Pith. sign in

REVIEW 2 major objections 4 minor 48 references

What Can We Learn from Inter-Annotator Variability in Skin Lesion Segmentation?

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper shows that skin lesions with lower inter-annotator agreement are more likely to be malignant, and that this disagreement signal can be predicted from the image and used to improve diagnostic accuracy.

desk verdict A genuinely useful dataset and a plausible IAA–malignancy link, but the headline statistical claim doesn't control for annotator pair/tool composition, so the central association is not yet fully established. read the letter →

arxiv 2508.09381 v1 pith:554XPFBO submitted 2025-08-12 cs.CV cs.AIcs.LG

Kumar Abhishek , Jeremy Kawahara , Ghassan Hamarneh This is my paper
classification cs.CVcs.AIcs.LG
keywords inter-annotatoragreementskinlesionsegmentationmalignancyassociationmulti-tasklearningdermoscopyDicesimilaritystochasticdominance
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 asks whether the amount of disagreement between human annotators when outlining skin lesions carries clinically meaningful information. It curates a large multi-annotator dataset, finds that malignant lesions have systematically lower inter-annotator agreement than benign ones, and shows this agreement can be predicted directly from a dermoscopic image. It then uses predicted agreement as an auxiliary training task, which improves diagnostic balanced accuracy by 4.2% on average across architectures and five datasets.

What carries the argument

The key mechanism is the per-image inter-annotator agreement score, computed as the average Dice similarity across all unique pairs of segmentation masks for that image. This single scalar condenses annotator variability and is treated as 'soft' clinical information that both correlates with malignancy and can be regressed from image features. The multi-task model shares a backbone between a diagnosis head and an IAA regression head, forcing the network to encode features of ambiguity that aid classification.

What would settle it

Recompute the benign-versus-malignant agreement comparison using only images whose masks were produced by the same pair of annotators, or after stratifying by annotator skill and annotation tool; if the association disappears, the reported link is an artifact of annotator assignment rather than lesion boundary ambiguity.

Watch

Extended reading notes

Core claim

The central discovery is a statistically significant association between inter-annotator agreement (IAA), measured as mean pairwise Dice score, and lesion malignancy. Using first-order stochastic dominance tests, the paper shows that the distribution of IAA for benign lesions dominates that for malignant lesions, meaning benign lesions tend to elicit higher annotator consensus. This association is further leveraged by training models to predict IAA from the image, achieving a mean absolute error of 0.108, and by using the predicted IAA as a soft clinical feature in a multi-task learning objective that improves diagnosis over the same model trained on diagnosis alone.

Load-bearing premise

The IAA score for each image is computed from a variable number of masks, with most images having only two masks annotated by different, non-overlapping annotators, and the statistical comparison does not control for which annotators segmented each image.

Editorial extensions

If this is right

  • If the association holds, inter-annotator agreement can serve as a practical proxy for ambiguous lesion boundaries, which are clinically linked to malignancy but difficult to formalize.
  • Predicting IAA directly from images allows this signal to be used at inference time without needing multiple annotations, making it feasible in routine clinical workflows.
  • The multi-task formulation suggests that adding an auxiliary regression task can improve diagnostic accuracy even on datasets that lack multiple annotations, as shown by fine-tuning experiments on four external datasets.
  • The released dataset provides a common ground for future studies on annotator variability in dermatology, enabling fair comparison of methods that model or exploit disagreement.

Reading between the lines

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

  • The association between IAA and malignancy might extend to other medical imaging domains where boundary ambiguity is a known marker of disease aggressiveness, such as lung nodules or breast masses.
  • The paper's IAA prediction model could potentially be used as an attention mechanism to highlight regions where clinicians should look most carefully, though the authors do not explore this.
  • A natural next test is whether the multi-task gain persists when the regression head is trained on a completely different set of annotators or when the number of masks per image is fixed, which would separate the effect of disagreement from annotator identity.
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

2 major / 4 minor

Summary. The paper curates IMA++, a multi-annotator skin lesion segmentation dataset with 5111 masks from 15 annotators on 2394 images, and reports three main results: (1) a statistically significant association between inter-annotator agreement (IAA), measured as Dice, and lesion malignancy, with benign lesions showing higher agreement; (2) IAA can be predicted directly from dermoscopic images with MAE ≈ 0.108; and (3) multi-task training with an IAA regression head improves diagnostic balanced accuracy by an average of 4.2% across model architectures and five datasets. The central scientific claim is that lesion boundary ambiguity, as captured by human segmentation disagreement, is linked to malignancy and can be used as a 'soft' clinical feature.

