Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

DMCIE: Diffusion Model with Concatenation of Inputs and Errors to Improve the Accuracy of the Segmentation of Brain Tumors in MRI Images

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read DMCIE is a two-stage diffusion framework that corrects brain tumor masks with error maps, reporting Dice 93.46 and HD95 5.94 mm on BraTS2020.

desk verdict The method's 'error map' is literally defined from ground truth, so the reported gains are almost certainly label leakage; the paper is cleanly written but its central experiment is invalid as presented. read the letter →

arxiv 2507.00983 v2 pith:HOY43SAE submitted 2025-07-01 eess.IV cs.CV

Sara Yavari , Rahul Nitin Pandya , Jacob Furst This is my paper · ORCID
classification eess.IVcs.CV
keywords braintumorsegmentationdiffusionmodelerrormapmulti-modalMRIBraTS2020DDPM3DU-NetDicescore
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 proposes DMCIE, a two-stage pipeline for binary whole-tumor segmentation in multi-modal brain MRI. A 3D U-Net first produces an initial mask, and a diffusion model then reconstructs the error between that mask and the ground truth, guided by the four MRI sequences concatenated with the error map. The corrected mask is the initial mask plus the reconstructed error. On the BraTS2020 dataset the method reports a Dice score of 93.46 and an HD95 of 5.94 mm, beating four reimplemented diffusion-based baselines on both metrics. The paper's intended contribution is to show that explicit error-guided diffusion, rather than pure generative reconstruction, improves boundary accuracy.

What carries the argument

The load-bearing object is the error map $E = M_I - M_{GT}$, which is concatenated with the four MRI modalities to form a five-channel conditioning input $x_{c,t} = \mathrm{concat}(c, x_t)$ for a denoising diffusion probabilistic model. The diffusion process follows the standard forward and reverse equations, but the conditioning differs: instead of generating a mask from noise, the model reconstructs the error $\hat{E}$, and the final result is obtained by adding it to the initial mask. The loss combines BCE and Dice on the predicted noise, steering the model toward correcting misclassified boundary regions.

What would settle it

On a held-out set, replace the ground-truth-derived error map with one produced by a separate network trained only on MRI and initial masks, and compare Dice and HD95; if the final corrected mask no longer improves over the initial U-Net, the reported gains depend on label information that is unavailable at deployment.

Watch

Extended reading notes

Core claim

DMCIE's central claim is that a diffusion model conditioned by concatenating a segmentation error map with the original MRI channels can correct an initial U-Net segmentation more accurately than diffusion baselines that refine masks or predict noise alone. The error map is defined as $E = M_I - M_{GT}$; the diffusion model denoises a noisy version of $E$ while receiving $x_{c,t} = \mathrm{concat}(c, x_t)$ at each timestep, with training loss $L = \mathrm{BCE}(\epsilon, \epsilon_\theta(x_{c,t}, t)) + \lambda \mathrm{Dice}(\epsilon, \epsilon_\theta(x_{c,t}, t))$. The final mask is $M_I^{\mathrm{corr}} = M_I + \hat{E}$. On BraTS2020 binary whole-tumor segmentation, the authors report Dice 93.46% and HD95 5.94 mm, outperforming their reimplemented CorrDiff, SF-Diff, MedSegDiff, and BerDiff baselines under identical data splits, preprocessing, and loss.

Load-bearing premise

The method's load-bearing premise is that the error map $E = M_I - M_{GT}$ can be formed without the ground truth at test time; the paper defines the error from ground truth and does not describe any estimator or surrogate for producing $E$ during inference.

Editorial extensions

If this is right

  • If the reported numbers hold, error-guided diffusion gives better whole-tumor Dice than four published diffusion segmentation approaches reimplemented on the same BraTS2020 setup.
  • The framework turns segmentation refinement into an error-reconstruction task, so the gains should concentrate at tumor boundaries, consistent with the reported HD95 of 5.94 mm.
  • Because the final mask is $M_I + \hat{E}$, any improvement over the 3D U-Net output is carried by the quality of the reconstructed error map.
  • The concatenation strategy is a simple conditioning mechanism that can be added to existing DDPM backbones without architectural redesign.

