CereVA

Visual Analysis of Functional Brain Connectivity

Michael de Ridder

, Karsten Klein

and Jinman Kim

School of IT, The University of Sydney, Sydney, Australia

School of IT, Monash University, Melbourne, Australia

Keywords:

Visual Analysis, Resting State Networks, Graph Layout, Interactive Visualization.

Abstract:

We present CereVA, a web-based interface for the visual analysis of brain activity data. CereVA combines 2D

and 3D visualizations and allows the user to interactively explore and compare brain activity data sets. The

web-based interface combines several linked graphical representations of the network data, allowing for tight

integration of different visualizations. The data is presented in the anatomical context within a 3D volume

rendering, by node-link visualizations of connectivity networks, and by a matrix view of the data. In addition,

our approach provides graph-theoretical analysis of the connectivity networks. Our solution supports several

analysis tasks, including the comparison of connectivity networks, the analysis of correlation patterns, and the

aggregation of networks, e.g. over a population.

1 INTRODUCTION

Unraveling the mysteries of the brain’s inner work-

ings is one of the great challenges in biomedical re-

search today. Recent efforts including the Human

Connectome Project, the Human Brain Project (Van

Essen et al., 2013; HBP, 2014) and the BRAIN ini-

tiative underline the importance of this task. A core

component of these initiatives is to study the patterns

that are evident in the brain activity correlation data,

known as functional connectivity networks. Differ-

ences in the networks between individuals over time,

and between individual networks and aggregated net-

works among the healthy population or groups of pa-

tients can be examined to detect and understand the

alterations related to disorders, e.g. in the context of

Alzheimer’s disease (Greicius et al., 2004). The anal-

ysis of such functional connectivity networks can fos-

ter the detection and understanding of brain activity

patterns (Smith et al., 2013).

Brain activity is commonly measured using neu-

roimaging scanners of functional magnetic resonance

imaging (fMRI). fMRI data is used to ﬁrst derive

time-course data, from which correlation matrices are

calculated, e.g. (van den Heuvel and Pol, 2010; Smith

et al., 2013). These matrices contain activity correla-

tion values between regions of the brain that are either

deﬁned by a brain parcellation, based on the anatomy,

or by the resolution of the imaging scan, e.g. the vox-

els of an fMRI session (Destrieux et al., 2010). Func-

tional networks are not directly given by this raw data,

but need to be modelled from the correlation matri-

ces. However, a large part of the correlation data is

regarded as noise as they are either redundant or does

not contain meaningful correlation. It is not obvi-

ous how to select the relevant entries from the matri-

ces where each individuals have unique correlations;

a constant ‘threshold value’ might not be sufﬁcient

to cover all biologically signiﬁcant correlations (Lee

et al., 2012). The need for interactive selection of the

threshold raises opportunities for the application of

visual analysis concepts, where modelling a weighted

graph that captures the main connectivity patterns in

the functional networks then permits graph theoreti-

cal and statistics methods to be employed for analysis

(Onoda and Yamaguchi, 2013; Wang et al., 2010).

While resting state neural networks, which oc-

cur when no conscious activity is being performed,

exhibit several speciﬁc properties that can be calcu-

lated based on the network topology, e.g. (Bullmore

and Sporns, 2009; Wang et al., 2010; Chang et al.,

2013), visualization of such networks and the under-

lying data is an important step in facilitating analysis.

Visualization allows the researcher to create insight

by exploiting the humans expertise and ability to spot

patterns, and can also help to cope with the impreci-

sion of the network model and the inﬂuence of noise

and uncertainty in the data. Research into the anal-

ysis of neural activity data often includes visualiza-

tion of anatomical and functional information, such as

brain and brain region depiction, e.g. (Nowke et al.,

2013), or neural network visualization, e.g. (Sorger

131

de Ridder M., Klein K. and Kim J..

CereVA - Visual Analysis of Functional Brain Connectivity.

