Pith. sign in

REVIEW 4 major objections 6 minor 41 references

Automatic Contouring of Spinal Vertebrae on X-Ray using a Novel Sandwich U-Net Architecture

T0 review · 4 major / 6 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 The paper claims a U-Net with ReLU down and AReLU up reaches 83.58% Dice on thoracic X-ray vertebra contouring, beating 80.13% for a standard U-Net.

desk verdict A plausible but not-yet-isolated empirical claim: the 4.1% Dice gain attributed to the ReLU/AReLU sandwich is confounded with an added down-sampling layer, and the statistics as reported cannot be checked. read the letter →

arxiv 2507.09158 v1 pith:XNH6EJDO submitted 2025-07-12 eess.IV cs.AIcs.CV

Sunil Munthumoduku Krishna Murthy , Kumar Rajamani , Srividya Tirunellai Rajamani , Yupei Li , Qiyang Sun , Bjoern W. Schuller This is my paper · ORCID
classification eess.IVcs.AIcs.CV
keywords U-NetAReLURevertebrasegmentationthoracicspineX-rayDicescoresemanticactivationfunction
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

Outlining individual vertebrae in spinal X-rays is done by hand today, which is slow and inconsistent, and automated contouring would help mobility assessment and surgical planning. This paper tries to establish that a small change to the standard U-Net—plain ReLU in the contracting half and attention-based ReLU (AReLU) in the expanding half—makes thoracic vertebra contouring more accurate. On 60 held-out anterior-posterior X-ray images from a public spine dataset, the proposed 'sandwich' U-Net reports a Dice score (the standard 0–100 overlap measure) of 83.58% versus 80.13% for a standard U-Net, with paired t-test p-values below 0.001 against all three comparators. A layer-by-layer experiment shows the score climbing as each decoder block is switched to AReLU, which the authors attribute to better preservation of vertebral edges and partial vertebrae. If the result holds, the improvement costs almost nothing in architecture complexity and could be folded into existing segmentation pipelines.

What carries the argument

The machinery is the sandwich U-Net itself: an encoder–decoder network with six ReLU-activated down-sampling blocks and five AReLU-activated up-sampling blocks. AReLU is the load-bearing component, a learnable activation that keeps negative inputs alive with a small clamped slope while scaling positive inputs with a sigmoid-driven gain, giving the decoder a trainable way to emphasize or suppress features during reconstruction. The paper's incremental experiment—adding AReLU to one additional decoder block at a time and watching the Dice score rise monotonically—is the evidence that ties this mechanism to the contour improvements.

What would settle it

Retrain the baseline U-Net and the sandwich U-Net on the same 300 anterior-posterior X-rays across several random splits and several seeds per split, then compare Dice distributions on one fixed 60-image test set; if the two distributions overlap substantially or the mean difference is not consistently above zero with non-overlapping confidence intervals, the claimed superiority fails.

Watch

Extended reading notes

Core claim

At its core, the paper claims that separating activation functions along the two halves of a U-Net makes each half specialize: the six down-sampling blocks, using ReLU, extract sparse, stable features, while the five up-sampling blocks, using AReLU, adaptively reconstruct details and sharpen boundaries. AReLU is a learnable attention-based ReLU defined as $F(x_i,\alpha,\beta)=R(x_i)+L(x_i,\alpha,\beta)$, where the negative branch has a clamped slope $C(\alpha)\in[0.01,0.99]$ and the positive branch is scaled by $1+\sigma(\beta)$. With $\alpha=\beta=0.9$, the model achieves 83.58% Dice on the 60-image test set versus 80.13% for the baseline U-Net, and the paper's step-by-step table shows Dice rising from 80.13% to 83.58% as zero through five decoder blocks are converted to AReLU. The authors conclude that the sandwich model produces contours that match ground truth more closely, especially at lower vertebral borders, reducing over- and under-segmentation.

Load-bearing premise

The load-bearing premise is that one random 80/20 split of the 300-image set and one training run per configuration are enough to trust the 3.5-point Dice gap as a real difference rather than split choice or run-to-run noise.

