Pith. sign in

REVIEW 3 major objections 4 minor 26 references

A geometry-aware Gaussian model synthesizes anatomically consistent multi-view spine projections across X-ray and ultrasound without explicit 3D reconstruction.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A radiative 3D Gaussian model with structure-prior loss reweighting synthesizes consistent multi-view spine projections for CT and ultrasound, beating neural-field baselines in quality and speed.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Solid radiative-Gaussian spine synthesizer with real speed/quality gains on CT; the ultrasound “cross-modality” claim is mostly DRR-of-volume, so scope the paper carefully. the 3 major comments →

arxiv 2607.04923 v1 pith:KU2REZMX submitted 2026-07-06 cs.CV

UniSpine-GS: An Efficient Physics-Aware Gaussian Framework for Cross-Modality Multi-view Spine Image Synthesis

Qiuhua Chen , Changning Yu , Na Huang , Chao Sun , Bo Du This is my paper
classification cs.CV
keywords Multi-view Spine ImagingCross-Modality 3D SynthesisX-Gaussian RepresentationMedical Image Synthesis3D Gaussian SplattingStructure Prior Weight MapNovel-view Synthesis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Spinal diagnosis benefits from three-dimensional assessment, yet dedicated 3D scanners are costly and models often fail when imaging physics change between X-ray transmission and ultrasound reflection. This paper claims that an explicit set of radiative 3D Gaussians, optimized under one shared projection operator, can render novel views that stay anatomically consistent across those modalities. A structure-prior weight map reweights the training loss so vertebral boundaries and fine texture remain sharp under sparse views. On a public CT spine set and a new fetal ultrasound volume set, the method reports large gains in image quality while training in minutes and rendering at over 100 frames per second. If the claim holds, multi-view spinal assessment becomes practical from cheaper, more accessible acquisitions.

Core claim

UniSpine-GS establishes that a geometry-aware radiative Gaussian representation, driven by a unified digitally reconstructed radiograph forward model and reweighted by a Structure Prior Weight Map, produces multi-view spine projections that are higher quality and far more efficient than neural-field baselines on both CTSpine3D and FeSpine3D.

What carries the argument

Structure Prior Weight Map (SPWM): a parameter-light map built from Sobel gradients and high-frequency residuals of the training projection, which reweights a Charbonnier loss (with warm-up and adaptive gating) so optimization focuses on vertebral boundaries and texture-rich regions under a shared radiative Gaussian renderer.

Load-bearing premise

The method treats a 3D ultrasound volume as a simple projectable scalar field and reuses the same X-ray-style projection operator, assuming that proxy is close enough for true multi-view ultrasound synthesis.

What would settle it

Acquire real multi-angle 2D ultrasound sweeps of the same fetal spines used in FeSpine3D and check whether UniSpine-GS novel views match those real acquisitions as closely as they match the synthetic DRR projections; a large quality drop would falsify the cross-modality claim.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Multi-view spinal assessment becomes feasible without expensive dedicated 3D imaging hardware.
  • One pipeline can serve both X-ray/CT projection settings and ultrasound volume projections under the same geometry-aware representation.
  • Training finishes in minutes and rendering exceeds 100 fps, making interactive clinical use practical.
  • Sparse-view supervision yields sharper vertebral boundaries and fewer low-contrast artifacts than neural-field baselines.
  • The new FeSpine3D dataset enables further public work on fetal spine multi-view synthesis.

Where Pith is reading between the lines

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

  • If the DRR proxy for ultrasound does not match real multi-angle acoustic formation, clinics would still need modality-specific forward models before live fetal deployment.
  • The same structure-weighted Gaussian idea could transfer to other sparse-view medical projection tasks where boundary fidelity matters more than raw intensity match.
  • Releasing FeSpine3D may turn it into a default benchmark for ultrasound novel-view methods even if later work replaces the Gaussian backbone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. UniSpine-GS proposes an explicit radiative 3D Gaussian representation for multi-view 2D spine projection synthesis across CT/X-ray and ultrasound. Building on X-Gaussian, it uses angle-pose cuboid initialization (ACUI), a Radiation Intensity Response Function (RIRF), and a unified differentiable DRR-style transmittance compositor (Eqs. 3–5). A Structure Prior Weight Map (SPWM) reweights a hybrid SSIM–Charbonnier loss toward edges and high-frequency regions with warm-up and quantile gating. The method is evaluated on CTSpine3D and a new FeSpine3D fetal ultrasound volume set, reporting large gains in PSNR/SSIM and orders-of-magnitude speedups over IntraTomo, NeRF, TensoRF, and NAF (Table 1), with ablations attributing gains to RIRF+ACUI and SPWM (Table 2). Code is released.

