Brain-computer interface with language model-electroencephalography fusion for locked-in syndrome

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Noninvasive brain-computer interface (BCI) systems exist for locked-in syndrome (LIS).
• Current systems lack integrated statistical language models for text generation.
• This gap limits communication efficiency for individuals with LIS.

Purpose of the Study:

• To develop and evaluate a noninvasive BCI system for LIS communication.
• The system integrates rapid serial visual presentation (RSVP) of symbols.
• It employs electroencephalography (EEG) and a unique classifier with language model fusion.

Main Methods:

• Developed the RSVP Keyboard with letters presented at 2.5/sec.
• Utilized machine learning for EEG-based classification and Bayesian fusion for language model integration.
• Assessed system performance with nine individuals with LIS and nine healthy controls through word generation tasks at five difficulty levels.

Main Results:

• Six individuals with LIS and nine controls completed the study.
• All LIS participants mastered basic spelling (Level 1).
• One LIS participant achieved advanced spelling (Level 5), comparable to control performance.

Conclusions:

• An EEG-based BCI system incorporating a statistical language model shows promise for individuals with incomplete LIS.
• The developed system enhances communication capabilities for those with severe motor impairments.
• Further system enhancements are under investigation to improve performance and usability.