DOI: 10.5220/0005305901310138

In Proceedings of the 6th International Conference on Information Visualization Theory and Applications (IVAPP-2015), pages 131-138

ISBN: 978-989-758-088-8

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

et al., 2013). An investigation into these visualiza-

tions by Margulies et al. (Margulies et al., 2013) ar-

gues that the best views intuitively combine network

and anatomical data without cluttering the anatomi-

cal context, which could potentially present mislead-

ing information. This holds, in particular, when sup-

porting a guided interactive exploration of activity

data. The visualization can be accompanied by a net-

work analysis that helps to identify and exploit the

structural characteristics of the network (Wang et al.,

2010). The overarching goals of these works are

comparison across the population and between indi-

viduals to derive global patterns and variations, e.g.

caused by disease, and to analyze the spatiotemporal

dynamics in order to better understand the structure-

function relationship underlying the connectivity pat-

terns. Faithful network representations are thus re-

quired that convey patterns, as well as strong, clear

links to the anatomical context.

We present a Cerebral Visual Analytics tool,

CereVA, a preliminary approach that facilitates the

analysis of brain activity data by providing a web-

based visual interface for interactive exploratory anal-

ysis of the data. The interface combines several

linked graphical representations of the network data,

allowing for tight integration of different visualiza-

tion views. The data is presented in the anatomi-

cal context within a 3D volume rendering, by node-

link visualizations of connectivity networks, and by

a matrix view of the data. In addition, our approach

provides graph-theoretical analysis of the connectiv-

ity networks. Our solution supports several analysis

tasks, including the comparison of connectivity net-

works, the analysis of correlation patterns, and the

aggregation of networks, e.g. over an aggregated pop-

ulation.

2 APPLICATION AND

VISUALIZATION TASKS AND

CHALLENGES

Important analysis tasks for resting state activity data

include:

A) identiﬁcation of relevant differences between in-

dividual or aggregated networks

B) identiﬁcation of signiﬁcant patterns in a set, e.g.

across the whole population

C) identiﬁcation of network classes, e.g. for healthy

individuals, disorders and behavioural patterns

D) investigating changes over time, e.g. during de-

velopment (BSA, 2014) or due to aging

These tasks lead to challenges including:

i) modelling weighted graphs from the correlation

matrix that reﬂect the main correlation character-

istics

ii) modelling aggregated or aligned networks that

represent a whole set of networks, e.g. the ag-

gregate across the population, covering the overall

characteristics well without a large deviation

iii) supporting tasks A and B by intuitive interactive

visualizations for comparison to show dissimilar-

ities and similarities

iv) supporting task C by including automated analy-

sis in the visual interface, e.g. classiﬁcation and

clustering

v) supporting task D by allowing the analysis of time

series data.

Therefore an intuitive visualization that combines

both functional and anatomical contexts is suitable to

facilitate an exploratory analysis of functional con-

nectivity networks. Several methods of combining

network visualization with imaging data are have

been investigated in recent research (B

ottger et al.,

2014; Sorger et al., 2013; Hagmann et al., 2008).

In addition, there is evidence that matrix-based visu-

alizations can be advantageous for several compari-

son tasks (Alper et al., 2013). Matrices are also well

suited for use in small multiples visualizations, which

can be exploited to represent temporal dynamics or

for comparison tasks. Thus incorporating a matrix-

based view into a visualization approach can facili-

tate comparison as long as it is smoothly integrated

into the interaction concept and does not increase the

complexity of the visualization.

3 CereVA

As the underlying mechanisms of brain function and

the impact of diseases are not yet fully understood, an

interactive visual analysis process is proposed to facil-

itate knowledge discovery. This approach combines

the strengths of the human expert to discover pat-

terns and to exploit existing knowledge with the com-

putational power of automated analysis and guided

navigation. CereVA provides the researcher with an

interactive visual interface that supports such a pro-

cess, combining visualizations of the activity data, the

anatomy and results from analysis. We aim to provide

a system of tools that addresses the tasks described

in Section 2, where we focus on tasks A, B, and C.

CereVA’s design was developed according to require-

ment analysis that included consultations with, and

IVAPP2015-InternationalConferenceonInformationVisualizationTheoryandApplications

132

feedback from, potential users, including an expert.

We provide the implementation of our concept in a

web-based application to allow users anywhere, any-

time access.

