Bayesian brain edge-based connectivity (BBeC): a Bayesian model for brain edge-based connectivity inference

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Magnetic resonance imaging (MRI) based brain connectivity analysis is vital for understanding neurological mechanisms.
• Edge-based connectivity inference faces the curse of dimensionality with high-dimensional covariance matrices.
• Existing methods struggle with latent topological structures, leading to inaccurate parameter estimation and unstable inference.

Purpose of the Study:

• To develop a robust Bayesian hierarchical model for inferring brain network topology and high-dimensional covariance structures.
• To address the limitations of existing methods in handling high-dimensional data and unknown latent structures.
• To provide an efficient and reliable tool for large-scale neuroimaging studies.

Main Methods:

• Proposed a Bayesian hierarchical model utilizing a finite-dimensional Dirichlet distribution to model latent network topology.
• Reformulated covariance matrix structure for guaranteed positive definiteness.
• Employed a Metropolis-Hastings algorithm for simultaneous inference of network topology and correlation parameters, with optimized likelihood function calculation for reduced time complexity.

Main Results:

• Simulations demonstrated accurate recovery of network topology and correlation parameters.
• The proposed model outperformed Graphical Lasso and Dirichlet process-based models in estimation accuracy and convergence stability.
• Application to the Alzheimer's Disease Neuroimaging Initiative dataset successfully identified relevant structural subnetworks, consistent with literature findings.

Conclusions:

• Introduced an effective Bayesian framework for inferring brain network topology and high-dimensional covariance structures.
• The model successfully reduces parameter dimensionality and ensures covariance matrix positive definiteness.
• The framework offers an efficient and reliable tool for investigating intrinsic brain connectivity in large-scale neuroimaging studies.