pith. sign in

arxiv: 2605.17343 · v1 · pith:QSPGTFVEnew · submitted 2026-05-17 · 💻 cs.CV

GraphMAR: Geometry-Aware Graph Learning Framework for Spatially Adaptive CT Metal Artifact Reduction

Zilong Li , Chenglong Ma , Yiming Lei , Yuanlin Li , Jing Han , Jiannan Liu , Huidong Xie , Junping Zhang
show 2 more authors
Yi Zhang Hongming Shan
This is my paper

Pith reviewed 2026-05-20 13:56 UTC · model grok-4.3

classification 💻 cs.CV
keywords CT metal artifact reductiongraph learningmixture of expertsartifact localizationimage domain MARgeometric modelingspatially adaptive processing
0
0 comments X

The pith

GraphMAR aligns routing maps with geometric density graphs from metal masks to localize and reduce CT artifacts in the image domain

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

This paper develops GraphMAR to reduce metal-induced streaking artifacts in CT scans using only reconstructed images rather than raw projection data. It builds a geometric graph directly from the metal mask to identify regions where artifacts are likely to form based on the relative positions of implants. A graph-routed mixture-of-experts network then processes different spatial areas with specialized sub-networks while aligning its internal routing decisions to the geometric map. The result is both improved restoration and an explicit, readable indication of where and why artifacts were corrected. Readers would care because this makes advanced artifact handling feasible in routine clinical workflows that lack access to sinogram traces.

Core claim

The paper claims that a geometric density graph constructed from the metal mask in the reconstructed image serves as a workable image-domain stand-in for sinogram metal traces, enabling a GraphMoE module to build a polar-coordinate artifact graph in feature space, route experts adaptively across regions, and produce explicit localization by aligning learned routing maps with the geometric prior.

What carries the argument

The geometric density graph derived from inter-implant geometry in the metal mask, which localizes artifact-prone regions, together with GraphMoE, the graph-routed mixture-of-experts module that builds a polar-coordinate artifact graph to adaptively assign experts to different spatial regions.

If this is right

  • Region-adaptive processing yields higher restoration quality than uniform image-domain methods on both simulated and real-world datasets.
  • Alignment of routing maps with the geometric density graph supplies explicit, interpretable artifact localization in the image domain.
  • The framework operates without raw sinogram data, making it applicable to standard clinical reconstructed images.
  • Graph-based modeling introduces a new mechanism for handling spatially varying artifact patterns induced by multiple implants.

Where Pith is reading between the lines

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

  • The same mask-to-graph construction could be tested on artifacts from non-metallic high-density objects such as bone or contrast agents.
  • Routing maps produced by the method could serve as attention priors for other downstream tasks like segmentation in metal-affected regions.
  • Extending the polar-coordinate artifact graph to include temporal information might support correction in dynamic or 4D CT acquisitions.

Load-bearing premise

A geometric graph built solely from the metal mask in the reconstructed image can reliably identify artifact-prone regions without access to the original projection measurements.

What would settle it

A test set of real CT scans with precisely measured metal implant positions where the geometric density graph shows no correspondence to the actual locations or severity of streaking artifacts after reconstruction.

Figures

Figures reproduced from arXiv: 2605.17343 by Chenglong Ma, Hongming Shan, Huidong Xie, JianNan Liu, Jing Han, Junping Zhang, Yiming Lei, Yi Zhang, Yuanlin Li, Zilong Li.

Figure 1
Figure 1. Figure 1: Illustration of different CT MAR paradigms. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of GraphMAR, which contains multiple GraphMoE modules integrated into the backbone model. First, a polar-coordinate artifact graph is [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The procedure of constructing the geometric graph and geometric [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the polar coordinate and the two types of edges in the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison of different methods on the DDMAR dataset. Methods that are competitive in Table [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of different methods on the simulated dental dataset. Methods that are competitive in Table [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of different methods on the real-world dental CT with severe artifacts. Methods that are competitive in Table [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison of different configurations of GraphMAR in Table [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: shows the artifact attention GA produced by Graph￾MoE, which highlights artifact-affected regions and aligns well with human observation. This provides an interpretable indication of where the model focuses during MAR, improving the reliability of MAR results. Furthermore, GA enables selective restoration that con￾ventional end-to-end MAR methods cannot offer: existing methods process the entire image uni… view at source ↗
Figure 11
Figure 11. Figure 11: Clinical application of artifact attention. The artifact attention [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

