Inter-regional correlation estimators for functional magnetic resonance imaging

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectivity Analysis

Background:

• Functional magnetic resonance imaging (fMRI) functional connectivity is crucial for understanding brain function.
• Current methods often rely on voxel averaging, which can be sensitive to noise and region size.
• A variety of estimators exist, each with different assumptions and robustness properties.

Purpose of the Study:

• To systematically present and study the properties of nine functional connectivity estimators for fMRI data.
• To compare existing and novel estimators regarding spatial structure, noise robustness, and region size dependence.
• To introduce a new estimator with improved theoretical guarantees and empirical performance.

Main Methods:

• Systematic evaluation of nine functional connectivity estimators, including three existing and six novel methods.
• Analysis based on a spatial model of fMRI data, using synthetic, animal, and human datasets.
• Assessment of graph structure, repeatability, reproducibility, discriminability, region size dependence, and noise robustness.

Main Results:

• Common estimators like correlation of averages (ca) exhibit positive bias and significant dependence on region size and intra-correlation.
• A novel 'local correlation of averages' estimator demonstrates significantly lower dependence on region size and bias.
• The choice of estimator influences connectivity patterns, revealing a ventral-dorsal gradient related to region size and intra-correlation.

Conclusions:

• The selection of functional connectivity estimators critically impacts the resulting brain network structure and interpretation.
• The proposed 'local correlation of averages' offers a more robust and less biased alternative for fMRI connectivity analysis.
• Open-source software (R and Python) is provided to facilitate the adoption of these improved estimators.