pith. sign in

arxiv: 2301.12647 · v3 · submitted 2023-01-30 · ⚛️ physics.med-ph

3DMPR -- A robust morphological approach for applying phase retrieval in proximity to highly-attenuating objects in CT

J. A. Pollock , L. C. P. Croton , K. S. Morgan , K. J. Crossley , M. J. Wallace , G. A. Buckley , S. B. Hooper , M. J. Kitchen This is my paper

Pith reviewed 2026-05-24 09:38 UTC · model grok-4.3

classification ⚛️ physics.med-ph
keywords phase retrievalcomputed tomographymorphological operationsphase contrast imagingmulti-material samplessignal-to-noise ratiopropagation-based imagingsynchrotron CT
0
0 comments X

The pith

3DMPR combines 3D phase retrieval with morphological operations to enable strong noise suppression across all material boundaries in CT without over-blurring.

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

The paper introduces 3DMPR as a non-iterative method that applies propagation-based phase retrieval tuned to the strongest-contrast interface while using 3D morphological operations to shield every other material boundary from excessive smoothing. In tests on synchrotron images of a rabbit kitten brain inside its dense skull, this produced a 6.8-fold SNR gain in brain tissue compared with standard phase retrieval, without removing quantitative attenuation values or introducing new artifacts. The approach targets the common problem that conventional phase retrieval cannot be applied safely to regions bounded by more than one material and that choosing one boundary for tuning blurs the rest. If the method works as described, it would allow lower radiation doses for multi-material samples while retaining sharp edges at every interface.

Core claim

3D phase retrieval is combined with morphological operations to prevent over-blurring artefacts from being introduced, while avoiding the potentially long convergence times required by iterative approaches. This technique, entitled 3DMPR, was tested on phase contrast images of a rabbit kitten brain encased by the surrounding dense skull and provided a 6.8-fold improvement in the signal-to-noise ratio of brain tissue over the standard phase retrieval procedure, without over-smoothing the images.

What carries the argument

3DMPR, the integration of 3D phase retrieval with 3D morphological operations that identify and protect material boundaries during strong phase retrieval.

If this is right

  • Phase retrieval can be applied at full strength to samples containing more than two materials.
  • Quantitative attenuation information remains available at every interface after strong phase retrieval.
  • Computation remains non-iterative and does not require training data or long convergence times.
  • Radiation dose can be lowered for the same image quality in multi-material specimens.

Where Pith is reading between the lines

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

  • The same boundary-protection logic could be tested on clinical cone-beam CT systems where skull and brain are imaged together.
  • Extending the morphological mask to time-resolved data might allow dynamic samples with moving dense objects.
  • If the morphological step is replaced by a different edge detector, the SNR gain may change for samples with gradual density transitions.

Load-bearing premise

The morphological operations correctly identify and protect every material boundary in 3D without introducing new artifacts or removing quantitative attenuation information.

What would settle it

Reconstruction of a known multi-material phantom with ground-truth attenuation values shows either residual blurring at non-protected boundaries or deviation from expected quantitative values after 3DMPR processing.

Figures

Figures reproduced from arXiv: 2301.12647 by G. A. Buckley, J. A. Pollock, K. J. Crossley, K. S. Morgan, L. C. P. Croton, M. J. Kitchen, M. J. Wallace, S. B. Hooper.

Figure 1
Figure 1. Figure 1: Demonstration of phase-retrieval algorithms applied [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The masking and phase retrieval process, using a rabbit brain cross-section for demonstration. Images are labelled [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Quantification of edge resolution in a 3 mm thick aluminium rod submerged in water after phase retrieval was applied to material A, water. (a) and (b) show slices through the aluminium cylinder from phase retrievals performed with (3) and 3DMPR, respectively. (c) Plots an azimuthally aver￾aged line profile across the aluminium-water interface, com￾paring the raw phase contrast profile to that of the conven… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of our masked phase retrieval method [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The masking and phase retrieval process conceptu [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
read the original abstract

