pith. sign in

arxiv: 2603.11344 · v2 · submitted 2026-03-11 · 📡 eess.IV · q-bio.QM

Hybrid eTFCE-GRF: Exact Cluster-Size Retrieval with Analytical p-Values for Voxel-Based Morphometry

Don Yin , Hao Chen , Takeshi Miki , Enyu Yang This is my paper

Pith reviewed 2026-05-15 12:22 UTC · model grok-4.3

classification 📡 eess.IV q-bio.QM
keywords TFCEvoxel-based morphometrycluster enhancementGaussian random fieldsunion-findanalytical p-valuesneuroimagingpermutation-free
0
0 comments X

The pith

Union-find structure plus Gaussian random field theory yields exact cluster sizes and analytical p-values without permutations for voxel-based morphometry.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a hybrid statistical method for voxel-based morphometry that retrieves exact cluster sizes at any threshold using a union-find structure built in one pass over sorted voxels. It then applies closed-form Gaussian random field expressions to obtain analytical p-values directly, bypassing both threshold discretization and permutation testing. This approach controls family-wise error rates at nominal levels on synthetic data while achieving high concordance with reference methods and completing analyses orders of magnitude faster on real datasets like the UK Biobank. Sympathetic readers would care because it makes precise cluster-enhanced inference feasible for large-scale brain imaging studies without the computational burden of permutations.

Core claim

By merging the union-find hierarchy from exact TFCE with the analytical GRF p-value formulas from probabilistic TFCE, the hybrid method computes exact cluster sizes for continuous thresholds in a single pass and converts them to p-values without requiring permutations or discrete grids.

What carries the argument

Union-find data structure for building cluster hierarchy in one pass over sorted voxels, paired with GRF theory to convert exact sizes to analytical p-values.

If this is right

  • Family-wise error rate is controlled at the nominal level, with zero rejections out of 200 null cases on synthetic phantoms.
  • Statistical power matches baseline pTFCE with Dice overlap of at least 0.999 and concordance correlation above 0.99.
  • Significance maps on UK Biobank and IXI data form strict conservative subsets of reference pTFCE maps.
  • Whole-brain VBM analysis completes in approximately 85 seconds for the hybrid version, over 1000 times faster than permutation TFCE.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The single-pass exact-size retrieval could scale to much larger cohorts or higher-resolution scans while preserving precision.
  • The approach might be tested for robustness under non-Gaussian smoothness or extended to other modalities such as fMRI.
  • Direct comparison of cluster-size distributions at finely spaced thresholds would further confirm absence of discretization artifacts.

Load-bearing premise

Exact cluster sizes obtained from the union-find data structure at continuous thresholds remain compatible with the closed-form GRF p-value expressions without introducing bias.

What would settle it

Observing a family-wise error rate exceeding the nominal 5 percent level, such as more than 1.9 percent false positives across 200 null simulations on 64^3 synthetic phantoms, would indicate the sizes are incompatible with the GRF formulas.

Figures

Figures reproduced from arXiv: 2603.11344 by Don Yin, Enyu Yang, Hao Chen, Takeshi Miki.

Figure 1
Figure 1. Figure 1: Phantom specification used in the Monte Carlo validation. Three non [PITH_FULL_IMAGE:figures/full_fig_p025_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Demographic distributions showing the joint distribution of age and sex across [PITH_FULL_IMAGE:figures/full_fig_p026_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Null FWER calibration across 200 independent null realisations ( [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) Power curves showing Dice coefficient between detected and true signal [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Smoothness estimation validation across 50 null realisations. The estimated [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Cross-variant concordance of enhanced Z-maps. (a) Phantom data: scatter plot of hybrid versus baseline pTFCE enhanced Z-values for five matched-seed realisations (r = 0.992). (b) IXI site-effect map: pairwise scatter plots for the three pTFCE variants, with Pearson r annotated for each pair. 29 [PITH_FULL_IMAGE:figures/full_fig_p030_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Real brain pTFCE significance maps (Python baseline): raw test statistics [PITH_FULL_IMAGE:figures/full_fig_p031_7.png] view at source ↗
read the original abstract

