Pith. sign in

REVIEW 4 major objections 5 minor 29 references

MM2CT: MR-to-CT translation for multi-modal image fusion with mamba

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims that fusing T1- and T2-weighted MRI with a Mamba-based module produces synthetic CT images that score higher on PSNR and SSIM than single-modality translation, and that both the Mamba fusion and dynamic enhancement modules

desk verdict A plausible Mamba-based T1+T2 MR-to-CT fusion, but the experiments don't yet support the SOTA claim. read the letter →

arxiv 2508.05476 v1 pith:TVVZ4DCG submitted 2025-08-07 eess.IV

Chaohui Gong , Zhiying Wu , Zisheng Huang , Gaofeng Meng , Zhen Lei , Hongbin Liu This is my paper
classification eess.IV
keywords MR-to-CTtranslationmultimodalfusionMambastatespacemodelT1-weightedMRIT2-weighteddiffusionsynthesismedicalimage
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

What does the paper try to establish? That combining T1- and T2-weighted MR images as inputs, rather than using a single MR contrast, yields better synthetic CT images, and that this improvement is carried by a Mamba-based fusion module plus a dynamic enhancement module. On a public pelvis dataset, the proposed MM2CT scores 25.72 dB PSNR and 89.54 SSIM, compared with 25.36 dB and 88.97 for the strongest single-modality baseline and around 22.69/81.47 and 21.57/79.56 for cycle-consistency baselines. If true, this would let radiotherapy planning derive CT-equivalent images from routine T1/T2 scans without radiation exposure, and it would establish a specific architecture choice—state-space fusion with dynamic difference enhancement—as the driver of that gain.

What carries the argument

The argument rides on the Mamba-based fusion module. It alternates channel-swapped T1/T2 features through a cross-modal Mamba block—a selective state space model with a gating mechanism—to project both contrasts into a shared representation. A dynamic local convolution layer and a dynamic difference-aware attention module then amplify subtle differences between modality feature maps. The fused representation is injected as conditioning into a diffusion-based generator, which combines adversarial and cycle-consistency losses. This specific combination is what the paper says enables long-range dependency modeling without quadratic attention cost.

What would settle it

Retrain Syndiff and MM2CT on the same pelvis split with five random seeds each and report mean plus-or-minus standard deviation of PSNR and SSIM. If the 95% confidence intervals overlap, the claimed 0.36 dB PSNR improvement is not established. Additionally, run i2i-mamba, Mambadfuse, and FusionMamba on the identical split; if any of them matches or exceeds 25.72 dB PSNR or 89.54 SSIM, the state-of-the-art claim fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that its multi-modal network is the first to meaningfully exploit T1 and T2 together for unpaired MR-to-CT synthesis. It reports that the full MM2CT model reaches 25.72 dB PSNR and 89.54 SSIM on the test split of a public pelvis dataset, beating the best single-modality baseline (25.36 dB, 88.97) and the two cycle-consistency variants. The ablation in Table 2 shows that replacing the Mamba fusion module with the baseline's architecture drops PSNR back to 25.36, and removing only the dynamic enhancement module drops it to 25.49, which the paper interprets as each component contributing. The authors conclude that retaining both Mamba cross-modal fusion and dy

Load-bearing premise

The 0.36 dB PSNR and 0.57% SSIM improvement over the strongest baseline is treated as a real signal despite being within the range that random seed variation could produce, and the state-of-the-art claim would collapse if that difference is noise.

Editorial extensions

If this is right

  • If the reported gains are real, multi-modal MR-to-CT translation becomes a practical route to CT-free radiotherapy planning, since routine clinical MR protocols already include T1 and T2.
  • The fusion module conditions a diffusion-based generator, so the fused feature representation can in principle be paired with different generation backbones, not only the one used here.
  • The paper's ablation places most of the gain on the Mamba fusion module (0.13 dB over baseline) with the dynamic enhancement adding 0.23 dB, so future designs need both to reproduce the reported numbers.
  • Because the method is trained on unpaired data, it can be applied to existing clinical MR/CT archives without requiring registration, which is the common acquisition scenario.

