I²RiMA AI model detects mental stress from EEG with 82.78% accuracy

Cross-subject EEG stress detection has long been hindered by subject-dependent and frequency-specific neural patterns. Conventional Riemannian methods model spatial covariance primarily in the time domain, missing critical neural oscillations, while standard temporal tokenization fragments inter-slice coherence. To address this, researchers (Cheng He et al.) propose I²RiMA, which constructs spatial covariance matrices independently at each frequency point and maps them to the SPD tangent space, preserving channel-wise geometry alongside frequency-specific discriminative cues. It further introduces frequency cluster aggregation to select informative spectral components aligned with EEG rhythms, and an intra-inter slice attention module that adaptively integrates local slice-level spectral dynamics with global temporal context.

Tested on three EEG datasets, I²RiMA consistently outperforms five state-of-the-art baselines, achieving up to 82.78% balanced accuracy. Remarkably, it maintains high efficiency with only 1.60 million parameters and 31.95 million FLOPs, making it suitable for real-time and edge deployment. This breakthrough paves the way for practical, non-invasive mental stress monitoring in wearable devices and clinical settings, offering a robust solution to the long-standing challenge of cross-subject generalization in EEG analysis.