Explainable multi-frequency and multi-region fusion model for affective brain-computer interfaces by Tao Wang instant download

Explainable multi-frequency and multi-region fusion model for affective brain-computer interfaces by Tao Wang instant download

An affective brain-computer interface (aBCI) has demonstrated great potential in the field of emotionrecognition. However, existing aBCI models encounter significant challenges in explainability and the effectivefusion of multi-frequency and multi-region features, which greatly limits their practical applicability. Toaddress these issues, this paper proposes an explainable multi-frequency and multi-region fusion network(MFMR-FN), consisting of multi-frequency encoding and multi-region decoding networks. Specifically, in theencoding network, we leverage spectral graph theory and symmetric positive definite (SPD) matrix learningto adaptively encode EEG data into functional connectivity (FC) matrices with neurobiological information.Furthermore, a multi-frequency fusion algorithm, based on Riemannian geometry is designed to guide thenetwork in achieving cross-frequency feature fusion. In the decoding network, we introduce a multi-regionselection mechanism and a multi-scale Riemannian network, fusing brain network features from the wholebrain, hemispheres, and local regions for coarse-to-fine emotion decoding. We conducted extensive experimentson the emotion recognition dataset (SEED) and the depression detection dataset (MODMA). The results showthat MFMR-FN outperforms existing methods in multiple metrics and provides explainable FC features thatreveal brain network patterns under different emotions and abnormal connectivity in depression. The proposedMFMR-FN is expected to improve the practicality and reliability of aBCIs in real-world and clinical application