EEG_GLT-Net: Optimising EEG graphs for real-time motor imagery signals classification

Brain-Computer Interfaces (BCIs) connect the brain to external control devices, necessitating the accurate translation of brain signals such as from electroencephalography (EEG) into executable commands. EEG MI classification has numerous applications, including neurorehabilitation for stroke patients, control of assistive robotic devices, and advancements in neurofeedback systems. Graph Neural Networks (GCN) have been increasingly applied for classifying EEG Motor Imagery (MI) signals, primarily because they incorporates the spatial relationships among EEG channels, resulting in improved accuracy over traditional convolutional methods. However, existing methods for constructing adjacency matrices, such as Geodesic distances, Pearson Correlation Coefficient (PCC), and others, often rely on predefined inter-channel relationships. These methods not only demand high computational resources during inference but often achieve limited performance accuracy, particularly for single time-point EEG MI classification where rapid interpretation is crucial. To address this, our paper introduces the EEG Graph Lottery Ticket (EEG_GLT) algorithm, an innovative technique for constructing adjacency matrices for EEG channels. This method does not require pre-existing knowledge of inter-channel relationships, and it can be tailored to suit both individual subjects and GCN model architectures. We conducted an empirical study with 20 subjects and six different GCN architectures to compare the performance of our EEG_GLT adjacency matrix against both Geodesic and PCC adjacency matrices on time-resolved EEG MI dataset, PhysioNet dataset. Our EEG_GLT method consistently exceeded performance accuracy benchmarks. Additionally, we compared our model with state-of-the-art models, achieving superior results. EEG_GLT algorithm marks a breakthrough in development of optimal adjacency matrices, effectively boosting both computational accuracy and efficiency, making it well-suited for single time point classification of EEG MI signals that demand intensive computational resources.