Pith. sign in

REVIEW 3 major objections 22 references

Small intrinsic C-arm calibration errors change vertebral DRR appearance and hurt 2D–3D registration even when anatomy and pose stay fixed.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 10:52 UTC pith:TV4MMS67

load-bearing objection Useful controlled synthetic demo that intrinsic K errors move vertebral DRRs and hurt landmark registration (esp. rotation), but LAT>AP is confounded by non-isocentric magnification and the numbers do not fully agree with themselves. the 3 major comments →

arxiv 2607.10551 v1 pith:TV4MMS67 submitted 2026-07-12 eess.IV cs.CVcs.LGcs.NAmath.NAphysics.comp-ph

Projection-Domain Sensitivity Analysis of Vertebral DRRs Under Intrinsic Calibration Perturbation

Lin Li , Chaochao Zhou , Benjamin Aubert , Junlin Guo , Junchao Zhu This is my paper
classification eess.IV cs.CVcs.LGcs.NAmath.NAphysics.comp-ph
keywords 2D–3D registrationfluoroscopyC-arm calibrationdigitally reconstructed radiographsvertebral imagingprojection-domain analysiscalibration sensitivity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper asks whether standard reconstruction-based checks of fluoroscopic calibration are enough for spine work that lives in the projection domain—DRR generation and 2D–3D vertebral registration. Using CT-derived vertebrae and a fixed cone-beam pose, the authors generate digitally reconstructed radiographs under ground-truth and deliberately perturbed intrinsic parameters (mainly focal scale and piercing point) and measure what changes on the detector. They find that modest intrinsic errors move landmarks, warp contours, reduce silhouette overlap, and alter DRR intensity, with lateral views far more sensitive than anterior–posterior views. Those same projection inconsistencies then increase landmark-based registration error, especially rotation. The claim is that projection-domain consistency is a necessary complement to reprojection or reconstruction error when judging calibration for fluoroscopy-guided spinal imaging.

Core claim

Relatively small perturbations of intrinsic calibration parameters alone—while anatomy and extrinsic acquisition pose remain unchanged—produce measurable changes in vertebral projection geometry, contour morphology, landmark locations, and DRR appearance; the effect is strongly view-dependent (LAT much worse than AP) and propagates into degraded landmark-based 2D–3D registration accuracy, particularly rotational alignment.

What carries the argument

Controlled synthetic projection-domain sensitivity framework: DRRs from CT-derived vertebral models under fixed extrinsic pose, with ground-truth versus perturbed intrinsic matrix K (focal scale α_f and piercing-point shift Δc), scored by landmark displacement, contour distances, silhouette Dice/IoU, image difference, and landmark-based rigid 2D–3D registration error in AP and LAT.

Load-bearing premise

That clean synthetic non-isocentric cone-beam DRRs of rigid vertebrae, with known landmarks and no scatter, noise, or detector distortion, adequately represent the calibration risk that real fluoroscopy-guided spine registration would face.

What would settle it

Repeat the same intrinsic focal-length and piercing-point perturbations on real dual-view C-arm images of a vertebral phantom with known ground-truth pose and independent optical or tracked calibration, and check whether LAT still shows substantially larger landmark/contour shift and rotational registration error than AP once object–detector distance is matched.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Calibration protocols for spine fluoroscopy should report projection-domain metrics (landmark shift, contour/silhouette consistency), not only reprojection or reconstruction error.
  • DRR-based 2D–3D registration pipelines should treat intrinsic uncertainty as a source of rotational bias, especially when LAT or multi-view objectives are used.
  • View-dependent sensitivity implies LAT geometry and object–detector distance need tighter intrinsic tolerance than AP for the same registration budget.
  • The synthetic perturbation protocol can be used as a calibration-robustness test bench for DRR generators and registration optimizers before clinical deployment.

