Pith. sign in

REVIEW 2 major objections 1 minor 59 references

Specificity- and Calibration-Aware Breast Ultrasound Segmentation via Entropy-Guided Boundary Supervision

T0 review · 2 major / 1 minor · reviewed 2026-06-26 · grok-4.3

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

Pith's one-line read Entropy scaling of boundary penalties reduces false positives on lesion-free breast ultrasound images while keeping lesion segmentation accuracy unchanged.

desk verdict The entropy-weighted boundary loss delivers a statistically backed drop in false positives on BUSI's 20 no-lesion images while holding Dice steady, but the tiny single-dataset no-lesion set leaves generalization unproven. read the letter →

arxiv 2606.22308 v1 pith:J43LWMGK submitted 2026-06-21 eess.IV cs.CVcs.LG

Manar Alsaid , Mandip Shrestha , Mohammad Abbas This is my paper
classification eess.IVcs.CVcs.LG
keywords breastultrasoundlesionsegmentationboundarysupervisionentropyweightingfalsepositivesspecificitycalibrationU-Net
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 modifies the training loss so that boundary penalties are scaled by per-pixel predictive entropy and the ground-truth boundary map. This change concentrates gradient updates on uncertain lesion margins. On the BUSI test set the method produces statistically equivalent Dice scores on 97 lesion images yet lowers the number of false-positive activations on 20 no-lesion images from 14–19 down to 5. A later spatial temperature scaling step further lowers expected calibration error. The work therefore shows that specificity and calibration can be improved inside the training objective of a standard U-Net without sacrificing contour quality.

What carries the argument

Entropy-guided boundary supervision loss that multiplies the contour penalty term by per-pixel predictive entropy and the ground-truth boundary map.

What would settle it

Re-evaluate the same method on an independent multi-center breast ultrasound collection and test whether the false-positive count on no-lesion cases remains significantly lower than the two baselines.

Watch

Extended reading notes

Core claim

Rather than weighting every boundary pixel equally, the proposed loss scales contour penalties by per-pixel predictive entropy and the ground-truth boundary map, concentrating gradient emphasis on lesion margin locations where the network remains uncertain. This single change suppresses false-positive activations in no-lesion images while preserving segmentation quality on lesion-containing images.

Load-bearing premise

The no-lesion images and training setup in the BUSI dataset represent the false-positive failure modes that occur in real clinical breast ultrasound workflows.

Editorial extensions

If this is right

  • Dice scores on lesion images stay statistically indistinguishable from the no-boundary baseline.
  • False-positive activations on no-lesion images drop from 14–19 cases to 5 cases out of 20.
  • Post-hoc spatial temperature scaling can reduce expected calibration error from 0.0201 to 0.0095 without changing the segmentation masks.
  • The entropy weighting and boundary supervision operate as complementary training-level and inference-level refinements inside a U-Net.

Reading between the lines

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

  • The same loss modification could be tested on other ultrasound modalities where speckle and shadowing produce similar false positives.
  • If the reduction holds, clinical pipelines might rely less on separate false-positive filtering stages.
  • The approach leaves open whether entropy weighting needs dataset-specific retuning when scanner hardware or acquisition protocols change.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper proposes modifying the training loss for U-Net breast ultrasound segmentation by scaling boundary penalties with per-pixel predictive entropy and the ground-truth boundary map. This is claimed to reduce false-positive activations on no-lesion images while preserving Dice on lesion images. On the BUSI dataset, Dice is statistically equivalent (0.7624 vs. 0.7616, Wilcoxon p=0.27) across 97 lesion test images, while false-positive images drop to 5/20 from 14/20 and 19/20 (McNemar p=0.012, 0.0005) on 20 no-lesion images; a post-hoc spatial temperature scaling further lowers expected calibration error from 0.0201 to 0.0095.

