IEEE Consumer Electronics Magazine - March 2018 - 8

infinite impulse response, fourth-order
Butterworth bandpass filter with 4-Hz
bandwidth, i.e., 6-10 Hz, 8-12 Hz, and
up to 20-24 Hz. Then, the common spatial pattern (CSP) algorithm was applied
to each bin to contrast variances of the
EEG signals in the three states in accordance with one versus all strategy. For
each contrast condition, Team NeuroImaging took the two CSP components
corresponding to the minimum and the
maximum variance ratios, resulting in a
total of six CSP components per frequency bin altogether. The next (optional) layer of the pipeline was the least
absolute squares shrinkage operator feature selection. Finally, an instantaneous
variance estimator was applied to each
CSP component across a 500-ms time
window preceding the current time
point, and the resulting feature vectors
were presented as an input to the random forest [8] classifier.

A PACKAGE FOR ANALYSIS OF
PRESTIMULUS CHARACTERISTICS
OF THE EEG SIGNAL TO PREDICT
POSTSTIMULUS REACTIONS
Team 1.1.1.1.1 included Anna Stetsenko, Evgeny Blagoveshchensky, Olga

Neurotechnology has
rapidly advanced into
consumer spaces.
Vakhrameeva, Petr Vasiliev, and Alexey
Harauzov, from the Pavlov Institute of
Physiology, RAS, St. Petersburg, Russia. The Team 1.1.1.1.1 project is
described as follows. Recent practice in
cognitive neuroscience tends to combine different techniques for analyzing
different aspects of brain activity in one
behavioral task, to observe time-coarse
brain signals during task performance
(via EEG) and localization of active
zones [via magnetic resonance imaging
(MRI); magnetoencephalography (MEG)]
or changes in brain-signal dynamics/
behavioral responses during electrical
or magnetic stimulation [e.g., transcranial direct current stimulation (tDCS);
transcranial magnetic stimulation
(TMS)]. Many tools, e.g., EEGLAB, statistical parametric mapping, and
xjView, were made available to analyze
separately data obtained by different
methods, e.g., psychophysics, EEG,
MRI, MEG, and TMS. However, none

Spearman

p-Value
0.0923

0.7

0.0148

0.6

-0.0628

0.5

-0.1403

0.4

-0.2179

0.3

(a)

(b)

Spearman

p-Value
0.1904

0.7

0.0907

0.6

-0.009

0.5

-0.1086

0.4

-0.2083

0.3

(c)

(d)

fiGURE 1. The main results of correlations between pre- and poststimulus brain signals.
(a) The distribution of the Spearman correlation between prestimulus alpha oscillation
power and EMP over the scalp. (b) The p-values of the correlation coefficients from the
plot in (a). (c) The distribution of the Spearman correlation between prestimulus alpha
oscillation standard deviations and EMP over the scalp. (d) The p-values of the correlation
coefficients from the plot in (c).
8 IEEE Consumer Electronics Magazine

march 2018

of these methods would allow for the
integration and analysis of data obtained
from different techniques.
Team 1.1.1.1.1 aimed to create an
open-source package with a userfriendly graphical user interface, which
would allow nonprogramming or littleprogramming cognitive neuroscientists
to compare prestimulus characteristics
of the EEG signal with a variety of
poststimulus reactions. Poststimulus
reactions include, e.g., behavioral reactions (reaction times and/or accuracy of
performance) and event-related potential (ERP) components latency/amplitude. This reaction can also be in terms
of evoked motor potentials (EMPs) in
TMS paradigms, or any other data,
obtained with different techniques.
Phase synchronicity and amplitudes
of prestimulus oscillations (usually, but
not limited to, alpha band diapason)
affect poststimulus ERP components
and behavioral reactions. Analysis of
prestimulus microstate of the EEG signal can help to predict poststimulus
reaction. However, a huge variety of
new techniques (such as TMS, tDCS,
and so forth) are showing improvement over existing ones (by reducing
signal-to-noise ratio or the improvement of existing techniques for EEG
recording, which would allow researchers to observe effects in few-trial ERP
or even single-trial ERP). Thus, a new
tool for analysis of such data is needed for a large group of experts in cognitive neuroscience who are novices
in programming.
Team 1.1.1.1.1 put together a tool
package under MATLAB, using some
functions developed by the Berlin BCI
company. To test the package, a data set
by Z. Iscan et al. [1], was used. Results
are shown in Figure 1. This package is
designed for open-source access. Further
testing with different data sets (reaction
time data sets, accuracy of performance)
will be needed. New features are also
being added to the package, including
cross-channel correlation measurement
and autocorrelation of the EEG signal.

BCI INTERFACE FOR DATA LABELING
Team Vostok included Roman O.
Malashin, Svyatoslav V. Ponomarev,

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March 2018