Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Coarse lesion masks can spatially ground concept-bottleneck explanations in breast ultrasound, improving both diagnosis and the anatomical plausibility of concept evidence.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Using lesion masks to ground concept activations in anatomically relevant zones improves both diagnosis and spatial faithfulness of concept bottleneck models on breast ultrasound.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Useful methods paper, but the spatial-faithfulness claim is mostly self-measuring: the headline metric is the training loss. the 3 major comments →

arxiv 2607.20691 v1 pith:HX4K6ON2 submitted 2026-07-22 cs.CV cs.AI

Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis

Moshiur Rahman Tonmoy , Dunren Che , Haitham Y. Adarbah , Afzel Noore This is my paper
classification cs.CV cs.AI
keywords concept bottleneck modelsspatial groundingbreast ultrasoundweak supervisionBI-RADSexplainable AIdata qualitytrustworthiness
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 aims to show that concept-bottleneck models — which explain a diagnosis through human-readable clinical concepts — can be made spatially trustworthy without expensive pixel-level concept annotations. It does this by deriving two clinically motivated zones from coarse lesion masks (the lesion region itself and a band immediately beneath it for posterior acoustic effects) and training the concept maps to concentrate their evidence inside their assigned zone. On breast ultrasound, this spatially grounded CBM reportedly improves diagnostic AUROC from 0.869 to 0.892, concept macro-AUROC from 0.741 to 0.771, and the fraction of concept evidence inside the lesion region from 0.149 to 0.815. A sympathetic reader would care because it offers a data-centric, low-cost mechanism for making medical AI explanations align with how radiologists actually reason.

Core claim

The paper claims that SG-CBM, a concept-bottleneck model with a grouped weak spatial supervision objective, achieves simultaneous gains in diagnosis, concept prediction, and spatial faithfulness on breast ultrasound. Compared to an ungrounded CBM, SG-CBM raises diagnostic AUROC (0.869 → 0.892), concept macro-AUROC (0.741 → 0.771), and the Energy-in-Zone grounding metric (ROI: 0.149 → 0.815; Hit@1: 0.915). These gains come from coarse lesion masks alone, requiring only two zone definitions per image, and hold across five-fold stratified group cross-validation.

What carries the argument

The grounding objective is built from two zone masks derived from each lesion mask: an in-lesion ROI (morphology concepts) and a size-adaptive posterior band (posterior acoustic concepts). Concept activations are pushed into their assigned zones via a separation loss (in-zone vs out-of-zone mean activation, with a 10% margin) and a mass-concentration loss (fraction of activation mass inside the zone), combined with group weights so each concept contributes comparably.

Load-bearing premise

The framework assumes both that reliable lesion masks are available at training time and that posterior acoustic evidence reliably lives in a fixed-size band immediately beneath the lesion; if either fails, the grounding loss can penalize correct concept activations and the grounding metrics overstate faithfulness.

What would settle it

Measure posterior-concept AUROC on a held-out set of malignant cases where radiologists confirm posterior shadowing appears outside the defined band or is absent; if SG-CBM's posterior grounding loss degrades those concepts relative to vanilla CBM, the band assumption is wrong.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Spatial grounding can serve as a replacement for expensive pixel-level concept annotation in practice.
  • Evidence-alignment constraints regularize concept representations, improving rather than sacrificing diagnostic performance.
  • Grounding metrics make concept explanations auditable both semantically and spatially.
  • The Train-corrupt/Test-clean stress test provides a methodology for quantifying how annotation quality affects trustworthiness.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same zone-based mechanism could be transferred to other imaging tasks with available coarse anatomical priors, such as thyroid nodules or lung lesions, by replacing BI-RADS zones with the relevant anatomy.
  • The mild-erosion improvement hints that zone tightness itself is a tunable regularizer; a systematic sweep over dilation sizes could reveal an optimal supervision tightness independent of mask correctness.
  • Treating the grounding metrics as an online quality signal could let a deployment pipeline flag cases where automatic segmentation quality is low and route them to human review.
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

3 major / 5 minor

Summary. The paper proposes SG-CBM, a spatially grounded Concept Bottleneck Model for breast ultrasound diagnosis. A lesion mask is used to derive two zones: an in-lesion ROI for morphology-related concepts and a size-adaptive posterior band for posterior acoustic concepts. Concept maps are trained with a grouped grounding loss (separation plus mass concentration) while diagnosis is mediated by a linear classifier over concept logits. Experiments on the BrEaST dataset with 5-fold stratified group cross-validation report that SG-CBM improves diagnostic AUROC and concept macro-AUROC over a vanilla CBM, and markedly increases zone-based grounding metrics (ROI Energy 0.149→0.815). A Train-corrupt/Test-clean stress test is used to study sensitivity to mask quality.

