Pith. sign in

REVIEW 4 major objections 7 minor 82 references

Splat-Based Metal Artifact Reduction in Cone-Beam CT via Compact Attenuation Modeling

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper argues that a Gaussian-splatting CBCT reconstruction with one scalar material parameter per Gaussian, chosen on a quadratic Bezier mass-attenuation manifold, embeds polychromatic beam physics and removes metal artifacts without…

desk verdict A solid, mask-free polychromatic Gaussian-splatting MAR method with a clever low-dimensional material model, but the speedup is overstated and the dental-filling application is not supported by the material basis or the experiments. read the letter →

arxiv 2608.04764 v1 pith:7LUJI7XT submitted 2026-08-05 cs.CV cs.GR

Kiseok Choi , Jaemin Cho , Inchul Kim , Min H. Kim This is my paper · ORCID
classification cs.CVcs.GR
keywords metalartifactreductioncone-beamCTGaussiansplattingpolychromaticforwardmodelmassattenuationcoefficientBeziermaterialmanifoldbeamhardeningmask-freereconstruction
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

Metal artifacts in cone-beam CT arise because real X-ray beams are polychromatic, so the standard monochromatic reconstruction assumption fails around high-attenuation objects. This paper argues that the fix can be built into the reconstruction itself: a Gaussian-splatting volume where each Gaussian carries a density and a single material parameter $u_i$, and where the energy-dependent mass attenuation is a quadratic Bezier curve over water, aluminum, and iron. Jointly optimizing geometry and material against a polychromatic forward model removes beam-hardening streaks without a metal mask or paired supervision. The reported payoff is higher 3D PSNR/SSIM than neural-field baselines and convergence in roughly 19-42 minutes instead of 1-2 hours. If the claims hold, physics-aware artifact reduction becomes practical for full-resolution CBCT while retaining sharp structure near metal.

What carries the argument

The load-bearing object is the compact material parameter $u_i \in [0,1]$ together with the quadratic Bezier MAC model of Equation (7): one scalar per Gaussian selects a mass-attenuation curve on the low-dimensional manifold spanned by water, aluminum, and iron curves from a standards database of attenuation data. This collapses energy-dependent material optimization to a single continuous variable, which makes the polychromatic forward projection of Equation (8) tractable inside splatting and removes the need to segment metal. A custom differentiable CUDA projection and voxelization pipeline backpropagates through the spectral sum, and the total loss pairs $L_1$ and SSIM projection terms with a total-variation term on reconstructed attenuation at effective energy.

What would settle it

Reconstruct a phantom that contains a piece of dental amalgam, whose silver and tin components absorb strongly in the 10-90 keV range, alongside clean aluminum and iron test pieces; if beam-hardening streaks remain around the amalgam while the calibration metals are artifact-free, the manifold assumption fails. A direct check is to plot the amalgam's mass-attenuation curve and measure its distance to the quadratic Bezier manifold used by the model.

Watch

Extended reading notes

Core claim

The central claim is that a mask-free, polychromatic Gaussian-splatting reconstruction can explain metal-induced beam hardening by giving every Gaussian primitive a compact material parameter. The attenuation of primitive $i$ at energy $E$ is written as $\delta_i \mu_\rho(u_i, E)$ with $\mu_\rho(u_i, E) = (1-u_i)^2 b_s(E) + 2(1-u_i) u_i b_m(E) + u_i^2 b_f(E)$, a quadratic Bezier interpolation of the mass attenuation curves of water, aluminum, and iron. Inserting this into the polychromatic Beer-Lambert projection $P(\hat{x}) = -\log \sum_E \eta(E) \exp(-\sum_i f_P(\hat{x}|p_i,\Sigma_i)\delta_i \mu_\rho(u_i,E))$ and optimizing $u_i$ along with positions, covariances, and densities lets the model attribute the nonlinearity to material composition rather than to corrupted projection bins. On three synthetic phantoms the method reports the highest PSNR3D (27.40-28.96 dB) and SSIM3D (0.993-0.997), and on real scans it removes streaks while preserving fine texture that the neural-field baselines oversmooth.

Load-bearing premise

The load-bearing premise is that every material in the scan has an X-ray-blocking curve across energies that lies exactly on the simple curved surface spanned by the water, aluminum, and iron curves, so that one number per Gaussian can represent any tissue or metal present.

