Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Global Context Is All You Need for Parallel Efficient Tractography Parcellation

Valentin von Bornhaupt, Johannes Grün, and Justus Bisten, Tobias Bauer, Theodor Rüber, Thomas Schultz

TL;DR

This work questions the value of local context in tractography parcellation and introduces PETParc, a transformer-based approach that relies on global context by processing randomly partitioned sub-tractograms in parallel. A flip-invariant embedding (or flip augmentation) addresses streamline orientation without heavy local recomputation, enabling up to two orders of magnitude faster inference than prior methods while achieving equal or superior accuracy. Trained on a 1600-cluster dataset mapped to 42 bundles, PETParc generalizes to diverse cohorts and maintains robustness in pathological brains, including post-hemispherotomy cases. The combination of parallel, registration-free processing and strong performance positions PETParc as a practical tool for large-scale diffusion MRI tractography and clinical workflows.

Abstract

Whole-brain tractography in diffusion MRI is often followed by a parcellation in which each streamline is classified as belonging to a specific white matter bundle, or discarded as a false positive. Efficient parcellation is important both in large-scale studies, which have to process huge amounts of data, and in the clinic, where computational resources are often limited. TractCloud is a state-of-the-art approach that aims to maximize accuracy with a local-global representation. We demonstrate that the local context does not contribute to the accuracy of that approach, and is even detrimental when dealing with pathological cases. Based on this observation, we propose PETParc, a new method for Parallel Efficient Tractography Parcellation. PETParc is a transformer-based architecture in which the whole-brain tractogram is randomly partitioned into sub-tractograms whose streamlines are classified in parallel, while serving as global context for each other. This leads to a speedup of up to two orders of magnitude relative to TractCloud, and permits inference even on clinical workstations without a GPU. PETParc accounts for the lack of streamline orientation either via a novel flip-invariant embedding, or by simply using flips as part of data augmentation. Despite the speedup, results are often even better than those of prior methods. The code and pretrained model will be made public upon acceptance.

Global Context Is All You Need for Parallel Efficient Tractography Parcellation

TL;DR

This work questions the value of local context in tractography parcellation and introduces PETParc, a transformer-based approach that relies on global context by processing randomly partitioned sub-tractograms in parallel. A flip-invariant embedding (or flip augmentation) addresses streamline orientation without heavy local recomputation, enabling up to two orders of magnitude faster inference than prior methods while achieving equal or superior accuracy. Trained on a 1600-cluster dataset mapped to 42 bundles, PETParc generalizes to diverse cohorts and maintains robustness in pathological brains, including post-hemispherotomy cases. The combination of parallel, registration-free processing and strong performance positions PETParc as a practical tool for large-scale diffusion MRI tractography and clinical workflows.

Abstract

Whole-brain tractography in diffusion MRI is often followed by a parcellation in which each streamline is classified as belonging to a specific white matter bundle, or discarded as a false positive. Efficient parcellation is important both in large-scale studies, which have to process huge amounts of data, and in the clinic, where computational resources are often limited. TractCloud is a state-of-the-art approach that aims to maximize accuracy with a local-global representation. We demonstrate that the local context does not contribute to the accuracy of that approach, and is even detrimental when dealing with pathological cases. Based on this observation, we propose PETParc, a new method for Parallel Efficient Tractography Parcellation. PETParc is a transformer-based architecture in which the whole-brain tractogram is randomly partitioned into sub-tractograms whose streamlines are classified in parallel, while serving as global context for each other. This leads to a speedup of up to two orders of magnitude relative to TractCloud, and permits inference even on clinical workstations without a GPU. PETParc accounts for the lack of streamline orientation either via a novel flip-invariant embedding, or by simply using flips as part of data augmentation. Despite the speedup, results are often even better than those of prior methods. The code and pretrained model will be made public upon acceptance.

Paper Structure

This paper contains 13 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods and Data
  3. Data
  4. Preprocessing
  5. Flip-Invariant Embedding
  6. Model Training
  7. Experiments and Results
  8. Performance on test dataset
  9. Robustness against streamline flipping and random sub-tractogram partitioning
  10. Qualitative evaluation
  11. Conclusion
  12. Acknowledgments.
  13. Disclosure of Interests.

Figures (3)

  • Figure 1: Our pipeline consists of the four stages preprocessing (yellow), embedding/token preparation (orange), transformer encoder/classifier (red), and postprocessing (blue). Tract names can be found in the paper by Zhang et al. AnatomicallyCuratedFiber.
  • Figure 2: Comparison of the flip-augmented and flip-invariant versions of PETParc to TractCloud on four subjects from different studies. All models generalize well to different measurement schemes, different ages, and different health status.
  • Figure 3: Parcellation results of the corticospinal tract (CST) and arcuate fasciculus (AF) in three individuals post-hemispherotomy. Both variants of PETParc yield more complete parcellations compared to TCglo+loc and TCglo.