Reading between the lines

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

  • The paper leaves open how $E$ is obtained at inference; a deployable variant would need a network that predicts the error map from MRI alone, and until then the reported figures should be read as an upper bound when ground-truth-derived errors are used.
  • The same error-concatenation idea could be tested on the tumor-core and enhancing-tumor subregions, where error maps are sparser and boundary errors differ; the paper only evaluates binary whole-tumor masks.
  • An ablation that removes the error-map channel while keeping the four MRI channels would separate the benefit of explicit error guidance from the benefit of conditioning on the modalities.
  • If the reverse process is run with a predicted error map rather than the true one, the method becomes a learned correction of the U-Net output; comparing those two settings would quantify the leakage-free benefit.
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 manuscript proposes DMCIE, a two-stage framework for binary whole-tumor segmentation of BraTS2020 MRI volumes. A 3D U-Net first produces an initial mask MI; a DDPM is then trained to reconstruct an error map E defined in Eq. (1) as E = MI − MGT, with E concatenated to the four MRI modalities during the diffusion process. The final corrected mask is formed in Eq. (14) as MI_corr = MI + E_hat. The paper reports a Dice score of 93.46% and HD95 of 5.94 mm on BraTS2020, claiming improvement over four reimplemented diffusion baselines (Table 1).

Significance. The high-level idea of using a diffusion model to correct residual segmentation errors under multi-modal MRI conditioning is reasonable, and the paper makes the proposed architecture explicit with equations, a figure, and a comparison table. The direct comparison with four reimplemented diffusion baselines on BraTS2020 is also a useful experimental structure. However, the central contribution cannot be assessed as written: the error map is defined using the ground-truth mask, no test-time estimator for E is described, and the reconstruction formula is sign-inconsistent with Eq. (1). These issues undermine the validity of the reported Dice/HD95 numbers. The absence of code, error bars, and a detailed data split further weakens the empirical claims. In its current form, the paper does not provide a sound basis for its claimed state-of-the-art result.

major comments (3)
  1. [Section 3.2.1, Eq. (1), Figure 1] The error map is defined as E = MI − MGT, where MGT is the ground-truth mask. At inference time, MGT is unknown for a new patient, and the paper never describes an estimator, surrogate, or network that predicts E from the MRI alone. The abstract, the contributions list, and Figure 1 all state that E is computed from the difference between the prediction and the ground truth. Therefore, the method as written cannot be deployed, and any evaluation that uses E derived from MGT constitutes label leakage. The reported Dice score and HD95 in Table 1 are thus unsupported as evidence of segmentation skill.
  2. [Section 3.2.2, Eq. (14) with Eq. (1)] Even if the diffusion model reconstructs E perfectly, Eq. (14) gives MI_corr = MI + E_hat ≈ MI + (MI − MGT) = 2MI − MGT. This is not a corrected mask that approaches MGT; it is the opposite of a correction. If the intended relationship is corrective, the error map should be E = MGT − MI, and Eq. (1), Eq. (13), and Figure 1 must be revised accordingly. As written, the mathematical definition of the final corrected mask is internally inconsistent, independent of the inference-time availability of MGT.
  3. [Section 5, Table 1, Section 4.2] The empirical claims rest on what appears to be a single run, with no standard deviations, confidence intervals, or per-subject variability reported. The train/validation/test split is not described in terms of subject counts or identifiers, and no code or trained models are provided. Given that the method's defining input is derived from ground truth, the reported margins over the baselines (for example, 1.43 Dice points over SF-Diff) cannot be regarded as evidence of improvement without a corrected and reproducible protocol.
minor comments (4)
  1. [Section 5, Figure ordering] The text states 'As shown in Figure 2' in a passage that appears after the caption for Figure 3, and the figures are not referenced in numerical order; the figure numbering and in-text references should be corrected.
  2. [Section 3.2.2, Eq. (11)] The notation switches between x1 and xt for the noisy error map, and Eq. (11) uses x1 while the surrounding text discusses arbitrary timesteps; using xt consistently would clarify the concatenation step.
  3. [Section 4.1] The modality is called both 'FLAIR' and 'T2-FLAIR' in different places; the paper should specify the exact preprocessing and channel order used for the four sequences.
  4. [References] Reference [12] and reference [44] both describe 'CorrDiff' but with different author lists; the intended citation should be verified to avoid duplicate or conflicting entries.

