Generative topographic mapping applied to clustering and visualization of motor unit action potentials

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Accurate identification and visualization of motor unit action potentials (MUAPs) are crucial for understanding neuromuscular system control.
• Existing clustering methods may not fully capture the complexity of MUAP data in high-dimensional spaces.

Purpose of the Study:

• To introduce Generative Topographic Mapping (GTM) as a novel machine learning tool for clustering and visualizing MUAPs.
• To compare the performance of GTM against other clustering techniques like SOM, GMM, and NGN.
• To develop and evaluate a GTM-based visualization tool (GTM grid) for high-dimensional MUAP data.

Main Methods:

• Application of Generative Topographic Mapping (GTM) for MUAP clustering and visualization.
• Comparative analysis of GTM with Self-Organizing Map (SOM), Gaussian Mixture Model (GMM), and Neural-Gas Network (NGN).
• Development of a GTM grid for visualizing MUAPs in high-dimensional space, compared against Principal Component Analysis (PCA).

Main Results:

• GTM and SOM demonstrated superior performance in clustering MUAPs compared to GMM and NGN.
• GTM was identified as a principled and effective alternative to SOM for MUAP analysis.
• The GTM grid provided effective visualization of high-dimensional MUAP data.

Conclusions:

• GTM is a powerful machine learning technique for the clustering and visualization of MUAPs.
• GTM offers a robust and effective alternative to existing methods, enhancing the study of neuromuscular control.
• The GTM grid visualization tool aids in interpreting complex, high-dimensional MUAP datasets.