Comparative survey of multigraph integration methods for holistic brain connectivity mapping.

Area of Science:

• Network Neuroscience
• Computational Neuroscience
• Neuroimaging Analysis

Background:

• Mapping heterogeneous brain networks is a key challenge in network neuroscience.
• A connectional brain template (CBT) serves as a population-specific fingerprint, integrating diverse network data.
• Existing methods for creating CBTs from single-view or multi-view brain networks require further evaluation.

Purpose of the Study:

• To review state-of-the-art methods for estimating connectional brain templates (CBTs).
• To introduce evaluation metrics for comparing CBT representativeness and topological preservation.
• To assess the performance of different single-view and multi-view integration methods for CBT estimation.

Main Methods:

• Review of current methods for learning connectional brain templates (CBTs).
• Introduction of evaluation criteria: centeredness, biomarker-reproducibility, node-level similarity, global-level similarity, and distance-based similarity.
• Comparative analysis of single-view and multi-view network integration methods, including the deep graph normalizer (DGN).

Main Results:

• The deep graph normalizer (DGN) method significantly outperforms other multi-graph and single-view integration methods.
• DGN demonstrates superior performance in centeredness and graph-derived biomarker reproducibility across healthy and disordered datasets.
• DGN effectively preserves topological traits at both local and global graph levels.

Conclusions:

• The deep graph normalizer (DGN) is a highly effective method for constructing representative connectional brain templates (CBTs).
• DGN excels in capturing essential network properties, including centeredness, reproducibility, and topological structure.
• This work provides a framework for evaluating and advancing CBT estimation in network neuroscience.