Decoding Visual Perception from EEG Using Explainable Graph Neural Network

Brain decoding is an emerging area in the fields of neuroscience and machine learning. The goal of decoding is to utilize measured brain activity to understand the thoughts or sensations of individuals. In the fields of computer vision and machine learning, Graph Neural Networks (GNNs) have demonstrated considerable success. Furthermore, the integration of attention mechanisms in these networks provides a pathway for improved model explainability. This study employs GNNs in the analysis of electroencephalography (EEG), aiming to explore how our brain handles visual information and uncover functional brain networks. We utilize GNNExplainer, a tool designed for GNN interpretation, to pinpoint critical EEG channels and their interconnections relevant to visual EEG tasks. Our findings, which align with existing neuroscience literature, underscore the significance of specific channels. This implies that GNNs have the capability to provide valuable insights within the field of neuroscience research.