Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

A Novel Convolutional-Free Method for 3D Medical Imaging Segmentation

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

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

Pith's one-line read Proposal: a fully convolutional-free transformer with a joint thick-thin loss aims to beat CNN and hybrid baselines on 3D medical segmentation.

desk verdict A research proposal dressed as a completed paper: the abstract claims experimental superiority, but the body has no experiments, no results table, and no equation for the proposed loss. read the letter →

arxiv 2502.05396 v1 pith:TMP46WBM submitted 2025-02-08 eess.IV cs.CV

Canxuan Gang This is my paper · ORCID
classification eess.IVcs.CV
keywords 3Dmedicalimagesegmentationtransformerconvolution-freeself-attentionthick-thinsliceadaptationjointlossbrainhemorrhagebenchmarkCTimaging
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 proposes a fully convolutional-free transformer architecture for multi-semantic segmentation of 3D medical images, arguing that self-attention captures the long-range context that CNNs miss. It also introduces a joint weighted loss intended to let thick-slice annotations supervise thin-slice segmentation, addressing the scarcity of thin-slice ground truth. As supporting evidence it announces a private five-class brain-hemorrhage dataset with thin slices and corresponding thick slices. The manuscript describes the architecture and the planned evaluation using mIoU and Dice; the experimental results promised in the abstract are not included in this version.

What carries the argument

The central mechanism is a Vision-Transformer-style 3D pipeline: the input volume is cut into non-overlapping 3D patches, flattened and linearly embedded, added to sine-cosine position encodings, and processed by stacked transformer blocks with multi-head self-attention, residual connections, and layer normalization; an MLP with softmax decodes the mask for the center patch. The second mechanism is the joint weighted loss for thick-thin adaptation, defined through three supervision targets: prediction on thick slices equals the thick ground truth, the average mask intensity of corresponding thin slices equals the thick ground truth, and the average feature map of corresponding thin slices equals the thick feature map.

What would settle it

Train the proposed model with the joint loss on thick-slice annotations and evaluate it on thin-slice ground truth; if its Dice or mIoU does not exceed a baseline trained only on thick slices and tested on thin slices, the thin-thick adaptation claim is falsified. A more targeted check is to ablate the feature-map average term and show that removing it does not change thin-slice accuracy.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that an encoder-decoder built entirely from 3D patch embeddings, transformer encoder blocks, and an MLP softmax decoder—with no convolutional layers anywhere—can segment 3D medical images better than CNN-based and CNN-transformer hybrid models. It further claims that a three-part joint loss, combining direct thick-slice supervision, average intensity projection consistency between thin and thick masks, and average feature-map consistency, can adapt thick-slice annotations to thin-slice segmentation. The paper presents this as a solution to both the global-context limitation of CNNs and the practical shortage of thin-slice labels. The manuscript lays out the method and evaluation protocol; the reported superiority is asserted rather than demonstrated in the text.

Load-bearing premise

The thin-thick adaptation assumes that average intensity projection of thin slices, combined with radiologist thick-slice labels, preserves the information needed to supervise thin-slice segmentation, yet no equation or experiment demonstrates that this transfer actually works.

Editorial extensions

If this is right

  • If the architecture performs as claimed, pure transformer backbones could replace CNN encoders and decoders in 3D medical segmentation, removing the need for hybrid designs.
  • If the joint loss works, existing thick-slice annotations could be reused to train thin-slice segmenters, substantially lowering annotation cost.
  • The announced benchmark would let the community compare models on thin-slice multi-semantic segmentation, which the paper says is currently missing.
  • The joint-loss scheme could generalize to other slice-thickness or resolution mismatches in volumetric imaging.

