A Hybrid EEG-fNIRS Brain-Computer Interface Based on Dynamic Functional Connectivity and Long Short-Term Memory

In this study, we combined dynamic functional connectivity (dFC) and long short-term memory (LSTM) to obtain robust high classification accuracy on bimodal motor imagery signals. To characterize the information flow in brain cortical activity, we measured the variation of the functional interaction between different signal channels with dFC in the feature extraction stage. The variation measured by dFC was then decoded by LSTM for classification. In addition, we used the electroencephalograph (EEG) and functional near-infrared spectroscopy (fNIRS) bimodal signals acquired simultaneously to overcome the artifact susceptibility of EEG. The average classification accuracy of 20 subjects was 90.03%, significantly higher than the traditional common spatial pattern (CSP) method. The result demonstrates the effectiveness of our data processing model.