Decoding working memory task condition using magnetoencephalography source level long-range phase coupling patterns.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Neural oscillation phase coupling is vital for brain communication during cognitive tasks like working memory (WM).
• Previous studies often focused on limited brain regions or frequency bands, necessitating data-driven methods.
• Machine learning offers a powerful approach to analyze complex neuroimaging data and identify subtle patterns.

Purpose of the Study:

• To identify cross-subject phase coupling patterns in magnetoencephalography (MEG) data during a working memory (WM) task using machine learning.
• To reveal WM-related neural processes shared across individuals.
• To leverage machine learning for data-driven analysis of neural oscillations in cognitive tasks.

Main Methods:

• Analyzed MEG data from 83 subjects performing N-back WM tasks (2-back vs. 0-back).
• Estimated phase coupling patterns (multivariate phase slope index) across theta, alpha, beta, and gamma bands.
• Trained a linear support vector machine with across-subject cross-validation to classify task conditions, evaluating individual and combined frequency bands.

Main Results:

• Successfully classified WM vs. control conditions using phase coupling in theta (62% accuracy) and alpha (60% accuracy) bands.
• Multiband classification incorporating theta, alpha, and gamma bands improved accuracy to 71%.
• Feature selection identified important phase coupling patterns related to WM processing.

Conclusions:

• Decoding WM tasks using MEG source space functional connectivity is feasible.
• The approach identifies generalizable connectivity patterns shared across individuals.
• Results allow for meaningful interpretation of task-relevant phase coupling.