Reading between the lines

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

  • A practical implementation would need to specify the relative weights of the three loss terms and the tiling strategy for full-volume inference, since the decoder predicts only the center patch of each block.
  • The average-intensity projection assumption treats thin-slice structures as linearly collapsed into thick-slice intensities; small or low-contrast hemorrhages may violate this, so the loss may need a weighting or a learned projection to survive.
  • A quick falsification test would compare thin-slice Dice from the joint loss against a simple baseline that trains on thick slices and resamples predictions; the paper's claim requires the former to win.
  • If validated, the same convolution-free recipe with adjusted patch size and normalization could be tried on MRI or PET volumes, where long-range context also matters.
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 manuscript proposes a fully convolution-free transformer model for 3D medical image segmentation, a joint loss for adapting thick-slice annotations to thin-slice segmentation, and a benchmark dataset of thin-slice brain hemorrhage CT scans. The methodology (Section 3) presents standard transformer encoder-decoder equations, patch embedding, and an MLP decoder, and describes a thin-thick adaptation loss in qualitative terms. Section 4 describes evaluation datasets and metrics but contains no experimental results. The paper is structured as a research proposal: Section 5 lists experimentation as future work, and Section 7 concludes by restating research aims rather than reporting findings.

Significance. If the claimed results existed, the contributions would be timely and valuable: a purely attention-based 3D segmentation architecture without convolutional inductive bias, a practical approach to use thick-slice annotations to supervise thin-slice segmentation, and a public thin-slice brain hemorrhage benchmark. The paper also correctly identifies a real data-availability bottleneck in medical imaging segmentation. However, the significance is entirely prospective: no experiments, no dataset, no code, and no quantitative evidence are provided. The only concrete equations (4)–(10) restate standard transformer and self-attention formulas from the literature, so the paper's novel components are not demonstrated or even fully specified. The central claim of demonstrated superiority is unsupported by the manuscript's own contents.

major comments (4)
  1. [Abstract / §4.2 / §5 / §7] The abstract states that "Our experiments demonstrate the superiority of the proposed model over traditional and hybrid architectures," yet the manuscript contains no experiments, no results tables, no metric values, and no figures comparing segmentation outputs. Section 4.2 describes the comparative and ablation experiments exclusively in future tense ("will be conducted"), Section 5 lists "Experimentation and analysis" and "Run comparative experiments..." as future tasks, and Section 7 concludes by recalling research aims rather than reporting outcomes. The central claim of demonstrated superiority is therefore unsupported by the document itself and cannot be verified.
  2. [§3.6] The joint thin-thick loss, which is the only novel methodological component for domain adaptation, is specified only as three qualitative bullet points. No equation defines the combined loss, no weighting coefficients are given, and the constraints "the average of the mask intensity of corresponding thin slices should be the ground truth of thick slices" and "the average of the feature maps of corresponding thin slices should be the feature map of thick slices" are not formalized. Without a precise loss expression, the method is not reproducible, and no experiment is provided to show that average-intensity-projection (AIP) thick-slice supervision transfers correctly to thin-slice segmentation.
  3. [§4.1] The claimed benchmark dataset is not delivered in this manuscript. The text states that a private brain-hemorrhage dataset "will be publicly released" and that it "will serve as a new benchmark," but no dataset, download link, statistics, acquisition protocol, or annotation details are provided. The data contribution is therefore only a plan, not a contribution that can be evaluated.
  4. [§2.1–§3] The methodological novelty beyond the absence of convolutions is not substantiated. The transformer encoder equations (4)–(9) are the standard multi-head self-attention and feed-forward formulas from Vaswani et al. (2017), and the MLP decoder in Eq. (10) is a linear projection followed by softmax. The paper cites the prior convolution-free approach of Karimi et al. [22] and the hybrid nnFormer [60] but provides no comparison with them. No architectural detail unique to this proposal (e.g., how the extract block or patch embeddings are configured for medical volumes) is evaluated against any baseline, so the claim of a novel and superior architecture is not supported.
minor comments (5)
  1. [Title / Section 2.2] The heading "Research Significants" should be "Research Significance."
  2. [Throughout] There are numerous typos and grammatical errors, e.g., "medicial imaging segmentations" (Section 2.1), "which shown in figure 1" and "which shows in figure 2" (Sections 1 and 3). These should be corrected.
  3. [References] References [11] and [12] are identical (both "Dynamically weighted balanced loss..." by Fernando and Tsokos), so one should be removed or replaced with the correct citation for DiffuMask, which the text appears to intend.
  4. [§4.3] The heading "Evaluation Matrices" should be "Evaluation Metrics," and the definitions of Intersection and Union in Eqs. (11)–(13) are informal; these should be formalized as sums over voxel-wise indicators.
  5. [§1] The introduction lists several future activities (e.g., "The report will be divided into distinct sections...") in a way more appropriate for a research proposal than a completed paper; the authors should restructure the manuscript to report completed work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found; the main claims are unsupported by missing experiments, which is a correctness issue rather than circularity.

