Thresholding functional connectomes by means of mixture modeling.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Network Science

Background:

• Functional connectivity analysis in fMRI is crucial for understanding brain architecture.
• Current methods for estimating functional connectomes from partial correlations face limitations.
• Thresholding techniques like proportional thresholding and permutation testing are common but have drawbacks.

Purpose of the Study:

• To propose a novel data-driven thresholding approach for fMRI network matrices using mixture modeling.
• To enable the creation of subject-specific sparse functional connectomes.
• To offer an alternative to existing thresholding methods for improved accuracy and reproducibility.

Main Methods:

• Developed a mixture modeling approach to segment partial correlations into reliable and unreliable connections.
• Utilized pseudo-False Discovery Rates derived from an empirical null distribution for thresholding.
• Evaluated the method on synthetic fMRI data and real data from the Human Connectome Project.

Main Results:

• The mixture modeling approach demonstrated improved performance over traditional methods for thresholding connectomes.
• Highly reproducible results were obtained when applied to functional connectomes of the visual system.
• The method successfully identified functional decoupling between hemispheres in visual cortex, consistent with known lateralization.

Conclusions:

• Mixture modeling provides a robust and reproducible method for creating sparse functional connectomes from fMRI data.
• This data-driven approach allows for subject-specific analysis without requiring cohort-level inference.
• The findings support the utility of mixture modeling for advancing connectome research and understanding brain function.