Significance. If the association result is robust, it is clinically and methodologically interesting: it offers a quantitative link between a measurable property of annotation behavior and disease severity, and the released IMA++ dataset is the largest public multi-annotator skin lesion segmentation dataset to date. The paper also has concrete strengths: it uses formal FOSD and Mann–Whitney tests, reports effect sizes, evaluates 13 architectures for IAA prediction, and tests multi-task transfer on four external datasets. The code is released. However, the central association is established only from pooled data, and the downstream multi-task claim lacks a control for IAA-specificity. These issues need to be addressed before the paper's main conclusions can be taken at face value.

major comments (2)
  1. [Sec. 3.1, Fig. 2C] The central association test pools all benign and malignant IAA scores without accounting for the composition of annotators, tools, or skills per image. IMA++ has 2130/2394 images with exactly two masks and 15 unique annotators, so each image's IAA is estimated from a small, non-matched set of segmentations. Fig. 2C itself shows that same-tool and same-skill pairs have significantly higher Dice than mixed pairs. If malignancy status is imbalanced across these factors, the pooled FOSD/Mann–Whitney result (p<0.001) could reflect annotator-pair composition rather than lesion-boundary ambiguity. Please report the balance of tool/skill/annotator-pair types across benign vs. malignant images and provide a stratified or regression-adjusted analysis (e.g., within annotator-pair type, or a mixed-effects model with random annotator effects).
  2. [Sec. 3.3, Table 1] The conclusion that IAA prediction is a beneficial auxiliary task is not yet supported as IAA-specific. MMT adds both a second prediction head and an auxiliary loss to M2; the consistent gains in Tables 1–2 could arise from generic multi-task regularization or increased capacity. A control with the same architecture trained on a meaningless auxiliary target (e.g., shuffled IAA labels, or image rotation) is needed to attribute the improvement to the semantic content of IAA. Also clarify how the best α is selected and whether Table 1 reports test-set metrics for models selected on validation; the current text says 'α=0.9 performs best' without specifying the selection protocol.
minor comments (4)
  1. [Abstract / Sec. 3.3] The abstract's 4.2% average improvement is not stated or derived in the body. Although it can be reconstructed from Tables 1–2, the authors should report the computation explicitly.
  2. [Sec. 3.1] After rejecting H_mal⪰1ben, the reverse p=0.923 only means the data are consistent with benign dominance; it does not establish FOSD. The authors acknowledge this in Sec. 4 as future work, but the main-text phrasing 'support that ... first-order stochastically dominates' should be softened.
  3. [Fig. 3a] The Mann–Whitney U p-values in Fig. 3a are not defined in the text. Specify the null hypothesis and what exactly is being compared (e.g., predicted vs. true IAA scores, or model predictions across images).
  4. [Typos/formatting] Fig. 2C caption: 'Hausdoff' should be 'Hausdorff'; introduction: 'psuedopods' should be 'pseudopods'; Ref. 20: 'iomedical' should be 'biomedical'; table captions: 'meanstd. dev.' should be spaced as 'mean std. dev.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: IAA is empirically measured and the auxiliary-task gains are evaluated against independent diagnosis labels.

full rationale

The derivation chain is self-contained. IAA (Zi) is defined as the per-image mean pairwise Dice over the available segmentation masks (Sec 3.1), an empirical measurement, not a function of the diagnosis label Yi. The association test (Sec 3.1) compares the empirical distributions Zben and Zmal using Mann-Whitney U and FOSD; these distributions are estimated from the collected masks, and the malignancy labels are external to the agreement computation. The regression model M1 (Sec 2.2) is trained to predict Zi from the image, with the target being the measured agreement, so there is no construction by which the prediction is forced to equal the input except the usual supervised fitting on a held-out test set. The multi-task model MMT (Sec 2.3) is trained with a combined loss using diagnosis labels and IAA labels; the IAA labels are not derived from the diagnosis labels or from the model's own output, so the reported 4.2% balanced-accuracy improvement is an empirical result against external diagnosis targets. Self-citations (e.g., [1], [2]) appear only as dataset comparisons or related-work context, not as load-bearing evidence for the central claims. The skeptic's concern about annotator-pair/tool composition confounding the IAA-malignancy association is a legitimate measurement-validity/correctness issue, but it is not a circular-derivation issue: it does not amount to the conclusion being equivalent to its premises by definition. Therefore the circularity score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper relies on standard statistical and domain assumptions about annotation quality and label correctness. The main weakness is the uncontrolled annotator-assignment assumption, which is not an invented entity but a measurement assumption. No new physical or conceptual entities are introduced.