Editorial extensions

If this is right

  • No metal mask or paired supervision is needed: the material parameter $u_i$ can interpolate continuously from soft tissue to metal, so the polychromatic model itself explains the beam hardening.
  • On the synthetic Lung, Teeth, and Broccoli phantoms the method reports the highest PSNR3D and SSIM3D among FDK, LIMAR, NMAR, the neural-field baselines, and supervised post-processing networks.
  • Reconstruction finishes in 18-42 minutes on an RTX A6000, roughly 3-6 times faster than the neural-field baselines that require over an hour.
  • Quality stays stable when the number of spectral samples varies from 7 to 63 and when tube voltage varies across 60, 90, and 120 kVp.
  • The method generalizes across metal types (iron, titanium, aluminum) and across real specimens with dense texture, multiple materials, and head-like structure.

Reading between the lines

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

  • If the Bezier manifold assumption is the real source of the speedup, the same one-scalar material encoding could transfer to spectral or dual-energy CT, where per-voxel material fractions are also low-dimensional; the paper does not make this connection.
  • Dental amalgam contains silver and tin, whose K-edges fall inside the 10-90 keV operating band and whose mass-attenuation curves bend away from the water-aluminum-iron manifold, so amalgam-bearing scans are the natural stress test where residual artifacts should persist even if the method works perfectly on iron and titanium.
  • The stability across spectral sample counts suggests that adding extra basis curves for K-edge materials would extend the method more than refining the spectrum estimate; this is a testable prediction, not a claim of the paper.
  • Because the pipeline is mask-free and self-supervised, it could in principle be applied to limited-angle or dynamic CBCT without retraining, though the paper explicitly leaves those settings as future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes a metal artifact reduction (MAR) method for cone-beam CT that replaces a neural field with a Gaussian-splatting representation and augments each Gaussian with a scalar material parameter u_i. The scalar parameter is mapped through a quadratic Bézier curve whose three control vectors are fixed mass attenuation coefficient (MAC) curves for water, aluminum, and iron (Eq. 7). A polychromatic forward projection (Eq. 8) is then used with a differentiable reconstruction pipeline, optimizing geometry, density, and material parameters jointly without a metal mask. The authors report improved PSNR/SSIM on three synthetic datasets and qualitative improvements on six real scans over FDK, LIMAR, NMAR, Polyner, Park et al., and supervised baselines, and they report substantially lower runtimes (Table 1). The supplementary material contains manual gradient derivations and additional experiments.

Significance. If fully substantiated, this would be a practical contribution: mask-free, splat-based polychromatic MAR that preserves fine structure while suppressing beam-hardening artifacts, with a modest runtime advantage over neural-field baselines. The work includes a physically motivated forward model, explicit CUDA-level gradient derivations, an ablation isolating the effect of the polychromatic model, and a new dataset. The significance is moderated, however, by the restrictive one-dimensional material manifold, the lack of any quantitative real-data evaluation, and the fact that the central claimed speedup is not supported by the paper's own timing table.