Reading between the lines

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

  • The paper cites i2i-mamba, Mambadfuse, and FusionMamba as related multimodal methods but never evaluates them; a head-to-head comparison on the same dataset is the most direct test of the state-of-the-art claim and is left as an open extension.
  • The reported margins come from a single run on a 15-subject dataset; repeated runs with different seeds would show whether a 0.36 dB PSNR gap is larger than run-to-run variation, and this check is not present.
  • The ablation table reports only three configurations, so the claim that each component is 'necessary' is supported only by pairwise subtraction, not by a full factorial design; a configuration with the dynamic enhancement but without the Mamba module would clarify the interaction.
  • The module's generality could be tested by swapping the diffusion generator for a GAN-based one or applying the fusion block to PET-MR synthesis; if the PSNR gain persists, the mechanism is structural, not tied to diffusion conditioning.
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

4 major / 5 minor

Summary. The paper proposes MM2CT, a multi-modal MR-to-CT translation framework that combines T1- and T2-weighted MRI inputs using a Mamba-based fusion module plus a dynamic enhancement module, together with an adversarial diffusion-style generation module adapted from Syndiff. On a public pelvis dataset, the authors report PSNR 25.72 dB and SSIM 89.54, outperforming single-modality CycleGAN and Syndiff baselines. They conclude that the Mamba fusion and dynamic enhancement components are necessary for the improvement and claim state-of-the-art performance. The manuscript includes an architecture overview, loss formulation, quantitative comparison, and a two-row ablation study.

Significance. If the claims were fully supported, the paper would make a useful incremental contribution to multi-modal medical image synthesis: it applies a state-space fusion mechanism to MR-to-CT translation, releases code, and evaluates on a public benchmark with re-trained baselines. However, the evidence as presented is not yet sufficient for the stated claims. The comparison omits all multi-modal baselines, the reported improvements over the strongest baseline are small and are reported without uncertainty quantification, and the ablation table does not actually isolate the proposed components. The core architectural idea is plausible and the paper is clearly written at a high level, but the central 'state-of-the-art' and 'components are necessary' claims require stronger experimental support.

major comments (4)
  1. [Section 4.3, Table 1] The comparison set contains only single-modality baselines (T1->CT and T2->CT versions of CycleGAN and Syndiff). No method that consumes both T1 and T2 is evaluated, even though Refs. [2], [16], and [25] describe directly relevant multi-modal Mamba-based synthesis/fusion methods. Consequently, the 0.36 dB PSNR and 0.57% SSIM gains over Syndiff could be due simply to the extra input modality or increased model capacity, rather than to the proposed Mamba fusion or dynamic enhancement modules. The authors should compare against i2i-mamba [2], Mambadfuse [16], FusionMamba [25], and, minimally, a control obtained by feeding concatenated T1+T2 inputs to the Syndiff architecture.
  2. [Section 4.4, Table 2] The ablation study does not isolate the proposed components. The first row ('Mamba module: -, DE: -') reports PSNR 25.36 and SSIM 88.97, which are exactly the T1->CT Syndiff numbers from Table 1, not a version of MM2CT without the Mamba and DE modules. Thus Table 2 compares MM2CT against a single-modality baseline, not against an ablated MM2CT. With only the '+' rows differing by 0.13/0.23 dB PSNR and 0.48/0.09 SSIM, the conclusion that the Mamba fusion and dynamic enhancement modules are 'necessary' is not supported. The text also acknowledges in Section 4.4 that isolating DE is difficult, yet proceeds to claim strong validation; this limitation needs to be addressed with properly designed ablations, e.g., a no-fusion MM2CT with two-channel concatenation and a no-DE version.
  3. [Section 4.1, Section 4.3] The reported improvements over the strongest baseline are small (0.36 dB PSNR, 0.57% SSIM) and are based on a fixed split with a 4-subject test set, with no multiple random seeds, confidence intervals, or paired statistical tests. Without uncertainty quantification, the claimed state-of-the-art performance cannot be distinguished from run-to-run variation. The authors should report mean and standard deviation over at least several seeds, or use paired tests across test subjects/slices, and should qualify the state-of-the-art claim accordingly.
  4. [Section 3.1, Section 3.2] The method description is largely qualitative. The diffusive module is described as a 'forward diffusion process' but no diffusion equations, noise schedule, or sampling procedure are given; Eq. (1) only states the overall loss. Similarly, the cross-modal Mamba block and the dynamic difference-aware attention mechanism (Fig. 3) are described in words, but the exact tensor operations and gating formulation are not specified. Because the paper's novelty rests on these modules, this level of detail is insufficient for reproducibility. Adding formulations or precise pseudocode is necessary.
minor comments (5)
  1. [Section 2.1] The sentence 'CycleGAN uses cycle consistency loss ... [2]' cites Ref. [2], which is i2i-mamba, not CycleGAN. Please correct the citation.
  2. [Section 4.2] Typo: 'implemeted' should be 'implemented'.
  3. [Section 5] Typo in 'furtheßr' (should be 'further') in the final paragraph of the conclusion.
  4. [Table 2] The first row should be labeled explicitly as the single-modality Syndiff baseline, not as an MM2CT variant, to avoid confusion about what is being ablated.
  5. [Section 3.1] The terminology alternates between 'diffusive module' and 'diffusion module'; please use consistent terminology throughout.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the translation performance is measured on held-out test data and no prediction reduces to a fitted input or self-citation chain.

