Pith. sign in

REVIEW 3 major objections 5 minor 30 references

Cascaded 3D Diffusion Models for Whole-body 3D 18-F FDG PET/CT synthesis from Demographics

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Demographics alone can drive a cascaded pair of 3D diffusion models to synthesize whole-body PET/CT volumes whose organ volumes and metabolic uptake track real scans within a few percent.

desk verdict Novel demographics-to-PET/CT cascade, but the abstract's 3-5% claim is contradicted by Table 1's 12-37% heart SUVmax deviations and a misstated error direction. read the letter →

arxiv 2505.22489 v1 pith:44XQU75L submitted 2025-05-28 eess.IV cs.CVcs.GR

Siyeop Yoon , Sifan Song , Pengfei Jin , Matthew Tivnan , Yujin Oh , Sekeun Kim , Dufan Wu , Xiang Li
show 1 more author
Quanzheng Li
This is my paper
classification eess.IVcs.CVcs.GR
keywords PET/CTsynthesisdiffusionmodelssuper-resolutiondemographics-conditionedgenerationstandardizeduptakevalue3Dmedicalimagingdigitaltwindataaugmentation
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

This paper argues that age, sex, height, and weight are sufficient inputs for generating plausible whole-body 3D PET/CT scans. The aim is a scalable alternative to deterministic imaging phantoms, producing synthetic but realistic patient-like volumes for data augmentation, digital twin research, and virtual trials. The method uses a two-stage cascade: first a low-resolution anatomical and metabolic blueprint is generated from demographics alone, then a super-resolution diffusion model refines it. Evaluated against real 18F-FDG PET/CT scans, synthetic volumes reproduced organ volumes and standardized uptake values closely, with most metabolic deviations within 3–5% of real measurements across demographic subgroups.

What carries the argument

The load-bearing object is the two-stage cascade. Stage one is a score-based 3D diffusion model, conditioned on continuous demographic variables and binary sex, that solves a reverse-time stochastic differential equation to generate a low-resolution PET/CT volume ($56\times56\times96$ voxels), establishing global anatomy and approximate metabolism. Stage two is a super-resolution residual diffusion model trained patch-wise: it predicts the residual $R = I_{HR} - I_{LU}$ between a true high-resolution volume and the linearly upsampled low-resolution volume, conditioned on the upsampled volume, patch location, noise level, and demographics, then reconstructs $I_{SR} = I_{LU} + R$. The residual formulation lets the refinement stage focus on textures and edges while the low-resolution input enforces anatomical consistency.

What would settle it

Measure organ SUV mean, SUV max, and volume on synthetic cohorts generated from held-out demographic extremes, such as very high or low BMI, and compare with real subjects of the same demographics; if deviations exceed the reported 3–5% for those subgroups, demographic conditioning alone is insufficient for population coverage. A complementary check is to feed real downsampled volumes versus generated low-resolution volumes into the super-resolution stage and compare final-image error, isolating whether the first stage's output statistics break the refinement assumption.

Watch

Extended reading notes

Core claim

The central claim is that whole-body PET/CT volumes can be synthesized from demographic variables alone with enough anatomical and metabolic fidelity to stand in for real patient data at the cohort level. Rather than conditioning on an existing image, the framework conditions a score-based diffusion model directly on demographic attributes to produce a low-resolution global volume, then applies a super-resolution residual diffusion model to recover fine detail. The paper reports that liver, kidney, and heart volumes and SUV mean and SUV max distributions in synthetic cohorts track the real distributions, with most subgroup-level uptake deviations in the 3–5% range and most organ-volume differences not statistically significant. The intended consequence is a scalable, population-informed alternative to conventional phantoms for data augmentation and virtual clinical trials.

Load-bearing premise

The super-resolution stage is trained on real high-resolution volumes that were deliberately downsampled, but at inference it refines low-resolution volumes produced by the first-stage generative model; if those generated volumes differ systematically from real downsampled volumes, errors can compound and the reported organ-level fidelity may not hold.

Editorial extensions

If this is right

  • Synthetic cohorts can be generated for arbitrary demographic distributions, enabling virtual trials that test imaging systems or AI models without collecting new patient data.
  • Because most organ uptake deviations stay within 3–5%, the synthetic volumes are usable for population-level PET/CT analysis, including SUV-based studies.
  • The two-stage design separates global structure from local detail, so the super-resolution component can be retrained or swapped without regenerating the anatomical blueprint.
  • Stochastic diffusion sampling preserves more natural variability than deterministic flow-matching, making it preferable when population diversity matters.

