Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

Jiancheng Fang; Junlin Wang; Qiegen Liu; Shaoyu Wang; Weiwen Wu; Yikun Zhang

REVIEW 3 major objections 5 minor 55 references

Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Diff-NAF couples neural attenuation fields with a conditional diffusion model to achieve the best reported PSNR/SSIM on multi-source stationary CT reconstruction from 20 and 50 projection views.

desk verdict Legitimate NAF+diffusion iterative framework, but the headline 'best performance' claim is only fairly tested on 2 of 6 simulated datasets due to same-volume diffusion training. read the letter →

arxiv 2511.14310 v2 pith:AA3E73QV submitted 2025-11-18 cs.CV

Jiancheng Fang , Shaoyu Wang , Junlin Wang , Weiwen Wu , Yikun Zhang , Qiegen Liu This is my paper

classification cs.CV

keywords Ultra-sparse-viewCTreconstructionMulti-sourcestationaryNeuralattenuationfieldsConditionaldiffusionmodelsProjectionsynthesisPseudo-labelrefinementIterativeDR-domain

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

The paper claims that coupling a neural attenuation field (NAF) with a conditional diffusion model in the projection (DR) domain yields state-of-the-art 3D CT reconstructions from as few as 20 projection views. The framework—Diff-NAF—works by training a NAF on the sparse views, synthesizing new views with an angle-prior strategy, refining those synthetic projections with a dual-branch diffusion model, and feeding them back as pseudo-labels. On six datasets (simulated and real), the authors report higher PSNR/SSIM than FDK, SART, InTomo, NeRF, NeAT, NAF, and SAX-NeRF at both 20 and 50 views. The authors also claim this is the first deep coupling of NAF and diffusion in the DR-domain. A sympathetic reader would care because it suggests a practical route to fast, low-dose, stationary CT.

What carries the argument

The load-bearing components are (1) a Neural Attenuation Field (NAF) parameterizing the 3D attenuation volume as a continuous coordinate-to-attenuation function optimized against measured projections; (2) the DRPR module, a dual-branch conditional diffusion model that predicts both residual and noise to refine synthetic DR projections, paired with DRAT, an invertible affine transformation that aligns the dynamic range of synthetic projections with the diffusion model's expected input; and (3) APGPS, an angle-prior selection strategy that picks novel views away from already-known angles by maximizing a gradient-based dissimilarity score. The iterative reuse of refined projections as pseudo-la

What would settle it

Train the diffusion model on eight AAPM volumes and test on two never-seen volumes (with no full-view projections of the test volume in training), then compare Diff-NAF against SAX-NeRF under 20-view and 50-view settings. If the PSNR advantage on the held-out volumes falls below roughly 1 dB, the same-volume training protocol is likely responsible for the reported gains.

Watch

Extended reading notes

Core claim

Diff-NAF is presented as the first algorithm to deeply couple NAF and diffusion models in the DR-domain. The central mechanism is an iterative loop: a NAF trained on ultra-sparse projections synthesizes novel views; a Diffusion-driven Reuse Projection Refinement (DRPR) module—a dual-branch conditional diffusion model with dynamic range adaptation—corrects those synthetic projections; and the corrected projections are merged into the training set as pseudo-labels for the next NAF training round. The authors report that this loop outperforms seven baselines on every dataset tested, with gains from about 1 to 3 dB PSNR over the best prior method under 20-view and 50-view settings.

Load-bearing premise

The state-of-the-art claim depends on the diffusion model being trained on 720-view projections of the same volume later reconstructed (for four of six datasets), so the model has effectively seen the object's anatomy from other angles before reconstruction, making the test not fully out-of-sample.

Editorial extensions

If this is right

If Diff-NAF holds up, ultra-sparse-view CT with as few as 20 projections can produce reconstructions with PSNR/SSIM comparable to much denser sampling, enabling faster and lower-dose stationary CT.
The iterative pseudo-label refinement loop provides a template for combining physics-based implicit fields with generative priors in other inverse problems beyond CT.
The DRPR module's dual-branch diffusion, which jointly predicts residual and noise, offers a reusable component for projection-domain denoising and completion tasks.
APGPS's adaptive view selection can be used to choose informative new angles in any multi-view acquisition setting, potentially reducing the number of iterations needed.
The framework's success suggests that generative diffusion priors can serve as a general 'refiner' for any implicitly represented field, not just NAF.

Reading between the lines

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

The reported gains may be inflated for four of the six datasets because the diffusion model was trained on 720-view projections of the very same volume being reconstructed, meaning it has effectively memorized the anatomy from other angles; a fully held-out evaluation is needed to confirm generalization.
The same iterative coupling could be applied in the sinogram domain, as the authors mention for future work, which might improve inter-angle consistency and further stabilize ultra-sparse reconstruction.
A natural testable extension is to replace the diffusion refiner with a lighter-weight generative model (e.g., a normalizing flow) to see how much of the gain comes from the diffusion prior specifically versus the iterative pseudo-label loop.
The DRAT dynamic-range adaptation suggests that data distribution mismatches between synthetic and real projections are a key obstacle; similar invertible alignment layers could be dropped into other generative-refinement pipelines.

