Physiological Gaussian process priors for the hemodynamics in fMRI analysis

Area of Science:

• Neuroimaging
• Biophysics
• Statistical modeling

Background:

• Functional magnetic resonance imaging (fMRI) analysis is challenged by the unknown hemodynamic system linking neural activity to the Blood-Oxygen-Level-Dependent (BOLD) signal.
• Accurate inference from fMRI data requires understanding the complex relationship between neural processes and the measured BOLD response.

Purpose of the Study:

• To develop a novel Bayesian model for task-based fMRI that jointly estimates brain activity and hemodynamics.
• To address the limitations of linear time-invariant (LTI) models by incorporating non-linear and time-varying hemodynamic responses.

Main Methods:

• A nonparametric Gaussian process (GP) prior is applied directly to the predicted BOLD response, incorporating physiological information.
• The model allows for flexible, non-linear hemodynamic modeling, moving beyond traditional hemodynamic response function (HRF) approaches.
• Joint estimation of brain activity and hemodynamics is performed within a Bayesian framework.

Main Results:

• The proposed model effectively discriminates between active and non-active voxels in simulated data, even with deviations in the GP prior from true hemodynamics.
• Application to real fMRI data revealed time-varying hemodynamic dynamics.
• The model demonstrated superior detection of simulated activity compared to standard models without increasing false positives.

Conclusions:

• A new non-linear model for task fMRI hemodynamics has been developed.
• This model enhances the detection of active voxels and offers new avenues for investigating hemodynamic processes.
• The proposed Bayesian approach provides a more flexible and accurate method for fMRI data analysis.