NETWORK CLASSIFICATION WITH APPLICATIONS TO BRAIN CONNECTOMICS.

Area of Science:

• Neuroimaging
• Network Science
• Statistical Analysis

Background:

• Analysis of multiple networks, particularly brain networks from functional connectivity data, presents unique challenges.
• Existing methods often neglect network structure or edge weight information, limiting their effectiveness in classification tasks.
• Brain network classification holds potential for distinguishing between various brain disorders.

Purpose of the Study:

• To develop a computationally efficient classification method that leverages both network structure and individual edge weights.
• To create a parsimonious and interpretable representation of brain connectivity differences between classes.
• To address limitations of existing network analysis techniques in neuroimaging applications.

Main Methods:

• Proposed a graph classification method incorporating network structure through node and edge sparsity penalties.
• Utilized edge weights as predictors within a framework that respects the inherent network topology.
• Implemented the method using efficient convex optimization techniques.

Main Results:

• The developed method successfully classifies networks by considering both structure and edge weights.
• Demonstrated the approach's utility in analyzing functional connectivity data from fMRI studies.
• Achieved a parsimonious and interpretable representation of connectivity patterns relevant to schizophrenia.

Conclusions:

• The proposed graph classification method offers an effective way to analyze brain networks.
• This approach enhances the understanding of brain connectivity patterns and their relation to disorders.
• The method provides a valuable tool for neuroimaging research and clinical applications.