Brain–Computer Interfaces - Index of

106 E.W. Sellers et al.
fixation was verified by vertical and horizontal infrared corneal reflectance [26].
This implies that spatial (also called covert) attention contributes to the observed
effects.
Furthermore, it is clear that fixation alone is not sufficient to elicit a P300
response. Evidence for this is provided by numerous studies that present target and
non-target items at fixation in a Bernoulli series (e.g.,[16]). In Fig. 8, Rows 2 and
3 show average responses to a standard oddball experiment. The stimuli “X” and
“O” were presented at fixation, with a probability of .2 and .8, respectively. The
responses shown in Row 2 had a stimulation rate similar to that of P300 BCI experiments
(every 175 ms); the responses in Row 3 used a stimulation rate similar to
standard oddball experiments (every 1.5 s). The negative peak around 200 ms is
present in these oddball conditions, as it is in the 6 × 6 matrix speller condition
shown in Row 1. If fixation alone were responsible for this response, in Rows 2 and
3 the target and non-target items would produce equivalent responses because all
stimuli are presented at fixation in the oddball conditions. The responses are clearly
not the same. This implies that a visual P300 BCI is not simply classifying gaze
direction in a fashion analogous to the Sutter (48) visual evoked potential communication
system. A BCI that requires the user to move their eyes may be problematic
if they have lost significant eye muscle control, regardless of the type of EEG signal
being used.
It is also possible that occipital component represents the negative part of a dipole
in the temporal-parietal junction (the area of cortex where the temporal and parietal
lobes meet) that is related to the P300 [11, 42]. By this view, neither the positive
component at the midline nor the negative occipital component is generated
directly below their spatial peaks on the surface. Rather, both would be generated
at the temporal-parietal junction, an area known to be closely associated with the
regulation of attention.
A P300 BCI must be accurate to be a useful option for communication. Accurate
classification depends on effective feature extraction and on the translation algorithm
used for classification. Recently, we tested several alternative classification
methods including SWLDA, linear support vector machines, Gaussian support vector
machines, Pearson’s correlation method, and Fisher’s linear discriminant [21].
The results indicated that, while all methods attained useful levels of classification
performance, the SWLDA and Fisher’s linear discriminant methods performed
significantly better than the other three methods.
4 A BCI System for Home Use
In addition to working to improve SMR- and P300-based BCI performance, we are
also focusing on developing clinically practical BCI systems that can be used by
severely disabled people in their homes, on a daily basis, in a largely unsupervised
manner. The primary goals of this project are to demonstrate that the BCI system
can be used for everyday communication, and that using a BCI has a positive impact