A Bayesian Approach for Estimating Dynamic Functional Network Connectivity in fMRI Data

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectivity

Background:

• Dynamic functional connectivity (DFC) research is growing, but current methods often use arbitrary sliding windows.
• Existing approaches may not fully capture the temporal dynamics of brain interactions during fMRI experiments.

Purpose of the Study:

• To develop a principled Bayesian framework for estimating time-varying brain networks in fMRI.
• To integrate latent cognitive state classification with network estimation for improved DFC analysis.
• To differentiate task-related brain activity from dynamic functional connectivity patterns.

Main Methods:

• A hidden Markov model is employed to classify latent cognitive states over the fMRI time course.
• A Bayesian approach estimates time-varying networks within an integrated framework, leveraging information across the entire experiment.
• A super-graph structure is assumed to relate connectivity states, promoting edge consistency.

Main Results:

• The proposed method successfully decouples task-related activations from dynamic functional connectivity states in simulated fMRI data.
• Analysis of real fMRI sensorimotor task data revealed specific anatomical regions influencing executive control and attention network interactions.
• The Bayesian approach provides a more robust estimation of dynamic functional connectivity compared to traditional window-based methods.

Conclusions:

• The developed Bayesian method offers a principled and integrated framework for analyzing dynamic functional connectivity in fMRI.
• This approach enhances the understanding of how brain network interactions evolve over time and relate to cognitive states.
• The findings highlight the potential of this method for neuroimaging research, particularly in distinguishing task effects from intrinsic brain dynamics.