Circularity Check

2 steps flagged · score 9.0 of 10

DMCIE's corrected mask is MI + E_hat with E defined as MI - MGT; the reported Dice/HD95 gains reduce to reconstructing a ground-truth-derived input channel.

  1. self definitional [Section 3.2.1, Eq. (1) and x0 definition]
    "The error map E is defined as the pixel-wise subtraction of the initial segmentation mask MI and the ground truth segmentation maskMGT: E = MI − MGT (1) ... We denote the error map E as x0, which serves as the initial input for the diffusion process."

    The diffusion model's input x0 is defined from the ground-truth mask MGT. No test-time estimator or surrogate for E without MGT is described anywhere in the paper, so the only consistent reading is that label-derived information is fed into the model during evaluation. With Eq. 14, if the diffusion model reconstructs its input error, then MI_corr = MI + E_hat ≈ MI + (MI - MGT) = MGT by construction. The reported segmentation accuracy therefore measures auto-encoding fidelity of a ground-truth-derived channel, not independent prediction.

  2. fitted input called prediction [Section 3.2.2, Eq. (14)]
    "In the proposed DMCIE framework, the reconstructed error map ˆE is added to the initial maskMI to produce the final corrected mask, improving the accuracy of brain tumor segmentation. The corrected mask is computed as: MIcorr = MI + ˆE (14)"

    The final 'prediction' is obtained by adding back a reconstructed error map whose training target is E = MI - MGT. The model is fitted to reproduce the label-derived error, and the same quantity is then presented as the source of the Dice/HD95 improvement. This is not a prediction of segmentation from MRI alone; it is a reconstruction of a fitted input being renamed as a corrected mask. A fair comparison against baselines would require an inference-time error estimate that the paper does not define.

full rationale

The central claim collapses into a definitional construction. Eq. 1 defines E = MI - MGT, and Section 3.2.1 explicitly designates E as x0, the diffusion input. Section 3.2.2 then forms the final corrected mask as MI + E_hat (Eq. 14). If the diffusion model reconstructs its input error map, then MI_corr = MI + (MI - MGT) = MGT. The reported Dice 93.46 and HD95 5.94 are therefore not independent segmentation results; they are dominated by the model's ability to reproduce a ground-truth-derived input channel. No test-time procedure for obtaining E without MGT is given: the implementation section only states that training inputs include the error map, and the figures show the error map computed directly from the ground truth mask. Consequently, the improvement over the reimplemented baselines is forced by construction, and the comparison cannot support the paper's claims. The only other self-citation, ReCoSeg [37], appears in related-work context and is not load-bearing, so it does not independently contribute to the circularity. A score of 9 reflects that the central result reduces by definition; a small residual non-circular component could exist only if the error reconstruction were imperfect, but the paper provides no evaluation isolating that component.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a standard DDPM backbone, several undisclosed hyperparameters, and one unstated and likely invalid assumption: that a ground-truth-derived error map can be obtained at test time. No new physical or mathematical entities are introduced; DMCIE is an architecture, not a new object.

