-
Fig. 5. Network measures for the stream length and inverse node volume scaled mans across all established constraints (i.e. unscaled, streamline length
structural network. Color code visualizes the vulnerability of each node. Odd scaled, inverse node volume scaled and a scaled combined network).
numbers are assigned to left hemispheric nodes, even numbers to right hemi- We then compared the structural vestibular network with tracer-based
spheric nodes (1-2: Area 2v, 3-4: Area 3av, 5-6: Premotor, 7-8: CSv, 9-10: VIP, findings in non-human primates (Guldin and Grusser, 1998). We dis-
11-12: OP2, 13-14: Area 7, 15-16: VPS, 17-18: SMA and 19-20: hMST). covered clear differences between the human and non-human primate

Fig. 6. (upper left) Network measures AUC Degree centrality, AUC betweenness centrality and vulnerability for all nodes of the functional network. Odd numbers are
assigned to left hemispheric nodes, even numbers to right hemispheric nodes (1-2: Area 2v, 3-4: Area 3av, 5-6: Premotor, 7-8: CSv, 9-10: VIP, 11-12: OP2, 13-14: Area
7, 15-16: VPS, 17-18: SMA and 19-20: hMST); (upper right) Mean matrix of the functional community Louvain ordered network. Color of the matrix corresponds to
the clustering coefficients between nodes. For further analyses negative values were set to zero. (lower left) Connectogram of the functional network thresholded at
0.2, grey color indicates that a homologous primate region/connection was not considered; (lower middle) connectogram of the fCVN with hemispheric information
and a threshold at 0.4. For comparison reasons the sCVN connectogram with hemispheric information is shown on the lower right thresholded at 0.1. R corresponds
to right hemisphere nodes, L to the left hemisphere nodes.
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

vestibular structural connections. The robustness of the vestibular net- 4.3. Structural vs. functional CVN
work may explain why cerebral stroke patients with vertigo symptoms
at onset never develop a chronic vestibular dysfunction. Our main goal of this work was to characterize the vestibular net-
work itself without comparing it to other sensory subnetworks. As-
suming DWI-based tractography provides an accurate assessment of
4.1. Robust modularity of the cortical vestibular network across constraints
the streamline fiber orientations, our combined structural and func-
tional results suggest that the central vestibular information flow in hu-
Independent of the constraints and scaling that were put on the struc-
mans works bottom-up through projection and association fibers with-
tural networks, we consistently found three modules with identical com-
out many structural commissural connections and information flow be-
position. These modules were found irrespective of correcting or not cor-
tween the hemispheres. Structurally, we observed a robust intrahemi-
recting for stream length or node volume. These findings underpin the
spheric connectivity in most of the vestibular regions. Two modules con-
stability and robustness of the CVN network. Neither of the constraints
sisting of seven brain regions were found in both hemispheres (Area 2v,
resulted in a different separation of modules. The first two modules are
3av, IPS3, OP2, 7, VPS and hMST). Functionally however, all ten ho-
homogenous for both hemispheres and may therefore be responsible for
motopic nodes of the left and the right hemisphere were strongly con-
similar specialized processes. The third module of the structural CVN,
nected, despite the absence of structural connections (except premotor
consists of the premotor cortices, the SMA and CSv, and therefore rep-
and SMA regions). Therefore, we hypothesize that vestibular informa-
resents general sensorimotor processing. These modules did not divide
tion flow might spread bottom-up from either the brainstem or thalamic
the vestibular regions into specific known functional networks, such as
regions creating a high connectivity in the homotopic cortical nodes.
the somatomotor or dorsal attention network (Schaefer et al., 2018) but
formed modules across networks.
4.4. Cerebral laterality in the CVN
4.2. The corticocortical vestibular network in human and non-human
Our results suggest that the CVN has a right-sided dominance.
primates
Firstly, we observed this in the volume of the homotopic regions.