Significance. If the spatial-faithfulness claim were supported by an independent evaluation, the paper would make a useful data-centric contribution: it shows how coarse lesion delineations can provide weak spatial supervision for concept localization without pixel-level concept annotations. The Train-corrupt/Test-clean stress test is a sensible methodology for studying annotation-quality effects. However, the central quantitative claim of 'markedly increasing spatial alignment' rests on metrics that are essentially the training objective (Eq. 8 vs. Eq. 4), so the reported gains do not yet demonstrate anatomical plausibility or clinical faithfulness. The diagnostic and concept-quality gains are modest and lack significance testing. The authors partially acknowledge the bias of zone-based evaluation in the Discussion, but the abstract and Section III present the grounding results as definitive evidence.

major comments (3)
  1. [III-A, Eq. (8) vs. II-B, Eq. (4)] The primary spatial-grounding metric, Energy-in-Zone (Eq. 8), is exactly 1 - L_mass from Eq. (4), the very term minimized during training. Hit@1 and Top-5% overlap are thresholded variants of the same spatial-concentration objective (Eq. 3). Therefore the improvement from 0.149 to 0.815 in ROI Energy largely reflects the model being optimized for that quantity, not an independent measure of anatomical plausibility. The stress test in Table II uses clean test zones but still evaluates the same objective; it does not break this circularity. The authors should either obtain independent spatial-grounding evidence (e.g., pixel-level concept annotations, radiologist localization judgments, or an out-of-distribution dataset with such annotations) or explicitly re-scope the claims from 'anatomical faithfulness' to 'conformity with the imposed zone prior.'
  2. [Table I, Section III-B(a)] The diagnostic improvement over vanilla CBM is small and within overlapping standard deviations: AUROC 0.892±0.012 vs. 0.869±0.040; concept macro-AUROC improves only 0.741→0.771. No significance test is reported. Given that the paper's abstract claims SG-CBM 'improves diagnostic AUROC and concept macro-AUROC,' paired tests (e.g., DeLong or a paired permutation test across folds) with confidence intervals should be provided. Without them, the predictive and semantic gains are not statistically supported.
  3. [II-A, Posterior Band] The paper states in Section II-A that posterior acoustic features 'are not consistently present or visually salient in all malignant ultrasound scans, but for simplicity, we assume that a posterior band may be informative across cases.' This assumption is load-bearing for the posterior grounding claims: if the band does not contain the true acoustic phenomenon, or if the phenomenon is absent, L_grounding penalizes correct activations and the posterior grounding metrics overstate spatial faithfulness. The authors should report posterior grounding metrics stratified by the presence/absence of posterior enhancement/shadowing (using the BI-RADS labels already available), and ideally analyze failure cases where the posterior band is empty or too tight.
minor comments (5)
  1. [Discussion, last paragraph] The Discussion correctly notes that predefined zones 'may bias evaluation' and that zone alignment 'should be interpreted alongside, rather than as a substitute for, semantic concept correctness.' This caveat should appear prominently in the abstract and at the first presentation of the grounding results, not only in the final discussion.
  2. [II-A, Eq. (1)] The hyperparameters alpha=1.0, H_min=10, H_max=50 are chosen heuristically. A sentence on sensitivity to these parameters (or a small ablation) would improve reproducibility; currently no sensitivity analysis is reported.
  3. [Table II] The first row is labeled with a single dash under 'K'; it should be labeled explicitly (e.g., 'none' or 'clean') for readability. Also, K is already used for the number of concepts; using K again for the erosion kernel is confusing.
  4. [Fig. 2] The caption refers to 'the rightmost 2 panels' but the figure appears to show several concept summaries; the panel layout should be described more precisely so the qualitative grounding comparison can be verified.
  5. [General] No code availability statement or trained-model release is provided. Given the paper's emphasis on reproducibility and data-centric validation, sharing code and the exact zone-generation implementation would strengthen the contribution.