Significance. If the entropy-guided weighting generalizes, the method provides a targeted way to improve specificity and calibration in BUS segmentation without Dice degradation, using appropriate statistical tests (Wilcoxon, McNemar, Wilson intervals) on held-out data. This addresses a practical clinical failure mode and could be complementary to existing U-Net pipelines.

major comments (2)
  1. [Evaluation / Results (no-lesion subset)] The specificity claim rests on false-positive reduction observed in only 20 no-lesion BUSI images. This small hold-out and single-dataset evaluation (with no cross-scanner, multi-center, or sensitivity analysis on entropy threshold/spatial scaling) is load-bearing for the central claim that the approach addresses real-world false-positive modes; the lesion Dice equivalence offers no safeguard if artifact statistics shift.
  2. [Methods] Exact loss formulation, data splits, ablation details, and hyper-parameter choices for the entropy weighting are not fully specified in the provided description, preventing verification that the reported gains are not due to post-hoc tuning or limited generalization testing.
minor comments (1)
  1. [Abstract / Calibration step] The post-hoc spatial temperature scaling is presented as complementary but its interaction with the training loss should be clarified to avoid implying it is part of the core contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We respond to each major comment below, indicating where revisions will be made to improve clarity and address concerns about evaluation scope and reproducibility.

read point-by-point responses

  1. Referee: [Evaluation / Results (no-lesion subset)] The specificity claim rests on false-positive reduction observed in only 20 no-lesion BUSI images. This small hold-out and single-dataset evaluation (with no cross-scanner, multi-center, or sensitivity analysis on entropy threshold/spatial scaling) is load-bearing for the central claim that the approach addresses real-world false-positive modes; the lesion Dice equivalence offers no safeguard if artifact statistics shift.

    Authors: We acknowledge the limited size of the no-lesion subset (n=20) in BUSI, which is the full available test split for this class in the standard benchmark. The observed reduction (14/20 and 19/20 to 5/20) is supported by McNemar tests (p=0.012, 0.0005) and non-overlapping Wilson intervals, indicating a statistically and practically meaningful effect within this dataset. In revision we will add an explicit limitations paragraph discussing single-dataset constraints and the value of future multi-center studies. We will also include a sensitivity analysis varying the entropy threshold and spatial scaling factor to assess robustness. Cross-scanner validation is noted as future work as it requires new data. revision: partial

  2. Referee: [Methods] Exact loss formulation, data splits, ablation details, and hyper-parameter choices for the entropy weighting are not fully specified in the provided description, preventing verification that the reported gains are not due to post-hoc tuning or limited generalization testing.

    Authors: We agree that additional methodological detail is needed for full reproducibility. The revised manuscript will include the exact loss formulation (the per-pixel entropy-weighted boundary term multiplied by the ground-truth boundary map), the precise train/validation/test splits on BUSI, complete ablation tables comparing all variants, and the specific hyper-parameter values used for entropy scaling and temperature. These additions will allow independent verification that gains are not attributable to post-hoc choices. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity; empirical claims rest on held-out test evaluation

full rationale

The paper introduces an entropy-weighted boundary loss term and reports Dice preservation plus specificity gains via direct ablation on the BUSI held-out sets (97 lesion images, 20 no-lesion images). All quantitative results are computed from model outputs on unseen data using standard metrics and statistical tests; no equations reduce a claimed prediction to a fitted input by construction, no self-citations carry load-bearing uniqueness arguments, and no ansatz or renaming is smuggled in. The derivation chain is therefore self-contained against external benchmarks.

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

The central claim rests on standard deep-learning assumptions (U-Net suitability for segmentation, BUSI as a valid benchmark) plus the unstated premise that entropy from a trained model provides a reliable signal for boundary importance; no free parameters or invented entities are introduced in the abstract.

assumptions (2)
  • domain assumption U-Net architecture is an appropriate base model for the task
    The method is evaluated inside a U-Net framework without justification of alternatives.
  • domain assumption BUSI dataset splits and no-lesion images are representative of clinical false-positive scenarios
    Evaluation relies on this dataset without external validation cohorts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Specificity- and Calibration-Aware Breast Ultrasound Segmentation via Entropy-Guided Boundary Supervision." pith.science (2026). https://pith.science/paper/J43LWMGK

@misc{pith2026260622308,
  author       = {Pith},
  title        = {Pith review of: Specificity- and Calibration-Aware Breast Ultrasound Segmentation via Entropy-Guided Boundary Supervision},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J43LWMGK}},
  note         = {Machine review of arXiv:2606.22308}
}
read the original abstract

