Pith. sign in

REVIEW 3 major objections 4 minor 30 references

TumorGen: Boundary-Aware Tumor-Mask Synthesis with Rectified Flow Matching

T0 review · 3 major / 4 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 TumorGen claims that 3D tumor images and their soft boundary masks can be synthesized jointly from a rough bounding box using rectified flow matching, matching or beating two-stage diffusion baselines in realism at roughly one-sixth the…

desk verdict A plausible and efficient joint tumor-image-and-mask synthesis method that deserves peer review, but the healthy-subject generalization claim rests on an unspecified inference-time bounding-box procedure. read the letter →

arxiv 2505.24687 v1 pith:ZA72WJL6 submitted 2025-05-30 eess.IV cs.CV

Shengyuan Liu , Wenting Chen , Boyun Zheng , Wentao Pan , Xiang Li , Yixuan Yuan This is my paper
classification eess.IVcs.CV
keywords tumorsynthesisrectifiedflowmatching3DmedicalimagingPETgenerativemodelsboundary-awaremasksdataaugmentationimage-maskalignment
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 aims to make synthetic 3D tumor data practical for training diagnostic models: instead of first generating a pixel-perfect tumor mask and then an image in a slow two-stage diffusion pipeline, TumorGen generates image and mask at the same time from a single rough bounding box. It replaces strict binary masks with a boundary-aware pseudo-mask that allows soft, gradual tumor edges, and uses rectified flow matching to cut sampling from hundreds of steps to ten. On the PET datasets Hecktor and AutoPET, the authors report lower Fréchet Inception Distance than prior tumor-synthesis methods, and generated image-mask pairs remain segmentable by a pre-trained nnU-Net. The significance, if the results hold, is a far cheaper pathway to synthetic tumors for data-hungry cancer imaging models.

What carries the argument

Three components carry the argument. BA-PMG (Boundary-Aware Pseudo Mask Generation) derives a training-time bounding box by expanding the tight box around the ground-truth tumor with random offsets proportional to image dimensions; this is the object that relaxes strict binary supervision. SC-VFE (Spatial-Constraint Vector Field Estimator) jointly predicts tumor-image latents and mask latents with rectified flow matching, which learns a straight interpolation path between noise and data so only few sampling steps are needed, and applies two spatial losses: an $\ell^2$ constraint inside the bounding box and an SSIM constraint inside the generated tumor mask. VMR (VAE-Guided Mask Refiner) upsamples the coarse latent mask by adding hierarchical features from the 3D-VAE decoder, recovering boundary detail lost in the latent space.

What would settle it

Run a held-out experiment in which tumors are synthesized into healthy PET scans using bounding boxes from three independent radiologists and from a purely automatic detector, then measure FID and the DSC/NSD of the generated image-mask pairs against real tumors; if realism and alignment collapse when the box is not derived from ground-truth mask expansion, the 'only a rough bounding box' claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that tumor synthesis does not need a predefined binary segmentation; a flexible bounding box plus joint flow-matching generation of latents and masks is enough to produce realistic 3D tumors with soft boundaries, and this recipe is substantially faster than the previous diffusion-based two-stage approaches. More specifically, TumorGen reports average FID 52.035 on Hecktor versus 66.821 for DiffTumor and 72.433 versus 84.946 on AutoPET, with 0.218 seconds per 10-step sample, and the generated masks align with images closely enough that a segmentation network achieves DSC 0.694 and NSD 0.741 on Hecktor.

Load-bearing premise

The method presupposes that a rough 3D bounding box around the tumor is available or provided at generation time; the paper only explains how the box is derived when the ground-truth mask is known, leaving the healthy-subject case without a specified box-proposal mechanism.

Editorial extensions

If this is right

  • A single trained model outputs both the tumor image and a high-resolution mask in the same forward pass, so downstream segmentation or detection training no longer needs a separate mask-proposal stage.
  • Tumor synthesis for rare lesion types becomes practical on clinical budgets: 10-step sampling at 0.218 seconds per volume allows dataset expansion on a single GPU in minutes.
  • The healthy-subject generalization shown in Figure 4 implies the model can populate tumor-free PET scans with plausible lesions, a route to balancing class-imbalanced cancer datasets.
  • The boundary-aware pseudo-mask replaces binary supervision, so synthesized tumors exhibit graded transitions between diseased and healthy tissue rather than hard edges.

