Comparing multilayer brain networks between groups: Introducing graph metrics and recommendations.

Area of Science:

• Neuroimaging
• Network Science
• Computational Neuroscience

Background:

• Neuroimaging studies increasingly use multi-modal data, yet networks from different modalities are often analyzed in isolation.
• Functional brain networks, even within a single modality like magnetoencephalography (MEG), are frequently treated independently, overlooking interdependencies.

Purpose of the Study:

• To explore the utility of a multilayer network framework for integrating diverse neuroimaging data.
• To investigate how graph metrics can quantify multilayer network organization for robust group comparisons.
• To analyze and compare bias correction schemes in multilayer network analysis.

Main Methods:

• Utilized a multilayer network framework to integrate networks from multiple neuroimaging modalities.
• Analyzed biases common in single-layer network analysis and their impact on multilayer networks.
• Compared four bias correction schemes: Minimum Spanning Tree (MST), effective graph resistance, Efficiency Cost Optimisation (ECO), and Singular Value Decomposition (SVD).
• Validated schemes using generative models and empirical magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) data.

Main Results:

• Singular Value Decomposition (SVD) effectively corrected biases when applied to the entire multilayer network.
• ECO, MST, and SVD demonstrated bias correction capabilities when applied to individual network layers.
• All tested schemes showed sensitivity in identifying network topology changes in perturbed generative models and empirical data.

Conclusions:

• Uncorrected multilayer network analysis introduces biases that can compromise the reproducibility of group comparison results.
• The choice and application of bias correction schemes are critical for reliable analysis of integrated neuroimaging network data.
• Recommend implementing bias correction prior to multilayer network analysis for group comparisons in multi-modal neuroimaging research.