Recursive Bayesian Coding for BCIs.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-Computer Interfaces (BCIs) translate brain signals into commands.
• Current BCIs face challenges in accurately inferring user intent from noisy physiological states.
• Decoding accuracy is crucial for effective BCI applications like assistive technology.

Purpose of the Study:

• To develop an advanced framework for inferring task symbols from brain signals in BCIs.
• To improve the accuracy and robustness of BCI decoding, especially in spelling tasks.
• To introduce a novel coding scheme that maximizes mutual information for enhanced BCI performance.

Main Methods:

• A recursive Bayesian decision framework incorporating prior context distributions and accounting for classifier accuracy.
• Maximum Mutual Information (MMI) coding, a generalization of the Information Transfer Rate (ITR).
• Experimental validation using Steady-State Visually Evoked Potentials (SSVEP) in a "Shuffle" Speller task, comparing with traditional methods.

Main Results:

• The proposed recursive coding scheme with MMI coding demonstrated a 33% increase in letter accuracy compared to traditional decision tree methods.
• MMI coding effectively leverages classifier error asymmetry for improved SSVEP-based spelling.
• The new approach showed a slight increase in processing time (13% slower) but significantly boosted accuracy.

Conclusions:

• The developed recursive Bayesian framework and MMI coding offer a significant improvement in BCI accuracy for discrete tasks.
• This method is robust to errors and adaptable to various brain signal types (P300, SSVEP, MI).
• The findings suggest a promising direction for developing more reliable and efficient Brain-Computer Interfaces.