Pith. sign in

REVIEW 4 major objections 6 minor 67 references

Noise-Inspired Diffusion Model for Generalizable Low-Dose CT Reconstruction

T0 review · 4 major / 6 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 A diffusion model trained only on normal-dose CT data can reconstruct low-dose scans at dose levels never seen during training.

desk verdict A well-engineered dual-domain diffusion method whose core dose-generalization claim rests on one simulated noise model; real projection data is the missing test. read the letter →

arxiv 2506.22012 v1 pith:KAVAYMU5 submitted 2025-06-27 eess.IV cs.CV

Qi Gao , Zhihao Chen , Dong Zeng , Junping Zhang , Jianhua Ma , Hongming Shan This is my paper
classification eess.IVcs.CV
keywords low-doseCTreconstructiondiffusionmodelshiftedPoissonnoisedosegeneralizationprojection-domaindenoisingimage-domainrefinementtimestepmatchingdual-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 a single CT reconstruction pipeline, trained only on normal-dose projection and image data, can clean up low-dose scans across a range of doses, including doses never encountered in training. The model, NEED, cascades a shifted Poisson diffusion model that denoises raw projections according to the measured physics of low-dose CT with a doubly guided diffusion model that refines the image using both the noisy low-dose scan and the initial reconstruction. A time step matching rule chooses where each test scan enters the diffusion process, effectively adapting the denoising strength to the scan's dose. If the claim is right, clinical and research CT denoising would no longer need paired normal/low-dose data or retraining for every new protocol.

What carries the argument

The load-bearing object is the shifted Poisson degradation operator $D(y_0,t)=\mathcal{P}(\lambda_t y_0+\sigma_e^2)/\lambda_t$, which injects Poisson noise scaled by $\lambda_t$ plus electronic-noise variance $\sigma_e^2$ into normal-dose projections, so the forward diffusion mimics real pre-log low-dose measurements. Training a restoration network $R_\phi$ on normal-dose projections corrupted by this operator gives a denoiser matched to CT physics, and the inference recursion $y_{t-1}=(\lambda_t y_t+\mathcal{P}((\lambda_{t-1}-\lambda_t)\tilde{y}_0))/\lambda_{t-1}$ removes the shifted Poisson noise in steps. For images, the mechanism is double guidance: the noisy sample is modulated by the low-dose image, $\tilde{x}_t=s_1 x_t+(1-s_1)\sqrt{\bar\alpha_t}\,x_{\mathrm{ld}}$, and the denoised estimate is blended with the initial reconstruction, $x_0=s_2\tilde{x}_0+(1-s_2)\hat{x}_0$. Time step matching, $t^*_{\mathrm{prj}}=\arg\min_t |\lambda_t-I_{\mathrm{ld}}|$ for projections and $t^*_{\mathrm{img}}=\arg\min_t |\sqrt{1-\bar\alpha_t}-\hat{\sigma}_{\mathrm{ld}}|$ for images, converts the known dose level into a dose-adaptive sampling schedule.

What would settle it

Take a phantom or patient with a normal-dose reference scan, acquire true low-dose pre-log projections at a dose level not used in training, estimate $I_0$ from the scan itself, and run NEED. The central claim fails if the reconstruction develops visible artifacts, CT-value bias, or a significant drop in PSNR/SSIM relative to the simulated-dose results—or if the measured variance of the real projections is not consistent with the linear-in-signal shifted Poisson form assumed by the degradation operator.

Watch

Extended reading notes

Core claim

NEED's central claim is that the main barrier to generalizable low-dose CT reconstruction is a mismatch between the noise model assumed by diffusion models and the noise actually present in CT data. The paper addresses this mismatch in two domains: in projection space it replaces Gaussian degradation with a shifted Poisson degradation operator aligned to pre-log measurement statistics, and in image space it uses a normal-dose diffusion prior with double guidance from the noisy low-dose image and the initial reconstruction, which locates the prior more precisely. The time step matching strategy then selects the sampling start point from the incident photon count for projections and from the residual standard deviation for the image, and this is what lets one model cover multiple dose levels. On two public low-dose CT datasets, the paper reports that NEED outperforms self-supervised methods, unsupervised diffusion-based baselines, and supervised diffusion and CNN baselines in PSNR, SSIM, and RMSE, and that its reconstructions improve a downstream segmentation model's Dice scores.

Load-bearing premise

The evaluation assumes that low-dose projections made by forward-projecting normal-dose images and adding Poisson plus Gaussian noise with known parameters faithfully reproduce real scanner measurements, including scatter, beam hardening, and detector effects.

Editorial extensions

If this is right

  • A single NEED model trained once on normal-dose data can be applied at multiple low-dose levels without retraining or fine-tuning, including levels not seen in training.
  • Paired normal/low-dose CT collections are no longer a prerequisite for high-quality denoising; only normal-dose projections and images are needed.
  • Handling the projection-domain noise with the shifted Poisson model reduces CT-value drift and preserves lesion structure, which carries through to better downstream segmentation.
  • The time step matching strategy cuts sampling steps and inference time compared with full diffusion sampling, while matching or improving reconstruction quality.
  • Because the image-domain prior is trained only on normal-dose images, the same pipeline can be transferred to a new dataset with a different noise distribution directly, as demonstrated on the second public dataset.

