Item

Abstract

Brain-Computer Interfaces (BCI’s) enable us to infer intentional control signals from brain activity. The Visual Speller is a BCI based on event related potentials (ERP’s) in the electroencephalogram, such as the P300 (a positive deflection in the EEG about 300 ms after a rarely occuring stimulus). In the classical paradigm one trial (i.e. prediction of one symbol) consists of successive highlightings of one or more symbol(s) on a visual grid presented to the subject. The stimulus events in which the symbol of interest was highlighted will result in an enhanced ERP. This ERP, being stronger than the ERP’s elicited by non-target stimulus events, can be used for prediction of the letter the subject was focussing on using some machine learning algorithm, for example the support vector machine. The more symbols are highlighted simultaneously the faster the speller could potentially work. A stimulus code that uses few events per trial (and
thus shows many symbols at once) is called dense. The tradeoff against code density is that the signal to noise ratio becomes worse with increasing stimulus frequency: the
P300 signal is reported to be strongest when the target symbol frequency is lowest. The stimulus code in which only one symbol per stimulus event is presented, is a maximally
sparse code. It has been proposed that high bitrates of information transfer in a visual speller can best be achieved with sparse stimulus codes. However sparse codes have long trial durations. In order to improve the information transfer rate, we tried to use denser stimulus codes that present fewer stimulus events per trial. To investigate the effect of stimulus type on classification accuracy and the interdependence of stimulus code and type, we explored new stimulus types including ones exploiting recent findings in neuropsychology, such as change blindness and isoluminant color motion. We show that, using appropriate stimuli, denser codes, and hence fewer stimulus events, yield sufficient classification accuracy to achieve competitive bitrates.