pith. sign in

arxiv: 2606.02935 · v1 · pith:7DU3ZUF5new · submitted 2026-06-01 · 💻 cs.CV · cs.CE

CAD-to-CT Registration of Cylindrical Objects via Ellipse-Based Axis Estimation

Aleksander Ogonowski , Miko{\l}aj Mrozowski , Daniel Wi\k{e}cek , Arkadiusz \'Cwiek , Konrad Klimaszewski , Rafa{\l} Mo\.zd\.zonek , Adam Padee , Lech Raczy\'nski
show 4 more authors
Piotr Wasiuk Wojciech Wi\'slicki Micha{\l} Matusiak S{\l}awomir Wronka
This is my paper

Pith reviewed 2026-06-28 14:39 UTC · model grok-4.3

classification 💻 cs.CV cs.CE
keywords CAD registrationCT imagingellipse fittingaxis estimationRANSACPCAvolumetric overlapcylindrical objects
0
0 comments X

The pith

Ellipse fitting on CT slices followed by PCA on their centers recovers the 3D axis of cylindrical objects for CAD registration without intensity calibration or feature matching.

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

The paper develops a two-stage geometric method to register CAD models to CT scans of cylindrical objects such as ionization chambers. It detects elliptical cross-sections in multiple CT slices, fits ellipses to edge contours, removes outliers with RANSAC, and uses PCA on the centers to estimate the rotation axis. The voxelized CAD model is then oriented along this axis and translated to maximize volumetric overlap with the scan. This sidesteps failures of intensity-based or point-correspondence methods on uncalibrated or noisy data. The reported outcome is tilt and orientation errors below 0.1 degrees, supplying ground-truth geometry for machine-learning localization tasks.

Core claim

Detecting elliptical cross-sections across CT slices, fitting ellipses to contours, cleaning centers via RANSAC, and performing PCA yields the object's 3D rotation axis; the CAD model is voxelized, aligned to that axis, and shifted to maximize overlap, producing registration with tilt and orientation errors below 0.1 degrees.

What carries the argument

Ellipse-based axis estimation: fitting ellipses to edge-detected CT contours, RANSAC outlier removal on centers, and PCA to recover the 3D rotation axis.

If this is right

  • The aligned CAD model supplies ground-truth masks for training machine-learning object localization on volumetric CT data.
  • Registration succeeds on scans lacking intensity calibration references.
  • No explicit feature correspondence is required between idealized CAD surfaces and noisy CT voxels.
  • The method supports automated analysis pipelines in industrial CT workflows.

Where Pith is reading between the lines

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

  • The same ellipse-center PCA step could be tested on objects with mild deviations from perfect cylindrical symmetry to measure the breakdown threshold.
  • Varying the number of slices or RANSAC parameters might reveal trade-offs between axis accuracy and computation time on larger datasets.
  • The overlap-maximization stage could be replaced by other translation estimators to check whether the axis estimate alone is the dominant accuracy driver.

Load-bearing premise

The objects are cylindrical enough that their CT cross-sections yield reliably elliptical contours whose cleaned centers trace the true rotation axis.

What would settle it

Apply the pipeline to CT volumes of cylinders whose true tilt is known independently and measure whether the recovered axis deviates by more than 0.1 degrees.

Figures

Figures reproduced from arXiv: 2606.02935 by Adam Padee, Aleksander Ogonowski, Arkadiusz \'Cwiek, Daniel Wi\k{e}cek, Konrad Klimaszewski, Lech Raczy\'nski, Micha{\l} Matusiak, Miko{\l}aj Mrozowski, Piotr Wasiuk, Rafa{\l} Mo\.zd\.zonek, S{\l}awomir Wronka, Wojciech Wi\'slicki.

Figure 1
Figure 1. Figure 1: Challenge of threshold-based segmentation in CT data: (left) full CT slice showing ion [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the CAD-to-CT registration pipeline: (a) rotation axis estimation from [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Three orthogonal views of the ionization chamber CT scan showing axial, coronal, and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ellipse fitting pipeline on a single CT slice: (a) original CT slice [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geometric consistency validation: distances between inner and outer ellipse centers across [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: RANSAC outlier removal visualized through 2D projections: (a) XZ projection (side [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Schematic illustration of the PCA-based axis estimation. Black dots are inlier ellipse [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Polynomial fitting for translational alignment. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Registration result on simulated CT data. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Registration and segmentation results for an example real ionization chamber. The CT [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
read the original abstract

