HKD-MGIN: a physics informed graph neural network using heat kernel diffusion for mapping adolescent functional brain...

Area of Science:

• Neuroscience
• Computational Biology
• Developmental Psychology

Background:

• Adolescence is a critical neurodevelopmental period with significant changes in brain function and organization.
• Understanding these dynamic changes is crucial for identifying developmental trajectories and potential disruptions.

Purpose of the Study:

• To develop and validate a novel graph-based learning framework for modeling adolescent brain development.
• To capture local and global functional interactions in the brain across adolescent stages.
• To apply the framework to predictive tasks such as brain age estimation and sex classification.

Main Methods:

• Introduction of the Heat Kernel Diffused Multi-task Graph Isomorphism Network (HKD-MGIN).
• Integration of multi-task functional magnetic resonance imaging (fMRI) data using heat kernel diffusion.
• Application of a graph isomorphism network architecture for representation learning.

Main Results:

• HKD-MGIN achieved strong performance in brain age estimation (RMSE=1.864) and high accuracy in sex classification (ACC=0.802, AUC=0.834).
• The model identified functional circuits associated with cognitive maturation, demonstrating interpretability.
• The framework effectively captures hierarchical and distributed brain connectivity patterns.

Conclusions:

• Heat kernel-based graph learning is effective for modeling adolescent brain dynamics.
• HKD-MGIN offers a promising approach for identifying biomarkers of cognitive development.
• The framework supports individualized tracking of neurodevelopmental trajectories.