Hybrid Human-Machine Interface for Gait Decoding Through Bayesian Fusion of EEG and EMG Classifiers

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Electroencephalography (EEG) alone is unreliable for controlling walking rehabilitation devices in clinical settings.
• Hybrid human-machine interfaces (hHMIs) combine multiple biosignals to enhance performance.
• Decoding gait activity using hHMIs, particularly with EEG and electromyography (EMG), is an emerging area.

Purpose of the Study:

• To propose and evaluate a hybrid human-machine interface (hHMI) for decoding walking phases using Bayesian fusion of EEG and EMG signals.
• To assess the performance of the hHMI compared to single-signal approaches (EEG or EMG alone).
• To investigate the robustness of the hHMI under conditions of compromised muscle signal reliability.

Main Methods:

• Developed a hybrid human-machine interface (hHMI) integrating EEG and EMG signals.
• Employed Bayesian fusion techniques to combine information from both biosignals.
• Evaluated the classification performance for decoding walking phases of both legs.
• Assessed performance degradation under simulated temporary and permanent EMG alterations.

Main Results:

• The proposed hHMI significantly outperformed single-signal (EEG or EMG) counterparts.
• The hybrid approach maintained high and stable performance (over 80% accuracy) even with permanent EMG degradation.
• Demonstrated smooth performance degradation with temporary EMG alteration (over 75% accuracy at 30% EMG amplitude).

Conclusions:

• Hybrid interfaces combining EEG and EMG show significant potential for reliable gait decoding in rehabilitation.
• The proposed hHMI enhances usability and stability, even with compromised muscle signals.
• Hybrid interfaces may be crucial for wider clinical application of locomotion restoration technologies.