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.

Desk Editor's Note

read the letter

Here's the quick take: this is a legitimate and interesting combination of NAF and diffusion for sparse-view CT, but the headline empirical claim is only solidly supported by two of the six simulated datasets. The authors themselves disclose in Section IV-A that for jaw, head, box, and foot they train the diffusion model on 720-view projections of the exact same volume being reconstructed. Non-overlapping training/inference views do not remove the leak: the model has seen the object's full anatomy from other angles and can use that to correct NAF-synthesized projections. That makes the reported PSNR/SSIM gains on those four datasets partly a self-fulfilling prediction. The two AAPM volumes (L067, L096), trained on eight separate volumes, are a fair out-of-sample test, and the results there are strong—roughly 2–3 dB over the best baseline. The real chicken-wing data also looks reasonable, with the diffusion model trained on four other specimens and a held-out one, though it is qualitative only.

What is actually new: the iterative pseudo-label loop coupling NAF and a dual-branch residual/noise diffusion in the DR domain, plus the APGPS view-selection heuristic and the DRAT dynamic-range adaptation. The ablation studies show each component contributes something, and the source code is promised, which helps reproducibility. The paper would be worth refereeing seriously, but it needs a major revision: either redo the four single-volume datasets with a leave-one-out protocol, or report them as a self-supervised/within-object test and not as evidence of generalization.

Soft spots in proportion: the diffusion formulation (Eq. 18–23) is under-specified—\bar\alpha_t and \bar\beta_t are used without being defined, and the “β_tε” variance term is non-standard. No error bars anywhere, which matters for 2 dB claims. The comparison baselines are run by the authors, and there is no comparison to the most recent diffusion-based CT methods beyond those cited. The iterative scheme also has several free parameters (a=4, w1/w2, number of iterations) that appear picked without sensitivity analysis beyond the one loop-count curve.

Who this is for: people working on neural-field CT, generative priors for sparse-view reconstruction, and stationary CT hardware. They should read it with the same-volume caveat in mind. It deserves a serious referee, but not acceptance as-is.

Referee Report

3 major / 5 minor

Summary. Diff-NAF is an iterative framework that combines a Neural Attenuation Field (NAF) with a dual-branch conditional diffusion model for ultra-sparse-view CT reconstruction in multi-source stationary systems. The method alternates between training NAF on real sparse-view projections, generating novel-view projections via an angle-prior strategy (APGPS), refining those projections with a diffusion-based DRPR module, and adding the refined projections as pseudo-labels for the next NAF retraining round. Quantitative evaluation is reported for 20-view and 50-view simulated acquisitions on six volumes (two AAPM chest CT volumes and four single-volume datasets: box, foot, head, jaw), plus a qualitative real chicken-wing experiment. The paper claims state-of-the-art PSNR/SSIM results relative to FDK, SART, InTomo, NeRF, NeAT, NAF, and SAX-NeRF, and claims to be the first deep coupling of NAF and diffusion in the DR-domain.

Significance. If fully validated, the proposed integration of a generative diffusion prior with a physical neural attenuation field could be a useful contribution to ultra-sparse-view CT, where pure model-based methods blur details and pure generative methods break volumetric consistency. The paper has several strengths: the iterative pseudo-label loop is clearly described in Algorithm 1; the ablation study on the two AAPM volumes (Tables III--V, Figs. 11--12) is internally consistent and supports the claim that the DRPR module provides a real gain over NAF, NeRF, and SAX-NeRF baselines; and the source code is promised publicly. The concern is that the headline claim of 'best performance on multiple simulated volumes and real projection data' is not currently supported by the evidence, because four of the six simulated test volumes are also used to train the diffusion model, and the real-data evaluation has no baselines. In addition, the diffusion process underlying DRPR is under-specified, making the method difficult to reproduce. The central idea is defensible, but the claims and presentation need substantial revision.

major comments (3)

