A regression framework for brain network distance metrics.

Area of Science:

• Neuroimaging and statistical analysis of brain networks.
• Development of advanced statistical methodologies for neuroscientific research.

Background:

• Analyzing brain networks is crucial in neuroimaging, but robust statistical methods for group comparisons and trait associations are lacking.
• Previous work established a permutation testing framework for two-group comparisons.

Purpose of the Study:

• To advance statistical methods for assessing brain network differences in relation to continuous phenotypes.
• To develop a framework that controls for confounding variables in neuroimaging studies.
• To explore similarity metrics and statistical inference methods for brain network comparison.

Main Methods:

• Proposed an innovative regression framework to link brain network distances/similarities with continuous and categorical covariates.
• Adapted and evaluated standard statistical methods including F-tests (with and without ILE), FGLS, and permutation testing.
• Compared the proposed framework with existing multivariate distance matrix regression (MDMR) methods using simulation studies.

Main Results:

• Simulation studies assessed the performance of various estimation and inference approaches within the new framework.
• The framework demonstrated utility in analyzing the relationship between fluid intelligence and brain network distances.
• Identified effective similarity metrics and statistical inference methods for brain network analysis.

Conclusions:

• The developed regression framework provides a powerful tool for relating brain network properties to continuous phenotypes while accounting for confounders.
• This advancement offers improved statistical rigor for neuroimaging research, enabling deeper insights into brain-behavior relationships.
• The framework's application to Human Connectome Project data highlights its practical relevance and potential for future discoveries.