Comparing dynamic causal models

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Dynamic Causal Models (DCMs) infer effective connectivity from functional magnetic resonance imaging (fMRI) data.
• Model structure assumptions, particularly regarding human brain anatomical connectivity, introduce significant degrees of freedom.
• Evaluating competing hypotheses about neural connections and their modulation is crucial but lacks formal procedures.

Purpose of the Study:

• To demonstrate the application of Bayes factors for comparing Dynamic Causal Models (DCMs).
• To guide model structure selection for both intrinsic connectivity and contextual modulation.
• To provide a formal framework for evaluating competing scientific theories of neural networks.

Main Methods:

• Utilized Bayes factors as a statistical tool for model comparison within the DCM framework.
• Applied Bayes factors to assess intrinsic connectivity patterns and the modulation of connections by experimental factors.
• Integrated Bayes factors with DCM to rigorously compare different hypotheses about neural architecture.

Main Results:

• Bayes factors effectively guide choices in defining DCM structure, addressing intrinsic connectivity.
• The method facilitates the evaluation of how experimental manipulations modulate specific neural connections.
• Demonstrated a formal procedure for comparing competing hypotheses regarding neural network connectivity.

Conclusions:

• The combined use of Bayes factors and DCMs enables robust evaluation of scientific theories on neural network architecture.
• This approach enhances the inferential power of fMRI data analysis for understanding perception and cognition.
• Provides a principled way to select between competing models of effective connectivity in the brain.