Reading between the lines

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

  • An implicit extension is that the cascade can sample demographic combinations that are rare or absent in the training data; the paper does not test extrapolation, so those samples would need separate validation before use.
  • The organ-level validation focuses on normal uptake in liver, heart, and kidney; pathological findings such as tumors or focal lesions are not evaluated, so lesion-level fidelity remains an open question.
  • The same global-to-local residual design could transfer to other modality pairs or imaging tasks where a cheap global prior plus residual refinement replaces full-resolution generation.
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

3 major / 5 minor

Summary. The manuscript proposes a two-stage cascaded 3D diffusion framework for synthesizing whole-body 18F-FDG PET/CT volumes conditioned only on demographic variables (age, sex, height, weight). The first stage generates a low-resolution PET/CT volume with a score-based diffusion or flow-matching model, and the second stage applies a patch-wise residual super-resolution diffusion model to refine spatial detail. The framework is trained on 565 AutoPET subjects and evaluated on 200 held-out subjects using organ volumes (liver, heart, kidneys) and SUVmean/SUVmax, with subgroup analyses by sex. The central claim is that demographic-only synthesis can produce anatomically and metabolically plausible PET/CT volumes suitable for virtual trials and data augmentation.

Significance. If the quantitative claims are supportable, the paper would be a useful step toward demographic-driven whole-body PET/CT synthesis, with potential applications in data augmentation, digital twins, and virtual clinical trials. The strengths include a held-out evaluation on AutoPET, a comparison of diffusion and flow-matching backbones under the same cascade, and organ-level clinical metrics rather than only qualitative examples. However, the headline claim in the abstract is currently overstated: Table 1 contains many deviations far outside the stated 3-5% range, and the text misreports the direction of the heart SUVmax error. The core idea is not invalid, but the paper needs substantial correction and additional analysis of the cascade's distribution shift before its significance can be assessed.

major comments (3)
  1. [Abstract and Table 1] The abstract claims that "most deviations in metabolic uptake values remained within 3-5% of the ground truth in sub-group analysis." Table 1 does not support this claim. For example, diffusion-model heart SUVmax deviations are +36.8% (male) and +12.2% (female), and flow-model liver SUVmax is -27.7% (male) and -23.1% (female). Counting only SUV comparisons across both methods, fewer than half of the 24 values are within 5% in absolute value. In addition, the text in Section 3 states that "a significant underestimation of heart SUVmax was observed in the male group," whereas Table 1 shows the synthetic male heart SUVmax is higher than AutoPET (13.96 vs 10.20 for diffusion; +36.8%). The abstract and results text must be corrected to match the table, or the claim should be restricted to specific metrics and subgroups for which it is accurate.
  2. [Section 3 and Table 1] The statistical basis for the concordance claim is not reported. The text says sub-groups "did not show statistically significant differences (p>0.05) in liver and kidney volumes," and Table 1 marks p<0.05, but no test name, normality assumption, or multiple-comparison correction is given. With 36 organ-metric comparisons (3 organs x 3 metrics x 2 sexes x 2 models), uncorrected p-values are likely to produce false positives. The authors should specify the test (e.g., two-sample t-test or Mann-Whitney), state how normality was assessed, and describe how multiple comparisons were handled.
  3. [Section 2.1] The super-resolution model is trained with low-resolution priors I_LU obtained by interpolating real high-resolution volumes, but at inference I_LU is produced by the first-stage generative model from demographics. The paper does not analyze whether generated low-resolution volumes have the same statistics as downsampled real volumes, nor does it measure how errors from the global stage propagate through the residual refinement. Because the "high-fidelity" claim depends on the second stage behaving as trained, this distribution shift should be examined, for example by comparing stage-1 outputs to real downsampled volumes or by ablating the cascade with real low-resolution inputs.
minor comments (5)
  1. [Section 2.2] There are typos in the implementation details: "The raining was performed" should be "The training was performed," and the phrase "16,252K images" is ambiguous (likely 16,252 or 16.252 million) and should be clarified.
  2. [Table 1] Table 1 has formatting issues: the female heart volume row shows "0.55 0.09" without a plus-minus symbol, and the female heart SUVmax flow value is listed as "25.%" instead of "25.0%."
  3. [Section 2.1] The term "EDM2 Solver [10,11]" is used without explanation; please define the solver or cite the specific algorithm step from the referenced works.
  4. [Section 2.2] The statement that "failures or poor-quality segmentations were not observed" should be supported by a quantitative criterion or a representative figure, since segmentation quality directly affects the reported organ metrics.
  5. [Title and Section 2.2] The paper uses "18-F FDG" in the title and abstract but "18F-FDG" in Section 2.2; please standardize the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported SUV and volume deviations are measured outcomes on held-out AutoPET subjects, and the method's self-citations concern technical components rather than the central claim.

