Brain-computer interface (BCI) systems like the visual P300 speller may be used to replace communication abilities in people who are neurologically "locked-in" – typically unable to move or speak. These individuals, who cannot use commercially available augmentative and alternative communication (AAC) or speech output devices, are a key target user population for this BCI, which leverages a user’s brain signals to spell characters. Presently, BCIs are primarily used in the research sphere, as their communication rates remain impractical for everyday use. BCI users are typically required to spend a considerable amount of time training a BCI prior to operation, and must be willing to accept a fairly slow communication rate. In order to address the overarching goal of transitioning P300 speller systems from the lab into the clinic or home, a team of basic scientists, engineers, and clinicians has been assembled.
Reliable P300 detection is central to our aims of reducing BCI training time and improving achievable communication rates. Improving the training procedure, visual interface, and stimulus selection and character selection strategies are key aspects to achieving these goals. Recently, we have developed techniques to optimize P300 speller stimulus selection in order to minimize the amount of time needed to spell a character. We are currently focused on leveraging deep weighted transfer learning to reduce BCI training times and adaptive classification to update BCI parameters according to changing brain signal properties over time during extended use.