Where Pith is reading between the lines

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

  • If LAT sensitivity is driven mainly by larger projected scale in the non-isocentric setup, matching SID/SDD between views might shrink the AP–LAT gap without changing anatomy.
  • Intensity-based registration (NCC/MI/gradient correlation) may be even more brittle than landmark registration under the same intrinsic errors, because the heatmaps already concentrate differences on high-gradient cortical edges.
  • Self-calibration methods that accept multiple K solutions with similar reprojection error may silently select geometries that are reconstruction-acceptable but projection-hostile for vertebral DRRs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 0 minor

Summary. This paper presents a controlled synthetic framework for quantifying how intrinsic C-arm calibration perturbations (primarily focal scale and piercing-point shift) affect vertebral DRR appearance and landmark-based 2D–3D registration when anatomy and extrinsic pose are held fixed. Using CT-derived lumbar vertebrae, cone-beam projection, and complementary metrics (landmark displacement, contour distances, Dice/IoU, raw DRR difference maps, and rigid registration error), the authors report that small intrinsic perturbations produce measurable projection-domain changes, that LAT views are substantially more sensitive than AP views, and that these inconsistencies degrade registration accuracy—especially rotational alignment. They argue that projection-domain consistency is a useful complement to conventional reconstruction-based calibration metrics for fluoroscopy-guided spinal applications.

Significance. If the results hold under clinically relevant conditions, the work fills a genuine gap: most calibration literature evaluates reprojection or reconstruction error, while DRR-based 2D–3D registration depends on projection-domain consistency. The controlled isolation of intrinsic K perturbations with fixed anatomy/pose is a clean experimental design, and the multi-metric reporting (landmarks, silhouettes, registration) is a practical contribution for assessing calibration robustness in vertebral imaging. The finding that reconstruction-preserving intrinsic ambiguity can still matter for registration is useful for image-guided spine workflows. Strengths include explicit disclosure of non-isocentric geometry, a clear perturbation model (Eqs. 7–15), and an honest limitations section. The main significance is methodological and cautionary rather than a new clinical algorithm.

major comments (3)
  1. The central claim that sensitivity is 'strongly view dependent' with LAT substantially more sensitive than AP (Abstract; Contributions 3; §3.7 hypothesis; §5.5; Conclusion) is confounded by the non-isocentric setup. §4.1 and Fig. 2 state that the vertebra is closer to the detector in LAT and therefore appears larger; Discussion explicitly notes that the same intrinsic ΔK then produces larger pixel displacements because the anatomy occupies more detector area. Landmark shifts (Table 1), Dice/IoU (Fig. 4), and registration errors are reported in absolute detector pixels without scale normalization (e.g., by projected vertebral size or object–detector distance) and without a matched-SID/isocentric control. The AP/LAT gap therefore cannot be cleanly attributed to 'overlapping anatomy and compressed depth geometry' alone. Either add scale-normalized metrics and/or an isocentric comparison, or
  2. §5.4 text and Table 1 are numerically inconsistent for the same experiment. The text reports LAT inferior-endplate and left-pedicle shifts of approximately 21.79 px and 21.92 px, and other landmarks in the 9.95–15.53 px range; Table 1 lists Inferior endplate 15.5, Left pedicle 8.9, Right pedicle 18.3, Spinous tip 18.8, Center 6.5. These cannot both be correct for the same focal-length perturbation. Please reconcile the table, narrative, and any underlying figure annotations, and state the exact perturbation magnitude used for Table 1.
  3. The downstream-registration claim (Abstract; Contribution 4; §3.6; §5.6) is supported only for landmark-based registration with known 3D–2D correspondences (Kabsch and direct reprojection objectives, Eqs. 21–26). Intensity-based DRR–fluoroscopy registration—the clinically dominant setting—is not evaluated. Residual reprojection error under a mismatched K is partly expected by construction when the same landmarks define both the objective and the error metric. To sustain the claim that calibration-induced projection inconsistency 'propagates to downstream registration,' either (i) include at least one intensity-based similarity experiment (NCC/GC/MI) under perturbed K, or (ii) clearly limit the claim in the abstract and conclusion to landmark-based pose recovery under known correspondences.

Circularity Check

1 steps flagged

No load-bearing circularity; results are direct measurements of controlled synthetic perturbations against known ground truth, with only minor non-essential self-citation for motivation.

specific steps
  1. self citation load bearing [Section 1 Introduction, paragraph on recent studies]
    "Recent studies [1, 2] have suggested that fluoroscopic calibration uncertainty can affect the underlying imaging geometry used for DRR generation and 2D–3D registration. In this study, calibration variation specifically refers to controlled perturbations of intrinsic calibration parameters..."