full rationale

The paper is an empirical deep-learning submission. The central result (Table 1) is a quantitative comparison on a fixed public split where the test-set PSNR/SSIM of MM2CT is computed from network outputs versus ground-truth CT; no parameter is fitted to the test metric and then reported as a prediction. The method borrows components (Mamba blocks, channel swapping, cross-modal Mamba) with explicit citations to [16], [25], and [10], and the loss settings are copied from [20], but these are external prior works used as building blocks, not self-citations. The authors' own prior work appears only as related references and is not load-bearing for the SOTA claim. The ablation in Table 2 is imperfect—the first row is Syndiff rather than MM2CT without Mamba/DE—but that is an experimental-design/control concern, not circularity: the reported numbers still come from held-out evaluation, and the paper does not redefine any metric to force its conclusion. Therefore no circular step meets the evidentiary bar of Eq. X = Eq. Y by construction or a fitted parameter renamed as prediction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard deep learning assumptions and the availability of a public dataset. No new theoretical entities are introduced. The main free parameters are borrowed from the prior Syndiff work.

free parameters (2)
  • λl1 = same as Syndiff [20]
    Loss weight for L1 loss, set to the value in [20] rather than derived.
  • λgan = same as Syndiff [20]
    Loss weight for GAN loss, set to the value in [20].
assumptions (4)
  • domain assumption Mamba blocks capture long-range dependencies effectively in medical images.
    Used to justify the choice of Mamba over CNNs/Transformers in Section 2.2 and 3.2.
  • domain assumption T1 and T2 MRI carry complementary information beneficial for CT synthesis.
    The entire fusion strategy depends on this; stated in Section 1 and 3.2.
  • domain assumption Unsupervised training on unpaired images transfers to paired registered evaluation.
    The model is trained on unpaired data but evaluated on registered pairs; this assumption underlies the evaluation in Section 4.1.
  • domain assumption The diffusion forward process acts as a low-pass filter.
    Invoked in Section 3.1 to motivate the diffusive module, but no formal derivation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MM2CT: MR-to-CT translation for multi-modal image fusion with mamba." pith.science (2026). https://pith.science/paper/TVVZ4DCG

@misc{pith2026250805476,
  author       = {Pith},
  title        = {Pith review of: MM2CT: MR-to-CT translation for multi-modal image fusion with mamba},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TVVZ4DCG}},
  note         = {Machine review of arXiv:2508.05476}
}
read the original abstract

Magnetic resonance (MR)-to-computed tomography (CT) translation offers significant advantages, including the elimination of radiation exposure associated with CT scans and the mitigation of imaging artifacts caused by patient motion. The existing approaches are based on single-modality MR-to-CT translation, with limited research exploring multimodal fusion. To address this limitation, we introduce Multi-modal MR to CT (MM2CT) translation method by leveraging multimodal T1- and T2-weighted MRI data, an innovative Mamba-based framework for multi-modal medical image synthesis. Mamba effectively overcomes the limited local receptive field in CNNs and the high computational complexity issues in Transformers. MM2CT leverages this advantage to maintain long-range dependencies modeling capabilities while achieving multi-modal MR feature integration. Additionally, we incorporate a dynamic local convolution module and a dynamic enhancement module to improve MRI-to-CT synthesis. The experiments on a public pelvis dataset demonstrate that MM2CT achieves state-of-the-art performance in terms of Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR). Our code is publicly available at https://github.com/Gots-ch/MM2CT.

