Toward Early Stopping Detection for Non-binary c-VEP-Based BCIs: A Pilot Study

Abstract

Code-modulated visual evoked potentials (c-VEPs) have potential as a reliable and non-invasive control signal for brain-computer interfaces (BCIs). However, these systems need to become more user-friendly. Non-binary codes have been proposed to reduce visual fatigue, but there is still a lack of adaptive methods to shorten trial durations. To address this, we propose a nonparametric early stopping algorithm for the non-binary circular shifting paradigm. The algorithm analyzes the distribution of unattended commands’ correlations and stops stimulation when the most probable correlation is considered an outlier. This proposal was evaluated offline with 15 healthy participants using p-ary maximal length sequences encoded with shades of gray. Results showed that the algorithm could stop stimulation in under two seconds for all sequences, achieving mean accuracies over 95%. The highest performances were achieved by bases p=2 and p=5, attaining 98.3% accuracy with ITRs of 164.8 bpm and 121.7 bpm, respectively. The proposed algorithm reduces required cycles without compromising accuracy for c-VEP-based BCI systems.