Multi-dynamic modelling reveals strongly time-varying resting fMRI correlations.

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Brain network activity underlies cognition and behavior.
• Time-varying functional connectivity (FC) in resting fMRI predicts traits and conditions.
• Current FC methods conflate changes in mean activity and FC, potentially overestimating FC stability.

Purpose of the Study:

• To develop a novel method that separately models changes in mean brain activity and functional connectivity.
• To address the limitations of sliding window approaches in estimating time-varying FC.
• To improve the accuracy and interpretability of dynamic functional connectivity analysis.

Main Methods:

• Proposed the Multi-dynamic Adversarial Generator Encoder (MAGE) model.
• MAGE models network dynamics with long-range time dependencies.
• Estimated MAGE using Generative Adversarial Networks on resting-state fMRI data.

Main Results:

• Separating mean activity from FC changes revealed significantly stronger temporal fluctuations in FC.
• The MAGE model demonstrated improved estimation of time-varying FC compared to traditional methods.
• The refined FC measures were more predictive of individual behavioral variability.

Conclusions:

• The proposed method accurately disentangles mean activity and FC modulations.
• This approach provides a more sensitive and accurate measure of dynamic functional connectivity.
• Improved dynamic FC analysis enhances understanding of individual differences in behavior and brain function.