Figures

Figures reproduced from arXiv: 2508.05476 by the authors.

Figure 1
Figure 1. The network architecture. It includes three key components: GAN-based gen￾eration module, feature fusion module and diffusive module. The process of translating CT into MR is similar. multi-contrast MR images into corresponding CT images, ensuring high-quality and accurate transformations. The multi-modal feature fusion module aims to integrate information from multiple MRI modalities. 3.1 Translation Module The con… view at source ↗
Figure 2
Figure 2. Mamba block used in Fusion module The convolutional layers provide a direct and effective approach for captur￾ing local semantic details. We employ two convolutional layers, each with a 3×3 kernel size and stride of 1. The internal structure of the Mamba block, as illus￾trated in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Enhance module architecture and Fusion module 3.3 Objective Loss In this study, our model is trained by incorporating both traditional adversarial loss and cycle consistency loss, with the discriminator network implemented us￾ing a PatchGAN-like architecture [11]. These loss functions jointly constitute the model’s overall objective function, ensuring the quality of generated images and maintaining consistent featur… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The MR-to-CT translation results of different models are shown along with the source image (Source) and the true target (GT). The result demonstrated on the pelvic dataset for mutl-modal MRI-to-CT translation. From left to right, input T1, input T2, synthetic image CT …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 23 canonical work pages

  1. [2]

    arXiv preprint arXiv:2405.14022 (2024)

    Atli, O.F., Kabas, B., Arslan, F., Demirtas, A.C., Yurt, M., Dalmaz, O., Çukur, T.: I2i-mamba: Multi-modal medical image synthesis via selective state space mod- eling. arXiv preprint arXiv:2405.14022 (2024)

    arXiv 2024

  2. [16]

    arXiv preprint arXiv:2404.08406 (2024)

    Li, Z., Pan, H., Zhang, K., Wang, Y., Yu, F.: Mambadfuse: A mamba-based dual- phase model for multi-modality image fusion. arXiv preprint arXiv:2404.08406 (2024)

    arXiv 2024

  3. [25]

    Visual Intelligence2(1), 37 (2024)

    Xie, X., Cui, Y., Tan, T., Zheng, X., Yu, Z.: Fusionmamba: Dynamic feature en- hancement for multimodal image fusion with mamba. Visual Intelligence2(1), 37 (2024)

    work page 2024

  4. [1]

    Computerized Medical Imaging and Graphics79, 101684 (2020)

    Armanious,K.,Jiang,C.,Fischer,M.,Küstner,T.,Hepp,T.,Nikolaou,K.,Gatidis, S., Yang, B.: MedGAN: Medical image translation using GANs. Computerized Medical Imaging and Graphics79, 101684 (2020)

    work page 2020

  5. [3]

    Nature Medicine29(12), 3111–3119 (2023)

    BoschdeBaseaGomez,M.,Thierry-Chef,I.,Harbron,R.,Hauptmann,M.,Byrnes, G., Bernier, M.O., Le Cornet, L., Dabin, J., Ferro, G., Istad, T.S., et al.: Risk of hematological malignancies from ct radiation exposure in children, adolescents and young adults. Nature Medicine29(12), 3111–3119 (2023)

    work page 2023

  6. [4]

    In: Medical Image Computing and Computer Assisted Intervention - MICCAI, 2025

    Chen, L., Wu, Z., Lei, T., Bai, X., Feng, M., Wang, Y., Meng, G., Lei, Z., Liu, H.: F2paseg: Feature fusion for pituitary anatomy segmentation in endoscopic surgery. In: Medical Image Computing and Computer Assisted Intervention - MICCAI, 2025

    work page 2025

  7. [5]

