Improving Donchin's Matrix Speller by Exploiting Natural P300 Amplitude Variations
P300 potentials are used increasingly frequently in BCI (also our
analogue BCIs are based on P300s and their modulations). While
significant improvements have been made in the detection of P300s, prior to this project nobody had exploited the variability in shape and timing of
P300 waves within the famous matrix speller devised by Farwell and Donchin. Within our Analogue Evolutionary Brain-Computer Interfaces EPSRC
Project (EP/F033818/1) we were very familiar with P300 amplitude variations. Thus, we decided to fill this gap.
Firstly, we documented and exploited the modulation in the amplitude of
P300s related to the number of non-targets preceding a target in
Donchin's classical BCI speller. Our approach uses an appropriately weighted average of the responses
produced by a classifier during multiple stimulus presentations,
instead of the traditional plain average. In this way, we weigh more
heavily events that are likely to be more informative, thereby
increasing the accuracy of classification. Later we also modelled mathematically and tested on a much larger set of subjects the approach.
Tests showed that our approach provides a marked statistically significant improvement in accuracy over the top-performing algorithm in the literature. The method and theoretical models we proposed are also general and can easily be used in other P300-based BCIs.