major comments (4)
  1. [Section 4, Eq. (7)] The one-parameter quadratic Bézier curve used as the material model can only represent MACs that are convex combinations of the three fixed control curves. Materials with a K-edge inside the operating spectrum—such as dental amalgam, which the Introduction explicitly motivates and which contains silver (K-edge 25.5 keV) and tin (K-edge 29.2 keV)—cannot be represented by any u_i because the control vectors contain no such edge. All evaluated metals (Fe, Ti, Al) have K-edges below 10 keV, so the experiments never exercise this failure mode. The Limitation section (Section 7) acknowledges that 'highly unusual materials' deviate, but this dismissal is not appropriate for a clinically common material that is a stated motivation. Please either restrict the applicability claim, add a synthetic or real experiment with an in-band K-edge material (e.g., iodine or a silver-tin alloy), or extend the basis to a higher-dimensional representation that can accommodate absorption edges.
  2. [Section 5.3, Table 1] The text states that 'our method achieves an order-of-magnitude speedup,' but the numbers in Table 1 do not support that wording. For Lung, Ours is 41m40s versus 1h14m34s for Polyner, a 1.8× speedup; for Teeth it is 1.8×; for Broccoli it is 3.8×. The largest speedup in the table is Broccoli (3.8×), and none of the nine scenes reaches 10×. Please revise the claim in Section 5.3 (and any abstract/summary wording that implies order-of-magnitude gains) to state the actual acceleration factors, or report timings that genuinely approach 10×.
  3. [Section 5.1 and Supplementary Section 6] The synthetic evaluation uses datasets generated by the authors' own prior pipeline [5], which shares the same polychromatic projection assumptions as the proposed forward model. This creates a circularity risk: the quantitative advantage on the synthetic scenes may partly reflect that the data was produced with the same physics and sampling conventions. The real-data evaluation is purely qualitative, with no reference-based metric. Please clarify what exactly is shared with [5], and add at least one independent source of quantitative validation—for example, a synthetic dataset from an independent simulator, or a quantitative real-data comparison using a registered metal-free scan as a reference (as is done for the qualitative figures).
  4. [Supplementary Section 6] The PSNR and SSIM metrics are computed 'over all regions excluding the metal mask.' Because one of the central claims is preservation of structure near metal boundaries, this evaluation protocol excludes the most relevant region of interest. Please report metrics in a dilated band around the metal mask (e.g., 5–10 voxels) in addition to the full non-metal region, and discuss any differences in the behavior of the compared methods in that band.
minor comments (7)
  1. [Section 4, Eq. (7)] The construction of the Bézier control vectors b_s, b_m, b_f is underspecified: it is not clear whether they are element-wise minima/medians/maxima across a selected set of NIST materials or specific materials chosen by hand. Please provide the exact definition and the list of materials used.
  2. [Figure 3] The caption refers to 'minimum, intermediate, and maximum MAC vectors' but the figure does not show which curves correspond to these vectors or how the approximation error is quantified. Adding an explicit error plot or a table of per-material approximation errors would make the claim more verifiable.
  3. [Eq. (8)] The notation P(\hat{x}, E) is confusing because E is both the summation index and an argument on the left-hand side; the right-hand side does not depend on E as a free variable after summation. Please use P(\hat{x}) or define E as the integration variable explicitly.
  4. [Table 1] The runtime comparison does not state the convergence criteria or the number of iterations used for each method. Without this, it is unclear whether the reported times reflect equivalent optimization effort. Please report iteration counts or a convergence threshold.
  5. [Section 5.3] The text says 'All measurements are performed on an Intel Xeon 4214R CPU and an NVIDIA RTX A6000 GPU,' but no version numbers or software versions are given for the CUDA/PyTorch stack, which can affect timing. Please add the exact environment.
  6. [Supplementary, Eqs. (2)–(5)] The variable C in the gradient expressions is not defined explicitly in the main text or in the supplementary. Although it is inferable from the projection formula, please provide a definition to avoid ambiguity.
  7. [References] Reference [5] has a title very similar to the current paper ('Splat-based metal artifact reduction in cone-beam CT via polychromatic modeling') and is by overlapping authors. Please clarify in Section 1 how this submission extends or differs from that prior work, to help readers place the novelty.

Circularity Check

2 steps flagged · score 4.0 of 10

Quantitative validation rests on the authors' own splat-based polychromatic synthetic pipeline and a material basis built from the same metals used in the tests, but real scans and honest limitations keep the reconstruction claim from collapsing into its inputs.

  1. self citation load bearing [Section 5.1 (Dataset) and Table 2]
    "We evaluate our method on three synthetic CBCT phantoms—Lung, Teeth, and Broccoli—constructed from the LIDC [2], X-plant [29], and ZCB100 [18] datasets, each containing inserted metal objects. The datasets are generated using the synthetic projection pipeline introduced in prior work [5]."

    Reference [5] is the same authors' prior work titled 'Splat-based metal artifact reduction in cone-beam ct via polychromatic modeling.' The present method is also a splat-based polychromatic forward model (Eq. 8). The synthetic projections that produce the quantitative PSNR/SSIM gains in Table 2 are generated by this same-family pipeline, so the evaluation partly measures how well the inverse model matches its own generative model rather than an external benchmark. The real-data experiments are qualitative only, so the quantitative superiority claim is load-bearing on this self-citation. This is a validation circularity, not a derivation collapse.

  2. other [Section 4 (Compact material attenuation), Section 5 (Implementation), Section 7 (Limitation)]
    "The Bézier MAC basis is constructed from the NIST [26] attenuation curves of water, iron, and aluminum, assuming an operating spectrum of 10–90 keV. ... To validate this formulation, we compare real MAC curves from NIST with their Bézier-approximated counterparts (Figure 3). The interpolated curves reproduce both the magnitude and the shape of the energy-dependent attenuation with high fidelity, indicating that a scalar parameter ui is sufficient to approximate the physically relevant variation of MACs."