Secondly, across all four structural networks, the connections were
We quantified the density of the connections between our prede-
significantly stronger in the right hemisphere compared to the left.
fined ROIs and compared them to previously reported connections in
Karolis et al. (2019) recently investigated the functional lateralization
the macaque brain (Guldin and Grusser, 1998). Directly neighboring re-
in combination with structural corpus callosum connectivity across four
gions of interest are shown to be (densely) connected in both species
different axes including communication, emotion, perception/action
(Area2v - Area3av, OP2/PIVC - VPS, VPS - Area7). The first two mod-
and decision. The vestibular network was not part of the laterality pro-
ules of the sCVN show similar connectivity to the macaque, suggesting
file. Our results might help to close this gap: Within our vestibular net-
a homology for the vestibular system between primate species.
work only two brain structures, premotor and SMA, were presented with
We additionally found clear differences between the two species’ net-
interhemispheric connections via the corpus callosum and there seems
works. In the macaque brain, the PIVC is densely connected within the
to be a lateralization to the right. These results are in line with the re-
vestibular network. Dense structural connections of the human homol-
ported results of Karolis et al. (2019) tested with other sensory modali-
ogous OP2 region with other vestibular regions of interest are scarce.
ties.
Functionally however, human area OP2 is very well interconnected with
other vestibular regions. Indeed, the whole functional connectogram of
humans shares more similarities with the macaques structural connec- 4.5. Remission of central vestibular symptoms due to cortical lesions
togram than the structural network of humans (but see following para-
graph for potential limitations). However, even within the same species, In about 5% of patients with dizziness, imbalance or vertigo symp-
a structural connection does not imply functional connectivity and vice toms an acute ischemic stroke is diagnosed (Kim et al., 2018). With le-
versa. Consequently, interpretations and comparisons of structural and sions in one node or one connection of the vestibular network patients
functional connectivity should be treated with caution. can suffer from vertigo in the acute stage prior and shortly after stroke
While the clear differences between the structural networks of hu- (see review Brandt and Dieterich, 2017). Chronic central vestibular dis-
mans and macaques might be due to the investigation of two different orders however do not exist. That was the motivation for us to investi-
species, they might as well arise due to the different methods used to gate the degree of vulnerability within the cortical vestibular network.
obtain the structural networks. DWI-based tractography cannot detect Across our delineated networks we found characteristics, that might at
the underlying anatomical connectivity on a micro scale and can only least partly explain the generally transient nature of vestibular symp-
give us an estimation of the fiber orientation. In general, the interpre- toms following a cerebral stroke. Hereto, the high degree of preexisting
tation of streamlines must be treated with caution: the resolution is too intrahemispheric structural connectivity might help to overcome lesion-
poor to detect the orientation of all fibers (Jones et al., 2013). With bet- associated symptoms occuring from stroke or other etiologies. In the
ter resolution the chance of multiple orientations of fibers within one acute state, functional connectivity might be reduced between homo-
voxel decreases, but it will not reach the level of cellular organization. topic regions leading to vertigo. But we infer from our results that de-
Conclusions about directionality cannot be drawn with DWI and graph pending on which edge or node is damaged, this information flow can
theoretical methods (Klauser et al., 2015). Hence, at the moment, we be quickly rebuilt through an alternative route. The structural cortico-
lack the appropriate in vivo methods to quantify and compare connec- cortical network with the detected modules forms a good basis for this
tions from the human brain to other species. Tracer studies, although on a corticocortical level.
they have the cellular specificity, only a limited number of injections Our findings with respect to functional vulnerability being signif-
possible per animal and it is time-consuming work, and therefore all icantly higher in right hemispheric nodes are in line with previously
connections are typically not studied. Within our pool of regions two reported higher prevalence of vestibular symptoms due to right hemi-
regions SMA and VIP were not reported in the macaques’ network by spheric lesions (Eguchi et al., 2019). Further, Abe and colleagues re-
Guldin and Grusser (1998). More work is needed to compare in vivo MRI ported longer rehabilitation time in the Pusher Syndrome (a disorder
and tracer injections within the same species across species to poten- of postural balance that manifests as a pushing away toward the con-
tially overcome this obstacle as already pointed out by van den Heuvel tralesional side in unilateral stroke) following right hemispheric lesions
et al. (2016). Limits of DWI in direct comparison with tracer injections (2012). Whether there is a general difference in remission of vestibular
are discussed elsewhere (Donahue et al., 2016; Reveley et al., 2015). symptoms depending on the lesion site needs to be further investigated
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

by comparing vulnerability profiles of the nodes with lesion mapping Supplementary materials
and vestibular symptom status.