full rationale

This manuscript is a proposal-style paper. It restates the standard transformer equations (Eqs. 4-9, from Vaswani et al.) and describes a qualitative joint loss (Section 3.6) with no experimental results. There is no fitted parameter that is later renamed as a prediction, and no equation is derived from an assumption that already contains the conclusion. The abstract's claim that 'experiments demonstrate the superiority' is contradicted by Sections 4.2 and 5, where the experiments are described in future tense and listed as future work; this is an unsupported-evidence and correctness problem, not a circularity issue. Citations to Zeyu Zhang et al., including Ref [58] (Thin-thick Adapter), are background or motivation and are not used as load-bearing proof of the proposed method's validity. The proposed loss is not derived from Ref [58]. Therefore no circular step satisfies the required specific reduction, such as an equation that reduces by construction or a fitted parameter renamed as a prediction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper does not fit any parameters because it contains no experiments. The load-bearing premises are the validity of the standard transformer for 3D patches, the sufficiency of AIP-generated thick slices for supervision, the consistency of the radiologist labels, and the claim that a convolution-free design can compete with CNNs. These are all assumed rather than checked.

free parameters (3)
  • Transformer hyperparameters (patch size w, block size W, embedding dimension D, number of layers k, number of heads h)
    The paper does not specify numeric values; these are design choices required to implement the model and they will affect the reported performance, but no values or search details are given.
  • Joint loss weighting coefficients
    Section 3.6 says a "joint weighted loss" is used without giving the weights or the functional form, so the relative importance of the three supervision constraints is unspecified.
  • Number of patches n = W/w
    Determined by block and patch sizes, not specified.
assumptions (4)
  • standard math Standard transformer self-attention and position encoding functions from Vaswani et al. are valid feature extractors for 3D patches.
    Section 3.2 to 3.4 reproduce the Transformer equations verbatim without proof or adaptation analysis.
  • domain assumption Thick slices generated by average intensity projection (AIP) from thin slices preserve sufficient anatomical information for supervision.
    Section 3.6 states the AIP generation procedure, but no experiment is provided to validate that thick-slice supervision transfers to thin-slice segmentation.
  • domain assumption Radiologist-labeled thick annotations on AIP thick slices are consistent with the thin-slice ground truth.
    The paper assumes the labels used for the adapted supervision are correct, without reporting inter-observer agreement or quality control.
  • ad hoc to paper A purely attention-based encoder-decoder can match CNN-based 3D segmentation without convolutional inductive bias.
    This is the paper's main architectural premise; it is asserted in the introduction and related work but not demonstrated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Novel Convolutional-Free Method for 3D Medical Imaging Segmentation." pith.science (2026). https://pith.science/paper/TMP46WBM

@misc{pith2026250205396,
  author       = {Pith},
  title        = {Pith review of: A Novel Convolutional-Free Method for 3D Medical Imaging Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TMP46WBM}},
  note         = {Machine review of arXiv:2502.05396}
}
read the original abstract