Circularity Check

1 steps flagged

Spatial-faithfulness result is by construction: Energy-in-Zone is the mass-concentration training loss, so the large grounding gains are not an independent validation.

specific steps
  1. fitted input called prediction [Section III-A Eq. (8) vs Section II-B Eqs. (4)-(7)]
    "L_mass = 1/|K| Σ_k (1 − Σ A_k Z / Σ A_k + ε). ... Energy(k) = Σ A_k Z / Σ A_k + ε ... higher values indicate that the model’s spatial evidence for concept k is concentrated in the expected anatomical region ... SG-CBM substantially increases activation concentration within target zones, achieving ROI Energy 0.815 and Hit@1 0.915."

    Energy-in-Zone (Eq. 8) is exactly 1 − L_mass per concept, and L_mass is minimized with λ_s=2 in the final objective L = λ_c L_concept + λ_y L_cls + λ_s L_grounding (Eq. 7). The paper also states: "we treat off-zone activation as an evidence-quality violation and penalize it through separation and mass concentration losses." Thus the reported increase in ROI Energy (0.149→0.815) and posterior Energy (0.099→0.585) is the direct optimization target, not an independent measure of anatomical plausibility. No radiologist-annotated concept localization or other external spatial-grounding ground truth is provided. The claimed "markedly increasing spatial alignment" is therefore substantially by construction.

full rationale

The only strong effect SG-CBM reports beyond the vanilla CBM is spatial alignment; diagnostic AUROC improves modestly (0.869±0.040 → 0.892±0.012, overlapping stds, no significance test) and concept AUROC improves 0.741→0.771. The headline spatial-faithfulness result reduces to the training objective: Eq. 8 is the complement of Eq. 4, and the model is explicitly optimized to maximize activation mass inside the same zones. Hit@1 and Top-5% Overlap are additional views of the same concentrated-activation objective. The paper itself concedes the limitation: in Discussion, "this design inherently encodes spatial priors that may bias evaluation... activation within a zone indicates spatial alignment of evidence and should be interpreted alongside, rather than as a substitute for, semantic concept correctness [13]"; and in Table II it uses test-clean masks "to avoid self-consistency bias in zone-based grounding metrics." These admissions corroborate that the grounding metrics are self-consistent with training rather than independent evidence. The posterior-band heuristic (Sec. II-A) is likewise an imposed assumption; the paper acknowledges "we rely on heuristic design and do not explicitly incorporate ultrasound physics," so posterior metrics measure conformity to that assumption, not clinical truth. No load-bearing self-citation chain or imported uniqueness theorem is involved. Score reflects that the central spatial-alignment claim is by construction while diagnostic and concept AUROC remain partially independent.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced. The main design constructs are the posterior band and the zone grouping, which are assumptions/parameters rather than independent entities. The free parameters are the manually chosen zone and loss hyperparameters, and the axioms are the anatomical priors and metric validity assumptions listed above.

free parameters (2)
  • posterior band height parameters (α=1.0, H_min=10, H_max=50) = α=1.0, H_min=10, H_max=50
    Defines the size-adaptive posterior band in Eq. (1); hand-set without ablation. Directly determines the spatial supervision region for posterior concepts.
  • grounding loss hyperparameters (λ_sep=1, λ_mass=1, λ_s=2, margin δ=10%) = λ_sep=1, λ_mass=1, λ_s=2, δ=0.10
    Weights in Eqs. (5)–(7) chosen by hand; no sensitivity analysis is reported, and the strength of spatial grounding depends on these values.
axioms (4)
  • domain assumption Posterior acoustic phenomena are informative in a fixed band immediately below the lesion for all cases.
    Explicitly stated in Section II-A: 'we assume that a posterior band may be informative across cases in this work.' This is load-bearing for the posterior grounding loss and metrics.
  • domain assumption Lesion masks are available at training time and are spatially aligned with the concept maps.
    The dataset D includes masks m_i; zones are derived from masks and resized to concept-map resolution (Section II-A). The method's grounding signal is entirely mask-dependent.
  • domain assumption Concepts can be partitioned into ROI-localized and posterior-localized groups (morphology vs. posterior phenomena).
    ROI concepts (shape, margin, echo pattern) and posterior concepts (enhancement, shadowing) are assigned to zones in Section III-A; this grouping is a modeling choice that the evaluation then uses.
  • domain assumption Zone-based metrics (Energy-in-Zone, Hit@1, Top-5% overlap) are a valid proxy for spatial faithfulness of concept evidence.
    The metrics measure activation fraction inside predefined zones, which is the same notion as the training objective. The paper itself notes (Discussion) that this 'indicates spatial alignment of evidence' and should be interpreted alongside semantic correctness, not as a substitute.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis." pith.science (2026). https://pith.science/paper/HX4K6ON2

