A Measurement of Motor Recovery for Motor Imagery-based BCI using EEG Coherence Analysis

Motor imagery-based BCI (MI-BCI) technology possesses the potential to be a post-stroke rehabilitation tool. To ensure the optimal use of the MI-BCI technology for stroke rehabilitation, the ability to measure the motor recovery patterns is important. In this study, the relationship between the EEG recorded during, and the changes in the recovery patterns before and after MI-BCI rehabilitation is investigated. Nine stroke patients underwent 10 sessions of 1 hour MI-BCI rehabilitation with robotic feedback for 2 weeks, 5 times a week. The coherence index (0 <= CI <= 1), which is an EEG metric comparing the coherences of the EEG in the ipsilesioned hemisphere with that in the contralesioned hemisphere, was computed for each session for the first week. Pre- and post-rehabilitation motor functions were measured with the Fugl-Meyer assessment (FMA). The number of sessions with CI greater than a unique subject-dependent baseline value. correlated with the change in the FMA scores (R = 0.712, p = 0.031). Subsequently, a leave-one-out approach resulted in a prediction mean squared error (MSE) of 15.1 using the established relationship. This result is better compared to using the initial FMA score as a predictor, which gave a MSE value of 18.6. This suggests that CI computed from EEG may have a prognostic value for measuring the motor recovery for MI-BCI.