Accurate registration of CAD models to CT scans is essential for establishing ground truth geometry in volumetric imaging. Obtaining reliable object masks is of growing importance in machine learning settings; as recent architectures grow more capable, huge datasets are required to fully utilise their capabilities. Traditional intensity-based methods fail when CT grayscale values lack calibration references, while point-based algorithms (e.g., ICP, RANSAC) require feature correspondence unavailable between idealized CAD geometry and noisy volumetric CT data. We propose a two-stage geometric registration method for cylindrical objects (ionization chambers) that takes advantage of the distinctive geometric features of the objects. First, we estimate the 3D rotation axis by detecting elliptical cross-sections across CT slices, fitting ellipses to edge-detected contours, and performing PCA on the fitted ellipse centers after RANSAC outlier removal. Second, we voxelize the CAD model, orient it along the detected axis, and maximize volumetric overlap with the CT scan through translational adjustment. This approach achieves robust registration with tilt and orientation errors below $0.1^\circ$ without intensity calibration or feature matching. Once registered, the aligned CAD model provides ground truth geometry for applications including machine learning-based object localization and automated analysis in industrial CT workflows.

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 claims to introduce a two-stage geometric method for CAD-to-CT registration of cylindrical objects. The first stage detects elliptical cross-sections in CT slices, fits ellipses to edge contours, removes outliers with RANSAC, and estimates the axis via PCA on the centers. The second stage orients the voxelized CAD model to this axis and adjusts translation to maximize overlap. It asserts sub-0.1° errors in tilt and orientation without calibration or matching.

Significance. If the accuracy claims are verified through experiments, this method could significantly aid in creating large ground-truth datasets for ML in industrial CT by providing a robust registration technique that does not rely on intensity values or point correspondences, addressing key challenges in the field.

major comments (1)
  1. [Abstract] Abstract: The central claim of achieving tilt and orientation errors below 0.1° is unsupported by any data, experiments, error analysis, or validation in the manuscript. The description provides no details on CT resolution, number of slices, ground truth acquisition, or quantitative metrics, making it impossible to assess whether the ellipse fitting and PCA steps deliver the required precision under realistic CT noise and partial volume effects.
minor comments (2)
  1. The manuscript would benefit from including pseudocode or a diagram of the pipeline for clarity.
  2. References to related work on ellipse fitting in CT or cylindrical registration are missing.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful review and for highlighting the need for explicit validation of the accuracy claims. We agree that the abstract's quantitative assertion requires supporting experimental details and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of achieving tilt and orientation errors below 0.1° is unsupported by any data, experiments, error analysis, or validation in the manuscript. The description provides no details on CT resolution, number of slices, ground truth acquisition, or quantitative metrics, making it impossible to assess whether the ellipse fitting and PCA steps deliver the required precision under realistic CT noise and partial volume effects.

    Authors: We agree that the current manuscript does not provide the requested experimental details, error analysis, or validation metrics to support the sub-0.1° claim. The abstract statement is therefore not substantiated as written. We will revise by (1) expanding the abstract to reference the validation protocol, (2) adding a dedicated experimental section that specifies CT voxel resolution, slice count, ground-truth acquisition via known CAD poses, and quantitative error metrics (tilt/orientation RMSE), and (3) including analysis of robustness to realistic CT noise and partial-volume effects. These additions will be placed in the results and discussion sections so that the accuracy claim can be directly assessed. revision: yes

Circularity Check

0 steps flagged

No circularity: procedural geometric pipeline with no derivations or self-referential fits

full rationale

The paper presents a two-stage registration algorithm: ellipse detection and fitting on CT slices, RANSAC on centers, PCA for axis estimation, followed by CAD voxelization and translational overlap maximization. No equations, parameter fits, or predictions are described that reduce to their own inputs by construction. The <0.1° error claim is asserted as an empirical outcome of the pipeline rather than a derived quantity from fitted values. No self-citations, uniqueness theorems, or ansatzes are invoked. The method is self-contained as a sequence of standard geometric operations (edge detection, ellipse fitting, PCA, overlap search) without load-bearing reductions to prior author work or internal definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities can be extracted or audited from the provided text.

pith-pipeline@v0.9.1-grok · 5824 in / 1155 out tokens · 28210 ms · 2026-06-28T14:39:38.983817+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

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

  1. [1]

    Aiger, N

    D. Aiger, N. J. Mitra, and D. Cohen-Or. 4-points congruent sets for robust pairwise surface registration.ACM Transactions on Graphics, 27(3):1–10, 2008

    2008

  2. [2]