Reading between the lines

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

  • A natural extension is to treat $I_0$ as an unknown and estimate it from the test projection itself, which would remove the need for scanner calibration data and make the method applicable when exposure metadata is unavailable.
  • The same cascade design—physics-matched diffusion denoising in a raw-data domain followed by a clean-data diffusion prior refined with double guidance—should transfer to other Poisson-limited imaging problems such as PET, SPECT, or photon-counting CT, provided pre-log data are available.
  • If real scanner noise departs from the shifted Poisson model, the time step matching rule could be recalibrated empirically per scanner by fitting $\lambda_t$ to measured noise-variance curves, preserving the dose-adaptive behavior without retraining the full model.
  • The paper's evaluation on simulated projections leaves open how much of the improvement survives real scatter and beam hardening; a direct test on real pre-log data would be the decisive next step.
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 / 6 minor

Summary. The paper proposes NEED, a cascade of two diffusion models for low-dose CT reconstruction. SPDiff denoises pre-log projection data with a shifted-Poisson cold-diffusion operator D(y0,t)=P(λ_t y0 + σ_e^2)/λ_t, and DGDiff refines the FBP reconstruction using a Gaussian diffusion prior with double guidance from the LDCT image and from the initial reconstruction. A time-step matching strategy selects the starting sampling step from the incident photon count I_ld in the projection domain and from σ_ld in the image domain. The method is trained only on normal-dose projections and images, then tested on simulated low-dose data at 50%, 25%, 12.5%, and 10% dose on Mayo 2016 and on Mayo 2020. The paper reports consistent improvements over self-supervised, iterative, and unsupervised diffusion baselines in PSNR/SSIM/RMSE, additional perceptual metrics, downstream MedSAM segmentation, and inference time.

Significance. If the central claims hold, the contribution is practically significant: NEED would offer dose-agnostic low-dose CT reconstruction trained only on normal-dose data, with an explicit physics-inspired projection-domain degradation and faster sampling than single-domain diffusion baselines. The paper has clear strengths: source code is released; the sampling update in Eq. (10) is internally consistent with the shifted-Poisson degradation; the ablation studies isolate the contributions of SPDiff, DGDiff, and time-step matching; and the diffusion baselines share the same pretrained Gaussian diffusion model, which makes the comparison fair at the architectural level. The downstream MedSAM evaluation adds a useful clinically oriented check.

major comments (4)
  1. [Sec. 4.1, Tables 2, 5, 7] The headline claim of generalization to unseen dose levels is validated only on projections synthesized with the same Poisson+Gaussian model from which SPDiff's degradation operator is constructed. For Mayo 2016, the pre-log projection data are obtained by forward-projecting reconstructed NDCT images and adding Poisson+Gaussian noise with I0=2.5e5 and σ_e^2=10; the SPDiff degradation in Eq. (9) uses the same σ_e^2 and a λ_t schedule spanning the range of I_ld. The 'unseen dose' experiments therefore test interpolation within the λ_t schedule, not robustness to deviations from the assumed noise model. Since Section 5.5 acknowledges that real projections contain scatter, detector nonlinearities, and other effects absent from this model, the clinical transferability of the dose-generalization claim is not established. I ask for validation on real measured pre-log data or, at minimum, a hold-out simulation with different σ_e^2, beam-hardening, or detector response.
  2. [Sec. 4.1, Mayo 2020 experiments] The Mayo 2020 'unseen dataset' test is also not based on measured projection data. The text states that the authors use the officially provided LDCT images and 'leverage the projection geometry outlined in Table 1 to obtain the corresponding pre-log LDCT projection.' Since those LDCT images are themselves simulated (from post-log signal-dependent Gaussian noise with bowtie filtration and automatic exposure control), the resulting pre-log projections are another simulation layer, not independent external validation. The claim that NEED 'can be effectively extended to various unseen dose levels' and the comparison in Table 5 would be substantially strengthened if the official Mayo 2020 raw projection data were used instead of re-projecting reconstructed images.
  3. [Sec. 4.3 vs Sec. 4.6, Table 2] There is a contradiction in the description of the Noise2Noise baseline. Section 4.3 says 'we simulate an additional set of 25% and 10% dose CT images from the Mayo 2016 dataset as training labels,' while Section 4.6 says 'Noise2Noise, Noise2Sim, and SSDDNet are trained using only 25% dose data.' If Noise2Noise received 10% paired training data, its 10% row in Table 2 is a seen-dose result, not a zero-shot generalization point, and the comparison at 10% dose is unfair. The authors should clarify exactly which dose levels were used for each baseline and, if 10% labels were used, retrain Noise2Noise on 25% only for the generalization comparison.
  4. [Sec. 3.2, Eq. (13)] The doubly guided sampling formula is presented as an exact posterior factorization, but the two guidance terms are conditioned on x_t rather than on x_{t-1}, and p(x_g|x_t) is then replaced by an L1-distance modulation with hand-set weights s1 and s2. This is a heuristic guided-sampling procedure, not a derivation. The ablations in Table 8 show empirical value, but the paper should either derive the approximation more carefully (for example, as a variational or annealed update) or explicitly state that Eq. (13) is an ad-hoc guided sampler. Without this clarification, readers cannot tell how much of the reported gain comes from the noise-aligned SPDiff versus from tuning the guidance weights.
minor comments (6)
  1. [Sec. 5.2] Section 5.2 contains a word-for-word duplicated paragraph: the passage beginning 'Our NEED leverages the number of X-ray incident photons I0...' appears twice in succession; one copy should be deleted.
  2. [Sec. 4.8, Table 7] The sentence 'SPDiff (τ) and SPDi ff (T) yield identical results at this dose level' is confusing because T is not defined for SPDiff; the total diffusion steps are τ=10, so either use SPDiff (τ) consistently or define T separately.
  3. [Fig. 12] The 'perceptual loss curve' is referred to but never defined; please state the exact loss used and how the intermediate reconstructed images are compared.
  4. [Eq. (9)] The notation P(λ_t y0 + σ_e^2)/λ_t conflates a shifted-Poisson random variable with a constant addition; defining the transition as m_t ∼ P(λ_t y0) + σ_e^2, y_t = m_t/λ_t would avoid ambiguity.
  5. [Algorithm 1, line 8] Line 8 uses y_1 as the input to R_phi, but if t*_prj=1 the loop at lines 4-7 is skipped and y_1=y_ld; an explicit note that this is the intended behavior would improve reproducibility.
  6. [Sec. 4.1] The values I0=2.5e5 and σ_e^2=10 are given without explicit units or normalization conventions for y_0; stating these conventions is important for reproducing the simulation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: NEED's training uses only normal-dose data, the time-step matching rule is driven by the physical photon-count parameter I0 rather than a fitted constant, and the central reconstruction chain is self-contained. The main limitation is external validity of simulated test data, not logical circularity.

