Improving brain-computer interface performance with optimized frequency interaction and enhancement techniques:...

This work aims to increase the classification accuracy of motor imagery-based brain-computer interface (MI-BCI) by employing Cross-Frequency Coupling (CFC) and using spontaneous EEG as an input for the features to increase the system's robustness.

Using a benchmark MI-BCI dataset, we examined 25 participants who completed two trials of a motor imagery task split into two classes. Our methodology involved preprocessing EEG data, using Phase-Amplitude Coupling (PAC) to extract CFC characteristics and Particle Swarm Optimization (PSO) to identify the optimal channels. The XGBoost method was utilized to classify the data, and 10-fold cross-validation was employed to verify the results. They are integrated into a single pipeline, named CFC-PSO-XGBoost (CPX).

With an average classification accuracy of 76.7 % ± 1.0 %, with only eight EEG channels, the suggested approach (CPX) outperformed cutting-edge techniques like CSP (60.2 % ± 12.4 %), FBCSP (63.5 % ± 13.5 %), FBCNet (68.8 % ± 14.6 %), and EEGNet. This significant improvement demonstrates the effectiveness of CFC features and PSO for channel selection in MI-BCI classification. Furthermore, the method was evaluated on the public BCI Competition IV-2a dataset, achieving an average multi-class classification accuracy of 78.3 % (95 % CI: 74.85-81.76 %), confirming the scalability and robustness of CPX on external benchmarks.

CPX leveraging spontaneous EEG signals and CFC features significantly improves classification accuracy. We anticipate this methodology will be a robust and practical solution in BCI applications, providing better brain-to-device communication with low-channel utilization and considerable performance metrics.