    P. R. Almond, P. J. Biggs, B. M. Coursey, W. F. Hanson, M. S. Huq, R. Nath, and D. W. O. Rogers. AAPM’s TG-51 protocol for clinical reference dosimetry of high-energy photon and electron beams.Medical Physics, 26(9):1847–1870, 1999

    1999

  3. [3]

    J. F. Barrett and N. Keat. Artifacts in CT: Recognition and avoidance.Radiographics, 24(6):1679–1691, 2004

    2004

  4. [4]

    P. J. Besl and N. D. McKay. A method for registration of 3-D shapes.IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239–256, 1992

    1992

  5. [5]

    Buratti, J

    A. Buratti, J. Bredemann, M. Pavan, R. Schmitt, and S. Carmignato. Applications of CT for non-destructive testing and materials characterization.Procedia CIRP, 78:250–255, 2018

    2018

  6. [6]

    CadQuery: A parametric 3D CAD scripting framework.https: //github.com/CadQuery/cadquery, 2023

    CadQuery Contributors. CadQuery: A parametric 3D CAD scripting framework.https: //github.com/CadQuery/cadquery, 2023

    2023

  7. [7]

    Carmignato, D

    S. Carmignato, D. Dreossi, L. Mancini, F. Marinello, G. Tromba, and E. Savio. Accuracy of industrial computed tomography measurements: Experimental results from an international comparison.CIRP Annals – Manufacturing Technology, 61(1):491–494, 2012

    2012

  8. [8]

    Dewulf, Y

    W. Dewulf, Y. Tan, and K. Kiekens. Sense and non-sense of beam hardening correction in CT metrology.CIRP Annals – Manufacturing Technology, 61(1):495–498, 2012

    2012

  9. [9]

    Drost, M

    B. Drost, M. Ulrich, N. Navab, and S. Ilic. Model globally, match locally: Efficient and robust 3D object recognition. InIEEE Computer Society Conference on Computer Vision and Pattern Recognition, pages 998–1005, 2010

    2010

  10. [10]

    Ferrucci, R

    M. Ferrucci, R. K. Leach, C. Giusca, S. Carmignato, and W. Dewulf. Towards geometrical calibration of X-ray computed tomography systems – A review.Measurement Science and Technology, 26(9):092003, 2015

    2015

  11. [11]

    M. A. Fischler and R. C. Bolles. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

    1981

  12. [12]

    Fitzgibbon, M

    A. Fitzgibbon, M. Pilu, and R. B. Fisher. Direct least square fitting of ellipses.IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 21(5):476–480, 1999

    1999

  13. [13]

    A. B. Forbes. Least-squares best-fit geometric elements. Technical Report DITC 140/89, National Physical Laboratory, Teddington, UK, 1989. 17

    1989

  14. [14]

    FreeCAD: An Open Source Parametric 3D CAD Modeler.https: //www.freecad.org, 2024

    FreeCAD Contributors. FreeCAD: An Open Source Parametric 3D CAD Modeler.https: //www.freecad.org, 2024

    2024

  15. [15]

    J. H. Hubbell and S. M. Seltzer. Tables of X-ray mass attenuation coefficients and mass energy- absorption coefficients. Technical Report NISTIR 5632, National Institute of Standards and Technology, 1995

    1995

  16. [16]

    J. S. Jørgensen, E. Ametova, G. Burca, et al. Core Imaging Library – Part I: a versatile Python framework for tomographic imaging.Philosophical Transactions of the Royal Society A, 379(2204):20200192, 2021

    2021

  17. [17]

    Klein, M

    S. Klein, M. Staring, K. Murphy, M. A. Viergever, and J. P. Pluim. elastix: A toolbox for intensity-based medical image registration.IEEE Transactions on Medical Imaging, 29(1):196– 205, 2010

    2010

  18. [18]

    Kruth, M

    J.-P. Kruth, M. Bartscher, S. Carmignato, R. Schmitt, L. De Chiffre, and A. Weckenmann. Computed tomography for dimensional metrology.CIRP Annals – Manufacturing Technology, 60(2):821–842, 2011

    2011

  19. [19]

    J. J. Lifton, A. A. Malcolm, and J. W. McBride. An experimental study on the influence of scatter and beam hardening in X-ray CT for dimensional metrology.Measurement Science and Technology, 27(1):015007, 2016

    2016

  20. [20]

    Litjens, T

    G. Litjens, T. Kooi, B. E. Bejnordi, et al. A survey on deep learning in medical image analysis. Medical Image Analysis, 42:60–88, 2017

    2017

  21. [21]

