Diffusion, Convection and Erosion on SE(3)/({0} \\times SO(2)) and their\n Application to the Enhancement of Crossing Fibers

Remco Duits, Eric Creusen, Arpan Ghosh, Tom Dela HaijearXiv 2011 · DOI · arXiv

Abstract

In this article we study both left-invariant (convection-)diffusions and left-invariant Hamilton-Jacobi equations on the space $SE(3)/(\{0\} \times SO(2))$ of 3D-positions and orientations naturally embedded in the group $SE(3)$ of 3D-rigid body movements. The general motivation for these (convection-)diffusions and erosions is to obtain crossing-preserving fiber enhancement on probability densities defined on the space of positions and orientations. The linear left-invariant (convection-)diffusions are forward Kolmogorov equations of Brownian motions on $SE(3)/(\{0\}\times SO(2))$ and can be solved by convolution with the corresponding Green's functions or by a finite difference scheme. The left-invariant Hamilton-Jacobi equations are Bellman equations of cost processes on $SE(3)/(\{0\}\times SO(2))$ and they are solved by a morphological convolution with the corresponding Green's functions. Furthermore, we consider pseudo-linear scale spaces on the space of positions and orientations that combines dilation and diffusion in a single evolution. In our design and analysis for appropriate linear, non-linear, morphological and pseudo-linear scale spaces on $SE(3)/(\{0\}\times SO(2))$ we employ the underlying differential geometry on $SE(3)$, where the frame of left-invariant vector fields serves as a moving frame of reference. Furthermore, we will present new and simpler finite difference schemes for our diffusions, which are clear improvements of our previous finite difference schemes. We apply our theory to the enhancement of fibres in magnetic resonance imaging (MRI) techniques for imaging water diffusion processes in brain white matter. We provide experiments of our crossing-preserving evolutions on neural images of a human brain containing crossing fibers.

Introduction: Addressing the Crossing Fiber Problem

Mapping the structural connectivity of the human brain is a cornerstone of modern neuroscience. This is primarily achieved through Diffusion Weighted MRI (DW-MRI), which tracks the diffusion of water molecules along axonal bundles. However, standard clinical techniques like Diffusion Tensor Imaging (DTI) face a fundamental limitation: they assume water diffuses in a single ellipsoidal shape per voxel. Consequently, DTI cannot accurately represent regions where neural fibers cross, kiss, or diverge, often resulting in ambiguous "blobs" at critical junctions.

Comparison of DTI and HARDI glyphs in fiber tracking. DTI (left) fails to resolve crossing fibers, while HARDI (right) captures multiple orientation lobes.

High Angular Resolution Diffusion Imaging (HARDI) was developed to overcome this by capturing more complex orientation distributions. While HARDI provides the necessary detail to resolve crossings, it is highly sensitive to noise and computationally demanding. This paper proposes a unified geometric framework to enhance these complex 3D images, sharpening fiber profiles while effectively denoising the data.

Methodology: Geometry on the Lie Group $SE(3)$

The core innovation of this work is treating brain fiber data as living on a specific mathematical space: $\mathbb{R}^3 \rtimes S^2$ (positions and orientations), embedded into the Lie group $SE(3)$ of 3D rigid body motions. By leveraging this geometry, the authors develop "left-invariant" operators that ensure the enhancement process remains consistent regardless of the coordinate system's orientation.

Key components of the framework include:

  • Hypo-elliptic Diffusion: This process smooths data specifically along the direction of the fibers. By "filling in the gaps" longitudinally without blurring the fibers into the background transversally, it preserves the integrity of thin neural structures.
  • Hamilton-Jacobi Equations (Erosion): Morphological operators are employed to "erode" or sharpen fiber profiles, making them crisper and easier to track in subsequent analysis.
  • Moving Frame Concept: Using left-invariant vector fields, the mathematical framework effectively attaches a coordinate system to the fiber itself. As a fiber curves, the coordinate system follows, allowing for precise local smoothing and sharpening.

Key Results and Insights

The researchers evaluated their framework on both synthetic noisy datasets and real-world human brain MRI data, specifically targeting areas like the corona radiata where fibers are known to cross extensively.

HARDI glyphs representing probability densities of fiber orientations ($b = 1000 s/mm^2, N_o = 49$).

Because the diffusion occurs in a 5D space (3D position + 2D orientation), fibers crossing at the same spatial point but with different angles remain separated in the mathematical domain. This prevents the "bleeding" effect seen in traditional spatial smoothing. The results demonstrate that the proposed nonlinear adaptive diffusions can successfully remove spurious crossings caused by noise while keeping actual biological structures sharp. Furthermore, the theory suggests a path toward "DTI extrapolation," where crossing information might be mathematically inferred from simpler DTI scans, potentially reducing clinical scan times.

Conclusion

By unifying linear diffusion and morphological erosion within a rigorous geometric framework on $SE(3)$, this work provides a powerful tool for neuroimaging. While the mathematical complexity is high, the benefits—superior crossing representation and robust denoising—are significant. Future work will focus on integrating these enhancement techniques directly into tractography algorithms to improve the accuracy of full-brain connectivity maps.

Summary updated

Abstract

In this article we study both left-invariant (convection-)diffusions and\nleft-invariant Hamilton-Jacobi equations on the space SE(3)/({0} \\times SO(2))\nof 3D-positions and orientations naturally embedded in the group SE(3) of\n3D-rigid body movements. The general motivation for these\n(convection-)diffusions and erosions is to obtain crossing-preserving fiber\nenhancement on probability densities defined on the space of positions and\norientations. The linear left-invariant (convection-)diffusions are forward\nKolmogorov equations of Brownian motions on SE(3)/({0}\\timesSO(2)) and can be\nsolved by convolution with the corresponding Green's functions or by a finite\ndifference scheme. The left-invariant Hamilton-Jacobi equations are Bellman\nequations of cost processes on SE(3)/({0}\\timesSO(2)) and they are solved by a\nmorphological convolution with the corresponding Green's functions.\nFurthermore, we consider pseudo-linear scale spaces on the space of positions\nand orientations that combines dilation and diffusion in a single evolution. In\nour design and analysis for appropriate linear, non-linear, morphological and\npseudo-linear scale spaces on SE(3)/({0}\\timesSO(2)) we employ the underlying\ndifferential geometry on SE(3), where the frame of left-invariant vector fields\nserves as a moving frame of reference. Furthermore, we will present new and\nsimpler finite difference schemes for our diffusions, which are clear\nimprovements of our previous finite difference schemes. We apply our theory to\nthe enhancement of fibres in magnetic resonance imaging (MRI) techniques for\nimaging water diffusion processes in brain white matter. We provide experiments\nof our crossing-preserving evolutions on neural images of a human brain\ncontaining crossing fibers.\n
BibTeX
@misc{duits2011diffusion,
  author = {Remco Duits and Eric Creusen and Arpan Ghosh and Tom Dela Haije},
  title = {{Diffusion, Convection and Erosion on SE(3)/(\{0\} \textbackslash\{\}\textbackslash\{\}times SO(2)) and their\textbackslash\{\}n Application to the Enhancement of Crossing Fibers}},
  year = {2011},
  eprint = {1103.0656},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1103.0656},
}
Figure