Hyperspherical von Mises-Fisher mixture (HvMF) modelling of high angular resolution diffusion MRI

Area of Science:

• Diffusion MRI
• Neuroimaging
• Computational Neuroscience
• Biomedical Engineering

Background:

• High-angular resolution diffusion imaging (HARDI) generates complex orientation distribution functions (ODFs) that require sophisticated modeling.
• Accurate partitioning of single and multiple fibre populations within voxels is crucial for understanding white matter architecture.
• Existing methods face challenges in robustly modeling and differentiating complex fibre configurations.

Purpose of the Study:

• To introduce a novel mapping of unit vectors onto a 5D hypersphere for modeling and partitioning ODFs from HARDI data.
• To explore the application of a von Mises-Fisher mixture model for analyzing directional samples in diffusion MRI.
• To establish a method for differentiating between single and multiple fibre regions using penalized-likelihood model selection.

Main Methods:

• Utilized a unit vector mapping onto a 5D hypersphere to represent and partition ODFs.
• Employed a von Mises-Fisher (vMF) mixture model for directional data analysis.
• Performed error analysis and applied penalized-likelihood model selection for fibre region differentiation.

Main Results:

• Demonstrated the utility of the hypersphere mapping for ODF modeling and partitioning.
• Linked the maximum likelihood estimate (MLE) of the second moment of the HvMF (Hypergeometric von Mises-Fisher) probability density function to fractional anisotropy.
• Successfully differentiated single and multiple fibre regions using the proposed penalized-likelihood method.

Conclusions:

• The proposed hypersphere mapping and vMF mixture model provide a robust framework for HARDI data analysis.
• The method offers improved accuracy in estimating fibre orientations and partitioning complex white matter regions.
• This approach has significant implications for quantitative analysis and tractography in diffusion MRI studies.