Reading between the lines

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

  • The authors do not compare DiffTumor at 10 diffusion steps, so part of the speed gap is sampling-schedule choice; a head-to-head at equal step count would isolate the rectified-flow contribution.
  • A testable extension is to couple TumorGen with an automatic lesion-proposal network, which would test whether the bounding box can be fully automatic rather than user-supplied on healthy subjects.
  • The same three-component design could be adapted to CT or MRI, since the VAE latent space and flow-matching objective are modality-agnostic; the BA-PMG box offsets would need re-tuning to organ-specific tumor sizes.
  • Because VMR conditions mask refinement on decoder features, the refined mask inherits image details, suggesting the module could double as an anomaly-localization head for generated images.
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 / 4 minor

Summary. The paper presents TumorGen, a 3D PET tumor synthesis framework with three components: a Boundary-Aware Pseudo Mask Generation module (BA-PMG) that expands ground-truth tumor masks into bounding boxes, a Spatial-Constraint Vector Field Estimator (SC-VFE) that jointly synthesizes tumor latents and masks by rectified flow matching under coarse- and fine-grained spatial constraints, and a VAE-Guided Mask Refiner (VMR) that refines predicted masks with decoder features. Experiments on Hecktor and AutoPET report improved FID, DSC/NSD, and faster sampling relative to existing tumor synthesis baselines, including a 10-step rectified flow run at 0.218 s per sample. The paper also claims generalization to healthy subjects by synthesizing tumors in tumor-free images.

Significance. If the method works as described, it offers a practical advance in medical tumor synthesis: simultaneous image-mask generation with few-step rectified flow could substantially reduce synthesis cost while maintaining image-mask alignment. The use of two public datasets and standard metrics (FID, DSC, NSD) and the explicit inference-time comparison are strengths, as is the direct measurement of sampling speed. However, two central claims are currently undersupported: the inference-time bounding-box mechanism for healthy-subject synthesis is not defined, and the stated soft-boundary motivation is contradicted by the binary mask reconstruction loss. These issues are fixable in revision and do not by themselves invalidate the reported comparative FID scores on tumor-bearing scans, but they must be addressed before the paper can be accepted.

major comments (3)
  1. [Section 2.1, Figure 4] The inference-time bounding box is not specified. BA-PMG computes the tight box B' from the coordinates of tumor voxels in the ground-truth mask m1 and then expands it (Section 2.1). During inference on tumor-bearing scans it is unclear whether the ground-truth mask is still used to obtain B; for healthy subjects in Figure 4 no tumor mask exists, so there is no described mechanism to place the box. This makes the generalizability experiment in Figure 4 non-reproducible and the claim of synthesizing tumors in healthy subjects unsupported. Please specify the protocol (user-drawn box, random placement, detector, or ground-truth-derived box), report sensitivity to box placement, and temper the claims of automatic generalization accordingly.
  2. [Section 2.3, Eq. (6)] The boundary-aware/soft-boundary claim is contradicted by the mask objective. The paper's motivation (Section 1) emphasizes soft, gradual tumor boundaries, but the refined mask M' is trained with Lrec = ||f_theta(hat m1) - M||^2 against the binary ground-truth mask M (defined in the Overview), and the final output is a binary mask. The bounding box in BA-PMG provides surrounding context but does not model gradual transitions in the mask. This is load-bearing because the boundary-aware framing is central to the paper's contribution. Please either replace the binary reconstruction loss with a soft-boundary formulation (e.g., distance-transform encoded labels) and evaluate with boundary-aware metrics, or revise the claims to describe bounding-box-conditioned binary mask synthesis.
  3. [Section 3.1, Section 2] Several implementation details needed for reproducibility are missing. The values of alpha (BA-PMG offset scaling) and lambda_1, lambda_2, lambda_3 (overall objective in Eq. (7)) are not reported, and the number of synthetic samples used to compute FID is not stated. These omissions do not invalidate the reported comparisons, but they prevent replication and make the sensitivity of the method to its free parameters impossible to assess. Please provide the hyperparameter values and the FID evaluation protocol in the revised manuscript.
minor comments (4)
  1. [Table 1, Section 3.2] There is a citation inconsistency for MedFusion: Table 1 labels it [23], while Section 3.2 cites MedFusion as [18]; reference [23] in the bibliography is Singan-Seg, not MedFusion.
  2. [Figure 2 caption] The caption calls BA-PMG 'automatic tumor bounding box generation', but the method computes the bounding box from a ground-truth tumor mask, so it is not automatic without annotation. Please rephrase the caption to match the actual procedure.
  3. [Equation (2)] Equation (2) uses the same symbol epsilon for z_t and m_t, which could imply that the same noise sample is used for the image and mask latents; please clarify whether the noises are shared or independently sampled.
  4. [Equation (5)] The fine-grained constraint uses SSIM on latent-space tensors, but the paper does not specify how SSIM is computed across the latent channels; please state the SSIM variant and parameter settings.

