Pith. sign in

REVIEW 3 major objections 4 minor 33 references

Dual Semantic-Aware Network for Noise Suppressed Ultrasound Video Segmentation

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

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

Pith's one-line read DSANet segments noisy ultrasound video by matching channels between adjacent frames, not pixels, and the paper reports top IoU and Dice on four benchmarks with faster inference than video-based alternatives.

desk verdict The paper's central AFSA equation is degenerate (cosine similarity on scalars collapses to ±1), so the main mechanism is not actually described and the method is unreproducible as written; that said, the overall architecture and experiments are coherent, making this a fixable-but-flawed submission rather than a hopeless one. read the letter →

arxiv 2507.07443 v1 pith:5VYIV3YG submitted 2025-07-10 cs.CV

Ling Zhou , Runtian Yuan , Yi Liu , Yuejie Zhang , Rui Feng , Shang Gao This is my paper
classification cs.CV
keywords ultrasoundvideosegmentationspecklenoisesuppressionchannel-wisesemanticsimilaritytemporalfeaturefusionlocal-globalmedicalimagereal-timeinference
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 DSANet, a network for segmenting organs and lesions in ultrasound video, and claims that it suppresses speckle noise by computing semantic similarities between adjacent frames at the channel level instead of matching individual pixels. On four benchmark ultrasound video datasets, the authors report that DSANet reaches the best MAE, IoU, and Dice scores among twelve compared methods, beating the next-best by roughly one percentage point in IoU and Dice. They also report that the model runs faster than all video-based baselines and some image-based ones because it avoids pixel-level dependencies. If the claim holds, it would show that a lightweight channel-level temporal fusion can replace expensive pixel-level cross-attention while improving noise robustness.

What carries the argument

The central mechanism is the channel-level semantic similarity computation in the AFSA module: it reshapes the two adjacent frame feature maps to $\mathbb{R}^{C\times HW}$, computes a cosine-similarity matrix $S$ that weights each activation by its consistency with the corresponding activation in the other frame, and applies $S$ element-wise to both frames before global-average-pooling and multiplicative fusion. The LGSA module then performs channel splitting and cross-reassembly, concatenating the first half of the global feature channels with the second half of the local feature channels and vice versa, so that temporally conditioned context and frame-specific spatial detail are recombined at every scale. The intended role of this machinery is to let speckle noise average out in coarse channel-level comparisons, yielding temporal consistency without pixel-level correspondences and with lower computational cost.

What would settle it

Run the released code on several clips from ERUS10K and log the entries of the similarity matrix S inside the AFSA module; if all entries turn out to be exactly $\pm 1$, the matrix cannot order features by semantic similarity and the reported gains must be attributed to the surrounding pooling, gating, or fusion operations rather than to the described alignment. If, instead, the implementation compares whole channel activation vectors per channel, then the paper's stated mechanism is identifiable and its central claim can be checked directly.

Watch

Extended reading notes

Core claim

The paper's central claim is that ultrasound video segmentation can be made materially more robust to speckle noise by keeping inter-frame interaction at the channel level. The Adjacent-Frame Semantic-Aware (AFSA) module builds a semantic similarity matrix between two consecutive frames and uses it to refine both feature maps, then pools and multiplies the refined features to produce a temporally enhanced representation. The Local-and-Global Semantic-Aware (LGSA) module splits the channels of these temporally fused global features and the unconditional local features from the same frame, recombines complementary halves, and upsamples the result into the final segmentation. With Dice, weighted binary cross-entropy, and weighted IoU losses applied to the target frame and its two predecessors, DSANet reports the highest scores on the ERUS10K, Prostate, CAMUS, and VTUS benchmarks, outperforming the strongest prior method by close to one percent in IoU and Dice while reaching between 33.8 and 38.3 frames per second.

Load-bearing premise

The method's noise-suppression story rests on the assumption that the similarity values AFSA computes between adjacent frames carry real semantic information; taken literally, the published equation computes cosine similarity between single numbers, which gives only plus or minus one, so a meaningful version of the operation must exist in the code or in a corrected formula.

