pith. sign in

arxiv: 2604.27552 · v1 · submitted 2026-04-30 · 💻 cs.CV

Residual Gaussian Splatting for Ultra Sparse-View CBCT Reconstruction

Jian Lin , Jiancheng Fang , Shaoyu Wang , Changan Lai , Yikun Zhang , Yang Chen , Qiegen Liu This is my paper

Pith reviewed 2026-05-07 08:10 UTC · model grok-4.3

classification 💻 cs.CV
keywords residual gaussian splattingsparse-view cbct reconstruction3d gaussian splattingwavelet multi-resolutionspectral decouplingartifact suppressiondetail preservationmedical tomography
0
0 comments X

The pith

Residual Gaussian Splatting recovers fine anatomical details from ultra-sparse CBCT views by separating geometric structure from residual textures.

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

The paper tries to establish that conventional 3D Gaussian splatting loses high-frequency anatomical details under ultra sparse X-ray views because of spectral bias toward smooth solutions. It proposes Residual Gaussian Splatting to fix this by splitting the volumetric representation into a low-frequency geometric base component and a high-frequency residual detail component, then optimizing them jointly with a spectral-spatial strategy. This decomposition sidesteps the mismatch between bipolar wavelet coefficients and the required non-negative X-ray attenuation values. A sympathetic reader would care because fewer projection views directly lower radiation dose to patients while still delivering clear images of fine structures such as trabecular bone and blood vessels. If the claim holds, it shows a practical route to high-fidelity 3D medical imaging without the usual dense sampling requirement.

Core claim

By introducing a spectrally-decoupled Gaussian representation that stratifies the volumetric field into a geometric base component and a residual detail component, Residual Gaussian Splatting converts explicit high-frequency fitting into a physically consistent implicit residual compensation task. A spectral-spatial collaborative optimization strategy then coordinates geometric anchoring with texture refinement to prevent spectral crosstalk, enabling the method to produce reconstructions that capture highly refined geometric textures while maintaining non-negative X-ray attenuation on clinical datasets.

What carries the argument

Spectrally-decoupled Gaussian representation that stratifies the volumetric field into a geometric base component and a residual detail component, paired with spectral-spatial collaborative optimization to handle high-frequency residuals.

If this is right

  • Reconstructed images capture highly refined geometric textures in complex trabecular and vascular structures.
  • The approach resolves the trade-off between artifact suppression and detail preservation.
  • It yields superior visual fidelity compared to existing neural rendering baselines under ultra sparse-view conditions.
  • The method maintains physical consistency with non-negative X-ray attenuation across clinical datasets.

Where Pith is reading between the lines

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

  • The same base-plus-residual split could be tested on other explicit representations such as voxel grids or neural fields for similar inverse problems.
  • If validated on larger cohorts, the technique could support clinical protocols that acquire CBCT with substantially fewer projections and lower patient dose.
  • Applying the residual compensation stage to 4D or motion-affected data might address temporal inconsistencies in dynamic imaging.
  • The wavelet-inspired multi-resolution handling suggests extensions to multi-modal fusion tasks where spectral separation is also needed.

Load-bearing premise

The decomposition into geometric base and residual detail components together with the joint optimization strategy can block spectral crosstalk and keep all attenuation values non-negative.

What would settle it

If the reconstructed volumes on clinical CBCT test sets still exhibit over-smoothing of trabecular patterns or produce negative attenuation values when compared to dense-view ground truth, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2604.27552 by Changan Lai, Jiancheng Fang, Jian Lin, Qiegen Liu, Shaoyu Wang, Yang Chen, Yikun Zhang.