Editorial extensions

If this is right

  • A team with an existing U-Net could adopt the change by swapping only the decoder activation functions, with no new data collection and no added inference-time memory cost beyond the AReLU computation.
  • At the reported accuracy, automated contours could support quantitative measurements of vertebral alignment, curvature, and spacing for early diagnosis of scoliosis and kyphosis and for planning spinal fusion.
  • The reported p-values below 0.001 for SegNet, EfficientNet, and U-Net comparisons are the paper's evidence that the Dice gap is not a within-test chance fluctuation.
  • Because the improvement appears most strongly at vertebral borders and partial vertebrae, the method targets exactly the cases where manual contouring is slowest and most inconsistent.

Reading between the lines

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

  • The paper changes two things at once—the decoder activations and the number of down-sampling blocks—so the AReLU contribution alone is not isolated; a follow-up that deepens the baseline equally would separate the two effects.
  • A transfer test on lateral spine X-ray views or CT slices would show whether path-split activation is a general encoder–decoder property or specific to anterior-posterior thoracic images.
  • Claimed edge benefits would be better captured by boundary metrics like Hausdorff distance or average surface distance, since Dice is fairly insensitive to small contour shifts.
  • Replacing AReLU with a simpler learnable slope such as PReLU in the decoder would reveal whether the attention gating or merely the learnable slope drives the gain.
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 / 6 minor

Summary. The paper proposes a modified U-Net, called a 'Sandwich U-Net', that uses ReLU activations in the encoder/contracting path and AReLU activations in the decoder/expanding path for segmenting thoracic vertebrae in AP-view X-ray images. On a 60-image held-out test set, the authors report a Dice score of 83.58% for the proposed model versus 80.13% for a standard U-Net, along with a layer-by-layer ablation of incremental AReLU adoption and paired t-test p-values below 0.001. The paper concludes that integrating dual activation functions improves segmentation accuracy and contour quality for vertebral contouring.

Significance. If the reported improvement is genuine and attributable to the activation change, the paper demonstrates a simple and low-cost architectural modification that yields a modest but consistent improvement of roughly 3.5 Dice points on a clinically relevant segmentation task. The use of a public dataset and a held-out test set is a positive feature, and the layer-by-layer ablation in Table 1 is a useful presentation of the effect of progressively introducing AReLU. However, the significance is currently limited by the absence of variance estimates, the unresolved confound between activation choice and network depth, and the incomplete statistical reporting; the contribution would be substantially strengthened by an isolated ablation and rigorous multi-run evaluation.

major comments (4)
  1. [Sections 4.1, 5.4, Table 1 and Table 3] The central claim that AReLU in the decoder produces the 3.45-point Dice improvement is confounded with a depth change: Section 4.1 states that the proposed architecture includes 'six down-sampling layers and five up-sampling layers' and explicitly justifies 'the inclusion of an additional down-sampling layer' as part of the proposal, while Section 3.3 does not specify the number of down-sampling layers in the standard U-Net. Consequently, Table 3 compares models that differ in both depth and activation, and Table 1's row for '0 AReLU' is ambiguous as to whether it represents the shallow baseline or a deeper all-ReLU variant. Please provide an ablation that fixes network depth while varying only the activation functions, and report the layer counts of all compared models.
  2. [Section 5.5, Table 4] The paired t-test results are not backed by a described statistical procedure. Table 3 provides a single aggregated Dice score per model; if the test is performed on per-image Dice scores, the paper must describe the pairing, report the number of paired samples, and give the mean and standard deviation (or confidence intervals) of the per-image scores for each model. If only one score per model exists, a paired t-test cannot be applied and the reported p-values are unsupported. Please either supply the full statistical details or remove the significance claims.
  3. [Sections 3.2, 5.2, 5.3] All results are based on a single random 80/20 train/validation split and one training run per configuration, with no seeds, repeated runs, or variance estimates reported. Since training stochasticity and split choice can shift Dice scores by several points, the reported 3.45-point gap cannot be distinguished from run-to-run variation. Please report the mean and standard deviation over multiple runs (e.g., five random seeds) for at least the baseline and proposed models, or provide bootstrap confidence intervals.
  4. [Section 4.4, Table 2] The grid search for alpha and beta initializations is reported as selecting the values that give 'the best accuracy', but the manuscript does not state whether this selection was made on the validation set or the test set. If the test set was used for selection, the reported test Dice of 83.58% is optimistically biased by selection. Please clarify that the hyperparameter selection was performed solely on validation data, and if not, re-evaluate on a truly held-out test set.
minor comments (6)
  1. [Section 4.1] The sentence 'the second half of the "U" shown in Fig. 3 depicts the novel U-Net architecture' is grammatically incomplete; it should be split into two sentences for clarity.
  2. [Sections 1 and 4.3] Describing AReLU as 'attention based' may be misleading because the equations define a learnable parametric activation function, not an attention mechanism over feature maps; please clarify the relationship to reference [5] and justify the terminology.
  3. [Equation (1)] The text says 'C(·) clamps the input variable into [0.01, 0.99]', but in the equation C is applied to the parameter alpha rather than to the input xi; please reword to avoid confusion.
  4. [Section 5.2] The text 'Python III' should be 'Python 3', and 'google co-lab' should be 'Google Colab' for correctness.
  5. [Sections 5.2 and 5.3] The manuscript gives no training hyperparameters (learning rate, batch size, number of filters, etc.) for any of the models, and it does not include a code or data availability statement; such details are needed for reproducibility.
  6. [Tables 1 and 2] Tables 1 and 2 report Dice scores without standard deviations or an indication of whether the numbers are from the validation or test set; adding these details would help readers judge the reliability of the comparisons.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central Dice comparison is measured on a held-out test set, AReLU is external prior work, and no reported result is equal by construction to a fitted parameter.

