Constructing Connectome Atlas by Graph Laplacian Learning.

Area of Science:

• Neuroimaging
• Network Neuroscience
• Computational Psychiatry

Background:

• Whole-brain functional connectivity can be visualized using advanced neuroimaging and network theory.
• Conventional structural atlases are vital for population-based studies.
• A common connectome atlas is needed to interpret brain disorder-related cognition and behaviors.

Purpose of the Study:

• To address the challenge of constructing a connectome atlas from graph-structured network data.
• To develop a novel learning-based graph inference model for connectome atlas construction.

Main Methods:

• Embedding brain networks into Euclidean space using diffusion mapping.
• Developing an iterative learning-based graph inference model.
• Aligning individual brain networks to a common spectral space and learning a consensus graph Laplacian.

Main Results:

• The proposed method successfully constructs a connectome atlas.
• Statistically significant improvements were observed compared to non-learning-based methods.
• The approach was validated on neuroimaging data from neurodegenerative and neuropediatric disorder populations.

Conclusions:

• The novel learning-based graph inference model provides a robust framework for connectome atlas construction.
• This method overcomes limitations of traditional atlas construction for graph-based neuroimaging data.
• The developed connectome atlas facilitates better interpretation of brain connectivity in neurological and psychiatric disorders.