Hand Force Estimation from Acoustic Myography Using Deep Wavelet Scattering Transform and Long Short-Term Memory.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• Surface electromyogram (sEMG) signals are crucial for powered prosthetics but face processing challenges.
• Alternative control signals are needed for widespread clinical implementation of upper limb prostheses.

Purpose of the Study:

• To investigate Acoustic Myography (AMG) as a novel control signal for upper limb prosthetics.
• To develop and validate a method for estimating muscle force from AMG signals.

Main Methods:

• Acquired AMG signals using high-sensitivity array microphones and custom housing.
• Applied Wavelet Scattering Transform (WST) for feature extraction from AMG signals.
• Utilized a Long Short-Term Memory (LSTM) neural network to predict force from AMG features.

Main Results:

• The WST-LSTM model achieved an average Normalized Root Mean Square Error (NRMSE) of approximately 8%.
• The model demonstrated robustness across varying window sizes and testing schemes.
• AMG signals were reliably correlated with measured force changes using a hand dynamometer.

Conclusions:

• Acoustic Myography (AMG) signals can be effectively used to estimate muscle force levels.
• The WST-LSTM model offers a robust and accurate method for prosthetic control.
• This research paves the way for more natural and accurate human-machine interfaces in prosthetics.