An integrative Bayesian approach to matrix-based analysis in neuroimaging.

Area of Science:

• Neuroscience
• Brain Imaging Analysis
• Statistical Modeling

Background:

• Understanding brain region correlations is crucial for mapping neural networks.
• Current matrix-based analysis methods, like general linear models (GLMs), often ignore interregional relationships, leading to inefficient statistical inference.

Purpose of the Study:

• To develop a Bayesian multilevel (BML) modeling framework for simultaneous analysis of brain regions, region pairs, and subjects.
• To improve statistical inference efficiency in matrix-based analysis of brain connectivity.

Main Methods:

• Developed a Bayesian multilevel (BML) modeling framework.
• Integrated analyses of all regions, region pairs, and subjects simultaneously.
• Applied the framework to functional magnetic resonance imaging (fMRI) data from a cognitive-emotional task.

Main Results:

• The BML framework quantitatively characterizes intricate relationships across regions and region pairs.
• It resolves multiple testing issues inherent in conventional GLMs.
• It enables inferences on effects typically treated as random and estimates region importance.

Conclusions:

• The developed BML methodology offers a more efficient and comprehensive approach to matrix-based analysis in neuroscience.
• This framework enhances the understanding of brain network organization and interregional relationships.
• The associated program, MBA, is available within the AFNI suite.