full rationale

The paper's central claim is an empirical architecture comparison: a standard U-Net with ReLU versus a proposed U-Net with ReLU in the encoder and AReLU in the decoder. The reported numbers (80.13% vs 83.58% Dice) come from a 60-image test set that is separate from the 240 training and validation images, so the improvement is not forced by fitting the test set. The AReLU activation is defined by Equations 1-4 and is taken from prior work by Chen et al. [5], not derived tautologically from the paper's own objective; the paper applies it in the decoder and measures its effect. The alpha and beta initializations (Table 2) are selection choices made before the final model evaluation, which is ordinary hyperparameter selection rather than a fitted input being renamed as a prediction. The layer-by-layer AReLU progression in Table 1 and the paired t-tests in Table 4 are post-hoc statistical comparisons, not self-referential derivations. The self-citations appearing in the reference list are in related-work or methodological contexts and are not load-bearing for the main claim. A reader could object that the proposed architecture also adds an extra down-sampling layer relative to a canonical U-Net, and that the ablation in Table 1 does not isolate activation from depth; that is a genuine experimental-design limitation and a threat to causal attribution, but it is not circularity because the test-set Dice values would remain what they are regardless of how the architectures are described. No equation in the paper reduces to its own input, and no predicted quantity is defined as the fitted parameter itself.

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

The paper introduces no new theoretical entities. The central claim rests on the quality of manual annotations, the suitability of AReLU from prior work, and the representativeness of a single data split. The two tuneable initializations (alpha, beta) are the only explicit free parameters selected by validation performance.

free parameters (2)
  • alpha_init (AReLU clamp parameter) = 0.9
    Selected by grid search over the validation set in Section 4.4 (Table 2).
  • beta_init (AReLU shift parameter) = 0.9
    Selected by grid search over the validation set in Section 4.4 (Table 2).
assumptions (3)
  • domain assumption Manual polygon annotations produced by the authors with physician guidance are accurate ground-truth segmentations.
    Section 3.2 Stage 1; no inter-annotator agreement or quality metric is reported.
  • domain assumption AReLU as defined by Chen et al. [5] is a valid attention-based activation suitable for decoder feature reconstruction.
    Section 4.3 relies on [5] for properties (faster training, resistance to vanishing gradients).
  • domain assumption The single random split of 300 images (192/48/60) yields representative training, validation, and test sets.
    Section 3.2; no cross-validation or repeated splits.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automatic Contouring of Spinal Vertebrae on X-Ray using a Novel Sandwich U-Net Architecture." pith.science (2026). https://pith.science/paper/XNH6EJDO

@misc{pith2026250709158,
  author       = {Pith},
  title        = {Pith review of: Automatic Contouring of Spinal Vertebrae on X-Ray using a Novel Sandwich U-Net Architecture},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XNH6EJDO}},
  note         = {Machine review of arXiv:2507.09158}
}
read the original abstract

