Multi-label segmentation of white matter structures: application to neonatal brains

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Developmental Neuroscience

Background:

• Accurate segmentation of neonatal brain white matter bundles is vital for understanding brain development and predicting psychiatric disorders.
• The complexity of white matter anatomy and diffusion-weighted MRI resolution pose challenges, as multiple fiber bundles can occupy a single voxel.
• Existing methods struggle to fully capture the intricate white matter architecture in neonates.

Purpose of the Study:

• To develop a novel supervised algorithm for accurate, voxel-wise segmentation of neonatal white matter bundles.
• To assign one or multiple anatomical labels to each voxel, reflecting complex white matter anatomy.
• To provide tools for understanding neonatal brain development and detecting abnormalities.

Main Methods:

• Development of a supervised multi-label k-nearest neighbor (ML-kNN) classification algorithm operating in Riemannian diffusion tensor spaces.
• Utilizing diffusion tensors on the Log-Euclidean Riemannian manifold of symmetric positive definite (SPD) matrices as the feature space.
• Employing the maximum a posteriori (MAP) principle for label prediction, leveraging neighbor information without assuming feature distributions.

Main Results:

• The ML-kNN algorithm automatically determines the number of white matter bundles at each location.
• The method provides accurate anatomical annotation of neonatal white matter.
• Binary masks for individual white matter bundles are generated, facilitating tract-based statistical analysis.

Conclusions:

• The developed ML-kNN approach offers accurate and automated segmentation of neonatal white matter bundles.
• This method enhances the understanding of neonatal brain development and aids in the early detection of potential psychiatric disorders.
• The generated anatomical labels and binary masks are valuable for clinical studies and further research.