    In: IEEE/RSJ International Conference on Intelligent Robots and Sys- tems, IROS 2024, Abu Dhabi, United Arab Emirates, October 14-18, 2024

    Chen, Z., Zhang, Z., Guo, W., Luo, X., Bai, L., Wu, J., Ren, H., Liu, H.: Asi- seg: Audio-driven surgical instrument segmentation with surgeon intention under- standing. In: IEEE/RSJ International Conference on Intelligent Robots and Sys- tems, IROS 2024, Abu Dhabi, United Arab Emirates, October 14-18, 2024. pp. 13773–13779. IEEE (2024). https://doi.org/1...

    arXiv 2024

  8. [6]

    IEEE Transactions on Medical Imaging41(10), 2598–2614 (2022)

    Dalmaz, O., Yurt, M., Çukur, T.: Resvit: residual vision transformers for multi- modal medical image synthesis. IEEE Transactions on Medical Imaging41(10), 2598–2614 (2022)

    work page 2022

Show all 29 references
  1. [7]

    Medical Image Analysis p

    Dayarathna, S., Islam, K.T., Uribe, S., Yang, G., Hayat, M., Chen, Z.: Deep learn- ing based synthesis of mri, ct and pet: Review and analysis. Medical Image Analysis p. 103046 (2023)

    2023

  2. [8]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Georgescu, M.I., Ionescu, R.T., Miron, A.I., Savencu, O., Ristea, N.C., Verga, N., Khan, F.S.: Multimodal multi-head convolutional attention with various kernel sizes for medical image super-resolution. In: Proceedings of the IEEE/CVF winter conference on applications of compu...

    2023

  3. [9]

    In: 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI)

    Gong, C., Huang, Z., Wu, Z., Zhao, M., Xu, F., Bai, X., Feng, M., Granados, A., Meng, G., Lei, Z., Liu, H.: Mr-to-ct translation using frequency-separated diffu- sion models. In: 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI)

    2025

  4. [10]

    In: BMVC

    Huang, P.K., Ni, H.Y., Ni, Y., Hsu, C.T.: Learnable descriptive convolutional net- work for face anti-spoofing. In: BMVC. vol. 2, p. 7 (2022) 10 C. Gong et al

    2022

  5. [11]

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

    Isola, P., Zhu, J.Y., Zhou, T., Efros, A.A.: Image-to-image translation with condi- tional adversarial networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 1125–1134 (2017)

    2017

  6. [12]

    In: European conference on computer vision

    Johnson, J., Alahi, A., Fei-Fei, L.: Perceptual losses for real-time style transfer and super-resolution. In: European conference on computer vision. pp. 694–711. Springer (2016)

    2016

  7. [13]

    In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion

    Li, G., Lyu, J., Wang, C., Dou, Q., Qin, J.: Wavtrans: Synergizing wavelet and cross-attention transformer for multi-contrast mri super-resolution. In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion. pp. 463–473. Springer (2022)

    2022

  8. [14]

    In: Medical Image Computing and Com- puter Assisted Intervention–MICCAI 2019: 22nd International Conference, Shen- zhen, China, October 13–17, 2019, Proceedings, Part IV 22

    Li, H., Paetzold, J.C., Sekuboyina, A., Kofler, F., Zhang, J., Kirschke, J.S., Wiestler, B., Menze, B.: Diamondgan: unified multi-modal generative adversarial networks for mri sequences synthesis. In: Medical Image Computing and Com- puter Assisted Intervention–MICCAI 2019: 22...

