Labeling of ambiguous subvoxel fibre bundle configurations in high angular resolution diffusion MRI

Peter Savadjiev, Jennifer S. W. Campbell, Maxime Descoteaux, Rachid Deriche, G. Bruce Pike, Kaleem SiddiqiNeuroImage 2008 · DOI

TL;DR

This paper introduces a novel algorithm to distinguish between different complex arrangements of white matter fiber bundles within a single MRI voxel, a common challenge in diffusion MRI. By inferring local curve geometries, their method accurately labels these configurations, leading to improved accuracy in brain white matter tractography.

What Problem Does This Paper Address?

Diffusion MRI (dMRI) is a powerful tool for visualizing the brain's white matter, which is composed of millions of nerve fibers. These fibers are organized into bundles, and dMRI techniques can map their general direction. However, a significant challenge arises when multiple fiber bundles pass through a single imaging voxel (the smallest unit of the image). This "subvoxel" complexity, where fibers can cross, fan out, or curve, leads to ambiguous signal patterns. Traditional dMRI methods, like diffusion tensor imaging (DTI), struggle to resolve these ambiguities, often resulting in inaccurate tractography (reconstruction of fiber pathways). While more advanced methods like High Angular Resolution Diffusion (HARD) MRI can estimate more detailed fiber orientation distributions (ODFs), they still face difficulties in distinguishing between subtle but functionally important configurations, such as a single fiber bundle curving versus a bundle fanning out into multiple directions. This ambiguity can lead to errors in mapping brain connections, impacting our understanding of brain function and disease.

Methods

The core of this paper's contribution lies in its novel labeling scheme, built upon a pre-existing 3D curve inference algorithm. Here's a breakdown:

  • 3D Curve Inference: This algorithm, previously developed by the authors, estimates local geometric properties (like curvature and torsion) of 3D curves from dMRI orientation data. It works by identifying "co-helical triplets" of fiber directions within a local neighborhood. Essentially, it infers the shape of the underlying fiber tracts as helical arcs.
  • Labeling Scheme: The key innovation is using the inferred helical curves to classify voxels into three categories:
  • Single-fibre bundles: These may exhibit curvature and torsion.
  • Fanning fibre tracts: A single bundle fans into multiple directions within the voxel.
  • Crossing fibre tracts: Two or more bundles cross. The algorithm leverages both the shape of the ODF (e.g., number of peaks) and the geometric model inferred by the 3D curve inference to make these distinctions. A crucial aspect is the recovery of "fan polarity," which indicates the direction of fanning, essential for correct tractography.

  • Tractography Integration: The developed labeling information is then used to guide a deterministic streamline tractography algorithm. This means that when the algorithm encounters a voxel labeled as a "fan," it can make informed decisions about whether to follow all directions within the fan or just the primary direction, based on the fan's polarity and the direction of tract propagation.

Datasets

The study utilized dMRI data acquired from a single healthy human subject. The specific acquisition parameters are detailed, including the use of a Siemens 3-T Trio MR scanner with an 8-channel phased-array head coil and a specific diffusion encoding scheme (N=99 directions, b=3000 s/mm²). A T1-weighted anatomical scan was also acquired for anatomical reference. No new benchmarks were introduced; the focus was on applying the developed method to existing dMRI data.

Key Insights and Ideas

  • Geometry over Orientation: The paper convincingly argues that simply knowing the orientation of fibers within a voxel is insufficient. Understanding the local geometry (curving vs. fanning) is critical for accurate tractography.
  • Resolving Ambiguity: The method effectively disambiguates situations where different fiber geometries produce similar ODF shapes, a long-standing problem in the field.
  • Fan Polarity is Key: Recovering the direction of fanning ("fan polarity") is shown to be crucial for correctly guiding tractography, especially when fibers are diverging.
  • Bridging the Gap: The 3D curve inference provides a powerful intermediate step, translating ODF data into a more interpretable geometric representation that can then be used for classification.

Key Results

  • Successful Labeling: The algorithm successfully labeled voxels into the categories of single-fibre tracts, fanning tracts, and crossing tracts, with visual results appearing anatomically plausible.
  • Improved Tractography: When integrated into a tractography algorithm, the labeling information led to improved reconstruction of fiber pathways, particularly in regions with complex fanning structures like the corpus callosum and internal capsule. The paper shows examples where tractography with labeling recovers more extensive and accurate connections to cortical areas compared to tractography without labeling.
  • Qualitative Demonstration: The results are primarily demonstrated through qualitative visual comparisons of tractography outputs, highlighting the enhanced recovery of fanning pathways.

Limitations and Future Work

  • Single Subject Data: The tractography experiments were performed on data from a single subject. While the labeling results are presented for a region of interest, further validation on larger datasets and across different subjects would be beneficial.
  • Deterministic Tractography: The current implementation uses a deterministic streamline propagation approach. The authors suggest that their labeling methodology could also be incorporated into probabilistic tractography frameworks, which might further improve robustness by accounting for uncertainty.
  • Bottlenecks and Branches: Future work is planned to address other complex configurations like "bottlenecks" (where fibers narrow) and "branches" (where a single tract splits), which are also challenging for current tractography methods.
  • Broader Applications: The authors envision their method being useful for studying white matter disease and understanding functional deficits by providing more accurate connectivity maps. They also suggest its utility in interpreting anisotropy index changes and in fMRI activation studies.

Summary updated

BibTeX
@article{savadjiev2008labeling,
  author = {Peter Savadjiev and Jennifer S. W. Campbell and Maxime Descoteaux and Rachid Deriche and G. Bruce Pike and Kaleem Siddiqi},
  title = {{Labeling of ambiguous subvoxel fibre bundle configurations in high angular resolution diffusion MRI}},
  journal = {NeuroImage},
  year = {2008},
  doi = {10.1016/j.neuroimage.2008.01.028},
}
Figure