Evaluation of Efficient and Performance-Retaining cVEP-BCI Class Augmentation

Brain-computer interfaces (BCIs) enable direct communication between the brain and external devices. BCI spellers are one of their most significant applications, allowing users to operate a keyboard with their brain. BCIs based on code-modulated visual evoked potentials (cVEPs) offer the potential for class augmentation due to the cyclical nature of their stimuli. This approach allows for the use of less training data while creating additional classes in the spelling interface, without requiring additional data or causing significant performance loss. To definitively address concerns regarding the potential limitations of this technique, we derived four extra classes from a single class in our BCI speller and evaluated their performance. The results demonstrated no performance loss and no evidence of anomalously poor participant performance. This research advocates for class augmentation as a standard practice to reduce calibration time and improve the scalability, accessibility, and overall effectiveness of cVEP-based BCI spellers.