@misc{pith2026260720691,
  author       = {Pith},
  title        = {Pith review of: Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HX4K6ON2}},
  note         = {Machine review of arXiv:2607.20691}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Concept Bottleneck Models provide interpretable-by-design predictions by mediating diagnosis through human-understandable concepts, but in medical imaging, their trustworthiness is often limited by the quality and granularity of available supervision. In particular, predicted concept activations can be driven by irrelevant regions, leading to spatially unfaithful explanations. We study a data-centric spatially grounded Concept Bottleneck Model (SG-CBM) that leverages coarse lesion delineations as weak supervision to encourage anatomically plausible concept evidence. For breast ultrasound, we derive two clinically motivated zones from each lesion mask: (i) an in-lesion region of interest for morphology-related concepts and (ii) a posterior acoustic band for posterior phenomena. We train concept maps using a grouped spatial grounding objective and preserve semantic faithfulness with a linear bottleneck classifier. Across five-fold stratified group cross-validation, the proposed SG-CBM improves diagnostic AUROC and concept macro-AUROC while markedly increasing spatial alignment of concept evidence. We also perform a Train-corrupt/Test-clean annotation-quality stress test to quantify the impact of supervision quality on diagnosis and spatial faithfulness. Overall, the results underscore the need for data-quality-aware supervision design and systematic trustworthiness validation for deployable healthcare AI systems.

Figures

Figures reproduced from arXiv: 2607.20691 by Afzel Noore, Dunren Che, Haitham Y. Adarbah, Moshiur Rahman Tonmoy.

Figure 1
Figure 1. Figure 1: High-level architecture of SG-CBM. The ROI zone (red) and posterior [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of spatial grounding, illustrating the averaged sigmoid activations of ROI and posterior-related concepts on the rightmost 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Loss Invariance Determines What Concept Layers Encode: Volume Grounding in Echocardiography

    cs.AI 2026-07 conditional novelty 5.0

    Because ejection fraction is a ratio, an EF-only objective leaves the volume concept layer determined only up to rescaling, and the ungrounded layer collapses to near-zero volume spread despite decent EF accuracy.

Reference graph

Works this paper leans on

22 extracted references · 4 canonical work pages · cited by 1 Pith paper

  1. [1]

    Application of deep learning on automated breast ultrasound: Current developments, challenges, and opportunities,

    R. Wang, Z. Wang, Y . Xiao, X. Liu, G. Tan, and J. Liu, “Application of deep learning on automated breast ultrasound: Current developments, challenges, and opportunities,”Meta-Radiology, vol. 3, no. 2, p. 100138, 2025. [Online]. Available: https://www.sciencedirect. com/science/article/pii/S2950162825000062

    2025

  2. [2]

    A review of methods for trustworthy ai in medical imaging: The future-ai guidelines,

    H. Kondylakis, R. Osuala, X. Puig-Bosch, N. Lazrak, O. Diaz, K. Kushibar, I. Chouvarda, S. Charalambous, M. P. Starmans, S. Colantonio, N. Tachos, S. Joshi, H. C. Woodruff, Z. Salahuddin, G. Tsakou, S. Auss ´o, L. C. Alberich, N. Papanikolaou, P. Lambin, K. Marias, M. Tsiknakis, D. I. Fotiadis, L. Mart ´ı-Bonmat´ı, and K. Lekadir, “A review of methods for...

    arXiv 2026

  3. [3]

    Post-hoc vs ante-hoc explanations: xai design guidelines for data scientists,

    C. O. Retzlaff, A. Angerschmid, A. Saranti, D. Schneeberger, R. R ¨ottger, H. M ¨uller, and A. Holzinger, “Post-hoc vs ante-hoc explanations: xai design guidelines for data scientists,”Cognitive Systems Research, vol. 86, p. 101243, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1389041724000378