Significance. If the claims hold under clinically realistic multi-view acquisition, the work would offer a practical, efficient alternative to expensive 3D hardware for multi-view spine assessment and a unified representation across modalities. Strengths include public code, clear efficiency numbers (e.g., ~113–148 fps and short training on a single 3090), a parameter-light structure-guided loss, and a new FeSpine3D resource. The CT/X-ray results are a solid contribution to radiative Gaussian novel-view synthesis. The cross-modality clinical significance, however, depends on whether the ultrasound evaluation protocol matches real multi-view ultrasound formation; that assumption is currently the main limit on impact.

major comments (3)
  1. [§2, §3.1, Abstract, Conclusion] Cross-modality ultrasound claim vs. evaluation protocol (§2 intro, §2.2 Eqs. 3–5, §3.1, Conclusion): FeSpine3D multi-view labels are generated by treating each 3D US volume as a projectable scalar field and rendering with the same cone-beam DRR/transmittance operator used for CT. Real ultrasound is reflection/scattering-driven and view-dependent, not a transmission integral. The large FeSpine3D gains in Table 1 therefore measure fit to DRR-of-US-volume, not clinical multi-view US synthesis. The abstract/title framing of a physics-aware unified cross-modality solution overstates what the experiments support. Either reframe claims to “DRR-style multi-view synthesis from CT and US volumes,” move the acoustic-proxy limitation into the abstract and evaluation design, or add evidence on real multi-view ultrasound (or a more acoustic-consistent forward model).
  2. [Table 1, Table 2] Table 1 vs Table 2 inconsistency on FeSpine3D: UniSpine-GS is reported as 40.35 PSNR / 0.9815 SSIM in Table 1 but 40.82 PSNR / 0.9846 SSIM in Table 2 under the full model. Training times are also reported in minutes (Table 1) versus seconds (Table 2) without a clear mapping. These discrepancies undermine confidence in the headline numbers. Please reconcile the protocols (same splits, iterations, seeds) and report a single consistent set of results, ideally with multi-seed means and standard deviations.
  3. [§3.1–3.2, Tables 1–2] Statistical and experimental rigor (Tables 1–2, §3.1–3.2): All quality metrics are single-point averages with no error bars, no multi-seed runs, and no description of train/test view sampling (number of input views, angular spacing, held-out protocol). Given free parameters for SPWM (α, λhf, q, warm-up), densification, and learning rates, sensitivity or multi-run statistics are needed to support “significantly outperforms … across all metrics.” At minimum, report view counts, split protocol, and variance over seeds or subjects.
minor comments (4)
  1. [§2, §2.3, Tables] Typographical/formatting issues: “densificationpruning” (missing hyphen/space, §2), “Structure Prior W eight Map” (stray space, §2.3), “qualityefficiency” (missing hyphen, §3.2/Conclusion), and mixed “F eSpine3D” spacing in tables.
  2. [Fig. 1, §2.3] Fig. 1 caption says SPWM is “derived from the Camera Viewpoint image”; §2.3 states it is computed from the ground-truth training projection. Align caption and text.
  3. [§1, §3.2] Baselines are neural-field methods; a short discussion of other radiative/Gaussian medical NVS methods (including X-Gaussian as the direct backbone) would better situate novelty of SPWM versus the inherited RIRF+ACUI stack.
  4. [§3.1] Implementation lists many hyperparameters (§3.1) without a sensitivity study; even a brief note on robustness of α, q, and warm-up would help reproducibility beyond the released code.

Circularity Check

0 steps flagged

No circularity: empirical Gaussian fitting scored on held-out views; SPWM is training-only loss reweighting, not a definition of the reported metrics.

full rationale

UniSpine-GS is a standard empirical novel-view synthesis pipeline. A set of radiative Gaussians is optimized against multi-view training projections via a hybrid SSIM + weighted Charbonnier loss (Eqs. 8–9) and evaluated with unweighted PSNR/SSIM on held-out views (Table 1). SPWM (Eqs. 6–7) is computed only from ground-truth training images as a soft loss reweighting with warm-up; it does not enter the evaluation metrics and does not define the predicted images. ACUI and the radiative forward model (RIRF, transmittance compositing Eqs. 2–5) are adopted from X-Gaussian (Cai et al., different author set) and disclosed as such; they are not self-derived uniqueness claims. Overlapping-author citation of Veganet [20] is incidental and not load-bearing for the quantitative claims. No fitted scalar is renamed as a prediction of a closely related quantity, and no result reduces by construction to its inputs. The ultrasound DRR-proxy validity issue is a domain-transfer / correctness concern, not circularity under this rubric.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 3 invented entities