    The representational claim is validated on the same NIST curves used to define the basis in Eq. (7): water and iron are endpoints, so their MACs are reproduced exactly by construction (u=0 and u=1), and aluminum is the intermediate control point used to shape the curve. No out-of-manifold material with a K-edge inside 10–90 keV (e.g., silver or tin in dental amalgam) is tested. Section 7 concedes the model 'cannot represent highly unusual materials.' The synthetic test metals Fe, Ti, Al and all real inserts are Fe, i.e., the same material family as the basis, so the successful experiments do not independently falsify the one-scalar sufficiency claim.

full rationale

The core reconstruction pipeline is not derived from its target: Eq. (8) is a physically motivated polychromatic Beer-Lambert forward model, and the per-Gaussian parameters are optimized against measured projections, not against the reference volumes. The reported gains over Polyner and Park et al. on real walnut/garlic/etc. scans provide some independent (though qualitative) support. However, the quantitative evidence is weakened by two self-referential components: (i) the synthetic benchmark images are produced by the authors' own prior splat-based polychromatic pipeline [5], and (ii) the compact MAC manifold is constructed from, and validated on, the same water-iron-aluminum family used in the synthetic and real metal tests. The paper's own Limitation section acknowledges that materials with MAC curves far from this manifold cannot be represented, which further confirms that the favorable numbers do not establish general clinical coverage (e.g., dental amalgam with in-band Ag/Sn K-edges). Because the method still has real-data grounding and the limitation is disclosed, the circularity is partial rather than total.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim depends on the low-dimensional MAC manifold assumption and on the trustworthiness of the self-generated synthetic evaluation, plus a handful of hand-set hyperparameters. No new physical entities are introduced; the scalar u per Gaussian is an optimization variable. The method also depends on the standard Beer-Lambert law, NIST MAC tables, and the Gaussian splatting projection approximation.

free parameters (4)
  • Loss weight λ0 = 0.25
    Used for SSIM loss in Eq. (10); chosen by hand for all experiments, no sensitivity study shown.
  • Loss weight λ1 = 3.0
    Used for TV loss in Eq. (10); chosen by hand, no sensitivity study shown.
  • Spectral bin count N = 15
    Chosen as operating point; Table 3 shows quality is stable across 7 to 63 bins.
  • Bezier basis MAC vectors = NIST curves for water, aluminum, iron over 10-90 keV
    Hand-chosen control points; matches the Fe, Ti, Al test metals used in the synthetic evaluation, making the benchmark partly matched to the model.
assumptions (5)
  • domain assumption The discrete polychromatic Beer-Lambert model (Eq. 8) with a single normalized source spectrum η(E) describes the measured intensity.
    Used in Section 4 and supplementary Eq. (1); the authors acknowledge in Limitation that the real spectrum is simulated with SPEKTR and cannot be measured directly, so spectrum error propagates into the whole reconstruction.
  • ad hoc to paper MAC curves of all materials in a target scan lie on the quadratic Bezier manifold spanned by water, iron, and aluminum.
    Eq. (7) and Figure 3; the paper asserts this as an empirical observation but provides no quantitative fit error, and its own Limitation admits unusual materials deviate.
  • domain assumption Gaussian splatting analytic projection (Eq. 6) accurately approximates the line integral of the attenuation field.
    Carried over from R2-Gaussian [39]; requires enough Gaussian primitives with appropriate covariances to represent compact metals and fine structures.
  • domain assumption The synthetic CBCT projection pipeline of prior work [5] produces realistic metal artifacts and a valid effective-energy ground truth.
    Section 5.1; the pipeline comes from a same-group paper and is not independently validated in this text.
  • domain assumption Effective-energy monochromatic ground truth with metal-mask exclusion is an appropriate evaluation protocol for MAR.
    Supplementary Section 6; this discards spectral information that the polychromatic model is designed to recover and can inflate agreement when the metal region is excluded.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Splat-Based Metal Artifact Reduction in Cone-Beam CT via Compact Attenuation Modeling." pith.science (2026). https://pith.science/paper/7LUJI7XT

@misc{pith2026260804764,
  author       = {Pith},
  title        = {Pith review of: Splat-Based Metal Artifact Reduction in Cone-Beam CT via Compact Attenuation Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7LUJI7XT}},
  note         = {Machine review of arXiv:2608.04764}
}
read the original abstract

X-ray computed tomography (CT) suffers from severe metal artifacts when high-attenuation objects such as dental fillings or orthopedic implants are present. These artifacts originate from the polychromatic nature of X-rays, where attenuation varies strongly with photon energy and material composition, breaking the monochromatic assumption used by conventional reconstruction algorithms. Recent neural rendering approaches attempt to address this mismatch through differentiable polychromatic projection models, but they still struggle with smoothness bias, loss of fine structures, and prohibitive computation when extended to large-scale cone-beam CT. We introduce a splat-based metal artifact reduction framework that incorporates a physically grounded polychromatic forward model into a continuous Gaussian representation for cone-beam CT. Each Gaussian encodes the energy-dependent attenuation of the underlying material using a compact material parameterization, which enables efficient joint optimization of geometric and material properties without relying on a metal mask. This compact attenuation formulation captures the essential variation across biological tissues and metallic implants, allowing our model to explain metal-induced nonlinearity while preserving high-frequency structure. Experiments on simulated and real cone-beam CT scans show that our method converges significantly faster and suppresses metal artifacts more effectively than existing reconstruction and neural field-based approaches.

