Effect of initial fMRI data modeling on the connectivity reported between brain areas.

Area of Science:

• Neuroimaging data analysis
• Statistical modeling in neuroscience

Background:

• Neuroimaging analysis commonly employs variance partitioning, often using the General Linear Model (GLM).
• Data-driven methods like Principal Component Analysis (PCA) and Structural Equation Modeling (SEM) for connectivity also rely on variance partitioning.
• Explicitly defining variance partitions is critical for accurate interpretation of neuroimaging data.

Purpose of the Study:

• To highlight the importance of precisely describing variance partitions in neuroimaging data analysis.
• To demonstrate how different subspace selections in variance partitioning affect connectivity inferences.
• To illustrate the quantitative impact of variance partitions on correlation metrics in fMRI studies.

Main Methods:

• Utilizing simulated neuroimaging data with condition and block effects.
• Applying the General Linear Model (GLM) framework for variance partitioning.
• Analyzing functional and effective connectivity using different subspace selections.
• Examining correlations between voxels and error terms based on design matrix partitions.
• Validating findings with a real fMRI study on biological motion.

Main Results:

• Selecting different variance partitions (subspaces) qualitatively and quantitatively alters sample covariances.
• The choice of subspace significantly influences correlations between brain regions.
• The partition of the design matrix affects correlation matrices derived from error terms.
• Demonstrated a clear, quantitative effect of variance partitions on inter-regional correlations in fMRI data.

Conclusions:

• The precise definition and selection of variance partitions are paramount in neuroimaging analysis.
• Inconsistent or unspecified subspace choices can lead to misleading conclusions about brain connectivity.
• Methodological rigor in variance partitioning is essential for reliable functional and effective connectivity findings.