Pith. sign in

REVIEW 5 major objections 5 minor 34 references

Leveraging Semantic Asymmetry for Precise Gross Tumor Volume Segmentation of Nasopharyngeal Carcinoma in Planning CT

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a 3D Siamese segmentation framework exploiting the left-right asymmetry of nasopharyngeal tumors can segment gross tumor volume directly on non-contrast planning CT, improving external Dice by 4.4% over the…

desk verdict Useful external benchmark, but Equation (1) is mis-signed, so the paper's core asymmetry mechanism is currently unverifiable as written. read the letter →

arxiv 2411.18290 v3 pith:ZOCMKDKV submitted 2024-11-27 eess.IV cs.CV

Zi Li , Ying Chen , Zeli Chen , Yanzhou Su , Tai Ma , Tony C. W. Mok , Yan-Jie Zhou , Yunhai Bai
show 7 more authors
Zhinlin Zheng Le Lu Yirui Wang Jia Ge Xianghua Ye Senxiang Yan Dakai Jin
This is my paper · ORCID
classification eess.IVcs.CV
keywords nasopharyngealcarcinomagrosstumorvolumeplanningCTsemanticasymmetrySiamesenetworkcontrastivelearningmarginlossmedicalimagesegmentation
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 tries to establish that nasopharyngeal carcinoma tumors can be segmented directly from non-contrast planning CT, the modality actually used for radiotherapy planning, without relying on MRIs that must be registered to CT. The core idea is that a healthy nasopharynx is bilaterally symmetric, and the tumor is precisely the structure that breaks that symmetry. The authors build a Siamese segmentation network that compares each CT with its mirror image and trains it to pull symmetric (non-tumor) voxel features together and push asymmetric (tumor) voxel features apart. On their in-house test set the method reaches 81.22% Dice, and on the external SegRap2023 dataset 66.80% Dice, outperforming ten CNN, Transformer, and Mamba baselines. If true, this would give radiation oncologists a planning-CT-only delineation tool, avoiding the registration errors of MRI-to-CT transfer.

What carries the argument

The load-bearing mechanism is a Siamese encoder-decoder paired with a voxel-wise margin loss over asymmetric regions. The margin loss (Eq. 1) acts on features after a non-linear projection head: for voxels flagged by the asymmetrical mask $m = s - s \cap s'$, it minimizes the $ℓ^2$ distance between the original and flipped feature maps $E$ and $E'$; for all other voxels it maximizes that distance up to a margin $t$ (set to 20). The asymmetric mask is computed from the ground-truth segmentation by subtracting the intersection of the mask with its flip, isolating the portion of the tumor that breaks bilateral symmetry.

What would settle it

Take a cohort of NPC patients, run SATS and nnUNet on scans that are intentionally rotated by a few degrees around the head axis (simulating pose asymmetry), and measure Dice; if the margin loss truly depends on perfect bilateral symmetry, SATS's Dice should fall disproportionately relative to nnUNet as the rotation angle grows.

Watch

Extended reading notes

Core claim

The central discovery claimed is that tumor-induced semantic asymmetry is a sufficient and generalizable cue for gross tumor volume segmentation in non-contrast planning CT. The method, SATS, assumes the healthy nasopharynx is bilaterally symmetric. It uses a shared-weights Siamese encoder-decoder to process the CT and its left-right flip, and a voxel-wise margin loss (Eq. 1) that decreases the distance between corresponding features in non-tumor regions and increases it in tumor regions, where the asymmetrical mask $m = s - s \cap s'$ locates tumor voxels that do not survive the flip. A non-linear projection head is required; the ablation shows margin loss alone hurts (61.50% vs 63.44% external Dice), and only with the projection head does it give +4.98% Dice. The authors report 81.22% internal and 66.80% external Dice, with at least 2% absolute improvement over all baselines in external testing, and interpret the external gain as evidence of better generalizability of the asymmetry signal.

Load-bearing premise

The method assumes that, once scans are aligned to a manually chosen symmetric atlas, any remaining asymmetry in the nasopharyngeal region is caused by the tumor rather than by normal anatomy, patient pose, or imaging artifacts.

