Biomedical Engineering Reference
In-Depth Information
spatial pattern (CSP) [21] and independent component analysis (ICA) methods [22]
have been proved to be very successful in forming a more informative virtual EEG
channel by combining multiple real EEG channels, as has been done for SMR-based
BCI.
For most of the cases, the output of the signal processing is a set of features that
can be used for further pattern classification. The task of pattern classification of a
BCI system is to find a suitable classifier and to optimize it for classifying the EEG
data into predefined brain states, that is, a logical value of class label. The process
usually consists of two phases: offline training phase and online operating phase.
The parameters of the classifier are trained offline with given training samples with
class labels and then tested in the online BCI operating session. Various classifiers
have been exploited in BCI research [23], among which the Fisher discriminant anal-
ysis and SVM classifiers bear the merit of robustness and better generalization abil-
ity. When considering pattern classification methods, keep in mind that the brain is
an adaptive and dynamic system during interaction with computer programs. Basi-
cally, a linear classifier with low complexity is more likely to have good generaliza-
tion ability and be more stable than nonlinear ones, such as a multilayer neural
network.
8.1.2.3 BCI Output: Real-Time Feedback of Control Results
As shown in Figure 8.1, two links are used to interface the brain and external
devices. The BCI core as described earlier comprised of a set of amplifier and com-
puter equipment with the proper program installed can be considered as a “hard
link.” Meanwhile, the feedback of control results is perceived by one of the BCI
user's sensory pathway, such as the visual, auditory, or tactile pathway, which
serves as a “soft link” to help the user adjust the brain activity for facilitating the BCI
operation.
As discussed before, the BCI user needs to produce specific brain activity to drive
the BCI system. The feedback tells the user how to modify their brain's encoding in
order to improve the output, as happens during a natural movement control through
the normal muscular pathway. It is the feedback that closes the loop of the BCI,
resulting in a stable control system. Many experimental data have shown that, with-
out feedback, BCI performance and robustness are much lower than in the feedback
case [12, 24]. From this perspective, the performance of a BCI system is not only
determined by the quality of the BCI translation algorithm, but also greatly affected
by the BCI user's skill of modulating his or her brain activity. Thus, a proper design
for the presentation of feedback could be a crucial point that can make a difference
in terms of BCI performance.
8.1.3 Oscillatory EEG as a Robust BCI Signal
Evoked potentials, early visual/auditory evoked potentials like P100 or late poten-
tials like P300, are low-frequency components, typically in the range of tens of
microvolts in amplitude. As a transient brain response, an evoked potential is usu-
ally phase locked to the onset of an external stimulus or event [25], although oscilla-
tory EEG, such as SSVEP or SMR, has a relatively higher frequency and larger
Search WWH ::
Custom Search