While questions asked in the investigation of

disease-related anomalies cannot be answered purely

by automated analysis, the results of the analysis can

still be used to guide the interactive exploration of

the data. The visualizations in our system are there-

fore enriched with the functional activity data and in-

formation derived from it. For the exploratory anal-

ysis, we facilitate interactive navigation through the

given data sets, in particular selection and compari-

son of networks of interest. Our interaction concept

is thus designed to support the following workﬂow: a

user can either visualize a single network or compare

networks with each other, wherein the networks are

either individual subject networks or aggregated net-

works that are derived from a set of networks, e.g. the

network created by computing the average connectiv-

ity value for each pair of nodes. Results from data

and network analysis are shown to the user and also

mapped onto the network and anatomical visualiza-

tions. By ﬁltering and selecting, the user can investi-

gate the structure and patterns of the activity data.

The visualization components, the underlying

data, the view enrichment, and the interaction concept

are described in the following sections.

3.1 Data Processing and View

Enrichment

Providing the results of an automated data analysis

and mapping these results in an intuitive way onto the

visualizations is important for supporting an efﬁcient

interactive exploration of the activity data.

Network Deﬁnitions. To cope with challenges i)

and ii), our concept includes both automated network

creation, and networks based on thresholds interac-

tively deﬁned by the user. Three different types of

networks are used in our approach, data, aggregation,

and comparison networks. All networks contain the

same nodes, brain regions speciﬁed in the input data.

A data network is simply a network modeled using

a single correlation matrix to create the edges. An

aggregation network provides a population view —

we create networks that represent the median and the

average pairwise correlation values. A comparison

network is modeled based on differences and similar-

ities between two individuals, including aggregation

matrices.

Threshold Guidance. In general a constant thresh-

old value will not be sufﬁcient for the decision of

which correlations are signiﬁcant, as this might vary

across different experiments, individuals, or even time

points for one individual. Useful generic models that

best support a speciﬁc task or question at hand are

however not known, and thus our system guides the

user in the decision for a correlation value interval

of interest. In order to support the user in selecting

a threshold, the distribution of correlation values is

shown as a heat map. The user can select a correla-

tion value interval of interest using a slider element

on which the above information is displayed. Stan-

dard characteristics such as connectivity, number of

edges and density are calculated for a number of user-

selectable threshold values and can be used guide se-

lection of which correlation threshold might be suit-

able to model the network.

Network Comparison. For correlation values

within the interval an edge is then modelled in the

network, see Figure 1 top right. In a basic implemen-

tation for the comparison of two individual networks,

we create a difference network with an edge mod-

elled if the difference of correlations between two

nodes is between user-deﬁned threshold values. By

using this simple model, the dissimilarities between

networks are emphasized. As an alternative, when

using the circular layout, we include information

on both similar and dissimilar correlation values. A

high value in either network with a small correlation

difference to the other network leads to the creation

of an edge in the comparison network, whereby the

visual representation is altered to show the different

characteristics, as shown in Fig. 3. With this model,

both common patterns and outlier correlations in both

networks are shown, however, the resulting network

is also more dense and thus can be more difﬁcult to

interpret.

Graph Analysis. Analysis methods can be classi-

ﬁed into three major categories, the analysis might be

based on the raw data, taking into account all correla-

tion values, might use the weighted graph modeled

from the data, for example after some user-deﬁned

correlation value threshold is taken into account, or

at the most abstract level, only the graph’s topology

might be used. In particular for the graph-based vari-

ants a large number of measures exist for analyzing

the characteristics. These can be global values, for

example the characteristic path length, or local val-

ues, such as the node strength. As we have a labeled

network with annotations for the different nodes and

their brain regions, we assume that purely topological

CereVA-VisualAnalysisofFunctionalBrainConnectivity

133

Figure 1: Overview of the visualization interface: Two representations are shown simultaneously, a 3D volume rendering and

a 2D visualization of the activity correlation, here a network visualization derived from a single correlation matrix. Sliders

can be used to deﬁne the range of correlation values that is modeled into the network. The subnetwork consisting of a selected

node (green) and its neighbors (red) is highlighted in the same color as the corresponding region in the 3D rendering, and the

axes in 3D view are positioned on the selected region.

measures, in particular when they express some value

that is aggregated over the whole graph, will not al-

ways adequately reﬂect the brain network character-

istics. However, even though networks with similar

topological properties may not be similar in terms of

neural activity, we still calculate these values as an

indicator to separate largely different networks.

Network properties that are calculated and con-

sidered useful for the analysis include the clustering

coefﬁcient, the characteristic path length, the node

degree (weighted, unweighted, and distribution), the

number of isolated nodes, and centrality measures.