Editorial extensions

If this is right

  • Segmentation can be done directly on non-contrast planning CT, eliminating the need for MRI-to-CT registration during delineation and the errors it introduces.
  • The external test improvement (66.80% vs 64.40% Dice for nnUNet) suggests the symmetry signal transfers across institutions better than texture-only features.
  • The projection head is not optional: removing it turns the margin loss from a help into a harm, so any reuse of the method must keep the non-linear projection.
  • Tumors that are symmetric about the midline (small, central lesions) are the method's known weakness, which the paper's failure-case analysis acknowledges.

Reading between the lines

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

  • The same bilateral-symmetry premise could transfer to other lateralized head-and-neck structures or to organs like the breast and pelvis, where prior symmetry work exists, but the margin loss would need to handle anatomy whose healthy state is not perfectly symmetric.
  • The reliance on a manually selected symmetric atlas and affine alignment is a practical bottleneck; a fully automated alignment could make the method more deployable, and one testable extension is to replace the atlas step with a learned canonical-space registration.
  • The margin loss implicitly assumes that any non-pathological asymmetry (vessels, lymph nodes, patient pose) is either removed by alignment or filtered by the projection head; a stress test with deliberately misaligned or asymmetric patients would reveal how much of the gain is architecture versus this premise.
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

5 major / 5 minor

Summary. The paper proposes SATS, a Siamese encoder-decoder method for segmenting nasopharyngeal carcinoma gross tumor volume (GTV) directly from non-contrast planning CT. The method relies on the premise that the healthy nasopharynx is bilaterally symmetric and that tumors break this symmetry. A voxel-wise margin loss is introduced to minimize feature distance between original and flipped images in symmetric regions and, according to the abstract and introduction, to maximize distance in asymmetric tumor regions. The model is trained on an in-house dataset of 163 patients and evaluated on an internal test split and the external SegRap2023 dataset, with comparisons against ten CNN, Transformer, and Mamba baselines. The paper reports the best DSC and HD95 on both internal and external test sets, with ablation studies on the projection head and margin loss, plus a failure-case analysis.

Significance. The clinical problem is well motivated: direct segmentation in non-contrast planning CT could avoid MRI-to-CT registration errors in radiotherapy workflow. If the proposed method were correctly specified and the reported numbers reproducible, this would be a meaningful contribution: the first NPC GTV segmentation on non-contrast pCT, external validation on a public dataset, and a systematic comparison against ten strong baselines. The paper also honestly discusses failure cases. However, the central equation describing the margin loss inverts the paper's core mechanism, and several quantitative claims in the text do not match the tables. As written, the validity of the central claim is not established, so the significance is conditional on a substantial correction and verification.

major comments (5)
  1. [Section 2.2, Eq. (1)] Equation (1) is the core of the method and the current text is a material misstatement of the algorithm.
  2. [Section 3.2, Table 1] Please verify all improvement figures against the underlying data and correct the text and Abstract.
  3. [Section 3.3, Table 2] The ablation study is a central piece of evidence for the contribution of the proposed loss; the current numbers are not reproducible from the table.
  4. [Section 3.2, Table 1] The internal-testing 'leading performance' claim is not statistically supported against the closest baselines.
  5. [Section 3.1, Implementation] Without sensitivity analysis, the reader cannot assess how robust the central mechanism is to hyperparameter choices.
minor comments (5)
  1. [Abstract and Section 3.2] The reported metric is not present in the corresponding table.
  2. [Section 2.1] The preprocessing is under-specified.
  3. [Section 2.1] Clarify that the 'detection' is done with ground-truth labels during training.
  4. [Section 3.3, Figure 6] Discrepancy in the number of external test cases.
  5. [Throughout] Minor formatting issue.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the margin loss is a supervised auxiliary objective built from training labels, and the central performance claim is tested on held-out and external data; the one self-citation is preprocessing-only and not load-bearing.

