Supervised brain node and network construction under voxel-level functional imaging.

Area of Science:

• Neuroscience
• Computational Biology
• Brain Imaging Analysis

Background:

• Understanding brain functional organization is key to behavior prediction using brain connectivity.
• Current methods often use a two-step process: defining brain regions then linking connectivity to outcomes.
• Existing unsupervised approaches for node partitioning show limitations in predictive efficiency.

Purpose of the Study:

• To introduce Supervised Brain Parcellation (SBP), a novel brain parcellation scheme.
• To enhance the prediction of behavioral outcomes by directly informing parcellation with the downstream predictive task.
• To improve the efficiency and accuracy of connectome-based predictive modeling.

Main Methods:

• SBP clusters voxels into nodes using voxel-level functional time courses (resting-state or task-based fMRI).
• The clustering maximizes the correlation between inter-node connections and behavioral outcomes.
• The approach ensures intra-node homogeneity while optimizing for predictive power.

Main Results:

• SBP significantly improves out-of-sample connectome-based predictive performance.
• Performance gains were observed across different brain atlases and datasets (ABCD, HCP).
• SBP outperforms conventional step-wise methods in predictive accuracy.

Conclusions:

• Supervised Brain Parcellation offers a more effective approach to mapping brain functional architecture.
• SBP enhances the development of informative network neuromarkers for clinical applications.
• This method advances our understanding of the relationship between brain connectivity and behavior.