henhaobio
Sign in

bioRxivNeurosciencePreprint

Anisotropic conductivity modeling for tDCS in Parkinsons disease using multidimensional diffusion MRI

Santiago Osorio JuradoiD, Mikael SkorpiliD, Per SvenningssoniD, Rodrigo Moreno AiD, Christoffer OlssoniD

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

Abstract

Transcranial direct current stimulation (tDCS) dose depends on how brain conductivity is modeled. White matter anisotropy is conventionally estimated from single-shell diffusion tensor imaging (DTI). Multidimensional diffusion MRI (MD-dMRI), specifically q-space trajectory imaging (QTI), instead gives a mean tensor expected to carry less kurtosis bias. Our primary question was whether replacing the conventional single-shell tensor with this mean tensor would change the predicted field. We built, to our knowledge, the first MD-dMRI tDCS conductivity model and compared it against DTI and isotropic models in 29 participants (12 with Parkinsons disease, 17 controls) across four montages, with the same mesh, electrodes, and solver. The three models agreed within a few percent. The two anisotropic models differed mainly in tensor orientation (about 21 degrees in white matter), with small differences in field magnitude. Field did not differ between patients and controls in any region or montage (which was an exploratory, underpowered comparison). Whole-brain electric field correlated with MR elastography stiffness (partial r = +0.58) but attenuated to non-significance once cerebrospinal fluid morphology was accounted for (r = +0.06 to +0.09). With no ground-truth field or conductivity available, the study establishes the feasibility of the MD-dMRI model and characterizes field sensitivity rather than improved dosimetry accuracy. The choice of diffusion tensor is second order for dose, which is primarily influenced by individual anatomy. For Parkinsons disease, modeling efforts should focus on cerebrospinal fluid- and atrophy-aware head models and dose normalization, rather than a more complex diffusion tensor. HighlightsO_LIIndividual anatomy, more than the conductivity tensor, governs tDCS dose. C_LIO_LIA first tDCS head model from multidimensional diffusion MRI (QTI). C_LIO_LIMD-dMRI, single-shell DTI and isotropic fields agreed within a few percent. C_LIO_LIThe anisotropic models differ mainly in orientation; their fields agree closely. C_LI

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (20 Aug 2026) · CC BY 4.0

tDCS fields from a multidimensional diffusion MRI conductivity model stayed within a few percent of DTI and isotropic predictions in 29 participants; whole-brain field instead tracked CSF fraction.

Transcranial direct current stimulation (tDCS) dose in Parkinson's disease (PD) depends on an unmeasurable electric field, and white-matter conductivity is usually derived from single-shell diffusion tensor imaging (DTI), which the paper calls biased by non-Gaussian diffusion. The authors say this is the first tDCS conductivity model built from a q-space trajectory imaging (QTI) mean tensor; they test whether this tensor, expected to be less kurtosis-biased, changes the predicted dose.

Key results

  1. ShownReplacing the single-shell DTI tensor with the QTI mean tensor changed the 95th-percentile field magnitude (p95 |E|) by only a few percent. The sign depended on tissue: QTI was slightly higher in the cortical lobes and lower in the white-matter lobes, corpus callosum and pons.

    Primary motor cortex (M1) pad montage, n = 29: the within-subject p95 |E| difference was significant in 10 of 11 primary regions, all except the mesencephalon (Wilcoxon, Benjamini-Hochberg q < 0.05). The sign pattern held in all four montages except HD-DLPFC occipital cortex; 9 of 11 regions were significant in each other montage. The magnitude is confounded by acquisition and smoothing. · 3.2 Conductivity model comparison (MD-dMRI…; 4.4 Limitations

  2. ShownThe isotropic, DTI and QTI conductivity models gave nearly the same predicted field. Differences between models were small compared with both field magnitude and the spread between subjects.

    M1 montage, n = 29: cortical-lobe models were within a few percent of each other. In white-matter and deep regions, the anisotropic models sat slightly above isotropic, with QTI below DTI; differences stayed small relative to field magnitude and the between-subject interquartile spread. Representative subject: median absolute model difference on the gray-matter surface 0.0026 V/m (1.8% of median |E|). · 3.3 Model difference relative to field magnitude

  3. ShownThe two anisotropic tensors differed mainly in principal-axis orientation, and sharing a single registration did not remove this difference. Even so, the size of the angle did not predict how much the field changed between models.

    Median principal-eigenvector angle: about 21 degrees in the white-matter lobes, 13 to 18 degrees in the corpus callosum and brainstem, and about 45 degrees in gray matter (60 degrees is chance for random axes). With a shared registration: 22 vs 21 degrees. Within-region correlation between the angle and the change in p95 |E|: r = -0.05 (95% subject-clustered bootstrap CI [-0.20, +0.12]; 29 subjects, 7 regions). · 3.5 Orientation divergence between the two…

  4. SuggestedPredicted whole-brain field correlated positively with magnetic resonance elastography (MRE) stiffness, but the association disappeared once CSF volume was controlled for. This fits a shared morphology effect rather than stiffness acting on the field.

