ABSTRACT

Stimulus-driven brain–computer interfaces (BCIs), such as the P300 speller, rely on eliciting and detecting event-related potentials (ERPs) that are embedded in noisy electroencephalography data. However, these BCIs are currently limited by their relatively slow spelling speeds due to repetitive data measurements to increase the signal-to-noise ratios of the elicited ERPs for improved selection accuracy. In addition, psycho-physiological factors such as refractory effects limit the ability to elicit a strong ERP response with every target stimulus event presentation. The stimulus presentation pattern encodes information about an intended message a user wishes to communicate. The role of the BCI is to translate the user’s attention-modulated responses to the stimulus events into a selection that conveys the user’s intent. Consequently, a BCI can be approached from an information-theoretic perspective in order to determine how best to reliably encode information to be robust to noisy channel transmission errors. In this chapter, we present a principled approach to design the stimulus presentation paradigm for the P300 speller that exploits an information-theoretic approach to maximize the information content that is presented to the user. We use a probabilistic performance prediction method to evaluate and compare the performances of different stimulus presentation configurations during the design optimization process. We select a final configuration that maximizes BCI performance while minimizing refractory effects. We present results with online BCI use, which demonstrate significant performance improvements with our performance-based stimulus presentation paradigm compared to the conventional method of stimulus presentation.