IEEE Signal Processing - May 2018 - 92

Using dense EEG from 124 Parkinson's patients, we recently reported
progressive disruption in functional connectivity among three patient groups:
cognitively intact patients, patients with
mild cognitive deficits, and patients
with severe cognitive deficits. Our findings indicate that functional connectivity decreases with the worsening of
cognitive performance, suggesting that
it can potentially be used to devise novel
neuromarkers of cognitive impairment
in Parkinson's patients [11].

Functional Connectivity

wMNE

dSPM

Inverse Solution

PLV

Discussion

EEG involves measuring the brain's
electrical activity by using electrodes
positioned on the scalp. A key feature
0.8
r2
PLV
of EEG is its intrinsically excellent
Reference Network
time resolution that makes it unique for
0.6
tracking the fast reconfiguration of
functional networks of neuronal assem0.4
blies distributed in the cerebral cortex.
Emerging evidence shows that function0.2
al brain connectivity computed at the
scalp level (the electrode space) does
0
not allow for the relevant interpretation
dSPM
wMNE
of anatomically interacting areas, as
(b)
(c)
estimates are severely corrupted by the
volume-conduction effect (see [6] and
FIGURE 6. (a) Brain networks obtained using two different inverse (wMNE and dSPM) and functional
[7] for recent comments). An efficient
connectivity (r2 and PLV) methods. (b) The original network (ground truth). (c) Values (mean ± stansolution, described in this review, is to
dard deviation) of the similarity index computed between the network identified for each combination
compute the functional connectivity at
and the reference "epileptogenic" network used to simulate dense-EEG data. (Image adapted from
[38].)
the level of the brain sources (the source
space). This EEG source connectivity
method
combines
the
excellent resolution of EEG with a
[47]. In this section, we highlight some recent results obtained
superior to exceptional spatial resolution, depending on the
with this approach in Parkinson's.
granularity (coarse to fine grain) of the source model that is
Dense EEG (122 channels) source connectivity was used
used to solve the EEG inverse problem and subsequently idenby Herz et al. [48] in patients with Parkinson's. The results
tify networks at the cortical level.
revealed the effect of dopamine in the reconfiguration of prefrontal-premotor connectivity. Using MEG source connectivity, decreases in alpha1 (8-10 Hz) and alpha2 (10-13 Hz)
Spatial leakage
frequency band connectivity were observed in Parkinson's
A critical issue often raised in the computation of connectivipatients. Most of the alterations were located in the tempoty at the source level is the spatial leakage. As source estiral regions. In a four-year longitudinal study, the same team
mates are spatially correlated, a leakage of inferred sources
also applied MEG source connectivity to 70 Parkinson's
into their local neighborhood often occurs. When the connecpatients to track the resting state of networks, with the goal
tivity method ignores this effect, false connectivity values
of assessing the possible follow-up on the disease progression
computed between distant sources may be interpreted as
[49]. The authors reported a progressive decrease in the local
functional connectivity, although they only reflect the fact
clustering network measure in multiple frequency bands,
that sources share components of the same sensor signal. To
together with a decrease in path lengths at the alpha2 freaddress this issue, several strategies have been proposed to
quency band. These alterations were related to a worsening
remove zero-lag correlations before performing connectivity
in motor function and cognitive performance. This study was
analyses. Other studies suggest that only the long-range conthe first to show that network measures (such as local/global
nections should be kept. However, these solutions may supefficiency) may lead to promising neuromarkers of progrespress important correlations that might occur at zero lag or
sion in Parkinson's.
even between close regions.
Network Similarity Index

(a)

IEEE Signal Processing Magazine

May 2018

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