A posteriori model validation for the temporal order of directed functional connectivity maps

Area of Science:

• Neuroscience
• Brain Imaging
• Network Analysis

Background:

• A posteriori model validation for temporal order in neural directed functional connectivity is underutilized.
• Many models assume sequential independence of residuals, yet this assumption is rarely tested post-hoc.
• Functional magnetic resonance imaging (fMRI) data analysis often involves complex connectivity mapping.

Purpose of the Study:

• To apply and demonstrate an a posteriori model validation procedure for directed functional connectivity maps derived from fMRI data.
• To test the procedure on simulated, task-related, and resting-state fMRI datasets.
• To assess the impact of unmodeled temporal dependencies on connectivity map accuracy.

Main Methods:

• Directed functional connectivity was analyzed using the unified structural equation modeling (SEM) family.
• White noise tests on one-step-ahead prediction errors were employed for validation.
• Lagrange Multiplier tests were used for decision criteria to revise connectivity maps.
• The procedure was applied to single-subject simulated, single-subject task-related, and multi-subject resting-state fMRI data.

Main Results:

• The validation procedure effectively detected unmodeled sequential dependencies and recovered higher-order simulated connections.
• The method proved capable of handling task-related input in functional connectivity analysis.
• A significant portion (44%) of subjects in resting-state data required second-order connections for white noise residuals, indicating prevalent higher-order lags.

Conclusions:

• Temporal validation is essential for directed functional connectivity analyses to prevent biased parameter estimates due to unmodeled higher-order dependencies.
• Higher-order temporal lags appear to be common in resting-state brain activity.
• The developed validation procedure offers broad methodological relevance for neuroimaging research.