Pith. sign in

REVIEW 3 major objections 6 minor 20 references

Super-Resolution Radiography by Mechanical Supersampling and Model-Based Iterative Reconstruction using High-Z Photon-Counting Detectors

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Mechanical supersampling with a high-Z photon-counting detector plus model-based iterative reconstruction raises radiographic spatial resolution by roughly threefold at clinical flux rates and dose.

desk verdict A credible proof-of-concept for mechanical supersampling on high-Z PCDs at clinical flux; the resolution gain is likely real, but the factor-of-three quantitative claim needs uncertainty analysis and a ground-truth check. read the letter →

arxiv 2507.20793 v1 pith:AVNKWDQL submitted 2025-07-28 physics.med-ph

Moritz Weigt , Karl Hugo E. Helin , Nicolas Karl Fix , Peter Bronsert , Frank Goldschmidtböing , Thomas Thuering , Matthias Habl , Spyridon Gkoumas
show 6 more authors
Thomas Stein Paul Martin Fleing Thomas Billoud Jakob Neubauer Dominik von Elverfeldt Martin Peter Pichotka
This is my paper · ORCID
classification physics.med-ph PACS 87.59.-e87.59.Dj
keywords photon-countingdetectorssuper-resolutionmechanicalsupersamplingiterativereconstructionMLEMdigitalmammographyedgespreadfunctionhigh-Zsemiconductor
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that planar radiography can be pushed well beyond the pixel-pitch limit of a high-Z photon-counting detector without the flux restrictions of event-based super-resolution, by physically translating the detector in sub-pixel steps during exposure and merging the frames with an iterative, model-based reconstruction. The authors show in phantom studies that this mechanical supersampling plus distance-driven MLEM roughly triples the effective spatial resolution (MTF10%) and narrows the 10–90% edge-spread width from 93.75 µm to 31.25 µm relative to interpolation-based upscaling. They further demonstrate that a prototype GaAs detector so processed yields sharper, more homogeneous images of a mammography phantom than a clinical mammography system, at comparable or lower dose. The clinical significance, if the result holds, is that micron-scale mammographic detail such as microcalcifications could be obtained without geometric magnification, large detector areas, or increased radiation exposure.

What carries the argument

The carrying mechanism is a two-stage pipeline. First, the detector is translated along a trajectory spanning several pixels while frames are accumulated; sub-pixel displacements are estimated afterward with enhanced correlation coefficient (ECC) registration, so no mechanical precision is needed. Second, the low-resolution frames are plugged into an inverse problem treated as pseudo-laminographic, solved with a maximum-likelihood expectation-maximization (MLEM) algorithm whose forward projector is a distance-driven ray tracer that models the PCD pixel response. The long-range trajectory additionally acts as a moving average over pixel-level response variations, which suppresses ASIC borders, sensor defects, and spectral inhomogeneities that would otherwise masquerade as detail.

What would settle it

Take a physical phantom with line pairs or a sharp edge, and image it twice: once with the supersampling pipeline presented here, and once with a detector whose true pixel pitch equals the claimed effective resolution (about 25 µm). If the supersampled image resolves features that the genuinely finer-pitched detector does not, or if the measured edge-spread width changes when the trajectory model is changed without touching the hardware, then the apparent resolution gain is a reconstruction artifact rather than recovered information.

Watch

Extended reading notes

Core claim

The paper establishes that mechanically supersampled acquisition combined with model-based iterative reconstruction produces a near threefold improvement in effective spatial resolution in planar radiography with high-Z photon-counting detectors operating in true photon-counting mode at clinical flux rates. The central evidence is the edge-spread measurement: the 10–90% rise of a tungsten edge falls from 93.75 µm (linear interpolation of registered frames) to 31.25 µm with the proposed iterative deconvolution, and the MTF10% threshold improves by a comparable factor. Long-range supersampling over many pixels is shown to average out sensor defects, ASIC boundary pixels, and inter-pixel spectral variations, yielding more homogeneous absorption and spectral line profiles, at the cost of some edge sharpness relative to short-range sampling. The authors position the result as making micron-scale radiography clinically viable, with digital mammography as the most immediate application.