In addition, we take into account the anatomical

and functional annotations to calculate several values

for comparison of networks. We calculate an out-

lier ﬁngerprint, based on the partition of the brain

into nine brain lobes derived from (Destrieux et al.,

2010). For each network we emphasize the amount to

which correlations per lobe differ signiﬁcantly from

the whole network set. To this end, we ﬁrst identify

the connections with a correlation that differs more

than the standard deviation from the median. For each

node, we sum up the number of such connections, and

identify the nodes for which that value again differs

more than the standard deviation from the respective

median number. As the nodes can be mapped to brain

lobes, we can aggregate the node values for each lobe

and thus create a ﬁngerprint of length 9 that represents

the distribution of outlier nodes. The level of opac-

ity of a block in the ﬁngerprint indicates the outlier

level, i.e. the stronger the color, the stronger the out-

lier characteristics regarding our measure. This ﬁn-

gerprint then can be used to identify similar networks,

see Figure 4 right.

3.2 Visualization Components and

Interaction

Visualization Components. The main visualiza-

tion components for our approach are the 3D brain

rendering view and an abstract correlation data view,

which can be either a graph-based network view or

a matrix view, see Figures 1 and 2. Adding a net-

work visualization directly into the volume rendering

can make it difﬁcult for the user to clearly perceive

the network structures. We conjecture that a side-by-

side visualization of network or matrix view and brain

anatomy will give the user a good overview and at

the same time allows for an intuitive interactive ex-

ploration of activity data. The assumption is based

on the idea that the 3D brain rendering allows intu-

itive orientation regarding the anatomical position and

functional annotation of the network nodes, while the

network can clearly depict the connection topology

and will thus help to detect patterns better than with

the 3D view alone. Hence our concept is based on

such a combination of a 3D brain rendering and a 2D

graph or matrix visualization, and we link the views

to enable the user to switch between views during in-

teractive exploration without a large cognitive effort.

In addition, our application features several inter-

active elements. These include pop-out panels for

the selection of networks and view details. Three

small subviews for the 3D rendering allow slice-based

browsing through the images, see Figure 1, bottom

left.

As part of our concept the precalculated network

properties for the currently selected network, or the

differences of these properties in comparison mode,

can be shown in a pop-out dashboard at the bottom of

IVAPP2015-InternationalConferenceonInformationVisualizationTheoryandApplications

134

Figure 2: Matrix view of the correlation data. The user can select correlations which then are highlighted both in the matrix

and the volume visualization. Red and blue color represent stronger correlation in one of the matrices under comparison.

Network statistics for a selected correlation threshold values are shown.

the screen. These values help to characterize a single

network, and can also be used for network compari-

son and classiﬁcation.

We show a graphical representation of the outlier

ﬁngerprint next to each network’s name in the net-

work list to enable the user to efﬁciently select most

relevant networks for analysis and comparison, see

Figure 4 right. Each brain lobe that contains outlier

nodes is depicted by a character that is colored ac-

cording to the number of outlier nodes. Blue color

represents nodes with a signiﬁcant number of out-

lier nodes higher than the median, yellow represents

smaller numbers, and the intensity is varied depend-

ing on the number.

Interaction and Navigation. In the exploratory

workﬂow, a user can freely select networks for vi-

sualization and comparison. A correlation value in-

terval can be chosen by using the slider element lo-

cated on top of the network view, allowing the user

to deﬁne the range of the values used to model the

current network. Whenever the user adjusts the slider

setting, the corresponding network is displayed in the

network view. To support the user in choosing a use-

ful threshold, the distribution of correlation values is

mapped onto the slider by means of a color shading.

In comparison mode, two shadings show the different

distributions in the networks on the slider, see Fig. 3.

All views are coordinated and linked by a global

selection mechanism. In the 2D and 3D views, users

can select regions, either represented as a node in the

2D view, or as a rendered region in the 3D view. Se-

lecting elements in the visualization facilitates focus

on substructures of the network, e.g. the neighbor-

hood of a node in the network, or the subnetwork

corresponding to a brain region. Selection triggers a

highlighting by a change in colors for the selected re-

gion and its neighbors in the current correlation net-

work in the 2D and 3D views.

The visual cues to highlight elements during the

interactive exploration process are kept consistent

over all views. The selected node together with its lo-

cal neighborhood are emphasized in the network view

by a change in the color, see Figure 1, and the corre-

sponding brain regions are colored in the same way.