Figures

Figures reproduced from arXiv: 2608.04764 by the authors.

Figure 1
Figure 1. Qualitative comparison of CBCT reconstructions for a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Method overview. Instead of using the conventional monochromatic attenuation assumption, we introduce an energy-dependent [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) Representative mass attenuation coefficients (MACs) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Axial (yellow), coronal (pink), and sagittal (green) slices from the three synthetic scenes are shown, with metal regions high [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Real-data reconstruction results. FDK [8] produces strong streaking and shading artifacts. LIMAR [15] reduces some of these effects but leaves noticeable residuals. OSCNet [33] fails to generalize to unseen acquisition conditions and produces incorrect intensity distri…
Figure 1
Figure 1. Figure 1: Energy spectra generated using the SPEKTR simulator for various CT system configurations. Each curve represents the [PITH_FULL_IMAGE:figures/full_fig_p012_1.png]
Figure 2
Figure 2. Figure 2: Evaluation of reconstruction robustness under varying tube voltages (60, 90, and 120 kVp). Despite substantial changes in the [PITH_FULL_IMAGE:figures/full_fig_p014_2.png]
Figure 3
Figure 3. Figure 3: Qualitative comparison on additional slices for real scenes. [PITH_FULL_IMAGE:figures/full_fig_p015_3.png]
Figure 4
Figure 4. Figure 4: Qualitative comparison on additional synthetic scenes. [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

82 extracted references · 68 canonical work pages

  1. [5]

    , author=

    Reduction of CT artifacts caused by metallic implants. , author=. Radiology , volume=

    work page

  2. [1]

    W. C. Röntgen , title =. Sitzungsberichte der Physikalisch-Medizinischen Gesellschaft zu Würzburg , volume =

  3. [2]

    Computer Graphics and Image Processing , volume=

    The reconstruction of fan-beam data by filtering the back-projection , author=. Computer Graphics and Image Processing , volume=. 1979 , publisher=

    1979

  4. [3]

    Journal of the Optical Society of America A , volume=

    Practical cone-beam algorithm , author=. Journal of the Optical Society of America A , volume=. 1984 , publisher=

    work page 1984

  5. [4]

    2001 , publisher=

    Principles of computerized tomographic imaging , author=. 2001 , publisher=

    2001

  6. [6]

    Medical physics , volume=

    Normalized metal artifact reduction (NMAR) in computed tomography , author=. Medical physics , volume=. 2010 , publisher=

    work page 2010

  7. [7]

    Unsupervised Polychromatic Neural Representation for

    Qing Wu and Lixuan Chen and Ce Wang and Hongjiang Wei and S Kevin Zhou and Jingyi Yu and Yuyao Zhang , booktitle=. Unsupervised Polychromatic Neural Representation for

    work page

  8. [8]

    Medical Physics , publisher=

    Implicit neural representation-based method for metal-induced beam hardening artifact reduction in X-ray CT imaging , author=. Medical Physics , publisher=

    work page

Show all 82 references
  1. [9]

    IEEE Transactions on Visualization and Computer Graphics , volume=

    EWA splatting , author=. IEEE Transactions on Visualization and Computer Graphics , volume=. 2002 , publisher=

    2002

  2. [10]

    Medical physics , volume=

    A streak artifact reduction algorithm in sparse-view CT using a self-supervised neural representation , author=. Medical physics , volume=. 2022 , publisher=

    2022

  3. [11]

    Physics in Medicine & Biology , volume=

    Metal artifact reduction method based on a constrained beam-hardening estimator for polychromatic x-ray CT , author=. Physics in Medicine & Biology , volume=. 2021 , publisher=

    2021

  4. [12]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    NAF: neural attenuation fields for sparse-view CBCT reconstruction , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2022 , organization=

    2022

  5. [13]

    CVPR , year=

    Structure-Aware Sparse-View X-ray 3D Reconstruction , author=. CVPR , year=

  6. [14]

