Decoding neural activity preceding balance loss during standing with a lower-limb exoskeleton using an interpretable...

Area of Science:

• Neuroscience
• Robotics
• Biomechanics

Background:

• Falls are a major cause of mortality in older adults.
• Wearable robotic systems aim to restore lower-limb function but require effective fall prevention.
• Early detection of balance loss is crucial for fall prevention in robotic systems.

Purpose of the Study:

• To investigate electroencephalography (EEG) as a predictor of impending falls.
• To understand the brain's response to perturbations for fall avoidance.
• To explore the utility of EEG in conjunction with robotic exoskeletons for fall prevention.

Main Methods:

• Acquired EEG, electromyography (EMG), and center of pressure (COP) data during mechanical perturbations in participants wearing an exoskeleton.
• Utilized a convolutional neural network to predict balance perturbations from single-trial EEG.
• Employed dynamic functional connectivity analysis and GradCAM for model interpretation.
• Developed a gated recurrent unit model for continuous-time decoding of COP trajectories from EEG.

Main Results:

• Perturbation Evoked Potentials (PEP) were detected in EEG 75-134 ms post-perturbation, preceding EMG and COP changes.
• The convolutional neural network achieved a 75.0% F-score in predicting perturbations from EEG.
• Model explanations confirmed reliance on PEP components, not artifacts.
• EEG-based decoding of COP trajectories yielded a Pearson's correlation coefficient of 0.7.

Conclusions:

• EEG signals contain short-latency neural information predictive of impending falls.
• This neural information precedes physical balance recovery responses.
• EEG shows potential for developing brain-machine interfaces for fall prevention in exoskeletons.