Electromyogram-based neural network control of transhumeral prostheses

Area of Science:

• Biomedical Engineering
• Neuroprosthetics
• Rehabilitation Robotics

Background:

• Upper-limb amputation, especially at or above the elbow, significantly impairs patient function.
• Current prosthetic control methods often lack the dexterity for intuitive, multi-joint movement.
• Advanced control strategies are needed to restore natural upper-limb functionality.

Purpose of the Study:

• To assess the feasibility of predicting dynamic arm movements using electromyographic (EMG) signals.
• To determine if EMG signals from intact muscles can enable control of prosthetic arms.
• To improve functional outcomes for patients with transhumeral amputations.

Main Methods:

• Recorded movement kinematics and EMG signals from seven muscles during various arm movements.
• Utilized time-delayed artificial neural networks trained offline to predict arm trajectories.
• Analyzed EMG features from different muscle subsets to evaluate predictive effectiveness.

Main Results:

• Successfully predicted elbow flexion/extension and forearm pronation/supination trajectories.
• Achieved average root-mean-square errors of 15.7° for flexion/extension and 24.9° for pronation/supination.
• Obtained R(2) values of approximately 0.81 for flexion/extension and 0.46 for pronation/supination.

Conclusions:

• EMG-based prediction of dynamic arm movements is feasible for transhumeral amputees.
• This approach holds promise for developing intuitive, multi-degree-of-freedom prosthetic control.
• Further integration with wireless telemetry could lead to advanced, fully implanted prosthetic systems.