Editorial extensions

If this is right

  • Channel-level temporal fusion can replace pixel-level cross-attention in video segmentation, giving a speed advantage while improving IoU and Dice on all four tested ultrasound benchmarks.
  • Applying auxiliary segmentation losses to the two preceding frames as well as the target frame helps temporal consistency and stabilizes predictions in noisy clips.
  • Under simulated multiplicative Gamma speckle noise with a lower look number, the model's accuracy degrades less when AFSA and LGSA are included than for the baseline, supporting the noise-suppression explanation.
  • Because the architecture builds on a U-Net-style backbone and does not require optical flow or dense attention, it can plausibly be retrained for other video segmentation tasks at similar resolutions.

Reading between the lines

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

  • The channel-level design, if it works as described, should transfer to other speckle-dominated imaging modalities such as optical coherence tomography or radar, where pixel-level alignment across frames is unreliable; the paper only evaluates ultrasound.
  • The reported FPS comparison is made with only three input frames, while some video baselines may use longer clips; a fully fair efficiency comparison would equalize the temporal context length.
  • Read literally, Eq. (1) computes cosine similarity between two scalar activations, which yields only $\pm 1$; the empirical gains may therefore come from the pooling-and-gating structure or from an unstated implementation rather than from the printed formula.
  • A direct extension would be to replace the scalar comparison in AFSA with a true per-channel cosine similarity between whole channel maps and measure whether the reported gains persist, grow, or shrink.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript proposes DSANet, a two-stage architecture for ultrasound video lesion and organ segmentation. The first stage, AFSA, computes a channel-level semantic similarity matrix between adjacent frames and uses it to refine features; the second stage, LGSA, splits and reassembles channels from unconditional local and conditional global features. Training applies Dice, weighted BCE, and weighted IoU losses to the target frame and two preceding frames. Experiments compare against 12 image- and video-based methods on ERUS10K, Prostate, CAMUS, and VTUS, reporting MAE, IoU, Dice, and FPS, with ablation studies and a synthetic multiplicative Gamma noise robustness evaluation.

Significance. If the method as described were correct, the channel-level temporal fusion idea would be a plausible efficiency- and robustness-motivated contribution to ultrasound video segmentation, and the paper provides broad comparisons, ablations, and qualitative visualizations. The FPS results and the public code link are useful. However, the core AFSA equation is degenerate as written, uncertainty quantification is absent, and the synthetic noise model is not validated, so the central claims are currently not established.

major comments (3)
  1. [§III-B2, Eq. (1)] As written, after reshaping to C×HW, F_{t-1}[i,j] and F_t[i,j] are scalar activations; the cosine similarity between two nonzero scalars is simply sign(F_{t-1}[i,j])·sign(F_t[i,j]), so the "semantic similarity matrix" S is a sign matrix rather than a graded similarity. Substituting into Eq. (2) gives F'_t = sign(F_{t-1}) element-wise multiplied by |F_t|, which only flips signs and cannot implement the claimed weighting of temporally consistent channels or speckle suppression. If the intended operation is cosine similarity between channel vectors across frames, S should be C×C (or another non-elementwise form) and Eq. (2) would be dimensionally inconsistent with the stated C×HW matrix. Since f_a = F_t_final is the sole input to LGSA, the entire architecture depends on this operation; please provide a corrected, dimensionally consistent formulation or clear pseudocode for the actual implementation.
  2. [§IV-C, Tables I and II] The headline claims of "nearly 1%" IoU/Dice improvements and the ablation gains are reported as point estimates without error bars, confidence intervals, or significance tests from repeated runs or test splits. The small margins, combined with the discrepancy between "Ours" in Table I (IoU 63.3, Dice 76.5 on ERUS10K) and "Full model" in Table II (IoU 63.2, Dice 76.4), make it difficult to distinguish genuine improvement from training or evaluation variability. Please report standard deviations across multiple seeds and, ideally, paired statistical tests for the main comparisons.
  3. [§IV-D2, Table II] The noise-robustness evaluation injects synthetic multiplicative Gamma noise with look numbers L=25 and L=20, but no evidence is provided that this synthetic degradation matches real ultrasound speckle statistics, no sample corrupted images are shown, and no comparison is made against independently acquired noisy ultrasound data. Since noise robustness is a central claim of the paper, the protocol needs validation, at minimum qualitative examples of corrupted frames and a sensitivity analysis over the noise parameters.
minor comments (4)
  1. [Table I] The dataset name is spelled "EKUS10K" in the table header but "ERUS10K" in the text and abstract; please correct this inconsistency.
  2. [§III-D, Eq. (6)] The notation L_tar = L_aux1 = L_aux2 = L_dice + L_wbce + L_iou is confusing because it suggests the three terms are identical before summation; please define L_tar, L_aux1, and L_aux2 separately and specify any weighting coefficients.
  3. [§III-B2, Eq. (3)] The two multiplicative pooling terms, Pooling(F'_{t-1}) and Pooling(F'_t), are not motivated and the pooling type (average or max) is not specified; please clarify the role of each term.
  4. [§IV-B] The experimental protocol is under-specified: the train/validation/test splits for each dataset, the clip sampling strategy, and the metric implementation are not described, which hampers reproducibility of the reported numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DSANet's performance claims rest on external benchmark evaluations and its own ablations, with self-citations used only as motivation.