full rationale

The paper's derivation chain is not circular. The two-stage model is trained on AutoPET (565 subjects) and evaluated on a held-out 200-subject test cohort from the same dataset; the abstract's '3-5%' uptake deviation is a measured outcome of that comparison, not a parameter fitted to the reported metrics. No equation in Section 2 defines the evaluation quantity in terms of the network's own output; the residual R = IHR - ILU is a supervised target from real high-resolution volumes, and the test-time cascade (generated LR -> interpolated ILU -> refined ISR) does not reduce to the training objective by construction. The self-citations [28,29] support methodological components (patch-wise training, residual diffusion) and are not invoked to justify the central claim of anatomical/metabolic fidelity. There is a real limitation in the train/inference distribution shift of the super-resolution stage, and Table 1 contradicts the 3-5% claim for heart SUVmax (12.2-36.8%), but these are correctness and robustness concerns, not circularity.

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

The paper introduces no fitted scalar parameters; the network weights are learned from the AutoPET training set. The evaluation metrics (organ volumes, SUV values) are computed from segmentations and are not fitted. The main assumptions are standard diffusion-model math, the sufficiency of four demographic variables, the distribution match between real and generated low-resolution priors for the super-resolution stage, and the reliability of TotalSegmentator on synthetic images.

assumptions (4)
  • standard math Denoising score matching approximates the true score function of the data distribution.
    Used to train the conditional score networks in Eq. (2), following the EDM and score-based framework of Song et al. and Karras et al., which are standard in generative modeling.
  • domain assumption Demographic variables (age, sex, height, weight) carry enough information to condition the full-body PET/CT distribution for the AutoPET population.
    The entire synthesis pipeline is conditioned only on these four variables (Section 2.2). The paper provides no external validation that they capture population anatomy and metabolism beyond the AutoPET cohort, and the model must sample the residual variability.
  • domain assumption The super-resolution model trained on interpolated real low-resolution volumes also correctly refines low-resolution volumes synthesized by the global model.
    In training, I_LU is produced by subsampling real HR volumes; at inference it comes from the generated I_LR (Section 2.1). The distribution shift between real and generated low-resolution priors is not measured.
  • domain assumption TotalSegmentator segmentation is accurate enough on both real and synthetic CT to make organ volume and SUV measurements comparable.
    All metrics in Table 1 are derived from TotalSegmentator masks (Section 2.2). The paper asserts no segmentation failures occurred, but provides no supporting evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cascaded 3D Diffusion Models for Whole-body 3D 18-F FDG PET/CT synthesis from Demographics." pith.science (2026). https://pith.science/paper/44XQU75L

@misc{pith2026250522489,
  author       = {Pith},
  title        = {Pith review of: Cascaded 3D Diffusion Models for Whole-body 3D 18-F FDG PET/CT synthesis from Demographics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/44XQU75L}},
  note         = {Machine review of arXiv:2505.22489}
}
read the original abstract

We propose a cascaded 3D diffusion model framework to synthesize high-fidelity 3D PET/CT volumes directly from demographic variables, addressing the growing need for realistic digital twins in oncologic imaging, virtual trials, and AI-driven data augmentation. Unlike deterministic phantoms, which rely on predefined anatomical and metabolic templates, our method employs a two-stage generative process. An initial score-based diffusion model synthesizes low-resolution PET/CT volumes from demographic variables alone, providing global anatomical structures and approximate metabolic activity. This is followed by a super-resolution residual diffusion model that refines spatial resolution. Our framework was trained on 18-F FDG PET/CT scans from the AutoPET dataset and evaluated using organ-wise volume and standardized uptake value (SUV) distributions, comparing synthetic and real data between demographic subgroups. The organ-wise comparison demonstrated strong concordance between synthetic and real images. In particular, most deviations in metabolic uptake values remained within 3-5% of the ground truth in subgroup analysis. These findings highlight the potential of cascaded 3D diffusion models to generate anatomically and metabolically accurate PET/CT images, offering a robust alternative to traditional phantoms and enabling scalable, population-informed synthetic imaging for clinical and research applications.

Figures

Figures reproduced from arXiv: 2505.22489 by the authors.