[Section IV-A, Tables I and II] The evaluation protocol for the box, foot, head, and jaw datasets leaks the target object into the diffusion model. The text states: 'For the jaw, head, box, and foot datasets, each provides a single volume. Due to limited data, the same 3D volume is used for training the diffusion model and for iterative reconstruction; however, the training and inference view sets are non-overlapping.' Because the DRPR diffusion model is trained on 720-view projections of the exact volume that is later reconstructed, it has already seen the full 3D attenuation field from other angles. Non-overlapping views do not remove this object-specific leak: during refinement of NAF-synthesized DR projections, the diffusion model can correct them using memorized anatomy from the same object, which none of the comparison methods can do. Consequently, the Diff-NAF entries for these four datasets in Tables I and II d
[Section III-C, Eqs. (18)--(23)] The DRPR diffusion model is not fully specified. Equation (19) writes P_t^res ~ N(α_t P_res, β_t ε), but β_t ε is not a covariance matrix: ε was introduced as a random noise term, not as a fixed variance parameter. Equation (20) introduces cumulative coefficients ᾱ_t and β̄_t without defining how they are computed from the schedules α_t and β_t. Equation (21) factorizes q(P_{1:T}|P_0,P_res) but does not give the form of q(P_t|P_{t-1},P_res) consistently with Eq. (20), and Eq. (22) uses differences (ᾱ_t − ᾱ_{t−1}) and (β̄_t − β̄_{t−1}) without a derivation of the reverse step. Finally, the loss weights λ_res and λ_ε in Eq. (23) are never specified. These are not cosmetic issues: the definition of the forward noising process and the reverse sampling rule are load-bearing for the generative model, and the current equations do not allow a reader to re-implement or verify the method. Pleas
[Section IV-B and Abstract] The abstract states that 'Experimental results on multiple simulated 3D CT volumes and real projection data demonstrate that Diff-NAF achieves the best performance under ultra-sparse-view conditions.' The real-data section, however, presents only reconstructions produced by Diff-NAF (Fig. 8) with no comparison to any baseline and no quantitative metric. A single qualitative display cannot demonstrate 'best performance' on real projection data. The authors should either add comparative reconstructions and quantitative assessment on the real dataset (even without ground truth, e.g., using established no-reference metrics or a baseline comparison) or revise the claim to say that the real-data result is a qualitative feasibility demonstration.

minor comments (5)

[Section III-B, Eq. (17)] The weights w1 and w2 are introduced as free parameters but no setting, schedule, or sensitivity analysis is given. Since the pseudo-label set grows with iterations while the real set stays fixed, it would be helpful to explain how these weights are chosen and whether the results are sensitive to them.
[Section III-C, Eq. (24)] The DRAT affine transform is described as 'invertible linear transformations' with a positive regularization term, but the estimation procedure for Γ and b from the dataset is not given. Please specify how these parameters are computed, e.g., from min/max or mean/standard deviation of the training projections.
[Section IV-C, APGPS ablation] The text describing APGPS says it is 'similar to central sampling but with learned offset adjustments,' but Eq. (26) uses a deterministic midpoint-centered interval with fixed a=4 and Eq. (27) selects the view with maximum gradient dissimilarity; there is no learned offset. Please align the description with the actual algorithm.
[Algorithm 1, line 8] The notation {I_pred^{(θ*)}}_{Nθ}^{u=1} is inconsistent: the sub/superscript indices do not match the definition of N_θ used elsewhere. Also, the detector-pixel count should be N_θ^* as in Eq. (13).
[Section IV-A] The text says 'five public datasets and one real projection dataset,' then enumerates the AAPM dataset plus four single volumes (jaw, head, box, foot) and the real data. Please clarify whether 'five public datasets' includes the four Open Science volumes individually and the AAPM dataset, because the phrasing is ambiguous.

Circularity Check

1 steps flagged · score 6.0 of 10

Same-volume DRPR training leaks target anatomy for 4 of 6 datasets; central 'best performance' claim rests on only two AAPM volumes.

fitted input called prediction [Section IV-A, Datasets and Evaluation Metrics]
"For all datasets, 720-view DR projections are used for training... For the jaw, head, box, and foot datasets, each provides a single volume. Due to limited data, the same 3D volume is used for training the diffusion model and for iterative reconstruction; however, the training and inference view sets are non-overlapping, so the DR projections used for training and those refined during iterative reconstruction come from different views."

The DRPR diffusion model is fitted to 720-view DR projections of the exact volume being reconstructed. During inference it refines NAF-synthesized novel-view projections of that same volume; the 'refined' projections and resulting pseudo-labels therefore carry object-specific information from the training set. Non-overlapping views do not remove this leakage: the model has already seen the object's anatomy from all other angles. Tables I and II for Box, Foot, Head, and Jaw thus do not provide an out-of-sample test of the claimed best performance; only the two AAPM volumes with separate training subjects do.

full rationale

The core algorithm (Eqs. 4-27, Algorithm 1) is internally self-contained: NAF training, APGPS view selection, and DRPR diffusion correction are defined without circular dependence among the equations. No derivation step reduces a predicted quantity to an input by construction. The circularity is in the evaluation protocol underpinning the abstract claim. Section IV-A states that for jaw, head, box, and foot, the same single 3D volume is used to train the DRPR diffusion model and to run iterative reconstruction. The diffusion model therefore has access to the target object's full-angular projection data, and its correction of novel projections is a fitted-input prediction rather than an independent generative output. The paper's appeal to non-overlapping view sets is insufficient because the training views cover the same object's anatomy from other angles. Consequently 4 of 6 datasets in Tables I and II cannot support the 'best performance' claim. The two AAPM test volumes, trained on eight separate volumes, and the held-out chicken-wing specimen provide fair evidence, so the paper is only partially circular. Score 6 reflects this central leakage without alleging circularity in the method's equations or self-citation.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the NAF representation, the diffusion prior, and the validity of pseudo-labels; none are derived from first principles. No new physical entities are postulated.