Circularity Check

0 steps flagged · score 2.0 of 10

No circularity in the core derivation; one non-load-bearing self-citation and an unspecified bounding-box protocol for healthy subjects.

full rationale

The derivation chain is self-contained. The rectified flow matching objective (Eq. 3) trains the vector field estimator v_theta on the joint latent x1 = [z1, m1] against standard Gaussian noise; the coarse and fine spatial constraints (Eqs. 4 and 5) and the mask reconstruction loss (Eq. 6) are supervised losses using ground-truth z1 and M, not fitted constants that later masquerade as predictions. The reported FID, DSC, and NSD numbers are computed on external public datasets (Hecktor-2021 and AutoPET) with standard metrics and an independently pre-trained nnU-Net, so the main comparative claims do not reduce to the method's own inputs. The only self-citation, Ref. [17] (Polyp-Gen, by two of the present authors), is used solely to support the general statement that tumors have soft, gradual boundaries; it does not justify any equation, architecture choice, or uniqueness claim and is therefore not load-bearing. One genuine weakness is the healthy-subject experiment (Fig. 4): BA-PMG (Sec. 2.1) computes its bounding box from the ground-truth tumor mask m1, and the paper never specifies how a box is obtained for a healthy image with no tumor mask, despite the caption claiming 'automatic tumor bounding box generation'. This is a reproducibility and generalization gap, not a circular reduction, because the method's output is not defined as its input and the comparative results on tumor-bearing scans do not depend on resolving it.

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

The central claim depends on the hyperparameters in Eq. 7 and on the assumption that a bounding box is available at inference; no new physical or domain entities are introduced.

free parameters (4)
  • lambda_1
    Weight for coarse spatial constraint in Eq. 7; value not reported.
  • lambda_2
    Weight for fine spatial constraint in Eq. 7; value not reported.
  • lambda_3
    Weight for mask reconstruction loss in Eq. 7; value not reported.
  • alpha
    Scaling factor controlling maximum bbox expansion in BA-PMG; value not reported.
assumptions (4)
  • standard math The rectified flow matching objective (Eq. 3) learns the correct vector field with L2 regression on (x1 - epsilon).
    Standard result from the cited flow matching literature (Esser et al., ICML 2024).
  • domain assumption PET image and tumor mask statistics in Hecktor and AutoPET are representative of the target distribution.
    The method is trained and evaluated on these datasets; no external validation on other modalities or centers.
  • domain assumption FID on middle slices is a valid proxy for 3D synthesis quality.
    The paper computes FID only on the middle slice of each plane following prior work [11], which may not capture volumetric consistency.
  • ad hoc to paper A user-supplied bounding box is available at inference to define the tumor region.
    BA-PMG as described derives the box from the ground-truth mask; its test-time counterpart is not specified in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TumorGen: Boundary-Aware Tumor-Mask Synthesis with Rectified Flow Matching." pith.science (2026). https://pith.science/paper/ZA72WJL6

@misc{pith2026250524687,
  author       = {Pith},
  title        = {Pith review of: TumorGen: Boundary-Aware Tumor-Mask Synthesis with Rectified Flow Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZA72WJL6}},
  note         = {Machine review of arXiv:2505.24687}
}
read the original abstract

Tumor data synthesis offers a promising solution to the shortage of annotated medical datasets. However, current approaches either limit tumor diversity by using predefined masks or employ computationally expensive two-stage processes with multiple denoising steps, causing computational inefficiency. Additionally, these methods typically rely on binary masks that fail to capture the gradual transitions characteristic of tumor boundaries. We present TumorGen, a novel Boundary-Aware Tumor-Mask Synthesis with Rectified Flow Matching for efficient 3D tumor synthesis with three key components: a Boundary-Aware Pseudo Mask Generation module that replaces strict binary masks with flexible bounding boxes; a Spatial-Constraint Vector Field Estimator that simultaneously synthesizes tumor latents and masks using rectified flow matching to ensure computational efficiency; and a VAE-guided mask refiner that enhances boundary realism. TumorGen significantly improves computational efficiency by requiring fewer sampling steps while maintaining pathological accuracy through coarse and fine-grained spatial constraints. Experimental results demonstrate TumorGen's superior performance over existing tumor synthesis methods in both efficiency and realism, offering a valuable contribution to AI-driven cancer diagnostics.

Figures

Figures reproduced from arXiv: 2505.24687 by the authors.