Figure 1
Figure 1. Overview of the cascaded 3D diffusion framework for demographic-driven PET/CT synthesis. (A) During training, demographics guide the global diffusion model to generate a low-resolution 3D PET/CT, which is then interpolated and re￾fined by a super-resolution diffusion model to produce high-resolution outputs. (B) For evaluation, the demographics from the testing set are matched and input to the same cascaded process.… view at source ↗
Figure 2
Figure 2. Representative examples of 18-F FDG PET/CT generated using cascaded 3D diffusion models. The images show CT, 18-FDG SUV, and 3D renderings for synthetic subjects of the same age (60 years) with different heights (male 175 cm vs. female 165 cm) and BMIs. The results demonstrate plausible anatomical and metabolic differ￾ences, including variations in adipose tissue distribution and PET signal heterogeneity. , where RΩ… view at source ↗
Figure 3
Figure 3. Representative slices from the AutoPET (left) compared with synthetic CT/PET generated by the flow-matching (middle) and the diffusion model (right). 3 Results [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 15 canonical work pages

  1. [1]

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

    Akrout, M., Gyepesi, B., Holló, P., Poór, A., Kincső, B., Solis, S., Cirone, K., Kawahara, J., Slade, D., Abid, L., et al.: Diffusion-based data augmentation for skin disease classification: Impact across original medical datasets to fully synthetic images. In: International Conference on Medical Image Computing and Computer- Assisted Intervention. pp. 99...

    2023

  2. [2]

    Stochastic Processes and their Applications 12(3), 313–326 (1982)

    Anderson, B.D.: Reverse-time diffusion equation models. Stochastic Processes and their Applications 12(3), 313–326 (1982)

    1982

  3. [3]

    Medical image analysis 80, 102479 (2022)

    Chung, H., Ye, J.C.: Score-based diffusion models for accelerated mri. Medical image analysis 80, 102479 (2022)

    work page 2022

  4. [4]

    Communications of the ACM 63(11), 139–144 (2020)

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial networks. Communications of the ACM 63(11), 139–144 (2020)

    2020

  5. [5]

    In: European Conference on Com- puter Vision

    Hamamci, I.E., Er, S., Sekuboyina, A., Simsar, E., Tezcan, A., Simsek, A.G., Esirgun, S.N., Almas, F., Doğan, I., Dasdelen, M.F., et al.: Generatect: Text- conditional generation of 3d chest ct volumes. In: European Conference on Com- puter Vision. pp. 126–143. Springer (2024)

    2024

  6. [6]

    Advances in neural information processing systems33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

    2020

  7. [7]

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

    Hu, Q., Li, H., Zhang, J.: Domain-adaptive 3d medical image synthesis: An efficient unsupervised approach. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 495–504. Springer (2022)

    work page 2022

  8. [8]

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

    Jiang, C., Pan, Y., Liu, M., Ma, L., Zhang, X., Liu, J., Xiong, X., Shen, D.: Pet- diffusion: Unsupervised pet enhancement based on the latent diffusion model. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 3–12. Springer (2023)

    work page 2023

Show all 30 references
  1. [9]

    npj Digital Medicine 7(1), 1–12 (2024)

    Kadry, K., Gupta, S., Nezami, F.R., Edelman, E.R.: Probing the limits and ca- pabilities of diffusion models for the anatomic editing of digital twins. npj Digital Medicine 7(1), 1–12 (2024)

    2024

  2. [10]

    Advances in Neural Information Processing Systems35, 26565–26577 (2022)

    Karras,T.,Aittala,M.,Aila,T.,Laine,S.:Elucidatingthedesignspaceofdiffusion- based generative models. Advances in Neural Information Processing Systems35, 26565–26577 (2022)

    2022

  3. [11]

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

    Karras, T., Aittala, M., Lehtinen, J., Hellsten, J., Aila, T., Laine, S.: Analyzing and improving the training dynamics of diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 24174– 24184 (2024) 10 Siyeop Yoon et al

    2024

  4. [12]

    Artificial intelligence in medicine 109, 101938 (2020)

    Kazeminia, S., Baur, C., Kuijper, A., Van Ginneken, B., Navab, N., Albarqouni, S., Mukhopadhyay, A.: Gans for medical image analysis. Artificial intelligence in medicine 109, 101938 (2020)

    2020

  5. [13]

    Scientific Reports 13(1), 7303 (2023)

    Khader, F., Müller-Franzes, G., Tayebi Arasteh, S., Han, T., Haarburger, C., Schulze-Hagen, M., Schad, P., Engelhardt, S., Baeßler, B., Foersch, S., et al.: De- noising diffusion probabilistic models for 3d medical image generation. Scientific Reports 13(1), 7303 (2023)

    2023

  6. [14]

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

    Kim, J., Park, H.: Adaptive latent diffusion model for 3d medical image to image translation: Multi-modal magnetic resonance imaging study. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 7604–7613 (2024)

    2024

  7. [15]

    Kingma, D.P., Welling, M., et al.: Auto-encoding variational bayes (2013)

