Simultaneously Probing Functional and Structural Brain
Connectivity in Real-time
Fibernavigator: An Interactive Tool for Brain Visualization
Maxime Chamberland
1,4
, Maxime Descoteaux
2,4
, Kevin Whittingstall
1,4
and David Fortin
3,4
1
Department of Nuclear Medicine and Radiobioloy, Faculty of Medicine and Health Science,
Université de Sherbrooke, Sherbrooke, Canada
2
Computer Science Departement, Faculty of Science, Université de Sherbrooke, Sherbrooke, Canada
3
Division of Neurosurgery and Neuro-Oncology, Faculty of Medicine and Health Science,
Université de Sherbrooke, Sherbrooke, Canada
4
Centre d’Imagerie Moléculaire de Sherbrooke (CIMS), Centre de Recherche Clinique CHUS, Sherbrooke, Canada
1 OBJECTIVES
The human brain can be viewed as a collection of
parallel networks. Those highly specialized
networks can be conceptualized to as a set of nodes
(gray matter functional areas) linked together by
edges (for example white matter axonal structure).
Functional MRI (fMRI) can provide 4D whole-brain
images that indicates changes in cortical blood flow,
volume and oxygen ratio as well (Blood-
Oxygenation-Level-Dependant or BOLD signal)
caused by cerebral activity across time (Bandettini et
al. 1993; Kwong et al. 1992; Turner 1992). The
spontaneous low fluctuations (< 0.1 Hz) present in
the BOLD signal allow the detection of temporally
correlated spatial patterns, also known as Resting
State Networks (RSNs) when the brain is at rest
(Biswal et al. 1995; Damoiseaux et al. 2006). A
common method of obtaining those networks is to
extract the BOLD time course from an a priori
region of interest (ROI) and perform the temporal
correlation with all other voxels of the brain. The
result is a correlation map or a functional
connectivity map based on the location of the seed
ROI.
Some have proposed a tool for voxel-wise brain
connectivity visualization but it often requires the
pre-calculation of a correlation matrix to be held in
memory (Dixhoorn 2012). Great effort was also
made towards GPU implementation of functional
connectivity exploration (Eklund et al. 2011)
However, the proposed software restrict the user
from placing their reference ROI directly into the 3D
space which greatly reduces the level of
interactivity. Another tool was proposed for
neurosurgical application which quickly allows the
user to interrogate data for pre-surgical planning
(Böttger et al. 2011). Here, the user is forced to
move the ROI solely on 2D anatomical slices, thus
only revealing activations present on those selected
slices.
In this work, we propose an interactive tool for
the exploration of functional connectivity in a fully
3D fashion, which can be coupled with our existing
real-time fiber tracking module inside the
Fibernavigator (Chamberland et al. 2014). Using a
healthy volunteer dataset, we qualitatively
demonstrate how both functional and structural
modules can be merged together for efficient brain
mapping exploration.
2 METHODS
Our new real-time functional exploration tool is
implemented on CPU and runs on a single core
computer, which does not require any specific
hardware. It works on any functional data (e.g
resting-state) that is preferably pre-processed. For
anatomical reference, the user has to first load a
subject-specific T1-weighted image. By placing a
ROI in the 3D environment, one can instantaneously
activate the functional correlation module while
dragging the ROI everywhere in the brain. The mean
signal is first extracted from the voxels encompassed
by the ROI, and then statistically compared to the
rest of the brain. The correlation coefficient (CC)
between voxels x and y is denoted as:
CC = cov(x,y)/σ
x
σ
y
(1)
where cov(x, y) is the covariance of any two signals
and σ
x
σ
y
are the standard deviation of the time
series. CCs are then converted to z-scores and
Chamberland M., Descoteaux M., Whittingstall K. and Fortin D..
Simultaneously Probing Functional and Structural Brain Connectivity in Real-time - Fibernavigator: An Interactive Tool for Brain Visualization.
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
rendered at each voxel as small particles like in the
following OpenGL example:
glEnable(GL_BLEND) //Activate blending
glBlendFunc(...) //Blending function
glEnable(GL_POINT_SPRITE) //Render ON
glPointSize(size * z_Score) //Pt size
glColor4f(r, g, b, alpha * z_Score)
glBegin(GL_POINTS)
glVertex3f(x, y, z) //3D points
glEnd();
glDisable(GL_POINT_SPRITE) //Render OFF
glDisable(GL_BLEND) //Deactivate blend
To reduce cluttering, the opacity (alpha) of each
particle is weighted by its z-score value. A transfer
function is also responsible for the mapping between
z-scores and color display (using a hot colormap).
Interactive z-threshold and cluster-level sliders are
also available for visualization purpose. Finally, the
user can save and export the generated activation
map into a 3D nifti file. As one can see, the
computation step is performed in the native space
(i.e fMRI space) while the rendering stage is done at
the anatomical level (T1-space) using the
transformation matrix coming from fMRI-T1
registration of the datasets.
2.1 Datasets
Continuous functional recording was carried out on
a Siemens 1.5 Tesla (T) imaging system using a
standard echo-planar imaging (EPI) sequence. 35
axial slices were obtained with a 64 x 64 matrix,
TR/TE 2730/40 msec, for a voxel size of 3.4 x 3.4 x
4.2 mm. Diffusion data was acquired using a single-
shot echo-planar (EPI) spin echo sequence (TR/TE =
11700/98 ms, GRAPPA factor 2), with b-value of
1000 s/mm² and 64 uniform directions (128 x 128, 2
mm isotropic spatial resolution, upsampled to 1
mm³).
