Page 48 - Read Online

Page 44 Valerio et al. Neuroimmunol Neuroinflammation 2021;8:42-9 I http://dx.doi.org/10.20517/2347-8659.2020.30

processed with a set of match filters, each comparing the signal with a scaled version of a prototype
biphasic waveform. The identified complexes were then automatically reviewed, excluding outliers and
waveforms showing repetitive firings, as some waveforms could have a shape similar to that of an SBC, but
they could not satisfy some properties indicated in previous publications [14,17,18] .

Source localization
Source localization in EEG refers to the detection of the sources inside the brain that generate the electrical
activity acquired on the scalp. When the available electrodes are in a small number (as in our cases, in
which either 12 or 18 channels were available for the two cases, respectively), the source detection may
[20]
have low accuracy , but can provide useful information on the brain areas that are most involved in the
inflammatory activity.

From the mathematical point of view, the dipoles inside the brain that produce a scalp potential that best
fits the original data are sought. The problem can be written as follows:

M = GD + n (1)

where each row of the matrix M contains a measured EEG, G is the Lead-Field matrix that describes
the response of the activation of N different dipoles, whose level of activity (collected in D) should be
estimated, and n is an additive noise, assumed spatially and temporally white. Different methods have been
proposed in the literature to solve this problem [19,21] . In this study, the minimum norm estimation (MNE)
[22]
was used . It searches for the solution with minimum power, by minimizing the following regularized
functional

2
2
U(D) = ll M - GD ll + all D ll (2)
where a is a regularization parameter to constrain the power of the solution (chosen in this study to be
equal to the mean of the eigenvalues of G G divided by 2,500; however the estimation was stable to a
T
variation of a by an order of magnitude). It brings to the following solution to recover the sources:

T
-1 T
D MNE = (G G + aI ) G M (3)
N
where I is the identity matrix of dimension N × N.
N
Localization of sources of SBCs
The waveforms of interest are concentrated in a low-frequency band. After visual inspection of the
portions containing SBCs, each EEG trace was then band-pass filtered between 0.5 and 5 Hz (Chebychev
filter of order 6 of type I), and the common-mode was removed. This filter provided clean EEG traces,
focusing the source detection mostly on the waveforms of interest. Half second long windows centered
on the identified SBCs were concatenated to generate the rows of matrix M in equation (1). Then, MNE
[23]
(FieldTrip implementation ) provided a discrete brain model made of equivalent current dipole sources,
containing the mean activation over time for each source location. The same procedure was applied to EEG
data with the same duration obtained concatenating windows not including SBCs, to estimate the average
background activity. The difference between the medians of dipole intensities during SBC onsets and in the
background was then investigated (checking significant differences with the Wilcoxon rank-sum test for
equal medians, with significance level P = 0.001).

First case
A 4-year-old subject was considered. At the time of the presentation, the patient presented with fever,