Segmentation of 3D medical images is a critical task for accurate diagnosis and treatment planning. Convolutional neural networks (CNNs) have dominated the field, achieving significant success in 3D medical image segmentation. However, CNNs struggle with capturing long-range dependencies and global context, limiting their performance, particularly for fine and complex structures. Recent transformer-based models, such as TransUNet and nnFormer, have demonstrated promise in addressing these limitations, though they still rely on hybrid CNN-transformer architectures. This paper introduces a novel, fully convolutional-free model based on transformer architecture and self-attention mechanisms for 3D medical image segmentation. Our approach focuses on improving multi-semantic segmentation accuracy and addressing domain adaptation challenges between thick and thin slice CT images. We propose a joint loss function that facilitates effective segmentation of thin slices based on thick slice annotations, overcoming limitations in dataset availability. Furthermore, we present a benchmark dataset for multi-semantic segmentation on thin slices, addressing a gap in current medical imaging research. Our experiments demonstrate the superiority of the proposed model over traditional and hybrid architectures, offering new insights into the future of convolution-free medical image segmentation.

Figures

Figures reproduced from arXiv: 2502.05396 by the authors.

Figure 1
Figure 1. Self-attention architecture from Transformer [ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Proposed pipeline In general, the related works mentioned above primarily utilize either convolutional architecture or incorpo￾rate transformer layers as part of the encoder-decoder architecture. In contrast, our proposed model deviates significantly from this approach by introducing a convolutional-free architecture. This novel approach aims to explore the potential of non-convolutional architectures for medical im… view at source ↗

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. Barlow-Swin: Toward a novel siamese-based segmentation architecture using Swin-Transformers

    cs.CV 2025-09 reject novelty 4.0 of 10

    Barlow-Swin is a hybrid medical segmenter that pairs a Barlow Twins-pretrained Swin encoder with a U-Net-like decoder, claiming competitive accuracy with fewer parameters.

Reference graph

Works this paper leans on

62 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [12]

    K. R. M. Fernando and C. P. Tsokos. Dynamically weighted balanced loss: Class imbalanced learning and confidence calibration of deep neural networks. IEEE Transactions on Neural Networks and Learning, 33(7):2940–2951, 2022

    work page 2022

  2. [22]

    Convolution-free medical image segmenta- tion using transformer networks

    Davood Karimi, Serge Didenko Vasylechko, and Ali Gholipour. Convolution-free medical image segmenta- tion using transformer networks. In MICCAI 2021, 2021

    work page 2021

  3. [60]

    nnformer: Interleaved transformer for volumetric segmentation

    Hong-Yu Zhou, Jiansen Guo, Yinghao Zhang, Lequan Yu, Liansheng Wang, and Yizhou Yu. nnformer: Interleaved transformer for volumetric segmentation. CoRR, abs/2109.03201, 2021

    arXiv 2021

  4. [1]

    Retrospective reconstruction of three-dimensional radiotherapy treatment plans of the thorax from two dimensional planning data

    Boxwala AA and Rosenman JG. Retrospective reconstruction of three-dimensional radiotherapy treatment plans of the thorax from two dimensional planning data. Int J Radiat Oncol Biol Phys., 28(4):1009–15, 1994

    work page 1994

  5. [2]

    Antonelli, A

    M. Antonelli, A. Reinke, S. Bakas, and et al. The medical segmentation decathlon. Nat Commun, 13(4128), 2022

    work page 2022

  6. [3]

    Badrinarayanan, A

    V. Badrinarayanan, A. Kendall, and R. Cipolla. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(12):2481– 2495, 2017

    work page 2017

  7. [4]

    Medical ai for early detection of lung cancer: A survey

    Guohui Cai, Ying Cai, Zeyu Zhang, Yuanzhouhan Cao, Lin Wu, Daji Ergu, Zhinbin Liao, and Yang Zhao. Medical ai for early detection of lung cancer: A survey. arXiv preprint arXiv:2410.14769, 2024

    arXiv 2024

  8. [5]

    Msdet: Receptive field enhanced multiscale detection for tiny pulmonary nodule

    Guohui Cai, Ying Cai, Zeyu Zhang, Daji Ergu, Yuanzhouhan Cao, Binbin Hu, Zhibin Liao, and Yang Zhao. Msdet: Receptive field enhanced multiscale detection for tiny pulmonary nodule. arXiv preprint arXiv:2409.14028, 2024

    arXiv 2024

Show all 62 references
  1. [6]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020