X-ray imaging is a fast, precise and non-invasive method of imaging which, when combined with computed tomography, provides detailed 3D rendering of samples. Incorporating propagation-based phase contrast can vastly improve data quality for weakly attenuating samples via phase retrieval, allowing radiation exposure to be reduced. However, applying phase retrieval to multi-material samples commonly requires choice of which material boundary to tune the reconstruction. Selecting the boundary with strongest phase contrast increases noise suppression, but at the detriment of over-blurring other interfaces and potentially removing quantitative sample information. Additionally, conventional phase-retrieval algorithms cannot be used for regions bounded by more than one material, requiring alternative methods. Here we present a computationally-efficient, non-iterative nor AI-mediated method for applying strong phase retrieval, whilst preserving sharp boundaries for all materials within the sample. 3D phase retrieval is combined with morphological operations to prevent over-blurring artefacts from being introduced, while avoiding the potentially long convergence times required by iterative approaches. This technique, entitled 3DMPR, was tested on phase contrast images of a rabbit kitten brain encased by the surrounding dense skull. Using 24kVp synchrotron radiation with a 5m propagation distance, 3DMPR provided a 6.8-fold improvement in the signal-to-noise ratio (SNR) of brain tissue over the standard phase retrieval procedure, without over-smoothing the images.

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

2 major / 0 minor

Summary. The manuscript introduces 3DMPR, a non-iterative method that combines 3D propagation-based phase retrieval with morphological operations to enable strong phase retrieval on multi-material samples while protecting all interfaces from over-blurring. It is demonstrated on a single synchrotron CT dataset of a rabbit-kitten brain inside its skull acquired at 24 kVp with 5 m propagation distance, reporting a 6.8-fold SNR improvement in brain tissue relative to standard phase retrieval without loss of quantitative information.

Significance. If the morphological masking step reliably identifies every material boundary in 3D and leaves attenuation coefficients unbiased, the technique would provide a computationally efficient route to quantitative phase-contrast imaging of complex biomedical samples, avoiding both the parameter tuning of single-boundary retrieval and the runtime of iterative alternatives.

major comments (2)
  1. [Abstract] Abstract: the central empirical claim of a 6.8-fold SNR gain is stated without any description of how SNR was computed, without error bars or replicate statistics, and without comparison to iterative phase-retrieval baselines, rendering the magnitude of the reported improvement impossible to evaluate.
  2. [Abstract] Abstract: the claim that the method works for regions bounded by more than one material rests on the untested premise that 3D morphological operations correctly label every interface (including triple points) without introducing new artifacts or removing quantitative attenuation values; no ground-truth phantom, no error maps at multi-material junctions, and no post-masking attenuation-coefficient checks are described.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We will revise the abstract and methods to clarify the SNR computation and add quantitative checks on attenuation values after masking. Responses to each major comment follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central empirical claim of a 6.8-fold SNR gain is stated without any description of how SNR was computed, without error bars or replicate statistics, and without comparison to iterative phase-retrieval baselines, rendering the magnitude of the reported improvement impossible to evaluate.

    Authors: We agree the SNR computation requires explicit description. SNR was calculated as the ratio of mean intensity (in a brain-tissue ROI) to the standard deviation (in a homogeneous sub-region of the same tissue) on the single synchrotron dataset; the 6.8-fold factor is the ratio of these two SNR values. We will add this definition to the revised abstract and methods. Because the work uses one acquisition, replicate statistics and error bars are unavailable; this is a limitation of the current demonstration rather than a statistical oversight. Iterative baselines are outside the paper's scope, which emphasizes non-iterative efficiency and interface protection, but a brief runtime comparison can be noted in the discussion. revision: partial

  2. Referee: [Abstract] Abstract: the claim that the method works for regions bounded by more than one material rests on the untested premise that 3D morphological operations correctly label every interface (including triple points) without introducing new artifacts or removing quantitative attenuation values; no ground-truth phantom, no error maps at multi-material junctions, and no post-masking attenuation-coefficient checks are described.

