Network alignment and similarity reveal atlas-based topological differences in structural connectomes

Area of Science:

• Neuroimaging
• Graph Theory
• Computational Neuroscience

Background:

• Brain connectivity is modeled as a graph (connectome) using diffusion MRI tractography.
• Selecting an appropriate brain atlas is crucial for accurate structural connectivity studies.
• Existing methods for atlas selection lack robustness in capturing network topology across subjects.

Purpose of the Study:

• To develop and validate novel methods for assessing brain atlas robustness in structural connectivity analysis.
• To introduce a new graph similarity measure and an alignment technique for connectomes.
• To provide a data-driven strategy for selecting optimal brain atlases for neuroimaging research.

Main Methods:

• Introduced the graph Jaccard index (GJI), a novel set-based similarity measure for graphs.
• Devised WL-align, a connectome alignment technique adapted from the Weisfeiler-Leman graph-isomorphism test.
• Validated GJI and WL-align using data from the Human Connectome Project.

Main Results:

• The graph Jaccard index (GJI) demonstrated superior mathematical properties compared to previous graph similarity measures.
• WL-align effectively aligned connectomes, enabling robust assessment of network topology.
• The study successfully inferred a strategy for selecting suitable brain atlases based on atlas robustness.

Conclusions:

• The GJI and WL-align offer a robust framework for evaluating brain atlas suitability in structural connectivity studies.
• These novel methods enhance the reliability of connectome analysis and atlas selection.
• The findings contribute to advancing neuroimaging research by improving the foundation for structural brain network analysis.