Offline analysis of context contribution to ERP-based typing BCI performance

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Brain-computer interfaces (BCIs) enable communication through electroencephalography (EEG) signals.
• Current BCI typing systems are limited by slow speeds due to the need for multiple signal repetitions for accuracy.
• Improving typing speed without compromising accuracy is crucial for practical BCI applications.

Purpose of the Study:

• To increase the symbol rate of EEG-based BCI typing systems.
• To investigate the use of context information, specifically language models (LMs), to enhance BCI performance.
• To evaluate the fusion of EEG features with LM evidence for improved intent detection.

Main Methods:

• Utilized event-related potentials (ERPs) from EEG to detect user intent.
• Employed Bayesian fusion of an n-gram symbol model with EEG features.
• Used regularized discriminant analysis for ERP feature extraction.
• Evaluated target detection accuracies with varying LM orders and ERP repetition counts.

Main Results:

• Language models significantly improve letter classification accuracy in BCI typing.
• A single-trial ERP detection combined with a 4-gram LM achieved performance comparable to 3-trial ERP classification for non-initial letters.
• The fusion of EEG and LM evidence demonstrated substantial benefits.

Conclusions:

• Fusing evidence from EEG and language models presents a significant opportunity to enhance BCI typing symbol rates.
• Contextual information from LMs can substantially boost the efficiency of BCI communication.
• This integrated approach offers a promising path toward faster and more accurate BCI systems.