    ECCV , year=

    Radiative gaussian splatting for efficient x-ray novel view synthesis , author=. ECCV , year=

  7. [15]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    R ^2 -Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  8. [16]

    arXiv preprint arXiv:2307.05801 , year=

    Differentiable forward projector for X-ray computed tomography , author=. arXiv preprint arXiv:2307.05801 , year=

    arXiv

  9. [17]

    Ultramicroscopy , volume=

    The ASTRA Toolbox: A platform for advanced algorithm development in electron tomography , author=. Ultramicroscopy , volume=. 2015 , publisher=

    2015

  10. [18]

    Optics express , volume=

    Fast and flexible X-ray tomography using the ASTRA toolbox , author=. Optics express , volume=. 2016 , publisher=

    2016

  11. [19]

    Biomedical Physics & Engineering Express , volume=

    TIGRE: a MATLAB-GPU toolbox for CBCT image reconstruction , author=. Biomedical Physics & Engineering Express , volume=. 2016 , publisher=

    2016

  12. [20]

    arXiv preprint arXiv:2404.03126 , year=

    Gaspct: Gaussian splatting for novel ct projection view synthesis , author=. arXiv preprint arXiv:2404.03126 , year=

    arXiv

  13. [21]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Intratomo: self-supervised learning-based tomography via sinogram synthesis and prediction , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  14. [22]

    GS-CPR: Efficient Camera Pose Refinement via 3D Gaussian Splatting , author=

  15. [23]

    NeurIPS , year=

    Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis , author=. NeurIPS , year=

  16. [24]

    arXiv preprint arXiv:2502.09563 , year=

    Self-Calibrating Gaussian Splatting for Large Field of View Reconstruction , author=. arXiv preprint arXiv:2502.09563 , year=

    arXiv

  17. [25]

    arXiv preprint arXiv:2502.19800 , year=

    No Parameters, No Problem: 3D Gaussian Splatting without Camera Intrinsics and Extrinsics , author=. arXiv preprint arXiv:2502.19800 , year=

  18. [26]

    , author=

    3d gaussian splatting for real-time radiance field rendering. , author=. ACM Trans. Graph. , volume=

  19. [27]

    Medical physics , volume=

    Empirical cupping correction: a first-order raw data precorrection for cone-beam computed tomography , author=. Medical physics , volume=. 2006 , publisher=

    2006

  20. [28]

    and Slaney, Malcolm , title =

    Kak, Avinash C. and Slaney, Malcolm , title =. 2001 , doi =

    2001

  21. [29]

    Online Geometric Calibration of Cone-Beam Computed Tomography for Arbitrary Imaging Objects , year=

    Meng, Yuanzheng and Gong, Hui and Yang, Xiaoquan , journal=. Online Geometric Calibration of Cone-Beam Computed Tomography for Arbitrary Imaging Objects , year=

  22. [30]

    A. H. Andersen and A. C. Kak , title =. Ultrasonic Imaging , volume =

  23. [31]

    Journal of Theoretical Biology , volume=

    Iterative methods for the three-dimensional reconstruction of an object from projections , author=. Journal of Theoretical Biology , volume=. 1972 , publisher=

    1972

  24. [32]

    1970 , issn =

    Algebraic Reconstruction Techniques (ART) for three-dimensional electron microscopy and X-ray photography , journal =. 1970 , issn =

    1970

  25. [33]

    Physics in Medicine & Biology , volume=

    Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization , author=. Physics in Medicine & Biology , volume=. 2008 , publisher=

    2008

  26. [34]

    2012 , month =

    Liu, Yan and Ma, Jianhua and Fan, Yi and Liang, Zhengrong , title =. 2012 , month =

    2012

  27. [35]

    Amir Beck and Marc Teboulle , title =

  28. [36]

    ICCV , year =

    Yoonwoo Jeong and Seokjun Ahn and Christopher Choy and Animashree Anandkumar and Minsu Cho and and Jaesik Park , title =. ICCV , year =

  29. [37]

    2020 , booktitle=

    NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis , author=. 2020 , booktitle=

    2020

  30. [38]

    ACM Trans

    Thomas M\"uller and Alex Evans and Christoph Schied and Alexander Keller , title =. ACM Trans. Graph. , issue_date =. 2022 , pages =

    2022

  31. [39]

    Attention is All you Need , volume =

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N and Kaiser, ukasz and Polosukhin, Illia , booktitle =. Attention is All you Need , volume =

  32. [40]

    MedNeRF: Medical Neural Radiance Fields for Reconstructing 3D-aware CT-Projections from a Single X-ray , booktitle =