full rationale

The derivation chain is an empirical supervised-learning pipeline, not a mathematical derivation that reduces to its inputs. The asymmetric-region mask m = s - s∩s' (Section 2.1) is computed from the training ground-truth mask s and is used only to weight the margin loss (Eq. 1) during training; at inference only the shared encoder-decoder is used (Section 2.3), so the reported segmentation output is not a restatement of m or s. The central claim is accordingly testable on held-out internal data and on the independent external SegRap2023 dataset (Table 1), where SATS is compared against ten baselines; this is external evidence, not a re-display of the paper's own fitted values. The only self-citation in the method, [33] (Same++), is used for head-neck position normalization before training; even if this preprocessing were imperfect, the consequence would be a robustness or correctness concern, not a circular reduction of the performance claim. The failure-case passage in Section 3.3 explicitly acknowledges that the method can misclassify symmetric lesions as asymmetric, which confirms that the symmetry premise is an empirical, falsifiable assumption rather than a definitionally guaranteed outcome. One non-circular correctness issue should be flagged separately: as printed, Eq. (1) assigns the minimization term ||E - E'||^2 to the m=1 (tumor-asymmetry) region, which is the opposite of the abstract's statement that the loss encourages a larger distance for tumor areas; if the implemented loss matches the printed equation, the reported +4.98 DSC ablation gain cannot be explained by the advertised asymmetry mechanism. This is a material correctness or typographical issue, not a circularity, and does not raise the circularity score. The score of 2 reflects only the minor, non-load-bearing self-citation in the preprocessing step.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the anatomical symmetry assumption, the label-derived asymmetry mask, and the affine atlas registration. The margin and loss weight are untested hyperparameters; no new physical entities are introduced.

free parameters (2)
  • margin t = 20
    Distance margin in Eq. (1); set without reported sensitivity analysis.
  • loss weight beta = 1
    Weight balancing the margin loss in the overall loss; no sensitivity study reported.
assumptions (3)
  • domain assumption Healthy nasopharyngeal region is bilaterally symmetric after affine alignment to a manually selected atlas.
    Assumed in Section 2.1 and Figure 1; the entire asymmetry-based loss depends on this premise.
  • domain assumption The asymmetry mask m = s - (s ∩ s'), computed from ground-truth mask s, isolates tumor-relevant asymmetry rather than non-pathological asymmetry (vessels, lymph nodes, pose).
    Section 2.1 defines m from labels; if non-pathological asymmetry dominates, the margin loss distracts the model. The paper notes lymphatic involvement can create additional asymmetric structures.
  • domain assumption Affine registration using Same++ [33] produces scans symmetric along the central sagittal plane.
    Section 2.1 preprocessing relies on this; alignment error propagates into the margin loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Leveraging Semantic Asymmetry for Precise Gross Tumor Volume Segmentation of Nasopharyngeal Carcinoma in Planning CT." pith.science (2026). https://pith.science/paper/ZOCMKDKV

@misc{pith2026241118290,
  author       = {Pith},
  title        = {Pith review of: Leveraging Semantic Asymmetry for Precise Gross Tumor Volume Segmentation of Nasopharyngeal Carcinoma in Planning CT},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZOCMKDKV}},
  note         = {Machine review of arXiv:2411.18290}
}
read the original abstract

In the radiation therapy of nasopharyngeal carcinoma (NPC), clinicians typically delineate the gross tumor volume (GTV) using non-contrast planning computed tomography to ensure accurate radiation dose delivery. However, the low contrast between tumors and adjacent normal tissues necessitates that radiation oncologists manually delineate the tumors, often relying on diagnostic MRI for guidance. % In this study, we propose a novel approach to directly segment NPC gross tumors on non-contrast planning CT images, circumventing potential registration errors when aligning MRI or MRI-derived tumor masks to planning CT. To address the low contrast issues between tumors and adjacent normal structures in planning CT, we introduce a 3D Semantic Asymmetry Tumor segmentation (SATs) method. Specifically, we posit that a healthy nasopharyngeal region is characteristically bilaterally symmetric, whereas the emergence of nasopharyngeal carcinoma disrupts this symmetry. Then, we propose a Siamese contrastive learning segmentation framework that minimizes the voxel-wise distance between original and flipped areas without tumor and encourages a larger distance between original and flipped areas with tumor. Thus, our approach enhances the sensitivity of features to semantic asymmetries. % Extensive experiments demonstrate that the proposed SATs achieves the leading NPC GTV segmentation performance in both internal and external testing, \emph{e.g.}, with at least 2\% absolute Dice score improvement and 12\% average distance error reduction when compared to other state-of-the-art methods in the external testing.