    Authors: The rabbit-kitten brain dataset contains multiple material boundaries (brain-skull, brain-soft tissue) and the visual results show all interfaces remain sharp after 3DMPR. We will add post-masking attenuation-coefficient checks in the revision to confirm no systematic bias is introduced. Error maps at junctions can be supplied if the raw data permit. The study does not include a dedicated ground-truth phantom; validation is performed on the real multi-material biomedical sample. The 3D morphological operations are constructed to label all boundaries, including triple points, by operating on the full 3D volume rather than 2D slices. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical method with independent validation on real data

full rationale

The paper describes a new 3DMPR technique that combines standard phase retrieval with 3D morphological operations to handle multi-material samples. The central result is an empirical SNR improvement (6.8-fold) measured on a real rabbit-kitten brain dataset at 24 kVp / 5 m propagation. No equations, fitted parameters, or self-citations are presented that reduce the reported gain or the multi-material claim to a tautology or input by construction. The method is framed as a practical, non-iterative alternative whose performance is tested directly on experimental images rather than derived from prior self-referential assumptions. This is a standard empirical methods paper with no load-bearing derivation chain that collapses into its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The method rests on standard assumptions of propagation-based phase retrieval (paraxial approximation, monochromatic beam, known propagation distance) plus the untested premise that morphological operations can segment all interfaces accurately in noisy 3D data. No free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption Phase retrieval can be applied in 3D across the entire volume before morphological correction.
    Invoked when the paper states that 3D phase retrieval is combined with morphological operations.
  • domain assumption Morphological operations can identify material boundaries without prior knowledge of which material produces the strongest phase contrast.
    Required for the claim that the method works for regions bounded by more than one material.

pith-pipeline@v0.9.0 · 5830 in / 1391 out tokens · 44187 ms · 2026-05-24T09:38:50.677474+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

19 extracted references · 19 canonical work pages

  1. [1]

    Innovations in tho- racic imaging: CT, radiomics, AI and x-ray velocime- try

    Rozemarijn Vliegenthart, Andreas Fouras, Colin Ja- cobs, and Nickolas Papanikolaou. Innovations in tho- racic imaging: CT, radiomics, AI and x-ray velocime- try. Respirology (Carlton, Vic.), 27(10):818–833, Octo- ber 2022

    work page 2022

  2. [2]

    Kitchen, Genevieve A

    Marcus J. Kitchen, Genevieve A. Buckley, Timur E. Gureyev, Megan J. Wallace, Nico Andres-Thio, Kentaro Uesugi, Naoto Yagi, and Stuart B. Hooper. CT dose re- duction factors in the thousands using X-ray phase con- trast. Scientific Reports , 7(1):15953, November 2017. Number: 1 Publisher: Nature Publishing Group

    work page 2017

  3. [3]

    Arhatari, Yakov I

    Benedicta D. Arhatari, Yakov I. Nesterets, Seyedamir T. Taba, Anton Maksimenko, Christopher J. Hall, An- drew W. Stevenson, Daniel H ¨asermann, Sarah J. Lewis, 7 Robust propagation-based phase retrieval for CT in proximity to highly attenuating objects Matthew Dimmock, Darren Thompson, Sheridan C. Mayo, Harry M. Quiney, Timur E. Gureyev, and Patrick C. Br...

    work page 2021

  4. [4]

    unrea- sonable

    Timur E. Gureyev, Yakov I. Nesterets, Alexander Kozlov, David M. Paganin, and Harry M. Quiney. On the “unrea- sonable” effectiveness of transport of intensity imaging and optical deconvolution. JOSA A, 34(12):2251–2260, December 2017. Publisher: Optica Publishing Group

    work page 2017

  5. [5]

    Paganin, S

    D. Paganin, S. C. Mayo, T. E. Gureyev, P. R. Miller, and S. W. Wilkins. Simultaneous phase and amplitude extrac- tion from a single defocused image of a homogeneous object. Journal of Microscopy , 206(Pt 1):33–40, April 2002

    work page 2002

  6. [6]

    Thompson, Yakov I

    Darren A. Thompson, Yakov I. Nesterets, Konstantin M. Pavlov, and Timur E. Gureyev. Fast three-dimensional phase retrieval in propagation-based X-ray tomography. Journal of Synchrotron Radiation , 26(Pt 3):825–838, May 2019

    work page 2019

  7. [7]

    Interface-specific x-ray phase retrieval tomog- raphy of complex biological organs

    Mario Beltran, David Paganin, Karen Siu, Andreas Fouras, Stuart Hooper, David Reser, and Marcus Kitchen. Interface-specific x-ray phase retrieval tomog- raphy of complex biological organs. Physics in Medicine & Biology , 56(23):7353–7369, 2011. Publisher: IOP Publishing

    work page 2011

  8. [8]

