Calibrating EEG-based motor imagery brain-computer interface from passive movement

Abstract

EEG data from performing motor imagery are usually collected to calibrate a subject-specific model for classifying the EEG data during the evaluation phase of motor imagery Brain-Computer Interface (BCI). However, there is no direct objective measure to determine if a subject is performing motor imagery correctly for proper calibration. Studies have shown that passive movement, which is directly observable, induces Event-Related Synchronization patterns that are similar to those induced from motor imagery. Hence, this paper investigates the feasibility of calibrating EEG-based motor imagery BCI from passive movement. EEG data of 12 healthy subjects were collected during motor imagery and passive movement of the hand by a haptic knob robot. The calibration models using the Filter Bank Common Spatial Pattern algorithm on the EEG data from motor imagery were compared against using the EEG data from passive movement. The performances were compared based on the 10×10-fold cross-validation accuracies of the calibration data, and off-line session-to-session transfer kappa values to other sessions of motor imagery performed on another day. The results showed that the calibration performed using passive movement yielded higher model accuracy and off-line session-to-session transfer (73.6% and 0.354) than the calibration performed using motor imagery (71.3% and 0.311), and no significant differences were observed between the two groups (p=0.20, 0.23). Hence, this study shows that it is feasible to calibrate EEG-based motor imagery BCI from passive movement.

Enhanced Motor Imagery Decoding by Calibration Model-Assisted With Tactile ERD

“…Various strategies have been widely studied that can induce similar oscillatory dynamics to MI. These studies include active movement [31], passive movement calibraiton [32], [33], [34], and functional electrical stimulation calibration [35]. However, calibration performance improvement is limited in these studies.…”

Adaptation of motor imagery EEG classification model based on tensor decomposition

“…This poses a grand challenge especially to online brain signal detection and classification such as in BCIs [6][7][8][9]; even if a BCI performs well in calibration, it may suffer from a considerable performance drop over time. For example in [10], some BCI subjects who achieved classification accuracies around 85% in the calibration session obtained only accuracies around 65% in later sessions. Therefore, nonstationarity and its consequences must be well addressed before BCI can be applied in real-world applications out of the laboratory [11][12][13].…”