Lesion segmentation in breast ultrasound involves two related challenges. In images with lesions, speckle noise, low tissue contrast, and posterior acoustic shadowing cause boundary leakage and incomplete contour delineation. In images without lesions, those same artifacts generate false-positive activations in regions resembling solid lesion tissue. This study addresses both failure modes through a single modification to the training objective. Rather than weighting every boundary pixel equally, the proposed loss scales contour penalties by per-pixel predictive entropy and the ground-truth boundary map, concentrating gradient emphasis on lesion margin locations where the network remains uncertain. The loss was evaluated on the BUSI dataset through a controlled ablation against two baselines: a model without boundary supervision and a model with uniformly weighted boundary binary cross-entropy. Across 97 lesion-containing test images, mean Dice scores were statistically indistinguishable between the proposed method and the no-boundary baseline (0.7624 versus 0.7616, paired Wilcoxon p = 0.27), confirming that lesion segmentation quality is preserved. The primary effect appears in specificity. False-positive activations on 20 no-lesion test images fell from 14 of 20 and 19 of 20 for the two baselines to 5 of 20 with the proposed approach (McNemar p = 0.012 and 0.0005). Non-overlapping Wilson 95% confidence intervals confirm the difference is both statistically significant and practically substantial. A post-hoc spatial temperature scaling step further reduced expected calibration error from 0.0201 to 0.0095 without altering segmentation masks. Entropy-guided boundary supervision and spatial calibration thus function as complementary training-level and inference-level refinements that improve specificity and probability reliability within a U-Net framework.

Figures

Figures reproduced from arXiv: 2606.22308 by the authors.

Figure 1
Figure 1. Overview of the proposed framework. A breast ultrasound image is [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Sorted per-case Dice distributions for the BUSI test split (n = 117). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 5
Figure 5. Qualitative comparison of the three training configurations across four representative test cases. Columns from left to right: input image, ground [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Foreground reliability diagrams for the proposed entropy-guided boundary model under three post-hoc calibration settings: uncalibrated (ECE 0.0201), [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 4 canonical work pages

  1. [1]

    Abdellaoui, S

    C. Abdellaoui, S. Belkacem, and N. Messaoudi, ”Deep learning ar- chitectures for medical image segmentation: an organized analysis of CNN-based models and uses,” Bulletin of Electrical Engineering and Informatics, vol. 15, no. 1, pp. 424-437, 2026

    2026

  2. [2]

    Thomas, M

    C. Thomas, M. Byra, R. Marti, M. H. Yap, and R. Zwiggelaar, ”BUS- Set: A benchmark for quantitative evaluation of breast ultrasound segmentation networks with public datasets,” Medical Physics, vol. 50, no. 5, pp. 3223-3243, 2023

    2023

  3. [3]

    Al-Dhabyani, M

    W. Al-Dhabyani, M. Gomaa, H. Khaled, and A. Fahmy, ”Dataset of breast ultrasound images,” Data in Brief, vol. 28, p. 104863, 2020

    2020

  4. [4]

    X. Xiao, J. Zhang, Y . Shao, et al., ”Deep learning-based medical ultrasound image and video segmentation methods: overview, frontiers, and challenges,” Sensors, vol. 25, no. 8, p. 2361, 2025

    2025

  5. [5]

    Ronneberger, P

    O. Ronneberger, P. Fischer, and T. Brox, ”U-Net: Convolutional net- works for biomedical image segmentation,” in MICCAI, 2015, pp. 234- 241

    2015

  6. [6]

    Isensee, P

    F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Petersen, and K. H. Maier-Hein, ”nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,” Nature Methods, vol. 18, no. 2, pp. 203-211, 2021

    2021

  7. [7]

    Sulaiman et al., ”Attention based U-Net model for breast cancer segmentation using BUSI dataset,” Scientific Reports, vol

    A. Sulaiman et al., ”Attention based U-Net model for breast cancer segmentation using BUSI dataset,” Scientific Reports, vol. 14, p. 22422, 2024

    2024

  8. [8]

    Yang et al., ”Multilevel perception boundary-guided network for breast lesion segmentation in ultrasound images,” Medical Physics, 2025

    X. Yang et al., ”Multilevel perception boundary-guided network for breast lesion segmentation in ultrasound images,” Medical Physics, 2025

    2025

Show all 59 references
  1. [9]