full rationale

The claimed derivation chain is not circular. SPDiff is trained on normal-dose pre-log projections with a forward degradation operator (Eq. 9) that adds shifted Poisson noise, and the test low-dose projections are synthesized from the same Poisson+Gaussian family; this is a modeling and evaluation choice, not a fitted-input-called-prediction. The time-step matching strategy (Eqs. 16 and 17) selects a starting step from the known physical input I0 (and from the data-dependent sigma_ld), with no learned test-time parameter fitting to the target dose. DGDiff uses a Gaussian diffusion prior trained on normal-dose images, and its double guidance blends the LDCT image and the initial reconstruction explicitly (Eqs. 14 and 15), again with no hidden reduction of the output to a training target. The self-citations present (CoCoDiff for channel concatenation, CoreDiff as a compared baseline) are not load-bearing for the central claim; no uniqueness theorem or ansatz is imported from the authors' prior work to force the architecture. The strongest concern is that all dose-generalization tests are conducted on simulated projections that match the model's assumed degradation family, and the paper itself acknowledges in Section 5.5 that real measured projections include scatter, detector nonlinearities, and other effects not modeled. That is a limitation on clinical transferability, not a circularity in the derivation. Under the stated rules, the appropriate finding is a low circularity score.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The method contributes a new combination of noise-aligned diffusion processes and a test-time adaptation strategy, but it rests on several empirically chosen hyperparameters and on the central assumption that simulated Poisson+Gaussian projections stand in for real scanner data. No new physical entities are postulated.

free parameters (5)
  • lambda_1 = 3e5
    Upper endpoint of the shifted Poisson diffusion schedule, empirically set to cover the incident photon count range of the tested doses (Section 4.2).
  • lambda_tau = 2.5e4
    Lower endpoint of the schedule, corresponding to the 10% dose level where Ild = 2.5e4, empirically set (Section 4.2).
  • tau = 10
    Total number of shifted Poisson diffusion steps, chosen by hand rather than derived from a criterion (Section 4.2).
  • s1 = 0.7 * exp(5000 sigma_ld^2) / (10 + exp(5000 sigma_ld^2))
    Weight balancing the LDCT image guidance; constants 0.7, 5000, and 10 are empirically set, with the derivation deferred to a missing supplementary (Section 4.2).
  • s2 = 0.8
    Weight balancing the initial reconstruction guidance in the image refinement stage, fixed empirically (Section 4.2).
assumptions (6)
  • domain assumption Pre-log CT measurements follow a shifted Poisson model, i.e., m_ld ~ P(m(x)) + N(0, sigma_e^2), as stated in Eq. (1).
    This is the standard CT noise model used to design SPDiff's degradation operator and to simulate test data; real scanners may deviate with scatter, detector nonlinearity, and other effects.
  • ad hoc to paper Forward projection of reconstructed normal-dose CT images yields representative pre-log normal-dose projections.
    The Mayo 2016 dataset provides post-log data only, so the authors generate pre-log projections via fan-beam forward projection with ASTRA (Section 4.1).
  • ad hoc to paper The posterior factorization log p(x_{t-1} | x_t, x_ld, x_hat_0) = log p(x_{t-1} | x_t) + log p(x_ld | x_t) + log p(x_hat_0 | x_t) + C is valid, with guidance approximated by L1-distance modulation.
    Eq. (13) is a heuristic approximation; the paper provides no derivation for the CT image prior, and the associated weights s1 and s2 are tuned empirically.
  • domain assumption A Gaussian diffusion model pretrained on 256x256 normal-dose CT images, followed by a 256x256 to 512x512 super-resolution stage, is an adequate generative prior for low-dose image refinement.
    DGDiff relies on this cascaded prior; its adequacy is validated only on simulated test data and not on real clinical low-dose images.
  • standard math DDIM with 10 sampling steps and DPM-Solver give accurate approximations for the super-resolution diffusion sampling.
    These are standard accelerated samplers, but no error bound is given for this specific CT image distribution.
  • ad hoc to paper Time-step matching by nearest lambda_t to Ild correctly maps each unseen dose level to the appropriate denoising level.
    Eq. (16) assumes the noise level in normalized pre-log LDCT projections is primarily controlled by Ild and that the hand-set lambda schedule covers the full range of test doses.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Noise-Inspired Diffusion Model for Generalizable Low-Dose CT Reconstruction." pith.science (2026). https://pith.science/paper/KAVAYMU5

