↓ Skip to main content
BMN Lab

BundleWarp: Streamline-Based Nonlinear Registration of White Matter Tracts

BundleWarp introduces a streamline-specific nonlinear registration framework that directly warps white matter fiber bundles while preserving their critical topological features — enabling precise, anatomically faithful alignment across subjects for population-level brain studies.

The Problem
#

Existing white matter registration methods rely on image-based transformations (e.g., ANTs) or linear streamline alignment, both of which fail to capture the complex, nonlinear shape differences between individual subjects’ fiber bundles. This limits the sensitivity of population-level tractometry and bundle shape analysis.

Method: Two-Step Pipeline
#

BundleWarp solves registration in two stages:

1. iterLAP — Streamline Correspondence
Solves a many-to-one assignment problem to find the best matching between streamlines across two bundles — handling the practical reality that bundles from different subjects contain different numbers of streamlines.

2. mlCPD — Nonlinear Deformation
Applies memoryless Coherent Point Drift deformations to individual streamlines. A Gaussian kernel regularization ensures that streamlines move coherently as a group, preserving anatomical topology while achieving full nonlinear alignment.

A controllable regularization parameter λ lets users tune the degree of deformation:

  • λ = 0.3–0.5 → partial deformation, preserving native shape (recommended for clinical use)
  • λ < 0.001 → full deformation, maximizing bundle overlap (for shape difference quantification)

BundleWarp pipeline: input bundles → affine alignment → iterLAP streamline correspondence → mlCPD nonlinear deformation.

Validation
#

Dataset: 64 subjects (32 healthy controls, 32 Parkinson’s disease) from the Parkinson’s Progression Markers Initiative (PPMI)
Scale: 1,728 bundle pairs across 27 white matter tract types, each registered to HCP-842 atlas bundles

BundleWarp morphometry: deformation field magnitudes quantify along-tract shape differences between subjects, enabling fine-grained tractometry and bundle comparison.

Applications
#

  • Tractometry — improved segment-to-segment correspondence for along-tract scalar profiling
  • Bundle shape analysis — quantify structural differences between subjects using deformation field magnitudes
  • Atlas construction — build population-specific white matter atlases with better anatomical fidelity
  • Bundle segmentation — enhance RecoBundles model alignment

Publication
#

Published as: BundleWarp: Enhancing white matter tractometry and morphometry with precise neuronal mapping using streamline-based nonlinear registration. Medical Image Analysis, 2026. https://doi.org/10.1016/j.media.2026.104114

Code
#