    Kervadec, J

    H. Kervadec, J. Bouchtiba, C. Desrosiers, E. Granger, J. Dolz, and I. Ben Ayed, ”Boundary loss for highly unbalanced segmentation,” Medical Image Analysis, vol. 67, p. 101851, 2021

    2021

  2. [10]

    F. Sun, Z. Luo, and S. Li, ”Boundary difference over union loss for medical image segmentation,” in MICCAI, 2023, pp. 292-301

    2023

  3. [11]

    Xu et al., ”Boundary guidance network for medical image segmen- tation,” Scientific Reports, vol

    R. Xu et al., ”Boundary guidance network for medical image segmen- tation,” Scientific Reports, vol. 14, p. 17345, 2024

    2024

  4. [12]

    Kendall and Y

    A. Kendall and Y . Gal, ”What uncertainties do we need in Bayesian deep learning for computer vision?” in NeurIPS, 2017, pp. 5574-5584

    2017

  5. [13]

    Scalco et al., ”Uncertainty quantification in multi-class segmentation: comparison between Bayesian and non-Bayesian approaches in a clinical perspective,” Medical Physics, vol

    E. Scalco et al., ”Uncertainty quantification in multi-class segmentation: comparison between Bayesian and non-Bayesian approaches in a clinical perspective,” Medical Physics, vol. 51, no. 8, pp. 5460-5474, 2024

    2024

  6. [14]

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, ”On calibration of modern neural networks,” in ICML, 2017, pp. 1321-1330

    2017

  7. [15]

    Z. Ding, X. Han, P. Liu, and M. Niethammer, ”Local temperature scaling for probability calibration,” in ICCV , 2021, pp. 6889-6899

    2021

  8. [16]

    Vallez, G

    N. Vallez, G. Bueno, O. Deniz, M. A. Rienda, and C. Pastor, ”BUS- UCLM: Breast ultrasound lesion segmentation dataset,” Scientific Data, vol. 12, no. 242, 2025

    2025

  9. [17]

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, ”UNet++: A nested U-Net architecture for medical image segmentation,” in DLMIA/MICCAI Workshops, LNCS 11045, 2018, pp. 3-11

    2018

  10. [18]

    Oktay et al., ”Attention U-Net: Learning where to look for the pancreas,” arXiv:1804.03999, 2018

    O. Oktay et al., ”Attention U-Net: Learning where to look for the pancreas,” arXiv:1804.03999, 2018

    2018 arXiv

  11. [19]

    Chen et al., ”TransUNet: Transformers make strong encoders for medical image segmentation,” arXiv:2102.04306, 2021

    J. Chen et al., ”TransUNet: Transformers make strong encoders for medical image segmentation,” arXiv:2102.04306, 2021

    2021 arXiv

  12. [20]

    Cao et al., ”Swin-UNet: Unet-like pure transformer for medical image segmentation,” in ECCV Workshops, LNCS 13803, 2023, pp

    H. Cao et al., ”Swin-UNet: Unet-like pure transformer for medical image segmentation,” in ECCV Workshops, LNCS 13803, 2023, pp. 205-218

    2023

  13. [21]

    Abboud, H

    Z. Abboud, H. Lombaert, and S. Kadoury, ”Sparse Bayesian networks: efficient uncertainty quantification in medical image analysis,” in MIC- CAI, 2024

    2024

  14. [22]

    Xie et al., ”Entropy-guided contrastive learning for semi-supervised medical image segmentation,” IET Image Processing, 2024

    X. Xie et al., ”Entropy-guided contrastive learning for semi-supervised medical image segmentation,” IET Image Processing, 2024

    2024

  15. [23]

    M. H. Yap, G. Pons, J. Marti, S. Ganau, M. Sentis, R. Zwiggelaar, A. K. Davison, and R. Marti, ”Automated breast ultrasound lesions detection using convolutional neural networks,” IEEE J. Biomed. Health Inform., vol. 22, no. 4, pp. 1218-1226, 2018

    2018

  16. [24]

    Gomez-Flores, M

    W. Gomez-Flores, M. J. Gregorio-Calas, and W. Coelho de Albuquerque Pereira, ”BUS-BRA: A breast ultrasound dataset for assessing computer- aided diagnosis systems,” Medical Physics, vol. 51, no. 4, pp. 3110-3123, 2024

    2024

  17. [25]