@misc{pith2026250622012,
  author       = {Pith},
  title        = {Pith review of: Noise-Inspired Diffusion Model for Generalizable Low-Dose CT Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KAVAYMU5}},
  note         = {Machine review of arXiv:2506.22012}
}
read the original abstract

The generalization of deep learning-based low-dose computed tomography (CT) reconstruction models to doses unseen in the training data is important and remains challenging. Previous efforts heavily rely on paired data to improve the generalization performance and robustness through collecting either diverse CT data for re-training or a few test data for fine-tuning. Recently, diffusion models have shown promising and generalizable performance in low-dose CT (LDCT) reconstruction, however, they may produce unrealistic structures due to the CT image noise deviating from Gaussian distribution and imprecise prior information from the guidance of noisy LDCT images. In this paper, we propose a noise-inspired diffusion model for generalizable LDCT reconstruction, termed NEED, which tailors diffusion models for noise characteristics of each domain. First, we propose a novel shifted Poisson diffusion model to denoise projection data, which aligns the diffusion process with the noise model in pre-log LDCT projections. Second, we devise a doubly guided diffusion model to refine reconstructed images, which leverages LDCT images and initial reconstructions to more accurately locate prior information and enhance reconstruction fidelity. By cascading these two diffusion models for dual-domain reconstruction, our NEED requires only normal-dose data for training and can be effectively extended to various unseen dose levels during testing via a time step matching strategy. Extensive qualitative, quantitative, and segmentation-based evaluations on two datasets demonstrate that our NEED consistently outperforms state-of-the-art methods in reconstruction and generalization performance. Source code is made available at https://github.com/qgao21/NEED.

Figures

Figures reproduced from arXiv: 2506.22012 by the authors.

Figure 1
Figure 1. Overview of CT reconstruction and different learning-based reconstruction paradigms. (a) CT reconstruction pipeline. (b) Illustra￾tion of the differences in training strategies among supervised methods, self-supervised methods, and our NEED. Bindman et al., 2009; Sodickson et al., 2009). In clinical prac￾tice, reducing radiation dose often involves lowering the tube current. However, this reduction inevitably leads … view at source ↗
Figure 2
Figure 2. Overview of our NEED for generalizable low-dose CT reconstruction. Our NEED cascades two components: 1) a shifted Poisson diffusion model (SPDiff) for pre-log LDCT projection data denoising and 2) a doubly guided diffusion model (DGDiff) for reconstructed image refinement. Then we utilize a time step matching strategy to control the denoising and refinement while reducing sampling steps. Although the Gaussian diffus… view at source ↗
Figure 3
Figure 3. Qualitative results of a 25% dose CT image from the Mayo 2016 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) PWLS, (d) Noise2Noise, (e) Noise2Sim, (f) SSDDNet, (g) DR2, (h) GDP, (i) Dn-Dp, (j) SPDiff (ours), and (k) NEED (ours). The display window is [-160, 240] HU. The red ROI is zoomed in for visual comparison and the orange arrow points to one lesion. et al., 2022). Both samplers produced similar results, w… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Qualitative results of a 10% dose CT image from the Mayo 2016 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) PWLS, (d) Noise2Noise, (e) Noise2Sim, (f) SSDDNet, (g) DR2, (h) GDP, (i) Dn-Dp, (j) SPDiff (ours), and (k) NEED (ours). The display window is [-160, 240] …
Figure 5
Figure 5. Figure 5: Visualization of lesion ROIs segmentation in Figs. 3 and 4. Seg￾mentation results are denoted in yellow for NDCT images and in green for images reconstructed by different methods. (a) FBP, (b) PWLS, (c) Noise2Noise, (d) Noise2Sim, (e) SSDDNet, (f) DR2, (g) GDP, (h) Dn￾…
Figure 6
Figure 6. Figure 6: Qualitative results of a selected ROI across seen and unseen dose levels from the Mayo 2016 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) Noise2Noise, (d) Noise2Sim, (e) SSDDNet, (f) DR2, (g) GDP, (h) Dn-Dp, and (i) NEED (ours). The orange arrow and the purple a…
Figure 7
Figure 7. Figure 7: Mean CT values and standard deviation (std) of each ROI in [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Qualitative results on the unseen Mayo 2020 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) PWLS, (d) Noise2Sim, (e) SSDDNet, (f) DR2, (g) GDP, (h) Dn-Dp, and (i) NEED (ours). The display window for chest images is [-1350, 150] HU and for abdomen images is [-160, …
Figure 9
Figure 9. Figure 9: Visualization of ROIs segmentation in [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Qualitative comparison of supervised methods, Noise2Sim, and NEED on seen and unseen doses. (a) NDCT image (Ground truth), (b) FBP, (c) RED-CNN, (d) CoreDiff, (e) Noise2Sim, and (f) NEED (ours). The display window for chest images is [-1350, 150] HU and for abdomen im…
Figure 11
Figure 11. Figure 11: Quantitative comparison of RED-CNN, CoreDiff, and NEED (ours) on seen and unseen datasets. 2016 dataset, all methods effectively reduce noise, RED-CNN exhibits over-smoothing, whereas CoreDiff and NEED retain more texture details. However, supervised methods degrade i…
Figure 12
Figure 12. Figure 12: Visualization of diffusion and sampling processes along with time step matching strategy for the SPDiff. The dose of the example LDCT data is 12.5%, with the perceptual loss between it and intermediate reconstructed images of the diffusion process shown in the lower l…
Figure 13
Figure 13. Figure 13: Qualitative results of our SPDiff and NEED starting sampling from different time steps on the Mayo 2016 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) SPDiff (t ∗ prj) (ours), (d) SPDiff (τ), (e) NEED (t ∗ prj +t ∗ img) (ours), and (f) NEED (t ∗ prj +T). The dis…
Figure 14
Figure 14. Figure 14: Qualitative analysis of proposed modules on 25% dose test data from the Mayo 2016 dataset. (a) NDCT image (Ground truth), (b) FBP, (c) SPDiff, (d) SGDiff, (e) SPDiff+SGDiff, and (f) NEED. The display window is [-160, 240] HU. paired ND/LDCT data for training, ensuring…
Figure 15
Figure 15. Figure 15: Qualitative results of DGDiff on the Mayo 2016 dataset using initial reconstructions xˆ0 provided by different algorithms. (a) NDCT image (Ground truth), (b) FBP, (c) NLM, (d) BM3D, (e) PWLS, (f) Noise2Sim, and (g) SPDiff. (c1) to (g1) show initial reconstructions fro…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 64 canonical work pages

