Neural-Muscular Coupling Analysis for Upper Limb Gesture Classification: A Multimodal EEG-EMG Fusion Framework

This study presents a novel multimodal fusion framework for upper-limb gesture classification by integrating electroencephalography (EEG) and electromyography (EMG) signals to capture the complete neural–muscular dynamics of motor control. Conventional single-modality approaches are limited in their ability to represent the full spectrum of voluntary movement generation, as EEG encodes cortical motor intent while EMG reflects the resulting peripheral muscle activation. To address this limitation, the proposed framework extracts discriminative temporal and spectral features from both modalities and derives cross-modal interaction descriptors that characterize neural–muscular coupling through correlation, phase synchronization, and energy ratio analysis. A comprehensive machine learning pipeline was implemented, encompassing Random Forest, Artificial Neural Network, XGBoost, LightGBM, and an ensemble–based Stacking Classifier, to evaluate classification performance. The framework was validated using data from eleven subjects performing seven distinct hand gestures. Experimental results shows that the EEG–EMG fusion enhances classification accuracy, employing Stacking Classifier achieves$94.7 \% \pm 2.1 \%$, outperforming both EEG–only and EMG-only baselines. These findings confirm that multimodal integration effectively bridges cortical intent and muscular execution, providing a biologically informed and computationally robust approach for gesture recognition. The findings highlight the effectiveness of multimodal integration in enhancing gesture recognition accuracy and contribute to the development of robust brain– computer interfaces and advanced prosthetic control systems. EEG-EMG fusion, gesture classification, brain-computer interface, machine learning, multi-modal signal processing.