Empirical Bayes for DCM: A Group Inversion Scheme.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Modeling

Background:

• Estimating conserved functional brain architectures across subjects is crucial for understanding neural mechanisms.
• Integrating multi-subject electrophysiological data presents methodological challenges, particularly with nonlinear models.

Purpose of the Study:

• To present a novel empirical Bayesian scheme for group or hierarchical models within dynamic causal modeling (DCM).
• To address the methodological issue of integrating multi-subject measurements for inferring conserved functional brain architectures.

Main Methods:

• Generalization of random effects analyses to Bayesian inference for nonlinear electrophysiological time-series data.
• An empirical Bayesian scheme using iterative optimization of posterior densities with empirical priors from a second level.
• Leveraging approximate Bayesian inference for hierarchical models to efficiently estimate group effects in DCM.

Main Results:

• The proposed iterative scheme effectively addresses the local minima problem in nonlinear model inversion.
• Empirical priors shrink first-level parameter estimates towards a global maximum, yielding more robust and efficient estimates.
• The method enhances the estimation of both within- and between-subject effects.

Conclusions:

• The developed empirical Bayesian approach provides a robust solution for estimating conserved functional brain architectures from multi-subject data.
• This method improves the reliability and efficiency of dynamic causal modeling in group studies.
• The technique offers a significant advancement for analyzing complex electrophysiological time-series data in neuroscience.