    The premise that calibration uncertainty affects DRR/registration geometry is justified by two self-citations (same lead author). This is framing continuity rather than a load-bearing step: the subsequent synthetic experiments independently generate and measure the projection and registration effects, so the numerical results are not forced by the citations.

full rationale

The paper's derivation chain is an experimental sensitivity pipeline, not a theoretical derivation that reduces to its inputs by construction. Ground-truth and perturbed intrinsic matrices (Eqs. 3-4, 7-12) generate DRRs under fixed anatomy and extrinsic pose; landmark displacement (Eq. 16), Chamfer/Hausdorff, Dice/IoU (Eq. 18), and landmark-based registration errors (Eqs. 21-26) are then measured. These quantities are not fitted free parameters renamed as predictions, nor are they definitionally identical to the inputs. Self-citations [1,2] (same lead author) appear only in the introduction and related-work framing to motivate the question of projection-domain effects; they do not supply uniqueness theorems, ansätze, or numerical values that force the reported AP/LAT differences, Dice/IoU curves, landmark shifts, or registration rotational degradation. The non-isocentric geometry confound noted by the authors themselves is a validity/generalization issue, not circularity. Score 1 reflects only the minor, non-load-bearing self-citation continuity; the central empirical claims stand independently.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard projective geometry plus modeling choices that isolate intrinsic error in a synthetic non-isocentric C-arm. No new physical entities are introduced. Free parameters are experimental design knobs (perturbation ranges, nominal focal length, detector resolution) chosen by the authors rather than fitted to clinical outcomes; the claim is about measured sensitivity under those knobs, so they matter for scope but do not 'fit' the conclusion in a circular sense.

free parameters (4)
  • Ground-truth effective focal length = ~4500 px
    Stated as approximately 4500 pixels; sets the scale of all focal perturbations and projected anatomy size.
  • Focal-length perturbation range = 100–500 px
    Primary sensitivity axis; hand-chosen 100–500 pixel errors drive reported Dice/IoU and registration trends.
  • Principal-point / piercing displacement range = 10–50 px
    Secondary intrinsic perturbation; 10–50 pixels used in piercing-condition registration sweeps.
  • Detector resolution and imaging distances (SDD/SID, pixel spacing) = 1024×1024; SDD/SID fixed but not fully numeric in text
    Fixed cone-beam geometry parameters that determine magnification; partially described but not fully numerically specified for exact replication.
axioms (5)
  • domain assumption Pinhole projection model P = K[R|t] adequately represents C-arm geometry for DRR generation and landmark projection.
    Section 3.1 equations (1)–(4); all sensitivity is defined as change under K while R,t fixed.
  • domain assumption Extrinsic pose and rigid vertebral anatomy can be held exactly fixed while only intrinsic K is perturbed, isolating projection-domain effects.
    Methods 3.4 and Experiments 4.1; core isolation design of the study.
  • ad hoc to paper Non-isocentric AP/LAT source–object–detector geometry (vertebra closer to detector in LAT) is an acceptable model of clinical C-arm views for comparing view sensitivity.
    Section 4.1 and Figure 2; authors acknowledge this drives larger LAT apparent scale and thus larger displacements.
  • domain assumption Landmark-based reprojection minimization is a sufficient proxy to show that calibration-induced projection inconsistency harms 2D–3D registration.
    Section 3.6 and 4.5; intensity-based clinical registration is not evaluated.
  • domain assumption Ray-driven attenuation DRRs without scatter, detector distortion, noise, or beam hardening still reveal calibration sensitivity relevant to real fluoroscopy.
    Limitations 6.2; synthetic-only physics.

pith-pipeline@v1.1.0-grok45 · 18794 in / 3695 out tokens · 43849 ms · 2026-07-14T10:52:46.666147+00:00 · methodology

0 comments
read the original abstract

Accurate geometric calibration is essential for fluoroscopy-guided spinal imaging, digitally reconstructed radiograph (DRR) generation, and 2D--3D vertebral registration. Although calibration quality is typically evaluated using reconstruction-based metrics such as reprojection error, its influence on projection-domain consistency remains poorly understood. This study presents a synthetic framework for evaluating how intrinsic calibration perturbations affect vertebral fluoroscopic projections and downstream registration performance. CT-derived vertebral models and controlled cone-beam imaging geometry were used to generate DRRs with both ground-truth and perturbed intrinsic calibration parameters while maintaining identical anatomy and acquisition pose. Projection-domain changes were quantified using anatomical landmark displacement, contour distance, silhouette overlap, image similarity, and landmark-based 2D--3D registration accuracy in anterior--posterior (AP) and lateral (LAT) views. Results show that even small intrinsic calibration perturbations produce measurable changes in vertebral projection geometry, contour morphology, landmark localization, and DRR appearance. Sensitivity is strongly view dependent, with LAT projections exhibiting substantially greater deformation and anatomical displacement than AP projections. These projection inconsistencies also degrade downstream 2D--3D registration, particularly rotational alignment accuracy. The findings demonstrate that projection-domain consistency complements conventional reconstruction-based calibration metrics and provides a practical framework for assessing calibration robustness. This approach may improve the reliability of DRR generation and fluoroscopy-guided vertebral registration in image-guided spinal applications.