Supplementary material associated with this article can be found, in
4.6. Methodological considerations the online version, at doi:10.1016/j.neuroimage.2020.117362.

References
We want to emphasize the quality of our generated dataset as we did
not only include a reasonable number of subjects, but also guaranteed Abe, H., Kondo, T., Oouchida, Y., Suzukamo, Y., Fujiwara, S., Izumi, S., 2012. Prevalence
a state-of-the-art DWI data quality. DWI data quality benefitted from and length of recovery of pusher syndrome based on cerebral hemispheric lesion side
three shells (b-values of 1000, 1750 & 2500 s/mm2 ) and 150 directions in patients with acute stroke. Stroke 43, 1654–1656.
Avants, B., Tustison, N., Song, G., 2009. Advanced normalization tools, ANTS 1.0. Source-
certainly improved the sensitivity of the fiber orientation (Jones et al.,
forge. Jun.
2013). Previous studies of the vestibular system that have used sin- Betzel, R.F., Griffa, A., Hagmann, P., Mišić, B., 2019. Distance-dependent consensus
gle shell DWI data on bias-field prone scanners at lower resolutions thresholds for generating group-representative structural brain networks. Netw. Neu-
than ours, make ambitious claims with regard to the corticothalamic rosci. 3, 475–496.
Blondel, V.D., Guillaume, J.-L., Lambiotte, R., Lefebvre, E., 2008. Fast unfolding of com-
or even brain stem-cortex connections within the vestibular network munities in large networks. J. Stat. Mech. Theory Exp. 2008, P10008.
(e.g. Kirsch et al., 2016; Wirth et al., 2018). We fear that the inves- Brandt, T., Dieterich, M., 2017. The dizzy patient: don’t forget disorders of the central
tigators were neglecting limitations of the signal-quality of their data vestibular system. Nat. Rev. Neurol. 13, 352–362.
Brandt, T., Strupp, M., Arbusow, V., Dieringer, N., 1997. Plasticity of the vestibular system:
and the applied analytical methods. Because of this awareness of the central compensation and sensory substitution for vestibular deficits. Adv. Neurol. 73,
profound limitations of DWI for deeper brain structures, we completely 297–309.
abstained from investigating any subcortical connectivity patterns. We de Reus, M.A., van den Heuvel, M.P., 2013. Estimating false positives and negatives in
brain networks. Neuroimage 70, 402–409.
also refrained from a whole-brain network analysis, and instead focussed Dhollander, T., Raffelt, D., Connelly, A., 2016. Unsupervised 3-tissue response function
specifically on the cortical vestibular brain regions because no whole- estimation from single-shell or multi-shell diffusion MR data without a co-registered
brain parcellation currently available represented the vestibular brain T1 image. ISMRM Workshop on Breaking the Barriers of Diffusion MRI.
Dieterich, M., Bense, S., Lutz, S., Drzezga, A., Stephan, T., Bartenstein, P., Brandt, T., 2003.
regions with the specificity and anatomical precision that our vestibu- Dominance for vestibular cortical function in the non-dominant hemisphere. Cereb.
lar atlas does. We therefore focused this analysis on relative comparisons Cortex 13, 994–1007.
between different networks and between hemispheres. Donahue, C.J., Sotiropoulos, S.N., Jbabdi, S., Hernandez-Fernandez, M., Behrens, T.E.,
Dyrby, T.B., Coalson, T., Kennedy, H., Knoblauch, K., Van Essen, D.C., Glasser, M.F.,
2016. Using diffusion tractography to predict cortical connection strength and dis-
Conclusion tance: a quantitative comparison with tracers in the monkey. J. Neurosci. 36,
6758–6770.