Load-bearing premise

The key load-bearing premise is that the distance-driven ray model accurately describes what the detector actually measures, including charge sharing and sensor field distortions; the paper's own observation of ringing artifacts and geometric warping at supersampled resolution shows this premise is only approximately true.

Editorial extensions

If this is right

  • Digital mammography could detect smaller microcalcifications and sharper spiculations without geometric magnification or a large-area detector, at comparable or reduced dose.
  • The resolution gain relaxes the pixel-pitch-versus-spectral-pixel-size trade-off in high-Z PCDs, since larger pixels can be used and supersampling recovers the lost detail.
  • Long-range supersampling improves spectral fidelity enough to support quantitative contrast-agent imaging from a single projection.
  • Continuous 'ballistic' detector motion with burst readout could bring a complete supersampled scan into the 100–400 ms range of clinical mammography exposures.
  • Explicit spectral field mapping, proposed for future work, may cure the geometric distortion and ringing artifacts that currently limit short-range supersampling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If charge sharing is included in the forward model as a threshold-dependent term, the same pipeline could plausibly push resolution toward the sensor's intrinsic edge-response limit, perhaps beyond the threefold factor demonstrated here.
  • The ECC-based shift estimation should generalize to patient motion compensation, not just detector motion, which could make the method robust in live breast imaging.
  • The spectral-averaging benefit is a testable precondition for energy-resolved phase-contrast imaging; a bench experiment comparing spectral homogeneity with and without long-range supersampling would settle it.
  • A reader trial on clinical mammograms (supersampled versus conventional, dose-matched) would be the decisive test of whether the improved phantom resolution translates into better detection of microcalcifications.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript proposes a super-resolution radiography method (PC-SSI) that combines mechanical detector motion over sub-pixel steps, ECC-based registration for shift estimation, and MLEM reconstruction with a distance-driven ray tracer. The authors report a near threefold improvement in effective spatial resolution, quantified by a 10-90% edge spread function (31.25 µm versus 93.75 µm for linear interpolation) and by MTF10% analysis, along with qualitative improvements in bar-pattern visibility, suppression of pixel/ASIC defects for long-range trajectories, and a comparison with a clinical mammography system at comparable or lower dose. The central quantitative claim rests on single ESF and MTF measurements without error bars, and the forward model does not include charge sharing or sensor field distortions that the authors themselves identify as sources of artifacts; these points need additional validation.

Significance. If the reported resolution gain is real, the method is practically significant: it operates in frame-based photon-counting mode at clinically relevant flux, avoids the flux limitations of event-based centroiding, and could reduce the need for geometric magnification and large-area detectors in mammography. The qualitative phantom results, the dose-conscious acquisition design, and the explicit comparison with a clinical system are valuable strengths. The method's feasibility is plausible, but the quantitative threefold resolution claim is not yet established because the evidence is a single-edge ESF and a single-region MTF, both produced by an unregularized iterative reconstruction whose forward model omits known detector physics.

