Hierarchical Bayesian Inference for Community Detection and Connectivity of Functional Brain Networks.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Network Science

Background:

• Functional magnetic resonance imaging (fMRI) studies often analyze hierarchically organized brain networks.
• Existing methods for estimating network community structure may not adequately account for inter-subject variability.
• Understanding brain network segregation and integration is crucial for cognitive and behavioral research.

Purpose of the Study:

• To develop a new multilayer community detection method for functional brain networks.
• To robustly estimate community structure at both individual and group levels, preserving inter-subject variability.
• To provide a more accurate and reliable alternative to existing modularity-based models.

Main Methods:

• A novel multilayer community detection method based on the Bayesian latent block model (LBM).
• Development of a community structure-based multivariate Gaussian generative model for synthetic data simulation.
• Validation using split-half reproducibility on working memory task fMRI data from the Human Connectome Project.

Main Results:

• The proposed Bayesian LBM method accurately detects community memberships in synthetic data, consistent with predefined node labels.
• The method demonstrates robustness and reliability in analyzing real fMRI data, outperforming traditional modularity models.
• The approach successfully retains individual network variability in both synthetic and real data analyses.

Conclusions:

• The Bayesian LBM offers a superior method for estimating functional brain network community structure.
• This approach enhances the analysis of brain networks by accounting for individual differences.
• The developed method provides a more accurate and reliable tool for neuroimaging research.