    2020

  2. [7]

    Yuille, and Yuyin Zhou

    Jieneng Chen, Yongyi Lu, Qihang Yu, Xiangde Luo, Ehsan Adeli, Yan Wang, Le Lu, Alan L. Yuille, and Yuyin Zhou. Transunet: Transformers make strong encoders for medical image segmentation. CoRR, abs/2102.04306, 2021

    2021 arXiv

  3. [8]

    Schwing, Alexander Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, and Rohit Girdhar. Masked- attention mask transformer for universal image segmentation. In CVPR, 2022

    2022

  4. [9]

    J. Deng, W. Dong, R. Socher, L. J. Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR 2009, 2009

    2009

  5. [10]

    An image is worth 16x16 words: Transform- ers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, et al. An image is worth 16x16 words: Transform- ers for image recognition at scale. In ICLR, 2021

    2021

  6. [13]

    Esa: Annotation- efficient active learning for semantic segmentation

    Jinchao Ge, Zeyu Zhang, Minh Hieu Phan, Bowen Zhang, Akide Liu, and Yang Zhao. Esa: Annotation- efficient active learning for semantic segmentation. arXiv preprint arXiv:2408.13491, 2024

    2024 arXiv

  7. [14]

    Non-contrast computed tomography in acute ischaemic stroke: A pictorial review

    Radhiana H, Syazarina SO, Shahizon Azura MM, Hilwati H, and Sobri MA. Non-contrast computed tomography in acute ischaemic stroke: A pictorial review. Med J Malaysia, 68(1):93–100, 2013

    2013

  8. [15]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In CVPR, 2022

    2022

  9. [16]

    Momentum contrast for unsupervised visual representation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. In CVPR, 2020

    2020

  10. [17]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR 2016, 2016

    2016

  11. [18]

    Can rotational thromboe- lastometry rapidly identify theragnostic targets in isolated traumatic brain injury? Emergency Medicine Australasia, 37(1):e14480, 2025

    Abhiram D Hiwase, Christopher D Ovenden, Lola M Kaukas, Mark Finnis, Zeyu Zhang, Stephanie O’Connor, Ngee Foo, Benjamin Reddi, Adam J Wells, and Daniel Y Ellis. Can rotational thromboe- lastometry rapidly identify theragnostic targets in isolated traumatic brain injury? Emerge...

    2025

  12. [19]

    Jaeger, Simon A

    Fabian Isensee, Paul F. Jaeger, Simon A. A. Kohl, and Klaus H. Maier-Hein. nnu-net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods, 18:203–211, 2021. 11

    2021

  13. [20]

    Sine activated low-rank matrices for parameter efficient learning

    Yiping Ji, Hemanth Saratchandran, Cameron Gordon, Zeyu Zhang, and Simon Lucey. Sine activated low-rank matrices for parameter efficient learning. arXiv preprint arXiv:2403.19243, 2024

    2024 arXiv

  14. [21]

    Impact of slice thickness, pixel size, and ct dose on the performance of automatic contouring algorithms

    Huang K, Rhee DJ, Ger R, Layman R, Yang J, Cardenas CE, and Court LE. Impact of slice thickness, pixel size, and ct dose on the performance of automatic contouring algorithms. Appl Clin Med Phys., 22(5):168–174, 2021

    2021

  15. [23]

    Quantifying translation-invariance in convolutional neural networks

    Eric Kauderer-Abrams. Quantifying translation-invariance in convolutional neural networks. CoRR, abs/1801.01450, 2018

    2018 arXiv

  16. [24]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS 2012, 2012

    2012

  17. [25]

    Deep learning for medical image segmentation

    Matthew Lai. Deep learning for medical image segmentation. CoRR, abs/1505.02000, 2015

    2015 arXiv

  18. [26]

    Lecun, L

    Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998

    1998

  19. [27]

