Brain Connectivity Measures via Direct Sub-Finslerian Front Propagation on the 5D Sphere Bundle of Positions and Directions

Jorg M. Portegies, Stephan Meesters, Pauly Ossenblok, A. Fuster, Luc Florack, Remco DuitsMathematics 2019 · DOI · OpenAlex

Introduction

Understanding the physical connectivity between brain regions is fundamental to neuroscience. Diffusion-Weighted MRI (dMRI) is the primary tool for mapping white matter tracts, but traditional 3D tractography methods face significant challenges. Specifically, they often struggle with "fiber crossings"—regions where multiple bundles intersect. In standard 3D space, algorithms frequently "leak" at these junctions, erroneously jumping between distinct tracts.

Methodology: Lifting to 5D

To address the limitations of 3D methods, this research proposes a novel framework that "lifts" the problem into a 5D sphere bundle ($M = \mathbb{R}^3 \times S^2$). By defining every point by its 3D position and its 2D orientation, crossing fibers that overlap in 3D space become distinct, separable points in 5D.

Figure 1: Comparison of conventional 3D propagation (left) showing 'leakage' at crossings, versus the proposed 5D lifting (right) which maintains tract integrity.

Sub-Finslerian Geometry and Anisotropic Fast Marching

The core of the method relies on sub-Finslerian geometry, which defines a distance metric penalizing both path length and curvature. This ensures that the identified connections are biologically plausible, favoring smooth trajectories over sharp, unnatural turns.

Furthermore, the authors employ an asymmetric Finsler function to enforce "forward-only" propagation. This prevents the formation of unnatural cusps or backtracking during the front propagation process. The numerical implementation uses Anisotropic Fast Marching (FM) to efficiently solve the eikonal equation, tracking how a wave-front expands through the complex 5D architecture of the brain.

Connectivity Metric

Unlike traditional methods that count streamlines, this approach defines connectivity ($\kappa$) as a scale-invariant ratio. It compares the "cost" of a path supported by the Fiber Orientation Distribution (FOD) data against a baseline cost, providing a direct measure of structural connectivity from the geodesic distance field.

Experimental Validation

The method was validated using two primary datasets:

  1. ISBI HARDI Challenge 2013: A synthetic "digital phantom" containing 27 complex fiber bundles used to establish ground truth accuracy.
  2. Human Connectome Project (HCP): Real-world dMRI data from 10 healthy volunteers, focusing on the connectivity of the Anterior Nucleus of the Thalamus (ANT).
Figure 2: Multi-view visualization of complex fiber bundles resolved using the proposed 5D front propagation method.

Key Results

  • Accuracy: On the ISBI challenge, the method correctly identified the "true" connection as the highest rank in 21 out of 54 cases, significantly outperforming traditional 3D methods in complex crossing scenarios.
  • Anatomical Consistency: In HCP data, the algorithm's results aligned with known neuroanatomy, correctly identifying high connectivity between the ANT and the cingulate gyrus while showing low connectivity to the accumbens.
  • Robustness: The asymmetric metric successfully prevented spatial cusps, leading to more stable and biologically realistic results than symmetric Riemannian approaches.

Conclusion

By treating direction as a coordinate rather than a property, this 5D front propagation method effectively "untangles" the brain's complex wiring. While computationally more demanding than 3D approaches, it offers a mathematically grounded and more accurate metric for structural connectivity, with potential applications in optimizing clinical interventions like Deep Brain Stimulation.

Summary updated

BibTeX
@inproceedings{portegies2019brain,
  author = {Jorg M. Portegies and Stephan Meesters and Pauly Ossenblok and A. Fuster and Luc Florack and Remco Duits},
  title = {{Brain Connectivity Measures via Direct Sub-Finslerian Front Propagation on the 5D Sphere Bundle of Positions and Directions}},
  booktitle = {Mathematics},
  year = {2019},
  doi = {10.1007/978-3-030-05831-9\_24},
}

Loading related content...

Figure