FBNetGen: Task-aware GNN-based fMRI Analysis via Functional Brain Network Generation.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Functional magnetic resonance imaging (fMRI) is crucial for brain function research.
• Functional brain networks derived from fMRI hold potential for clinical predictions.
• Existing methods for fMRI network analysis are often noisy, task-agnostic, and incompatible with deep Graph Neural Networks (GNNs).

Purpose of the Study:

• To develop FBNETGEN, a task-aware and interpretable framework for deep brain network generation from fMRI data.
• To enable end-to-end training of GNN models for clinical predictions using fMRI-derived brain networks.
• To enhance the interpretability of fMRI analysis by highlighting prediction-relevant brain regions.

Main Methods:

• FBNETGEN integrates region of interest (ROI) feature extraction, brain network generation, and clinical prediction using GNNs.
• A novel graph generator transforms raw fMRI time-series data into task-oriented brain networks.
• The framework is trained end-to-end, guided by specific prediction tasks.

Main Results:

• FBNETGEN demonstrates superior effectiveness and interpretability compared to traditional methods.
• Experiments on the Adolescent Brain Cognitive Development (ABCD) and PNC fMRI datasets validate the framework's performance.
• The generated learnable graphs provide insights by identifying key brain regions for predictions.

Conclusions:

• FBNETGEN offers a powerful and interpretable approach for network-based fMRI analysis.
• The framework effectively leverages GNNs for clinical predictions from fMRI data.
• Task-aware brain network generation is key to unlocking the full potential of GNNs in neuroscience.