Figures

Figures reproduced from arXiv: 2411.18290 by the authors.

Figure 1
Figure 1. (a) Our SATS model begins with the encoder-decoder module, which uses shared weights to process two input signals and encode them into a discrimi￾native representation. This representation is then further processed through a non-linear projection module and a distance metric learning module to produce feature maps. (b) A graphical representation of our metric learning strategy. Cir￾cles indicate individual CT images… view at source ↗
Figure 2
Figure 2. Left: the data partitioning situation of the main comparing experiment. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Example CT slices with tumor segmentation overlays (red color) using [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Example CT slices with tumor segmentation overlays (red color) using [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Dice score compared to the baselines shown for each test subject. in external testing by 0.61% DSC. This result indicates the strong performance of CNN-based nnUNet over transformer-based segmentation models. 3.3 Ablation Studies Effect of projection head and margin lo…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 31 canonical work pages

  1. [30]

    Frontiers in oncology7, 315 (2017) 2

    Men, K., Chen, X., Zhang, Y., Zhang, T., Dai, J., Yi, J., Li, Y.: Deep deconvolu- tional neural network for target segmentation of nasopharyngeal cancer in planning computed tomography images. Frontiers in oncology7, 315 (2017) 2

    work page 2017

  2. [1]

    Matching in the Wild: Learning Anatomical Embeddings for Multi-Modality Images

    Bai, X., Bai, F., Huo, X., et al.: Matching in the wild: Learning anatomical em- beddings for multi-modality images. CoRRabs/2307.03535 (2023) 5

    work page Pith review arXiv 2023

  3. [2]

    Biomedical Signal Processing and Control64, 102246 (2021) 2

    Bai, X., Hu, Y., Gong, G., Yin, Y., Xia, Y.: A deep learning approach to segmenta- tion of nasopharyngeal carcinoma using computed tomography. Biomedical Signal Processing and Control64, 102246 (2021) 2

    work page 2021

  4. [3]

    In: ECCV

    Chen, H., Wang, Y., Zheng, K., Li, W., Chang, C., Harrison, A.P., et al.: Anatomy- aware siamese network: Exploiting semantic asymmetryfor accurate pelvic fracture detection in x-ray images. In: ECCV. vol. 12368, pp. 239–255 (2020) 2

    work page 2020

  5. [4]

    Neurocomputing394, 27–40 (2020) 2

    Chen, H., Qi, Y., Yin, Y., et al.: Mmfnet: A multi-modality MRI fusion network for segmentation of nasopharyngeal carcinoma. Neurocomputing394, 27–40 (2020) 2

    work page 2020

  6. [5]

    Medical Image Analysis p

    Chen, J., Mei, J., Li, X., Lu, Y., Yu, Q., Wei, Q., et al.: Transunet: Rethinking the u-net architecture design for medical image segmentation through the lens of transformers. Medical Image Analysis p. 103280 (2024) 6

    work page 2024

  7. [6]

    The Lancet394(10192), 64–80 (2019) 2

    Chen, Y.P., Chan, A.T., Le, Q.T., Blanchard, P., Sun, Y., Ma, J.: Nasopharyngeal carcinoma. The Lancet394(10192), 64–80 (2019) 2

    work page 2019

  8. [7]

    The Lancet 387(10022), 1012–1024 (2016) 1 10 Z

    Chua, M.L., Wee, J.T., Hui, E.P., Chan, A.T.: Nasopharyngeal carcinoma. The Lancet 387(10022), 1012–1024 (2016) 1 10 Z. Li et al

    work page 2016

Show all 34 references
  1. [8]

    In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries

    Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H.R., Xu, D.: Swin UNETR: swin transformers for semantic segmentation of brain tumors in MRI images. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. vol. 12962, pp. 272–284 (2021) 6

    2021

  2. [9]

    In: IEEE Winter Conference on Applications of Computer Vision

    Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B.A., et al.: UNETR: transformers for 3d medical image segmentation. In: IEEE Winter Conference on Applications of Computer Vision. pp. 1748–1758 (2022) 6

