Fine-scale striatal parcellation using diffusion MRI tractography and graph neural networks.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• The striatum, a basal ganglia component, is vital for brain functions via cortical interactions.
• Its complex subdivisions necessitate precise segmentation for accurate study.
• Current segmentation methods may lack the fine-scale resolution required.

Purpose of the Study:

• To introduce a novel deep clustering pipeline for automated, fine-scale striatal parcellation.
• To leverage diffusion MRI (dMRI) tractography for detailed connectivity mapping.
• To improve the accuracy and anatomical fidelity of striatal segmentation.

Main Methods:

• Utilized voxel-based probabilistic fiber tractography and fiber-tract embedding.
• Employed Graph Neural Networks (GNNs) for accurate striatal graph representation.
• Incorporated a Transformer-based GraphConv autoencoder and a joint loss mechanism for segmentation refinement.

Main Results:

• The pipeline successfully achieved fine-scale parcellation of the striatum.
• The GNNs and joint loss mechanism enhanced segmentation precision and anatomical fidelity.
• Integration with self-attention mechanisms improved segmentation robustness.

Conclusions:

• The novel deep clustering pipeline offers a precise method for striatal segmentation using dMRI.
• This technique provides new insights into the striatum's role in cognition and behavior.
• The approach holds potential for clinical applications in neurological disorders.