Structure can predict function in the human brain: a graph neural network deep learning model of functional...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Graph Theory

Background:

• Functional connectivity and graph theory measures are vital in brain research.
• The relationship between functional and structural brain connectivity is not fully understood.
• Graph neural networks (GNNs) offer a novel approach for analyzing network data.

Purpose of the Study:

• To apply GNN deep learning to predict functional connectivity from structural connectivity.
• To investigate the predictive power of GNNs for both mean and individual-level functional connectivity and centrality.
• To establish a new benchmark for predicting functional connectivity from structural data.

Main Methods:

• Utilized a GNN deep learning model.
• Applied the model to structural connectivity data from 998 Human Connectome Project participants.
• Analyzed the prediction accuracy for mean and individual-level functional connectivity and centrality.

Main Results:

• GNNs explained 89% of the variance in mean functional connectivity.
• GNNs explained 56% of the variance in individual-level functional connectivity.
• GNNs explained 99% of the variance in mean functional centrality and 81% in individual-level functional centrality.
• Functional centrality is robustly predictable from structural connectivity.

Conclusions:

• GNNs provide a powerful tool for modeling brain connectivity.
• Structural connectivity can robustly predict functional centrality.
• This method sets a new benchmark and may enable functional connectivity prediction in non-fMRI-capable individuals.