Delay-Aligned Network Boosts Decoding in Hybrid Brain–Computer Interfaces

A newly developed framework known as the Neurovascular Coupling-Aware Delay Alignment and Subspace Disentanglement Network (NVC-DSDNet) improves multimodal neural decoding in hybrid brain–computer interfaces while maintaining neurophysiological interpretability. Hybrid brain–computer interfaces unite the rapid electrophysiological activity recorded through electroencephalography (EEG) with hemodynamic responses captured by functional near-infrared spectroscopy (fNIRS). However, effective fusion across these two modalities has faced significant obstacles due to neurovascular latency differences and entangled multimodal feature spaces.

To resolve these integration challenges, NVC-DSDNet couples delay-aware attention mechanisms with shared and private neural subspaces. Within this framework, EEG responses operate as temporal anchors that dynamically retrieve task-relevant fNIRS features across candidate delay intervals. Concurrently, a Neural Feature Subspace Decomposition process separates shared neural intent from modality-specific information, isolating cross-modal features during representation learning.

The network incorporates an adaptive multi-head decision module that calculates sample-level weights for unimodal and fused predictions. When evaluated under subject-independent five-fold cross-validation, NVC-DSDNet achieved decoding accuracies of 72.67 percent on motor imagery tasks, 82.79 percent on mental arithmetic tasks, and 76.17 percent on word generation tasks.

The model’s learned delay weights exhibited distinct temporal dynamics corresponding to hemodynamic response profiles across cognitive conditions. These learned delay weights peaked at 4.5 to 5.5 seconds during motor imagery tasks and shifted to between 6.0 and 9.0 seconds for mental arithmetic tasks, matching task-dependent hemodynamic time courses. The findings demonstrate that coupling-aware temporal alignment and feature subspace disentanglement effectively enhance multimodal decoding performance while preserving interpretability grounded in neurovascular physiology.

Subject of Research: Multimodal neural decoding in hybrid EEG–fNIRS brain–computer interfaces

Article Title: NVC-DSDNet: neurovascular coupling-aware delay alignment and subspace disentanglement network for hybrid EEG–fNIRS brain–computer interfaces

Article References: Wang, S., Gao, Y., Zhang, Z., Shao, X., & Qin, Y. (2026). NVC-DSDNet: neurovascular coupling-aware delay alignment and subspace disentanglement network for hybrid EEG–fNIRS brain–computer interfaces. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02580-1

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02580-1

Keywords: brain–computer interfaces, EEG, fNIRS, neurovascular coupling, delay alignment, subspace disentanglement, multimodal decoding, neural decoding, NVC-DSDNet, neurovascular, coupling-aware, delay