All colors are user-adjustable.

The volume rendering allows the user to freely

navigate within the 3D visualization that shows the

brain’s anatomy, rotating and zooming to the region

of interest (ROI). The volume axes in the 3D view are

automatically positioned to indicate the selected fo-

cus region, thereby also updating the slice views to

show the current selection. By interacting with the

slice subviews, the user can move the volume axes

and scroll through all individual slices. An ROI at-

las can be mapped onto the 3D rendering that allows

links back from selected brain regions to the network

visualization.

By changing the setting in the pop-out panels,

the user can select different networks for visualiza-

tion and comparison, and also toggle the visibility

of the volume rendering of white matter and pial

for each hemisphere depending on the user’s require-

ments. Both the network and the matrix view sup-

port a comparison mode that allows visualization dif-

ferences between selected networks. A color coding

scheme is used to show differences and similarities,

see Fig. 3 for the network view. For the compari-

son network described in Section 3.1, the edges with

a difference value within the threshold are drawn in

red and blue, where the color is determined by which

network has the higher correlation value. Strong cor-

relations occurring in both networks are drawn in yel-

CereVA-VisualAnalysisofFunctionalBrainConnectivity

135

Figure 3: Network comparison combining similarities and dissimilarities for two networks. Differences above the user-deﬁned

threshold value are shown in blue and red, common patterns in yellow.

Figure 4: Using the outlier ﬁngerprint to analyze correla-

tion data. The ﬁngerprint shows signiﬁcant differences and

similarity both of a network to aggregated networks and be-

tween several pairs of individual networks, e.g. the net-

works labeled 3E6J9LC6 and SVP271T6 (selected for vi-

sual comparison) are quite similar regarding the outliers,

whereas SVP271T6 and Q7C341EB differ largely.

low. This helps the user to distinguish between net-

works that are similar besides a few differences and

networks that vary a large amount. The opacity of

the graphical edge representation depends on the dif-

ference values, such that more prominent patterns are

emphasized. This way, differences and their ampli-

tude are highlighted in the 2D view, allowing reﬁned

ﬁltering or selection of ROIs.

Network Layout. CereVA uses a circular layout for

graph visualization as the default layout, as it is estab-

lished in biomedical network visualization, including

recent iterations of brain network visualization tools

(Margulies et al., 2013). The view allows relatively

compact visualizations with a static aspect ratio, facil-

itating the side-by-side presentation. In order to min-

imize the effort to switch between 2D and 3D view,

we use a ﬁxed node order that reﬂects the anatomical

location — clockwise from the top it shows left hemi-

sphere, right hemisphere, then central regions. The

membership and vicinity of network nodes with re-

spect to the brain regions are shown in the 3D view.

For an improved visualization of the connection pat-

terns, we apply edge bundling, where the tension dy-

namically adjusts for the number of visible edges, to

visually aggregate edges with a similar route. While

the single edge representations still exist in the visu-

alization, the bundling allows better identiﬁcation of

global connectivity patterns by joining similar edge

sections to bundles, while also reducing visual clut-

ter.

We assume that for an in-depth visual analy-

sis of more complex patterns, different layout meth-

ods might prove more valuable, and are therefore

investigating the use of other methods for ﬁltered

subnetworks. As such layouts might be subject to

constraints, constraint-based methods of the Web-

CoLa (Webcola, 2014) library might be best suited

to provide ﬂexible layouts including data-driven con-

straints, e.g. to emphasize a spatial clustering of the

brain regions in the network layout.

3.3 Implementation

Our implementation allows the user to visualize rest-

ing state networks in the context of the brain’s

anatomy, and to compare networks using a node-link

representation. In addition, the raw correlation data

can be visualized and compared in a matrix view. The

comparison can be done visually in both the node-link

and the matrix view, while data- and network-based

indicators highlight network similarities and differ-

ences, facilitating analysis tasks A and B.

We created CereVA as a web-based solution, as

IVAPP2015-InternationalConferenceonInformationVisualizationTheoryandApplications

136

it can be easily made available and updated, without

the need to download and install an application. As a

web-based solution, we expect it to have advantages

when collaborative work on a data set is required, and

when interactive analysis of larger data sets can be

supported by cloud-based computing.