3 RESULTS
Figure 1 illustrates the functional connectivity tool
on its own. It shows the well-known “Default Mode
Network” (Raichle et al. 2001) obtained by placing a
4 mm³ ROI at the junction of the posterior cingulate
cortex and precuneus (z-score > 2). As one can see,
the expected 4 nodes of the network are well
represented, namely the medial prefrontal cortex and
the bilateral inferior parietal lobes.
Figure 2 demonstrates how the functional
connectivity tool can be coupled with our existing
real-time fiber tracking module. A 20 x 10 x 10 mm
ROI was interactively placed in the left lateral motor
cortex, revealing associated right activations
(namely known as the motor RSN, z-score > 2) with
underlying corticospinal tract and corpus callosum.
Tractography parameters are shown in Table 1.
Figure 1: Default Mode Network. Left: Seed ROI (circled)
interactively placed in the posterior cingulate cortex.
Middle: 3D rendering of co-activated nodes. Right: 2D
view of the activation map generated.
Figure 2: Motor activation and associated fiber pathway
(corticospinal tract / corpus callosum). By lauching seeds
and CC from the ROI (blue), the user can observe a joint
functional/structural network, without the need of
precomputing CC matrices.
Table 1: Tractography parameters.
Step size 1.0 mm
Max. Angle (θ) 35°
Threshold 0.15 (Frac.Anis.)
Min. Length 10 mm
Max. Length 200 mm
# of seeds 500
Experimentation was done on a laptop with the
following specs: System: Linux Mint 32-bit, Video
card: Geforce GT 640M memory 2GB, NVIDIA
Driver: 306.97, CPU: Intel(R)Core(TM) i7-3632QM
@ 2,20GHz, 16GB RAM. Mean frame-per-second
(FPS) ratio remained above 20 when moving the
ROI in the 3D space which indicates no latency.
4 DISCUSSION
We present here a new real-time interactive feature
of the Fibernavigator that allows fast visualization
of functional and structural organization of the brain
in a 3D fashion. It gives convincing results on the fly
and is an important tool to better understand how
connections lies behind functional networks. It can
also serve as a quality assurance tool at the
individual level for close inspection of data prior
launching massive analysis.
Since the 2 modules work independently, it
allows the user to either look at structural,
functional, or as described here, combined brain
connectivity. If one is interested in visualizing brain
connectivity without performing real-time fiber
tracking, it is possible to load a set of precomputed
tracts. In this case, the ROI will serve as a selection
object, only displaying streamlines that pass through
it.
Finally, our real-time technique will shortly be
introduced in a clinical setup and is achievable
without complex GPU programming. One possible
extension would be to not only initiate tractography
from the draggable ROI but also launch seed from
the generated functional clusters to fully visualize
the total extent of the underlying network.
SUPPLEMENTARY MATERIAL
Supplementary video data showing the real-time
functional tool in action can be found online at:
www.youtube.com/watch?v=HmlxktmVSPA.
REFERENCES
Bandettini, P.A. et al., 1993. Processing strategies for
time-course data sets in functional MRI of the human
brain. Magnetic Resonance in Medicine, 30(2),
pp.161–173.
Biswal, B. et al., 1995. Functional connectivity in the
motor cortex of resting human brain using echo-planar
MRI. Magnetic Resonance in Medicine, 34(4),
pp.537–541.
Böttger, J. et al., 2011. A software tool for interactive
exploration of intrinsic functional connectivity opens
new perspectives for brain surgery. Acta
neurochirurgica, 153(8), pp.1561–72.
Chamberland, M. et al., 2014. Real-time multi-peak
tractography for instantaneous connectivity display.
Frontiers in neuroinformatics, 8, p.59.
Chamberland, M. & Descoteaux, M., 2012. Real-time
fiber tractography: interactive parameter tuning for
neurosurgical interventions. In Human brain mapping.
The Organization for Humain Brain Mapping.
Damoiseaux, J.S. et al., 2006. Consistent resting-state
networks across healthy subjects. Proceedings of the
National Academy of Sciences of the United States of
America, 103(37), pp.13848–13853.
Dixhoorn, A.F. van and M.J. and L.B. van and B.C.P.,
2012. BrainCove: A Tool for Voxel-wise fMRI Brain
Connectivity Visualization. In Timo Ropinski and
Anders Ynnerman and Charl Botha and Jos Roerdink,
ed. Eurographics Workshop on Visual Computing for
Biology and Medicine. Eurographics Association, pp.
99–106.
Eklund, A. et al., 2011. A GPU accelerated interactive
interface for exploratory functional connectivity
analysis of FMRI data. In 2011 18th IEEE
International Conference on Image Processing. IEEE,
pp. 1589–1592.
Kwong, K.K. et al., 1992. Dynamic magnetic resonance
imaging of human brain activity during primary
sensory stimulation. Proc Natl Acad Sci U S A, 89(12),
pp.5675–5679.
Raichle, M.E. et al., 2001. A default mode of brain
function. Proceedings of the National Academy of
Sciences of the United States of America, 98(2),
pp.676–82.
Turner, R., 1992. Magnetic resonance imaging of brain
function. Am J Physiol Imaging, 7(3-4), pp.136–145.