Figure 1
Figure 1. Figure 1: Schematic illustration of the motivation and core concept of the view at source ↗
Figure 2
Figure 2. Figure 2: Schematic illustration of the CBCT imaging geometry. view at source ↗
Figure 3
Figure 3. Figure 3: Visual and quantitative analysis demonstrating the inherent spectral view at source ↗
Figure 4
Figure 4. Figure 4: Schematic overview of the proposed RGS framework. (a) Spectrally-decoupled Gaussian initialization. Sparse raw projections are decomposed via view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison of reconstructed chest and abdomen slices from the AAPM dataset under a 20-view sparse configuration. Magnified regions of view at source ↗
Figure 6
Figure 6. Figure 6: Visual ablation study on the wavelet-based spectral prior under a 40- view at source ↗
Figure 7
Figure 7. Figure 7: Quantitative frequency analysis evaluating the effectiveness of the view at source ↗
Figure 9
Figure 9. Figure 9: Visual comparison of reconstructed slices for a real rhinoceros beetle view at source ↗
Figure 8
Figure 8. Figure 8: Visual ablation study evaluating the impact of curriculum optimization view at source ↗
read the original abstract

While 3D Gaussian splatting (3DGS) offers explicit and efficient scene representations for cone-beam computed tomography reconstruction, conventional photometric optimization inherently suffers from spectral bias under ultra sparse-view conditions, leading to over-smoothing and a loss of high-frequency anatomical details. Since wavelet transforms provide rich high-frequency information and have been widely utilized to enhance sparse reconstruction, this work integrates wavelet multi-resolution analysis with 3DGS. To circumvent the mathematical mismatch between the strict non-negativity of physical X-ray attenuation and the bipolar nature of high-frequency wavelet coefficients, we propose Residual Gaussian Splatting (RGS). Methodologically, we introduce a spectrally-decoupled Gaussian representation that stratifies the volumetric field into a geometric base component and a residual detail component. This decomposition systematically transforms explicit high-frequency fitting into a physically consistent, implicit residual compensation task. Furthermore, we devise a spectral-spatial collaborative optimization strategy to coordinate the interplay between geometric anchoring and texture refinement, effectively preventing spectral crosstalk. Extensive experiments on clinical datasets demonstrate that RGS enables the reconstructed images to capture highly refined geometric textures. It successfully resolves the trade-off between artifact suppression and detail preservation, yielding superior visual fidelity in complex trabecular and vascular structures compared to existing neural rendering baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Residual Gaussian Splatting (RGS) for ultra sparse-view CBCT reconstruction. It augments 3D Gaussian splatting with wavelet multi-resolution analysis via a spectrally-decoupled representation that decomposes the volumetric attenuation field into a geometric base component and a bipolar residual detail component. This converts explicit high-frequency fitting into an implicit residual compensation task. A spectral-spatial collaborative optimization strategy is introduced to coordinate the components and avoid crosstalk. The authors claim that experiments on clinical datasets demonstrate superior capture of refined geometric textures, resolution of the artifact-detail trade-off, and better visual fidelity in trabecular and vascular structures relative to existing neural rendering baselines.

Significance. If the central claims are substantiated, the work could meaningfully advance explicit scene representations for medical CT by reconciling the efficiency of 3DGS with wavelet-derived high-frequency content while respecting non-negativity. This addresses a practical limitation in sparse-view CBCT, where radiation dose reduction is clinically important, and offers a concrete decomposition strategy that may generalize beyond the current setting. The approach is technically novel relative to prior 3DGS adaptations in tomography.

