Representing diffusion MRI in 5-D simplifies regularization and segmentation of white matter tracts

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Diffusion magnetic resonance imaging (dMRI) is crucial for mapping white matter tracts.
• Current 3D space representations struggle to disentangle complex, crossing fiber tracts.
• High angular resolution dMRI (HARDI) generates rich data requiring advanced processing.

Purpose of the Study:

• To introduce a new five-dimensional (5-D) space representation for HARDI data.
• To demonstrate how this 5-D space simplifies regularization and segmentation of dMRI.
• To explore direct segmentation of white matter structures as distinct bundles in 5-D space.

Main Methods:

• Developed a non-Euclidean 5-D space incorporating position and orientation.
• Applied Chan-Vese method with edge-less active contours for segmentation in 5-D.
• Utilized total variation functional for regularization across multiple scales in 5-D.
• Validated methods on synthetic data and real human brain HARDI data (Diffusion Spectrum Imaging).

Main Results:

• The 5-D representation effectively disentangles crossing fiber tracts, overcoming 3D limitations.
• Regularization and segmentation tasks were significantly simplified in the 5-D space.
• Demonstrated successful denoising and facilitated segmentation through multi-scale analysis.
• Achieved direct segmentation of white matter structures as separate bundles.

Conclusions:

• The proposed 5-D space representation offers a powerful framework for dMRI analysis.
• This high-dimensional approach simplifies complex image processing tasks like regularization and segmentation.
• The method shows promise for more accurate and detailed mapping of brain white matter architecture.