  1. [1]

    , author Borgnia, E

    author Bansal, A. , author Borgnia, E. , author Chu, H.M. , author Li, J. , author Kazemi, H. , author Huang, F. , author Goldblum, M. , author Geiping, J. , author Goldstein, T. , year 2023 . title Cold diffusion: Inverting arbitrary image transforms without noise , in: booktitle Proc. Adv. Neural Inf. Process. Syst

    work page 2023

  2. [2]

    , author Coll, B

    author Buades, A. , author Coll, B. , author Morel, J.M. , year 2005 . title A non-local algorithm for image denoising , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , pp. pages 60--65

    work page 2005

  3. [3]

    , year 2008

    author Buzug, T. , year 2008 . title Computed Tomography: F rom Photon Statistics to Modern Cone-Beam CT . publisher Springer Berlin, Heidelberg

    work page 2008

  4. [4]

    , author Leng, S

    author Chen, B. , author Leng, S. , author Yu, L. , author Holmes III, D. , author Fletcher, J. , author McCollough, C. , year 2016 . title An open library of CT patient projection data , in: booktitle Proc. SPIE , pp. pages 330--335

    work page 2016

  5. [5]

    , author Zhang, Y

    author Chen, H. , author Zhang, Y. , author Kalra, M.K. , author Lin, F. , author Chen, Y. , author Liao, P. , author Zhou, J. , author Wang, G. , year 2017 . title Low-dose CT with a residual encoder-decoder convolutional neural network . journal IEEE Trans. Med. Imag. volume 36 , pages 2524--2535

    work page 2017

  6. [6]

    , author Foi, A

    author Dabov, K. , author Foi, A. , author Katkovnik, V. , author Egiazarian, K. , year 2007 . title Image denoising by sparse 3-D transform-domain collaborative filtering . journal IEEE Trans. Image Process. volume 16 , pages 2080--2095

    work page 2007

  7. [7]

    , author Kite, T.D

    author Damera-Venkata, N. , author Kite, T.D. , author Geisler, W.S. , author Evans, B.L. , author Bovik, A.C. , year 2000 . title Image quality assessment based on a degradation model . journal IEEE Trans. Image Process. volume 9 , pages 636--650

    work page 2000

  8. [8]

    , author Nichol, A

    author Dhariwal, P. , author Nichol, A. , year 2021 . title Diffusion models beat GANs on image synthesis , in: booktitle Proc. Adv. Neural Inf. Process. Syst. , pp. pages 8780--8794

    work page 2021

Show all 67 references
  1. [9]

    , author Pelc, N.J

    author Divel, S.E. , author Pelc, N.J. , year 2019 . title Accurate image domain noise insertion in CT images . journal IEEE Trans. Med. Imag. volume 39 , pages 1906--1916

    2019

  2. [10]

    , author Lyu, Z

    author Fei, B. , author Lyu, Z. , author Pan, L. , author Zhang, J. , author Yang, W. , author Luo, T. , author Zhang, B. , author Dai, B. , year 2023 . title Generative diffusion prior for unified image restoration and enhancement , in: booktitle Proc. IEEE Conf. Comput. Vis....

    2023

  3. [11]

    , author Lee, T.C

    author Fu, L. , author Lee, T.C. , author Kim, S.M. , author Alessio, A.M. , author Kinahan, P.E. , author Chang, Z. , author Sauer, K. , author Kalra, M.K. , author De Man, B. , year 2016 . title Comparison between pre-log and post-log statistical models in ultra-low-dose CT ...

    2016

  4. [12]

    , author Dong, S

    author Fu, Y. , author Dong, S. , author Niu, M. , author Xue, L. , author Guo, H. , author Huang, Y. , author Xu, Y. , author Yu, T. , author Shi, K. , author Yang, Q. , author Shi, Y. , author Zhang, H. , author Tian, M. , author Zhuo, C. , year 2023 . title AIGAN : A ttenti...

    2023

  5. [13]

    , author Li, Z

    author Gao, Q. , author Li, Z. , author Zhang, J. , author Zhang, Y. , author Shan, H. , year 2024 . title CoreDiff : C ontextual error-modulated generalized diffusion model for low-dose CT denoising and generalization . journal IEEE Trans. Med. Imag. volume 43 , pages 745--759

    2024

  6. [14]

    , author Shan, H

    author Gao, Q. , author Shan, H. , year 2022 . title CoCoDiff : a contextual conditional diffusion model for low-dose CT image denoising , in: booktitle Proc. SPIE

    2022

  7. [15]

    , author Tirer, T

    author Garber, T. , author Tirer, T. , year 2024 . title Image restoration by denoising diffusion models with iteratively preconditioned guidance , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , pp. pages 25245--25254

    2024

  8. [16]

    , author He, Y

    author Ge, R. , author He, Y. , author Xia, C. , author Sun, H. , author Zhang, Y. , author Hu, D. , author Chen, S. , author Chen, Y. , author Li, S. , author Zhang, D. , year 2022 . title DDPNet : A novel dual-domain parallel network for low-dose CT reconstruction , in: book...

