A Systematic Review of EEG and EMG Integration Techniques in Brain-Computer Interfaces: Enhancing Grasp Classification Accuracy

Introduction: Brain-computer interfaces (BCIs) using electroencephalography (EEG) and electromyography (EMG) have shown great potential in decoding motor intentions for grasping tasks. Integrating EEG and EMG signals can enhance classification accuracy for controlling neuroprosthetics, especially in grasp-related activities. This systematic review aims to evaluate the current techniques used for EEG-EMG integration in BCIs and their impact on improving grasp classification accuracy.

Methods: A systematic search of relevant studies from 2010 to 2023 was conducted. Studies that incorporated both EEG and EMG signals for grasp intention detection and classification in BCIs were included. Data on signal acquisition, feature extraction, classification methods, and performance metrics (e.g., accuracy) were extracted. Meta-analysis was performed using Python and R to assess the pooled classification accuracy improvements when EEG and EMG signals were combined compared to EEG or EMG alone. Subgroup analyses were conducted based on BCI tasks, feature selection methods, and classifiers used.

Results: A total of 50 studies met the inclusion criteria. Integrating EEG and EMG signals yielded a significant improvement in classification accuracy, with a pooled increase of 15% (95% CI: 12%-18%) compared to EEG-only systems and 10% (95% CI: 8%-13%) over EMG-only systems. The most commonly used classification techniques were support vector machines (SVM) and convolutional neural networks (CNNs). Studies focusing on complex hand grasps showed the greatest accuracy improvement. Hybrid feature extraction methods combining time-domain and frequency-domain features were found to be the most effective.

Conclusion : The integration of EEG and EMG signals in BCIs enhances the accuracy of grasp classification tasks, providing a more reliable control for neuroprosthetic devices. Future research should focus on optimizing real-time implementation and exploring new classifiers to further boost performance in dynamic environments.