The central performance claim rests on standard differentiable rendering math, domain assumptions about known cone-beam geometry and DRR-as-proxy for ultrasound, many hand-chosen training/SPWM hyperparameters, and the invented SPWM module plus FeSpine3D corpus. Most representational machinery is imported from X-Gaussian rather than derived here.

free parameters (8)
  • SPWM strength α
    Hand-set to 0.8; scales how much structure attention reweights the Charbonnier loss and directly affects reported boundary fidelity.
  • SPWM high-frequency balance λhf
    Set to 0.5 when fusing Sobel gradients with Gaussian-blur residuals into attention map A.
  • SPWM adaptive quantile q
    q=0.85 sets the soft gate threshold τ on A; changes which pixels receive extra weight.
  • SPWM warm-up schedule (ts, duration)
    Warm-up starts at iteration 2000 for 5000 iterations; controls when structure weighting ramps up and affects optimization stability claims.
  • Gaussian attribute learning rates
    Feature 0.002, opacity 0.008, scaling 0.005, rotation 0.001; chosen by hand for 20k-iteration Adam training.
  • Densification schedule and gradient threshold
    Densify iterations 500–8000 every 200 steps with threshold 1.5e-4; controls final Gaussian count and quality/efficiency trade-off.
  • Cuboid sampling interval d and dimensions S1,S2,S3
    ACUI initializes Gaussian centers on a cuboid grid; spacing and box size are geometry choices that seed the representation.
  • Hybrid loss SSIM weight λdssim
    Balances weighted Charbonnier vs SSIM in Eq. (8); value not numerically fixed in text but is a free training knob.
axioms (6)
  • domain assumption A spine can be adequately represented for multi-view projection synthesis by a finite set of anisotropic 3D Gaussians with radiative features.
    Foundational model §2.2 Eq. (1); imported from 3DGS/X-Gaussian practice without independent anatomical proof.
  • domain assumption Known cone-beam scanner geometry yields accurate camera extrinsics/intrinsics for ACUI without SfM.
    §2.1; required for pose-free initialization and the unified forward operator.
  • domain assumption Transmittance-based radiative compositing (DRR-style) is a valid unified forward model for both X-ray and the ultrasound experiments.
    §2 intro and Eqs. (3)–(5); authors note ultrasound is only a pragmatic proxy, not exact acoustics.
  • ad hoc to paper Sobel gradients plus high-frequency blur residuals identify anatomically informative pixels worth up-weighting.
    SPWM §2.3 Eq. (6); structure prior is engineered, not derived from imaging physics.
  • domain assumption Held-out synthetic projections rendered from the same 3D volumes are a sufficient test of multi-view synthesis quality for the clinical motivation.
    §3.1 evaluation protocol; bridges reported metrics to the cost-effective clinical narrative.
  • standard math Standard autodiff optimization with densification/pruning converges to a geometry-aware representation under sparse views.
    Training pipeline following X-Gaussian/3DGS; ordinary gradient-based fitting assumption.
invented entities (3)
  • SPWM (Structure Prior Weight Map) no independent evidence
    purpose: Reweight reconstruction loss toward vertebral boundaries and texture-rich regions with warm-up and quantile gating.
    Main paper-specific module (§2.3–2.4); independent evidence is only the ablation gains on the same benchmarks, not an external physical measurement.
  • UniSpine-GS framework no independent evidence
    purpose: Name the end-to-end radiative Gaussian pipeline for cross-modality multi-view spine synthesis.
    System packaging of X-Gaussian components plus SPWM; not a new physical entity.
  • FeSpine3D dataset no independent evidence
    purpose: Provide 100 fetal ultrasound volumes for multi-view synthesis evaluation.
    New constructed corpus (§3.1); value depends on public release and clinical protocol documentation beyond de-identification note.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of UniSpine-GS: An Efficient Physics-Aware Gaussian Framework for Cross-Modality Multi-view Spine Image Synthesis." pith.science (2026). https://pith.science/paper/KU2REZMX