    2022

  9. [17]

    , author Ren, L

    author Gong, H. , author Ren, L. , author Hsieh, S.S. , author McCollough, C.H. , author Yu, L. , year 2021 . title Deep learning enabled ultra-fast-pitch acquisition in clinical x-ray computed tomography . journal Med. Phys. volume 48 , pages 5712--5726

    2021

  10. [18]

    , author Jain, A

    author Ho, J. , author Jain, A. , author Abbeel, P. , year 2020 . title Denoising diffusion probabilistic models , in: booktitle Proc. Adv. Neural Inf. Process. Syst. , pp. pages 6840--6851

    2020

  11. [19]

    , author Saharia, C

    author Ho, J. , author Saharia, C. , author Chan, W. , author Fleet, D.J. , author Norouzi, M. , author Salimans, T. , year 2022 . title Cascaded diffusion models for high fidelity image generation . journal J. Mach. Learn. Res. volume 23 , pages 2249--2281

    2022

  12. [20]

    , author Young, S

    author Hoffman, J. , author Young, S. , author Noo, F. , author McNitt-Gray, M. , year 2016 . title FreeCT \_ wFBP : a robust, efficient, open-source implementation of weighted filtered backprojection for helical, fan-beam CT . journal Med. Phys. volume 43 , pages 1411--1420

    2016

  13. [21]

    , year 2003

    author Hsieh, J. , year 2003 . title Computed tomography: principles, design, artifacts, and recent advances . publisher SPIE

    2003

  14. [22]

    , author Zhang, J

    author Huang, Z. , author Zhang, J. , author Zhang, Y. , author Shan, H. , year 2022 . title DU - GAN : Generative adversarial networks with dual-domain U - Net based discriminators for low-dose CT denoising . journal IEEE Trans. Instrum. Meas. volume 71 , pages 4500512

    2022

  15. [23]

    , author Elad, M

    author Kawar, B. , author Elad, M. , author Ermon, S. , author Song, J. , year 2022 . title Denoising diffusion restoration models . journal Proc. Adv. Neural Inf. Process. Syst. volume 35 , pages 23593--23606

    2022

  16. [24]

    , author Aghdam, E.K

    author Kazerouni, A. , author Aghdam, E.K. , author Heidari, M. , author Azad, R. , author Fayyaz, M. , author Hacihaliloglu, I. , author Merhof, D. , year 2023 . title Diffusion models in medical imaging: A comprehensive survey . journal Med. Image Anal. volume 88 , pages 102846

    2023

  17. [25]

    , author Munkberg, J

    author Lehtinen, J. , author Munkberg, J. , author Hasselgren, J. , author Laine, S. , author Karras, T. , author Aittala, M. , author Aila, T. , year 2018 . title Noise2Noise : L earning image restoration without clean data , in: booktitle Proc. Int. Conf. Mach. Learn. , pp. ...

    2018

  18. [26]

    , author Bian, Z

    author Li, D. , author Bian, Z. , author Li, S. , author He, J. , author Zeng, D. , author Ma, J. , year 2022 . title Noise characteristics modeled unsupervised network for robust CT image reconstruction . journal IEEE Trans. Med. Imag. volume 41 , pages 3849--3861

    2022

  19. [27]

    , author Li, X

    author Li, T. , author Li, X. , author Wang, J. , author Wen, J. , author Lu, H. , author Hsieh, J. , author Liang, Z. , year 2004 . title Nonlinear sinogram smoothing for low-dose X -ray CT . journal IEEE Trans. Nucl. Sci. volume 51 , pages 2505--2513

    2004

  20. [28]

    , author Xie, Y

    author Liu, X. , author Xie, Y. , author Liu, C. , author Cheng, J. , author Diao, S. , author Tan, S. , author Liang, X. , year 2024 . title Diffusion probabilistic priors for zero-shot low-dose CT image denoising . journal Med. Phys

    2024

  21. [29]

    , author Zhou, Y

    author Lu, C. , author Zhou, Y. , author Bao, F. , author Chen, J. , author Li, C. , author Zhu, J. , year 2022 . title DPM-Solver : A fast ode solver for diffusion probabilistic model sampling in around 10 steps , in: booktitle Proc. Adv. Neural Inf. Process. Syst. , pp. page...

    2022

  22. [30]

    , author Gao, Q

    author Lu, Z. , author Gao, Q. , author Wang, T. , author Yang, Z. , author Wang, Z. , author Yu, H. , author Chen, H. , author Zhou, J. , author Shan, H. , author Zhang, Y. , year 2024 . title PrideDiff : P hysics-regularized generalized diffusion model for CT reconstruction ...