free parameters (6)
  • loss balance lambda
    Balances Dice loss against BCE in Eq. 13; the paper says it is tuned on validation performance but reports no value, so the result depends on an undisclosed choice.
  • diffusion timesteps T = 1000
    Standard DDPM choice for the noise schedule, but no ablation shows that 1000 steps is optimal or that fewer steps give the same result.
  • noise schedule endpoints beta_1 and beta_T
    The paper mentions a linear noise schedule over 1000 timesteps but does not report the variance endpoints, which affect the DDPM behavior.
  • U-Net learning rate = 7e-4
    Hyperparameter for the initial mask generator; the initial mask directly determines the error map and therefore the whole pipeline.
  • diffusion learning rate and weight decay = 3e-4, 1e-5
    Training hyperparameters for the DDPM, reported with no sensitivity analysis.
  • preprocessing slice cropping and resizing = keep 78 slices, resize to 120x120
    Discarding the uppermost 26 and lowermost 80 axial slices and resizing changes the tumor distribution seen by the model; baselines were re-run with the same preprocessing, but robustness to this choice is not shown.
assumptions (4)
  • standard math DDPM forward and reverse equations from Ho et al. (2020) are valid as stated.
    The paper reproduces the standard DDPM derivation of Eq. 2 through Eq. 7 and Eq. 10; this is background math, not a new contribution.
  • domain assumption A 3D U-Net trained with hybrid Dice and BCE loss produces an initial mask that a diffusion refiner can improve.
    No ablation compares DMCIE with the raw 3D U-Net output, so the paper assumes the two-stage correction adds value rather than demonstrating it.
  • ad hoc to paper The error map E is available at inference time for new patients.
    Eq. 1 defines E using MGT, but no inference-time estimator is described. This unstated premise is required for Eq. 14 to be computable; if false, the method collapses, and if E is taken from ground truth, the evaluation is circular.
  • domain assumption BraTS2020 ground-truth annotations and the binary whole-tumor merge are a valid target for the clinical claim.
    The paper merges all tumor subregions into one foreground class and evaluates only whole-tumor segmentation, so the clinical claim is limited to tumor localization, not subtype delineation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DMCIE: Diffusion Model with Concatenation of Inputs and Errors to Improve the Accuracy of the Segmentation of Brain Tumors in MRI Images." pith.science (2026). https://pith.science/paper/HOY43SAE

@misc{pith2026250700983,
  author       = {Pith},
  title        = {Pith review of: DMCIE: Diffusion Model with Concatenation of Inputs and Errors to Improve the Accuracy of the Segmentation of Brain Tumors in MRI Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HOY43SAE}},
  note         = {Machine review of arXiv:2507.00983}
}
read the original abstract

Accurate segmentation of brain tumors in MRI scans is essential for reliable clinical diagnosis and effective treatment planning. Recently, diffusion models have demonstrated remarkable effectiveness in image generation and segmentation tasks. This paper introduces a novel approach to corrective segmentation based on diffusion models. We propose DMCIE (Diffusion Model with Concatenation of Inputs and Errors), a novel framework for accurate brain tumor segmentation in multi-modal MRI scans. We employ a 3D U-Net to generate an initial segmentation mask, from which an error map is generated by identifying the differences between the prediction and the ground truth. The error map, concatenated with the original MRI images, are used to guide a diffusion model. Using multimodal MRI inputs (T1, T1ce, T2, FLAIR), DMCIE effectively enhances segmentation accuracy by focusing on misclassified regions, guided by the original inputs. Evaluated on the BraTS2020 dataset, DMCIE outperforms several state-of-the-art diffusion-based segmentation methods, achieving a Dice Score of 93.46 and an HD95 of 5.94 mm. These results highlight the effectiveness of error-guided diffusion in producing precise and reliable brain tumor segmentations.

Figures

Figures reproduced from arXiv: 2507.00983 by the authors.

Figure 1
Figure 1. 2 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 1
Figure 1. Overview of our proposed DMCIE framework for brain tumor [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Illustration of the DMCIE segmentation pipeline on BraTS2020 samples. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (1 more)
Figure 2
Figure 2. Figure 2: Visual illustration of the DMCIE segmentation pipeline using data [PITH_FULL_IMAGE:figures/full_fig_p013_2.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. ReCoSeg++:Extended Residual-Guided Cross-Modal Diffusion for Brain Tumor Segmentation

    eess.IV 2025-08 reject novelty 4.0 of 10

    ReCoSeg++ extends ReCoSeg to BraTS 2021, feeding diffusion-derived T1ce residual maps to a 2D U-Net and reporting 93.02 Dice and 86.7 IoU for whole-tumor segmentation.