With the present work, we described and characterized the cortico- Eguchi, S., Hirose, G., Miaki, M., 2019. Vestibular symptoms in acute hemispheric strokes.
J. Neurol..
cortical vestibular network. Modularity of the human structural cortic- Esteban, O., Birman, D., Schaer, M., Koyejo, O.O., Poldrack, R.A., Gorgolewski, K.J., 2017.
ocortical vestibular network is extremely robust. There is little overlap MRIQC: Advancing the automatic prediction of image quality in MRI from unseen
between the structural vestibular network in humans and monkeys. The sites. PLoS One 12, e0184661.
Feinberg, D.A., Moeller, S., Smith, S.M., Auerbach, E., Ramanna, S., Gunther, M.,
human structural corticocortical vestibular network features a stronger Glasser, M.F., Miller, K.L., Ugurbil, K., Yacoub, E., 2010. Multiplexed echo planar
intrahemispheric connectivity whereas the resting-state derived cortico- imaging for sub-second whole brain FMRI and fast diffusion imaging. PLoS One 5,
cortical vestibular network emphasizes a substantial functional connec- e15710.
Goulas, A., Bastiani, M., Bezgin, G., Uylings, H.B., Roebroeck, A., Stiers, P., 2014. Com-
tivity of homotopic nodes. A laterality preference for the right hemi- parative analysis of the macroscale structural connectivity in the macaque and human
sphere was observable in vestibular processing for the functional and brain. PLoS Comput. Biol. 10, e1003529.
structural analysis alike (stronger connectivity values, larger anatomi- Guldin, W.O., Grusser, O.J., 1998. Is there a vestibular cortex? Trends Neurosci. 21,
254–259.
cal nodes, higher functional vulnerability in the right cerebral cortex). Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R., Honey, C.J., Wedeen, V.J., Sporns, O.,
There are obviously still many steps pending to fully understand the 2008. Mapping the structural core of human cerebral cortex. PLoS Biol. 6, e159.
cortical vestibular organization and its embedding in and contribution Iturria-Medina, Y., Sotero, R.C., Canales-Rodriguez, E.J., Aleman-Gomez, Y., Melie-Gar-
cia, L., 2008. Studying the human brain anatomical network via diffusion-weighted
to the whole-brain network. We are optimistic that our results and sub-
MRI and graph theory. Neuroimage 40, 1064–1076.
sequent characterizations of the vestibular and adjacent subnetworks Jenkinson, M., Beckmann, C.F., Behrens, T.E., Woolrich, M.W., Smith, S.M., 2012. Fsl.
will further our understanding of the vestibular system in health and Neuroimage 62, 782–790.
disease. Jones, D.K., Knosche, T.R., Turner, R., 2013. White matter integrity, fiber count, and other
fallacies: the do’s and don’ts of diffusion MRI. Neuroimage 73, 239–254.
Karolis, V.R., Corbetta, M., Thiebaut de Schotten, M., 2019. The architecture of functional
Declaration of Competing Interest lateralisation and its relationship to callosal connectivity in the human brain. Nat.
Commun. 10, 1417.
Kim, Y., Faysel, M., Balucani, C., Yu, D., Gilles, N., Levine, S.R., 2018. Ischemic stroke
None. predictors in patients presenting with dizziness, imbalance, and vertigo. J. Stroke
Cerebrovasc. Dis. 27, 3419–3424.
Acknowledgement Kirsch, V., Keeser, D., Hergenroeder, T., Erat, O., Ertl-Wagner, B., Brandt, T., Dieterich, M.,
2016. Structural and functional connectivity mapping of the vestibular circuitry from
human brainstem to cortex. Brain Struct. Funct. 221, 1291–1308.
We thank all participants for taking part in the experiment. Further, Klauser, P., Whittle, S., Simmons, J.G., Byrne, M.L., Mundy, L.K., Patton, G.C., Fornito, A.,
we thank Mathias Hübner for his technical MRI facility assistance, Judita Allen, N.B., 2015. Reduced frontal white matter volume in children with early onset
of adrenarche. Psychoneuroendocrinology 52, 111–118.
