Bayesian modeling of dependence in brain connectivity data

Area of Science:

• Neuroscience
• Graph Theory
• Statistical Modeling

Background:

• Brain connectivity studies model brain areas as nodes and connections as edges to find neuropsychiatric phenotype-related patterns.
• Accurate group-level analysis requires modeling the complex dependence structure between multivariate connectivity edges.
• Existing methods struggle with high-dimensional covariance matrices, spatial information, and unknown network topology.

Purpose of the Study:

• To develop a novel Bayesian nonparametric model for analyzing brain connectivity.
• To unify information from brain network nodes, edges, and their covariance.
• To accurately estimate model parameters by incorporating underlying network topology.

Main Methods:

• Developed a Bayesian nonparametric model to construct the covariance matrix function based on network topology.
• Employed an efficient Markov chain Monte Carlo algorithm for parameter estimation.
• Applied the method to resting-state functional magnetic resonance imaging (fMRI) data and simulated data.

Main Results:

• The proposed model effectively unifies information from network structure and connectivity data.
• Demonstrated accurate parameter estimation for complex dependence structures.
• Successfully applied to schizophrenia resting-state fMRI data.

Conclusions:

• The new Bayesian nonparametric model offers a robust approach for analyzing brain connectivity.
• This method enhances the understanding of neuropsychiatric phenotype-related connectivity patterns.
• The approach is validated through application to real and simulated neuroimaging data.