free parameters (1)
  • loss weight alpha = 0.9 (selected from {0.1, 0.2, 0.5, 0.8, 0.9} on validation)
    Eq. 3; the multi-task superiority claim depends on this hyperparameter, which was chosen by validation performance.
assumptions (4)
  • domain assumption Dice coefficient averaged over pairwise masks is a valid measure of inter-annotator agreement
    Sec 3.1: 'we adopt the Dice metric, which is standard in IAA studies in medical imaging'.
  • domain assumption ISIC malignancy labels are correct ground truth
    The entire statistical analysis uses these labels without independent verification, as stated in Sec 3.1.
  • domain assumption The set of annotators per image is random with respect to lesion malignancy
    The malignancy comparison pools images without controlling for annotator identity or skill mix (Sec 3.1).
  • standard math FOSD test implementation (PySDTest) and bootstrap resampling assumptions hold
    Sec 2.1 uses PySDTest with 1000 bootstrap iterations at alpha=0.001.

how reviews work

0 comments
Cite this review

Pith. "Pith review of What Can We Learn from Inter-Annotator Variability in Skin Lesion Segmentation?." pith.science (2026). https://pith.science/paper/554XPFBO

@misc{pith2026250809381,
  author       = {Pith},
  title        = {Pith review of: What Can We Learn from Inter-Annotator Variability in Skin Lesion Segmentation?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/554XPFBO}},
  note         = {Machine review of arXiv:2508.09381}
}
read the original abstract

Medical image segmentation exhibits intra- and inter-annotator variability due to ambiguous object boundaries, annotator preferences, expertise, and tools, among other factors. Lesions with ambiguous boundaries, e.g., spiculated or infiltrative nodules, or irregular borders per the ABCD rule, are particularly prone to disagreement and are often associated with malignancy. In this work, we curate IMA++, the largest multi-annotator skin lesion segmentation dataset, on which we conduct an in-depth study of variability due to annotator, malignancy, tool, and skill factors. We find a statistically significant (p<0.001) association between inter-annotator agreement (IAA), measured using Dice, and the malignancy of skin lesions. We further show that IAA can be accurately predicted directly from dermoscopic images, achieving a mean absolute error of 0.108. Finally, we leverage this association by utilizing IAA as a "soft" clinical feature within a multi-task learning objective, yielding a 4.2% improvement in balanced accuracy averaged across multiple model architectures and across IMA++ and four public dermoscopic datasets. The code is available at https://github.com/sfu-mial/skin-IAV.

Figures

Figures reproduced from arXiv: 2508.09381 by the authors.

Figure 1
Figure 1. Regression (M1), diagnosis-only (M2), and multi-task (MMT) models. tensive evaluation on multiple datasets, that multi-task models jointly predicting diagnosis and IAA outperform diagnosis-only models. 2 Methods 2.1 Agreement distribution shift across disease classes Let (X , Y, S) denote a dataset of N images {Xi} N i=1, corresponding N diagnoses {Yi} N i=1, and N sets of multiple segmentation masks S = {{Sik} Ki k… view at source ↗
Figure 2
Figure 2. IMA++ dataset statistics: (A) number of segmentations per image, (B) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Predicting inter-annotator agreement (Dice) on the proposed IMA++ [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 43 canonical work pages

  1. [1]

    Scientific Reports11(1), 7769 (2021)

    Abhishek, K., Kawahara, J., Hamarneh, G.: Predicting the clinical management of skin lesions using deep learning. Scientific Reports11(1), 7769 (2021)

    work page 2021

  2. [2]

    In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention (MICCAI) ISIC Skin Image Analysis Workshop (ISIC)

    Abhishek, K., Kawahara, J., Hamarneh, G.: Segmentation style discovery: Appli- cation to skin lesion images. In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention (MICCAI) ISIC Skin Image Analysis Workshop (ISIC). pp. 24–34. Springer (2024)

    work page 2024

  3. [3]

    Comprehensive framework for evaluation of deep neural networks in detection and quantification of lymphoma from PET/CT images: clinical insights, pitfalls, and observer agreement analyses

    Ahamed, S., Xu, Y., Gowdy, C., Bloise, I., Wilson, D., Martineau, P., Bénard, F., Yousefirizi, F., Dodhia, R., Lavista, J.M., et al.: Comprehensive evaluation and insights into the use of deep neural networks to detect and quantify lymphoma lesions in PET/CT images. arXiv preprint arXiv:2311.09614 (2023)

    work page Pith review arXiv 2023

  4. [4]

    Artificial Intelligence Review54(1), 137–178 (2021)