In spinal vertebral mobility disease, accurately extracting and contouring vertebrae is essential for assessing mobility impairments and monitoring variations during flexion-extension movements. Precise vertebral contouring plays a crucial role in surgical planning; however, this process is traditionally performed manually by radiologists or surgeons, making it labour-intensive, time-consuming, and prone to human error. In particular, mobility disease analysis requires the individual contouring of each vertebra, which is both tedious and susceptible to inconsistencies. Automated methods provide a more efficient alternative, enabling vertebra identification, segmentation, and contouring with greater accuracy and reduced time consumption. In this study, we propose a novel U-Net variation designed to accurately segment thoracic vertebrae from anteroposterior view on X-Ray images. Our proposed approach, incorporating a ``sandwich" U-Net structure with dual activation functions, achieves a 4.1\% improvement in Dice score compared to the baseline U-Net model, enhancing segmentation accuracy while ensuring reliable vertebral contour extraction.

Figures

Figures reproduced from arXiv: 2507.09158 by the authors.

Figure 1
Figure 1. Annotated thoracic vertebrae on X-Ray image in AP view sample for one patient: Orig [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Algorithm Architecture Flow. 4 Proposed Methodology We propose a novel architecture in this section, the schematic flow chart of which is shown in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Proposed Novel U-Net architecture with a sandwich of ReLU and AReLU activation [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Plot of the Loss and Dice Score of the U-Net model. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Plot of the Loss and Dice Score of the optimised sandwich model. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Incremental addition of AReLU to each layer during up-sampling. The 6 rows, from [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Predicted contours by the U-Net model and the novel sandwich U-Net model. The [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Contours predicted on an unseen data using Proposed Novel U-Net Architecture. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 40 canonical work pages

  1. [1]

    Vijay Badrinarayanan, Alex Kendall, and Roberto Cipolla. 2017. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE transactions on pattern analysis and machine intelligence 39, 12 (2017), 2481–2495

    work page 2017

  2. [2]

    Koushik Biswas, Meghana Karri, and Ulas ¸ Ba˘gcı. 2023. A Non-monotonic Smooth Activation Func- tion. arXiv preprint arXiv:2310.10126 (2023)

    work page Pith review arXiv 2023

  3. [3]

    Hu Cao, Yueyue Wang, Joy Chen, Dongsheng Jiang, Xiaopeng Zhang, Qi Tian, and Manning Wang

    work page

  4. [4]

    Animesh Chaturvedi, N Apoorva, Mayank Sharan Awasthi, Shubhra Jyoti, DP Akarsha, S Brunda, and CS Soumya. 2022. Analyzing the Performance of Novel Activation Functions on Deep Learning Architectures. In Emerging Research in Computing, Information, Communication and Applications: Proceedings of ERCICA 2022. Springer, 903–915

    work page 2022

  5. [5]

    Dengsheng Chen, Jun Li, and Kai Xu. 2020. AReLU: Attention-based Rectified Linear Unit. arXiv preprint arXiv:2006.13858 (2020). 13

    work page Pith review arXiv 2020

  6. [6]

    Getao Du, Xu Cao, Jimin Liang, Xueli Chen, and Yonghua Zhan. 2020. Medical Image Segmentation based on U-Net: A Review. Journal of Imaging Science & Technology64, 2 (2020)

    work page 2020

  7. [7]

    Shiv Ram Dubey, Satish Kumar Singh, and Bidyut Baran Chaudhuri. 2022. Activation functions in deep learning: A comprehensive survey and benchmark. Neurocomputing 503 (2022), 92–108

    work page 2022

  8. [8]

    Ali Hatamizadeh, Yucheng Tang, Vishwesh Nath, Dong Yang, Andriy Myronenko, Bennett Landman, Holger R Roth, and Daguang Xu. 2022. Unetr: Transformers for 3d medical image segmentation. In Proceedings of the IEEE/CVF winter conference on applications of computer vision. 574–584

    work page 2022

Show all 41 references
  1. [9]

    Fabio Hellmann, Zhao Ren, Elisabeth Andr ´e, and Bj ¨orn W Schuller. 2021. Deformable dilated faster R-CNN for universal lesion detection in CT images. In 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). IEEE, 2896–2902

    2021

  2. [10]

    Huimin Huang, Lanfen Lin, Ruofeng Tong, Hongjie Hu, Qiaowei Zhang, Yutaro Iwamoto, Xianhua Han, Yen-Wei Chen, and Jian Wu. 2020. Unet 3+: A full-scale connected unet for medical image segmentation. In ICASSP 2020-2020 IEEE international conference on acoustics, speech and sign...