major comments (3)
  1. [Section III, Figs. 4 and 7] The central quantitative claim of a threefold resolution improvement is based on a single ESF from one tungsten edge and a single MTF from one image region, with no repeated acquisitions, error bars, or sensitivity analysis. The reported numbers (10-90% rise in 31.25 µm versus 93.75 µm, and MTF10% values) are load-bearing and need to be shown to be reproducible and stable against ROI selection, edge-angle estimation, and registration uncertainty. Please provide multiple independent measurements with spread, and state how the metrics were computed.
  2. [Section II.C, Section III, Section IV] The distance-driven ray tracer models geometric pixel sensitivity but not charge sharing or sensor field distortions, and the paper attributes the ringing in Fig. 3 and the geometric warping to these unmodeled effects. Because MLEM is run without regularization (Section IV), an incorrect forward model can be compensated by placing energy at high spatial frequencies, which can artificially narrow the 10-90% ESF and raise MTF10%. The comparison with linear/cubic interpolation therefore does not by itself demonstrate recovery of true object information. Please validate against known ground truth (for example, an independently measured edge profile, a phantom with known feature dimensions, or a high-resolution reference method), include a charge-sharing response in the forward model, or report how the ESF/MTF metrics change with iteration count, regularization, and energy threshold settings.
  3. [Section II.C and Section IV] The acquisition and reconstruction protocol is under-specified for reproducibility: no MLEM iteration count or stopping criterion, no TV regularization weight (even though it was disabled), no ECC registration parameters, no exact trajectory or dwell pattern (only approximate strides in figure captions), and no description of the ASIC-border interpolation parameters. Since these choices directly affect the resolution metrics, the full protocol should be reported or the code/data should be made available.
minor comments (6)
  1. [Fig. 7] "Module Transfer Function" should be "Modulation Transfer Function".
  2. [Section III] There are spelling errors such as "imhomongeneities" and "magni ication" in the Fig. 8 caption; the manuscript should be proofread.
  3. [Section IV] The ballistic imaging paragraph refers to a "5 kW X-ray source" while the experimental settings are described elsewhere as 5 W; please clarify which value is intended.
  4. [Section III] The acronym "PCD SSI" is used without definition; elsewhere the method is called "PC-SSI".
  5. [Section II.B and Fig. 8] The phantom is referred to as both "Gammex 156 Mammographic Accreditation Phantom" and "Gammex ACR MAP"; please use consistent terminology.
  6. [Fig. 3] The caption's "TH40kV" is ambiguous; please specify the tube setting or threshold condition it denotes.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the resolution gain is measured from reconstructed images rather than fitted by the model, and the self-citations are not load-bearing.

full rationale

The central claim is an experimental measurement: the ESF in Fig. 4 and the MTF10% in Fig. 7 are computed from reconstructed phantom images, not obtained by fitting model parameters to those metrics. The forward model is a standard distance-driven projector (ref. [11], De Man and Basu) used in a pseudo-laminographic MLEM scheme; shift estimation uses the external ECC method (ref. [10]); STC calibration uses external references ([8], [9]). The only self-citations are [12]-[14], cited for the MLEM/TV framework, but the paper states 'we disabled regularization for the results presented here in order to isolate and clearly demonstrate the effects of mechanical supersampling,' so those citations are not load-bearing for the headline result. The skeptic's concern that unmodeled charge sharing and field distortions may inflate the measured sharpness is a correctness or validity risk, not circular reasoning: no equation in the paper defines the resolution metric in terms of an input that was fitted to that metric, and the ringing/geometric artifacts are explicitly attributed to detector physics rather than encoded as the claimed improvement. The missing [23,28] citations in the workflow list (Section II.C, step 3) are an editorial defect, not a circular step.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The method rests on the accuracy of the forward model, the registration, and the calibration, rather than on fitted parameters. The free parameters are acquisition protocol choices (step size, grid count, iteration count) that affect the resolution/artifact trade-off. No new physical entities are introduced.

