Contrastive voxel clustering for multiscale modeling of brain network

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Resting-state functional MRI (fMRI) analyzes brain connectivity but often simplifies data by averaging signals within brain regions.
• This averaging overlooks the heterogeneity of signals within each Region of Interest (ROI), limiting comprehensive brain functionality analysis.

Purpose of the Study:

• To introduce a novel multi-stage self-supervised learning framework for multiscale brain network analysis.
• To delineate brain functionality from the voxel level up to the sample level, capturing finer details than traditional methods.

Main Methods:

• A Contrastive Voxel Clustering (CVC) module was developed to learn voxel-level features and clustering assignments simultaneously.
• A Brain ROI-based Graph Neural Network (BR-GNN) was employed to extract functional connectivity at the ROI level, integrating voxel-level insights.

Main Results:

• The proposed framework effectively delineates brain functionality across multiple scales (voxel, ROI, sample).
• Experiments demonstrated the method's effectiveness and generalization ability on two datasets, validating voxel-level clustering and ROI-level functional characteristics.

Conclusions:

• The developed multiscale modeling framework offers a more comprehensive approach to brain functional connectivity analysis using fMRI data.
• This method provides a foundation for future applications, including brain disease identification.