180
Bioinformatics of the Brain
are the preferred paradigms, regarding reactive BCI. A compressive review by
Xu et al. [4] overviewed various encoding paradigms for BCI systems.
7.3.3
Pre-processing of EEG Signals
The pre-processing stage prepares the recorded signals in a suitable form for
further steps by removing any unwanted components embedded within the
EEG signal and reorganizing the data. Effective pre-processing contributes to
an elevation in signal quality, thereby resulting in enhanced feature separa-
bility and improved performance in classification [31].
Although EEG allows us to measure brain electrical signals, it does not
solely display these signals. Many noises, referred to as artifacts, can occur
during EEG recording, and these noises can interfere with the signals ob-
tained from the brain. Some of these noises are related to the individual be-
ing recorded (body movement, eye movements and/or blinks, sweating, etc.),
while others arise from technical reasons (50/60 Hz artifact, cable movement,
improper placement of electrodes, etc.) [32]. Various methods are used to clean
EEG data from noise [33, 34].
The most important sources of physiological artifacts in BCI systems are
Electrooculography (EOG) and electromyography (EMG) artifacts, including
eye and body movements. Especially for eye movements and blinks, Indepen-
dent Component Analysis (ICA) is the most preferred methodology to exclude
blinking patterns from the EEG signals. Although ICA is also used for EMG
artifacts, automatic artifact rejection algorithms based on defined criteria or
manual rejections by identifying specific regions affected by artifacts are the
common ways to eliminate muscle artifacts. However, both ways of cleaning
processes are challenging since automatic techniques could cause the loss of
valuable data whereas manual rejection is more labor intensive and some-
times subjective evolution [31]. Therefore, it could be more reliable if cleaning
processes are done by experts.
Moreover, technical problems could occur during the recording of EEG and
destroy the actual brain activity data. One of them is line interference, also
called 50/60 Hz noise. These kinds of noise can be eliminated from the data
by using Notch Filtering. The second technical problem most often encoun-
tered is damaged electrodes which could not record the signal accurately. In
that case, instead of losing the data completely, interpolation techniques that
fill in missing or rejected data points by estimating values from neighboring
electrodes are used.
During the EEG recording, utilizing a physical reference is essential and
the signal at each electrode is determined by subtracting the electric potential
at its designated location from the electric potential at the location of the ref-
erence electrode. In the pre-processing phase, the reference electrode used in
recording could be reorganized by applying a re-referencing process to data.
Re-referencing is preferred for various purposes, including noise reduction,
adjusting for head model variability, and Adaptation to Analysis Techniques.