    2020

  3. [11]

    Huimin Huang, Lanfen Lin, Ruofeng Tong, Hongjie Hu, Qiaowei Zhang, Yutaro Iwamoto, Xianhua Han, Yen-Wei Chen, and Jian Wu. 2020. WNET: An end-to-end atlas-guided and boundary-enhanced network for medical image segmentation. In 2020 IEEE 17th International Symposium on Biomedic...

    2020

  4. [12]

    Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Koehler, Tobias Norajitra, Sebastian Wirkert, and Klaus H

    Fabian Isensee, Jens Petersen, Andre Klein, David Zimmerer, Paul F. Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Koehler, Tobias Norajitra, Sebastian Wirkert, and Klaus H. Maier-Hein. 2019. Abstract: nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation. ...

    2019

  5. [13]

    Ameya D Jagtap and George Em Karniadakis. 2023. How important are activation functions in regres- sion and classification? A survey, performance comparison, and future directions. Journal of Machine Learning for Modeling and Computing 4, 1 (2023)

    2023

  6. [14]

    Yuan Jiang and Zhi-Hua Zhou. 2004. SOM ensemble-based image segmentation. Neural Processing Letters 20 (2004), 171–178

    2004

  7. [15]

    M Krithika Alias AnbuDevi and K Suganthi. 2022. Review of semantic segmentation of medical images using modified architectures of UNET. Diagnostics 12, 12 (2022), 3064

    2022

  8. [16]

    Vladim ´ır Kunc. 2024. Exploring the Relationship: Transformative Adaptive Activation Functions in Comparison to Other Activation Functions. arXiv preprint arXiv:2402.09249 (2024)

    2024 arXiv

  9. [17]

    Bingtao Liang, Chen Tang, Wei Zhang, Min Xu, and Tianbo Wu. 2023. N-Net: an UNet architecture with dual encoder for medical image segmentation. Signal, Image and Video Processing17, 6 (2023), 3073–3081

    2023

  10. [18]

    Guosheng Lin, Anton Milan, Chunhua Shen, and Ian D. Reid. 2017. RefineNet: Multi-path Refinement Networks for High-Resolution Semantic Segmentation. In2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017. IEEE Computer...

    2017

  11. [19]

    Marina Adriana Mercioni and Stefan Holban. 2023. A Brief Review of the Most Recent Activation Functions for Neural Networks. In 2023 17th International Conference on Engineering of Modern Electric Systems (EMES). IEEE, 1–4

    2023

  12. [20]

    Shervin Minaee, Yuri Boykov, Fatih Porikli, Antonio Plaza, Nasser Kehtarnavaz, and Demetri Ter- zopoulos. 2021. Image segmentation using deep learning: A survey. IEEE transactions on pattern analysis and machine intelligence 44, 7 (2021), 3523–3542

    2021

  13. [21]

    Kumar Rajamani, Sahana D Gowda, Vishwa Nedunoori Tej, and Srividya Tirunellai Rajamani. 2022. Deformable attention (DANet) for semantic image segmentation. In 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). IEEE, 3781–3784

    2022

  14. [22]

    ElagiriRamalingam, and M.P

    Kumar Rajamani, Priya Rani, H Siebert, R. ElagiriRamalingam, and M.P. Heinrich. 2022. Attention- augmented U-Net (AA-U-Net) for semantic segmentation. In Signal, Image and Video Processing

    2022

  15. [23]

    Srividya Tirunellai Rajamani, Kumar Rajamani, and Bj ¨orn W Schuller. 2023. A novel and simple approach to regularise attention frameworks and its efficacy in segmentation. In 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC...

