Is Pearson's correlation coefficient enough for functional connectivity in fMRI?

Area of Science:

• Neuroscience
• Data Analysis
• Brain Imaging

Background:

• Functional connectivity (FC) quantifies brain region interactions using statistical dependencies.
• Pearson's correlation coefficient (Pearson's r) captures linear relationships, potentially missing nonlinear dynamics.
• Multiscale Graph Correlation (MGC) assesses both linear and nonlinear dependencies across multiple scales.

Purpose of the Study:

• To systematically compare Pearson's r and MGC for measuring functional connectivity.
• To evaluate reliability, sensitivity to data quantity, and ability to detect experimental condition changes and brain-behavior associations.
• To explore the utility of MGC in identifying nonlinear interactions and optimal scales in brain networks.

Main Methods:

• Comparison of Pearson's r and MGC on fMRI datasets.
• Assessment of reliability, data quantity sensitivity, and performance in distinct experimental conditions (anesthesia).
• Evaluation of brain-behavior association detection capabilities.

Main Results:

• Pearson's r and MGC showed similar spatial patterns and alignment for within-network FC, with global optimal scales.
• Local optimal scales between networks suggested nonlinear FC dependencies, particularly detected by MGC.
• Pearson's r demonstrated higher overall reliability, while both methods improved with more data.
• MGC revealed state-dependent optimal scale variability under anesthesia, indicating sensitivity to altered brain states.
• MGC required more computational resources and did not outperform Pearson's r in brain-behavior association tasks.

Conclusions:

• Pearson's r is sufficient for standard FC analysis due to its reliability and efficiency.
• MGC provides nuanced insights into nonlinear brain dynamics and state changes, valuable for specific research questions.
• Researchers should weigh MGC's benefits against its computational cost and complexity for FC quantification.