    J. Ma, J. Chen, M. Ng, R. Huang, Y . Li, C. Li, X. Yang, and A. L. Martel, ”Loss odyssey in medical image segmentation,” Medical Image Analysis, vol. 71, p. 102035, 2021

    2021

  18. [26]

    Karimi and S

    D. Karimi and S. E. Salcudean, ”Reducing the Hausdorff distance in medical image segmentation with convolutional neural networks,” IEEE Trans. Med. Imaging, vol. 39, no. 2, pp. 499-513, 2020

    2020

  19. [27]

    Jungo, F

    A. Jungo, F. Balsiger, and M. Reyes, ”Analyzing the quality and challenges of uncertainty estimations for brain tumor segmentation,” Frontiers in Neuroscience, vol. 14, p. 282, 2020

    2020

  20. [28]

    Mehrtash, W

    A. Mehrtash, W. M. Wells, C. M. Tempany, P. Abolmaesumi, and T. Kapur, ”Confidence calibration and predictive uncertainty estimation for deep medical image segmentation,” IEEE Trans. Med. Imaging, vol. 39, no. 12, pp. 3868-3878, 2020

    2020

  21. [29]

    K. He, X. Zhang, S. Ren, and J. Sun, ”Deep residual learning for image recognition,” in CVPR, 2016, pp. 770-778

    2016

  22. [30]

    Loshchilov and F

    I. Loshchilov and F. Hutter, ”Decoupled weight decay regularization,” in ICLR, 2019

    2019

  23. [31]

    Gal and Z

    Y . Gal and Z. Ghahramani, ”Dropout as a Bayesian approximation: Representing model uncertainty in deep learning,” in ICML, 2016, pp. 1050-1059

    2016

  24. [32]

    Salehi, A

    A. Salehi, A. Erdogmus, and A. Gholipour, ”Tversky loss function for image segmentation using 3D fully convolutional deep networks,” in MLMI/MICCAI, LNCS 10541, 2017, pp. 379-387

    2017

  25. [33]

    A. L. Simpson et al., ”A large annotated medical image dataset for the development and evaluation of segmentation algorithms,” arXiv:1902.09063, 2019

    1902 arXiv

  26. [34]

    G. W. Brier, ”Verification of forecasts expressed in terms of probability,” Monthly Weather Review, vol. 78, no. 1, pp. 1-3, 1950

    1950

  27. [35]

    Rohlfing, ”Image similarity and tissue overlaps as surrogates for image registration accuracy: widely used but unreliable,” IEEE Trans

    T. Rohlfing, ”Image similarity and tissue overlaps as surrogates for image registration accuracy: widely used but unreliable,” IEEE Trans. Med. Imaging, vol. 31, no. 2, pp. 153-163, 2012

    2012

  28. [36]

    Al-Dhabyani, M

    W. Al-Dhabyani, M. Gomaa, H. Khaled, and A. Fahmy, ”Deep learning approaches for data augmentation and classification of breast masses using ultrasound images,” International Journal of Advanced Computer Science and Applications, vol. 10, no. 5, pp. 618–627, 2019, doi: 10.1456...

    2019 doi

  29. [37]

    A. T. Stavros, C. Thickman, C. L. Rapp, M. A. Dennis, S. H. Parker, and G. A. Sisney, ”Solid breast nodules: use of sonography to distinguish between benign and malignant lesions,” Radiology, vol. 196, no. 1, pp. 123-134, 1995

    1995

  30. [38]

    C. M. Rumack, S. R. Wilson, J. W. Charboneau, and D. Levine, Diagnostic Ultrasound, 4th ed. Philadelphia, PA: Elsevier Mosby, 2011

    2011

  31. [39]

    Wilcoxon, ”Individual comparisons by ranking methods,” Biometrics Bulletin, vol

    F. Wilcoxon, ”Individual comparisons by ranking methods,” Biometrics Bulletin, vol. 1, no. 6, pp. 80-83, 1945

    1945

  32. [40]

    McNemar, ”Note on the sampling error of the difference between correlated proportions or percentages,” Psychometrika, vol

    Q. McNemar, ”Note on the sampling error of the difference between correlated proportions or percentages,” Psychometrika, vol. 12, no. 2, pp. 153-157, 1947

    1947

  33. [41]