free parameters (7)

Diffusion schedules α_t, β_t = not specified
Eq (18)-(21) use α_t and β_t to control residual and noise diffusion, but no schedule formula or values are given; cumulative \barα_t, \barβ_t are never defined.
Loss weights λ_res, λ_ε = not specified
Eq (23) defines residual and noise losses with weights λ_res and λ_ε; numerical values not provided.
Weights w1, w2 (real vs pseudo projections) = not specified
Eq (17) weights measured and pseudo projections; values not given.
APGPS interval divisor a = 4
Eq (26) sets a=4 'based on past experience' (Section III-D).
DRAT scaling Γ, shift b = estimated per dataset
Eq (24)-(25) uses affine mapping with parameters estimated from dataset; no estimation procedure or regularization term specified.
Number of outer iterations K = 2-4
Fig. 12 shows PSNR/SSIM plateau after 2-3 iterations; practical choice of 2-4 iterations is a tuned parameter.
NAF optimization steps per iteration = 1500
Section IV-A: 'The density prediction network is optimized for 1,500 gradient steps in each outer iteration.'

assumptions (5)

standard math Generalized Beer-Lambert law (Eq 1)
Physical model for X-ray attenuation used to generate simulated projections and to define NAF ray integrals.
domain assumption NAF can represent a continuous attenuation field from sparse projections
Adopts Zha et al. NAF assumption that a coordinate MLP with hash encoding can fit the attenuation volume; Section III-B.
domain assumption A diffusion model trained on full-view DR projections can refine NAF-synthesized projections
The DRPR module assumes learned projection priors correct the missing-view synthesis; no proof or analysis of error distribution.
ad hoc to paper Refined pseudo-projections are valid training labels for NAF
The iterative loop (Algorithm 1) adds P_pseudo^(k) to the training set with only fixed weights; validity of pseudo-labels is assumed, not validated.
ad hoc to paper Forward diffusion process Eq (18)-(21) with residual P_res and cumulative coefficients is well-defined
The paper does not define \barα_t, \barβ_t or prove the marginal P_t formula; the variance notation 'β_tϵ' is not a covariance matrix.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model." pith.science (2026). https://pith.science/paper/AA3E73QV

@misc{pith2026251114310,
  author       = {Pith},
  title        = {Pith review of: Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AA3E73QV}},
  note         = {Machine review of arXiv:2511.14310}
}

read the original abstract

Multi-source stationary computed tomography (CT) has recently attracted attention for its ability to achieve rapid image reconstruction, making it suitable for time-sensitive clinical and industrial applications. However, practical systems are often constrained by ultra-sparse-view sampling, which significantly degrades reconstruction quality. Traditional methods struggle under ultra-sparse-view settings, where interpolation becomes inaccurate and the resulting reconstructions are unsatisfactory. To address this challenge, this study proposes Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework tailored for multi-source stationary CT under ultra-sparse-view conditions. Diff-NAF combines a Neural Attenuation Field representation with a dual-branch conditional diffusion model. The process begins by training an initial NAF using ultra-sparse-view projections. New projections are then generated through an Angle-Prior Guided Projection Synthesis strategy that exploits inter view priors, and are subsequently refined by a Diffusion-driven Reuse Projection Refinement Module. The refined projections are incorporated as pseudo-labels into the training set for the next iteration. Through iterative refinement, Diff-NAF progressively enhances projection completeness and reconstruction fidelity under ultra-sparse-view conditions, ultimately yielding high-quality CT reconstructions. Experimental results on multiple simulated 3D CT volumes and real projection data demonstrate that Diff-NAF achieves the best performance under ultra-sparse-view conditions.

Figures

Figures reproduced from arXiv: 2511.14310 by the authors.

**Figure 2.** Schematic of a multi-source stationary CT configuration. Multiple [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Motivation schematic. We consider a setting where NAF is trained on [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (9 more)

**Figure 4.** Figure 4: Overview of the proposed Diff-NAF framework. (a) Iterative reconstruction framework: starting from initial sparse-view projections, the NAF is [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 5.** Figure 5: Illustration of the dual-branch diffusion correction process. The [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: Qualitative CT reconstruction comparison under the 50-view setting. Representative slices reconstructed by eight methods are shown. The vertical [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Qualitative CT reconstruction comparison under the 20-view setting. Representative slices reconstructed by eight methods are shown. The vertical [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

**Figure 8.** Figure 8: CT reconstructions of a chicken wing placed in a glass tube under [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

**Figure 10.** Figure 10: Box plots of novel-view DR projection refinement. Statistics are [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

**Figure 9.** Figure 9: Residual analysis of novel-view DR projection synthesis under the [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

**Figure 11.** Figure 11: Ablation study results. Left: performance of the DRPR module [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]

**Figure 12.** Figure 12: Relationship between the number of outer iterations and quantitative [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

