Magnetoencephalography 12 subcortical structures of the AAL

0 Comment

Magnetoencephalography (MEG)

recording was performed inside a magnetically shielded room (Vacuumschmelze
GmbH, Hanau, Germany) with a whole-head system (Elekta Neuromag Oy, Helsinki,
Finland), which utilizes 306 channels, 102 of which are magnetometers and 204
gradiometers. Patients were placed in supine position with no task and with
closed eyes for the identification and localisation of interictal epileptiform
activity. The sampling frequency was 1250Hz and the data were filtered online
with a 410Hz anti-aliasing filter and a 0.1Hz high-pass filter. Using a three
dimensional digitizer (Fastrak; Polhemus, Colchester, VT. U.S.A.) head location
and scalp outline were digitized. Head localisation was performed with four to
five head-localisation coils. Scalp surface was co-registered with the anatomic
MRI of the patient using surface-matching.
By spatially filtering the raw data offline, artefacts were removed. This was
done using the temporal extension of Signal Space Seperation (tSSS)(BRON), a function of
MaxFilter software (Elekta Neuromag Oy; version 2.1)

Atlas based Beamforming

Using an
atlas based  beamforming approach,
modified from Hillebrand et al. (BRON), neuronal activity was reconstructed. Time series of
neuronal activation were reconstructed to the centroids (Voxels located in the
middle of a Region of Interest) of 78 cortical Regions of Interest (ROIs) and
12 subcortical structures of the AAL (automated anatomic labelling) atlas(BRON). Centroids were inversely
transformed to the co-registered MRI of the patient. Then,  for each centroid, time series were
reconstructed using a scalar beamformer (Elekta Neuromag Oy; beamformer;
version 2.2.10). 

The lead
fields, which are obtained by compiling a spherical head model on the basis of
the anatomic MRI of the patient, data covariance and noise covariance determine
the beamformer weights. These weights function as a spatial filter and are
calculated for each centroid separately to minimise the loss of signals
originating from the ROI (centroid?)
and to mitigate influence by all
other signals. To calculate the covariance, all data was used (typically 15
minutes), filtered in the broadband (0.5-48Hz).
To simulate noise covariance when estimating the optimum source orientation for
the beamformer weights, a unity matrix was used. Finally, the broadband data
was projected through the normalised beamformer weights to obtain time series
for all ROIs.

Network and connectivity

For each
patient, MEG data was analysed in the broadband (0.5-48Hz) using Brainwave (version
available from to gain insight on the connective
properties of their brain. Based on the aforementioned virtual electrode time
series, a functional network was constructed. Functional connectivity was
estimated using the phase lag index (PLI), which indicates asymmetry in the
distribution of instantaneous phase differences between two time series. PLI
ranges from 0 – 1, zero meaning no connectivity or noise and one indicating
full synchronisation. Using the PLI, a minimum spanning tree (MST) was
constructed between ROIs, which served as nodes. A MST is formed by adding the
highest Phase Lag Indexes to an empty network, while not including loops,
connecting all nodes in the network. Because the number of edges remains
constant when the number of nodes is kept constant, direct comparison across
groups is possible. Using the MST, Betweenness Centrality (BC) was calculated
for each node. BC is a value between 0 and 1, which is calculated based on the
number of shortest paths in a network that pass through it, and can be used in
identifying hubs in a network. Additionally, the degree and eccentricity of the
network was calculated. MST, degree and eccentricity of the network were
calculated with the average of 205 epochs, with 4096 samples per epoch and a
sample frequency of 1250. The 205 epochs used all contained epileptiform
activity and artefacts.