Computed tomography (CT) metal artifact reduction (MAR) aims to reduce the severe streaking artifacts induced by metallic implants and other high-density objects. Effective MAR generally requires both accurate artifact localization and artifact removal. Sinogram-domain methods can exploit explicit geometric cues, such as metal traces, to identify metal-corrupted measurements, while requiring raw projection data, which is often unavailable in clinical and practical scenarios. Image-domain methods are more flexible and widely applicable, yet they usually lack comparable geometric guidance, limiting their ability to localize artifacts and leading to suboptimal results. To address this limitation, we propose GraphMAR, a geometry-aware learning framework for explicit artifact identification and spatially adaptive MAR in the image domain. The key idea is to introduce graph-based geometric modeling as an image-domain analogue of sinogram metal traces. Specifically, we first construct a geometric graph from the metal mask and derive a geometric density graph that coarsely localizes artifact-prone regions according to inter-implant geometry. We then design GraphMoE, a graph-routed mixture-of-experts module that builds a polar-coordinate artifact graph in feature space and adaptively routes different experts to different spatial regions for MAR. By aligning the learned routing maps with the geometric density graph, GraphMAR provides explicit and interpretable artifact localization while enabling region-adaptive artifact reduction. Experiments on both simulated and real-world datasets demonstrate that GraphMAR achieves superior MAR performance compared with existing methods. To the best of our knowledge, this is the first work to introduce graph-based modeling for CT MAR and to enable explicit artifact identification in the image domain, improving both restoration quality and interpretability.

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 manuscript proposes GraphMAR, a geometry-aware graph learning framework for CT metal artifact reduction (MAR) in the image domain. It constructs a geometric density graph from the metal mask to localize artifact-prone regions based on inter-implant geometry and introduces the GraphMoE module, which builds a polar-coordinate artifact graph in feature space to adaptively route experts for region-specific artifact reduction. By aligning learned routing maps with the geometric density graph, the method aims to provide explicit and interpretable artifact localization. Experiments on simulated and real-world datasets are claimed to show superior performance over existing methods, positioning it as the first graph-based approach for CT MAR.

Significance. If the results hold, this work could significantly advance image-domain MAR by providing a geometric modeling approach that mimics sinogram-domain cues without needing raw projection data. The explicit interpretability through routing maps aligned with geometric density is a strength, potentially improving both restoration quality and clinical trust. The introduction of graph-based modeling and GraphMoE for adaptive processing represents a novel direction in the field.

major comments (1)
  1. [Abstract and §3] Abstract and §3 (geometric density graph construction): The assumption that a geometric density graph built solely from the metal mask in the reconstructed image serves as a reliable image-domain analogue to sinogram metal traces is load-bearing for the central claim of explicit and interpretable artifact localization. Artifact streaks arise from specific ray paths under CT geometry and polychromatic spectrum; the image-domain graph using only inter-implant distances or densities omits scanner angles, source-detector geometry, and attenuation coefficients. This is least secure for multi-implant or off-center cases, as highlighted by the stress-test concern. A direct validation (e.g., overlap with ground-truth artifact maps from projection data or ablation across implant configurations) is needed to support the claim.
minor comments (2)
  1. [Abstract] Abstract: The statement 'to the best of our knowledge, this is the first work to introduce graph-based modeling for CT MAR' requires a more exhaustive related-work discussion to substantiate novelty.
  2. [Experiments] Experiments section: Claims of superior performance should include specific quantitative metrics (e.g., PSNR, SSIM), error bars, statistical significance tests, and ablation studies on the geometric density graph and GraphMoE components.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the constructive major comment. We address the concern regarding the geometric density graph point by point below.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (geometric density graph construction): The assumption that a geometric density graph built solely from the metal mask in the reconstructed image serves as a reliable image-domain analogue to sinogram metal traces is load-bearing for the central claim of explicit and interpretable artifact localization. Artifact streaks arise from specific ray paths under CT geometry and polychromatic spectrum; the image-domain graph using only inter-implant distances or densities omits scanner angles, source-detector geometry, and attenuation coefficients. This is least secure for multi-implant or off-center cases, as highlighted by the stress-test concern. A direct validation (e.g., overlap with ground-truth artifact maps from projection data or ablation across implant configurations) is needed to support the claim.

