Non-parametric Bayesian graph models reveal community structure in resting state fMRI.

Area of Science:

• Neuroimaging
• Network Science
• Computational Neuroscience

Background:

• Resting state functional magnetic resonance imaging (rs-fMRI) analysis increasingly employs network models.
• Node clustering is crucial for interpreting communication patterns within these networks.
• Evaluating different clustering models for rs-fMRI data is essential for robust analysis.

Purpose of the Study:

• To compare three nonparametric Bayesian models for node clustering in complex networks.
• To assess the predictive ability and reproducibility of these models using rs-fMRI data.
• To identify the most effective model for uncovering community structure in brain networks.

Main Methods:

• Investigated three generative Bayesian models: Infinite Relational Model (IRM), Bayesian Community Detection (BCD), and Infinite Diagonal Model (IDM).
• Compared these models against non-Bayesian methods: Infomap, Louvain modularity, and hierarchical clustering.
• Evaluated model performance on three independent rs-fMRI datasets from healthy volunteers.

Main Results:

• The Bayesian Community Detection (BCD) model demonstrated superior performance in predicting unseen data.
• BCD also exhibited the highest reproducibility of clustering results across different datasets.
• The findings suggest that rs-fMRI data possesses a community structure that BCD effectively captures.

Conclusions:

• The Bayesian Community Detection (BCD) model is highly effective for node clustering in rs-fMRI network analysis.
• The results underscore the importance of modeling heterogeneous between-cluster link probabilities for accurate brain network analysis.
• This study validates the utility of Bayesian approaches for understanding functional brain connectivity.