    E. B. Wilson, ”Probable inference, the law of succession, and statistical inference,” Journal of the American Statistical Association, vol. 22, no. 158, pp. 209-212, 1927

    1927

  34. [42]

    Cohen, Statistical Power Analysis for the Behavioral Sciences, 2nd ed

    J. Cohen, Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates, 1988

    1988

  35. [43]

    Shrivastava, A

    A. Shrivastava, A. Gupta, and R. Girshick, ”Training region-based object detectors with online hard example mining,” in CVPR, 2016, pp. 761- 769

    2016

  36. [44]

    T. Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, ”Focal loss for dense object detection,” in ICCV , 2017, pp. 2980-2988

    2017

  37. [45]

    E. A. Krupinski, ”Current perspectives in medical image perception,” Attention, Perception, and Psychophysics, vol. 72, no. 5, pp. 1205-1217, 2010

    2010

  38. [46]

    Tajbakhsh et al., ”Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation,” Medical Image Analysis, vol

    N. Tajbakhsh et al., ”Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation,” Medical Image Analysis, vol. 63, p. 101693, 2020

    2020

  39. [47]

    Bouthillier, C

    P. Bouthillier, C. Laurent, and P. Vincent, ”Unreproducible research is reproducible,” in ICML, 2019, pp. 725-734

    2019

  40. [48]

    W. A. Berg et al., ”Diagnostic accuracy of mammography, clinical examination, US, and MR imaging in preoperative assessment of breast cancer,” Radiology, vol. 233, no. 3, pp. 830-849, 2004

    2004

  41. [49]

    Dodge and N

    J. Dodge and N. Gane, ”Show your work: Improved reporting of experimental results,” in EMNLP, 2019, pp. 2185-2194

    2019

  42. [50]

    Fornasa, ”Ultrasound-based breast lesion segmentation for surgical planning,” in Medical Imaging: Image Processing, SPIE, 2019

    M. Fornasa, ”Ultrasound-based breast lesion segmentation for surgical planning,” in Medical Imaging: Image Processing, SPIE, 2019

    2019

  43. [51]

    V . Buda, M. Maki, and M. A. Mazurowski, ”A systematic study of the class imbalance problem in convolutional neural networks,” Neural Networks, vol. 106, pp. 249-259, 2018

    2018

  44. [52]

    P. Luc, C. Couprie, S. Chintala, and J. Verbeek, ”Semantic segmentation using adversarial networks,” in NIPS Workshop on Adversarial Training, 2016

    2016

  45. [53]

    S. Meng, H. Zhang, C. Li, and Y . Zheng, ”Acoustic shadow detection in ultrasound images using deep learning,” in IEEE International Ultra- sonics Symposium, 2020

    2020

  46. [54]

    T. F. Cootes, C. J. Taylor, D. H. Cooper, and J. Graham, ”Active shape models – their training and application,” Computer Vision and Image Understanding, vol. 61, no. 1, pp. 38-59, 1995

    1995

  47. [55]

    Avendi, A

    S. Avendi, A. Kheradvar, and H. Jafarkhani, ”A combined deep-learning and deformable-model approach to fully automatic segmentation of the left ventricle in cardiac MRI,” Medical Image Analysis, vol. 30, pp. 108-119, 2016

    2016

  48. [56]

    S. Liu, D. Johns, and A. Davison, ”End-to-end multi-task learning with attention,” in CVPR, 2019, pp. 1871-1880

    2019

  49. [57]

    Ovadia et al., ”Can you trust your model’s uncertainty? Evaluating predictive uncertainty under dataset shift,” in NeurIPS, 2019, pp

    S. Ovadia et al., ”Can you trust your model’s uncertainty? Evaluating predictive uncertainty under dataset shift,” in NeurIPS, 2019, pp. 13991- 14002

    2019

  50. [58]

    Ranschaert, S

    R. Ranschaert, S. Morozov, and P. Algra, Artificial Intelligence in Medical Imaging. Cham, Switzerland: Springer, 2019

    2019

  51. [59]

    Food and Drug Administration, ”Artificial intelligence and machine learning in software as a medical device,” FDA Discussion Paper, 2021

    U.S. Food and Drug Administration, ”Artificial intelligence and machine learning in software as a medical device,” FDA Discussion Paper, 2021. [Online]. Available: https://www.fda.gov/media/145022/download

    2021

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.