Feasibility of EMG-based neural network controller for an upper extremity neuroprosthesis

Area of Science:

• Biomedical Engineering
• Neuroprosthetics
• Rehabilitation Technology

Background:

• Spinal cord injury (SCI) at the C5/C6 level results in significant loss of upper extremity function.
• Functional electrical stimulation (FES) offers a potential solution to restore movement by activating paralyzed muscles.
• Existing neuroprostheses primarily focus on hand function, leaving shoulder and elbow control unaddressed.

Purpose of the Study:

• To design an artificial neural network (ANN) controller for FES-assisted shoulder and elbow movement in C5/C6 SCI.
• To predict the required stimulation levels for paralyzed muscles based on voluntary muscle activity.
• To enhance functional outcomes beyond current hand neuroprostheses.

Main Methods:

• Developed an ANN controller trained on simulated musculoskeletal arm models reflecting C5 SCI and FES.
• Used kinematic data from able-bodied subjects to drive inverse dynamic simulations.
• Identified a reduced set of voluntary muscles as input for the ANN to predict paralyzed muscle activations.
• Validated the controller's accuracy using Root Mean Square (RMS) error.

Main Results:

• The ANN controller successfully predicted FES-required muscle activations for paralyzed upper extremity muscles.
• The controller achieved a prediction error of less than 3.6% RMS.
• The system effectively translated voluntary muscle signals into commands for FES.

Conclusions:

• An ANN-based controller can accurately predict FES stimulation for restoring shoulder and elbow function in C5/C6 SCI.
• This approach holds promise for significantly improving upper limb functionality in individuals with SCI.
• The developed controller represents a step towards more comprehensive neuroprosthetic solutions for SCI.