Figures

Figures reproduced from arXiv: 2607.10551 by Benjamin Aubert, Chaochao Zhou, Junchao Zhu, Junlin Guo, Lin Li.

Figure 1
Figure 1. Figure 1: Overview of the proposed projection-domain analysis pipeline. A CT-derived vertebral volume is first [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic illustration of the simulated AP and LAT fluoroscopic imaging setup. The vertebral model [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AP and LAT raw DRR difference heatmaps under focal length perturbations of 100 and 500 pixels. The [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Projection silhouette consistency under increasing intrinsic calibration perturbation magnitude. AP projec [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of vertebral landmark displacement between AP and LAT projections under intrinsic calibration [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Kabsch-based registration error under varying focal scale perturbation and piercing conditions. [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Direct registration error under varying focal scale perturbation and piercing conditions. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗

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

22 extracted references

  1. [1]

    Rigid 2d–3d registration in medical imaging: A survey.Preprints, May 2026

    Lin Li, Chaochao Zhou, Benjamin Albert, Junlin Guo, and Junchao Zhu. Rigid 2d–3d registration in medical imaging: A survey.Preprints, May 2026

    2026

  2. [2]

    Intrinsic tolerance in C-arm imaging: how extrinsic reoptimization preserves 3D reconstruction accuracy

    Lin Li, Benjamin Aubert, Paul Kemper, and Aric Plumley. Intrinsic tolerance in C-arm imaging: how extrinsic reoptimization preserves 3D reconstruction accuracy. In Maryam E. Rettmann and Pierre Jannin, editors,Medical Imaging 2026: Image-Guided Procedures, Robotic Interventions, and Modeling, volume 13927, page 139271T. International Society for Optics an...

    2026

  3. [3]

    J. B. Antoine Maintz and Max A. Viergever. A survey of medical image registration.Medical Image Analysis, 2(1):1–36, 1998

    1998

  4. [4]

    Image registration methods: a survey.Image and Vision Computing, 21(11):977– 1000, 2003

    Barbara Zitov ´a and Jan Flusser. Image registration methods: a survey.Image and Vision Computing, 21(11):977– 1000, 2003

    2003

  5. [5]

    Markelj, D

    P. Markelj, D. Toma ˇzeviˇc, B. Likar, and F. Pernu ˇs. A review of 3d/2d registration methods for image-guided interventions.Medical Image Analysis, 16(3):642–661, 2012

    2012

  6. [6]

    Visual prompt-agnostic evolution

    Junze Wang, Lei Fan, Dezheng Zhang, Weipeng Jing, Donglin Di, Yang Song, Sidong Liu, and Cong Cong. Visual prompt-agnostic evolution. InThe F ourteenth International Conference on Learning Representations, 2026

    2026

  7. [7]

    Penney, J

    Graeme P. Penney, J. Weese, J. A. Little, P. Desmedt, D. L. G. Hill, and D. J. Hawkes. A comparison of similarity measures for use in 2-d/3-d medical image registration.IEEE Transactions on Medical Imaging, 17(4):586–595, 1998

    1998

  8. [8]

    Paul Viola and William M. Wells. Alignment by maximization of mutual information.International Journal of Computer Vision, 24(2):137–154, 1997

    1997

  9. [9]

    Yoshito Otake, Amy S. Wang, J. Webster Stayman, Jeffrey H. Siewerdsen, Ajoy J. Khanna, Jean-Paul Wolinsky, Ziya L. Gokaslan, Nicholas Theodore, and Gregory D. Hager. Automatic localization of vertebral levels in x- ray fluoroscopy using 3d-2d registration: a tool to reduce wrong-site surgery.Physics in Medicine & Biology, 57(17):5485–5508, 2012

    2012

  10. [10]

    Pedicle screw pairing and registration for screw pose estimation from dual C-arm images using CAD models

    Yehyun Suh, Lin Li, Aric Plumley, Chaochao Zhou, Daniel Moyer, and Kongbin Kang. Pedicle screw pairing and registration for screw pose estimation from dual C-arm images using CAD models. In Maryam E. Rettmann and Pierre Jannin, editors,Medical Imaging 2026: Image-Guided Procedures, Robotic Interventions, and Modeling, volume 13927, page 139271W. Internati...

