LI-DSN: A Layer-wise Interactive Dual-Stream Network for EEG Decoding

TL;DR

LI-DSN introduces a layer-wise interactive dual-stream neural network with novel attention mechanisms for improved EEG decoding across multiple applications, outperforming existing models.

Contribution

It proposes a novel layer-wise interactive dual-stream network with attention mechanisms for EEG decoding, addressing late-fusion limitations and enhancing feature utilization.

Findings

LI-DSN outperforms 13 SOTA models across eight EEG datasets.

The model improves robustness and decoding accuracy in various EEG tasks.

Extensive experiments validate the effectiveness of the proposed approach.

Abstract

Electroencephalography (EEG) provides a non-invasive window into brain activity, offering high temporal resolution crucial for understanding and interacting with neural processes through brain-computer interfaces (BCIs). Current dual-stream neural networks for EEG often process temporal and spatial features independently through parallel branches, delaying their integration until a final, late-stage fusion. This design inherently leads to an "information silo" problem, precluding intermediate cross-stream refinement and hindering spatial-temporal decompositions essential for full feature utilization. We propose LI-DSN, a layer-wise interactive dual-stream network that facilitates progressive, cross-stream communication at each layer, thereby overcoming the limitations of late-fusion paradigms. LI-DSN introduces a novel Temporal-Spatial Integration Attention (TSIA) mechanism, which constructs a Spatial Affinity Correlation Matrix (SACM) to capture inter-electrode spatial structural relationships and a Temporal Channel Aggregation Matrix (TCAM) to integrate cosine-gated temporal dynamics under spatial guidance. Furthermore, we employ an adaptive fusion strategy with learnable channel weights to optimize the integration of dual-stream features. Extensive experiments across eight diverse EEG datasets, encompassing motor imagery (MI) classification, emotion recognition, and steady-state visual evoked potentials (SSVEP), consistently demonstrate that LI-DSN significantly outperforms 13 state-of-the-art (SOTA) baseline models, showcasing its superior robustness and decoding performance. The code will be publicized after acceptance.