@misc{pith2026260704923,
  author       = {Pith},
  title        = {Pith review of: UniSpine-GS: An Efficient Physics-Aware Gaussian Framework for Cross-Modality Multi-view Spine Image Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KU2REZMX}},
  note         = {Machine review of arXiv:2607.04923}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The diagnosis of spinal diseases is often assisted by 3D imaging techniques in clinical practice. However, precise 3D spinal assessment is limited by the high costs of 3D imaging hardware and the challenges posed by the physical differences between imaging modalities, which hinder the generalizability of models. To address these issues, we propose UniSpine-GS, an efficient, physics-aware Gaussian framework designed for novel-view projection rendering in multi-view spine imaging via a 3D-aware representation. Instead of performing explicit 3D reconstruction, our approach learns a geometry-aware Gaussian representation that ensures anatomical consistency across different views. We introduce SPWM, a structure-guided loss reweighting strategy to improve boundary fidelity and local details. We evaluate our method on the CTSpine3D dataset and a newly constructed 3D fetal ultrasound dataset, FeSpine3D. Our results demonstrate that UniSpine-GS significantly outperforms existing methods across all metrics, offering a practical and cost-effective solution for unified multi-view medical imaging. Our code is publicly available at https://github.com/orangeisland66/UniSpine-GS.

Figures

Figures reproduced from arXiv: 2607.04923 by Bo Du, Changning Yu, Chao Sun, Na Huang, Qiuhua Chen.

Figure 1
Figure 1. Figure 1: Overview of UniSpine-GS. We optimize a radiative Gaussian representation for multi-view spine synthesis using a unified forward operator. An SPWM derived from the Camera Viewpoint image reweights the reconstruction loss, together with an SSIM term, to emphasize structurally informative regions. 2.1 Angle-pose Cuboid Initialization (ACUI) Efficient optimization requires reliable initialization of camera pos… view at source ↗
Figure 2
Figure 2. Figure 2: Visualization results on (a) CTSpine3D and (b) FeSpine3D datasets. 4 Conclusion We propose UniSpine-GS, an efficient physics-aware Gaussian framework for cross-modality multi-view spine image synthesis. With an explicit geometry- [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

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

26 extracted references · 4 linked inside Pith

  1. [1]

    IEEE Transactions on Circuits and Systems for Video Technology 35(7), 6832–6852 (2025)

    Bao, Y., Ding, T., Huo, J., Liu, Y., Li, Y., Li, W., Gao, Y., Luo, J.: 3d gaussian splatting: Survey, technologies, challenges, and opportunities. IEEE Transactions on Circuits and Systems for Video Technology 35(7), 6832–6852 (2025)

    2025

  2. [2]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Zip-nerf: Anti-aliased grid-based neural radiance fields. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19697–19705 (2023)

    2023

  3. [3]

    Biomedical Physics & Engineering Express 2(5), 055010 (2016)

    Biguri, A., Dosanjh, M., Hancock, S., Soleimani, M.: Tigre: a matlab-gpu toolbox for cbct image reconstruction. Biomedical Physics & Engineering Express 2(5), 055010 (2016)

    2016

  4. [4]

    In: ECCV (2024)

    Cai, Y., Liang, Y., Wang, J., Wang, A., Zhang, Y., Yang, X., Zhou, Z., Yuille, A.: Radiative gaussian splatting for efficient x-ray novel view synthesis. In: ECCV (2024)

    2024

  5. [5]

    In: CVPR (2024)

    Cai, Y., Wang, J., Yuille, A., Zhou, Z., Wang, A.: Structure-aware sparse-view x-ray 3d reconstruction. In: CVPR (2024)

    2024

  6. [6]

    In: European conference on computer vision

    Chen, A., Xu, Z., Geiger, A., Yu, J., Su, H.: Tensorf: Tensorial radiance fields. In: European conference on computer vision. pp. 333–350. Springer (2022)

    2022

  7. [7]

    arXiv preprint arXiv:2403.11247 (2024)

    Deng, T., Chen, Y., Zhang, L., Yang, J., Yuan, S., Liu, J., Wang, D., Wang, H., Chen, W.: Compact 3d gaussian splatting for dense visual slam. arXiv preprint arXiv:2403.11247 (2024)

    Pith/arXiv arXiv 2024

  8. [8]

    arXiv preprint arXiv:2105.14711 (2021) 10 Q

    Deng, Y., Wang, C., Hui, Y., Li, Q., Li, J., Luo, S., Sun, M., Quan, Q., Yang, S., Hao, Y., et al.: Ctspine1k: A large-scale dataset for spinal vertebrae segmentation in computed tomography. arXiv preprint arXiv:2105.14711 (2021) 10 Q. Chen, C. Yu et al