major comments (2)
  1. [§3 (spectrally-decoupled Gaussian representation)] §3 (spectrally-decoupled Gaussian representation): the claim that the decomposition 'systematically transforms explicit high-frequency fitting into a physically consistent, implicit residual compensation task' and maintains 'physical consistency with non-negative X-ray attenuation' is load-bearing. The residual component is explicitly bipolar (high-frequency wavelet coefficients), yet the manuscript provides no explicit non-negativity mechanism (ReLU, softplus, projection, or loss penalty) on the summed field. In ultra-sparse regimes the photometric loss can still drive negative densities in low-signal voxels; without such a constraint or a proof that optimization preserves non-negativity, the physical-consistency guarantee does not follow from the construction.
  2. [§5 (experiments)] §5 (experiments): the abstract and results assert 'superior visual fidelity' and successful resolution of the artifact-detail trade-off on clinical datasets, but no quantitative metrics (PSNR, SSIM, MAE, or equivalent), error bars, or statistical tests are reported. Without these, together with ablations isolating the base/residual split and the collaborative optimization, it is impossible to verify that observed improvements exceed what could be obtained by post-hoc tuning of existing 3DGS baselines.
minor comments (2)
  1. [Abstract] The abstract would be strengthened by stating the precise number of views (e.g., '4–8 views') and naming the clinical datasets, allowing readers to immediately gauge the ultra-sparse regime.
  2. [§3] Notation for the base and residual Gaussians (e.g., how the wavelet coefficients are mapped to residual splat parameters) should be introduced with explicit equations rather than descriptive prose only.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the insightful comments on our manuscript. We address each of the major comments below and will make the necessary revisions to strengthen the paper.

read point-by-point responses

  1. Referee: §3 (spectrally-decoupled Gaussian representation): the claim that the decomposition 'systematically transforms explicit high-frequency fitting into a physically consistent, implicit residual compensation task' and maintains 'physical consistency with non-negative X-ray attenuation' is load-bearing. The residual component is explicitly bipolar (high-frequency wavelet coefficients), yet the manuscript provides no explicit non-negativity mechanism (ReLU, softplus, projection, or loss penalty) on the summed field. In ultra-sparse regimes the photometric loss can still drive negative densities in low-signal voxels; without such a constraint or a proof that optimization preserves non-negativity, the physical-consistency guarantee does not follow from the construction.

    Authors: We thank the referee for highlighting this critical aspect of physical consistency. Upon re-examination, we recognize that while the base component is designed to be non-negative, an explicit safeguard on the summed field is indeed beneficial for ultra-sparse scenarios. In the revised manuscript, we will add a non-negativity enforcement mechanism, specifically applying a ReLU activation to the final attenuation values after summing the base and residual components, along with a regularization term in the loss to penalize any residual negative values. This will be detailed in the updated §3, including a short proof sketch that the optimization maintains non-negativity. The core decomposition remains valid as it converts high-frequency fitting to residual compensation within this constrained framework. revision: yes

  2. Referee: §5 (experiments): the abstract and results assert 'superior visual fidelity' and successful resolution of the artifact-detail trade-off on clinical datasets, but no quantitative metrics (PSNR, SSIM, MAE, or equivalent), error bars, or statistical tests are reported. Without these, together with ablations isolating the base/residual split and the collaborative optimization, it is impossible to verify that observed improvements exceed what could be obtained by post-hoc tuning of existing 3DGS baselines.

    Authors: We agree that quantitative metrics and ablations are necessary to fully substantiate the claims. The original submission prioritized visual and qualitative analysis to demonstrate the resolution of the artifact-detail trade-off in complex anatomical structures. For the revision, we will incorporate PSNR, SSIM, and MAE metrics with error bars computed over the clinical dataset cases. We will also add ablation experiments that isolate the effects of the base/residual decomposition and the spectral-spatial collaborative optimization. These will be compared to appropriately tuned 3DGS baselines to show the specific contributions of our method. The updated results section will include tables summarizing these quantitative findings and statistical significance where applicable. revision: yes

Circularity Check

0 steps flagged

Derivation chain is self-contained; new decomposition and optimization add independent content

full rationale

The paper begins from the established spectral bias of photometric 3DGS optimization under ultra-sparse views and the external literature on wavelets for sparse reconstruction. It then identifies the non-negativity mismatch with bipolar wavelet coefficients and introduces an explicit methodological choice: a spectrally-decoupled Gaussian representation that splits the field into a geometric base component plus a residual detail component. This split is presented as transforming the fitting task into implicit residual compensation, accompanied by a newly devised spectral-spatial collaborative optimization. No equation or claim reduces a 'prediction' or result to a fitted parameter or prior definition by construction. No self-citation is invoked as a load-bearing uniqueness theorem, no ansatz is smuggled via prior work, and no renaming of a known empirical pattern occurs. The central construction (base + residual + collaborative optimization) is therefore an independent modeling decision rather than a tautological re-expression of inputs. The derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 1 invented entities

