Neurophysiological screening of individual variability for robust decoding in c-VEP-based BCI

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Code-modulated visual evoked-potential (c-VEP) brain-computer interfaces (BCIs) offer high data rates and minimal calibration.
• A significant challenge is the performance drop when BCI models are transferred between different users.
• Understanding inter-participant variability is crucial for robust BCI applications.

Purpose of the Study:

• To identify neurophysiological predictors of performance variability in c-VEP BCIs.
• To develop a decoding pipeline that maintains accuracy across users in a burst-c-VEP paradigm.
• To enable rapid and robust deployment of c-VEP BCIs.

Main Methods:

• Analyzed neurophysiological data from 24 participants performing a burst-c-VEP task.
• Identified predictors including inter-epoch correlation, VEP amplitude, and specific brainwave bandpowers (α, θ, δ).
• Compared three preprocessing schemes and three decoders (CNN, Riemannian xDAWN-LDA, GREEN) for intra- and cross-participant accuracy.

Main Results:

• Five neurophysiological predictors were found to correlate with performance differences between high and low performers.
• Subject-specific alignment combined with the GREEN decoder achieved 93% trial-level accuracy.
• This approach eliminated the typical 15-20% cross-participant transfer loss and kept calibration under 1 minute.

Conclusions:

• Rapid user screening using identified neurophysiological predictors can quickly identify suitable candidates.
• A lightweight, user-specific decoding pipeline (Subject-specific alignment + GREEN) ensures robust and accurate c-VEP control.
• This optimized approach facilitates fast deployment and reliable performance of c-VEP BCIs across diverse users.