full rationale

The paper's central claims are empirical rather than derivational: DSANet is trained and evaluated on four external ultrasound-video benchmarks, and the reported IoU, Dice, MAE, and FPS numbers are held-out evaluation results, not quantities fitted to those benchmarks. The noise-robustness study uses fixed look numbers L=20 and L=25 as experimental conditions, not as parameters tuned to produce the claimed outcome. Citations to the authors' own prior work, notably Ref. [9], are used as general motivation for channel-level feature fusion and holistic local-feature preservation, but the load-bearing evidence for the method's effectiveness is provided by the paper's own ablations and comparisons in Tables I and II. The degenerate scalar cosine-similarity computation in Eq. (1) is a serious correctness and reproducibility concern, since it reduces to a sign comparison rather than a graded semantic similarity, but this is not circularity: the operation is a function of its inputs and does not assume the output that it is used to predict. No fitted constant is renamed as a prediction, no derivation is shown to be equivalent to its inputs by construction, and no uniqueness claim is imported from the authors' prior work. Therefore no circular step is identified.

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

The central claim relies on the assumption that channel-level semantics are stable across frames and that the synthetic Gamma noise model represents real speckle; neither is independently established.

free parameters (4)
  • Learning rate = 1e-4
    Adam optimizer hyperparameter chosen by hand; no sensitivity analysis.
  • Input frame count T = 3
    Model uses target frame plus two predecessors; no ablation on longer context.
  • Noise look numbers L = 20 and 25
    Arbitrary settings for synthetic Gamma noise; no justification of clinical relevance.
  • Channel split ratio in LGSA = C/2
    Channels split into two equal halves; no ablation on other ratios.
assumptions (3)
  • domain assumption Channel activations encode stable semantic attributes that are consistent across adjacent frames.
    Stated in Section III-B1 as motivation for channel-level temporal modeling; not directly validated.
  • domain assumption Multiplicative Gamma noise with look numbers 20 and 25 models real ultrasound speckle noise.
    Used in Section IV-D2 for noise robustness evaluation; no calibration against real noisy ultrasound.
  • ad hoc to paper Cosine similarity between scalar activations (Eq. 1) is a valid measure of semantic similarity.
    The formula as written results in sign agreement (±1), which is not a standard semantic similarity; this is a load-bearing assumption of the AFSA module.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual Semantic-Aware Network for Noise Suppressed Ultrasound Video Segmentation." pith.science (2026). https://pith.science/paper/5VYIV3YG

@misc{pith2026250707443,
  author       = {Pith},
  title        = {Pith review of: Dual Semantic-Aware Network for Noise Suppressed Ultrasound Video Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5VYIV3YG}},
  note         = {Machine review of arXiv:2507.07443}
}
read the original abstract

