Cross-comparison of three electromyogram decomposition algorithms assessed with experimental and simulated data

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• The reliability of clinical and scientific information derived from electromyogram (EMG) decomposition relies heavily on the accuracy of automated algorithms.
• Evaluating the performance of these algorithms is essential for their effective application in research and clinical settings.

Purpose of the Study:

• To assess the agreement and accuracy of three publicly available EMG decomposition algorithms: EMGlab, Fuzzy Expert, and Montreal.
• To investigate the relationship between algorithm performance and the Decomposability Index.

Main Methods:

• Utilized experimental EMG data from tibialis anterior and biceps brachii muscles across varying contraction levels (10-50% MVC) and simulated data.
• Assessed algorithm performance using pairwise agreement on experimental data and accuracy against known decompositions for simulated data.
• Included data from 12 subjects for quadrifilar needle EMGs and 10 controls and 10 patients for single-channel needle EMGs.

Main Results:

• For quadrifilar data, median agreement between Montreal and Fuzzy Expert decreased from 95% at 10% MVC to 64% at 50% MVC.
• For single-channel data, median agreements between algorithm pairs were high (∼97% for controls, ∼92% for patients).
• Algorithm accuracy on simulated data generally exceeded agreement metrics on experimental data, correlating strongly with the Decomposability Index.

Conclusions:

• The performance of EMG decomposition algorithms varies and is influenced by factors such as contraction intensity and data characteristics.
• High agreement between algorithms on simulated data indicates high accuracy, suggesting the Decomposability Index is a reliable predictor of performance.
• Further validation and comparison of EMG decomposition algorithms are necessary for robust clinical and scientific applications.