free parameters (2)
  • Supersampling step size and grid dimensions = e.g., 0.125 pixel vertical x 14 pixel horizontal for Gammex; 0.15 pixel stride 14x14 and 1.1 pixel stride 11x11/22x22…
    Chosen by hand per experiment; the stride and grid extent control the trade-off between resolution gain and artifact averaging, not fitted to the outcome metrics.
  • MLEM iteration count = Not reported
    The reconstruction uses MLEM with the number of iterations unspecified in the text; this is a tuning parameter that affects the resolution/noise balance of the recovered image.
assumptions (5)
  • domain assumption Distance-driven ray tracer accurately models PCD pixel response including charge sharing
    Invoked in Section II.C and the Discussion; ringing artifacts in Fig. 3 and threshold-dependent ringing in the Discussion indicate the model does not fully capture charge sharing.
  • domain assumption Pseudo-laminographic geometry with negligible parallax holds for the acquisition trajectories
    Stated in Section II.C ('A pseudo-laminographic inverse problem, with negligible parallax'); if false, the forward model used in reconstruction is inaccurate.
  • domain assumption ECC registration provides accurate sub-pixel shift estimates from image content alone, without mechanical precision
    Stated in the Abstract and Section II.C; the entire supersampling relies on registration accuracy.
  • domain assumption STC calibration maps counts to equivalent thickness and compensates for inter-pixel spectral variations
    Invoked in Section II.C; relies on the calibration material (PMMA) having attenuation behavior close to the imaged object.
  • standard math Poisson noise model and MLEM convergence as the basis for the iterative reconstruction
    MLEM is standard (Shepp-Vardi); the paper invokes it in Section II.C without derivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Super-Resolution Radiography by Mechanical Supersampling and Model-Based Iterative Reconstruction using High-Z Photon-Counting Detectors." pith.science (2026). https://pith.science/paper/AVNKWDQL

@misc{pith2026250720793,
  author       = {Pith},
  title        = {Pith review of: Super-Resolution Radiography by Mechanical Supersampling and Model-Based Iterative Reconstruction using High-Z Photon-Counting Detectors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AVNKWDQL}},
  note         = {Machine review of arXiv:2507.20793}
}
read the original abstract

This study presents a practical and dose-efficient strategy for resolution enhancement in planar radiography, based on mechanically supersampled acquisition with high-Z photon-counting detectors (PCDs). Unlike prior event-based or cluster methods, our approach operates in true photon-counting mode and supports clinical flux rates. Using detector trajectories spanning multiple pixels and image registration-based shift estimation, we achieve sub-pixel sampling without requiring mechanical precision, while also compensating for motion and geometric instabilities. An iterative reconstruction framework based on Maximum Likelihood Expectation Maximization (MLEM) with a distance-driven ray model further enhances resolution and noise robustness. Long-range supersampling additionally mitigates pixel defects and spectral inhomogeneities inherent to high-Z detectors. Phantom studies demonstrate substantial resolution improvement and image uniformity. In comparison with a clinical mammography system, the method reveals sharper detail and more homogeneous contrast at comparable or reduced dose. The resolution gain also reduces the need for geometric magnification, enabling smaller and more cost-effective PCDs. These results establish mechanically supersampled radiography as a clinically viable approach for micron-scale imaging, with strong potential for digital mammography and other high-resolution applications and with scan times compatible with clinical workflow.

Figures

Figures reproduced from arXiv: 2507.20793 by the authors.

Figure 1
Figure 1. Schematic of the proposed imaging strategy. Short range supersampling and consecutive iterative reconstruction results in spatial resolution enhancement, whereas long range supersampling additionally reduces geometrical distortions and improves spectral image fidelity. I. INTRODUCTION hoton-counting detectors (PCDs) offer significant benefits over energy-integrating detectors (EIDs), particularly in terms of spatial… view at source ↗
Figure 3
Figure 3. Scan of micro-CT resolution phantom; prototype DECTRIS SANTIS GaAs (1280x512 px), TH40kV; Tube MXR-Microbox, 100kVp, 100muA, magnification ~2x. Comparison of phantom scans without (a) and with super-sampling (b) acquired with identical photon statistics. In both cases 21x21 frames were acquired, with and without super-sampling motion (total displacement of ~1 pixel), respectively. The effect of the spatial resolutio… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

  1. [12]

    Iterative CBCT reconstruction - algorithms for a spectroscopic Medipix -Micro-CT, https://freidok.uni -freiburg.de/data/9562,