    2023

  16. [24]

    Srividya Tirunellai Rajamani, Kumar Rajamani, Ashwin Venkateshvaran, Andreas Triantafyllopoulos, Alexander Kathan, and Bj ¨orn W Schuller. 2023. Towards detecting and addressing corner cases in deep learning based medical image segmentation. IEEE Access (2023)

    2023

  17. [25]

    Andrinandrasana David Rasamoelina, Ivan C ´ık, Peter Sincak, Mari´an Mach, and Luk´aˇs Hruˇska. 2022. A large-scale study of activation functions in modern deep neural network architectures for efficient convergence. Inteligencia Artificial 25, 70 (2022), 95–109

    2022

  18. [26]

    Rituparna Sarma and Yogesh Kumar Gupta. 2021. A comparative study of new and existing seg- mentation techniques. In IOP conference series: materials science and engineering , V ol. 1022. IOP Publishing, 012027

    2021

  19. [27]

    Longfeng Shen, Qiong Wang, Yingjie Zhang, Fenglan Qin, Hengjun Jin, and Wei Zhao. 2023. DSKCA-UNet: Dynamic selective kernel channel attention for medical image segmentation.Medicine 102, 39 (2023), e35328

    2023

  20. [28]

    Ningthoujam Johny Singh and Kishorjit Nongmeikapam. 2023. Semantic segmentation of satellite images using deep-UNet. Arabian Journal for Science and Engineering 48, 2 (2023), 1193–1205

    2023

  21. [29]

    Hao Su, Shunjun Wei, Min Yan, Chen Wang, Jun Shi, and Xiaoling Zhang. 2019. Object detection and instance segmentation in remote sensing imagery based on precise mask R-CNN. In IGARSS 2019- 2019 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 1454–1457

    2019

  22. [30]

    Zhongbin Su, Wei Li, Zheng Ma, and Rui Gao. 2022. An improved U-Net method for the semantic segmentation of remote sensing images. Applied Intelligence 52, 3 (2022), 3276–3288

    2022

  23. [31]

    Fan Sun, Zhiming Luo, and Shaozi Li. 2023. Boundary difference over union loss for medical im- age segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 292–301

    2023

  24. [32]

    Mingxing Tan and Quoc Le. 2019. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning. PMLR, 6105–6114. 15

    2019

  25. [33]

    Dongyue Wu, Zilin Guo, Aoyan Li, Changqian Yu, Changxin Gao, and Nong Sang. 2023. Conditional boundary loss for semantic segmentation. IEEE Transactions on Image Processing(2023)

    2023

  26. [34]

    Changqian Yu, Jingbo Wang, Chao Peng, Changxin Gao, Gang Yu, and Nong Sang. 2018. Learning a Discriminative Feature Network for Semantic Segmentation. In 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE Computer Society

    2018

  27. [35]

    Chengyu Yuan, Hao Xiong, Guoqing Shangguan, Hualei Shen, Dong Liu, Haojie Zhang, Zhonghua Liu, Kun Qian, Bin Hu, Bj ¨orn W Schuller, et al. 2024. Deep Fusion of Shifted MLP and CNN for Medical Image Segmentation. In ICASSP 2024-2024 IEEE International Conference on Acoustics, ...

    2024

  28. [36]

    Kaidong Zhang and Dong Liu. 2023. Customized segment anything model for medical image segmen- tation. arXiv preprint arXiv:2304.13785 (2023)

    2023 arXiv

  29. [37]

    Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. 2021. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. In Proceedings of the IEEE/CVF conferen...

    2021

  30. [38]

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang. 2018. UNet++: A Nested U-Net Architecture for Medical Image Segmentation. InDeep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. 3–11

    2018

  31. [39]

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang. 2019. Unet++: Redesigning skip connections to exploit multiscale features in image segmentation. IEEE transactions on medical imaging 39, 6 (2019), 1856–1867

    2019

  32. [40]

    Guohua Zhu, Zewen Piao, and Suk Chan Kim. 2020. Tooth detection and segmentation with mask R- CNN. In 2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC). IEEE, 070–072. 16

    2020

  33. [2022]

    InEuropean conference on computer vision

    Swin-unet: Unet-like pure transformer for medical image segmentation. InEuropean conference on computer vision. Springer, 205–218

Pith tools

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