Cross-paradigm data alignment to improve the calibration of asynchronous BCI systems in EEG-based speech imagery

The brain-computer interfaces (BCIs) based on speech imagery with asynchronous (self-paced) paradigms enable users to directly access and manipulate BCIs with more freedom. Compared with the indirect BCIs with traditional synchronous (cue-based) paradigms, the calibration time of asynchronous paradigms was much longer and with the unbalanced number of task states and idle states. This work aimed to improve the calibration of asynchronous BCI systems by applying a data alignment (DA) approach on cue-based and self-paced paradigms. The cue-based paradigm was regarded as the calibration paradigm and the self-paced paradigm was the testing paradigm. The data alignment approach based on the parallel transport mapped their features on the same tangent space. The logistic regression was used as the classifier to classify task states and idle states. The average result with DA was 7.52% higher than that without DA (baseline), which were 78.45% and 70.92%, respectively. Specially, the best classification accuracy was for 91.82% with DA, and the largest improvement in accuracy was 22.92%. These results suggest that it is practical to use a synchronous paradigm as calibration paradigm in asynchronous BCI systems and the data alignment approach has positive impacts on the classification of task states and idle states.