Empowering Movement: A Human-in-the-Loop Proportional Controller for Assisted Elbow Flexion in Brachial Plexus Injury

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Neuroprosthetics

Background:

• Powered orthotic devices offer potential but face adoption barriers due to simple control systems.
• Current controllers (e.g., bang-bang) limit nuanced interaction with the user.
• Developing proportional controllers is challenging due to patient-specific injury and recovery variations.

Purpose of the Study:

• To evaluate the feasibility of a proportional controller for a powered myoelectric elbow orthosis (PMEO).
• To investigate the use of neural network regression (NNR) for real-time EMG-to-torque control.
• To assess the potential for field deployment of NNR-based control in clinical settings.

Main Methods:

• Collected elbow flexion torque and EMG data from 31 brachial plexus injury (BPI) participants.
• Utilized a 2-layer neural network regression model with Tanh activation and cross-validation.
• Trained NNR models on participant-specific EMG features (e.g., RMS, variance) to predict elbow torque.

Main Results:

• NNR models trained on standard computers within clinically relevant times (median 82 seconds).
• Low prediction latency (median 7 milliseconds) achieved for real-time control.
• Achieved a median test Root Mean Squared Error (RMSE) of 0.39 Nm, with some models showing R2 > 0.7.

Conclusions:

• NNR is a viable method for creating proportional controllers for PMEOs.
• The training process is efficient and can be performed in a clinical setting.
• This approach enhances the human-machine connection for individuals with BPI, paving the way for field-deployable solutions.