Abstract
This research study presents a novel approach for decoding brain activity measured by electroencephalogram (EEG) signals in brain-computer interface (BCI) systems. Traditional methods overlook the topological relationship among electrodes, which is crucial for understanding brain dynamics. To address this, we propose a deep learning framework based on graph convolutional neural networks (GCNs) with an attention mechanism, aiming to enhance the decoding performance of raw EEG signals during motor imagery tasks. Our approach constructs a GCNs-Net by utilizing graph convolutional layers to learn generalized features from EEG data, considering the functional topological relationship of electrodes. Pooling layers are employed for dimensionality reduction, and a fully-connected softmax layer generates final predictions. The evaluation results demonstrate that our approach achieves a high averaged accu-racy of 95% for group levels, showing adaptability and robustness to individual variability. Moreover, the results are reproducible across multiple experiments for cross-validation, highlighting the potential of our method in improving BCI approaches.