henhaobio
Sign in

Human Brain MappingNeurosciencePeer-reviewed journal article

White matter geometry confounds Diffusion Tensor Imaging Along Perivascular Space (DTI-ALPS) measures

Kurt G. SchillingiD, Allen T. NewtoniD, Chantal M. W. TaxiD, Markus NilssoniD, Maxime ChamberlandiD, Adam Anderson, Bennett A. LandmaniD, Maxime DescoteauxiD

PDFDOIPreprintData
In-depth full-text digest available · Jump to digest ↓

Abstract

IntroductionThe perivascular space (PVS) is integral to glymphatic function, facilitating fluid exchange and waste clearance in the brain. Diffusion Tensor Imaging Along the Perivascular Space (DTI-ALPS) has been proposed as a non-invasive marker of perivascular diffusion, yet its specificity remains unclear. ALPS measures assume that radial asymmetry in white matter diffusivity predominantly reflects PVS contributions. However, anatomical and microstructural confounds may influence these metrics. MethodsWe systematically evaluated potential biases in ALPS-derived measures using high-resolution, multi-shell diffusion MRI from the Human Connectome Project (HCP) and high-field imaging. Specifically, we examined (1) the prevalence of radial asymmetry across white matter, (2) the influence of crossing fibers on ALPS indices, (3) the impact of axonal undulations and dispersion, and (4) the spatial alignment of vasculature with white matter in ALPS-associated regions. ResultsRadial asymmetry is widespread across white matter and persists even at high b-values, suggesting a dominant contribution from axonal geometry rather than faster PVS-specific diffusion. Crossing fibers significantly inflate ALPS indices, with greater radial asymmetry observed in regions with a greater prevalence of crossing fibers. Furthermore, anisotropic axonal dispersion and undulations introduce systematic asymmetry independent of perivascular diffusion. Finally, high-resolution vascular imaging reveals substantial heterogeneity in medullary vein orientation, challenging the assumption that PVS consistently aligns with the left-right axis in ALPS regions. ConclusionALPS indices are significantly influenced by white matter microstructure, including fiber crossings, undulations, and dispersion. These findings suggest that ALPS-derived metrics may not provide a direct measure of glymphatic function but rather reflect underlying axonal geometry. Interpretations of ALPS-derived metrics as biomarkers of glymphatic function must consider these anatomical complexities, and future studies should integrate advanced modeling approaches to disentangle perivascular contributions from white matter structure.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (19 Apr 2025) · CC BY 4.0

Radial diffusivity asymmetry driving DTI-ALPS indices pervades white matter, persists at high b-values, and tracks crossing fibers, axonal undulation and dispersion rather than perivascular diffusion.

The ALPS-index has been adopted widely as a low-effort, DTI-only proxy for glymphatic clearance in aging and neurodegeneration, yet it presumes radially symmetric white matter diffusion and left-right medullary veins orthogonal to fiber tracts. Building on reports of brain-wide asymmetry and crossing-fiber inflation, this work tests each assumption with HCP multi-shell data, simulations, and 7T vein imaging, showing the index is heavily confounded by axonal geometry.

Key results

  1. ShownRadial diffusivity asymmetry (λ2/λ3), which the ALPS-index mathematically reduces to and strongly correlates with, appears in every white matter region examined rather than only ALPS regions of interest (ROIs), is spatially coherent rather than noise-driven, and persists at high b-values.

    In a representative HCP subject, λ2/λ3 was often above 1.5-2 voxel-wise; in HCP-YA (N=105), most of 35 JHU ROIs plus the 4 ALPS ROIs averaged ~1.3-1.8, above noise expectations, with coherent V2 orientation, at all b-values up to 3000 s/mm2; similar in HCP-A (N=50). A supplementary 10% PVS compartment model had its asymmetric signal effectively nulled at b = 3000 s/mm^2. · There is widespread radial asymmetry in…; There is widespread radial asymmetry in…; Importantly, the ALPS-index itself is…; Perivascular space size and shape and diffusion…

  2. ShownFiber crossings inflate radial asymmetry: simulations reproduce the observed values, ALPS ROIs are dominated by crossing-fiber voxels, and asymmetry rises with crossing prevalence and angle, although it persists in single-fiber voxels.

    Two-population simulations (angles 0-90°, fractions 0-0.5, 1000 voxels per configuration, SNR~20-30) gave λ2/λ3 frequently above 2 for orthogonal, high-fraction crossings. Most JHU ROIs other than corpus callosum segments had very low single fiber fraction, with crossing angles mostly 60-90 degrees; in ALPS ROIs asymmetry rose as single fiber fraction fell and crossing angle grew. · Simulations; Single and Crossing Fibers; Most white matter regions contain crossing…; Crossing fibers contribute to radial asymmetry

  3. SuggestedAxonal undulation and anisotropic orientation dispersion can generate asymmetry of the observed magnitude, and NODDI-derived dispersion anisotropy covaries with radial asymmetry even within single-fiber voxels.

    Monte Carlo microscale (A=4um, L=24um) and macroscale (A=50um, L=300um) undulations gave λ2/λ3 of 1.36-1.48 and 2.06-2.43, respectively, varying with b-value. BIC favored or matched Bingham over Watson NODDI across white matter; in single-fiber voxels DAB was ~0.3-0.5 in most ROIs, ALPS ROIs included, and higher DAB accompanied higher asymmetry. · Simulations; Modeling anisotropic orientation dispersion; Most white matter regions contain anisotropic…; Dispersion and Undulation contribute to radial…

  4. SuggestedMedullary veins in ALPS regions run mostly left-right and near-orthogonal to the principal fiber direction, but their orientation is heterogeneous voxel to voxel and the secondary eigenvector does not consistently align with them.