Ultrasound imaging is a prevalent diagnostic tool known for its simplicity and non-invasiveness. However, its inherent characteristics often introduce substantial noise, posing considerable challenges for automated lesion or organ segmentation in ultrasound video sequences. To address these limitations, we propose the Dual Semantic-Aware Network (DSANet), a novel framework designed to enhance noise robustness in ultrasound video segmentation by fostering mutual semantic awareness between local and global features. Specifically, we introduce an Adjacent-Frame Semantic-Aware (AFSA) module, which constructs a channel-wise similarity matrix to guide feature fusion across adjacent frames, effectively mitigating the impact of random noise without relying on pixel-level relationships. Additionally, we propose a Local-and-Global Semantic-Aware (LGSA) module that reorganizes and fuses temporal unconditional local features, which capture spatial details independently at each frame, with conditional global features that incorporate temporal context from adjacent frames. This integration facilitates multi-level semantic representation, significantly improving the model's resilience to noise interference. Extensive evaluations on four benchmark datasets demonstrate that DSANet substantially outperforms state-of-the-art methods in segmentation accuracy. Moreover, since our model avoids pixel-level feature dependencies, it achieves significantly higher inference FPS than video-based methods, and even surpasses some image-based models. Code can be found in \href{https://github.com/ZhouL2001/DSANet}{DSANet}

Figures

Figures reproduced from arXiv: 2507.07443 by the authors.

Figure 1
Figure 1. Ultrasound images are formed from the echoes of coherent acoustic [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison between prior work and our method. (a) Previous methods [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of the proposed Dual Semantic-Aware Network (DSANet). The Adjacent-Frame Semantic-Aware (AFSA) module captures channel-level [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: (a) Heatmap visualization of the results generated by the AFSA module. (b) Heatmap visualization of the results generated by the LGSA module. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of DSANet with video-based SOTA methods. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Visual Comparison of component effectiveness. (a) Frame, (b) Ground [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Visual Comparison between direct global and local-and-global [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 24 canonical work pages

  1. [1]

    T. L. Szabo,Diagnostic ultrasound imaging: inside out. Academic press, 2013

    2013

  2. [2]

    Detection performance theory for ultrasound imaging systems,

    R. J. Zemp, M. D. Parry, C. K. Abbey, and M. F. Insana, “Detection performance theory for ultrasound imaging systems,”IEEE transactions on medical imaging, vol. 24, no. 3, pp. 300–310, 2005

    work page 2005

  3. [3]

    Spatial prediction filtering of acoustic clutter and random noise in medical ultrasound imaging,

    J. Shin and L. Huang, “Spatial prediction filtering of acoustic clutter and random noise in medical ultrasound imaging,”IEEE transactions on medical imaging, vol. 36, no. 2, pp. 396–406, 2016

    work page 2016

  4. [4]

    Shifting more attention to breast lesion segmentation in ultrasound videos,

    J. Lin, Q. Dai, L. Zhu, H. Fu, Q. Wang, W. Li, W. Rao, X. Huang, and L. Wang, “Shifting more attention to breast lesion segmentation in ultrasound videos,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2023, pp. 497–507

    work page 2023

  5. [5]

    Rethinking breast lesion segmentation in ultrasound: a new video dataset and a baseline network,

    J. Li, Q. Zheng, M. Li, P. Liu, Q. Wang, L. Sun, and L. Zhu, “Rethinking breast lesion segmentation in ultrasound: a new video dataset and a baseline network,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2022, pp. 391–400

    work page 2022

  6. [6]

    Lgr- net: Local-global reciprocal network for uterine fibroid segmentation in ultrasound videos,

    H. Xu, Y . Yang, A. I. Aviles-Rivero, G. Yang, J. Qin, and L. Zhu, “Lgr- net: Local-global reciprocal network for uterine fibroid segmentation in ultrasound videos,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2024, pp. 667–677

    work page 2024

  7. [7]

    Towards a benchmark for colorectal cancer segmentation in endorectal ultrasound videos: Dataset and model development,

    Y . Jiang, Y . Hu, Z. Zhang, J. Wei, C.-M. Feng, X. Tang, X. Wan, Y . Liu, S. Cui, and Z. Li, “Towards a benchmark for colorectal cancer segmentation in endorectal ultrasound videos: Dataset and model development,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2024, pp. 732–742

    work page 2024

  8. [8]

