Practical Online Brain-computer Interface System Based On Motion-onset Visual Responses

Brain-computer interface (BCI) is a newly developed approach of the multi-modal human-computer interaction. Different from the previous human-computer interaction, BCI could provide a non-muscular output, which allows direct interaction between human brain and the outside environment. Although the proof-of-concept of BCI systems was implemented decades ago, several major challenges are still to be addressed when moving towards a practical online BCI system, such as ease of the system setup, the adaptability of an online system, and the design of the user interface when presented in a real-world application.Motion onset visual evoked potential (mVEP) has been introduced into BCI research recently by our group. Based on our previous studies, this paper intensively studied the properties that could affect the performance of the mVEP-based BCI, such as the refractory effect, the stimulus pattern, and the background contrast. Specifically, based on the robust N200 component in varied contrast, and the corresponding strategy of channel selection, the proposed mVEP-based BCI system could achieve a stable accuracy even in a low contrast.Feature extraction and classification of the EEG are critical factors in the online BCI. In order to classify the unbalanced dataset in the ERP-based BCI, an improvement was made to the current binary support vector machine (SVM). Also, based on the expectation of the decision time and accuracies, two adaptive decision criteria were proposed, which was shown to effectively improve the performance of the online BCI. What is more, an algorithm based on one-class SVM was proposed in this study for asynchronous BCI. By generalizing a previous probability model, our approach provides an promising way to estimate and detect the'idle state'within a statistical framework.Furthermore, the first online BCI system based on mVEP was presented in this study. Using the EEG signal recorded from only a single channel, an average information transfer rate of 42.1 bits/min was achieved among 12 subjects. In addition, an online application for web searching and browsing was developed based on this paradigm. The promising results, that all of the 12 subjects were able to operate the system with high performance, validate the feasibility of a practical mVEP-based BCI, which could be embedded in screen elements and accommodate diverse interfaces, to provide a better human-computer interaction with higher efficiency and more friendly interface.Moreover, by employing the independent component analysis, this study analyzed the cognitive component of selective attention under mVEP setting. The preliminary results suggest the feasibility of an independent BCI based on mVEP.