Comparing dynamic causal models using AIC, BIC and free energy.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Model selection is crucial in neuroimaging, particularly for brain connectivity analysis.
• Standard practice involves fitting multiple models and comparing them using criteria like AIC and BIC.
• Dynamic Causal Models (DCMs) are increasingly used for inferring brain connectivity.

Purpose of the Study:

• To compare the performance of three model selection criteria: Akaike's Information Criterion (AIC), Bayesian Information Criterion (BIC), and variational Free Energy.
• To evaluate these criteria within the context of General Linear Models (GLMs) and Dynamic Causal Models (DCMs).
• To determine the most effective criterion for comparing DCMs in neuroimaging.

Main Methods:

• A simulation study was conducted to assess the relative merits of AIC, BIC, and Free Energy.
• Model performance was evaluated using both General Linear Models (GLMs) and Dynamic Causal Models (DCMs).
• Statistical comparisons were made based on the simulation outcomes.

Main Results:

• Variational Free Energy exhibited superior model selection ability compared to AIC and BIC.
• Differences in performance were observed between GLMs and DCMs.
• Free Energy proved most effective for the comparison of DCMs.

Conclusions:

• Variational Free Energy is recommended as the preferred model selection criterion for Dynamic Causal Models (DCMs) in neuroimaging.
• The findings support the use of Free Energy for robust brain connectivity analysis.
• This study highlights the importance of choosing appropriate model selection criteria for accurate neuroimaging data interpretation.