    Cross-domain object detection through coarse-to-fine feature adaptation,

    Y . Zheng, D. Huang, S. Liu, and Y . Wang, “Cross-domain object detection through coarse-to-fine feature adaptation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 13 766–13 775

    work page 2020

Show all 33 references
  1. [9]

    Tcgnet: Type-correlation guidance for salient object detection,

    Y . Liu, L. Zhou, G. Wu, S. Xu, and J. Han, “Tcgnet: Type-correlation guidance for salient object detection,”IEEE Transactions on Intelligent Transportation Systems, 2023

    2023

  2. [10]

    Thyroid region prior guided attention for ultrasound segmentation of thyroid nodules,

    H. Gong, J. Chen, G. Chen, H. Li, G. Li, and F. Chen, “Thyroid region prior guided attention for ultrasound segmentation of thyroid nodules,” Computers in biology and medicine, vol. 155, p. 106389, 2023

    2023

  3. [11]

    Res2net: A new multi-scale backbone architecture,

    S.-H. Gao, M.-M. Cheng, K. Zhao, X.-Y . Zhang, M.-H. Yang, and P. Torr, “Res2net: A new multi-scale backbone architecture,”IEEE transactions on pattern analysis and machine intelligence, vol. 43, no. 2, pp. 652–662, 2019

    2019

  4. [12]

    Microsegnet: A deep learning approach for prostate segmentation on micro-ultrasound images,

    H. Jiang, M. Imran, P. Muralidharan, A. Patel, J. Pensa, M. Liang, T. Benidir, J. R. Grajo, J. P. Joseph, R. Terryet al., “Microsegnet: A deep learning approach for prostate segmentation on micro-ultrasound images,”Computerized Medical Imaging and Graphics, vol. 112, p. 102326, 2024

    2024

  5. [13]

    F 3net: fusion, feedback and focus for salient object detection,

    J. Wei, S. Wang, and Q. Huang, “F 3net: fusion, feedback and focus for salient object detection,” inProceedings of the AAAI conference on artificial intelligence, vol. 34, no. 07, 2020, pp. 12 321–12 328

    2020

  6. [14]

    Deep learning for segmentation using an open large-scale dataset in 2d echocardiography,

    S. Leclerc, E. Smistad, J. Pedrosa, A. Østvik, F. Cervenansky, F. Es- pinosa, T. Espeland, E. A. R. Berg, P.-M. Jodoin, T. Grenieret al., “Deep learning for segmentation using an open large-scale dataset in 2d echocardiography,”IEEE transactions on medical imaging, vol. 38, no...

    2019

  7. [15]

    Pranet: Parallel reverse attention network for polyp segmentation,

    D.-P. Fan, G.-P. Ji, T. Zhou, G. Chen, H. Fu, J. Shen, and L. Shao, “Pranet: Parallel reverse attention network for polyp segmentation,” in International conference on medical image computing and computer- assisted intervention. Springer, 2020, pp. 263–273

    2020

  8. [16]

    Progressively normalized self-attention network for video polyp seg- mentation,

    G.-P. Ji, Y .-C. Chou, D.-P. Fan, G. Chen, H. Fu, D. Jha, and L. Shao, “Progressively normalized self-attention network for video polyp seg- mentation,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2021, pp. 142–152

    2021

  9. [17]

    Lesion-aware dynamic kernel for polyp segmentation,

    R. Zhang, P. Lai, X. Wan, D.-J. Fan, F. Gao, X.-J. Wu, and G. Li, “Lesion-aware dynamic kernel for polyp segmentation,” inInternational Conference on Medical Image Computing and Computer-Assisted Inter- vention. Springer, 2022, pp. 99–109

    2022

  10. [18]

    Lssnet: A method for colon polyp segmentation based on local feature supplementation and shallow fea- ture supplementation,

    W. Wang, H. Sun, and X. Wang, “Lssnet: A method for colon polyp segmentation based on local feature supplementation and shallow fea- ture supplementation,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2024, pp. 446–456

    2024

  11. [19]

    Sali: Short- term alignment and long-term interaction network for colonoscopy video polyp segmentation,

    Q. Hu, Z. Yi, Y . Zhou, F. Peng, M. Liu, Q. Li, and Z. Wang, “Sali: Short- term alignment and long-term interaction network for colonoscopy video polyp segmentation,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2024, pp. 531–541

