Computational NeuroscienceThe predictive role of pre-cue EEG rhythms on MI-based BCI classification performance

- A novel coefficient computed from pre-cue EEG rhythms over different regions of the brain is proposed.
- The feasibility of predicting the performance of motor imagery-based BCI based on the proposed coefficient is examined.
- Significant positive correlation between the proposed coefficient and accuracies is achieved.
- The results suggest that having higher frontal theta and lower posterior alpha prior to performing motor imagery may result in better BCI performance.

BackgroundOne of the main issues in motor imagery-based (MI-based) brain-computer interface (BCI) systems is a large variation in the classification performance of BCI users. However, the exact reason of low performance of some users is still under investigation. Having some prior knowledge about the performance of users may be helpful in understanding possible reasons of performance variations.New methodIn this study a novel coefficient from pre-cue EEG rhythms is proposed. The proposed coefficient is computed from the spectral power of pre-cue EEG data for specific rhythms over different regions of the brain. The feasibility of predicting the classification performance of the MI-based BCI users from the proposed coefficient is investigated.ResultsGroup level analysis on N = 17 healthy subjects showed that there is a significant correlation r = 0.53 (p = 0.02) between the proposed coefficient and the cross-validation accuracies of the subjects in performing MI. The results showed that subjects with higher cross-validation accuracies have yielded significantly higher values of the proposed coefficient and vice versa.Comparison with existing methodsIn comparison with other previous predictors, this coefficient captures spatial information from the brain in addition to spectral information.ConclusionThe result of using the proposed coefficient suggests that having higher frontal theta and lower posterior alpha prior to performing MI may enhance the BCI classification performance. This finding reveals prospect of designing a novel experiment to prepare the user towards improved motor imagery performance.