A Voxel-Spacing-Aware Extension of PyRadiomics for Anisotropic Texture Analysis
David Corral Fontecha, Juan Miranda Bautista, Pablo Menendez Fernández-Miranda, Andrea Trapote Fernandez, Lara Lloret Iglesias, Jose A. Vega
cs.CV · cs.AI
Quick verdict
Presents a spacing-aware PyRadiomics extension avoiding interpolation; sound synthetic-phantom software validation only, code not yet publicly released.
Summary
This method paper describes a voxel-spacing-aware extension of PyRadiomics that incorporates physical voxel geometry into texture-feature extraction for anisotropic CT/MRI without interpolating gray levels. Activated opt-in via weightingNorm='voxelSpacing', the framework modifies feature families selectively: GLCM receives anisotropy-relative feature-level angular weighting, NGTDM uses weighted neighborhood averaging, and GLRLM/GLDM/GLSZM are computed on a finite-volume zero-order-hold lattice built by replicating voxels along anisotropic axes. Validation uses deterministic synthetic phantoms: the modified code reproduces stock PyRadiomics exactly when disabled (75/75 features, zero difference), is exactly equivalent under isotropic spacing, and under anisotropic spacing changes features with family medians of 10.1% (GLCM) to 49.1% (GLDM). Against 2:1:1, 3:1:1, and 5:1:1 phantoms (225 comparisons), the method altered 192/225 (85.3%) — the same count as nearest-neighbor resampling — with median differences of similar magnitude to linear/B-spline resampling but distinct values. Profiling across expansion factors R=1–7 shows runtimes under 1.75 s and peak memory up to 90.55 MB; non-integer spacing ratios incur rounding errors of −20.0% to +14.3%. Main limitations: no clinical data, no IBSI conformance testing of the new mode, code not yet publicly released, and clinical evaluation deferred to a companion preprint by the same authors.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
4 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- The modified implementation reproduces standard PyRadiomics exactly when the spacing-aware mode is disabled and is exactly equivalent under isotropic spacing across all 75 texture features.
- Under anisotropic spacing, the extension selectively modifies texture families (all GLCM/GLRLM/NGTDM features, 11/14 GLDM, 8/16 GLSZM) with median relative differences of 10.1–49.1%.
- The spacing-aware outputs are numerically distinct from nearest-neighbor, linear, and B-spline resampling while never generating interpolated gray levels.
- Computational overhead is moderate within tested settings: runtime below 1.75 s and peak memory up to 90.55 MB at the largest tested configuration (R=7).
- The framework is a backward-compatible technical basis for spacing-aware radiomics and makes no claim of IBSI conformance for the new mode.
Strengths
Exact backward-compatibility verification
The modified overlay passes 3568/3568 upstream PyRadiomics tests and reproduces stock output exactly (mean/max absolute difference 0/0) across all 75 texture features with the mode disabled, plus exact isotropic equivalence (§3.1–3.2, Table 2).
Feature-family-specific design with mathematical justification
GLCM weighting is applied only at feature-level angular aggregation (Eq. 11) because Haralick features are nonlinear in the co-occurrence matrix, preserving directional distributions; NGTDM embeds weights in the neighborhood mean (Eq. 14); GLRLM/GLDM/GLSZM keep integer matrix axes on a zero-order-hold lattice (Eqs. 6–10, Table 1).
Approximations are quantified, not hidden
Honest scope discipline
Weaknesses
Synthetic-only, under-specified validation data
Code not publicly available
Descriptive statistics only, with an unresolved NN-resampling comparison
Design choices argued analytically but never ablated empirically
Filter-type adjustments unvalidated
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a software/method paper presenting a voxel-spacing-aware extension of PyRadiomics, implemented across the Python frontend, C extension wrapper (_cmatrices.c), and computational backend (cmatrices.c). Its claims, each tied to reported evidence: (1) exact backward compatibility with stock PyRadiomics when the mode is off (§3.1–3.2: 3568/3568 upstream tests; 75/75 features with 0/0 mean/max absolute difference); (2) exact equivalence under isotropic spacing; (3) selective, family-specific activation under anisotropic spacing (family medians 10.1% GLCM to 49.1% GLDM, §3.2); (4) numerical distinctness from nearest-neighbor, linear, and B-spline resampling without interpolated gray levels (225 comparisons, §3.2); and (5) moderate cost (runtime <1.75 s, peak 90.55 MB at R=7, Table 2). The authors scope the work carefully: they explicitly disclaim IBSI conformance of the new mode (§2.4, §4) and defer all clinical evaluation to ref. [10]. Within those boundaries the claims are supported; the boundaries themselves are the review issue.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
Does this PyRadiomics extension replace isotropic resampling?
It offers an alternative rather than a replacement: voxel spacing is incorporated during feature computation (directional weights for GLCM/NGTDM, a replicated lattice for GLRLM/GLDM/GLSZM), so gray levels are never interpolated. The paper shows it changes features by magnitudes similar to resampling but with different values, and does not claim superiority — clinical comparison is deferred to a companion study.
Is the spacing-aware mode IBSI-compliant?
No. The paper explicitly states the spacing-aware outputs are methodological extensions, not IBSI reference values, and no independent IBSI benchmark of the new mode was run. With the mode off, the library matches standard PyRadiomics and passes its 3568-test regression suite.
Can I download the code?
Not yet at publication time. The paper says the modifications are being prepared for release and intended as a contribution to the PyRadiomics project; meanwhile source is available from the corresponding author upon reasonable request.
What happens with non-integer spacing ratios such as 2.5:1?
The finite-volume lattice needs integer repeat factors, so 2.5:1:1 is represented as 2:1:1 (a −20.0% through-plane representation error) and 3.5:1:1 as 4:1:1 (+14.3%). These errors are logged, warned above 10%, and quantified in the sensitivity analysis; integer ratios incur no error.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure 1 is extracted as garbled schematic text; its visual layout could not be assessed.
- Tables 1–2 appear as inline text; row/column alignment was inferred from surrounding prose.
- The 3568-test figure and upstream-suite coverage are taken from the paper's own report and cannot be independently verified from the text.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.