Huber and Alex Knorr for methodological discussions and Judita Huber Knecht, S., Deppe, M., Drager, B., Bobe, L., Lohmann, H., Ringelstein, E., Henningsen, H.,
for calculating the overlap with the Schaefer400 atlas. 2000. Language lateralization in healthy right-handers. Brain 123 (Pt 1), 74–81.
Kucyi, A., Davis, K.D., 2015. The dynamic pain connectome. Trends Neurosci. 38, 86–95.
Lopez, C., Blanke, O., Mast, F.W., 2012. The human vestibular cortex revealed by co-
Funding
ordinate-based activation likelihood estimation meta-analysis. Neuroscience 212,
159–179.
This work was supported by the Federal Ministry of Education and Milardi, D., Cacciola, A., Calamuneri, A., Ghilardi, M.F., Caminiti, F., Cascio, F., An-
dronaco, V., Anastasi, G., Mormina, E., Arrigo, A., Bruschetta, D., Quartarone, A.,
Research (BMBF 01 EO 0901) and a stipend of the Graduate School of
2017. The olfactory system revealed: non-invasive mapping by using constrained
Systemic Neurosciences grant DFG-GSC 82/3 (to TMR), a research fel- spherical deconvolution tractography in healthy humans. Front. Neuroanat. 11, 32.
lowship (11U6414N and 11U6416N) and a research grant to stay abroad Misic, B., Betzel, R.F., Griffa, A., de Reus, M.A., He, Y., Zuo, X.N., van den Heuvel, M.P.,
(V400717N) by the Research Foundation Flanders/FWO (to AVO), an Hagmann, P., Sporns, O., Zatorre, R.J., 2018. Network-based asymmetry of the human
auditory system. Cereb. Cortex 28, 2655–2664.
Amelia Earhart fellowship by Zonta International (to AVO) and BELSPO Oh, S.W., Harris, J.A., Ng, L., Winslow, B., Cain, N., Mihalas, S., Wang, Q., Lau, C., Kuan, L.,
Prodex (to AVO). Henry, A.M., Mortrud, M.T., Ouellette, B., Nguyen, T.N., Sorensen, S.A., Slaughter-
T.M. Raiser, V.L. Flanagin and M. Duering et al. NeuroImage 223 (2020) 117362

beck, C.R., Wakeman, W., Li, Y., Feng, D., Ho, A., Nicholas, E., Hirokawa, K.E., Sporns, O., Tononi, G., Kotter, R., 2005. The human connectome: a structural description
Bohn, P., Joines, K.M., Peng, H., Hawrylycz, M.J., Phillips, J.W., Hohmann, J.G., of the human brain. PLoS Comput. Biol. 1, e42.
Wohnoutka, P., Gerfen, C.R., Koch, C., Bernard, A., Dang, C., Jones, A.R., Zeng, H., Tournier, J.D., Calamante, F., Connelly, A., 2007. Robust determination of the fibre orien-
2014. A mesoscale connectome of the mouse brain. Nature 508, 207–214. tation distribution in diffusion MRI: non-negativity constrained super-resolved spher-
Orban, G.A., Van Essen, D., Vanduffel, W., 2004. Comparative mapping of higher visual ical deconvolution. Neuroimage 35, 1459–1472.
areas in monkeys and humans. Trends Cogn. Sci. 8, 315–324. Tournier, J.D., Calamante, F., Connelly, A., 2012. MRtrix: diffusion tractography in cross-
Reveley, C., Seth, A.K., Pierpaoli, C., Silva, A.C., Yu, D., Saunders, R.C., Leopold, D.A., ing fiber regions. Int. J. Imaging Syst. Technol. 22, 53–66.