    Partial r = +0.58 (q = 0.002) controlling for group; r = +0.579 to +0.592 across the three conductivity models; r = +0.48 (q = 0.018) after adding age; r = +0.09 (q = 0.66) controlling for CSF volume; r = +0.06 controlling for CSF fraction. Field vs CSF fraction: r = -0.66, and partial r = -0.41 controlling for stiffness. Stiffness and CSF fraction: r = -0.84. n = 29. · 3.6 Association between predicted field and MRE…; 4.2 Individual anatomy, not the conductivity…

  5. SuggestedUnder the QTI model, PD patients and controls did not differ in modeled field in any primary region or montage. The authors treat this comparison as exploratory and underpowered.

    0 of 11 regions per montage differed (age-adjusted residuals, Mann-Whitney, Benjamini-Hochberg). Largest Cohen's d = 0.63; minimum q = 0.48. With 12 PD vs 17 controls, the test had 80% power only for d of about 1.1. · 3.4 Group comparison (PD vs HC); 2.8 Electric field metrics and statistics

Methods, data and reproducibility

For each subject, SimNIBS 4.6 charm built a finite-element head mesh from T1/T2 MRI. tDCS fields were simulated for four montages (5 x 5 cm pads and 4 x 1 high-definition rings over M1 and dorsolateral prefrontal cortex, 2 mA) under isotropic, single-shell DTI and QTI mean-tensor conductivity models, the anisotropic ones via volume-normalized mapping; regional p95 |E| was compared with paired Wilcoxon tests.

  • Parkinson's disease cohort from Olsson et al. (2025): PD patients scanned ON medication and healthy controls, Philips Ingenia CX 3T · human · n = 29 (12 PD, 17 HC)
  • QTI multidimensional diffusion MRI (linear and spherical b-tensor encoding, b = 0 to 2000 s/mm², 2.5 mm isotropic) · human · n = 29
  • Single-shell diffusion MRI (80 directions, b = 1500 s/mm², 2.0 mm isotropic), separate scan in the same session · human · n = 29
  • T1- and T2-weighted structural MRI (1 mm isotropic) used for head segmentation and meshing · human · n = 29
  • MR elastography whole-brain stiffness (|G*|) maps from Olsson et al. (2025) · human · n = 29
Methods & tools
SimNIBS 4.6 (charm, dwi2cond), FSL 6.0.7.22 (dtifit, FLIRT, vecreg, topup, eddy), Synb0-DisCo, MD-dMRI framework (MATLAB R2026a), FreeSurfer 7.2 recon-all, Iglesias brainstem substructures module, CIT168 probabilistic atlas, Python 3.11.14 (NumPy 2.3.0, SciPy 1.17.1)
Software & environment
SimNIBS 4.6.0; FSL 6.0.7.22; FreeSurfer 7.2; MD-dMRI framework in MATLAB R2026a; Python 3.11.14 with NumPy 2.3.0 and SciPy 1.17.1.
Reproducibility gaps
The MRI data are under ethical and legal restrictions and are not public. Researchers who meet the legal and ethical requirements can request anonymized data from the corresponding author.
Limitations (6)
  • The DTI and QTI tensors came from separate scans that differed in resolution (2.0 vs 2.5 mm) and smoothing, so the white-matter magnitude difference is partly a preprocessing artifact. Only the orientation difference is attributed to the tensor estimate. · stated by the authors
  • Smoothing (Gaussian, σ = 0.7 voxels) made the covariance fit stable but reduces QTI white-matter anisotropy. The unsmoothed fit gave noise-driven anisotropy and failed positive-definiteness in roughly one voxel in five. Neither setting gives an unbiased magnitude. · stated by the authors
  • The orientation divergence may partly reflect angular sampling: 80 directions for DTI vs 6 to 21 per shell in the QTI linear encoding. Neither tensor's orientation is shown to be more accurate. · stated by the authors
  • The cohort is small and mostly male (12 PD, 17 controls). This limits sex-stratified generalization, and the group comparison had 80% power only for Cohen's d of about 1.1. · stated by the authors
  • No measured field or conductivity is available to validate the simulations. The link with stiffness comes from a single time point, and CSF volume, age and stiffness are intercorrelated. A shared morphological driver therefore fits the data, but a mediation pathway is not proven. · stated by the authors
  • Scalar conductivities were fixed at SimNIBS 4.6 defaults (white matter 0.126, gray matter 0.275 S/m) and only the volume-normalized mapping was used. The paper's own discussion names scalar conductivity uncertainty as dominant, but the study does not compare against it. · noted in this summary

Read this if you build personalized tDCS or other transcranial electrical stimulation head models and want to know whether tensor-valued diffusion encoding is worth adding over standard single-shell DTI.

Citations
0

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
THISSalvador '22Laakso '18Caulfield '19Nandi '22Rawji '17Johnstone '21
72% semantic matchNeuroscience
Ricardo Salvador, Maria Chiara Biagi, Manuel Pérez Pelegrí+5 more·bioRxiv·Jan 22, 2022· 5 citations