Directed functional connectivity using dynamic graphical models.

Area of Science:

• Neuroimaging
• Network Neuroscience
• Computational Neuroscience

Background:

• Characterizing brain networks requires advanced neuroimaging methods to model spatio-temporal activity patterns.
• Existing methods often struggle with dynamic and directed functional connectivity, particularly in large-scale networks.

Purpose of the Study:

• To introduce and validate dynamic graphical models (DGMs) for estimating dynamic, directed functional connectivity.
• To assess the reliability and accuracy of DGMs in the presence of hemodynamic lag confounds.
• To apply DGMs to human and mouse resting-state fMRI data for network characterization.

Main Methods:

• Developed dynamic graphical models (DGMs), a multivariate graphical model with time-varying coefficients for instantaneous directed relationships.
• Utilized network simulations with systematic hemodynamic lags (0.4-0.8s) to test directionality estimation.
• Applied DGMs to human resting-state fMRI (N=500) and mouse fMRI data.

Main Results:

• DGMs demonstrated 72%-77% sensitivity in detecting true directionality despite lag confounds, with no increase in false positives.
• Identified consistent influence of the default mode network on cerebellar, limbic, and auditory/temporal networks in humans.
• Revealed a reciprocal relationship between visual medial and lateral networks and a plausible hippocampal-cingulate pathway in mice.

Conclusions:

• DGMs provide a valid and reliable method for estimating dynamic, directed functional connectivity in neuroimaging.
• The method is robust to hemodynamic lag confounds and applicable to large-scale brain networks.
• DGMs offer valuable insights into brain network organization and function across species.