    Pith/arXiv arXiv 2021

  9. [9]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Feng, C., Yu, W., Cheng, X., Tang, Z., Zhang, J., Yuan, L., Tian, Y.: Ae-nerf: Augmenting event-based neural radiance fields for non-ideal conditions and larger scenes. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 2924–2932 (2025)

    2025

  10. [10]

    In: Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition

    Fu, Y., Liu, S., Kulkarni, A., Kautz, J., Efros, A.A., Wang, X.: Colmap-free 3d gaussian splatting. In: Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition. pp. 20796–20805 (2024)

    2024

  11. [11]

    Medical image analysis 88, 102846 (2023)

    Kazerouni, A., Aghdam, E.K., Heidari, M., Azad, R., Fayyaz, M., Hacihaliloglu, I., Merhof, D.: Diffusion models in medical imaging: A comprehensive survey. Medical image analysis 88, 102846 (2023)

    2023

  12. [12]

    ACM Trans

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G., et al.: 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph. 42(4), 139–1 (2023)

    2023

  13. [13]

    Tomography 10(8), 1192–1204 (2024)

    Kumar, S., Awadhiya, B., Ratnakumar, R., Thalengala, A., Areeckal, A.S., Nan- jappa, Y.: A review of 3d modalities used for the diagnosis of scoliosis. Tomography 10(8), 1192–1204 (2024)

    2024

  14. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Liang, Z., Zhang, Q., Feng, Y., Shan, Y., Jia, K.: Gs-ir: 3d gaussian splatting for inverse rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 21644–21653 (June 2024)

    2024

  15. [15]

    Information Fusion 115, 102752 (2025)

    Liu, D., Wang, Z., Chen, P.: Dsem-nerf: Multimodal feature fusion and global–local attention for enhanced 3d scene reconstruction. Information Fusion 115, 102752 (2025)

    2025

  16. [16]

    Commu- nications of the ACM 65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM 65(1), 99–106 (2021)

    2021

  17. [17]

    ACM transactions on graphics (TOG) 41(4), 1–15 (2022)

    Müller, T., Evans, A., Schied, C., Keller, A.: Instant neural graphics primitives with a multiresolution hash encoding. ACM transactions on graphics (TOG) 41(4), 1–15 (2022)

    2022

  18. [18]

    arXiv preprint arXiv:2309.16653 (2023)

    Tang, J., Ren, J., Zhou, H., Liu, Z., Zeng, G.: Dreamgaussian: Generative gaussian splatting for efficient 3d content creation. arXiv preprint arXiv:2309.16653 (2023)

    Pith/arXiv arXiv 2023

  19. [19]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., Wang, X.: 4d gaussian splatting for real-time dynamic scene rendering. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 20310– 20320 (2024)

    2024

  20. [20]

    In: 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

    Yu, C., Huang, N., Su, H., Sun, C., Du, B.: Veganet: Conditional multi-view medi- cal image correction via x-gaussian modeling for efficient perspective alignment. In: 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). pp. 3177–3183. IEEE (2025)

    2025

  21. [21]

    In: ICCV (2025)

    Yu, W., Cai, Y., Zha, R., Fan, Z., Li, C., Yuan, Y.: X2-gaussian: 4d radiative gaus- sian splatting for continuous-time tomographic reconstruction. In: ICCV (2025)

    2025

  22. [22]

    arXiv preprint arXiv:2312.10070 (2023)

    Yugay, V., Li, Y., Gevers, T., Oswald, M.R.: Gaussian-slam: Photo-realistic dense slam with gaussian splatting. arXiv preprint arXiv:2312.10070 (2023)

    Pith/arXiv arXiv 2023

  23. [23]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

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

    1960

  24. [24]

    In: NeurIPS (2024)

    Zha, R., Lin, T.J., Cai, Y., Cao, J., Zhang, Y., Li, H.: R2-gaussian: Rectifying radiative gaussian splatting for tomographic reconstruction. In: NeurIPS (2024)

    2024

  25. [25]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Zha, R., Zhang, Y., Li, H.: Naf: neural attenuation fields for sparse-view cbct reconstruction. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 442–452. Springer (2022) UniSpine-GS 11

    2022

  26. [26]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Zhou, Y., Wu, C., Wang, X., Zheng, Y.: Uxdiff: Synthesis of x-ray image from ultra- sound coronal image of spine with diffusion probabilistic network. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 3–12. Springer (2023)

    2023

This paper was first reviewed by grok-4.5 on July 11, 2026.