    2022

  3. [10]

    In: MICCAI

    He, Y., Nath, V., Yang, D., Tang, Y., Myronenko, A., Xu, D.: Swinunetr-v2: Stronger swin transformers with stagewise convolutions for 3d medical image seg- mentation. In: MICCAI. vol. 14223, pp. 416–426 (2023) 6

    2023

  4. [11]

    In: MICCAI

    Heinrich, M.P., Jenkinson, M., Brady, S.M., Schnabel, J.A.: Globally optimal de- formable registration on a minimum spanning tree using dense displacement sam- pling. In: MICCAI. pp. 115–122 (2012) 5

    2012

  5. [12]

    In: MICCAI

    Huang, J.b., Zhuo, E., Li, H., Liu, L., Cai, H., Ou, Y.: Achieving accurate segmen- tation of nasopharyngeal carcinoma in mr images through recurrent attention. In: MICCAI. pp. 494–502 (2019) 2

    2019

  6. [13]

    In: MICCAI

    Huang, J., Li, H., Li, G., Wan, X.: Attentive symmetric autoencoder for brain MRI segmentation. In: MICCAI. vol. 13435, pp. 203–213 (2022) 2

    2022

  7. [14]

    In: MICCAI

    Huang, W., Liu, W., Zhang, X., Yin, X., Han, X., Li, C., Gao, Y., et al.: Lidia: Precise liver tumor diagnosis on multi-phase contrast-enhanced ct via iterative fusion and asymmetric contrastive learning. In: MICCAI. pp. 394–404 (2024) 2

    2024

  8. [15]

    CoRR abs/2304.06716 (2023) 6

    Huang, Z., Wang, H., Deng, Z., Ye, J., Su, Y., et al.: Stu-net: Scalable and trans- ferable medical image segmentation models empowered by large-scale supervised pre-training. CoRR abs/2304.06716 (2023) 6

    2023 arXiv

  9. [16]

    In: MICCAI

    Isensee, F., Wald, T., Ulrich, C., Baumgartner, M., Roy, S., Maier-Hein, K., Jaeger, P.F.: nnu-net revisited: A call for rigorous validation in 3d medical image segmen- tation. In: MICCAI. pp. 488–498 (2024) 6

    2024

  10. [17]

    Oral Oncology110, 104862 (2020) 2

    Ke, L., Deng, Y., Xia, W., Qiang, M., et al.: Development of a self-constrained 3d densenet model in automatic detection and segmentation of nasopharyngeal carcinoma using magnetic resonance images. Oral Oncology110, 104862 (2020) 2

    2020

  11. [18]

    Radiotherapy and Oncology126(1), 25–36 (2018) 6

    Lee, A.W., Ng, W.T., Pan, J.J., Poh, S.S., Ahn, Y.C., AlHussain, H.o.: Interna- tional guideline for the delineation of the clinical target volumes (ctv) for nasopha- ryngeal carcinoma. Radiotherapy and Oncology126(1), 25–36 (2018) 6

    2018

  12. [19]

    In: MICCAI

    Li, C., Zhang, X., Gao, Y., Yin, X., Lu, L., et al.: Improved esophageal varices assessment from non-contrast ct scans. In: MICCAI. pp. 349–359 (2024) 2

    2024

  13. [20]

    Technology in cancer research & treatment18, 153–160 (2019) 2

    Li, S., Xiao, J., He, L., Peng, X., Yuan, X.: The tumor target segmentation of nasopharyngeal cancer in ct images based on deep learning methods. Technology in cancer research & treatment18, 153–160 (2019) 2

    2019

  14. [21]

    IEEE Transactions on Medical Imaging41(7), 1639–1650 (2022) 2

    Li, Y., Dan, T., Li, H., Chen, J., Peng, H., Liu, L., Cai, H.: Npcnet: Jointly seg- ment primary nasopharyngeal carcinoma tumors and metastatic lymph nodes in mr images. IEEE Transactions on Medical Imaging41(7), 1639–1650 (2022) 2

    2022

  15. [22]

    International Journal of Radiation Oncology* Biology* Physics113(4), 893–902 (2022) 2

    Liao, W., He, J., Luo, X., Wu, M., Shen, Y., et al.: Automatic delineation of gross tumor volume based on magnetic resonance imaging by performing a novel semisu- pervised learning framework in nasopharyngeal carcinoma. International Journal of Radiation Oncology* Biology* Phy...

