An Online Self-Paced Brain-Computer Interface Onset Detection Based on Sound-Production Imagery Applied to Real-Life Scenarios

This research investigated an online onset detection (i.e., ON state detection in asynchronous BCIs) method for BCIs by opening a message when it arrives in two different daily-life task scenarios (watching video and reading text). A new soundproduction related cognitive task (Sound-production imagery, SI) was tested. Blind-source separation with canonical correlation analysis was used for artefact handling. Autoregressive coefficients, band power, common spatial patterns and discrete wavelet transform were used for feature extraction to cover all time, frequency, and spatial time-frequency domain. Linear discriminant analysis was used for classification. The averaged true-positive rate with six subjects was 88.9% in the watching video scenario and 78.9% in the reading text case. The average false-positive rates were 4.2% and 3.9%, respectively. In terms of task response speed, SI task recognition took 4.45s on average for an onset. From these results, the new SI task showed promising results for an online self-paced onset detection system compared to other similar studies.