    2024

  23. [31]

    , author He, Y

    author Ma, J. , author He, Y. , author Li, F. , author Han, L. , author You, C. , author Wang, B. , year 2024 . title Segment anything in medical images . journal Nat. Commun. volume 15 , pages 654

    2024

  24. [32]

    , author Huang, J

    author Ma, J. , author Huang, J. , author Feng, Q. , author Zhang, H. , author Lu, H. , author Liang, Z. , author Chen, W. , year 2011 . title Low-dose computed tomography image restoration using previous normal-dose scan . journal Med. Phys. volume 38 , pages 5713--5731

    2011

  25. [33]

    , author Liang, Z

    author Ma, J. , author Liang, Z. , author Fan, Y. , author Liu, Y. , author Huang, J. , author Chen, W. , author Lu, H. , year 2012 . title Variance analysis of x-ray CT sinograms in the presence of electronic noise background . journal Med. Phys. volume 39 , pages 4051--4065

    2012

  26. [34]

    , author Ren, Y

    author Ma, Y.J. , author Ren, Y. , author Feng, P. , author He, P. , author Guo, X.D. , author Wei, B. , year 2021 . title Sinogram denoising via attention residual dense convolutional neural network for low-dose computed tomography . journal Nucl. Sci. Tech. volume 32 , pages 1--14

    2021

  27. [35]

    , author Rioux, J

    author Mason, A. , author Rioux, J. , author Clarke, S.E. , author Costa, A. , author Schmidt, M. , author Keough, V. , author Huynh, T. , author Beyea, S. , year 2019 . title Comparison of objective image quality metrics to expert radiologists’ scoring of diagnostic quality o...

    2019

  28. [36]

    , author Chen, B

    author Moen, T.R. , author Chen, B. , author Holmes III, D.R. , author Duan, X. , author Yu, Z. , author Yu, L. , author Leng, S. , author Fletcher, J.G. , author McCollough, C.H. , year 2021 . title Low-dose CT image and projection dataset . journal Med. Phys. volume 48 , pag...

    2021

  29. [37]

    , author Li, M

    author Niu, C. , author Li, M. , author Fan, F. , author Wu, W. , author Guo, X. , author Lyu, Q. , author Wang, G. , year 2022 a. title Noise suppression with similarity-based self-supervised deep learning . journal IEEE Trans. Med. Imag. volume 42 , pages 1590--1602

    2022

  30. [38]

    , author Li, M

    author Niu, C. , author Li, M. , author Guo, X. , author Wang, G. , year 2022 b. title Self-supervised dual-domain network for low-dose CT denoising , in: booktitle Proc. SPIE

    2022

  31. [39]

    , year 2020

    author Ronchetti, M. , year 2020 . title Torchradon: F ast differentiable routines for computed tomography . journal arXiv preprint arXiv:2009.14788

    2020 arXiv

  32. [40]

    , author Padole, A

    author Shan, H. , author Padole, A. , author Homayounieh, F. , author Kruger, U. , author Khera, R.D. , author Nitiwarangkul, C. , author Kalra, M.K. , author Wang, G. , year 2019 . title Competitive performance of a modularized deep neural network compared to commercial algor...

    2019

  33. [41]

    , author Zhang, Y

    author Shan, H. , author Zhang, Y. , author Yang, Q. , author Kruger, U. , author Kalra, M.K. , author Sun, L. , author Cong, W. , author Wang, G. , year 2018 . title 3- D convolutional encoder-decoder network for low-dose CT via transfer learning from a 2- D trained network ....

    2018

  34. [42]

    , author Awate, S.P

    author Sharma, V. , author Awate, S.P. , year 2024 . title Adversarial EM for variational deep learning: A pplication to semi-supervised image quality enhancement in low-dose PET and low-dose CT . journal Med. Image Anal. volume 97 , pages 103291

    2024

  35. [43]

    , author Bovik, A.C

    author Sheikh, H.R. , author Bovik, A.C. , year 2006 . title Image information and visual quality . journal IEEE Trans. Image Process. volume 15 , pages 430--444

    2006

  36. [44]

    , author Lipson, J

    author Smith-Bindman, R. , author Lipson, J. , author Marcus, R. , author Kim, K.P. , author Mahesh, M. , author Gould, R. , author De Gonz \'a lez, A.B. , author Miglioretti, D.L. , year 2009 . title Radiation dose associated with common computed tomography examinations and t...

    2009

  37. [45]

    , author Baeyens, P.F

    author Sodickson, A. , author Baeyens, P.F. , author Andriole, K.P. , author Prevedello, L.M. , author Nawfel, R.D. , author Hanson, R. , author Khorasani, R. , year 2009 . title Recurrent CT , cumulative radiation exposure, and associated radiation-induced cancer risks from C...

    2009

  38. [46]

    , author Weiss, E

    author Sohl-Dickstein, J. , author Weiss, E. , author Maheswaranathan, N. , author Ganguli, S. , year 2015 . title Deep unsupervised learning using nonequilibrium thermodynamics , in: booktitle Proc. Int. Conf. Mach. Learn. , pp. pages 2256--2265

    2015

  39. [47]

    , author Meng, C

    author Song, J. , author Meng, C. , author Ermon, S. , year 2021 a. title Denoising diffusion implicit models , in: booktitle Proc. Int. Conf. Learn. Represent

    2021

  40. [48]

    , author Sohl-Dickstein, J

    author Song, Y. , author Sohl-Dickstein, J. , author Kingma, D.P. , author Kumar, A. , author Ermon, S. , author Poole, B. , year 2021 b. title Score-based generative modeling through stochastic differential equations , in: booktitle Proc. Int. Conf. Learn. Represent

    2021

  41. [49]

    , author Palenstijn, W.J

    author Van Aarle, W. , author Palenstijn, W.J. , author Cant, J. , author Janssens, E. , author Bleichrodt, F. , author Dabravolski, A. , author De Beenhouwer, J. , author Joost Batenburg, K. , author Sijbers, J. , year 2016 . title Fast and flexible X -ray tomography using th...

