Benchmarking methods for mapping functional connectivity in the brain.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• The brain's networked architecture facilitates synchronized neuronal activity.
• Functional connectivity (FC) networks map these communication patterns using neuroimaging.
• Current FC research predominantly employs Pearson's correlation, overlooking alternative statistical measures.

Purpose of the Study:

• To investigate how the choice of pairwise statistic impacts the organization and interpretation of functional connectivity (FC) matrices.
• To benchmark canonical FC network features across a wide array of statistical methods.

Main Methods:

• Utilized a library of 239 pairwise statistics to analyze functional connectivity (FC) matrices.
• Evaluated FC network properties including hub mapping, weight-distance relationships, structure-function coupling, and correspondence with other neurophysiological networks.
• Assessed the capacity of different FC methods for individual fingerprinting and brain-behavior prediction.

Main Results:

• Observed substantial quantitative and qualitative variations in FC network organization depending on the chosen pairwise statistic.
• Identified specific measures, such as covariance, precision, and distance, as demonstrating desirable properties.
• These selected measures showed strong correspondence with structural connectivity and enhanced capabilities for individual differentiation and behavior prediction.

Conclusions:

• The selection of pairwise statistics critically influences the characteristics and interpretability of functional connectivity (FC) networks.
• Measures like covariance, precision, and distance offer advantages for understanding brain organization and individual differences.
• Optimizing FC mapping involves tailoring statistical methods to specific neurophysiological mechanisms and research questions.