    M. P. Pichotka, “Iterative CBCT reconstruction - algorithms for a spectroscopic Medipix -Micro-CT, https://freidok.uni -freiburg.de/data/9562,” 2014

    work page 2014

  2. [14]

    Dual energy CT inspection of a carbo n fibre reinforced plastic composite combined with metal components,

    D. Vavrik, J. Jakubek, I. Kumpova, and M. Pichotka, “Dual energy CT inspection of a carbo n fibre reinforced plastic composite combined with metal components,” Case Studies in Nondestructive Testing and Evaluation , vol. 6, pp. 47 –55, Nov. 2016

    work page 2016

  3. [1]

    The medipix3RX: A high resolution, zero dead -time pixel detector readout chip allowing spectroscopic imaging,

    R. Ballabriga, J. Alozy, G . Blaj, M. Campbell, M. Fiederle, E. Frojdh, E. H. M. Heijne, X. Llopart, M. Pichotka, S. Procz, L. Tlustos, and W. Wong, “The medipix3RX: A high resolution, zero dead -time pixel detector readout chip allowing spectroscopic imaging,” Journal of Instrumenta tion , vol. 8, no. 2, Feb. 2013

    work page 2013

  4. [2]

    Photon -counti ng MCP/Timepix detectors for soft X -ray imaging and spectroscop ic applications,

    A. S. Tremsin, J. V. Vallerga, O. H. W. Siegmund, J. Woods, L. E. De Long, J. T. Hastings, R. J. Koch, S. A. Morley, Y. De Chuang, and S. Roy, “Photon -counti ng MCP/Timepix detectors for soft X -ray imaging and spectroscop ic applications,” J Synchrotron Radiat , vol. 28, no. Pt 4, pp. 1069 –1080, Jul. 2021

    work page 2021

  5. [3]

    Subpixel resolution in CdTe Timepix3 pixel detectors,

    M. Khalil, E. S. Dreier, J. Kehres, J. Jakubek, and U. L. Olsen, “Subpixel resolution in CdTe Timepix3 pixel detectors,” J Synchrotron Radiat , vol. 25, no. 6, pp. 1650 –1657, Nov. 2018

    work page 2018

  6. [4]

    Semiconductor materials for X -ray detectors,

    D. Pennicard, B. Pirard, O. Tolbanov, and K. Iniewski, “Semiconductor materials for X -ray detectors,” MRS Bull, vol. 42, no. 6, pp. 445 –450, Jun. 2017

    work page 2017

  7. [5]

    Overview of GaAs und CdTe Pixel Detectors Using Medipix Electronics,

    M. Fiederle, S. Procz, E. Hamann, A. Fauler, and C. Fröjdh, “Overview of GaAs und CdTe Pixel Detectors Using Medipix Electronics,” Crystal Research and Technology , vol. 55, no. 9, Sep. 2020

    work page 2020

  8. [6]

    Photon counting multienergy x -ray imaging: Effect of the characteris tic x rays on detector performance,

    P. M. Shikhaliev, S. G. Fritz, and J. W. Chapman, “Photon counting multienergy x -ray imaging: Effect of the characteris tic x rays on detector performance,” Med Phys, vol. 36, no. 11, pp. 5107 –5119, 2009

    work page 2009

Show all 20 references
  1. [7]

    Pilot study on high - resolution radiological methods for the analysis of cerebrospinal fluid (CSF) shunt valves,

    M. P. Pichotka, M. Weigt, M. J. Shah, M. F. Russe, T. Stein, T. Billoud, J. Beck, J. Straehle, C. L. Schlett, D. v. Elverfeldt, and M. Reisert , “Pilot study on high - resolution radiological methods for the analysis of cerebrospinal fluid (CSF) shunt valves,” Z Med Phys , 2023

    2023

  2. [8]