    Asgari Taghanaki, S., Abhishek, K., Cohen, J.P., Cohen-Adad, J., Hamarneh, G.: Deep semantic segmentation of natural and medical images: A review. Artificial Intelligence Review54(1), 137–178 (2021)

    work page 2021

  5. [5]

    In: Color Medical Image Analysis, pp

    Ballerini, L., Fisher, R.B., Aldridge, B., Rees, J.: A color and texture based hi- erarchical K-NN approach to the classification of non-melanoma skin lesions. In: Color Medical Image Analysis, pp. 63–86. Springer (2013)

    work page 2013

  6. [6]

    Economet- rica 71(1), 71–104 (2003)

    Barrett, G.F., Donald, S.G.: Consistent tests for stochastic dominance. Economet- rica 71(1), 71–104 (2003)

    work page 2003

  7. [7]

    Dermoscopy Image Analysis10(1), 97–129 (2015)

    Celebi, M.E., Wen, Q., Iyatomi, H., Shimizu, K., Zhou, H., Schaefer, G.: A state- of-the-art survey on lesion border detection in dermoscopy images. Dermoscopy Image Analysis10(1), 97–129 (2015)

    work page 2015

  8. [8]

    In: 2018 IEEE Winter Conference on Applications of Computer Vision (WACV)

    Chattopadhay, A., Sarkar, A., Howlader, P., Balasubramanian, V.N.: Grad- CAM++: Generalized gradient-based visual explanations for deep convolutional networks. In: 2018 IEEE Winter Conference on Applications of Computer Vision (WACV). pp. 839–847. IEEE (2018) 10 Abhishek et al

    work page 2018

Show all 48 references
  1. [9]

    Cheplygina, V., Pluim, J.P.: Crowd disagreement about medical images is infor- mative. In: International Conference on Medical Image Computing and Computer- Assisted Intervention (MICCAI) International Workshop on Large-scale Anno- tation of Biomedical data and Expert Label Sy...

    2018

  2. [10]

    arXiv preprint arXiv:1902.03368 (2019)

    Codella, N., Rotemberg, V., Tschandl, P., Celebi, M.E., Dusza, S., Gutman, D., Helba, B., Kalloo, A., Liopyris, K., Marchetti, M., Kittler, H., Halpern, A.: Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the In- ternational Skin Imaging Collaboratio...

    2018 arXiv

  3. [11]

    In: 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018)

    Codella, N.C., Gutman, D., Celebi, M.E., Helba, B., Marchetti, M.A., Dusza, S.W., Kalloo, A., Liopyris, K., Mishra, N., Kittler, H., Halpern, A.: Skin lesion analysis toward melanoma detection: A challenge at the 2017 international symposium on biomedical imaging (ISBI), hoste...

    2017

  4. [12]

    Routledge (2013)

    Cohen, J.: Statistical power analysis for the behavioral sciences. Routledge (2013)

    2013

  5. [13]

    Skin Research and Technology18(3), 311–315 (2012)

    Fortina, A.B., Peserico, E., Silletti, A., Zattra, E.: Where’s the naevus? inter- operator variability in the localization of melanocytic lesion border. Skin Research and Technology18(3), 311–315 (2012)

    2012

  6. [14]

    The Spine Journal14(10), 2442–2448 (2014)

    Fu, M.C., Buerba, R.A., Long III, W.D., Blizzard, D.J., Lischuk, A.W., Haims, A.H., Grauer, J.N.: Interrater and intrarater agreements of magnetic resonance imaging findings in the lumbar spine: significant variability across degenerative conditions. The Spine Journal14(10), 2...

    2014

  7. [15]

    In: Proceedings of the IEEE International Conference on Computer Vision

    Girshick, R.: Fast R-CNN. In: Proceedings of the IEEE International Conference on Computer Vision. pp. 1440–1448 (2015)

    2015

  8. [16]

    In: Bragg, D.G., Rubin, P., Hricak, H

    Griff, S.K., Dershaw, D.D.: Chapter 16 - breast cancer. In: Bragg, D.G., Rubin, P., Hricak, H. (eds.) Oncologic Imaging, pp. 265–294. Elsevier (2002),https://doi. org/10.1016/B0-72-167494-1/50019-0

    2002 doi

  9. [17]

    Medical Image Analysis 94, 103141 (2024)

    Gut, D., Trombini, M., Kucybała, I., Krupa, K., Rozynek, M., Dellepiane, S., Tabor, Z., Wojciechowski, W.: Use of superpixels for improvement of inter-rater and intra-rater reliability during annotation of medical images. Medical Image Analysis 94, 103141 (2024)

    2024

  10. [18]

