BrainGB

BrainGB facilitates brain network analysis using Graph Neural Networks (GNNs) grounded in geometric deep learning to model structural and functional connectome data from neuroimaging.

Key Features:

• Unified Framework: Integrates components of brain network analysis to ensure consistency and standardization across studies.
• Modular Design: Modular architecture enables customization and implementation of specific GNN designs and architectures.
• Scalability: Supports large-scale analyses across diverse datasets, cohorts, and neuroimaging modalities.
• Reproducibility: Emphasizes reproducible research practices and provides model implementations and a Python package.

Scientific Applications:

• Neuroimaging network analysis: Enables systematic study of brain networks derived from structural and functional neuroimaging data using GNNs.
• Connectome modeling: Enhances modeling of complex connectome data to analyze brain connectivity patterns.
• Cross-cohort comparative studies: Facilitates large-scale and multi-cohort analyses to evaluate GNN performance and generalizability.

Methodology:

Uses standardized brain network construction pipelines from various neuroimaging modalities and modularizes GNN implementations to experiment with different architectures, employing Graph Neural Networks inspired by geometric deep learning and empirical evaluation across diverse datasets.