Multimodal Connectivity-based Cortical Segmentation with Graph Neural Networks.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Manual segmentation of the brain cortex from MRI is time-consuming and requires specialized expertise.
• Development of automated, accurate segmentation algorithms is crucial for efficient neuroimaging analysis.
• Graph Neural Networks (GNNs) offer a novel approach for analyzing complex brain data.

Purpose of the Study:

• To evaluate the efficacy of different GNN architectures (GCN, GAT, Graph U-Net) for brain cortex segmentation.
• To investigate the impact of multimodal data (sMRI and dMRI) on segmentation performance.
• To compare GNN-based segmentation with FreeSurfer for predicting demographic/clinical data.

Main Methods:

• Trained three GNN architectures (GCN, GAT, Graph U-Net) on structural brain connectivity.
• Utilized structural MRI (sMRI) and diffusion MRI (dMRI) derived attributes for multimodal segmentation.
• Evaluated performance using silver-standard cortical labels from FreeSurfer and compared with FreeSurfer's output.

Main Results:

• The Graph Attention Network (GAT) architecture achieved competitive Dice scores compared to non-graph methods.
• Incorporating structural connectivity from dMRI significantly enhanced segmentation accuracy over sMRI alone.
• GNN-based and FreeSurfer segmentations showed comparable performance in predicting demographic/clinical data.

Conclusions:

• GNNs, particularly GAT, are effective tools for automated brain cortex segmentation.
• Multimodal data integration (sMRI + dMRI) improves the accuracy of GNN-based brain segmentation.
• GNN approaches provide a viable alternative to traditional methods like FreeSurfer for neuroimaging analysis.