Structure-function coupling in the human connectome: A machine learning approach.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• The relationship between brain structure and function is a fundamental question in neuroscience.
• Current models suggest a modest coupling between the structural connectome and brain function.
• It remains unclear if this modest coupling is inherent or a limitation of existing models.

Purpose of the Study:

• To investigate the extent of coupling between human brain network structure and function.
• To determine if current brain network models underestimate the structure-function relationship.
• To develop advanced computational models for predicting brain function from structural connectomics.

Main Methods:

• Development of a novel deep learning framework to predict brain function from structural connectomes.
• Comparison of deep learning model performance against state-of-the-art biophysical models.
• Assessment of the predictive power of structure-derived brain function on cognitive performance.

Main Results:

• The deep learning framework achieved significantly higher prediction accuracies (group: R=0.9±0.1, individual: R=0.55±0.1) compared to existing biophysical models.
• Predicted brain function from structural connectomes explained substantial inter-individual variation in cognitive performance.
• Demonstrated a tighter coupling between human brain structure and function than previously reported.

Conclusions:

• The structure-function coupling in human brain networks is substantially stronger than previously suggested.
• Deep learning offers a powerful approach to overcome limitations of current brain network models.
• This work advances our understanding of brain network organization and its relation to cognition.