From Signal Stacking to Dynamic Coupling: A Critical Review of Wearable EEG–EMG Fusion Brain–Computer Interfaces for Stroke Rehabilitation

Abstract

This structured critical review examines wearable brain–computer interface (BCI) systems that integrate electroencephalographic (EEG) and electromyographic (EMG) signals for post-stroke motor rehabilitation. The central engineering problem is the spatio-temporal heterogeneity between cortical and muscular signals, which limits the reliability and generalizability of conventional EEG–EMG fusion. We review acquisition and synchronization methods, data-, feature-, and decision-level fusion, deep-learning architectures, wearable implementation, and clinically oriented closed-loop rehabilitation. Conventional fusion can exploit complementary information but usually treats the cross-modal relationship as fixed. By contrast, dynamic brain–muscle coupling is defined here as the explicit, time-resolved estimation of interaction strength, delay, directionality, or network topology between cortical regions and target muscles. Measurable candidates include time-resolved corticomuscular coherence, phase locking, lagged dependence, information-theoretic directionality, and dynamic graph connectivity. Coupling-aware and graph-based methods are promising, but clinical translation remains constrained by artifacts, inter-subject and cross-session variability, overfitting, limited clinical datasets, interpretability, synchronization error, and embedded-computing requirements. The review therefore proposes a transparent pathway from static signal stacking toward physiologically grounded, dynamically coupled, and adaptively controlled rehabilitation systems.