Page 142 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 139

• For workload indicators, it is common to record EEG for 1–5min, cal-
culate the PSD over that time period, extract features such as mean fre-
quency from the PSD, and use classification algorithms to translate those
features into different levels of workload (Novak et al., 2014). This
workload level is then assumed to apply to the entire 1–5-min time
period. Similarly to motor/mental imagery, popular classification algo-
rithms include, for example, linear discriminant analysis, support vector
machines, and artificial neural networks (Novak et al., 2014). However,
compared to motor/mental imagery, there has been little development
of advanced algorithms and little comparison of different algorithms to
each other. Thus, workload classification is still largely based on factors
such as ease of implementation and developers’ personal preferences.
The different paradigms can also be combined to some degree in order to
improve BCI performance. One classic example is to use SSVEPs to control
the elbow function of an artificial limb and motor imagery to control the
grasp function of the same limb (Horki et al., 2011). Similarly, a wheelchair
can be controlled by using motor imagery of the left and right hands to trig-
ger left/right turns and by using the P300 to control the acceleration (Long
et al., 2012). A different example is to use SSVEPs and the P300 response
simultaneously using a screen that shows P300 visual stimuli on one part
of the screen and SSVEP stimuli on another part of the screen
(Bi et al., 2014).

1.2 Electrocorticography and Intracortical Electrodes
The electrocorticogram (ECoG) is similar to the EEG, but is recorded
invasively with electrodes placed on the surface of the brain using a surgical
procedure. This results in a significantly higher SNR than in EEG; however,
due to invasiveness, the biomechatronic applications of ECoG are largely
limited to severely impaired users (e.g., tetraplegics). Similarly, intracortical
electrodes are placed inside the brain itself, resulting in an even higher SNR
than ECoG and allowing measurement of the electrical activity of small,
very specific regions of the brain. However, they are again very invasive
and are frequently rejected by the cortical tissue surrounding them, gradually
resulting in loss of the signal (Groothuis et al., 2014).
Signal processing for the ECoG and intracortical electrodes can be similar
to that seen in the EEG, but is characterized by less noise and higher pattern-
recognition accuracy. For example, while EEG is commonly bandpass-
filtered between 5 and 30Hz, the lower cutoff frequency for ECoG can