    2019

  9. [15]

    International Journal of Cognitive Computing in Engineering2, 21–29 (2021)

    Li, Y., Zhao, J., Lv, Z., Li, J.: Medical image fusion method by deep learning. International Journal of Cognitive Computing in Engineering2, 21–29 (2021)

    2021

  10. [17]

    IEEE Transactions on Emerging Topics in Computational Intelligence 7(1), 178–189 (2022)

    Li, Z., Xia, P., Tao, R., Niu, H., Li, B.: A new perspective on stabilizing gans training: Direct adversarial training. IEEE Transactions on Emerging Topics in Computational Intelligence 7(1), 178–189 (2022)

    2022

  11. [18]

    Advances in neural information processing systems 37, 103031–103063 (2025)

    Liu, Y., Tian, Y., Zhao, Y., Yu, H., Xie, L., Wang, Y., Ye, Q., Jiao, J., Liu, Y.: Vmamba: Visual state space model. Advances in neural information processing systems 37, 103031–103063 (2025)

    2025

  12. [19]

    Medical physics45(3), 1295–1300 (2018)

    Nyholm, T., Svensson, S., Andersson, S., Jonsson, J., Sohlin, M., Gustafsson, C., Kjellén, E., Söderström, K., Albertsson, P., Blomqvist, L., et al.: Mr and ct data with multiobserver delineations of organs in the pelvic area—part of the gold atlas project. Medical physics45(3...

    2018

  13. [20]

    IEEE Transactions on Medical Imaging (2023)

    Özbey, M., Dalmaz, O., Dar, S.U., Bedel, H.A., Özturk, Ş., Güngör, A., Çukur, T.: Unsupervised medical image translation with adversarial diffusion models. IEEE Transactions on Medical Imaging (2023)

    2023

  14. [21]

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

    Phan, V.M.H., Liao, Z., Verjans, J.W., To, M.S.: Structure-preserving synthesis: Maskgan for unpaired mr-ct translation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 56–65. Springer (2023)

    2023

  15. [22]

    arXiv preprint arXiv:2408.01129 (2024)

    Qu, H., Ning, L., An, R., Fan, W., Derr, T., Liu, H., Xu, X., Li, Q.: A survey of mamba. arXiv preprint arXiv:2408.01129 (2024)

    2024 arXiv

  16. [23]

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

    Wu, H., Fu, J., Ye, H., Zhong, Y., Zou, X., Zhou, J., Wang, Y.: Towards multi- modality fusion and prototype-based feature refinement for clinically significant prostate cancer classification in transrectal ultrasound. In: International Confer- ence on Medical Image Computing ...

  17. [24]

    Wu, Z., Lau, C.Y., Zhou, Q., Wu, J., Wang, Y., Liu, Q., Lei, Z., Liu, H.: Surgivisor: Transformer-based semi-supervised instrument seg- mentation for endoscopic surgery. Biomed. Signal Process. Control. 87(Part B), 105434 (2024). https://doi.org/10.1016/J.BSPC.2023.105434, htt...

    2024

  18. [26]

    Machine Intelligence Research

    Xu, F., Zhao, M., Wu, Z., Liu, H., Meng, G.: Ft-reg: Unsupervised multimodal medical image registration using dynamic feature translation. Machine Intelligence Research

  19. [27]

    In: Linguraru, M.G., Dou, Q., Feragen, A., Giannarou, S., Glocker, B., Lekadir, K., Schn- abel, J.A

    Xu, H., Wu, J., Cao, G., Chen, Z., Lei, Z., Liu, H.: Transforming surgical in- terventions with embodied intelligence for ultrasound robotics. In: Linguraru, M.G., Dou, Q., Feragen, A., Giannarou, S., Glocker, B., Lekadir, K., Schn- abel, J.A. (eds.) Medical Image Computing an...

    2024

  20. [28]

    IEEE Transactions on Medical Imaging 43(1), 76–95 (2023)

    Zhang, J., Zhang, S., Shen, X., Lukasiewicz, T., Xu, Z.: Multi-condos: Multimodal contrastivedomainsharinggenerativeadversarialnetworksforself-supervisedmed- ical image segmentation. IEEE Transactions on Medical Imaging 43(1), 76–95 (2023)

    2023

  21. [29]

    In: Proceedings of the IEEE Interna- tional Conference on Computer Vision

    Zhu, J.Y., Park, T., Isola, P., Efros, A.A.: Unpaired image-to-image translation using cycle-consistent adversarial networks. In: Proceedings of the IEEE Interna- tional Conference on Computer Vision. pp. 2223–2232 (2017)

    2017

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.