Sensitivity of quantitative diffusion MRI tractography and microstructure to anisotropic spatial sampling

Elyssa M. McMaster, Nancy R. Newlin, Chloe Cho, Gaurav Rudravaram, Adam M. Saunders, Aravind R. Krishnan, Lucas W. Remedios, Michael E. Kim, Hanliang Xu, Kurt G. Schilling, François Rheault, Laurie E. Cutting, Bennett A. Landman

TL;DR

This study investigates how the shape of voxels in diffusion MRI (dMRI) data, specifically when they are stretched in one direction (anisotropic), affects the accuracy of brain white matter mapping. The researchers found that anisotropic voxels introduce significant biases in microstructural measures like fractional anisotropy (FA) and mean diffusivity (MD), and while resampling to isotropic voxels can improve tractography repeatability for some white matter tracts, it doesn't fully recover the lost microstructural information.

What Problem Does This Paper Address?

Diffusion MRI (dMRI) is a powerful tool for visualizing and understanding the complex network of white matter tracts in the brain. These tracts are crucial for communication between different brain regions. However, the quality and resolution of dMRI data can vary, and a common issue is the use of "anisotropic" voxels, which are stretched in one direction (e.g., longer in the superior-inferior axis). This anisotropy can arise from various acquisition protocols and is prevalent in clinical settings. Previous work has shown that complex graph measures of brain connectivity can change with spatial sampling, but the specific biases introduced by anisotropic voxels on fundamental microstructural and macrostructural measures of white matter have not been well characterized. This paper aims to fill that gap by quantifying these biases and evaluating the effectiveness of resampling strategies to mitigate them.

Methods

The researchers used data from the Human Connectome Project – Young Adult (HCP-YA) dataset, which includes scan-rescan data from 44 healthy subjects. They started with a high-resolution isotropic dataset (1.25 mm isotropic) and then simulated anisotropic voxel data by down-sampling in the z-dimension in 0.25 mm increments, creating resolutions up to 1.25 x 1.25 x 2.75 mm³. They also explored up-sampling these anisotropic datasets to 1.25 mm isotropic and 1 mm isotropic.

Key techniques employed include:

  • Diffusion Tensor Imaging (DTI): Used to generate quantitative maps of fractional anisotropy (FA) and mean diffusivity (MD), which describe water diffusion properties at a microstructural level.
  • Tractography (using TractSeg): A probabilistic approach to reconstruct white matter pathways, generating metrics like bundle volume, length, and surface area.
  • Statistical Analysis: Wilcoxon Signed-Rank tests were used to compare metrics between different resolutions, and Cohen's d coefficient was employed to quantify the effect size of anisotropy.

Datasets

The study utilized the Human Connectome Project – Young Adult (HCP-YA) dataset, which consists of diffusion imaging data from 44 healthy subjects. This dataset is a well-established benchmark for neuroimaging research.

Key Insights and Ideas

  • Anisotropy is a significant source of bias: The study clearly demonstrates that anisotropic voxels introduce substantial biases in microstructural measures (FA and MD), with large effect sizes observed even with relatively small elongations.
  • Resampling isn't a magic bullet for microstructure: While up-sampling anisotropic voxels to an isotropic resolution (especially 1 mm isotropic) can improve the repeatability of tractography metrics (like bundle volume, length, and surface area) for some tracts, it does not fully recover the lost microstructural information. The biases introduced by anisotropy in FA and MD maps are not effectively corrected by simple resampling.
  • Tract direction matters: The impact of anisotropy and the effectiveness of resampling vary depending on the orientation of the white matter tract relative to the direction of voxel elongation. Some tracts, like the corticospinal tract (CST) and arcuate fasciculus (AF), show improved repeatability after resampling, while others, like the corpus callosum (CC), are less responsive.
  • Tractography is more robust to anisotropy than microstructure: The study highlights that tractography metrics, which represent macrostructure, tend to be more reproducible after resampling to isotropic voxels compared to the microstructural measures (FA and MD).

Key Results

  • Microstructural Metrics (FA & MD): Large effect sizes (Cohen's d ≥ 0.8) were observed for FA and MD across all bundles when comparing resolutions with varying degrees of anisotropy. Resampling to isotropic resolutions did not consistently improve the repeatability of these microstructural measures.
  • Macrostructural Metrics (Bundle Volume, Length, Surface Area): These metrics showed significant effects across different voxel sizes. Resampling to 1 mm isotropic improved consistency for the CST and moderately for the AF. However, the CC's surface area and volume measurements were still challenging to make consistent, even after resampling.
  • Qualitative Findings: Figure 4 visually demonstrates how anisotropy leads to thicker, less precise bundles in tractography. While resampling to 1 mm isotropic corrects some of this imprecision, it does not restore the fine detail in the FA maps, which appear smoothed in regions of high anisotropy.

Limitations and Future Work

  • Limited Tract Selection: The study focused on three specific white matter tracts. Future work could investigate a broader selection of bundles to understand how anisotropy affects different types of tracts.
  • Simulated Anisotropy: The study used down-sampling to simulate anisotropic voxels. Future research should include data from genuinely anisotropic acquisitions to evaluate real-world blurring characteristics.
  • Interpolation Method: The study primarily used cubic interpolation. Exploring more advanced interpolation techniques, including deep learning-based super-resolution methods, could potentially improve the recovery of anatomical detail.
  • Population: The study was limited to a healthy young adult population. Future work should investigate how age and disease status might influence the sensitivity to anisotropic voxels.
  • Computational Cost: Increasing resolution beyond 1 mm isotropic led to unrealistic increases in compute time, suggesting a trade-off between resolution and computational feasibility.

Summary updated

Abstract

Fractional anisotropy and mean diffusivity cannot be recovered with basic up-sampling from low quality data with gold-standard data with the methods selected for this study. However, the bundle measures across our selected regions of interest become more repeatable when voxels are resampled to 1 mm isotropic.
BibTeX
@inproceedings{mcmaster2025sensitivity,
  author = {Elyssa M. McMaster and Nancy R. Newlin and Chloe Cho and Gaurav Rudravaram and Adam M. Saunders and Aravind R. Krishnan and Lucas W. Remedios and Michael E. Kim and Hanliang Xu and Kurt G. Schilling and François Rheault and Laurie E. Cutting and Bennett A. Landman},
  title = {{Sensitivity of quantitative diffusion MRI tractography and microstructure to anisotropic spatial sampling}},
  booktitle = {Magnetic},
  year = {2025},
  doi = {10.1016/j.mri.2025.110539},
}
Figure