Threshold-free cluster enhancement (TFCE) integrates cluster extent across thresholds to improve voxel-wise neuroimaging inference, but permutation testing makes it prohibitively slow for large datasets. Probabilistic TFCE (pTFCE) uses analytical Gaussian random field (GRF) p-values but discretises the threshold grid. Exact TFCE (eTFCE) eliminates discretisation via a union-find data structure but still requires permutations. We combine eTFCE's union-find for exact cluster-size retrieval with pTFCE's analytical GRF inference. The union-find builds the cluster hierarchy in one pass over sorted voxels and enables exact size queries at any threshold; GRF theory then converts these sizes to analytical p-values without permutations. Validation on synthetic phantoms (64^3, 80 subjects): FWER controlled at nominal level (0/200 null rejections, 95% CI [0.0%, 1.9%]); power matches baseline pTFCE (Dice >= 0.999); smoothness error below 1%; concordance r > 0.99. On UK Biobank (N=500) and IXI (N=563), significance maps form strict subsets of reference R pTFCE, which supports conservative error control. Implemented in pytfce (pip install pytfce): baseline completes whole-brain VBM in ~5s (75x faster than R pTFCE), hybrid in ~85s (4.6x faster) with exact cluster sizes; both >1000x faster than permutation TFCE.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper proposes Hybrid eTFCE-GRF, which combines the union-find data structure from exact TFCE (eTFCE) to retrieve exact cluster sizes across any threshold with the analytical Gaussian random field (GRF) p-value formulas from probabilistic TFCE (pTFCE). This yields permutation-free, threshold-free cluster enhancement for voxel-based morphometry (VBM) that is substantially faster than prior methods while claiming exact FWER control.

Significance. If the central compatibility assumption holds, the work provides a practical advance for large-scale neuroimaging by reducing whole-brain VBM runtime from hours (permutation TFCE) to under two minutes while preserving power and delivering conservative maps on real data (UK Biobank N=500, IXI N=563). The open-source pytfce implementation and synthetic validation (exact FWER at 0/200 null cases, Dice >=0.999) are concrete strengths that would make the method immediately usable.

major comments (1)
  1. [Methods] Methods section on the union-find-to-GRF interface: the manuscript does not supply a derivation or error-propagation analysis showing that exact cluster sizes obtained from the union-find at continuous thresholds remain statistically compatible with the closed-form GRF expressions originally derived under discretized or permutation settings; this assumption is load-bearing for the claim of unbiased analytical p-values.
minor comments (2)
  1. [Abstract] Abstract: the statement 'smoothness error below 1%' is reported without specifying the smoothness estimator or the precise metric, making the claim difficult to interpret or reproduce.
  2. [Results] Results: the real-data concordance (r > 0.99) is given only as a scalar; reporting the spatial distribution of differences or the fraction of voxels that differ in significance would strengthen the conservative-map claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the single major comment below and commit to the corresponding revision.

read point-by-point responses

  1. Referee: [Methods] Methods section on the union-find-to-GRF interface: the manuscript does not supply a derivation or error-propagation analysis showing that exact cluster sizes obtained from the union-find at continuous thresholds remain statistically compatible with the closed-form GRF expressions originally derived under discretized or permutation settings; this assumption is load-bearing for the claim of unbiased analytical p-values.

    Authors: We agree that an explicit derivation is missing and that the compatibility assumption is central. The union-find computes the exact Lebesgue measure of the excursion set at any continuous threshold by merging connected components in sorted order, eliminating the discretization grid used in pTFCE. Because the GRF cluster-size tail probabilities depend only on the observed extent, the estimated smoothness, and the field dimension (not on the discretization method), substituting the exact size into the same closed-form expressions preserves validity under the identical null-field assumptions. In the revision we will insert a short Methods subsection that (i) recalls the GRF derivation, (ii) shows that the union-find size is the precise input required by those formulas, and (iii) provides a first-order error-propagation bound demonstrating that removal of discretization error can only reduce, not increase, bias relative to pTFCE. The existing synthetic null simulations (0/200 rejections) already supply empirical corroboration; the added derivation will make the theoretical claim explicit. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the hybrid derivation

full rationale

The paper presents a hybrid of two pre-existing components: eTFCE's union-find structure for exact cluster-size retrieval (one-pass hierarchy over sorted voxels) and pTFCE's closed-form GRF analytical p-values. The central step is the claim that exact sizes at continuous thresholds remain compatible with GRF expressions; this is treated as an empirical compatibility question and is supported by synthetic FWER control (0/200 null rejections), Dice equivalence, and real-data concordance (r > 0.99) rather than by algebraic reduction or self-reference. No equations are shown that define the output in terms of itself, no parameters are fitted on a subset and then relabeled as prediction, and no load-bearing uniqueness theorems or ansatzes are imported via self-citation. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method relies on standard GRF theory assumptions and the union-find algorithm without introducing new free parameters or invented entities in the described approach.

