Enhancing detection of steady-state visual evoked potentials using individual training data

Although the performance of steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) has improved gradually in the past decades, it still does not meet the requirement of a high communication speed in many applications. A major challenge is the interference of spontaneous background EEG activities in discriminating SSVEPs. An SSVEP BCI using frequency coding typically does not have a calibration procedure since the frequency of SSVEPs can be recognized by power spectrum density analysis (PSDA). However, the detection rate can be deteriorated by the spontaneous EEG activities within the same frequency range because phase information of SSVEPs is ignored in frequency detection. To address this problem, this study proposed to incorporate individual SSVEP training data into canonical correlation analysis (CCA) to improve the frequency detection of SSVEPs. An eight-class SSVEP dataset recorded from 10 subjects in a simulated online BCI experiment was used for performance evaluation. Compared to the standard CCA method, the proposed method obtained significantly improved detection accuracy (95.2% vs. 88.4%, p<;0.05) and information transfer rates (ITR) (104.6 bits/min vs. 89.1 bits/min, p<;0.05). The results suggest that the employment of individual SSVEP training data can significantly improve the detection rate and thereby facilitate the implementation of a high-speed BCI.