Brain–Computer Interfaces - Index of

Toward Ubiquitous BCIs 371
controlling a BCI. Furthermore, this “hybrid” task was not considered especially
difficult, according to subjects’ questionnaires [5]. We later validated this approach
in online work with 12 subjects, and then extended this work with a two dimensional
hybrid BCI. Subjects could move a cursor horizontally with SSVEP activity,
and simultaneously control vertical position with ERD activity [1, 12]. Related work
from TU Graz showed that subjects can use ERD activity to turn an SSVEP BCI on
or off [61]. [55] presents these and other examples of hybrid BCIs-all of which were
developed within the last year.
Figure 2 shows three examples of possible hybrid BCI systems. Panel A is based
on the Donchin matrix, the canonical approach to a P300 BCI [9, 75]. Subjects
focus on a particular letter or other character, called the target, and count each time
a row or column flash illuminates the target. For example, if the subject wanted to
spell the letter F, she would not count the column flashed in Panel A, but would
count each time the top row or rightmost column flashed. This system could be
hybridized if the rows and/or columns oscillated in addition to flashing. The added
information from SSVEP activity could improve accuracy or reduce selection time.
Panel B shows a variant on the Hex-O-Spell system that might include two arrows at
the top that flash or oscillate, allowing people to choose within an application with
ERD and use P300 or SSVEP to switch applications. These added signals might
also be used to confirm or jump to specific selections. Panel C shows a new BCI
system that could train users to combine P300, ERD, and SSVEP activity. Each box
is bracketed by two feedback bars that reflect the strength of the user’s brain activity
associated with that box. Three boxes contain oscillating checkerboxes, and thus the
feedback bars represent the SNR of the relevant EEG features. The other two boxes
depend on left or right hand motor imagery, and the feedback bars thus reflect ERD
over relevant areas and frequencies. With this system, users could send one of five
signals, such as moving a cursor in two dimensions plus select, with SSVEP and
ERD activity. Each box might also flash a word containing the position of that box.
If the user silently counts the flash, then this produces a P300 that could confirm
attention to that box. Like other hybrid BCIs, the added signal could allow the user
to choose from more possible targets, thus increasing N, or send information more
accurately. Some approaches might yield stronger brain activity. Furthermore, if the
BCI included software that could learn how to best combine the different signals,
then hybrid BCIs might work in people who lack one type of brain signal. This
could substantially improve BCI reliability.
Hardware integration has typically foundered on the same assumption that hampered
software and functional integration: that BCIs are standalone systems. The
necessary hardware, such as sensors, an amplifier, and computer, are typically
devoted only to BCI operation. This has begun to change as EEG sensors have been
mounted in glasses, headphones, gamer microphone systems, and other headwear
[65, 79]. As head-mounted devices become more ubiquitous, integrated EEG and
other sensors will become more common. This integration of BCI hardware, software,
and functionality is essential for wider BCI adoption. Integration is likely to
continue, yielding increasingly transparent and ubiquitous BCIs.