Event-Related Potential-based Brain Computer Interface Speller based on a Novel Sequential Protocol

Most BCI spellers based on visual stimulus presentation rely on the oddball effect, which causes the brain to respond with a P300 ERP to a rare random stimulus of interest (e.g., the flashing of the letter the participants intends to input). Naturally the information transfer rate (ITR) of a speller depends on how many such relevant stimuli one can react to in a given time. So, fairly short SOAs are commonly used, resulting in a reduced amplitude of P300s, very big deformations w.r.t. to its text-book shape, and the contamination from near targets (where a P300 like ERP may be elicited). All of which, hampers classification accuracy and correspondingly limits ITRs. In previous research on a BCI for mouse cursor control, a sequential non-oddball-based stimulation protocol was developed where 8 stimuli (representing different directions of desired movement) were arranged in a circle and were flashed sequentially. The colour of the flashes was randomly chosen and participants were asked to mentally name the colour of the attended stimulus. This produced better recognisable P300s and, so, significant improvements of AUC and ITR. In this thesis we apply this idea to a BCI speller, where 36 letters are organised around a circle and they are highlighted sequentially in either green or red and users need to mentally name the colour of the target letter. Each revolution required 2 seconds in one experiment and 3 seconds in another experiment. We compared this speller against a traditional 6 × 6 matrix speller where all letters are highlighted twice (row and column) also within 2 seconds or 3 seconds. All participants used both protocols in counterbalanced order. Results show that our sequential speller produces much bigger and cleaner P300s and, in the 2 second condition, this leads to a significantly higher classification accuracy and approximately doubles the ITR w.r.t. the Donchin’s speller.