Figure 1
Figure 1. (a) Previous two-stage tumor synthesis methods need to specify the pixel [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of TumorGen, a rectified flow-based 3D tumor synthesis framework with a Boundary-Aware Pseudo Mask Generation module (BA-PMG) to automatic tumor bounding box generation, a Spatial-Constraint Vector Field Estimator (SC-VFE) for simultaneous tumor image and mask synthesis, and a VAE-guided mask refiner (VMR) to enhance mask boundary. consists of three key components: a Boundary-Aware Pseudo Mask Generation mo… view at source ↗
Figure 3
Figure 3. Qualitative comparison among different synthesis methods. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Tumor images generated from healthy images. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 25 canonical work pages

  1. [1]

    In: 3D head and neck tumor segmentation in PET/CT challenge, pp

    Andrearczyk, V., Oreiller, V., Boughdad, S., Rest, C.C.L., Elhalawani, H., Jreige, M., Prior, J.O., Vallières, M., Visvikis, D., Hatt, M., et al.: Overview of the heck- tor challenge at miccai 2021: automatic head and neck tumor segmentation and outcome prediction in pet/ct images. In: 3D head and neck tumor segmentation in PET/CT challenge, pp. 1–37. Spr...

    2021

  2. [2]

    ACM Trans

    Avrahami, O., Fried, O., Lischinski, D.: Blended latent diffusion. ACM Trans. Graph.42(4), 1–11 (2023)

    work page 2023

  3. [3]

    Cao,K.,Xia,Y.,Yao,J.,Han,X.,Lambert,L.,Zhang,T.,Tang,W.,Jin,G.,Jiang, H., Fang, X., et al.: Large-scale pancreatic cancer detection via non-contrast ct and deep learning. Nat. Med.29(12), 3033–3043 (2023)

    work page 2023

  4. [4]

    In: CVPR

    Chen, Q., Chen, X., Song, H., Xiong, Z., Yuille, A., Wei, C., Zhou, Z.: Towards generalizable tumor synthesis. In: CVPR. pp. 11147–11158 (2024)

    work page 2024

  5. [5]

    NeurIPS31(2018)

    Chen, R.T., Rubanova, Y., Bettencourt, J., Duvenaud, D.K.: Neural ordinary dif- ferential equations. NeurIPS31(2018)

    work page 2018

  6. [6]

    In: ICML (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: ICML (2024)

    2024

  7. [7]

    Gatidis, S., Hepp, T., Früh, M., La Fougère, C., Nikolaou, K., Pfannenberg, C., Schölkopf, B., Küstner, T., Cyran, C., Rubin, D.: A whole-body fdg-pet/ct dataset with manually annotated tumor lesions. Sci. Data9(1), 601 (2022)

    work page 2022

  8. [8]

    CoRR (2024)

    Guo, P., Zhao, C., Yang, D., Xu, Z., Nath, V., Tang, Y., Simon, B., Belue, M., Harmon, S.A., Turkbey, B., et al.: Maisi: Medical ai for synthetic imaging. CoRR (2024)

    work page 2024

Show all 30 references
  1. [9]

    NeurIPS30 (2017)

    Heusel,M.,Ramsauer,H.,Unterthiner,T.,Nessler,B.,Hochreiter,S.:Ganstrained by a two time-scale update rule converge to a local nash equilibrium. NeurIPS30 (2017)

    2017

  2. [10]

    NeurIPS33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. NeurIPS33, 6840–6851 (2020)

    2020

  3. [11]

    In: MICCAI Workshop

    Hong, S., Marinescu, R., Dalca, A.V., Bonkhoff, A.K., Bretzner, M., Rost, N.S., Golland, P.: 3d-stylegan: A style-based generative adversarial network for gener- ative modeling of three-dimensional medical images. In: MICCAI Workshop. pp. 24–34. Springer (2021)

    2021

  4. [12]

    In: CVPR

    Hu, Q., Chen, Y., Xiao, J., Sun, S., Chen, J., Yuille, A.L., Zhou, Z.: Label-free liver tumor segmentation. In: CVPR. pp. 7422–7432 (2023)

    2023

  5. [13]

    Lancet Digital Health 3(4), e260–e265 (2021) 10 Authors Suppressed Due to Excessive Length

    Ibrahim, H., Liu, X., Zariffa, N., Morris, A.D., Denniston, A.K.: Health data poverty: an assailable barrier to equitable digital health care. Lancet Digital Health 3(4), e260–e265 (2021) 10 Authors Suppressed Due to Excessive Length

    2021

  6. [14]

    Isensee, F., Jaeger, P.F., Kohl, S.A., Petersen, J., Maier-Hein, K.H.: nnu-net: a self-configuring method for deep learning-based biomedical image segmentation. Nat. Methods18(2), 203–211 (2021)

