Riemannian spatio-temporal graph neural network for enhanced cognitive load detection using EEG

Cognitive load detection using EEG signals is crucial for real-time performance monitoring in high-stakes environments, such as aviation, healthcare, and education. Existing methods, however, often fail to effectively capture the complex spatial and temporal dependencies inherent in EEG data. Recent approaches have leveraged graph neural networks (GNNs) for spatial modeling, but many overlook the rich covariance structure of EEG signals and neglect non-local temporal dependencies. To address these challenges, a novel framework, the Riemannian Spatio-Temporal Graph Neural Network (RST-GNN), is proposed for multi-class cognitive load detection based on electroencephalography (EEG). This method integrates Riemannian manifold filtering with temporal graph learning to model both spatial and temporal dynamics jointly. Specifically, EEG signals are filtered using Riemannian spatial filtering to extract discriminative covariance features, which are then represented as nodes in a temporal graph. A Graph-BiMap module is used to extract structured features from these nodes. A temporal attention mechanism is then applied to fuse window-level features, thereby enhancing the model’s ability to capture evolving cognitive states. Experiments on two publicly available EEG datasets demonstrate that the proposed method consistently outperforms existing models across four cognitive load levels, achieving classification accuracies exceeding 96% with low inter-subject variability. These results highlight the robustness of the proposed method and its effectiveness for reliable cognitive load detection.