HIERARCHICAL BRAIN EMBEDDING USING EXPLAINABLE GRAPH LEARNING.

Area of Science:

• Neuroscience
• Machine Learning
• Graph Theory

Background:

• Brain networks are crucial for understanding human behavior and neurological disorders.
• Current deep graph learning models for brain networks often lack interpretability, hindering biological insight.
• Explainable AI is needed to bridge the gap between complex models and biological understanding.

Purpose of the Study:

• To develop a novel, interpretable deep graph learning model for brain network analysis.
• To extract hierarchical patterns from brain networks for enhanced biological insight.
• To predict standardized behavioral scores using functional brain network data.

Main Methods:

• Proposed a new explainable graph learning model: hierarchical brain embedding (HBE).
• HBE leverages network community structure to create interpretable hierarchical representations.
• Applied HBE to functional brain networks from the Human Connectome Project (1,000 subjects) using Independent Component Analysis (ICA).

Main Results:

• HBE demonstrated superior performance compared to state-of-the-art graph learning methods in predicting behavioral measures.
• The model successfully predicted scores for aggressivity, rule-breaking, and other standardized behaviors.
• Identified hierarchical brain network patterns associated with clinical symptoms, providing interpretable insights.

Conclusions:

• Hierarchical Brain Embedding (HBE) offers an interpretable approach to deep learning on brain networks.
• The method advances the prediction of behavioral traits from neuroimaging data.
• HBE facilitates the discovery of biologically meaningful hierarchical structures within brain networks.