Large-Scale Granger Causality Analysis on Resting-State Functional MRI | JoVE Visualize

Area of Science:

Neuroscience
Network Science
Data Analysis

Background:

Resting-state functional MRI (fMRI) allows non-invasive study of brain function.
Understanding information flow and network structure in the brain is crucial for neuroscience.
Existing multivariate approaches often lack voxel-level causal inference.

Purpose of the Study:

To develop and validate a novel method for measuring information flow between voxel time series in resting-state fMRI.
To recover the underlying functional network structure of the human brain.
To assess the effectiveness of the proposed method by comparing recovered networks with known brain regions and stimulation-based experiments.

Main Methods:

Integration of dimensionality reduction with predictive time series modeling for large-scale Granger Causality (lsGC) analysis.
Quantification of directed information flow at an individual voxel level.
Application of non-metric network clustering (Louvain method) to identify functional networks.
Validation using resting-state fMRI data, comparing recovered motor and visual cortex networks with those from a visuomotor stimulation experiment.

Main Results:

The lsGC method successfully identified directed information flow suggestive of causal influence between voxel time series.
Non-metric network clustering effectively segmented the brain into functionally connected networks.
The recovered motor and visual cortex networks from resting-state fMRI showed strong agreement (Dice Coefficient of 0.59) with networks derived from visuomotor stimulation data.
The study systematically analyzed the impact of dimensionality reduction on network recovery.

Conclusions:

The developed approach accurately measures multivariate causal influence between time series in resting-state fMRI data.
This method is effective for segmenting and identifying functionally connected brain networks without external stimulation.
The findings highlight the potential of lsGC analysis integrated with dimensionality reduction for uncovering brain network dynamics.