Predictive connectome subnetwork extraction with anatomical and connectivity priors.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• The human connectome, representing brain network structure and function, is crucial for understanding neurological conditions.
• Identifying specific subnetworks predictive of clinical variables remains a challenge.

Purpose of the Study:

• To develop a novel method for identifying anatomically plausible and predictive brain subnetworks.
• To enhance the accuracy of predicting clinical outcomes, developmental trajectories, and disease states using brain network data.

Main Methods:

• A sparse linear regression model was employed on structural (diffusion MRI) and functional (fMRI) brain networks.
• Novel priors, including backbone network, connectivity, and non-negativity constraints, were enforced to ensure anatomical plausibility and sparsity.
• The method was validated on datasets predicting neonatal neurodevelopment and autism spectrum disorder (ASD) from the ABIDE database.

Main Results:

• The proposed method successfully identified subnetworks predictive of cognitive/neuromotor outcomes in preterm neonates and ASD classification.
• Each novel prior individually improved prediction accuracy.
• The combined priors outperformed existing state-of-the-art prediction techniques.

Conclusions:

• The developed method effectively identifies predictive anatomical subnetworks of the human connectome.
• Incorporating anatomical and connectivity priors enhances the robustness and accuracy of brain network analysis for clinical applications.
• The learned subnetworks offer insights into the topological and functional underpinnings of neurodevelopmental and psychiatric conditions.