Learning a Hand Model From Dynamic Movements Using High-Density EMG and Convolutional Neural Networks.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Surface electromyography (sEMG) measures muscle electrical activity, reflecting motor commands.
• Decoding sEMG into intended movements is crucial for advanced prosthetics and human-computer interfaces.

Purpose of the Study:

• To develop and validate a deep learning method for decoding forearm sEMG into detailed human hand kinematics and kinetics.
• To investigate the neural encoding of hand movements within the sEMG signal.

Main Methods:

• Recorded hand kinematics/kinetics across 22 degrees of freedom during various grasps and digit movements.
• Utilized 320 non-invasive sEMG sensors on forearm muscles as input to a deep learning network.
• Analyzed full-bandwidth, monopolar, unfiltered EMG signals.

Main Results:

• The deep learning network accurately estimated hand kinematics and kinetics, outperforming existing methods.
• Analysis revealed the network maps sEMG activity to hand anatomy at the individual digit level.
• Distinct neural embeddings encoding specific hand movements were identified in full-bandwidth EMG signals, generalizing across participants.

Conclusions:

• The proposed deep learning approach provides a robust and intuitive interface for translating muscle signals into hand movements.
• This technology has the potential to significantly advance the control of assistive hand devices.