MEG Channel Selection Using Copula Entropy-Based Transfer Entropy for Motor Imagery BCI

Chao Tang, Dongyao Jiang, Badong Chen

Abstract: Multi-channel magnetoencephalography (MEG) data provides high spatiotemporal resolution for motor imagery (MI)-based brain-machine interfaces (BCIs). However, not all channels contribute to the performance of BCIs. Taking into account the importance of specific channels in measuring their causal relationships with other channels during MI tasks, a novel channel selection method using copula entropy-based transfer entropy (CTE) is proposed to select task-relevant channels. Experiments on a publicly available dataset validate the effectiveness of the proposed methods. Compared to using all channels, channel selection based on CTE can significantly (p < 0.05) improve single-session classification accuracy and greatly reduce the number of MEG channels. Cross-session classification also outperforms the competing method.