55 extracted references · 3 linked inside Pith

[1]

Clinical applications of cone-beam computed tomography in dental practice,

W. C. Scarfe, A. G. Farman, P. Sukovicet al., “Clinical applications of cone-beam computed tomography in dental practice,”Journal-Canadian Dental Association, vol. 72, no. 1, p. 75, 2006

2006
[2]

Signal sensing and reconstruction paradigms for a novel multi-source static computed tomography system,

A. Kowtal, A. Cramer, D. Wu, K. Yang, W. Krull, I. Gkioulekas, and R. Gupta, “Signal sensing and reconstruction paradigms for a novel multi-source static computed tomography system,” inICASSP 2020- 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 9274–9278

2020
[3]

Swinging multi-source industrial ct systems for aperiodic dynamic imaging,

W. Wu, H. Yu, C. Gong, and F. Liu, “Swinging multi-source industrial ct systems for aperiodic dynamic imaging,”Optics express, vol. 25, no. 20, pp. 24 215–24 235, 2017

2017
[4]

Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam ct,

J. Bian, J. H. Siewerdsen, X. Han, E. Y . Sidky, J. L. Prince, C. A. Pelizzari, and X. Pan, “Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam ct,”Physics in Medicine & Biology, vol. 55, no. 22, p. 6575, 2010

2010
[5]

Accurate image reconstruction from few-views and limited-angle data in divergent-beam ct,

E. Y . Sidky, C.-M. Kao, and X. Pan, “Accurate image reconstruction from few-views and limited-angle data in divergent-beam ct,”Journal of X-ray Science and Technology, vol. 14, no. 2, pp. 119–139, 2006

2006
[6]

Practical cone-beam algorithm,

L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beam algorithm,”Journal of the Optical Society of America A, vol. 1, no. 6, pp. 612–619, 1984

1984
[7]

Simultaneous algebraic reconstruc- tion technique (sart): a superior implementation of the art algorithm,

A. H. Andersen and A. C. Kak, “Simultaneous algebraic reconstruc- tion technique (sart): a superior implementation of the art algorithm,” Ultrasonic imaging, vol. 6, no. 1, pp. 81–94, 1984

1984
[8]

Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,

E. Y . Sidky and X. Pan, “Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,” Physics in Medicine & Biology, vol. 53, no. 17, p. 4777, 2008

2008

Show all 55 references

[9]

X-ray ct image reconstruction via wavelet frame based regularization and radon domain inpainting,

B. Dong, J. Li, and Z. Shen, “X-ray ct image reconstruction via wavelet frame based regularization and radon domain inpainting,”Journal of Scientific Computing, vol. 54, no. 2, pp. 333–349, 2013

2013
[10]

Low-dose x-ray ct reconstruction via dictionary learning,

Q. Xu, H. Yu, X. Mou, L. Zhang, J. Hsieh, and G. Wang, “Low-dose x-ray ct reconstruction via dictionary learning,”IEEE transactions on medical imaging, vol. 31, no. 9, pp. 1682–1697, 2012

2012
[11]

Image gradient l0-norm based piccs for swinging multi-source ct reconstruction,

H. Yu, W. Wu, P. Chen, C. Gong, J. Jiang, S. Wang, F. Liu, and H. Yu, “Image gradient l0-norm based piccs for swinging multi-source ct reconstruction,”Optics express, vol. 27, no. 4, pp. 5264–5279, 2019

2019
[12]

Deep-neural-network- based sinogram synthesis for sparse-view ct image reconstruction,

H. Lee, J. Lee, H. Kim, B. Cho, and S. Cho, “Deep-neural-network- based sinogram synthesis for sparse-view ct image reconstruction,”IEEE Transactions on Radiation and Plasma Medical Sciences, vol. 3, no. 2, pp. 109–119, 2018

2018
[13]

Sinogram interpolation for sparse- view micro-ct with deep learning neural network,

X. Dong, S. Vekhande, and G. Cao, “Sinogram interpolation for sparse- view micro-ct with deep learning neural network,” inMedical Imaging 2019: Physics of Medical Imaging, vol. 10948. SPIE, 2019, pp. 692– 698

2019
[14]

A sparse-view ct reconstruction method based on combination of densenet and decon- volution,

Z. Zhang, X. Liang, X. Dong, Y . Xie, and G. Cao, “A sparse-view ct reconstruction method based on combination of densenet and decon- volution,”IEEE transactions on medical imaging, vol. 37, no. 6, pp. 1407–1417, 2018

2018
[15]

Framing u-net via deep convolutional framelets: Application to sparse-view ct,

Y . Han and J. C. Ye, “Framing u-net via deep convolutional framelets: Application to sparse-view ct,”IEEE transactions on medical imaging, vol. 37, no. 6, pp. 1418–1429, 2018

2018
[16]

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,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 21 196–21 207

2023
[17]