    In N=4 subjects with 7T multi-echo SWI (super-resolved to 0.2 mm isotropic) and 3T multi-shell DWI, veins in ALPS regions sat typically 80-90 degrees from V1 but deviated up to ~30 degrees from left-right; few adequately sized regions stayed consistently <10 degrees from RL, and V2 did not reliably follow vein direction. · Geometrical configuration of vasculature and…; Data Acquisition; Investigating PVS and WM geometrical assumptions…; Orientation of PVS and WM

  5. SuggestedGlobal radial asymmetry declines with age alongside age-related shifts in fiber geometry, implying that reported ALPS-index aging effects may partly track axonal architecture.

    Linear regressions of each feature on age (results displayed only when FDR-corrected p<0.05): whole-white-matter radial asymmetry, crossing angle and DAB were negatively age-associated, single fiber fraction positively; regional trends varied across commissural, association and projection pathways. No coefficients are given in the text. · Asymmetry and Age; Radial asymmetry shows age-related changes, and…; Age-related axonal changes

Methods, data and reproducibility

Per-b-value DTI eigenvalue ratios (λ2/λ3) in HCP JHU and ALPS ROIs were related to crossing-fiber metrics from multi-shell multi-tissue spherical deconvolution, Monte Carlo undulation and multi-tensor crossing simulations, and Bingham versus Watson NODDI fits; vein orientations from 7T susceptibility-weighted imaging (SWI) were compared with DTI eigenvectors.

  • Human Connectome Project Young Adult (HCP-YA) diffusion MRI subset, ages 21 to 35; b = 1000, 2000, 3000 s/mm2, 90 directions each, 1.25 mm voxels · human · n = N=105
  • Human Connectome Project Aging (HCP-A) diffusion MRI subset, ages 35 to 90; b = 1500 and 3000 s/mm2, 1.5 mm voxels · human · n = N=50
  • Vanderbilt University Medical Center cohort imaged with 7T multi-echo SWI for veins and 3T multi-shell DWI (b = 500 to 3,000 s/mm2) for fiber orientation · human · n = N=4
  • Simulated two-fiber voxels: crossing angle 0-90° (1° steps), crossing fraction 0-0.5 (0.01 steps), Rician noise at SNR~20-30 · n = 1000 per geometric configuration
  • Monte Carlo random-walk simulations of sinusoidal axons, microscale (A=4um, L=24um) and macroscale (A=50um, L=300um), using the HCP encoding protocol
Methods & tools
MRtrix3 (weighted least-squares tensor fit, dwi2response dhollander, msmt_csd, sh2peaks), FSL v6.0.7.9 TOPUP and EDDY, HCP minimal preprocessing pipeline, NODDI with Watson and Bingham orientation distributions, Bayesian Information Criterion model comparison, JHU white matter atlas (35 ROIs) and JHU-ICBM-FA template, ALPS-specific ROIs (4 spheres, 5 mm diameter), Monte Carlo diffusion simulation of undulating axons, SMORE deep learning super-resolution, MPPCA and Patch2Self denoising, PreQual preprocessing pipeline, Adaptive non-local means and Frangi vesselness filters
Reagents & environment
FSL v6.0.7.9 (TOPUP, EDDY) within the HCP minimal preprocessing pipeline; MRtrix3, NODDI, SMORE and PreQual used without stated versions.
Reproducibility gaps
The only URL is a third-party ALPS ROI repository; no code for the authors' simulations or analyses is linked; availability of the Vanderbilt 7T/3T dataset (N=4) is not stated; PVS simulation details are in supplementary material.
Limitations (6)
  • Only two cross-sectional, high-quality research datasets were analyzed, unlike typical clinical 1.5T or 3T acquisitions; the authors argue crossing-fiber confounds would grow with the larger voxels of clinical scans. · stated by the authors
  • Only crossing fibers and dispersion/undulation were examined; larger-scale fanning, branching, and sheet-like fiber arrangements could also create radial asymmetry but were left unexamined. · stated by the authors
  • Assigning dispersion anisotropy to axons relies on NODDI model assumptions, and truly single-fiber voxels are difficult to isolate, so residual asymmetry in single-fiber voxels may partly reflect unresolved fiber populations. · stated by the authors
  • The supplementary PVS sensitivity simulation is simplified and not claimed to capture true compartment diffusivities, and PVS dimensions came from mouse pial artery measurements that may miss smaller spaces. · stated by the authors
  • Vein-fiber orientation comparisons rest on N=4 subjects and are reported mainly descriptively, without cross-subject summary statistics or tests of how much vein misalignment shifts ALPS values. · noted in this summary
  • No independent glymphatic measure (such as tracer or contrast-based clearance imaging) was acquired, so the study demonstrates confounding but cannot estimate how much ALPS variance, if any, reflects perivascular diffusion. · noted in this summary

Read this if you use or review DTI-ALPS as a glymphatic biomarker and need evidence on how crossing fibers, axonal dispersion, and vein orientation bias the index.

Citations
69
FWCI
39.60
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISWright '26Sacchi '26Sepehrband '19Ollen‐Bittl '25Zarkali '25Chen '25
81% semantic matchNeuroscience
David K. Wright·Imaging Neuroscience·Jan 1, 2026· 1 citations