Reference graph

Works this paper leans on

47 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    Wang et al

    J. Wang et al. Information bottleneck-based interpretable multitask network for breast cancer classification and segmentation.Medical Image Analysis, 83:102687, 2022

    work page 2022

  2. [2]

    A. S. Panayides et al. Ai in medical imaging informatics: Current challenges and future directions.IEEE Journal of Biomedical and Health Informatics, 24(7):1837–1857, Jul. 2020

    work page 2020

  3. [3]

    Isensee et al

    F. Isensee et al. nnu-net: A self-configuring method for deep learning-based biomedical image segmentation.Nature Methods, 18:203–211, 2020

    work page 2020

  4. [4]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020

    work page 2020

  5. [5]

    Ramesh, P

    A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen. Hierarchical text-conditional image generation with clip latents.arXiv preprint, 2022

    work page 2022

  6. [6]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High- resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684–10695, 2022

    work page 2022

  7. [7]

    R. Zhao, Z. Shi, and Z. Zou. High-resolution remote sensing image captioning based on structured attention. IEEE Transactions on Geoscience and Remote Sensing, 60:1–14, 2021

    work page 2021

  8. [8]

    Goodfellow et al

    I. Goodfellow et al. Generative adversarial networks.Communications of the ACM, 63(11):139–144, 2020

    work page 2020

Show all 47 references
  1. [9]

    Kawar et al

    B. Kawar et al. Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6007–6017, 2023

    2023

  2. [10]

    M. Sun, W. Huang, and Y. Zheng. Instance-aware diffusion model for gland segmentation in colon histology images. InProceedings of the Inter- national Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 662–672, 2023

    2023

  3. [11]

    Song et al

    Y. Song et al. Diffusion models in medical imaging: A comprehensive review. Medical Image Analysis, 87:102802, 2023

    2023

  4. [12]

    Corrdiff: Corrective diffusion model for accurate mri brain tumor segmentation.IEEE Journal of Biomedical and Health Informatics, 28(3):1587, 2024

    Wenqing Li, Wenhui Huang, and Yuanjie Zheng. Corrdiff: Corrective diffusion model for accurate mri brain tumor segmentation.IEEE Journal of Biomedical and Health Informatics, 28(3):1587, 2024. 14

    2024

  5. [13]

    van den Bent, Thierry Gorlia, Wolfgang Wick, Martin Bendszus, and Klaus H

    Philipp Kickingereder, Fabian Isensee, Irada Tursunova, Jens Petersen, Ulf Neuberger, Gianluca Brugnara, David Bonekamp, Marianne Schell, Tobias Kessler, Martha Foltyn, Inga Harting, Felix Sahm, Marcel Prager, Martha Nowosielski, Antje Wick, Marco Nolden, Alexander Radbruch, J...

    2019

  6. [14]

    Detection and localization of early-stage multiple brain tumors using a hybrid technique of patch-based processing, k-means clustering and object counting

    Mohamed Nasor and Walid Obaid. Detection and localization of early-stage multiple brain tumors using a hybrid technique of patch-based processing, k-means clustering and object counting. 2020

    2020

  7. [15]

    Brennan, Holly J

    Paul M. Brennan, Holly J. Butler, Loren Christie, Mark G. Hegarty, Michael D. Jenkinson, Catriona Keerie, John Norrie, Rachel O’Brien, David S. Palmer, Benjamin R. Smith, and Matthew J. Baker. Early diagnosis of brain tumours using a novel spectroscopic liquid biopsy. 2021

    2021

  8. [16]

    Automated multi-class mri brain tumor classification and segmentation using deformable attention and saliency mapping.Scientific Reports, 15(8114), 2025

    Erfan Zarenia, Amirhossein Akhlaghi Far, and Khosro Rezaee. Automated multi-class mri brain tumor classification and segmentation using deformable attention and saliency mapping.Scientific Reports, 15(8114), 2025

    2025

  9. [17]

    Brain tumor segmentation from mri images using handcrafted convolutional neural network

    Faizan Ullah, Muhammad Nadeem, Mohammad Abrar, Muna Al-Razgan, Taha Alfakih, Farhan Amin, and Abdu Salam. Brain tumor segmentation from mri images using handcrafted convolutional neural network. 2023

    2023

  10. [18]

    Chandramma, T

    Asma Alshuhail, Arastu Thakur, R. Chandramma, T. R. Mahesh, Ahlam Almusharraf, V. Vinoth Kumar, and Surbhi Bhatia Khan. Refining neural network algorithms for accurate brain tumor classification in mri imagery. BMC Medical Imaging, 24:118, 2024