Hybrid-domain neural network processing for sparse-view ct reconstruction,

D. Hu, J. Liu, T. Lv, Q. Zhao, Y . Zhang, G. Quan, J. Feng, Y . Chen, and L. Luo, “Hybrid-domain neural network processing for sparse-view ct reconstruction,”IEEE Transactions on Radiation and Plasma Medical Sciences, vol. 5, no. 1, pp. 88–98, 2020

2020
[18]

Freeseed: Frequency- band-aware and self-guided network for sparse-view ct reconstruc- tion,

C. Ma, Z. Li, J. Zhang, Y . Zhang, and H. Shan, “Freeseed: Frequency- band-aware and self-guided network for sparse-view ct reconstruc- tion,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2023, pp. 250–259

2023
[19]

Stabilizing deep tomographic recon- struction: Part a. hybrid framework and experimental results,

W. Wu, D. Hu, W. Cong, H. Shan, S. Wang, C. Niu, P. Yan, H. Yu, V . Vardhanabhuti, and G. Wang, “Stabilizing deep tomographic recon- struction: Part a. hybrid framework and experimental results,”Patterns, vol. 3, no. 5, 2022

2022
[20]

Drone: Dual-domain residual-based optimization network for sparse-view ct reconstruction,

W. Wu, D. Hu, C. Niu, H. Yu, V . Vardhanabhuti, and G. Wang, “Drone: Dual-domain residual-based optimization network for sparse-view ct reconstruction,”IEEE Transactions on Medical Imaging, vol. 40, no. 11, pp. 3002–3014, 2021

2021
[21]

Cair: Combining integrated attention with iterative optimization learning for sparse-view ct reconstruction,

W. Cheng, J. He, Y . Liu, H. Zhang, X. Wang, Y . Liu, P. Zhang, H. Chen, and Z. Gui, “Cair: Combining integrated attention with iterative optimization learning for sparse-view ct reconstruction,”Computers in Biology and Medicine, vol. 163, p. 107161, 2023

2023
[22]

Nerf: Representing scenes as neural radiance fields for view synthesis,

B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

2021
[23]

Piner: Prior-informed implicit neural representation learning for test-time adaptation in sparse-view ct re- construction,

B. Song, L. Shen, and L. Xing, “Piner: Prior-informed implicit neural representation learning for test-time adaptation in sparse-view ct re- construction,” inProceedings of the IEEE/CVF winter conference on applications of computer vision, 2023, pp. 1928–1938

2023
[24]

Implicit neural representation in medical imaging: A comparative survey,

A. Molaei, A. Aminimehr, A. Tavakoli, A. Kazerouni, B. Azad, R. Azad, and D. Merhof, “Implicit neural representation in medical imaging: A comparative survey,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 2381–2391

2023
[25]

Intratomo: self-supervised learning-based tomography via sinogram synthesis and prediction,

G. Zang, R. Idoughi, R. Li, P. Wonka, and W. Heidrich, “Intratomo: self-supervised learning-based tomography via sinogram synthesis and prediction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 1960–1970

2021
[26]

Dynamic ct reconstruction from limited views with implicit neural representations and parametric motion fields,

A. W. Reed, H. Kim, R. Anirudh, K. A. Mohan, K. Champley, J. Kang, and S. Jayasuriya, “Dynamic ct reconstruction from limited views with implicit neural representations and parametric motion fields,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2...

2021
[27]

Naf: neural attenuation fields for sparse- view cbct reconstruction,

R. Zha, Y . Zhang, and H. Li, “Naf: neural attenuation fields for sparse- view cbct reconstruction,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2022, pp. 442–452

2022
[28]

Diffusion models in medical imaging: JOURNAL OF IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING 13 A comprehensive survey,

A. Kazerouni, E. K. Aghdam, M. Heidari, R. Azad, M. Fayyaz, I. Hacihaliloglu, and D. Merhof, “Diffusion models in medical imaging: JOURNAL OF IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING 13 A comprehensive survey,”Medical image analysis, vol. 88, p. 102846, 2023

2023
[29]

Solving 3d inverse problems using pre-trained 2d diffusion models,

H. Chung, D. Ryu, M. T. McCann, M. L. Klasky, and J. C. Ye, “Solving 3d inverse problems using pre-trained 2d diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 542–22 551

2023
[30]

Wavelet-improved score-based generative model for medical imaging,

W. Wu, Y . Wang, Q. Liu, G. Wang, and J. Zhang, “Wavelet-improved score-based generative model for medical imaging,”IEEE transactions on medical imaging, vol. 43, no. 3, pp. 966–979, 2023

2023
[31]

Pridediff: Physics-regularized generalized diffusion model for ct reconstruction,

Z. Lu, Q. Gao, T. Wang, Z. Yang, Z. Wang, H. Yu, H. Chen, J. Zhou, H. Shan, and Y . Zhang, “Pridediff: Physics-regularized generalized diffusion model for ct reconstruction,”IEEE Transactions on Radiation and Plasma Medical Sciences, 2024

2024
[32]

Stage-by-stage wavelet op- timization refinement diffusion model for sparse-view ct reconstruction,

