Page 155 - Handbook of Biomechatronics

152 Domen Novak

matrix of letters, and the individual columns of the matrix light up one after
another. The user focuses on the letter that they wish to select, and this trig-
gers a P300 response when the column containing that letter lights up. Once
the correct column has been identified, the screen next lights up each indi-
vidual row of the matrix one after another; again, a P300 response is trig-
gered when the row containing the letter of interest lights up. The
selected letter is then added to the message and the process repeats with
the next letter that the user wishes to select. This matrix-based speller
achieved a mean letter selection accuracy of 95% and a mean information
transfer rate of 12bits/min. This principle is shown in Fig. 3.
Significant work on P300 spellers has been performed since their intro-
duction in the 1980s and has included innovations in both EEG processing
(e.g., improved P300 recognition) as well as user interface design. For exam-
ple, researchers have experimented with different letter layouts in both two
and three dimensions, have added “autocomplete” functions similar to those
on mobile phones, and have developed letter matrices for different languages
(Rezeika et al., 2018). In a particularly interesting variation, Kaufmann et al.
(2011) superimposed faces of different famous people such as Albert Einstein
over individual letters, allowing participants to focus on both faces and letters
for stronger P300 elicitation. Such improved P300 spellers now achieve
information transfer rates of up to 50–60bits/min in able-bodied users
(Rezeika et al., 2018), though it is often necessary to perform multiple iden-
tification trials per letter if the signal quality is low.
Aside from P300 spellers, spellers based on SSVEPs and motor imagery
have also been gaining in popularity. One of the best-known SSVEP spellers
is the Bremen BCI speller (Volosyak et al., 2010), which presents a virtual
keyboard on the screen next to five buttons flashing at different frequencies:
up, left, down, right, and select. Participants can use these buttons (via the
SSVEP BCI) to control a cursor on the keyboard and thus select the desired
letter. The letters are arranged according to their usage frequency in the
English language, and each selected letter is spoken out loud by the system
as a form of confirmation. As with P300 spellers, the interface can be
expanded with word prediction algorithms that automatically complete
the word and/or suggest the next word in the sentence. Furthermore, newer
versions of the Bremen speller have added visual feedback about the strength
of the SSVEP signal: when the speller detects that the user is looking at one
of the five buttons, that button’s size increases to indicate that a selection is
about to be made. Through such improvements, SSVEP spellers have
achieved information transfer rates of up to 300bits/min in able-bodied
users (Rezeika et al., 2018).