Predicting Neuromuscular Engagement to Improve Gait Training with a Robotic Ankle Exoskeleton

Area of Science:

• Robotics
• Biomedical Engineering
• Neuroscience

Background:

• Robotic rehabilitation efficacy depends on patient neuromuscular engagement.
• Cerebral palsy (CP) affects motor control, necessitating tailored rehabilitation strategies.
• Exoskeleton-assisted gait training requires optimizing muscle activation patterns.

Purpose of the Study:

• To predict ankle plantar flexor muscle recruitment during exoskeleton walking in individuals with CP using machine learning.
• To develop a personalized biofeedback system based on these predictions to enhance user engagement.
• To evaluate the impact of this biofeedback system on muscle recruitment during robotic rehabilitation.

Main Methods:

• Supervised machine learning, specifically multilayer perceptrons (MLPs), were trained using data from exoskeleton sensors.
• Artificial neural networks (ANNs) achieved 85-87% accuracy in predicting muscle recruitment from electromyography (EMG) data.
• Participants underwent gait training with real-time audio-visual biofeedback derived from the online MLP predictions.

Main Results:

• The MLPs accurately predicted neuromuscular recruitment patterns during exoskeleton-assisted walking.
• Biofeedback integration led to a significant increase in plantar flexor muscle recruitment (24 ± 16%) compared to resistance training alone.
• The personalized biofeedback framework demonstrated potential for improving patient engagement and rehabilitation outcomes.

Conclusions:

• Online machine learning models can effectively predict neuromuscular recruitment for robotic rehabilitation.
• Personalized biofeedback systems enhance patient engagement and muscle activation in CP individuals using exoskeletons.
• This approach shows promise for optimizing the effectiveness of robotic rehabilitation interventions.