    Sebastian Seung

    Kisuk Lee, Jonathan Zung, Peter Li, Viren Jain, and H. Sebastian Seung. Superhuman accuracy on the snemi3d connectomics challenge. CoRR, abs/1706.00120, 2017

    2017 arXiv

  20. [28]

    Downsampling for binary classification with a highly imbalanced dataset using active learning

    Wonjae Lee and Kangwon Seo. Downsampling for binary classification with a highly imbalanced dataset using active learning. Big Data Research, 28:100314, 2022

    2022

  21. [29]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In ICCV, 2021

    2021

  22. [30]

    Hough-cnn: Deep learning for segmentation of deep brain regions in mri and ultrasound

    Fausto Milletari, Seyed-Ahmad Ahmadi, Christine Kroll, and Annika Plate et al. Hough-cnn: Deep learning for segmentation of deep brain regions in mri and ultrasound. Computer Vision and Image Understanding, 164:92–102, 2017

    2017

  23. [31]

    Projectedex: Enhancing generation in explainable ai for prostate cancer

    Xuyin Qi, Zeyu Zhang, Aaron Berliano Handoko, Huazhan Zheng, Mingxi Chen, Ta Duc Huy, Vu Minh Hieu Phan, Lei Zhang, Linqi Cheng, Shiyu Jiang, et al. Projectedex: Enhancing generation in explainable ai for prostate cancer. arXiv preprint arXiv:2501.01392, 2025

    2025 arXiv

  24. [32]

    Medconv: Convolutions beat transformers on long-tailed bone density prediction

    Xuyin Qi, Zeyu Zhang, Huazhan Zheng, Mingxi Chen, Numan Kutaiba, Ruth Lim, Cherie Chiang, Zi En Tham, Xuan Ren, Wenxin Zhang, et al. Medconv: Convolutions beat transformers on long-tailed bone density prediction. arXiv preprint arXiv:2502.00631, 2025

    2025 arXiv

  25. [33]

    Stand-alone self-attention in vision models

    Prajit Ramachandran, Niki Parmar, Ashish Vaswani, Irwan Bello, Anselm Levskaya, and Jon Shlens. Stand-alone self-attention in vision models. In NIPS, 2019

    2019

  26. [34]

    Rau, S.C

    C.S. Rau, S.C. Wu, S.Y. Hsu, H.T. Liu, C.Y. Huang, T.M. Hsieh, S.E. Chou, W.T. Su, Y.W. Liu, and C.H. Hsieh. Concurrent types of intracranial hemorrhage are associated with a higher mortality rate in adult patients with traumatic subarachnoid hemorrhage: A cross-sectional retr...

    2019

  27. [35]

    Redmon, S

    J. Redmon, S. Divvala, R. Girshick, and A. Farhadi. You only look once: Unified, real-time object detection. In CVPR 2016, 2016

    2016

  28. [36]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015

    2015

  29. [37]

    L. Shen, W. Zhao, and L. Xing. Patient-specific reconstruction of volumetric computed tomography images from a single projection view via deep learning. Nat Biomed Eng, 3:880–888, 2019

    2019

  30. [38]

    Shirai, K

    K. Shirai, K. Nishiyama, T. Katsuda, Y. Ueda, M. Miyazaki, K. Tsujii, and S. Ueyama. Maximum intensity projection (mip) and average intensity projection (aip) in image guided stereotactic body radiation therapy (sbrt) for lung cancer. International Journal of Radiation Oncolog...

    2012

  31. [39]

    Very deep convolutional networks for large-scale image recogni- tion

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recogni- tion. In ICLR 2015, 2015

    2015

  32. [40]

    Segkan: High-resolution medical image segmentation with long-distance dependencies

    Shengbo Tan, Rundong Xue, Shipeng Luo, Zeyu Zhang, Xinran Wang, Lei Zhang, Daji Ergu, Zhang Yi, Yang Zhao, and Ying Cai. Segkan: High-resolution medical image segmentation with long-distance dependencies. arXiv preprint arXiv:2412.19990, 2024. 12

    2024 arXiv

  33. [41]

