IEEE Signal Processing - May 2018 - 90

Table 1. Some MATLAB-based toolboxes for preprocessing, source estimation,
functional/effective connectivity measures, and network analysis/visualization.
Software

Web Page

EEG
Preprocessing

EEG Inverse
Solution

Functional Connectivity/
Effective Connectivity

Brainstorm

http://neuroimage.usc.edu/brainstorm/



EEGLAB

https://sccn.ucsd.edu/eeglab/



FieldTrip

http://www.fieldtriptoolbox.org/



eConnectome http://econnectome.umn.edu/

Network
Measures

Network
Visualization



EEGNET

https://sites.google.com/site/eegnetworks/

Conn

https://www.nitrc.org/projects/conn/

FCT

https://sites.google.com/site/
functionalconnectivitytoolbox/



BCT

https://sites.google.com/site/bctnet/



BNV

https://www.nitrc.org/projects/bnv



GraphVar

https://www.nitrc.org/projects/graphvar/



NBS

https://www.nitrc.org/projects/nbs/



FCT: functional connectivity toolbox; NBS: network-based statistic.

BNSs. Details about this segmentation algorithm developed in
the context of the picture-naming task can be found in [3].
We were then interested in the difference in terms of network topology between both categories. We computed the clustering coefficient reflecting the local information processing,
as defined previously, for each brain region (R = 68, Desikan-
Killiany atlas), and we retained regions leading to significant differences between both conditions (t-test, p < 0.01,
corrected for multiple comparisons using the false discovery
rate method). Figure 4 shows that the clustering coefficient of
the network is not the same for both categories over time, suggesting that the process of information segregation in the brain
is different in the two conditions (meaningful versus meaningless images). Interestingly, the method also showed the implication of the temporal lobe for all of the states, which is widely
reported to be related to semantic processing in the brain.

EEG source connectivity in brain disorders
Converging evidence suggests that perturbations in the brain are
rarely limited to a single region. As the brain is a complex network of structurally and functionally interconnected regions,
local dysfunctions often propagate and affect other regions,
resulting in large-scale network alterations [2]. These dysfunctions likely occur at both the axonal and synaptic levels. A typical example is the rapid spread of partial (focal) epileptic activity
at the onset of seizures, which rapidly involves spatially distributed brain regions [36]. Along the same line but on a longer
timescale, it is now established that the progressive evolution in
Alzheimer's disease and Parkinson's disease is also related to
pathological changes in large-scale networks, although these
neurodegenerative diseases have a focal onset [37].
Therefore, from a clinical perspective, the demand is high
for noninvasive and easy-to-use methods to identify pathological networks such as those involved in epilepsy. In addition,
the demand is also great for novel neuromarkers that are able
90

to characterize network alterations and associated cognitive
deficits in Parkinson's and Alzheimer's patients, in particular,
at the early stages. In this context, EEG exhibits some major
strengths, since it is a noninvasive, easy-to-use, and clinically
available technique. Therefore, and as shown by our recent
studies [11], [38], EEG source connectivity methods could provide some responses to clinical demand, provided that appropriate information processing is performed. Next, we describe
the main applications of EEG source connectivity in neurological disorders.

Epilepsy
Ding et al. [40] were the first to apply functional connectivity
to EEG source signals in epileptic patients. The authors
showed that the method was able to distinguish the primary
sources responsible for the seizure generation from the secondary sources involved in the seizure propagation. In a follow-up
study, Lu et al. [41] applied the method to EEG recordings
(76 channels) performed in patients with partial epilepsy.
Those authors found that the EEG source connectivity method
led to correct seizure onset localization as compared to invasive
recordings. They also reported the need for a large number of
electrodes to better estimate the epileptic network.
Vecchio et al. [42] used EEG source connectivity combined
with graph theory-based analysis in patients suffering from
frontotemporal epilepsy. The authors reported a significant
increase of local connectivity (characterized by the clustering coefficient) and global connectivity (computed using the
characteristic path length) in the alpha band in the ipsilateral hemisphere as compared to the contralateral hemisphere.
With 16 patients, Coito et al. [43] investigated the directionality of the interactions between brain regions during interictal
spikes estimated with effective connectivity methods applied
at the source level. In addition to achieving good matching
with invasive recordings, [43] showed a relationship between

IEEE Signal Processing Magazine

May 2018

http://neuroimage.usc.edu/brainstorm/ https://sccn.ucsd.edu/eeglab/ http://www.fieldtriptoolbox.org/ http://econnectome.umn.edu/ https://sites.google.com/site/eegnetworks/ https://www.nitrc.org/projects/conn/ https://sites.google.com/site/functionalconnectivitytoolbox/ https://sites.google.com/site/bctnet/ https://www.nitrc.org/projects/bnv https://www.nitrc.org/projects/graphvar/ https://www.nitrc.org/projects/nbs/

Table of Contents for the Digital Edition of IEEE Signal Processing - May 2018