    2024

  11. [19]

    Patterson, and Huixiao Hong

    Md Kamrul Hasan Khan, Wenjing Guo, Jie Liu, Fan Dong, Zoe Li, Tucker A. Patterson, and Huixiao Hong. Machine learning and deep learning for brain tumor mri image segmentation.Experimental Biology and Medicine (Maywood), 248(21):1974–1992, 2023

    1974

  12. [20]

    W. Wang, C. Chen, M. Ding, J. Li, H. Yu, and S. Zha. Transbts: Multimodal brain tumor segmentation using transformer. InProceedings of the Inter- national Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 109–119, 2021

    2021

  13. [21]

    Liu et al

    Z. Liu et al. Canet: Context aware network for brain glioma segmentation. IEEE Transactions on Medical Imaging, 40(7):1763–1777, July 2021

    2021

  14. [22]

    Baid et al

    U. Baid et al. A novel approach for fully automatic intra-tumor segmen- tation with 3d u-net architecture for gliomas.Frontiers in Computational Neuroscience, 14, 2020. 15

    2020

  15. [23]

    Tutorial on variational autoencoders.arXiv preprint, 2016

    Carl Doersch. Tutorial on variational autoencoders.arXiv preprint, 2016

    2016

  16. [24]

    Huang, J

    W. Huang, J. J. Gu, P. Duan, S. Hou, and Y. Zheng. Exploiting probabilistic siamese visual tracking with a conditional variational autoencoder. InPro- ceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 14213–14219, 2021

    2021

  17. [25]

    3d mri brain tumor segmentation using autoencoder regularization

    Andriy Myronenko. 3d mri brain tumor segmentation using autoencoder regularization. In 4th International Workshop, BrainLes 2018, Held in Conjunction with MICCAI 2018, Lecture Notes in Computer Science, pages 311–320, 2019

    2018

  18. [26]

    A two-stage cascade model with variational autoencoders and attention gates for mri brain tumor segmentation.Brain- lesion, pages 435–447, 2020

    Chenggang Lyu and Hai Shu. A two-stage cascade model with variational autoencoders and attention gates for mri brain tumor segmentation.Brain- lesion, pages 435–447, 2020

    2020

  19. [27]

    Aswani and D

    K. Aswani and D. Menaka. A dual autoencoder and singular value decom- position based feature optimization for the segmentation of brain tumor from mri images.BMC Medical Imaging, 21:82, 2021

    2021

  20. [28]

    Drm-vae: A dual residual multi variational auto-encoder for brain tumor segmentation with missing modalities

    Yian Zhu, Shaoyu Wang, Yun Hu, and Xiao Ma. Drm-vae: A dual residual multi variational auto-encoder for brain tumor segmentation with missing modalities. In 2021 IEEE 4th International Conference on Electronics and Communication Engineering (ICECE), 2021

    2021

  21. [29]

    Generative adversarial networks.arXiv preprint, 2014

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.arXiv preprint, 2014

    2014

  22. [30]

    H. C. Shin, N. A. Tenenholtz, J. K. Rogers, C. G. Schwarz, M. L. Senjem, J. L. Gunter, K. P. Andriole, and M. Michalski. Medical image synthesis for data augmentation and anonymization using generative adversarial networks. In Lecture Notes in Computer Science, volume 11037, p...

    2018

  23. [31]

    Generative adversarial network in medical imaging: A review.Medical Image Analysis, 2019

    Xin Yi, Ekta Walia, and Paul Babyn. Generative adversarial network in medical imaging: A review.Medical Image Analysis, 2019

    2019

  24. [32]

    Vox2vox: 3d-gan for brain tumour segmentation.arXiv preprint, 2020

    Marco Domenico Cirillo, David Abramian, and Anders Eklund. Vox2vox: 3d-gan for brain tumour segmentation.arXiv preprint, 2020

    2020

  25. [33]

    Promptable counterfac- tual diffusion model for unified brain tumor segmentation and generation with mris