    Segstitch: Multidimensional transformer for robust and efficient medical imaging segmentation

    Shengbo Tan, Zeyu Zhang, Ying Cai, Daji Ergu, Lin Wu, Binbin Hu, Pengzhang Yu, and Yang Zhao. Segstitch: Multidimensional transformer for robust and efficient medical imaging segmentation. arXiv preprint arXiv:2408.00496, 2024

    2024 arXiv

  34. [42]

    Performance evaluation of deep learning networks for semantic segmentation of traffic stereo-pair images

    Vlad Taran, Nikita Gordienko, Yuriy Kochura, Yuri Gordienko, Alexandr Rokovyi, Oleg Alienin, and Sergii Stirenko. Performance evaluation of deep learning networks for semantic segmentation of traffic stereo-pair images. In International Conference on Computer Systems and Techn...

    2018

  35. [43]

    A survey of semantic segmentation

    Martin Thoma. A survey of semantic segmentation. CoRR, abs/1602.06541, 2016

    2016 arXiv

  36. [44]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, et al. Attention is all you need. In NIPS, 2017

    2017

  37. [45]

    Learning to model the tail

    Yu-Xiong Wang, Deva Ramanan, and Martial Hebert. Learning to model the tail. In NIPS, 2017

    2017

  38. [46]

    Bhsd: A 3d multi-class brain hemorrhage segmentation dataset

    Biao Wu, Yutong Xie, Zeyu Zhang, Jinchao Ge, Kaspar Yaxley, Suzan Bahadir, Qi Wu, Yifan Liu, and Minh-Son To. Bhsd: A 3d multi-class brain hemorrhage segmentation dataset. In International Workshop on Machine Learning in Medical Imaging, pages 147–156. Springer, 2023

    2023

  39. [47]

    Xlip: Cross-modal attention masked modelling for medical language-image pre-training

    Biao Wu, Yutong Xie, Zeyu Zhang, Minh Hieu Phan, Qi Chen, Ling Chen, and Qi Wu. Xlip: Cross-modal attention masked modelling for medical language-image pre-training. arXiv preprint arXiv:2407.19546, 2024

    2024 arXiv

  40. [48]

    Gamed-snake: Gradient- aware adaptive momentum evolution deep snake model for multi-organ segmentation

    Ruicheng Zhang, Haowei Guo, Zeyu Zhang, Puxin Yan, and Shen Zhao. Gamed-snake: Gradient- aware adaptive momentum evolution deep snake model for multi-organ segmentation. arXiv preprint arXiv:2501.12844, 2025

    2025 arXiv

  41. [49]

    Deep long-tailed learning: A survey

    Yifan Zhang, Bingyi Kang, Bryan Hooi, Shuicheng Yan, and Jiashi Feng. Deep long-tailed learning: A survey. CoRR, abs/2110.04596, 2021

    2021 arXiv

  42. [50]

    A deep learning approach to diabetes diagnosis

    Zeyu Zhang, Khandaker Asif Ahmed, Md Rakibul Hasan, Tom Gedeon, and Md Zakir Hossain. A deep learning approach to diabetes diagnosis. In Asian Conference on Intelligent Information and Database Systems, pages 87–99. Springer, 2024

    2024

  43. [51]

    Kmm: Key frame mask mamba for extended motion generation

    Zeyu Zhang, Hang Gao, Akide Liu, Qi Chen, Feng Chen, Yiran Wang, Danning Li, and Hao Tang. Kmm: Key frame mask mamba for extended motion generation. arXiv preprint arXiv:2411.06481, 2024

    2024 arXiv

  44. [52]

    Infinimotion: Mamba boosts memory in transformer for arbitrary long motion generation

    Zeyu Zhang, Akide Liu, Qi Chen, Feng Chen, Ian Reid, Richard Hartley, Bohan Zhuang, and Hao Tang. Infinimotion: Mamba boosts memory in transformer for arbitrary long motion generation. arXiv preprint arXiv:2407.10061, 2024

    2024 arXiv

  45. [53]