Our implementation uses several libraries for the

different parts of our interface. For the rendering of

3D visualizations in the browser we employed We-

bGL (WebGL, 2013). WebGL enables the browser di-

rect access the Graphics Processung Unit (GPU), re-

sulting in more seamless interaction and performance

compared to non-GPU implementations. The XTK li-

brary (XTK, 2014), a wrapper for WebGL, is used as

it is speciﬁcally designed to support rendering of 3D

medical image data. For the graph and matrix visu-

alization and layout we used the implementation pro-

vided by D3(D3, 2014), a well-established library for

web-based graph drawing.

4 DISCUSSION

To derive meaningful information from network visu-

alizations, the networks need to be modeled in an in-

tuitive way that facilitates knowledge discovery. Our

preliminary implementation allows the user to select

an interval for the correlation values that is used to

ﬁlter a subnetwork of the data set. While we think

that it is useful to allow the user such a ﬁltering in

an interactive fashion to focus on ROIs, the search for

a reasonable threshold should be guided by an auto-

mated analysis. We use a constant threshold to start

with, and give the user visual hints based on network

characteristics to support their search.

Network comparison is an active area of research,

and there is no obvious optimal visualization of brain

activity networks. While our solution presents the dif-

ference and similarities in a single cumulative graph-

based view, separate views, for example individ-

ual graphs representing speciﬁc characteristics (small

multiple graphs), might be useful to quickly detect

patterns and to identify outliers. We aim to inves-

tigate further whether calculations such as weighted

maximum common subgraphs help to better identify

common patterns. The reason for this suggestion is

that we could see that for several pairs of threshold-

ﬁltered networks, the dominant differences consisted

only of one or two additional edges. Our network

analysis so far does not make use of the fact that we

have labeled graphs, i.e. we have some speciﬁc brain

location associated with each node of the network. As

the combination of this information may prove useful,

we will include it in the analysis in future steps.

Aggregation, including of subnetworks, may re-

duce the complexity and emphasize patterns in the

poplation. However we have not seen enough evi-

dence so far of which network characteristics could be

subject to aggregation. We aim to answer this ques-

tion in ongoing discussions with collaborating neu-

roscientists. This might also help to further support

analysis task C, in particular when clustering and clas-

siﬁcation methods will be included, as this task is so

far mainly supported by giving the user information

on network and correlation characteristics. Finally

larger neighborhoods or subnetwork patterns than the

distance one neighborhood might be of interest for

highlighting or semi-automated selection.

5 OUTLOOK

In this paper, we presented CereVA, a preliminary

concept for the visual analysis of brain activity corre-

lation data. CereVA is implemented as a web service

that provides an interactive visual interface for data

exploration. Our approach currently supports a ba-

sic correlation matrix format as input, and we plan to

provide import and export of data and visualizations

in standard formats, e.g as provided by the large brain

research projects. Further, we aim to integrate addi-

tional knowledge that is available for example in such

public databases or publications.

In order to extend our solution towards a visual

analysis approach, we will embed additional infor-

mation for analysis and visualization. This will in-

clude an analysis of time series data, as we believe

that an analysis will provide further insight into the

variation of correlations for a single subject, and thus

allow experts to derive better characterizations of sim-

ilarities and differences across the population. Visu-

alizations and interactions based on a decomposition

of the network, e.g. using modularity clustering or

functional aggregation, will be considered for addi-

tion to the system. While the linked complementary

2D and 3D views demonstrated its capabilities to pro-

vide a 3D anatomical reference for the visualization

of activity networks, adding information from struc-

tural MRI and diffusion tensor imaging might greatly

improve the analysis of individual brain activity, e.g.

after brain damage.

To assess the usefulness of our system beyond the

feedback from domain experts during the develop-

ment, and to make decisions on the further implemen-

tation and features, we will perform a formal evalu-

tion with domain experts. This paper presented our

ﬁndings from the use of resting-state MRI networks,

however our system is not limited to this data type and

CereVA-VisualAnalysisofFunctionalBrainConnectivity

137

we will explore extension of our system for use with

different types of networks.

