Comparing hemodynamic models with DCM.

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• The Buxton model provides a biophysically plausible framework for blood oxygen level-dependent (BOLD) responses.
• This model is crucial for hemodynamic forward models in dynamic causal modeling (DCM) of fMRI data.
• Previous work by Obata et al. linearized the BOLD signal equation and proposed revised model coefficients.

Purpose of the Study:

• To determine the best formulation of the BOLD signal equation for empirically measured BOLD responses.
• To compare classical and revised hemodynamic models within a dynamic causal modeling (DCM) framework.
• To investigate the impact of different model variants on the analysis of functional interactions.

Main Methods:

• Developed a generalized hemodynamic model encompassing classical and revised Buxton models.
• Embedded variants of the BOLD signal equation into a DCM of visual cortex functional interactions.
• Employed Bayesian model selection to compare eight different hemodynamic models with factorial variations.

Main Results:

• Demonstrated that classical and revised models are special cases of the generalized model.
• Identified a non-linear model with revised coefficients and a free parameter (epsilon) as superior.
• Found that treating epsilon as a free parameter improved model fit for fMRI data.

Conclusions:

• The best model for empirical BOLD responses is non-linear with revised coefficients and a free parameter.
• This generalized framework allows for more accurate modeling of hemodynamic responses in fMRI.
• Findings advance the understanding of effective connectivity analysis using DCM.