    Abril Corona. MedNeRF: Medical Neural Radiance Fields for Reconstructing 3D-aware CT-Projections from a Single X-ray , booktitle =. 2022 , url =. doi:10.1109/EMBC48229.2022.9871757 , timestamp =

    2022

  33. [41]

    proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024 , year =

    Liu, Jiachen and Bai, Xiangzhi , title =. proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024 , year =

    2024

  34. [42]

    2015 , month =

    Ferrucci, Massimiliano and Leach, Richard K and Giusca, Claudiu and Carmignato, Simone and Dewulf, Wim , title =. 2015 , month =. doi:10.1088/0957-0233/26/9/092003 , url =

    2015 doi

  35. [43]

    Yang, Kai and Kwan, Alexander L. C. and Miller, DeWitt F and Boone, John M , title =. Medical Physics , issn =. 2006 , month =. doi:10.1118/1.2198187 , url =

    2006 doi

  36. [44]

    2023 , issn =

    Geometry calibration for a dental cone-beam CT system with an offset detector , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.precisioneng.2022.11.004 , url =

    2023 doi

  37. [45]

    Xu, Jingyan and Tsui, Benjamin M. W. , journal=. An Analytical Geometric Calibration Method for Circular Cone-Beam Geometry , year=

  38. [46]

    2016 , month =

    Ouadah, S and Stayman, J W and Gang, G J and Ehtiati, T and Siewerdsen, J H , title =. 2016 , month =. doi:10.1088/0031-9155/61/7/2613 , url =

    2016 doi

  39. [47]

    2008 , month =

    Kyriakou, Y and Lapp, R M and Hillebrand, L and Ertel, D and Kalender, W A , title =. 2008 , month =. doi:10.1088/0031-9155/53/22/001 , url =

    2008 doi

  40. [48]

    and Pfister, H

    Zwicker, M. and Pfister, H. and van Baar, J. and Gross, M. , journal=. EWA splatting , year=

  41. [49]

    arXiv preprint arXiv:1812.01537 , year=

    A micro lie theory for state estimation in robotics , author=. arXiv preprint arXiv:1812.01537 , year=

    arXiv

  42. [50]

    Kingma and Jimmy Ba , editor =

    Diederik P. Kingma and Jimmy Ba , editor =. Adam:. 3rd International Conference on Learning Representations,. 2015 , url =

    2015

  43. [51]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library , author =

  44. [52]

    Physical Principles of Medical Imaging , pages=

    Interaction of radiation with matter , author=. Physical Principles of Medical Imaging , pages=

  45. [53]

    2024 , publisher=

    Emerging Radiation Detection: Technology and Applications , author=. 2024 , publisher=

    2024

  46. [54]

    and Azevedo, Stephen G

    Champley, Kyle M. and Azevedo, Stephen G. and Seetho, Isaac M. and Glenn, Steven M. and McMichael, Larry D. and Smith, Jerel A. and Kallman, Jeffrey S. and Brown, William D. and Martz, Harry E. , journal=. Method to Extract System-Independent Material Properties From Dual-Ener...

  47. [55]

    Kevin and Luo, Jiebo , journal=

    Liao, Haofu and Lin, Wei-An and Zhou, S. Kevin and Luo, Jiebo , journal=. ADN: Artifact Disentanglement Network for Unsupervised Metal Artifact Reduction , year=

  48. [56]

    DICDNet: Deep Interpretable Convolutional Dictionary Network for Metal Artifact Reduction in CT Images , year=

    Wang, Hong and Li, Yuexiang and He, Nanjun and Ma, Kai and Meng, Deyu and Zheng, Yefeng , journal=. DICDNet: Deep Interpretable Convolutional Dictionary Network for Metal Artifact Reduction in CT Images , year=

  49. [57]

    OSCNet: Orientation-Shared Convolutional Network for CT Metal Artifact Learning , year=

    Wang, Hong and Xie, Qi and Zeng, Dong and Ma, Jianhua and Meng, Deyu and Zheng, Yefeng , journal=. OSCNet: Orientation-Shared Convolutional Network for CT Metal Artifact Learning , year=

  50. [58]

    The 31st International Joint Conference on Artificial Intelligence , year=

    Adaptive Convolutional Dictionary Network for CT Metal Artifact Reduction , author=. The 31st International Joint Conference on Artificial Intelligence , year=

  51. [59]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    InDuDoNet: An Interpretable Dual Domain Network for CT Metal Artifact Reduction , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2021 , organization=