The central claim rests on standard assumptions from 3D Gaussian splatting and wavelet theory plus the new residual decomposition; no numerical free parameters are stated in the abstract, and no new physical entities beyond the proposed representation are introduced.

axioms (3)
  • domain assumption Wavelet transforms provide rich high-frequency information useful for enhancing sparse reconstruction
    Invoked to motivate integration with 3DGS for detail preservation
  • domain assumption 3D Gaussian splatting provides explicit and efficient scene representations for CBCT
    Base representation whose photometric optimization suffers from spectral bias under sparse views
  • domain assumption X-ray attenuation coefficients are strictly non-negative
    Creates the mathematical mismatch with bipolar wavelet coefficients that the residual approach is designed to circumvent
invented entities (1)
  • Residual Gaussian Splatting (RGS) with spectrally-decoupled representation no independent evidence
    purpose: To transform explicit high-frequency fitting into a physically consistent implicit residual compensation task
    New stratified representation (geometric base + residual detail) introduced to reconcile wavelet coefficients with non-negative attenuation

pith-pipeline@v0.9.0 · 5530 in / 1683 out tokens · 65556 ms · 2026-05-07T08:10:42.695493+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages

  1. [1]

    Compressed sensing,

    D. Donoho, “Compressed sensing,”IEEE Transactions on Information Theory, vol. 52, no. 4, pp. 1289–1306, 2006

    work page 2006

  2. [2]

    Sparse MRI: The application of compressed sensing for rapid MR imaging,

    M. Lustig, D. Donoho, and J. M. Pauly, “Sparse MRI: The application of compressed sensing for rapid MR imaging,”Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine, vol. 58, no. 6, pp. 1182–1195, 2007

    work page 2007

  3. [3]

    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, Sep. 2012

    work page 2012

  4. [4]

    K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation,

    M. Aharon, M. Elad, and A. Bruckstein, “K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation,”IEEE Transactions on Signal Processing, vol. 54, no. 11, pp. 4311–4322, Nov. 2006

    work page 2006

  5. [5]

    A theory for multiresolution signal decomposition: The wavelet representation,

    S. Mallat, “A theory for multiresolution signal decomposition: The wavelet representation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 11, no. 7, pp. 674–693, Jul. 1989

    work page 1989

  6. [6]

    Deep learning for tomographic image reconstruction,

    G. Wang, J. C. Ye, and B. De Man, “Deep learning for tomographic image reconstruction,”Nature Machine Intelligence, vol. 2, no. 12, pp. 737–748, Dec. 2020

    work page 2020

  7. [7]

    Deep learning-based algorithms for low-dose CT imaging: A review,

    H. Chen, Q. Li, L. Zhou, and F. Li, “Deep learning-based algorithms for low-dose CT imaging: A review,”European Journal of Radiology, vol. 172, Mar. 2024

    work page 2024

  8. [8]

    Deep learning computed tomography,

    T. W ¨urfl, F. C. Ghesu, V . Christlein, and A. Maier, “Deep learning computed tomography,” inMedical Image Computing and Computer- Assisted Intervention - MICCAI 2016, S. Ourselin, L. Joskowicz, M. R. Sabuncu, G. Unal, and W. Wells, Eds. Cham: Springer International Publishing, 2016, pp. 432–440

    work page 2016

  9. [9]

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

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

    work page 2018

  10. [10]

    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, Mar. 2019

    work page 2019

  11. [11]

    Stage-by-stage wavelet optimization refinement diffusion model for sparse-view CT reconstruc- tion,

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

    work page 2024

  12. [12]

    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, Feb. 2024

    work page 2024

  13. [13]

