Developing RPC-Net: Leveraging High-Density Electromyography and Machine Learning for Improved Hand Position...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Technology

Background:

• Advanced prosthetic control is crucial for restoring function in individuals with limb loss.
• Translating biological signals into intuitive device control remains a significant challenge.

Purpose of the Study:

• To develop and evaluate RPC-Net (Recursive Prosthetic Control Network), a novel neural network for electromyographic (EMG) signal translation.
• To achieve accurate and computationally efficient conversion of EMG activity to hand position.

Main Methods:

• Employed a regression-based approach to map forearm EMG signals to hand kinematics.
• Tested RPC-Net's adaptability across different conditions and compared it to existing academic solutions.
• Investigated the impact of incorporating previous position data and reducing EMG input parameters.

Main Results:

• RPC-Net achieved high accuracy in predicting hand position from EMG data, surpassing other methods at similar computational costs.
• Including historical position data consistently improved prediction accuracy.
• Demonstrated robustness with fewer EMG electrodes and shorter input signals, suggesting potential for reduced computational load.

Conclusions:

• RPC-Net accurately translates forearm EMG activity into hand position, providing a practical and adaptable solution.
• The technology shows potential for clinical accessibility and enabling more natural prosthetic device control.
• This advancement could significantly improve the quality of life for individuals with limb loss.