    2024

  12. [20]

    Pyramidal feature shrinking for salient object detection,

    M. Ma, C. Xia, and J. Li, “Pyramidal feature shrinking for salient object detection,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 3, pp. 2311–2318, May 2021

    2021

  13. [21]

    Implicit motion handling for video camouflaged object detection,

    X. Cheng, H. Xiong, D.-P. Fan, Y . Zhong, M. Harandi, T. Drummond, and Z. Ge, “Implicit motion handling for video camouflaged object detection,” inCVPR, 2022

    2022

  14. [22]

    Channel split convolutional neural network (chasnet) for thermal image super-resolution,

    K. Prajapati, V . Chudasama, H. Patel, A. Sarvaiya, K. P. Upla, K. Raja, R. Ramachandra, and C. Busch, “Channel split convolutional neural network (chasnet) for thermal image super-resolution,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

    2021

  15. [23]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international con- ference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. ...

    2015

  16. [24]

    Enhancing row-column array (rca)-based 3d ultrasound vascular imaging with spatial-temporal similarity weighting,

    J. Zhang, C. Huang, U.-W. Lok, Z. Dong, H. Liu, P. Gong, P. Song, and S. Chen, “Enhancing row-column array (rca)-based 3d ultrasound vascular imaging with spatial-temporal similarity weighting,”IEEE Transactions on Medical Imaging, vol. 44, no. 1, pp. 297–309, 2025

    2025

  17. [25]

    Attention-guided learning with feature reconstruction for skin lesion diagnosis using clinical and ultrasound images,

    C. Xiao, A. Zhu, C. Xia, Z. Qiu, Y . Liu, C. Zhao, W. Ren, L. Wang, L. Dong, T. Wang, L. Guo, and B. Lei, “Attention-guided learning with feature reconstruction for skin lesion diagnosis using clinical and ultrasound images,”IEEE Transactions on Medical Imaging, vol. 44, no. 1...

    2025

  18. [26]

    Prior-knowledge embedded u-net-based fully automatic vessel wall volume measurement of the carotid artery in 3d ultrasound image,

    Z. Yue, J. Jiang, W. Hou, Q. Zhou, J. David Spence, A. Fenster, W. Qiu, and M. Ding, “Prior-knowledge embedded u-net-based fully automatic vessel wall volume measurement of the carotid artery in 3d ultrasound image,”IEEE Transactions on Medical Imaging, vol. 44, no. 2, pp. 711...

    2025

  19. [27]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural information processing systems, vol. 30, 2017

    2017

  20. [28]

    Automatic deep learning semantic segmentation of ul- trasound thyroid cineclips using recurrent fully convolutional networks,

    J. M. Webb, D. D. Meixner, S. A. Adusei, E. C. Polley, M. Fatemi, and A. Alizad, “Automatic deep learning semantic segmentation of ul- trasound thyroid cineclips using recurrent fully convolutional networks,” IEEE Access, vol. 9, pp. 5119–5127, 2020

    2020

  21. [29]

    Memsam: taming segment anything model for echocardiography video segmentation,

    X. Deng, H. Wu, R. Zeng, and J. Qin, “Memsam: taming segment anything model for echocardiography video segmentation,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 9622–9631

    2024

  22. [30]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

    2023

  23. [31]

    Vivim: a video vision mamba for ultrasound video segmentation,

    Y . Yang, Z. Xing, L. Yu, H. Fu, C. Huang, and L. Zhu, “Vivim: a video vision mamba for ultrasound video segmentation,”IEEE Transactions on Circuits and Systems for Video Technology, pp. 1–1, 2025

    2025

  24. [32]

    Sam 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R¨adle, C. Rolland, L. Gustafsonet al., “Sam 2: Segment anything in images and videos,”arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  25. [33]

    Medical sam 2: Segment medical images as video via segment anything model 2,

    J. Zhu, A. Hamdi, Y . Qi, Y . Jin, and J. Wu, “Medical sam 2: Segment medical images as video via segment anything model 2,”arXiv preprint arXiv:2408.00874, 2024

    2024 arXiv

Pith tools

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