Interpretable Supervised Muscle Network Decomposition by Multifactorial ANOVA-ICA.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Motor Control

Background:

• Functional muscular connectivity reveals muscle coordination and neural control during motor tasks.
• Multivariate linear decomposition methods identify fundamental variation modes in muscle networks.
• Existing methods lack explicit disentanglement of experimental factors, limiting interpretation.

Purpose of the Study:

• To introduce a multifactorial supervised decomposition technique (ANOVA-ICA) for enhanced muscle network analysis.
• To enable explicit association of identified modes with task or subject factors.
• To improve the interpretability of muscle network decomposition.

Main Methods:

• Developed a multifactorial supervised decomposition technique combining Analysis of Variance (ANOVA) with Independent Component Analysis (ICA).
• Applied the ANOVA-ICA method to intermuscular coherence networks derived from surface electromyography (sEMG).
• Tested on data from postural control (standing) and running training.

Main Results:

• The ANOVA-ICA framework successfully identified interpretable modes of systematic variation in muscle networks.
• Each mode was explicitly linked to task- or subject-related factors or their combinations.
• Multifactorial ANOVA modeling and ICA significantly improved decomposition interpretability compared to baseline methods.

Conclusions:

• The multifactorial supervised approach (ANOVA-ICA) offers a valid framework for muscle network decomposition.
• This method enhances the interpretability of muscle coordination patterns.
• Potential applications include motor neurophysiology and rehabilitation research.