DEEP DAG LEARNING OF EFFECTIVE BRAIN CONNECTIVITY FOR FMRI ANALYSIS

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Functional magnetic resonance imaging (fMRI) is a key tool for brain function analysis.
• Graph neural networks (GNNs) show promise in fMRI analysis but are limited by noisy traditional brain network construction.
• Existing methods using region of interest (ROI) similarities can hinder GNN performance.

Purpose of the Study:

• To introduce DABNet, a Deep DAG learning framework based on Brain Networks, for enhanced fMRI analysis.
• To improve the construction of functional brain networks for GNN models.
• To leverage Directed Acyclic Graph (DAG) learning for more effective brain connectivity generation.

Main Methods:

• Developed DABNet, a framework incorporating a brain network generator module.
• Utilized DAG learning to transform raw fMRI time-series data into robust brain connectivities.
• Validated the approach on two independent fMRI datasets.

Main Results:

• DABNet demonstrated superior performance in fMRI analysis compared to traditional methods.
• The generated brain networks effectively captured relevant brain connectivities.
• The framework successfully highlighted prediction-related brain regions, offering interpretability.

Conclusions:

• DABNet offers a significant advancement in applying GNNs to fMRI data.
• The DAG learning approach within DABNet generates more effective brain networks.
• The interpretability of DABNet aids in understanding prediction mechanisms in brain function analysis.