Related Experiment Videos
Adaptive fuzzy k-NN classifier for EMG signal decomposition
Sarbast Rasheed1, Daniel Stashuk, Mohamed Kamel
1Pattern Analysis and Machine Intelligence Lab, Department of Systems Design Engineering, University of Waterloo, Waterloo, Ont., Canada N2L 3G1. srasheed@engmail.uwaterloo.ca
Medical Engineering & Physics
|January 13, 2006
Summary
An adaptive fuzzy k-nearest neighbour classifier (AFNNC) improves EMG signal decomposition by better classifying motor unit potentials (MUPs). While slightly increasing errors, its performance gains are significant, especially with high MUP shape variability, making it suitable for clinical use.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Machine Learning
Background:
- Electromyography (EMG) signal decomposition is crucial for understanding muscle activity.
- Accurate identification and classification of motor unit potentials (MUPs) are challenging due to signal variability.
- Existing methods like adaptive template matching have limitations in handling MUP instability.
Purpose of the Study:
- To introduce and evaluate an adaptive fuzzy k-nearest neighbour classifier (AFNNC) for improved EMG signal decomposition.
- To compare the AFNNC's performance against an adaptive certainty classifier (ACC) using both synthetic and experimental EMG data.
- To assess the AFNNC's effectiveness in grouping MUPs based on shape and firing patterns.
Main Methods:
- Developed an AFNNC employing an adaptive assertion-based classification approach.
- Utilized a similarity criterion combining MUP shapes and firing pattern information (passive and active modes).
- Evaluated performance on synthetic signals with varying properties and experimental EMG signals, comparing with ACC.
Main Results:
- AFNNC demonstrated superior average classification performance across simulated and experimental EMG signals compared to ACC.
- AFNNC achieved improved correct classification rates (CCr) of 8.1% (independent simulated), 5% (related simulated), and 6% (experimental) over ACC.
- While AFNNC showed slightly increased error rates (Er), these were deemed acceptable for clinical applications, particularly excelling when MUP shape variability was high.
Conclusions:
- The AFNNC offers enhanced accuracy in EMG signal decomposition, outperforming the ACC.
- The k-nearest neighbour assignment paradigm effectively mitigates classification issues arising from MUP instability and shape variability.
- AFNNC shows promise for clinical applications requiring precise EMG signal decomposition.