Regression models for partially localized fMRI connectivity analyses.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectivity Analysis

Background:

• Resting-state fMRI connectivity analysis often assumes consistent brain region alignment across subjects, which is frequently not the case.
• Traditional methods like one-edge-at-a-time analyses or decomposition methods rely on this potentially flawed assumption.

Purpose of the Study:

• To develop and validate a subject-level regression model for explaining intra-subject variability in brain functional connectivity.
• To offer a more parsimonious and robust alternative to existing connectivity analysis methods by characterizing variation explained by different covariates.

Main Methods:

• Utilized subject-level regression models with covariates including geographic distance, homotopy, and functional network membership.
• Included region-specific covariates to account for consistent connectivity differences.
• Applied the model to Human Connectome Project data with 268 regions of interest (ROIs) across eight functional networks.

Main Results:

• Region covariates and network membership explained a high proportion of connectivity variation.
• Geographic distance and homotopy remained important predictors after accounting for the number of covariates.
• The connectivity regression model achieved high data repeatability, comparable to methods using full location information.

Conclusions:

• The developed connectivity regression model effectively explains intra-subject variability in functional connectivity.
• The findings suggest that fMRI connectivity analysis may be feasible in subject-space with reduced registration requirements.
• This approach holds promise for more flexible and potentially less computationally intensive neuroimaging analyses.