    2024

  4. [4]

    X-brain: Explainable recognition of brain tumors using robust deep attention cnn,

    M. R. Tonmoy, M. A. Shams, M. A. Adnan, M. Mridha, M. Safran, S. Alfarhood, and D. Che, “X-brain: Explainable recognition of brain tumors using robust deep attention cnn,”Biomedical Signal Processing and Control, vol. 100, p. 106988, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1746809424010462

    2025

  5. [5]

    Unmasking dementia detection by masking input gradients: A jsm approach to model interpretability and precision,

    Y . Mustafa and T. Luo, “Unmasking dementia detection by masking input gradients: A jsm approach to model interpretability and precision,” inAdvances in Knowledge Discovery and Data Mining, D.-N. Yang, X. Xie, V . S. Tseng, J. Pei, J.-W. Huang, and J. C.-W. Lin, Eds. Singapore: Springer Nature Singapore, 2024, pp. 75–90

    2024

  6. [6]

    Detecting and mitigating the clever hans effect in medical imaging: a scoping review,

    C. Vasquez-Venegas, C. Wu, S. Sundar, R. Proa, F. J. Beloy, J. R. Medina, M. Mcnichol, K. Parvataneni, N. Kurtzman, F. Mirshawka et al., “Detecting and mitigating the clever hans effect in medical imaging: a scoping review,”Journal of Imaging Informatics in Medicine, vol. 38, no. 4, pp. 2563–2579, 2025. [Online]. Available: https://doi.org/10.1007/s10278-...

    work page doi:10.1007/s10278-024-01335-z 2025

  7. [7]

    Concept bottleneck models,

    P. W. Koh, T. Nguyen, Y . S. Tang, S. Mussmann, E. Pierson, B. Kim, and P. Liang, “Concept bottleneck models,” inProceedings of the 37th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, H. D. III and A. Singh, Eds., vol

  8. [8]

    Learning a Clinically-Relevant Concept Bottleneck for Lesion Detec- tion in Breast Ultrasound ,

    A. Bunnell, Y . Glaser, D. Valdez, T. Wolfgruber, A. Altamirano, C. Zamora Gonz´alez, B. Y . Hernandez, P. Sadowski, and J. A. Shepherd, “ Learning a Clinically-Relevant Concept Bottleneck for Lesion Detec- tion in Breast Ultrasound ,” inproceedings of Medical Image Computing and Computer Assisted Intervention – MICCAI 2024, vol. LNCS 15003. Springer Natu...

    2024

  9. [9]

    Bi-rads-net: An explainable multi- task learning approach for cancer diagnosis in breast ultrasound images,

    B. Zhang, A. Vakanski, and M. Xian, “Bi-rads-net: An explainable multi- task learning approach for cancer diagnosis in breast ultrasound images,” in2021 IEEE 31st International Workshop on Machine Learning for Signal Processing (MLSP), 2021, pp. 1–6

    2021

  10. [10]

    Deep learning for describing breast ultrasound images with bi-rads terms,

    M. Carrilero-Mardones, M. Parras-Jurado, A. Nogales, J. P ´erez-Mart´ın, and F. J. D ´ıez, “Deep learning for describing breast ultrasound images with bi-rads terms,”Journal of Imaging Informatics in Medicine, vol. 37, no. 6, pp. 2940–2954, Dec 2024. [Online]. Available: https://doi.org/10.1007/s10278-024-01155-1

    work page doi:10.1007/s10278-024-01155-1 2024

  11. [11]

    A review of the segment anything model (sam) for medical image analysis: Accomplishments and perspectives,

    M. Ali, T. Wu, H. Hu, Q. Luo, D. Xu, W. Zheng, N. Jin, C. Yang, and J. Yao, “A review of the segment anything model (sam) for medical image analysis: Accomplishments and perspectives,”Computerized Medical Imaging and Graphics, vol. 119, p. 102473, 2025. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0895611124001502

    2025

  12. [12]

    On the concept trustworthiness in concept bottleneck models,

    Q. Huang, J. Song, J. Hu, H. Zhang, Y . Wang, and M. Song, “On the concept trustworthiness in concept bottleneck models,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 19, pp. 21 161–21 168, Mar. 2024. [Online]. Available: https://ojs.aaai.org/index.php/AAAI/article/view/30109