    Motion mamba: Efficient and long sequence motion generation

    Zeyu Zhang, Akide Liu, Ian Reid, Richard Hartley, Bohan Zhuang, and Hao Tang. Motion mamba: Efficient and long sequence motion generation. In European Conference on Computer Vision, pages 265–

  46. [54]

    Jointvit: Modeling oxygen saturation levels with joint supervision on long-tailed octa

    Zeyu Zhang, Xuyin Qi, Mingxi Chen, Guangxi Li, Ryan Pham, Ayub Qassim, Ella Berry, Zhibin Liao, Owen Siggs, Robert Mclaughlin, et al. Jointvit: Modeling oxygen saturation levels with joint supervision on long-tailed octa. In Annual Conference on Medical Image Understanding and...

    2024

  47. [55]

    Segreg: Segmenting oars by registering mr images and ct annotations

    Zeyu Zhang, Xuyin Qi, Bowen Zhang, Biao Wu, Hien Le, Bora Jeong, Zhibin Liao, Yunxiang Liu, Johan Verjans, Minh-Son To, et al. Segreg: Segmenting oars by registering mr images and ct annotations. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5. IEEE, 2024

    2024

  48. [56]

    Motion avatar: Generate human and animal avatars with arbitrary motion

    Zeyu Zhang, Yiran Wang, Biao Wu, Shuo Chen, Zhiyuan Zhang, Shiya Huang, Wenbo Zhang, Meng Fang, Ling Chen, and Yang Zhao. Motion avatar: Generate human and animal avatars with arbitrary motion. arXiv preprint arXiv:2405.11286, 2024

    2024 arXiv

  49. [57]

    Meddet: Generative adversarial distillation for efficient cervical disc herniation detection

    Zeyu Zhang, Nengmin Yi, Shengbo Tan, Ying Cai, Yi Yang, Lei Xu, Qingtai Li, Zhang Yi, Daji Ergu, and Yang Zhao. Meddet: Generative adversarial distillation for efficient cervical disc herniation detection. In 2024 IEEE International Conference on Bioinformatics and Biomedicine...

    2024

  50. [58]

    Thin-thick adapter: Segmenting thin scans using thick annotations

    Zeyu Zhang, Bowen Zhang, Abhiram Hiwase, Christen Barras, Feng Chen, Biao Wu, Adam James Wells, Daniel Y Ellis, Benjamin Reddi, Andrew William Burgan, Minh-Son To, Ian Reid, and Richard Hartley. Thin-thick adapter: Segmenting thin scans using thick annotations. OpenReview, 2023. 13

    2023

  51. [59]

    A landmark-based approach for instability prediction in distal radius fractures

    Yang Zhao, Zhibin Liao, Yunxiang Liu, Koen Oude Nijhuis, Britt Barvelink, Jasper Prijs, Joost Colaris, Mathieu Wijffels, Max Reijman, Zeyu Zhang, et al. A landmark-based approach for instability prediction in distal radius fractures. In 2024 IEEE International Symposium on Bio...

    2024

  52. [61]

    Unet++: A nested u-net architecture for medical image segmentation

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang. Unet++: A nested u-net architecture for medical image segmentation. In International Workshop on Deep Learning in Medical Image Analysis, volume 11045, pages 3–11, 2018

    2018

  53. [62]

    Zou, S.K

    K.H. Zou, S.K. Warfield, A. Bharatha, C.M. Tempany, M.R. Kaus, S.J. Haker, W.M. 3rd Wells, F.A. Jolesz, and R. Kikinis. Statistical validation of image segmentation quality based on a spatial overlap index. Acad Radiol, 11(2):178–89, 2004

    2004

  54. [63]

    Lienkamp, Thomas Brox, and Olaf Ronneberger

    ¨Ozg¨ un C ¸ i¸ cek, Ahmed Abdulkadir, Sonja S. Lienkamp, Thomas Brox, and Olaf Ronneberger. 3d u-net: Learning dense volumetric segmentation from sparse annotation. In MICCAI, 2016. 14

    2016

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.