From full calibration to zero training for a code-modulated visual evoked potentials for brain-computer interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) typically require extensive user- and session-specific calibration due to individual differences and electroencephalogram (EEG) signal variability.
• This calibration process involves a time-consuming passive training stage, delaying direct user operation.

Purpose of the Study:

• To develop and validate a calibration-free method for code-modulated visually evoked potential (cVEP)-based BCIs.
• To systematically reduce training data requirements, ultimately eliminating the need for a training stage.

Main Methods:

• A sophisticated encoding model was developed and compared against traditional event-related potential (ERP) techniques.
• The encoding model was calibrated using minimal data (single class or no data) and evaluated offline and online.

Main Results:

• The encoding model significantly reduced training data while maintaining classification performance and explained variance comparable to the ERP method.
• Excellent performance was achieved even with single-class or zero training data.
• The zero-training cVEP BCI demonstrated high communication rates in an online spelling task, confirming its practical feasibility.

Conclusions:

• This work presents the fastest zero-training cVEP BCI to date, achieving high communication speeds with minimal non-invasive EEG electrodes.
• The elimination of the training stage minimizes session time and enables practical "plug-and-play" BCI applications.
• The adopted neural encoding model effectively compresses data into event responses without sacrificing explanatory power.