Multivariate autoregressive modeling of fMRI time series

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Functional magnetic resonance imaging (fMRI) generates time series data reflecting brain activity.
• Understanding functional integration and interregional dependence is crucial for mapping brain networks.
• Existing methods may have limitations in characterizing complex temporal dynamics and feedback loops.

Purpose of the Study:

• To propose and validate multivariate autoregressive (MAR) models for inferring functional integration from fMRI time series.
• To extend linear MAR models to capture nonlinear interactions, modeling top-down modulatory processes.
• To demonstrate the capability of MAR models in characterizing interregional dependence, including feedback loops.

Main Methods:

• Application of multivariate autoregressive (MAR) models to fMRI time series data.
• Extension of linear MAR models to include bilinear terms for nonlinear interactions.
• Utilizing Bayesian methods for model order selection and parameter estimation.
• Demonstration with both synthetic and real fMRI data.

Main Results:

• MAR models effectively characterize interregional dependence in brain activity.
• The extended models can capture nonlinear interactions, important for modulatory processes.
• The approach allows for inference in models with feedback loops within a linear framework.
• Bayesian methods provide a robust framework for model parameter estimation and selection.

Conclusions:

• MAR models offer a powerful framework for analyzing functional integration in human brain fMRI data.
• The ability to model nonlinearities and feedback loops enhances the understanding of brain network dynamics.
• This approach provides a method for precise inference of brain connectivity patterns.