    Yiqing Shen, Guannan He, and Mathias Unberath. Promptable counterfac- tual diffusion model for unified brain tumor segmentation and generation with mris. arXiv preprint, 2024

    2024

  26. [34]

    Pohl, and Yu Zhang

    Haokai Zhao, Haowei Lou, Lina Yao, Wei Peng, Ehsan Adeli, Kilian M. Pohl, and Yu Zhang. Diffusion models for computational neuroimaging: A survey. arXiv preprint, 2024. 16

    2024

  27. [35]

    Jonas Wolleb, Robin Sandkühler, Florian Bieder, Pietro Valmaggia, and Pascal C. Cattin. Diffusion models for implicit image segmentation ensem- bles. In International Conference on Medical Imaging with Deep Learning (MIDL). PMLR, 2022

    2022

  28. [36]

    Accelerating diffusion models via pre-segmentation diffusion sampling for medical image segmentation.arXiv preprint, 2022

    Xutao Guo, Yanwu Yang, Chenfei Ye, Shang Lu, Yang Xiang, and Ting Ma. Accelerating diffusion models via pre-segmentation diffusion sampling for medical image segmentation.arXiv preprint, 2022

    2022

  29. [37]

    Recoseg: Residual guided cross-modal diffusion for efficient brain tumor segmentation

    Sara Yavari, Rahul Pandya, and Jacob Furst. Recoseg: Residual guided cross-modal diffusion for efficient brain tumor segmentation. InMedical Imaging with Deep Learning (MIDL), 2025. Accepted for publication

    2025

  30. [38]

    Med- segdiff: Medical image segmentation with diffusion probabilistic model

    Jiachen Wu, Hao Fang, Yuying Zhang, Yubo Yang, and Yufeng Xu. Med- segdiff: Medical image segmentation with diffusion probabilistic model. arXiv preprint, 2022

    2022

  31. [39]

    Medsegdiff-v2: Diffusion based medical image segmentation with transformer

    Junde Wu, Wei Ji, Rao Fu, Huazhu Fu, Min Xu, Yueming Jin, and Yanwu Xu. Medsegdiff-v2: Diffusion based medical image segmentation with transformer. arXiv preprint, 2023

    2023

  32. [40]

    Segdiff: Image segmentation with diffusion probabilistic models.arXiv preprint, 2021

    Tali Amit, Eliya Nachmani, Tom Shaharbany, and Lior Wolf. Segdiff: Image segmentation with diffusion probabilistic models.arXiv preprint, 2021

    2021

  33. [41]

    3d u-net: Learning dense volumetric segmentation from sparse annotation

    Özgün Çiçek, Ahmed Abdulkadir, Soeren S Lienkamp, Thomas Brox, and Olaf Ronneberger. 3d u-net: Learning dense volumetric segmentation from sparse annotation. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 424–432. Springer, 2016

    2016

  34. [42]

    A multi brain tumor region segmentation model based on 3d u-net.Applied Sciences, 13(16):9282, 2023

    Zhenwei Li, Xiaoqin Wu, and Xiaoli Yang. A multi brain tumor region segmentation model based on 3d u-net.Applied Sciences, 13(16):9282, 2023

    2023

  35. [43]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In Marina Meila and Tong Zhang, editors,Proceedings of the 38th International Conference on Machine Learning, volume 139, pages 8162–8171. PMLR, 2021

    2021

  36. [44]

    Babu and K

    R.V. Babu and K. Kruthika. Corrdiff: Corrective diffusion model for accurate mri brain tumor segmentation.arXiv preprint, 2023

    2023

  37. [45]

    Sharma et al

    R. Sharma et al. Sf-diff: Structure-aware diffusion model for brain tumor segmentation. Medical Image Analysis, 89:102873, 2024

    2024

  38. [46]

    Xie et al

    C. Xie et al. Medsegdiff: Medical image segmentation with diffusion proba- bilistic models. arXiv preprint, 2022

    2022

  39. [47]

    Nguyen et al

    H. Nguyen et al. Berdiff: Bernoulli diffusion for binary medical segmen- tation. In Medical Image Computing and Computer-Assisted Intervention (MICCAI). Springer, 2023. 17

    2023

Pith tools

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