REFERENCES

Alper, B., Bach, B., Riche, N. H., Isenberg, T., and Fekete,

J.-D. (2013). Weighted graph comparison techniques

for brain connectivity analysis. In Mackay, W. E.,

Brewster, S. A., and Bødker, S., editors, CHI, pages

483–492. ACM.

ottger, J., Schurade, R., Jakobsen, E., Schaefer, A., and

Margulies, D. S. (2014). Connexel visualization: a

software implementation of glyphs and edge-bundling

for dense connectivity data using braingl. Front Neu-

rosci, 8(15).

BSA (2014). Brainspan Atlas. See

http://www.brainspan.org/.

Bullmore, E. and Sporns, O. (2009). Complex brain net-

works: graph theoretical analysis of structural and

functional systems. Nat. Rev. Neurosci., 10(3):186–

198.

Chang, H.-M., Chiang, A.-S., Didimo, W., Lin, C.-Y., Li-

otta, G., and Montecchiani, F. (2013). On the robust-

ness of the drosophila neural network. In Network Sci-

ence Workshop (NSW), 2013 IEEE 2nd, pages 168–

171.

D3 (2014). Data Driven Documents. See http://d3js.org/.

Destrieux, C., Fischl, B., Dale, A., and Halgren, E. (2010).

Automatic parcellation of human cortical gyri and

sulci using standard anatomical nomenclature. Neu-

roImage, 53(1):1 – 15.

Greicius, M. D., Srivastava, G., Reiss, A. L., and Menon, V.

(2004). Default-mode network activity distinguishes

alzheimer’s disease from healthy aging: Evidence

from functional mri. Proceedings of the National

Academy of Sciences of the United States of America,

101(13):4637–4642.

Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R.,

Honey, C. J., Wedeen, V. J., and Sporns, O. (2008).

Mapping the structural core of human cerebral cortex.

PLoS Biol, 6(7).

HBP (2014). The Human Brain Project. See

https://www.humanbrainproject.eu/.

Lee, H., Kang, H., Chung, M. K., Kim, B. N., and Lee, D. S.

(2012). Persistent brain network homology from the

perspective of dendrogram. IEEE Trans Med Imaging,

31(12):2267–2277.

Margulies, D. S., B

ottger, J., Watanabe, A., and Gor-

golewski, K. J. (2013). Visualizing the human con-

nectome. NeuroImage, 80:445–461.

Nowke, C., Schmidt, M., van Albada, S., Eppler, J., Bakker,

R., Diesrnann, M., Hentschel, B., and Kuhlen, T.

(2013). Visnest – interactive analysis of neural ac-

tivity data. In Biological Data Visualization (BioVis),

2013 IEEE Symposium on, pages 65–72.

Onoda, K. and Yamaguchi, S. (2013). Small-worldness and

modularity of the resting-state functional brain net-

work decrease with aging. Neurosci. Lett., 556:104–

108.

Smith, S., Vidaurre, D., Beckmann, C., Glasser, M., Jenkin-

son, M., Miller, K., Nichols, T., Robinson, E., Salimi-

Khorshidi, G., Woolrich, M., Barch, D., Ugurbil, K.,

and DC, V. (2013). Functional connectomics from

resting-state fmri. Trends in Cognitive Sciences, 17.

Sorger, J., Buhler, K., Schulze, F., Liu, T., and Dickson, B.

(2013). neuromap – interactive graph-visualization of

the fruit ﬂy’s neural circuit. In Biological Data Visu-

alization (BioVis), 2013 IEEE Symposium on, pages

73–80.

van den Heuvel, M. P. and Pol, H. E. H. (2010). Explor-

ing the brain network: A review on resting-state fmri

functional connectivity. European Neuropsychophar-

macology, 20(8):519 – 534.

Van Essen, D. C., Smith, S. M., Barch, D. M., Behrens,

T. E., Yacoub, E., and Ugurbil, K. (2013). The wu-

minn human connectome project: An overview. Neu-

roImage, 80(0):62 – 79. Mapping the Connectome.

Wang, J., Zuo, X., and He, Y. (2010). Graph-based network

analysis of resting-state functional mri. Frontiers in

Systems Neuroscience, 4(16).

Webcola (2014). WebCoLa. See

http://marvl.infotech.monash.edu/webcola/.

WebGL (2013). WebGL. See

https://www.khronos.org/webgl/.

XTK (2014). The X Toolkit. See

https://github.com/xtk/X#publications.

IVAPP2015-InternationalConferenceonInformationVisualizationTheoryandApplications

138