Pre-processing

All data processing was completed by MATLAB R2019b, and the BBCI toolbox was also applied for further data processing (Benjamin et al., 2010). The original EEG signals were re-referenced firstly and filtered with a passband of 0.5–50 Hz. When we collect EEG signals, there will be some unavoidable noises, such as electrooculogram (EOG), electromyogram (EMG). These signals will cause interference to our brain electricity. Hence, in data processing, these interference signals need to be eliminated. Then we performed the independent component analysis (ICA)-based artifact rejection to remove the ocular artifacts. This method assumes that the observed signals are composed of clean EEG data and independent noise components.

Down-sampled the EEG signals to 200 Hz and filtered between 8 and 30 Hz. Then normalized by Z-score and segmented into epochs ranging from −3 to 7 s. After that, subtract the mean value of the period ranging from −3 to 0s to correct the baseline. Electrodes attached to the motor cortex (FCC5h, FCC6h, FCC3h, FCC4h, CCP5h, CCP6h, CCP3h, CCP4h) were selected for further analysis.

Raw fNIRS signal is filtered at 0.01–0.1 Hz to remove internal noise generated by physiological activities such as respiration and heartbeat. After that, the filtered data are converted to a concentration of deoxy- and oxy-hemoglobin (HbR and HbO) according to the modified Beer-Lambert law (Kocsis et al., 2006). Then, considering that the hemodynamic response of fNIRS is relatively slow, the HbO and HbR signals are divided into periods of 0–10 s. Finally, we conducted the baseline correction by diminishing the mean value from 0 to 3 s. Similar to EEG, twenty-four channels near the motor cortex were used for further NIRS data processing.