    Data processing and image reconstruction methods for pixel detectors,

    J. Jakubek, “Data processing and image reconstruction methods for pixel detectors,” Nucl Instrum Methods Phys Res A, vol. 576, no. 1, pp. 223 –234, 2007

    2007

  3. [9]

    Radiogram enhancement and linearization using the beam hardening correction method,

    D. Vavrik and J. Jakubek, “Radiogram enhancement and linearization using the beam hardening correction method,” Nucl Instrum Methods Phys Res A , vol. 607, no. 1, pp. 212 –214, 2009

    2009

  4. [10]

    Parametri c image alignment using enhanced correlation coeffici ent maximization,

    G. D. Evangelidis and E. Z. Psa rakis, “Parametri c image alignment using enhanced correlation coeffici ent maximization,” IEEE Trans Pattern Anal Mach Intell , vol. 30, no. 10, pp. 1858 –1865, 2008

    2008

  5. [11]

    Distance -driven projection and backprojection in three dimensio ns,

    B. De Man and S. Basu, “Distance -driven projection and backprojection in three dimensio ns,” Phys Med Biol, vol. 49, no. 11, pp. 2463 –2475, Jun. 2004

    2004

  6. [13]

    Spectroscopic micro-tomography of metallic -organic composites by means of photon -counting detectors,

    M. Pichotka, J. Jakubek, and D. Vavri k, “Spectroscopic micro-tomography of metallic -organic composites by means of photon -counting detectors,” Journal of Instrumentation , vol. 10, no. 12, Dec. 2015

    2015

  7. [15]

    Super - resolution X -ray imaging with hybrid pixel detector s using electromagnetic source stepping,

    T. Dreier, U. Lundström, and M. Bech, “Super - resolution X -ray imaging with hybrid pixel detector s using electromagnetic source stepping,” Journal of Instrumentation , vol. 15, no. 3, p. C03002, Mar. 2020

    2020

  8. [16]

    Accuracy of cone -beam computed tomography, digital mammography and digital breast tomosynthesis for microcalcifications and margins to microcalcifications in breast specimens,

    C. Neubauer, J. S. Yilmaz, P. Bronsert, M. Pichotka, F. Bamberg, M. Windfuhr -Blum, T. Erbes , and J. Neubauer, “Accuracy of cone -beam computed tomography, digital mammography and digital breast tomosynthesis for microcalcifications and margins to microcalcifications in breas...

    2022

  9. [17]

    Edge pixel response studies of edgeless silicon sensor technology for pixellated imaging detectors,

    D. Maneu ski, R. Bates, A. Blue, C. Buttar, K. Doonan, L. Eklund, E. N. Gimenez, D. Hynds, S. Kachkanov, J. Kalliopuska, T. McMullen, V. O’Shea, N. Tartoni, R. Plackett, S. Vahanen, and K. Wraight, “Edge pixel response studies of edgeless silicon sensor technology for pixellat...

    2015

  10. [18]

    Digital breast tomosynthesis: State of the art1,

    S. Vedantham, A. Karellas, G. R. Vijayaraghavan, and D. B. Kopans, “Digital breast tomosynthesis: State of the art1,” Radiology , vol. 277, no. 3, pp. 663–684, Dec. 2015

    2015

  11. [19]

    A review of breast tomosynthesis. Part I. The image acquisition process,

    I. Sechopoulos, “A review of breast tomosynthesis. Part I. The image acquisition process,” Med Phys , vol. 40, no. 1, 2013

    2013

  12. [20]

    Ultra -high resolution, multi -scale, c ontext -aware approach for detection of small cancers on mammography,

    K. Rangarajan, A. Gupta, S. Dasgupta, U. Marri, A. K. Gupta, S. Hari, S. Banerjee, and C. Arora, “Ultra -high resolution, multi -scale, c ontext -aware approach for detection of small cancers on mammography,” Scientific Reports 2022 12:1 , vol. 12, no. 1, pp. 1 –8, Jul. 2022

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.