axioms (1)
  • domain assumption Gaussian random field theory assumptions hold for converting exact cluster sizes to analytical p-values
    Invoked to replace permutation testing with closed-form p-value computation from cluster sizes.

pith-pipeline@v0.9.0 · 5593 in / 1294 out tokens · 33759 ms · 2026-05-15T12:22:44.366794+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages · 1 internal anchor

  1. [1]

    S. M. Smith, T. E. Nichols, Threshold-free cluster enhancement: Ad- dressing problems of smoothing, threshold dependence and localisation in cluster inference, NeuroImage 44 (1) (2009) 83–98.doi:10.1016/j. neuroimage.2008.03.061

    work page doi:10.1016/j 2009

  2. [2]

    K. J. Friston, K. J. Worsley, R. S. J. Frackowiak, J. C. Mazziotta, A. C. Evans, Assessing the significance of focal activations using their spatial extent, Human Brain Mapping 1 (3) (1994) 210–220.doi:10.1002/ hbm.460010306

    work page 1994

  3. [3]

    Ashburner, K

    J. Ashburner, K. J. Friston, Voxel-based morphometry—the methods, NeuroImage 11 (6) (2000) 805–821.doi:10.1006/nimg.2000.0582. 31 Figure S2: IXI site-level mean (left) and standard deviation (right) of registered T1- weighted intensity maps. The IOP site (GE 1.5T) shows higher mean intensity and inter- subject variability compared to the two Philips sites...

    work page doi:10.1006/nimg.2000.0582 2000

  4. [4]

    K. L. Miller, F. Alfaro-Almagro, N. K. Bangerter, D. L. Thomas, E. Ya- coub, J. Xu, A. J. Bartsch, S. Jbabdi, S. N. Sotiropoulos, J. L. R. An- dersson, et al., Multimodal population brain imaging in the UK Biobank prospective epidemiological study, Nature Neuroscience 19 (11) (2016) 1523–1536.doi:10.1038/nn.4393

    work page doi:10.1038/nn.4393 2016

  5. [5]

    Spisák, Z

    T. Spisák, Z. Spisák, M. Zunhammer, U. Bingel, S. Smith, T. Nichols, T. Kincses, Probabilistic TFCE: A generalized combination of cluster size and voxel intensity to increase statistical power, NeuroImage 185 (2019) 12–26.doi:10.1016/j.neuroimage.2018.09.078

    work page doi:10.1016/j.neuroimage.2018.09.078 2019

  6. [6]

    X. Chen, W. Weeda, T. E. Nichols, J. J. Goeman, eTFCE: Exact Threshold-Free Cluster Enhancement via Fast Cluster Retrieval (2026). arXiv:2603.03004. URLhttps://arxiv.org/abs/2603.03004

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  7. [7]

    D. Yin, H. Chen, pytfce: Fast probabilistic Threshold-Free Cluster En- hancement in Python (2026). URLhttps://github.com/Don-Yin/pytfce

    work page 2026

  8. [8]

    E. T. Bullmore, J. Suckling, S. Overmeyer, S. Rabe-Hesketh, E. Taylor, M. J. Brammer, Global, voxel, and cluster tests, by theory and permu- tation, for a difference between two groups of structural MR images of the brain, IEEE Transactions on Medical Imaging 18 (1) (1999) 32–42. doi:10.1109/42.750253

    work page doi:10.1109/42.750253 1999

  9. [9]

    A. M. Winkler, G. R. Ridgway, M. A. Webster, S. M. Smith, T. E. Nichols, Permutation inference for the general linear model, NeuroImage 92 (2014) 381–397.doi:10.1016/j.neuroimage.2014.01.060

    work page doi:10.1016/j.neuroimage.2014.01.060 2014

  10. [10]

    K. J. Worsley, S. Marrett, P. Neelin, A. C. Vandal, K. J. Friston, A. C. Evans, A unified statistical approach for determining significant signals in images of cerebral activation, Human Brain Mapping 4 (1) (1996) 58–73.doi:10.1002/(SICI)1097-0193(1996)4:1<58::AID-HBM4>3. 0.CO;2-O

    work page doi:10.1002/(sici)1097-0193(1996)4:1 1996

  11. [11]