    2022

  16. [23]

    Magnetic resonance imaging64, 190–199 (2019) 2

    Liu, C.F., Padhy, S., Ramachandran, S., et al.: Using deep siamese neural networks for detection of brain asymmetries associated with alzheimer’s disease and mild cognitive impairment. Magnetic resonance imaging64, 190–199 (2019) 2

    2019

  17. [24]

    In: MICCAI

    Liu, Y., Zhou, Z., Zhang, S., Luo, L., Zhang, Q., Zhang, F., et al.: From unilat- eral to bilateral learning: Detecting mammogram masses with contrasted bilateral network. In: MICCAI. vol. 11769, pp. 477–485 (2019) 2 Leveraging Semantic Asymmetry for NPC in Planning CT 11

    2019

  18. [25]

    In: MICCAI

    Luo, X., Liao, W., Chen, J., Song, T., Chen, Y., Zhang, S., et al.: Efficient semi- supervised gross target volume of nasopharyngeal carcinoma segmentation via un- certainty rectified pyramid consistency. In: MICCAI. pp. 318–329 (2021) 2

    2021

  19. [26]

    Radiotherapy and Oncology p

    Luo, X., Liao, W., He, Y., Tang, F., Wu, M., Shen, Y., et al.: Deep learning-based accurate delineation of primary gross tumor volume of nasopharyngeal carcinoma onheterogeneousmagneticresonanceimaging:alarge-scaleandmulti-centerstudy. Radiotherapy and Oncology p. 109480 (2023) 2

    2023

  20. [27]

    CoRRabs/2401.04722 (2024) 6

    Ma, J., Li, F., Wang, B.: U-mamba: Enhancing long-range dependency for biomed- ical image segmentation. CoRRabs/2401.04722 (2024) 6

    2024 arXiv

  21. [28]

    Physics in Medicine & Biology64(2), 025005 (2019) 2

    Ma, Z., Zhou, S., Wu, X., Zhang, H., Yan, W., et al.: Nasopharyngeal carcinoma segmentation based on enhanced convolutional neural networks using multi-modal metric learning. Physics in Medicine & Biology64(2), 025005 (2019) 2

    2019

  22. [29]

    Neurocomputing438, 211–222 (2021) 2

    Mei, H., Lei, W., Gu, R., Ye, S., Sun, Z., Zhang, S., et al.: Automatic segmentation of gross target volume of nasopharynx cancer using ensemble of multiscale deep neural networks with spatial attention. Neurocomputing438, 211–222 (2021) 2

    2021

  23. [31]

    American Journal of Roentgenology198(1), 11–18 (2012) 2

    Razek, A.A.K.A., King, A.: Mri and ct of nasopharyngeal carcinoma. American Journal of Roentgenology198(1), 11–18 (2012) 2

    2012

  24. [32]

    In: MICCAI

    Roy, S., Köhler, G., Ulrich, C., Baumgartner, M., Petersen, J., Isensee, F., et al.: Mednext: Transformer-driven scaling of convnets for medical image segmentation. In: MICCAI. vol. 14223, pp. 405–415 (2023) 6

    2023

  25. [33]

    ArXivabs/2311.14986 (2023) 3

    Tian, L., Li, Z., Liu, F., Bai, X., Ge, J., Lu, L., et al.: Same++: A self-supervised anatomical embeddings enhanced medical image registration framework using sta- ble sampling and regularized transformation. ArXivabs/2311.14986 (2023) 3

    2023 arXiv

  26. [34]

    IEEE Transactions on Medical Imaging 42(9), 2751–2762 (2023) 2

    Zeng, B., Wang, H., Xu, J., Tu, P., Joskowicz, L., Chen, X.: Two-stage structure- focused contrastive learning for automatic identification and localization of com- plex pelvic fractures. IEEE Transactions on Medical Imaging 42(9), 2751–2762 (2023) 2

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.