    2013

  8. [16]

    arXiv preprint arXiv:2210.02747 (2022)

    Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

    2022 arXiv

  9. [17]

    arXiv preprint arXiv:2412.06264 (2024)

    Lipman, Y., Havasi, M., Holderrieth, P., Shaul, N., Le, M., Karrer, B., Chen, R.T., Lopez-Paz, D., Ben-Hamu, H., Gat, I.: Flow matching guide and code. arXiv preprint arXiv:2412.06264 (2024)

    2024 arXiv

  10. [18]

    Medical Physics51(4), 2538–2548 (2024)

    Pan, S., Abouei, E., Wynne, J., Chang, C.W., Wang, T., Qiu, R.L., Li, Y., Peng, J., Roper, J., Patel, P., et al.: Synthetic ct generation from mri using 3d transformer- based denoising diffusion model. Medical Physics51(4), 2538–2548 (2024)

    2024

  11. [19]

    Radiology297(1), 6–14 (2020)

    Pianykh, O.S., Langs, G., Dewey, M., Enzmann, D.R., Herold, C.J., Schoenberg, S.O., Brink, J.A.: Continuous learning ai in radiology: implementation principles and early applications. Radiology297(1), 6–14 (2020)

    2020

  12. [20]

    arXiv preprint arXiv:2307.15208 (2023)

    Pinaya, W.H., Graham, M.S., Kerfoot, E., Tudosiu, P.D., Dafflon, J., Fernandez, V., Sanchez, P., Wolleb, J., Da Costa, P.F., Patel, A., et al.: Generative ai for medical imaging: extending the monai framework. arXiv preprint arXiv:2307.15208 (2023)

    2023 arXiv

  13. [21]

    In: MICCAI Workshop on Deep Generative Models

    Pinaya, W.H., Tudosiu, P.D., Dafflon, J., Da Costa, P.F., Fernandez, V., Nachev, P., Ourselin, S., Cardoso, M.J.: Brain imaging generation with latent diffusion models. In: MICCAI Workshop on Deep Generative Models. pp. 117–126. Springer (2022)

    2022

  14. [22]

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

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  15. [23]

    In: European Confer- ence on Computer Vision

    Schusterbauer, J., Gui, M., Ma, P., Stracke, N., Baumann, S.A., Hu, V.T., Ommer, B.: Fmboost: Boosting latent diffusion with flow matching. In: European Confer- ence on Computer Vision. pp. 338–355. Springer (2024)

    2024

  16. [24]

    arXiv preprint arXiv:2011.13456 (2020)

    Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score- based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 (2020)

    2020 arXiv

  17. [25]

    Cell Reports Medicine5(4) (2024)

    Wang, T., Yang, X.: Take ct, get pet free: Ai-powered breakthrough in lung cancer diagnosis and prognosis. Cell Reports Medicine5(4) (2024)

    2024

  18. [26]

    arXiv preprint arXiv:2304.12526 (2023)

    Wang, Z., Jiang, Y., Zheng, H., Wang, P., He, P., Wang, Z., Chen, W., Zhou, M.: Patch diffusion: Faster and more data-efficient training of diffusion models. arXiv preprint arXiv:2304.12526 (2023)

    2023 arXiv

  19. [27]

    Radiology: Artificial Intelligence 5(5), e230024 (2023) Title Suppressed Due to Excessive Length 11

    Wasserthal, J., Breit, H.C., Meyer, M.T., Pradella, M., Hinck, D., Sauter, A.W., Heye, T., Boll, D.T., Cyriac, J., Yang, S., et al.: Totalsegmentator: robust segmen- tation of 104 anatomic structures in ct images. Radiology: Artificial Intelligence 5(5), e230024 (2023) Title S...

    2023

  20. [28]

    In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

    Yoon, S., Pratap, J.S., Liu, W.C., Tivnan, M., Ren, H., Bhashyam, A., Li, Q., Chen, N., Li, X.: High-resolution 3d ct synthesis from bidirectional x-ray images using 3d diffusion model. In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI). pp. 1–4. IEEE (2024)

    2024

  21. [29]

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

    Yoon, S., Tivnan, M., Hu, R., Wang, Y., Son, Y.d., Wu, D., Li, X., Kim, K., Li, Q.: Volumetric conditional score-based residual diffusion model for pet/mr denoising. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 754–763. Spring...

    2024

  22. [30]

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

    Zhu, L., Xue, Z., Jin, Z., Liu, X., He, J., Liu, Z., Yu, L.: Make-a-volume: Leveraging latent diffusion models for cross-modality 3d brain mri synthesis. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 592–601. Springer (2023)

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.