    V. P. Nosko, Local structure of Gaussian random fields in the vicinity of high-level shines, Soviet Mathematics: Doklady 10 (1969) 1481–1484. 38

    work page 1969

  12. [12]

    R. E. Tarjan, Efficiency of a good but not linear set union algorithm, Journal of the ACM 22 (2) (1975) 215–225.doi:10.1145/321879. 321884

    work page doi:10.1145/321879 1975

  13. [13]

    K. J. Worsley, A. C. Evans, S. Marrett, P. Neelin, A three-dimensional statistical analysis for CBF activation studies in human brain, Journal of Cerebral Blood Flow & Metabolism 12 (6) (1992) 900–918.doi: 10.1038/jcbfm.1992.127

    work page doi:10.1038/jcbfm.1992.127 1992

  14. [14]

    S. J. Kiebel, J.-B. Poline, K. J. Friston, A. P. Holmes, K. J. Wors- ley, Robust smoothness estimation in statistical parametric maps using standardized residuals from the general linear model, NeuroImage 10 (6) (1999) 756–766.doi:10.1006/nimg.1999.0508

    work page doi:10.1006/nimg.1999.0508 1999

  15. [15]

    Alfaro-Almagro, M

    F. Alfaro-Almagro, M. Jenkinson, N. K. Bangerter, J. L. R. Andersson, L. Griffanti, G. Douaud, S. N. Sotiropoulos, S. Jbabdi, M. Hernandez- Fernandez, E. Vallee, et al., Image processing and quality control for the first 10,000 brain imaging datasets from UK Biobank, NeuroImage 166 (2018) 400–424.doi:10.1016/j.neuroimage.2017.10.034

    work page doi:10.1016/j.neuroimage.2017.10.034 2018

  16. [16]

    S. M. Smith, Fast robust automated brain extraction, Human Brain Mapping 17 (3) (2002) 143–155.doi:10.1002/hbm.10062

    work page doi:10.1002/hbm.10062 2002

  17. [17]

    Jenkinson, S

    M. Jenkinson, S. Smith, A global optimisation method for robust affine registration of brain images, Medical Image Analysis 5 (2) (2001) 143– 156.doi:10.1016/S1361-8415(01)00036-6

    work page doi:10.1016/s1361-8415(01)00036-6 2001

  18. [18]

    Jenkinson, P

    M. Jenkinson, P. Bannister, M. Brady, S. Smith, Improved optimization for the robust and accurate linear registration and motion correction of brain images, NeuroImage 17 (2) (2002) 825–841.doi:10.1006/nimg. 2002.1132

    work page doi:10.1006/nimg 2002

  19. [19]

    J. L. R. Andersson, M. Jenkinson, S. Smith, Non-linear registration, aka spatial normalisation, FMRIB Technical Report TR07JA2 (2007). URLhttps://www.fmrib.ox.ac.uk/datasets/techrep/tr07ja2/ tr07ja2.pdf

    work page 2007

  20. [20]

    A., Hekker, S., Stello, D., Guti ´errez-Soto, J., Handberg, R., Huber, D., et al

    Y. Benjamini, Y. Hochberg, Controlling the false discovery rate: A prac- tical and powerful approach to multiple testing, Journal of the Royal Statistical Society: Series B 57 (1) (1995) 289–300.doi:10.1111/j. 2517-6161.1995.tb02031.x. 39

    work page doi:10.1111/j 1995

  21. [21]

    C. D. Good, I. S. Johnsrude, J. Ashburner, R. N. A. Henson, K. J. Fris- ton, R. S. J. Frackowiak, A voxel-based morphometric study of ageing in 465 normal adult human brains, NeuroImage 14 (1) (2001) 21–36. doi:10.1006/nimg.2001.0786

    work page doi:10.1006/nimg.2001.0786 2001

  22. [22]

    Jenkinson, C

    M. Jenkinson, C. F. Beckmann, T. E. J. Behrens, M. W. Woolrich, S. M. Smith, FSL, NeuroImage 62 (2) (2012) 782–790.doi:10.1016/ j.neuroimage.2011.09.015

    work page 2012

  23. [23]

    Eklund, T

    A. Eklund, T. E. Nichols, H. Knutsson, Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates, Proceed- ings of the National Academy of Sciences 113 (28) (2016) 7900–7905. doi:10.1073/pnas.1602413113. 40

    work page doi:10.1073/pnas.1602413113 2016