Network Representation of fMRI Data Using Visibility Graphs: The Impact of Motion and Test-Retest Reliability

Area of Science:

• Neuroscience
• Graph Theory
• Data Mining

Background:

• Visibility graphs offer novel time-series analysis methods.
• Graph theoretical analysis of visibility graphs can yield new features for fMRI data mining.
• However, visibility graph features are underutilized in neuroscience due to concerns about noise robustness and test-retest reliability.

Purpose of the Study:

• To investigate the robustness and test-retest reliability of visibility graph properties in fMRI data.
• To assess the sensitivity of these graph features to motion artifacts.
• To characterize functional connectivity using visibility graph degree synchrony.

Main Methods:

• Analysis of visibility graph properties from fMRI data (N=1010) from the Human Connectome Project.
• Testing sensitivity to motion and evaluating test-retest reliability using Intraclass correlation coefficient (ICC).
• Characterizing connectivity strength via degree synchrony of visibility graphs.

Main Results:

• Strong correlations (r > 0.5) were observed between visibility graph properties (e.g., number of communities, average degrees) and motion.
• High test-retest reliability for average degrees (ICC=0.74), moderate for clustering coefficient (ICC=0.43) and average path length (ICC=0.41).
• Moderate to low correlations (r < 0.35) were found for functional connectivity measured by correlating visibility graph degrees.

Conclusions:

• fMRI motion significantly influences the robustness and reliability of visibility graph features.
• Robust motion correction strategies are essential before applying visibility graph analysis to fMRI data.
• Further research is needed to fully understand the application potential of visibility graph features in fMRI.