    2021

  52. [60]

    Medical physics , volume=

    The lung image database consortium (LIDC) and image database resource initiative (IDRI): a completed reference database of lung nodules on CT scans , author=. Medical physics , volume=. 2011 , publisher=

    2011

  53. [61]

    www. X-plant. org-The CT database of plant organs , author=. 6th Symposium on X-ray Computed Tomography: Inauguration of the KU Leuven XCT Core Facility, Location: Leuven, Belgium , year=

  54. [62]

    PloS one , volume=

    Automatic synthesis of panoramic radiographs from dental cone beam computed tomography data , author=. PloS one , volume=. 2016 , publisher=

    2016

  55. [63]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Dudonet: Dual domain network for ct metal artifact reduction , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  56. [64]

    IEEE transactions on medical imaging , volume=

    Convolutional neural network based metal artifact reduction in X-ray computed tomography , author=. IEEE transactions on medical imaging , volume=. 2018 , publisher=

    2018

  57. [65]

    NVIDIA, July , volume=

    Cuda c programming guide , author=. NVIDIA, July , volume=

  58. [66]

    OpenGATE Documentation , year =

  59. [67]

    Atomic Data and Nuclear Data Tables , volume=

    X-ray attenuation cross sections for energies 100 eV to 100 keV and elements Z= 1 to Z= 92 , author=. Atomic Data and Nuclear Data Tables , volume=. 1988 , publisher=

    1988

  60. [68]

    Sains Malaysiana , volume=

    A review of common beam hardening correction methods for industrial X-ray computed tomography , author=. Sains Malaysiana , volume=

  61. [69]

    Physics in Medicine & Biology , volume=

    MicroCT with energy-resolved photon-counting detectors , author=. Physics in Medicine & Biology , volume=. 2011 , publisher=

    2011

  62. [70]

    arXiv preprint arXiv:2405.07047 , year=

    Solving energy-independent density for ct metal artifact reduction via neural representation , author=. arXiv preprint arXiv:2405.07047 , year=

    arXiv

  63. [71]

    ACM Transactions on Graphics (TOG) , volume=

    Neat: Neural adaptive tomography , author=. ACM Transactions on Graphics (TOG) , volume=. 2022 , publisher=

    2022

  64. [72]

    Biomedical Signal Processing and Control , volume=

    A discretized tomographic image reconstruction based upon total variation regularization , author=. Biomedical Signal Processing and Control , volume=. 2017 , publisher=

    2017

  65. [73]

    Total Variation and Tight Frame Iteration Regularization Based Image Reconstruction for Short-scanning Few-view Computer Tomography , author=

  66. [74]

    Physics in Medicine & Biology , volume=

    Evaluation of hybrid SART+ OS+ TV iterative reconstruction algorithm for optical-CT gel dosimeter imaging , author=. Physics in Medicine & Biology , volume=. 2016 , publisher=

    2016

  67. [75]

    Quantitative imaging in medicine and surgery , volume=

    Quantification of metal artifacts in computed tomography: methodological considerations , author=. Quantitative imaging in medicine and surgery , volume=

  68. [76]

    Imaging Med , volume=

    CT artifacts: causes and reduction techniques , author=. Imaging Med , volume=

  69. [77]

    European Journal of Radiology , volume=

    Advances in metal artifact reduction in CT images: A review of traditional and novel metal artifact reduction techniques , author=. European Journal of Radiology , volume=. 2024 , publisher=

    2024

  70. [78]

    Medical Physics , volume=

    Technical Note: spektr 3.0—A computational tool for x-ray spectrum , author=. Medical Physics , volume=

  71. [79]

    Annals of the ICRP , volume=

    Basic anatomical and physiological data for use in radiological protection: reference values: ICRP Publication 89: Approved by the Commission in September 2001 , author=. Annals of the ICRP , volume=. 2002 , publisher=

    2001

  72. [80]

    The British journal of radiology , volume=

    Average soft-tissue and bone models for use in radiation dosimetry , author=. The British journal of radiology , volume=. 1987 , publisher=

    1987

  73. [81]

    IEEE transactions on medical imaging , volume=

    Multi-material decomposition using statistical image reconstruction for spectral CT , author=. IEEE transactions on medical imaging , volume=. 2014 , publisher=

    2014

  74. [82]

    Kim , title =

    Kiseok Choi and Inchul Kim and Jaemin Cho and Hyeongjun Cho and Min H. Kim , title =. Computer Graphics Forum (Proc. EUROGRAPHICS 2026) , year =

    2026

Pith tools

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