    Authors: We thank the referee for this insightful observation. We agree that the geometric density graph is an approximation that encodes inter-implant geometry from the metal mask but does not explicitly incorporate scanner angles, source-detector geometry, or polychromatic effects. Our design choice is motivated by the fact that, in the image domain, streaking artifacts predominantly manifest along paths connecting high-density implants, which the graph structure directly captures via distances and densities. Nevertheless, to strengthen the central claim of explicit localization, we will revise the manuscript to include additional validation experiments on the simulated datasets. These will report quantitative overlap (e.g., Dice or IoU) between the geometric density graph and ground-truth artifact maps derived from projection data, together with ablations across multi-implant and off-center configurations. We believe this will provide the requested direct evidence while preserving the method's applicability to real-world image-only scenarios. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper introduces novel components including a geometric density graph constructed from the metal mask, a polar-coordinate artifact graph, and the GraphMoE module. These are defined from the input metal mask and inter-implant geometry without reducing to fitted target quantities or prior self-citations by construction. The central claims rest on explicit modeling choices and experimental results rather than tautological redefinitions or load-bearing self-citations. This is a standard case of a self-contained proposal against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the premise that graph structures derived from metal masks can substitute for sinogram information and on the design of the GraphMoE module; these are introduced without independent external validation in the abstract.

axioms (1)
  • domain assumption A geometric graph derived from the metal mask can coarsely localize artifact-prone regions according to inter-implant geometry
    Invoked as the key idea that enables image-domain analogue to sinogram metal traces.
invented entities (2)
  • geometric density graph no independent evidence
    purpose: Coarsely localizes artifact-prone regions from inter-implant geometry
    New construct derived from metal mask; no independent evidence supplied in abstract.
  • GraphMoE module no independent evidence
    purpose: Builds polar-coordinate artifact graph and adaptively routes experts to spatial regions
    Novel module introduced to achieve region-adaptive MAR; no external validation mentioned.

pith-pipeline@v0.9.0 · 5855 in / 1524 out tokens · 62409 ms · 2026-05-20T13:56:59.247652+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 2 internal anchors

  1. [1]

    Metal artifact reduction in CT: where are we after four decades?

    L. Gjestebyet al., “Metal artifact reduction in CT: where are we after four decades?”IEEE Access, vol. 4, pp. 5826–5849, 2016

    work page 2016

  2. [2]

    Aapm ct metal artifact reduction grand challenge,

    E. Haneda, N. Peters, J. Zhang, G. Karageorgos, W. Xia, H. Paganetti, G. Wang, Y . Guo, J. Ma, H. S. Park, K. Jeon, F. Fan, M. Thies, and B. De Man, “Aapm ct metal artifact reduction grand challenge,”Medical Physics, vol. 52, no. 10, p. e70050, 2025

    work page 2025

  3. [3]

    Deep sinogram completion with image prior for metal artifact reduction in CT images,

    L. Yu, Z. Zhang, X. Li, and L. Xing, “Deep sinogram completion with image prior for metal artifact reduction in CT images,”IEEE Trans. Med. Imaging, vol. 40, no. 1, pp. 228–238, 2020

    work page 2020

  4. [4]

    DuDoNet: Dual domain network for CT metal artifact reduction,

    W.-A. Linet al., “DuDoNet: Dual domain network for CT metal artifact reduction,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., 2019, pp. 10 512–10 521

    work page 2019

  5. [5]

    Encoding metal mask projection for metal artifact reduction in computed tomography,

    Y . Lyu, W.-A. Lin, H. Liao, J. Lu, and S. K. Zhou, “Encoding metal mask projection for metal artifact reduction in computed tomography,” inProc. Int. Conf. Med. Image Comput. Comput.-Assisted Intervention, 2020, pp. 147–157

    work page 2020

  6. [6]

    InDuDoNet: An interpretable dual domain network for CT metal artifact reduction,

    H. Wanget al., “InDuDoNet: An interpretable dual domain network for CT metal artifact reduction,” inProc. Int. Conf. Med. Image Comput. Comput.-Assisted Intervention, 2021, pp. 107–118

    work page 2021

  7. [7]

    Quad-Net: Quad-domain network for CT metal artifact reduction,

    Z. Li, Q. Gao, Y . Wu, C. Niu, J. Zhang, M. Wang, G. Wang, and H. Shan, “Quad-Net: Quad-domain network for CT metal artifact reduction,” IEEE Trans. Med. Imaging, pp. 1–1, 2024

    work page 2024

  8. [8]

    An iterative approach to the beam hardening correction in cone beam CT,

    J. Hsieh, R. C. Molthen, C. A. Dawson, and R. H. Johnson, “An iterative approach to the beam hardening correction in cone beam CT,”Med. Phys., vol. 27, no. 1, pp. 23–29, 2000

    work page 2000

  9. [9]

    CT metal artifact reduction method cor- recting for beam hardening and missing projections,