K. Xu, S. Lu, B. Huang, W. Wu, and Q. Liu, “Stage-by-stage wavelet op- timization refinement diffusion model for sparse-view ct reconstruction,” IEEE Transactions on Medical Imaging, vol. 43, no. 10, pp. 3412–3424, 2024

2024
[33]

One-sample diffusion modeling in projection domain for low-dose ct imaging,

B. Huang, S. Lu, L. Zhang, B. Lin, W. Wu, and Q. Liu, “One-sample diffusion modeling in projection domain for low-dose ct imaging,”IEEE Transactions on Radiation and Plasma Medical Sciences, 2024

2024
[34]

Implicit neural prior-guided diffusion for spectral ct reconstruction,

Y . Wang, N. Liang, S. Wang, J. Guo, X. Zhang, Z. Zheng, A. Cai, L. Li, and B. Yan, “Implicit neural prior-guided diffusion for spectral ct reconstruction,”Medical Physics, vol. 52, no. 7, p. e17946, 2025

2025
[35]

Snafusion: Distilling 2d axial plane diffusion priors for sparse-view 3d cone-beam ct imaging,

X. Li, T. Cai, K. Shang, M. D. Butala, and G. Wang, “Snafusion: Distilling 2d axial plane diffusion priors for sparse-view 3d cone-beam ct imaging,” inMICCAI Workshop on Deep Generative Models. Springer, 2024, pp. 64–74

2024
[36]

Stationary computed tomography for space and other resource-constrained environments,

A. Cramer, J. Hecla, D. Wu, X. Lai, T. Boers, K. Yang, T. Moulton, S. Kenyon, Z. Arzoumanian, W. Krullet al., “Stationary computed tomography for space and other resource-constrained environments,” Scientific reports, vol. 8, no. 1, p. 14195, 2018

2018
[37]

Algebraic reconstruction techniques (art) for three-dimensional electron microscopy and x-ray photography,

R. Gordon, R. Bender, and G. T. Herman, “Algebraic reconstruction techniques (art) for three-dimensional electron microscopy and x-ray photography,”Journal of theoretical Biology, vol. 29, no. 3, pp. 471– 481, 1970

1970
[38]

Residual de- noising diffusion models,

J. Liu, Q. Wang, H. Fan, Y . Wang, Y . Tang, and L. Qu, “Residual de- noising diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 2773–2783

2024
[39]

Denoising diffusion probabilistic models,

J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840– 6851, 2020

2020
[40]

Score-based generative modeling through stochastic differ- ential equations,

Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differ- ential equations,”arXiv preprint arXiv:2011.13456, 2020

2011 arXiv
[41]

Generative modeling in sinogram domain for sparse-view ct reconstruction,

B. Guan, C. Yang, L. Zhang, S. Niu, M. Zhang, Y . Wang, W. Wu, and Q. Liu, “Generative modeling in sinogram domain for sparse-view ct reconstruction,”IEEE Transactions on Radiation and Plasma Medical Sciences, vol. 8, no. 2, pp. 195–207, 2023

2023
[42]

Ct-sdm: A sampling dif- fusion model for sparse-view ct reconstruction across various sampling rates,

L. Yang, J. Huang, G. Yang, and D. Zhang, “Ct-sdm: A sampling dif- fusion model for sparse-view ct reconstruction across various sampling rates,”IEEE Transactions on Medical Imaging, 2025

2025
[43]

Unsupervised ct metal artifact reduction by plugging diffusion priors in dual domains,

X. Liu, Y . Xie, S. Diao, S. Tan, and X. Liang, “Unsupervised ct metal artifact reduction by plugging diffusion priors in dual domains,”IEEE Transactions on Medical Imaging, vol. 43, no. 10, pp. 3533–3545, 2024

2024
[44]

Self-fusion sim- plex noise-based diffusion model for self-supervised low-dose digital radiography denoising,

Y . Wang, Z. Li, W. Wu, J. Zhang, and W. Wu, “Self-fusion sim- plex noise-based diffusion model for self-supervised low-dose digital radiography denoising,”IEEE Transactions on Instrumentation and Measurement, 2024

2024
[45]

Implicit neural representations with periodic activation functions,

V . Sitzmann, J. Martel, A. Bergman, D. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation functions,” Advances in neural information processing systems, vol. 33, pp. 7462– 7473, 2020

2020
[46]

Neu- ral reflectance fields for appearance acquisition,

S. Bi, Z. Xu, P. Srinivasan, B. Mildenhall, K. Sunkavalli, M. Ha ˇsan, Y . Hold-Geoffroy, D. Kriegman, and R. Ramamoorthi, “Neu- ral reflectance fields for appearance acquisition,”arXiv preprint arXiv:2008.03824, 2020

2008 arXiv
[47]

Neat: Neural adaptive tomography,

D. R ¨uckert, Y . Wang, R. Li, R. Idoughi, and W. Heidrich, “Neat: Neural adaptive tomography,”ACM Transactions on Graphics (TOG), vol. 41, no. 4, pp. 1–13, 2022