    2021

  7. [15]

    Kim, K., Cho, H., Na, Y., Ye, S.J., Lee, J., Ahn, S.S., Park, J.E., Kim, H.: A tumor inpainting framework for mri using automated masks based on channel-specific conditions across the volume. Biomed. Signal Process. Control105, 107579 (2025)

    2025

  8. [16]

    AAAI (2025)

    Lei, M., Wu, H., Lv, X., Wang, X.: Condseg: A general medical image segmentation framework via contrast-driven feature enhancement. AAAI (2025)

    2025

  9. [17]

    ICRA (2025)

    Liu,S.,Chen,Z.,Yang,Q.,Yu,W.,Dong,D.,Hu,J.,Yuan,Y.:Polyp-gen:Realistic and diverse polyp image generation for endoscopic dataset expansion. ICRA (2025)

    2025

  10. [18]

    Müller-Franzes, G., Niehues, J.M., Khader, F., Arasteh, S.T., Haarburger, C., Kuhl, C., Wang, T., Han, T., Nolte, T., Nebelung, S., et al.: A multimodal compar- ison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthe...

    2023

  11. [19]

    Pan, X., AbdulJabbar, K., Coelho-Lima, J., Grapa, A.I., Zhang, H., Cheung, A.H.K., Baena, J., Karasaki, T., Wilson, C.R., Sereno, M., et al.: The artificial intelligence-based model anorak improves histopathological grading of lung adeno- carcinoma. Nat. Cancer5(2), 347–363 (2024)

    2024

  12. [20]

    In: PILM Workshop

    Peng, L., Zhang, Z., Durak, G., Miller, F.H., Medetalibeyoglu, A., Wallace, M.B., Bagci, U.: Optimizing synthetic data for enhanced pancreatic tumor segmentation. In: PILM Workshop. pp. 35–44. Springer (2024)

    2024

  13. [21]

    In: ICLR (2020)

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. In: ICLR (2020)

    2020

  14. [22]

    Sun, L., Chen, J., Xu, Y., Gong, M., Yu, K., Batmanghelich, K.: Hierarchical amortized gan for 3d high resolution medical image synthesis. IEEE J. Biomed. Health Inform.26(8), 3966–3975 (2022)

    2022

  15. [23]

    PloS one17(5), e0267976 (2022)

    Thambawita, V., Salehi, P., Sheshkal, S.A., Hicks, S.A., Hammer, H.L., Parasa, S., Lange, T.d., Halvorsen, P., Riegler, M.A.: Singan-seg: Synthetic training data generation for medical image segmentation. PloS one17(5), e0267976 (2022)

    2022

  16. [24]

    Wang, J., Wang, K., Yu, Y., Lu, Y., Xiao, W., Sun, Z., Liu, F., Zou, Z., Gao, Y., Yang, L., et al.: Self-improving generative foundation model for synthetic medical image generation and clinical applications. Nat. Med. pp. 1–9 (2024)

    2024

  17. [25]

    Wang, J., Yu, Y., Tan, Y., Wan, H., Zheng, N., He, Z., Mao, L., Ren, W., Chen, K., Lin, Z., et al.: Artificial intelligence enables precision diagnosis of cervical cytology grades and cervical cancer. Nat. Commun.15(1), 4369 (2024)

    2024

  18. [26]

    Nature634(8035), 970–978 (2024)

    Wang, X., Zhao, J., Marostica, E., Yuan, W., Jin, J., Zhang, J., Li, R., Tang, H., Wang, K., Li, Y., et al.: A pathology foundation model for cancer diagnosis and prognosis prediction. Nature634(8035), 970–978 (2024)

    2024

  19. [27]

    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. Radiol. Artif. Intell.5(5) (2023)

    2023

  20. [28]

    ECCV (2024)

    Wu, L., Zhuang, J., Ni, X., Chen, H.: Freetumor: Advance tumor segmentation via large-scale tumor synthesis. ECCV (2024)

    2024

  21. [29]

    Yang, Y., Zhang, H., Gichoya, J.W., Katabi, D., Ghassemi, M.: The limits of fair medical imaging ai in real-world generalization. Nat. Med.30(10), 2838–2848 (2024)

    2024

  22. [30]

    In: International Workshop on Simulation and Synthesis in Medical Imaging

    Zhang, Z., Deng, H., Li, X.: Unsupervised liver tumor segmentation with pseudo anomaly synthesis. In: International Workshop on Simulation and Synthesis in Medical Imaging. pp. 86–96. Springer (2023)

    2023

Pith tools

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