    2016

  42. [50]

    , author Ye, J.C

    author Wang, G. , author Ye, J.C. , author De Man, B. , year 2020 . title Deep learning for tomographic image reconstruction . journal Nat. Mach. Intell. volume 2 , pages 737--748

    2020

  43. [51]

    , author Zhou, J

    author Wang, G. , author Zhou, J. , author Yu, Z. , author Wang, W. , author Qi, J. , year 2017 . title Hybrid pre-log and post-log image reconstruction for computed tomography . journal IEEE Trans. Med. Imag. volume 36 , pages 2457--2465

    2017

  44. [52]

    , author Li, T

    author Wang, J. , author Li, T. , author Lu, H. , author Liang, Z. , year 2006 . title Penalized weighted least-squares approach to sinogram noise reduction and image reconstruction for low-dose X -ray computed tomography . journal IEEE Trans. Med. Imag. volume 25 , pages 1272--1283

    2006

  45. [53]

    , author Bovik, A.C

    author Wang, Z. , author Bovik, A.C. , author Sheikh, H.R. , author Simoncelli, E.P. , year 2004 . title Image quality assessment: from error visibility to structural similarity . journal IEEE Trans. Image Process. volume 13 , pages 600--612

    2004

  46. [54]

    , author Zhang, Z

    author Wang, Z. , author Zhang, Z. , author Zhang, X. , author Zheng, H. , author Zhou, M. , author Zhang, Y. , author Wang, Y. , year 2023 . title DR2 : D iffusion-based robust degradation remover for blind face restoration , in: booktitle Proc. IEEE Conf. Comput. Vis. Patter...

    2023

  47. [55]

    , author Lu, Z

    author Xia, W. , author Lu, Z. , author Huang, Y. , author Liu, Y. , author Chen, H. , author Zhou, J. , author Zhang, Y. , year 2021 . title CT reconstruction with PDF : Parameter -dependent framework for data from multiple geometries and dose levels . journal IEEE Trans. Med...

    2021

  48. [56]

    , author Yurt, M

    author Xiang, T. , author Yurt, M. , author Syed, A.B. , author Setsompop, K. , author Chaudhari, A. , year 2023 . title DDM ^2 : Self-supervised diffusion MRI denoising with generative diffusion models , in: booktitle Proc. Int. Conf. Learn. Represent

    2023

  49. [57]

    , author Zeng, D

    author Xie, Q. , author Zeng, D. , author Zhao, Q. , author Meng, D. , author Xu, Z. , author Liang, Z. , author Ma, J. , year 2017 . title Robust low-dose CT sinogram preprocessing via exploiting noise-generating mechanism . journal IEEE Trans. Med. Imag. volume 36 , pages 2487--2498

    2017

  50. [58]

    , author Yuan, M

    author Xie, Y. , author Yuan, M. , author Dong, B. , author Li, Q. , year 2023 . title Diffusion model for generative image denoising . journal arXiv preprint arXiv:2302.02398

    2023 arXiv

  51. [59]

    , author Li, Z

    author Yang, L. , author Li, Z. , author Ge, R. , author Zhao, J. , author Si, H. , author Zhang, D. , year 2022 . title Low-dose CT denoising via sinogram inner-structure transformer . journal IEEE Trans. Med. Imag. volume 42 , pages 910--921

    2022

  52. [60]

    , author Zhao, Q

    author Yin, X. , author Zhao, Q. , author Liu, J. , author Yang, W. , author Yang, J. , author Quan, G. , author Chen, Y. , author Shu, H. , author Luo, L. , author Coatrieux, J.L. , year 2019 . title Domain progressive 3 D residual convolution network to improve low-dose CT i...

    2019

  53. [61]

    , author Zhang, H

    author Yu, J. , author Zhang, H. , author Zhang, P. , author Zhu, Y. , year 2023 . title Unsupervised learning-based dual-domain method for low-dose CT denoising . journal Phys. Med. Biol. volume 68 , pages 185010

    2023

  54. [62]

    , author Shiung, M

    author Yu, L. , author Shiung, M. , author Jondal, D. , author McCollough, C.H. , year 2012 . title Development and validation of a practical lower-dose-simulation tool for optimizing computed tomography scan protocols . journal J. Comput. Assist. Tomogr. volume 36 , pages 477--487

    2012

  55. [63]

    , author Huang, J

    author Zeng, D. , author Huang, J. , author Bian, Z. , author Niu, S. , author Zhang, H. , author Feng, Q. , author Liang, Z. , author Ma, J. , year 2015 . title A simple low-dose X -ray CT simulation from high-dose scan . journal IEEE Trans. Nucl. Sci. volume 62 , pages 2226--2233

    2015

  56. [64]

    , author Zhang, L

    author Zhang, L. , author Zhang, L. , author Mou, X. , author Zhang, D. , year 2011 . title FSIM : A feature similarity index for image quality assessment . journal IEEE Trans. Image Process. volume 20 , pages 2378--2386

    2011

  57. [65]

    , author Chen, X

    author Zhou, B. , author Chen, X. , author Zhou, S.K. , author Duncan, J.S. , author Liu, C. , year 2022 . title DuDoDR-Net : Dual-domain data consistent recurrent network for simultaneous sparse view and metal artifact reduction in computed tomography . journal Med. Image Ana...

    2022

  58. [66]

    , author Zhang, K

    author Zhu, Y. , author Zhang, K. , author Liang, J. , author Cao, J. , author Wen, B. , author Timofte, R. , author Van Gool, L. , year 2023 . title Denoising diffusion models for plug-and-play image restoration , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recognit. ...

    2023

  59. [67]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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