2022
[48]

Ineat: an artifact- suppressed and resolution-enhanced computed tomography through iter- ative neural adaptive tomography,

B. Xiong, C. Su, Z. Lin, Y . Zhou, and Z. Yu, “Ineat: an artifact- suppressed and resolution-enhanced computed tomography through iter- ative neural adaptive tomography,”Optics Express, vol. 32, no. 18, pp. 32 336–32 353, 2024

2024
[49]

Snaf: Sparse-view cbct reconstruction with neural attenuation fields,

Y . Fang, L. Mei, C. Li, Y . Liu, W. Wang, Z. Cui, and D. Shen, “Snaf: Sparse-view cbct reconstruction with neural attenuation fields,”arXiv preprint arXiv:2211.17048, 2022

2022 arXiv
[50]

Geometry- aware attenuation learning for sparse-view cbct reconstruction,

Z. Liu, Y . Fang, C. Li, H. Wu, Y . Liu, D. Shen, and Z. Cui, “Geometry- aware attenuation learning for sparse-view cbct reconstruction,”IEEE Transactions on Medical Imaging, 2024

2024
[51]

Structure-aware sparse-view x-ray 3d reconstruction,

Y . Cai, J. Wang, A. Yuille, Z. Zhou, and A. Wang, “Structure-aware sparse-view x-ray 3d reconstruction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 11 174–11 183

2024
[52]

On the spectral bias of neural networks,

N. Rahaman, A. Baratin, D. Arpit, F. Draxler, M. Lin, F. Hamprecht, Y . Bengio, and A. Courville, “On the spectral bias of neural networks,” inInternational conference on machine learning. PMLR, 2019, pp. 5301–5310

2019
[53]

Fourier features let networks learn high frequency functions in low dimensional domains,

M. Tancik, P. Srinivasan, B. Mildenhall, S. Fridovich-Keil, N. Raghavan, U. Singhal, R. Ramamoorthi, J. Barron, and R. Ng, “Fourier features let networks learn high frequency functions in low dimensional domains,” Advances in neural information processing systems, vol. 33, pp....

2020
[54]

American Association of Physicists in Medicine (AAPM)

(2017) Low dose ct grand challenge. American Association of Physicists in Medicine (AAPM). Accessed: Apr. 6, 2017. [Online]. Available: http://www.aapm.org/GrandChallenge/LowDoseCT/

2017
[55]

Scientific visualization datasets,

P. Klacansky, “Scientific visualization datasets,” 2022

2022

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

Clinical applications of cone-beam computed tomography in dental practice,

W. C. Scarfe, A. G. Farman, P. Sukovicet al., “Clinical applications of cone-beam computed tomography in dental practice,”Journal-Canadian Dental Association, vol. 72, no. 1, p. 75, 2006

2006

[2] [2]

Signal sensing and reconstruction paradigms for a novel multi-source static computed tomography system,

A. Kowtal, A. Cramer, D. Wu, K. Yang, W. Krull, I. Gkioulekas, and R. Gupta, “Signal sensing and reconstruction paradigms for a novel multi-source static computed tomography system,” inICASSP 2020- 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 9274–9278

2020

[3] [3]

Swinging multi-source industrial ct systems for aperiodic dynamic imaging,

W. Wu, H. Yu, C. Gong, and F. Liu, “Swinging multi-source industrial ct systems for aperiodic dynamic imaging,”Optics express, vol. 25, no. 20, pp. 24 215–24 235, 2017

2017

[4] [4]

Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam ct,

J. Bian, J. H. Siewerdsen, X. Han, E. Y . Sidky, J. L. Prince, C. A. Pelizzari, and X. Pan, “Evaluation of sparse-view reconstruction from flat-panel-detector cone-beam ct,”Physics in Medicine & Biology, vol. 55, no. 22, p. 6575, 2010

2010

[5] [5]

Accurate image reconstruction from few-views and limited-angle data in divergent-beam ct,

E. Y . Sidky, C.-M. Kao, and X. Pan, “Accurate image reconstruction from few-views and limited-angle data in divergent-beam ct,”Journal of X-ray Science and Technology, vol. 14, no. 2, pp. 119–139, 2006

2006

[6] [6]

Practical cone-beam algorithm,

L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beam algorithm,”Journal of the Optical Society of America A, vol. 1, no. 6, pp. 612–619, 1984

1984

[7] [7]

Simultaneous algebraic reconstruc- tion technique (sart): a superior implementation of the art algorithm,

A. H. Andersen and A. C. Kak, “Simultaneous algebraic reconstruc- tion technique (sart): a superior implementation of the art algorithm,” Ultrasonic imaging, vol. 6, no. 1, pp. 81–94, 1984

1984

[8] [8]

Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,

E. Y . Sidky and X. Pan, “Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,” Physics in Medicine & Biology, vol. 53, no. 17, p. 4777, 2008

2008