Graph alignment exploiting the spatial organization improves the similarity of brain networks

Area of Science:

• Neuroscience
• Computational Biology
• Brain Imaging Analysis

Background:

• Individual brain structure and function are shaped by genetics and environment.
• Brain imaging studies often average data, assuming consistent cortical organization across subjects.
• The assumption of consistent anatomical regions across individuals in brain atlases is questioned.

Purpose of the Study:

• To investigate the assumption of consistent structural properties of brain regions across subjects.
• To analyze the alignment of structural connectomes and identify connectivity misalignments.
• To evaluate the impact of parcellation granularity and initialization methods on cross-subject brain mapping.

Main Methods:

• Utilized a network representation of the brain, with nodes as regions and edges as structural relationships.
• Employed an unsupervised graph matching strategy to align structural connectomes of healthy subjects.
• Compared alignment results using four initialization methods (Spatial Adjacency, Identity, Barycenter, Random) and varying parcellation granularities.

Main Results:

• Applying an alignment strategy significantly improves cross-subject similarity for parcellations with over 100 parcels.
• Spatial Adjacency and Identity initializations, considered plausible priors, yield better alignment results.
• Permutations primarily occur between neighboring parcels, with no clear large-scale spatial patterns observed across the cortex.

Conclusions:

• Brain region correspondence across subjects is not perfectly maintained, necessitating alignment strategies for accurate analysis.
• Structural connectome alignment is crucial for studies relying on consistent anatomical parcellations, particularly with high-resolution atlases.
• Neighboring regions exhibit the most significant structural misalignments, suggesting localized variations in brain organization.