    Dual-domain collaborative diffusion sampling for multi-source stationary computed tomography reconstruction,

    Z. Li, D. Chang, Z. Zhang, F. Luo, Q. Liu, J. Zhang, G. Yang, and W. Wu, “Dual-domain collaborative diffusion sampling for multi-source stationary computed tomography reconstruction,”IEEE Transactions on Medical Imaging, vol. 43, no. 10, pp. 3398–3411, Oct. 2024

    work page 2024

  14. [14]

    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, Jan. 2021

    work page 2021

  15. [15]

    Genera- tive adversarial networks for noise reduction in low-dose CT,

    J. M. Wolterink, T. Leiner, M. A. Viergever, and I. I ˇsgum, “Genera- tive adversarial networks for noise reduction in low-dose CT,”IEEE Transactions on Medical Imaging, vol. 36, no. 12, pp. 2536–2545, Dec. 2017

    work page 2017

  16. [16]

    DuDoRNet: Learning a dual-domain recurrent network for fast MRI reconstruction with deep T1 prior,

    B. Zhou and S. K. Zhou, “DuDoRNet: Learning a dual-domain recurrent network for fast MRI reconstruction with deep T1 prior,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2020, pp. 4273–4282

    work page 2020

  17. [17]

    DuDoNet: Dual domain network for CT metal artifact reduction,

    W.-A. Lin, H. Liao, C. Peng, X. Sun, J. Zhang, J. Luo, R. Chellappa, and S. K. Zhou, “DuDoNet: Dual domain network for CT metal artifact reduction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 10 512–10 521

    work page 2019

  18. [18]

    Momentum-Net: Fast and convergent iterative neural network for inverse problems,

    I. Y . Chun, Z. Huang, H. Lim, and J. A. Fessler, “Momentum-Net: Fast and convergent iterative neural network for inverse problems,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 4, pp. 4915–4931, Apr. 2023

    work page 2023

  19. [19]

    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, Dec. 2021

    work page 2021

  20. [20]

    3D Gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3D Gaussian splatting for real-time radiance field rendering,” Aug. 2023

    work page 2023

  21. [21]

    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,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2022, L. Wang, Q. Dou, P. T. Fletcher, S. Speidel, and S. Li, Eds. Cham: Springer Nature Switzerland, 2022, pp. 442–452

    work page 2022

  22. [22]

    Radiative gaussian splatting for efficient x-ray novel view synthesis,

    Y . Cai, Y . Liang, J. Wang, A. Wang, Y . Zhang, X. Yang, Z. Zhou, and A. Yuille, “Radiative gaussian splatting for efficient x-ray novel view synthesis,” inComputer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds. Cham: Springer Nature Switzerland, 2025, pp. 283–299

    work page 2024

  23. [23]

    R$ˆ2$-Gaussian: Rectifying radiative Gaussian splatting for tomographic reconstruction,

    R. Zha, T. J. Lin, Y . Cai, J. Cao, Y . Zhang, and H. Li, “R$ˆ2$-Gaussian: Rectifying radiative Gaussian splatting for tomographic reconstruction,” Oct. 2024

    work page 2024

  24. [24]

    3DGR-CT: Sparse- view CT reconstruction with a 3D Gaussian representation,

    Y . Li, X. Fu, H. Li, S. Zhao, R. Jin, and S. K. Zhou, “3DGR-CT: Sparse- view CT reconstruction with a 3D Gaussian representation,”Medical Image Analysis, vol. 103, p. 103585, Jul. 2025

    work page 2025

  25. [25]

    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,” in Proceedings of the 36th International Conference on Machine Learning. PMLR, May 2019, pp. 5301–5310

    work page 2019

  26. [26]

    What is cone-beam CT and how does it work?

