Brain Connectivity Based Graph Convolutional Networks and Its Application to Infant Age Prediction.

Area of Science:

• Neuroimaging and Developmental Neuroscience
• Machine Learning and Artificial Intelligence
• Pediatric Brain Development

Background:

• Infancy is a critical period for brain development, with brain age serving as a key neuroimaging index.
• Deviations in brain age from chronological age can signal abnormal developmental trajectories.
• Resting-state functional magnetic resonance imaging (rs-fMRI) provides valuable data on brain connectivity.

Purpose of the Study:

• To develop a novel Graph Convolutional Network (GCN) model for accurate infant brain age prediction using rs-fMRI data.
• To address the limitations of existing node-based GCNs for dense brain connectivity graphs.
• To introduce an enhanced coarse-to-fine framework for improved developmental trajectory assessment.

Main Methods:

• Proposed an edge-based Graph Path Convolution (GPC) method, integrated into Brain Connectivity Graph Convolutional Networks (BC-GCN).
• Developed upgraded BC-GCN models (BC-GCN-Res, BC-GCN-SE) incorporating residual and attention modules.
• Implemented a two-stage coarse-to-fine framework with cross-group training for robust age prediction.

Main Results:

• The proposed BC-GCN-SE model with the coarse-to-fine framework achieved a mean absolute error of 49.9 days.
• This represents a significant reduction in prediction error compared to state-of-the-art methods (>70 days to 49.9 days).
• The coarse-to-fine framework demonstrated over 10 days of error reduction when applied to various models.

Conclusions:

• The BC-GCN model, particularly with the SE and coarse-to-fine framework, offers a powerful tool for infant brain age prediction.
• This approach enhances the ability to identify early deviations in brain development from neuroimaging data.
• The findings have implications for early diagnosis and intervention in pediatric neurodevelopmental disorders.