Conditional Generative Models for Simulation of EMG During Naturalistic Movements
IEEE Transactions on Neural Networks and Learning Systems
|August 14, 2024
Summary
This study introduces BioMime, a novel neural network that rapidly simulates electromyography (EMG) signals. This advancement significantly reduces computational costs, enabling dynamic movement analysis in motor neuroscience and human-machine interfaces.
Area of Science:
- Neuroscience
- Computational Biology
- Biomedical Engineering
Background:
- Electromyography (EMG) signal models are crucial for understanding neurophysiology and developing human-machine interfaces.
- Current finite element method (FEM) simulations are highly accurate but computationally expensive, limiting their use to static models.
- There is a need for computationally efficient methods to simulate EMG signals for dynamic movements.
Purpose of the Study:
- To develop a computationally efficient method for simulating EMG signals.
- To overcome the limitations of computationally expensive FEM simulations for dynamic movements.
- To introduce BioMime, a conditional generative model for mimicking advanced numerical EMG models.
Main Methods:
- Developed BioMime, a conditional generative neural network.
- Trained BioMime adversarially to generate motor unit (MU) activation potential waveforms.
- Utilized a smaller set of numerical model outputs for training and predictive interpolation.
Main Results:
- BioMime accurately predicts EMG signal waveforms.
- The model demonstrates high accuracy in interpolating between numerical model outputs.
- Achieved a dramatic reduction in computational load compared to traditional FEM simulations.
Conclusions:
- BioMime offers a computationally efficient solution for EMG signal simulation.
- The model enables rapid simulation of EMG signals during dynamic and naturalistic movements.
- This approach advances motor neuroscience research and the development of human-machine interfaces.
