A graph neural network framework for causal inference in brain networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging

Background:

• Understanding the dynamic functional interactions of the brain on its static structural network is a key challenge.
• The complexity of spatio-temporal dependencies in brain activity hinders a full comprehension of the structure-function relationship.

Purpose of the Study:

• To present a novel graph neural network (GNN) framework for modeling functional brain interactions based on anatomical structure.
• To integrate structural information from diffusion tensor imaging (DTI) with functional data from functional magnetic resonance imaging (fMRI) using GNNs.
• To develop a data-driven, multi-modal approach for measuring causal connectivity strength in brain networks.

Main Methods:

• Utilized a graph neural network (GNN) framework to process graph-structured spatio-temporal brain signals.
• Combined structural connectivity data (DTI) with functional activity data (fMRI).
• Assessed model accuracy by replicating empirical neural activation profiles and compared performance against Vector Auto Regression (VAR) models.

Main Results:

• The GNN framework successfully replicated observed neural activation patterns.
• GNNs demonstrated the ability to capture long-term dependencies and scale to large-scale brain networks.
• Learned GNN features generalized across different MRI scanner types and acquisition protocols, with pre-training improving performance on smaller datasets.

Conclusions:

• The proposed multi-modal GNN framework offers a novel perspective on the brain's structure-function relationship.
• This data-driven approach shows promise for characterizing information flow within brain networks.
• GNNs provide a powerful tool for analyzing complex spatio-temporal dynamics in neuroscience research.