    2026

  11. [11]

    Camera augmented mobile c-arm (camc): calibra- tion, accuracy study, and clinical applications.IEEE Transactions on Medical Imaging, 18(9):806–816, 1999

    Nassir Navab, Michael Mitschke, and Alireza Bani-Hashemi. Camera augmented mobile c-arm (camc): calibra- tion, accuracy study, and clinical applications.IEEE Transactions on Medical Imaging, 18(9):806–816, 1999

    1999

  12. [12]

    Siewerdsen

    Jeffrey H. Siewerdsen. Cone-beam ct with a flat-panel detector: from image science to image-guided surgery. Nuclear Instruments and Methods in Physics Research Section A, 648(Supplement 1):S241–S250, 2011

    2011

  13. [13]

    V olume-of-change cone-beam ct for image-guided surgery.Physics in Medicine & Biology, 56(21):6855–6871, 2011

    Wojciech Zbijewski et al. V olume-of-change cone-beam ct for image-guided surgery.Physics in Medicine & Biology, 56(21):6855–6871, 2011

    2011

  14. [14]

    KAN-DDPM: Kolmogorov-Arnold networks with diffusion denoising proba- bilistic models for MRI-to-CT synthesis

    Vanessa Su and Xiaofeng Yang. KAN-DDPM: Kolmogorov-Arnold networks with diffusion denoising proba- bilistic models for MRI-to-CT synthesis. In Shandong Wu, editor,Medical Imaging 2025: Imaging Informatics, volume 13411, page 1341110. International Society for Optics and Photonics, SPIE, 2025

    2025

  15. [15]

    Robert L. Siddon. Fast calculation of the exact radiological path for a three-dimensional ct array. InMedical Physics, volume 12, pages 252–255, 1985

    1985

  16. [16]

    Precision, accuracy and performance of gpu-based rigid multi-modal image registration

    Andreas Maier et al. Precision, accuracy and performance of gpu-based rigid multi-modal image registration. Medical Physics, 40(9):091907, 2013

    2013

  17. [17]

    Differentiable x-ray rendering and 2d–3d registration for medical imaging.Med- ical Image Analysis, 2022

    Vignesh Gopalakrishnan et al. Differentiable x-ray rendering and 2d–3d registration for medical imaging.Med- ical Image Analysis, 2022

    2022

  18. [18]

    Jane Wang, and Rui Liao

    Shun Miao, Z. Jane Wang, and Rui Liao. A cnn regression approach for real-time 2d/3d registration.IEEE Transactions on Medical Imaging, 35(5):1352–1363, 2016

    2016

  19. [19]

    Deepdrr – a catalyst for machine learning in fluoroscopy-guided procedures.Interna- tional Conference on Medical Image Computing and Computer-Assisted Intervention, 2018

    Mathias Unberath et al. Deepdrr – a catalyst for machine learning in fluoroscopy-guided procedures.Interna- tional Conference on Medical Image Computing and Computer-Assisted Intervention, 2018

    2018

  20. [20]

    TTT- SEG: a multi-structure segmentation on cardiac MRI using vision testing-time training network

    Vanessa Su, Xiaohan Yuan, Shaoyan Pan, Mingzhe Hu, Xiaoqian Chen, Sibo Tian, and Xiaofeng Yang. TTT- SEG: a multi-structure segmentation on cardiac MRI using vision testing-time training network. In Barjor S. Gimi and Andrzej Krol, editors,Medical Imaging 2025: Clinical and Biomedical Imaging, volume 13410, page 134101X. International Society for Optics a...

    2025

  21. [21]

    Hypergraph tversky- aware domain incremental learning for brain tumor segmentation with missing modalities

    Junze Wang, Lei Fan, Weipeng Jing, Donglin Di, Yang Song, Sidong Liu, and Cong Cong. Hypergraph tversky- aware domain incremental learning for brain tumor segmentation with missing modalities. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 283–293. Springer, 2025

    2025

  22. [22]

    Chaochao Zhou, Thomas Cha, Yun Peng, and Guoan Li. Transfer learning from an artificial radiograph-landmark dataset for registration of the anatomic skull model to dual fluoroscopic x-ray images.Computers in Biology and Medicine, 138:104923, 2021. 20

    2021