    2024

  13. [13]

    What’s in the bottle? a survey and roadmap of concept bottleneck models,

    P. Knab, D. Steinmann, C. Bartelt, K. Kersting, B. Schiele, T. Seidl, U. Schlegel, and W. Stammer, “What’s in the bottle? a survey and roadmap of concept bottleneck models,”Transactions on Machine Learning Research, 2026. [Online]. Available: https: //openreview.net/forum?id=IF5vnqxBEW

    2026

  14. [14]

    Identification of benign and malignant breast nodules on ultrasound: comparison of multiple deep learning models and model interpretation,

    X. Wen, H. Tu, B. Zhao, W. Zhou, Z. Yang, and L. Li, “Identification of benign and malignant breast nodules on ultrasound: comparison of multiple deep learning models and model interpretation,”Frontiers in Oncology, vol. V olume 15 - 2025, 2025. [Online]. Available: https://www.frontiersin.org/journals/ oncology/articles/10.3389/fonc.2025.1517278

    arXiv 2025

  15. [15]

    Weakly supervised lesion detection and diagnosis for breast cancers with partially annotated ultrasound images,

    J. Wang, L. Qiao, S. Zhou, J. Zhou, J. Wang, J. Li, S. Ying, C. Chang, and J. Shi, “Weakly supervised lesion detection and diagnosis for breast cancers with partially annotated ultrasound images,”IEEE Transactions on Medical Imaging, vol. 43, no. 7, pp. 2509–2521, 2024

    2024

  16. [16]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, “Focal loss for dense object detection,” inProceedings of the IEEE International Conference on Computer Vision (ICCV), Oct 2017

    2017

  17. [17]

    Curated benchmark dataset for ultrasound based breast lesion analysis,

    A. Pawłowska, A. ´Cwierz-Pie´nkowska, A. Domalik, D. Jagu ´s, P. Kasprzak, R. Matkowski, Ł. Fura, A. Nowicki, and N. ˙Zołek, “Curated benchmark dataset for ultrasound based breast lesion analysis,” Scientific Data, vol. 11, no. 1, p. 148, Jan 2024. [Online]. Available: https://doi.org/10.1038/s41597-024-02984-z

    work page doi:10.1038/s41597-024-02984-z 2024

  18. [18]

    A curated benchmark dataset for ultrasound based breast lesion analysis (breast-lesions-usg) (version 1) [dataset],

    A. Pawłowska, A. ´Cwierz-Pie´nkowska, A. Domalik, D. Jagu ´s, P. Kasprzak, R. Matkowski, Ł. Fura, A. Nowicki, and N. Zolek, “A curated benchmark dataset for ultrasound based breast lesion analysis (breast-lesions-usg) (version 1) [dataset],”The Cancer Imaging Archive,

  19. [19]

    Sam 3: Segment anything with concepts,

    N. Carion, L. Gustafson, Y .-T. Hu, S. Debnath, R. Hu, D. Suris, C. Ryali, K. V . Alwala, H. Khedr, A. Huang, J. Lei, T. Ma, B. Guo, A. Kalla, M. Marks, J. Greer, M. Wang, P. Sun, R. R ¨adle, T. Afouras, E. Mavroudi, K. Xu, T.-H. Wu, Y . Zhou, L. Momeni, R. Hazra, S. Ding, S. Vaze, F. Porcher, F. Li, S. Li, A. Kamath, H. K. Cheng, P. Doll ´ar, N. Ravi, K....

    Pith/arXiv arXiv 2025

  20. [20]

    A comprehensive survey on deep active learning in medical image analysis,

    H. Wang, Q. Jin, S. Li, S. Liu, M. Wang, and Z. Song, “A comprehensive survey on deep active learning in medical image analysis,”Medical Image Analysis, vol. 95, p. 103201, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1361841524001269

    2024

  21. [119]

    5338–5348

    PMLR, 13–18 Jul 2020, pp. 5338–5348. [Online]. Available: https://proceedings.mlr.press/v119/koh20a.html

    2020

  22. [2024]

    Available: https://doi.org/10.7937/9WKK-Q141

    [Online]. Available: https://doi.org/10.7937/9WKK-Q141

    work page doi:10.7937/9wkk-q141

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.