    W. C. Scarfe and A. G. Farman, “What is cone-beam CT and how does it work?”Dental Clinics of North America, vol. 52, no. 4, pp. 707–730, Oct. 2008

    work page 2008

  27. [27]

    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

    work page 2006

  28. [28]

    Practical cone-beam algorithm,

    L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beam algorithm,”JOSA A, vol. 1, no. 6, pp. 612–619, Jun. 1984

    work page 1984

  29. [29]

    EW A volume splatting,

    M. Zwicker, H. Pfister, J. van Baar, and M. Gross, “EW A volume splatting,” inProceedings Visualization, 2001. VIS ’01., Oct. 2001, pp. 29–538

    work page 2001

  30. [30]

    Mallat,A wavelet tour of signal processing

    S. Mallat,A wavelet tour of signal processing. Elsevier, Sep. 1999

    work page 1999

  31. [31]

    A fast iterative shrinkage-thresholding algo- rithm for linear inverse problems,

    A. Beck and M. Teboulle, “A fast iterative shrinkage-thresholding algo- rithm for linear inverse problems,”SIAM Journal on Imaging Sciences, vol. 2, no. 1, pp. 183–202, 2009

    work page 2009

  32. [32]

    Distributed optimization and statistical learning via the alternating direction method of multipliers,

    B. Stephen, P. Neal, C. Eric, P. Borja, and E. Jonathan, “Distributed optimization and statistical learning via the alternating direction method of multipliers,”Foundations and Trends in Information Retrieval, vol. 3, no. 1, pp. 1–122, Jul. 2011

    work page 2011

  33. [33]

    Improved compressed sensing-based algorithm for sparse-view CT image reconstruction,

    Z. Zhu, K. Wahid, P. Babyn, D. Cooper, I. Pratt, and Y . Carter, “Improved compressed sensing-based algorithm for sparse-view CT image reconstruction,”Computational and Mathematical Methods in Medicine, vol. 2013, no. 1, p. 185750, 2013

    work page 2013

  34. [34]

    Prior image constrained compressed sensing (PICCS): A method to accurately reconstruct dynamic CT im- ages from highly undersampled projection data sets,

    G.-H. Chen, J. Tang, and S. Leng, “Prior image constrained compressed sensing (PICCS): A method to accurately reconstruct dynamic CT im- ages from highly undersampled projection data sets,”Medical Physics, vol. 35, no. 2, pp. 660–663, 2008

    work page 2008

  35. [35]

    Sparse-view x-ray CT reconstruction via total generalized variation regularization,

    S. Niu, Y . Gao, Z. Bian, J. Huang, W. Chen, G. Yu, Z. Liang, and J. Ma, “Sparse-view x-ray CT reconstruction via total generalized variation regularization,”Physics in Medicine & Biology, vol. 59, no. 12, p. 2997, May 2014

    work page 2014

  36. [36]

    Compressed sensing: From research to clinical practice with deep neural networks: Shortening scan times for magnetic resonance imaging,

    C. M. Sandino, J. Y . Cheng, F. Chen, M. Mardani, J. M. Pauly, and S. S. Vasanawala, “Compressed sensing: From research to clinical practice with deep neural networks: Shortening scan times for magnetic resonance imaging,”IEEE Signal Processing Magazine, vol. 37, no. 1, pp. 117–127, Jan. 2020

    work page 2020

  37. [37]

    ISTA-Net: Interpretable optimization-inspired deep network for image compressive sensing,

    J. Zhang and B. Ghanem, “ISTA-Net: Interpretable optimization-inspired deep network for image compressive sensing,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 1828–1837

    work page 2018

  38. [38]

    State-of-the-art deep learning CT reconstruction algorithms in abdominal imaging,

    A. Mileto, L. Yu, J. W. Revels, S. Kamel, M. A. Shehata, J. J. Ibarra- Rovira, V . K. Wong, A. M. Roman-Colon, J. M. Lee, K. M. Elsayes, and C. T. Jensen, “State-of-the-art deep learning CT reconstruction algorithms in abdominal imaging,”RadioGraphics, vol. 44, no. 12, p. e240095, Dec. 2024

    work page 2024