Information Technology Reference
In-Depth Information
to track precisely all event onsets of each trial. These triggers were received by the
Mitsar system as a logic signal, synchronized with the EEG stream, and recorded as
a supplementary data channel.
8.8.4 Preprocessing
Data were
filtered in the 1
-
40 Hz band-pass region using an order four Butterworth
FIR
filter with linear phase response in the band-pass region. Ocular artifacts were
extracted using the SOBI algorithm (Belouchrani et al. 1997 ) available in the
EEGLAB toolbox (Delorme and Makeig 2004 ). One EOG source corresponding to
eyeblinks was suppressed for each subject. It was manually selected using both the
temporal shape of the source and its topography. All other artifacts were left into the
signal, so as to approximate the conditions of online analysis of EEG data acquired
during BCI operation.
8.8.5 Analysis in the Sensor Space
The analysis in the sensor space is the traditional analysis of the signal as recorded
at each electrode. We are interested in the analysis of the error versus correct trials.
We performed both the analysis of the event-related potential (ERP: both time- and
phase-locked: Lopes Da Silva 2005 ) and analysis of the event-related synchroni-
zation (ERS: time-locked, but not necessarily phase-locked: Pfurtscheller and Lopes
da Silva 1999 ). ERPs were analyzed contrasting the average potential obtained from
each subject at each electrode and time sample. ERSs were analyzed contrasting the
average time
frequency map obtained on each trial from each subject at each
electrode. In order to compute ERS, we employed a multitapering Hanning sliding
window (frequency dependent, with the taper equal to four cycles for each fre-
quency) over the 2
-
32 Hz band using a 1 Hz step, as implemented in the Fieldtrip
software (Oostenveld et al. 2011 ). ERSs were computed on time window [
-
0.5 s
1.2 s] using a time step of 0.03 s and a baseline de
ned as [
1 s 0 s] prestimulus.
The statistical analysis in the sensor space for contrasting
error
versus
cor-
rect
trials needs to be performed for each electrode, discrete frequency, and time
segment in the case of ERS and for each electrode and time segment for ERP data.
In order to account for the extreme multiple-comparison nature of the test, we
employed a permutation strategy. The test chosen is a slight modi
cation of the
supra-threshold cluster size permutation test originally proposed for neuroimaging
data by Holmes et al. ( 1996 ). Here, the statistic is not the supra-threshold cluster
size, but the supra-threshold cluster intensity, de
ned as the sum of the t values
within the supra-threshold clusters. As compared to the test described by Holmes
et al. ( 1996 ), such a statistic is in
uenced not only by the spatial extent of the
clusters, but also by the strength of the effect. The test is sensitive to effects that are
contiguous in space (adjacent electrodes), frequency, and time, in line with phys-
iological considerations. The family-wise error rate for multiple comparisons was
 
Search WWH ::




Custom Search