New EEG Model Improves Mental Stress Detection Accuracy

Detecting mental stress across different individuals using EEG signals is challenging because stress-related brain patterns vary significantly between people and across different brainwave frequencies. Traditional methods often focus on spatial covariance in the time domain, neglecting crucial neural oscillations, and can fragment temporal coherence. To overcome these limitations, a new model called I2RiMA (Intra-Inter Riemannian Manifold Attention Network) has been proposed. This network constructs spatial covariance matrices for each frequency independently, mapping them to a specific mathematical space (SPD tangent space) to preserve both channel geometry and frequency-specific information. I2RiMA also incorporates frequency cluster aggregation to identify and reduce redundant spectral components, aligning them with known EEG rhythms. Furthermore, an intra-inter slice attention module dynamically integrates local spectral dynamics with global temporal context. Experimental results across three datasets show I2RiMA consistently outperforms five state-of-the-art baselines, achieving higher accuracy with efficient computational resources.