    J. M. Verburg and J. Seco, “CT metal artifact reduction method cor- recting for beam hardening and missing projections,”Phys. Med. Biol., vol. 57, no. 9, p. 2803, 2012

    work page 2012

  10. [10]

    Metal artifact reduction for polychromatic x-ray CT based on a beam-hardening corrector,

    H. S. Park, D. Hwang, and J. K. Seo, “Metal artifact reduction for polychromatic x-ray CT based on a beam-hardening corrector,”IEEE Trans. Med. Imaging, vol. 35, no. 2, pp. 480–487, 2015

    work page 2015

  11. [11]

    Generalized multi- dimensional adaptive filtering for conventional and spiral single-slice, multi-slice, and cone-beam CT,

    M. Kachelriess, O. Watzke, and W. A. Kalender, “Generalized multi- dimensional adaptive filtering for conventional and spiral single-slice, multi-slice, and cone-beam CT,”Med. Phys., vol. 28, no. 4, pp. 475– 490, 2001

    work page 2001

  12. [12]

    Reduction of CT artifacts caused by metallic implants

    W. A. Kalender, R. Hebel, and J. Ebersberger, “Reduction of CT artifacts caused by metallic implants.”Radiology, vol. 164, no. 2, pp. 576–577, 1987

    work page 1987

  13. [13]

    Normalized metal artifact reduction NMAR in computed tomography,

    E. Meyer, R. Raupach, M. Lell, B. Schmidt, and M. Kachelrieß, “Normalized metal artifact reduction NMAR in computed tomography,” Med. Phys., vol. 37, no. 10, pp. 5482–5493, 2010

    work page 2010

  14. [14]

    Convolutional neural network based metal artifact reduction in x-ray computed tomography,

    Y . Zhang and H. Yu, “Convolutional neural network based metal artifact reduction in x-ray computed tomography,”IEEE Trans. Med. Imaging, vol. 37, no. 6, pp. 1370–1381, 2018

    work page 2018

  15. [15]

    Unsupervised CT metal artifact learning us- ing attention-guidedβ-CycleGAN,

    J. Lee, J. Gu, and J. C. Ye, “Unsupervised CT metal artifact learning us- ing attention-guidedβ-CycleGAN,”IEEE Trans. Med. Imaging, vol. 40, no. 12, pp. 3932–3944, 2021

    work page 2021

  16. [16]

    ADN: artifact disentanglement network for unsupervised metal artifact reduction,

    H. Liao, W. Lin, S. K. Zhou, and J. Luo, “ADN: artifact disentanglement network for unsupervised metal artifact reduction,”IEEE Trans. Med. Imaging, vol. 39, no. 3, pp. 634–643, 2019

    work page 2019

  17. [17]

    Adaptive convolutional dictionary network for CT metal artifact reduction,

    H. Wang, Y . Li, D. Meng, and Y . Zheng, “Adaptive convolutional dictionary network for CT metal artifact reduction,”arXiv:2205.07471, 2022

    work page arXiv 2022

  18. [18]

    DICDNet: deep interpretable convolutional dictionary network for metal artifact reduction in CT images,

    H. Wang, Y . Li, N. He, K. Ma, D. Meng, and Y . Zheng, “DICDNet: deep interpretable convolutional dictionary network for metal artifact reduction in CT images,”IEEE Trans. Med. Imaging, vol. 41, no. 4, pp. 869–880, 2021

    work page 2021

  19. [19]

    Dual-domain adaptive-scaling non-local network for CT metal artifact reduction,

    T. Wanget al., “Dual-domain adaptive-scaling non-local network for CT metal artifact reduction,” inProc. Int. Conf. Med. Image Comput. Comput.-Assisted Intervention, 2021, pp. 243–253. 13

    work page 2021

  20. [20]

    F2iflow for CT metal artifact reduction,

    J. Su, C. Wang, Y . Li, D. Liang, and K. Shang, “F2iflow for CT metal artifact reduction,”IEEE Trans. Comput. Imaging, 2024

    work page 2024

  21. [21]

    Radiologist-in-the-loop self-training for generalizable CT metal artifact reduction,

    C. Ma, Z. Li, Y . Li, J. Han, J. Zhang, Y . Zhang, J. Liu, and H. Shan, “Radiologist-in-the-loop self-training for generalizable CT metal artifact reduction,”IEEE Trans. Med. Imaging, 2025

    work page 2025

  22. [22]

    Fre- quency split metal artifact reduction (fsmar) in computed tomography,

    E. Meyer, R. Raupach, M. Lell, B. Schmidt, and M. Kachelrieß, “Fre- quency split metal artifact reduction (fsmar) in computed tomography,” Med. Phys., vol. 39, no. 4, pp. 1904–1916, 2012

    work page 1904

  23. [23]

