Related Experiment Videos
NNERVE: neural network extraction of repetitive vectors for electromyography--Part I: Algorithm
M H Hassoun1, C Wang, A R Spitzer
1Department of Electrical and Computer Engineering, Wayne State University, Detroit, MI 48202.
IEEE Transactions on Bio-Medical Engineering
|November 1, 1994
Summary
This study introduces a novel artificial neural network (ANN) for unsupervised electromyogram (EMG) signal decomposition. The ANN effectively identifies motor unit action potential (MUAP) waveforms and refines classification using firing information.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Computational Neuroscience
Background:
- Artificial neural networks (ANNs) offer robust signal processing for complex and noisy data.
- Electromyogram (EMG) signal decomposition is crucial for understanding neuromuscular activity.
- Unsupervised learning is necessary for EMG decomposition due to unknown motor unit action potential (MUAP) morphologies.
Purpose of the Study:
- To present a novel unsupervised approach for automated EMG signal decomposition using an ANN.
- To develop an ANN classifier capable of learning and identifying MUAP waveforms without prior knowledge.
- To refine MUAP classification and derive individual waveform shapes and firing patterns.
Main Methods:
- Utilized a multilayer perceptron neural network with a novel unsupervised training strategy.
- Employed an autoassociative learning task where the ANN learns repetitive MUAP waveform appearances.
- Developed a dynamic retrieval net classifier by feeding ANN outputs back to its input.
- Classified waveforms by comparing discovered feature vectors and refined results using MUAP firing information.
Main Results:
- The ANN successfully learned repetitive MUAP waveform appearances from filtered EMG signals.
- A dynamic retrieval net classifier was established for waveform classification.
- Feature vectors were discovered for each waveform, enabling classification by comparison.
- MUAP firing information was used to enhance the accuracy of the ANN classifier's results.
Conclusions:
- The proposed ANN approach enables unsupervised EMG signal decomposition.
- Individual MUAP waveform shapes and firing tables can be accurately derived.
- This method provides a robust tool for analyzing complex EMG signals.