    M. A. Beltran, D. M. Paganin, K. Uesugi, and M. J. Kitchen. 2D and 3D X-ray phase retrieval of multi- material objects using a single defocus distance. Optics Express, 18(7):6423–6436, March 2010. Publisher: Op- tica Publishing Group

    work page 2010

  9. [9]

    Imaging the Brain In Situ with Phase Contrast CT

    Linda Croton, Kaye Morgan, David Paganin, Lauren Kerr, Megan Wallace, Kelly Crossley, Gary Ruben, Suzanne Miller, Naoto Yagi, Kentaro Uesugi, Stuart Hooper, and Marcus Kitchen. Imaging the Brain In Situ with Phase Contrast CT. Microscopy and Microanalysis, 24(S2):352–353, August 2018. Publisher: Cambridge University Press

    work page 2018

  10. [10]

    Linda C. P. Croton, Kaye S. Morgan, David M. Paganin, Lauren T. Kerr, Megan J. Wallace, Kelly J. Crossley, Suzanne L. Miller, Naoto Yagi, Kentaro Uesugi, Stu- art B. Hooper, and Marcus J. Kitchen. In situ phase con- trast X-ray brain CT. Scientific Reports, 8(1):11412, July

    work page

  11. [11]

    Number: 1 Publisher: Nature Publishing Group

    work page

  12. [12]

    No¨el, and Franz Pfeiffer

    Lorenz Hehn, Kaye Morgan, Pidassa Bidola, Wolfgang Noichl, Regine Gradl, Martin Dierolf, Peter B. No¨el, and Franz Pfeiffer. Nonlinear statistical iterative reconstruc- tion for propagation-based phase-contrast tomography. APL Bioengineering , 2(1):016105, March 2018. Pub- lisher: American Institute of Physics

    work page 2018

  13. [13]

    Hammel, Roland Hessler, Kaye S

    Lorenz Hehn, Regine Gradl, Andrej V oss, Benedikt G¨unther, Martin Dierolf, Christoph Jud, Konstantin Willer, Sebastian Allner, J ¨org U. Hammel, Roland Hessler, Kaye S. Morgan, Julia Herzen, Werner Hem- mert, and Franz Pfeiffer. Propagation-based phase- contrast tomography of a guinea pig inner ear with cochlear implant using a model-based iterative re- ...

    work page 2018

  14. [14]

    C. T. Chantler, K. Olsen, R. A. Dragoset, J. Chang, A. R. Kishore, S. A. Kotochigova, and D. S. Zucker. X-Ray Form Factor, Attenuation, and Scattering Tables. NIST, 2005

    work page 2005

  15. [15]

    M. J. Berger, J. H. Hubbell, S. M. Seltzer, J. Chang, J. S. Coursey, R. Sukumar, D. S. Zucker, and K. Olsen. XCOM: Photon Cross Sections Database. NIST, 2010

    work page 2010

  16. [16]

    Joost Batenburg, and Jan Sijbers

    Wim van Aarle, Willem Jan Palenstijn, Jeroen Cant, Eline Janssens, Folkert Bleichrodt, Andrei Dabravolski, Jan De Beenhouwer, K. Joost Batenburg, and Jan Sijbers. Fast and flexible X-ray tomography using the ASTRA toolbox. Optics Express, 24(22):25129–25147, October

    work page

  17. [17]

    Publisher: Optica Publishing Group

    work page

  18. [18]

    Joost Baten- burg, and Jan Sijbers

    Wim van Aarle, Willem Jan Palenstijn, Jan De Been- houwer, Thomas Altantzis, Sara Bals, K. Joost Baten- burg, and Jan Sijbers. The ASTRA Toolbox: A platform for advanced algorithm development in electron tomog- raphy. Ultramicroscopy, 157:35–47, October 2015

    work page 2015

  19. [19]

    The xraylib li- brary for X-ray–matter interactions

    Tom Schoonjans, Antonio Brunetti, Bruno Golosio, Manuel Sanchez del Rio, Vicente Armando Sol ´e, Clau- dio Ferrero, and Laszlo Vincze. The xraylib li- brary for X-ray–matter interactions. Recent develop- ments. Spectrochimica Acta Part B: Atomic Spec- troscopy, 66(11):776–784, November 2011. 8

    work page 2011