    Frequency split metal artefact reduction in pelvic computed tomography,

    M. Lellet al., “Frequency split metal artefact reduction in pelvic computed tomography,”Eur. Radiol., vol. 23, pp. 2137–2145, 2013

    work page 2013

  24. [24]

    Normalized metal artifact reduction in head and neck computed tomography,

    M. M. Lellet al., “Normalized metal artifact reduction in head and neck computed tomography,”Invest. Radiol., vol. 47, no. 7, pp. 415– 421, 2012

    work page 2012

  25. [25]

    Vision gnn: An image is worth graph of nodes,

    K. Han, Y . Wang, J. Guo, Y . Tang, and E. Wu, “Vision gnn: An image is worth graph of nodes,”Proc. Adv. Neural Inf. Process. Syst., vol. 35, pp. 8291–8303, 2022

    work page 2022

  26. [26]

    Vision HGNN: An image is more than a graph of nodes,

    Y . Han, P. Wang, S. Kundu, Y . Ding, and Z. Wang, “Vision HGNN: An image is more than a graph of nodes,” inProc. IEEE/CVF Int. Conf. Comput. Vis., 2023, pp. 19 878–19 888

    work page 2023

  27. [27]

    Key-graph transformer for image restora- tion,

    B. Ren, Y . Li, J. Liang, R. Ranjan, M. Liu, R. Cucchiara, L. Van Gool, and N. Sebe, “Key-graph transformer for image restora- tion,”arXiv:2402.02634, 2024

    work page arXiv 2024

  28. [28]

    Dynamic attentive graph learning for image restoration,

    C. Mou, J. Zhang, and Z. Wu, “Dynamic attentive graph learning for image restoration,” inProc. IEEE/CVF Int. Conf. Comput. Vis., 2021, pp. 4328–4337

    work page 2021

  29. [29]

    Cross-scale internal graph neural network for image super-resolution,

    S. Zhou, J. Zhang, W. Zuo, and C. C. Loy, “Cross-scale internal graph neural network for image super-resolution,”Proc. Adv. Neural Inf. Process. Syst., vol. 33, pp. 3499–3509, 2020

    work page 2020

  30. [30]

    Scene graph driven text-prompt generation for image inpaint- ing,

    T. Shukla, P. Maheshwari, R. Singh, A. Shukla, K. Kulkarni, and P. Turaga, “Scene graph driven text-prompt generation for image inpaint- ing,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2023, pp. 759–768

    work page 2023

  31. [31]

    Graphfill: Deep image inpainting using graphs,

    S. Verma, A. Sharma, R. Sheshadri, and S. Raman, “Graphfill: Deep image inpainting using graphs,” inProc. Wint. Conf. Appli. Comput. Vis., 2024, pp. 4996–5006

    work page 2024

  32. [32]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” Proc. Int. Conf. Learn Represent., 2017

    work page 2017

  33. [33]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Ben- gio, “Graph attention networks,” Proc. Int. Conf. Learn Represent., 2018

    work page 2018

  34. [34]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,”Proc. Adv. Neural Inf. Pro- cess. Syst., vol. 25, 2012

    work page 2012

  35. [35]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Proc. Adv. Neural Inf. Process. Syst., vol. 30, 2017

    work page 2017

  36. [36]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gellyet al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  37. [37]

    Universal pre- training for generalizable incomplete-view ct reconstruction,

    C. Ma, Z. Li, J. He, J. Zhang, Y . Zhang, and H. Shan, “Universal pre- training for generalizable incomplete-view ct reconstruction,”Pattern Recognition, p. 113513, 2026

    work page 2026

  38. [38]

    Deep convo- lutional neural network for inverse problems in imaging,

    K. H. Jin, M. T. McCann, E. Froustey, and M. Unser, “Deep convo- lutional neural network for inverse problems in imaging,”IEEE Trans. Med. Imaging, vol. 26, no. 9, pp. 4509–4522, 2017

    work page 2017

  39. [39]

    Learning to distill global representation for sparse-view CT,

    Z. Li, C. Ma, J. Chen, J. Zhang, and H. Shan, “Learning to distill global representation for sparse-view CT,” inProc. IEEE/CVF Int. Conf. Comput. Vis., 2023, pp. 21 196–21 207

    work page 2023

  40. [40]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  41. [41]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. Image Process., vol. 13, no. 4, 2004

    work page 2004

  42. [42]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,”Proc. Adv. Neural Inf. Process. Syst., vol. 30, 2017

    work page 2017