Page 140 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 137

The placement of electrodes depends on the EEG paradigm used and has
a huge effect on BCI performance. While some researchers prefer to place
electrodes at evenly spaced location across the scalp (thus obtaining both rel-
evant and irrelevant information, which is useful for, e.g., filtering), elec-
trodes can also be placed only at locations relevant to the EEG paradigm
of interest. For example, electrodes for SSVEP detection are commonly
placed near the visual cortex (Nicolas-Alonso and Gomez-Gil, 2012), elec-
trodes for motor imagery are commonly placed near the motor cortex, and
electrodes for workload recognition are commonly placed near the frontal
lobe (Novak et al., 2014).

1.1.3 Signal Processing and Pattern Recognition
EEG signal processing generally begins with a bandpass filter that removes
very low-frequency artifacts as well as high-frequency noise. However,
many artifacts cannot be removed using simple bandpass filtering. For exam-
ple, eye artifacts such as blinks appear in EEG measured from the frontal lobe
since the eyes are located near the front of the brain, but these artifacts over-
lap with the frequency bands of the EEG (Vaughan et al., 1996). Similarly,
head movement causes artifacts in EEG measured from electrodes near the
back of the head due to activation of the neck muscles. These artifacts can be
reduced using secondary sensors. For example, eye artifacts can be removed
from the EEG by using the electrooculogram (EOG) as a reference for
noise-removal algorithms (Croft and Barry, 2000); similarly, head move-
ment can be detected using accelerometers or neck electromyography
(EMG) and used as a reference input to adaptive filtering algorithms. If sec-
ondary sensors are not available, we can instead use spatial-filtering methods
such as Laplacian filtering, which enhance localized activity while
suppressing components that are present in many signal channels (such as
blink artifacts, which are present in all signals measured from frontal areas).
Once the SNR has been improved, patterns corresponding to different
desired commands or mental states must be identified from the EEG. This
can be done in one of two different operating modes: synchronous or asyn-
chronous. In synchronous mode, commands are only accepted by the BCI at
specific times that are clearly communicated to the user (e.g., via visual sig-
nal). At each of these specific times, a window of the EEG is analyzed by the
BCI. In asynchronous mode, commands are accepted by the BCI at any
time, and a sliding window of the EEG signal (with lengths ranging from
250 to 1000ms for SSVEPs, P300, and motor imagery (Novak and
Riener, 2015) and 1–5min for workload indicators (Novak et al., 2014))