    Luk´ acs, R

    G. Luk´ acs, R. R. Martin, and A. D. Marshall. Faithful least-squares fitting of spheres, cylinders, cones and tori for reliable segmentation.Computer Vision – ECCV’98, pages 671–686, 1998

    1998

  22. [22]

    F. Maes, A. Collignon, D. Vandermeulen, G. Marchal, and P. Suetens. Multimodality image registration by maximization of mutual information.IEEE Transactions on Medical Imaging, 16(2):187–198, 1997

    1997

  23. [23]

    M¨ uller, A

    P. M¨ uller, A. Cantatore, J. L. Andreasen, J. Hiller, and L. De Chiffre. Computed tomography as a tool for tolerance verification of industrial parts.Procedia CIRP, 10:125–132, 2013

    2013

  24. [24]

    F. S. Nooruddin and G. Turk. Simplification and repair of polygonal models using volumetric techniques.IEEE Transactions on Visualization and Computer Graphics, 9(2):191–205, 2003

    2003

  25. [25]

    Nurunnabi, D

    A. Nurunnabi, D. Belton, and G. West. Robust cylinder fitting in three-dimensional point cloud data.International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 39:1–6, 2012

    2012

  26. [26]

    J. P. Pluim, J. A. Maintz, and M. A. Viergever. Mutual-information-based registration of medical images: A survey.IEEE Transactions on Medical Imaging, 22(8):986–1004, 2003

    2003

  27. [27]

    Po ludniowski et al

    G. Po ludniowski et al. Technical Note: SpekPy v2.0 — a Software Toolkit for Modeling X-ray Tube Spectra.Medical Physics, 48(7):3630–3637, 2021

    2021

  28. [28]

    Rabbani, F

    T. Rabbani, F. A. van den Heuvel, and G. Vosselmann. Automatic reconstruction of industrial installations using point clouds and images.International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 36(3):62–67, 2006. 18

    2006

  29. [29]

    Ronneberger, P

    O. Ronneberger, P. Fischer, and T. Brox. U-Net: Convolutional networks for biomedical image segmentation. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 234–241. Springer, 2015

    2015

  30. [30]

    Rusinkiewicz and M

    S. Rusinkiewicz and M. Levoy. Efficient variants of the ICP algorithm.Proceedings Third International Conference on 3-D Digital Imaging and Modeling, pages 145–152, 2001

    2001

  31. [31]

    R. B. Rusu, N. Blodow, and M. Beetz. Fast point feature histograms (FPFH) for 3D reg- istration. InIEEE International Conference on Robotics and Automation, pages 3212–3217, 2009

    2009

  32. [32]

    Scharr.Optimal operators in digital image processing

    H. Scharr.Optimal operators in digital image processing. PhD thesis, Universit¨ at Heidelberg, 2000

    2000

  33. [33]

    C. M. Shakarji. Least-squares fitting algorithms of the NIST algorithm testing system.Journal of Research of the National Institute of Standards and Technology, 103(6):633, 1998

    1998

  34. [34]

    Suzuki and K

    S. Suzuki and K. Abe. Topological structural analysis of digitized binary images by border following.Computer Vision, Graphics, and Image Processing, 30(1):32–46, 1985

    1985

  35. [35]

    A. A. Taha and A. Hanbury. Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool.BMC Medical Imaging, 15:29, 2015

    2015

  36. [36]

    Tan and Q

    M. Tan and Q. Le. EfficientNet: Rethinking model scaling for convolutional neural networks. InInternational Conference on Machine Learning, pages 6105–6114, 2019

    2019

  37. [37]

    Y. Tan, K. Kiekens, F. Welkenhuyzen, J. Angel, L. De Chiffre, J.-P. Kruth, and W. Dewulf. Comparison of surface detection methods to evaluate cone beam computed tomography data for three dimensional metrology.Optics and Lasers in Engineering, 76:11–21, 2016

    2016

  38. [38]

    J. Yang, H. Li, D. Campbell, and Y. Jia. Go-ICP: A globally optimal solution to 3D ICP point-set registration.IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(11):2241–2254, 2016

    2016

  39. [39]

    Q.-Y. Zhou, J. Park, and V. Koltun. Open3D: A Modern Library for 3D Data Processing. arXiv preprint arXiv:1801.09847, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  40. [40]

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang. UNet++: A nested U-Net architecture for medical image segmentation. InDeep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, pages 3–11. Springer, 2018

    2018

  41. [41]

    Zitov´ a and J

    B. Zitov´ a and J. Flusser. Image registration methods: A survey.Image and Vision Computing, 21(11):977–1000, 2003. 19

    2003