Language-Model Assisted Brain Computer Interface for Typing: A Comparison of Matrix and Rapid Serial Visual...

Area of Science:

• Neuroscience and Biomedical Engineering
• Human-Computer Interaction

Background:

• Noninvasive electroencephalography (EEG)-based brain-computer interfaces (BCIs) commonly use event-related potentials (ERPs) for intent detection.
• Various symbol presentation paradigms are employed in EEG-based BCI typing systems to elicit ERPs.

Purpose of the Study:

• To experimentally assess the speed, signal quality, and accuracy of a language-model-assisted BCI typing system.
• To compare three distinct presentation paradigms: row-column presentation (RCP), single-character presentation (SCP), and rapid serial visual presentation (RSVP).

Main Methods:

• An experimental study was conducted evaluating a language-model-assisted BCI typing system.
• Three visual stimulus presentation paradigms (RCP, SCP, RSVP) were utilized to induce ERPs for character selection.
• Key performance metrics including typing speed, signal quality, and classification accuracy were analyzed.

Main Results:

• Signal quality and classification accuracy were found to be comparable between the matrix-based (RCP/SCP) and RSVP paradigms.
• Matrix-based paradigms generally allowed for lower inter-trial-interval (ITI) values.
• Optimal presentation paradigm and ITI configuration demonstrated user dependency.

Conclusions:

• Both matrix-based and RSVP paradigms offer viable options for BCI typing systems regarding signal quality and accuracy.
• The optimal configuration for BCI typing performance is user-specific.
• Future BCI typing systems should consider offering users a choice of presentation paradigms and variable ITI settings.