Ye, F.Q., 2015. Superficial white matter fiber systems impede detection of long-range Tournier, J.D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M., Christi-
cortical connections in diffusion MR tractography. Proc. Natl. Acad. Sci. U S A 112, aens, D., Jeurissen, B., Yeh, C.H., Connelly, A., 2019. MRtrix3: A fast, flexible and
E2820–E2828. open software framework for medical image processing and visualisation. Neuroim-
Rosa, M.G., Tweedale, R., 2005. Brain maps, great and small: lessons from comparative age, 116137.
studies of primate visual cortical organization. Philos. Trans. R. Soc. Lond. B Biol Sci van den Heuvel, M.P., Bullmore, E.T., Sporns, O., 2016. Comparative connectomics.
360, 665–691. Trends Cogn. Sci. 20, 345–361.
Rubinov, M., Sporns, O., 2010. Complex network measures of brain connectivity: uses and Van Essen, D.C., Smith, S.M., Barch, D.M., Behrens, T.E., Yacoub, E., Ugurbil, K., Con-
interpretations. Neuroimage 52, 1059–1069. sortium, W.U.-M.H., 2013. The WU-Minn human connectome project: an overview.
Rubinov, M., Sporns, O., 2011. Weight-conserving characterization of complex functional Neuroimage 80, 62–79.
brain networks. Neuroimage 56, 2068–2079. Wang, Q.X., Sporns, O., Burkhalter, A., 2012. Network analysis of corticocortical connec-
Ruehl, R.M., Stephan, T., Dieterich, M., zu Eulenburg, P., 2017. Towards a human vestibu- tions reveals ventral and dorsal processing streams in mouse visual cortex. J. Neurosci.
lar cortex – Manifold confounders hamper the delineation of vestibular responses in 32, 4386–4399.
functional neuroimaging. Clin. Neurophysiol. 128, e331–e332. Whitfield-Gabrieli, S., Nieto-Castanon, A., 2012. Conn: a functional connectivity toolbox
Schaefer, A., Kong, R., Gordon, E.M., Laumann, T.O., Zuo, X.N., Holmes, A.J., Eick- for correlated and anticorrelated brain networks. Brain Connect. 2, 125–141.
hoff, S.B., Yeo, B.T.T., 2018. Local-global parcellation of the human cerebral cortex Wirth, A.M., Frank, S.M., Greenlee, M.W., Beer, A.L., 2018. White matter connectivity of
from intrinsic functional connectivity MRI. Cereb. Cortex 28, 3095–3114. the visual-vestibular cortex examined by diffusion-weighted imaging. Brain Connect.
Smith, A.T., Greenlee, M.W., DeAngelis, G.C., Angelaki, Dora E., 2017. Distributed visual– 8, 235–244.
vestibular processing in the cerebral cortex of man and macaque. 30, 91. zu Eulenburg, P., Caspers, S., Roski, C., Eickhoff, S.B., 2012. Meta-analytical definition and
Smith, R.E., Tournier, J.D., Calamante, F., Connelly, A., 2012. Anatomically-constrained functional connectivity of the human vestibular cortex. Neuroimage 60, 162–169.
tractography: improved diffusion MRI streamlines tractography through effective use zu Eulenburg, P., Muller-Forell, W., Dieterich, M., 2013. On the recall of vestibular sen-
of anatomical information. Neuroimage 62, 1924–1938. sations. Brain Struct. Funct. 218, 255–267.
Smith, R.E., Tournier, J.D., Calamante, F., Connelly, A., 2013. SIFT: spherical-deconvolu- zu Eulenburg, P., Stephan, T., Dieterich, M., Ruehl, R.M., 2018a. The human vestibular
tion informed filtering of tractograms. Neuroimage 67, 298–312. cortex. OHBM 2018 Organization for Human Brain Mapping, Republic of Singapore.
Smith, R.E., Tournier, J.D., Calamante, F., Connelly, A., 2015. SIFT2: enabling dense quan- zu Eulenburg, P., Stephan, T., Dieterich, M., Ruehl, R.M., 2018b. The Human Vestibular
titative assessment of brain white matter connectivity using streamlines tractography. Cortex, 30th Bárány Society meeting, Uppsala, Sweden.
Neuroimage 119, 338–351.