An active recursive state estimation framework for brain-interfaced typing systems

摘要

Typing systems driven by noninvasive electroencephalogram (EEG)-based brain-computer interfaces (BCIs) can help people with severe communication disorders (including locked-in state) communicate. These systems mainly suffer from lack of sufficient accuracy and speed due to inefficient querying to surpass a hard pre-defined threshold. We introduce a novel recursive state estimation framework for BCI-based typing systems using active querying and stopping. Previously, we proposed a history-based objective called Momentum which is a function of posterior changes across sequences. In this paper, we first extend the definition of the Momentum, propose a unified framework that employs this extended Momentum objective both for querying and stopping. To provide a practical example, we employ a language-model-assisted EEG-based BCI typing system called RSVP Keyboard. Our results show that proposed framework on average improves the information transfer rate (ITR) and accuracy at least 52% and 8.7%, respectively, when compared to alternative approaches (random or mutual information).