    Computers in Biology and Medicine 155, 106624 (2023)

    Hasan, M.K., Ahamad, M.A., Yap, C.H., Yang, G.: A survey, review, and future trends of skin lesion segmentation and classification. Computers in Biology and Medicine 155, 106624 (2023)

    2023

  11. [19]

    Scientific Data11(1), 641 (2024)

    Hernández-Pérez, C., Combalia, M., Podlipnik, S., Codella, N.C., Rotemberg, V., Halpern, A.C., Reiter, O., Carrera, C., Barreiro, A., Helba, B., Puig, S., Vilaplana, V., Malvehy, J.: BCN20000: Dermoscopic lesions in the wild. Scientific Data11(1), 641 (2024)

    2024

  12. [20]

    IEEE Journal of iomedical and Health Informatics23(2), 538–546 (2018)

    Kawahara, J., Daneshvar, S., Argenziano, G., Hamarneh, G.: Seven-point checklist and skin lesion classification using multitask multimodal neural nets. IEEE Journal of iomedical and Health Informatics23(2), 538–546 (2018)

    2018

  13. [21]

    BMC Bioinformatics17(Suppl 13), 367 (2016)

    Kaya, S., Bayraktar, M., Kockara, S., Mete, M., Halic, T., Field, H.E., Wong, H.K.: Abrupt skin lesion border cutoff measurement for malignancy detection in dermoscopy images. BMC Bioinformatics17(Suppl 13), 367 (2016)

    2016

  14. [22]

    Artificial Intelligence in Medicine56(2), 69–90 (2012)

    Korotkov, K., Garcia, R.: Computerized analysis of pigmented skin lesions: A re- view. Artificial Intelligence in Medicine56(2), 69–90 (2012)

    2012

  15. [23]

    arXiv preprint arXiv:2307.10694 (2023) Inter-Annotator Variability in Skin Lesion Segmentation 11

    Lee,K.,Whang,Y.J.:PySDTest:aPython/Statapackageforstochasticdominance tests. arXiv preprint arXiv:2307.10694 (2023) Inter-Annotator Variability in Skin Lesion Segmentation 11

    2023 arXiv

  16. [24]

    In: Proc

    Li, X., Aldridge, B., Rees, J., Fisher, R.: Estimating the ground truth from multiple individual segmentations with application to skin lesion segmentation. In: Proc. Medical Image Understanding and Analysis Conference, UK. vol. 1, pp. 101–106 (2010)

    2010

  17. [25]

    In: Proceedings of the IEEE International Conference on Computer Vision

    Lin, T.Y., Goyal, P., Girshick, R., He, K., Dollár, P.: Focal loss for dense object de- tection. In: Proceedings of the IEEE International Conference on Computer Vision. pp. 2980–2988 (2017)

    2017

  18. [26]

    Computer Methods and Programs in Biomedicine215, 106620 (2022)

    Lucieri, A., Bajwa, M.N., Braun, S.A., Malik, M.I., Dengel, A., Ahmed, S.: ExAID: A multimodal explanation framework for computer-aided diagnosis of skin lesions. Computer Methods and Programs in Biomedicine215, 106620 (2022)

    2022

  19. [27]

    The Annals of Mathematical Statistics pp

    Mann, H.B., Whitney, D.R.: On a test of whether one of two random variables is stochastically larger than the other. The Annals of Mathematical Statistics pp. 50–60 (1947)

    1947

  20. [28]

    Journal of Machine Learning Research17(81), 1–32 (2016)

    Maurer, A., Pontil, M., Romera-Paredes, B.: The benefit of multitask representa- tion learning. Journal of Machine Learning Research17(81), 1–32 (2016)

    2016

  21. [29]

    Med- ical Image Analysis88, 102863 (2023)

    Mirikharaji, Z., Abhishek, K., Bissoto, A., Barata, C., Avila, S., Valle, E., Celebi, M.E., Hamarneh, G.: A survey on deep learning for skin lesion segmentation. Med- ical Image Analysis88, 102863 (2023)

    2023

  22. [30]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) ISIC Skin Image Analysis Workshop (ISIC)

    Mirikharaji, Z., Abhishek, K., Izadi, S., Hamarneh, G.: D-LEMA: Deep learn- ing ensembles from multiple annotations-application to skin lesion segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) ISIC Skin Image Analysis Wo...

    2021

  23. [31]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshop on Safe Artificial Intelligence for All Domains (SAIAD)

    Patrício, C., Neves, J.C., Teixeira, L.F.: Coherent concept-based explanations in medical image and its application to skin lesion diagnosis. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshop on Safe Artificial Intelligence f...

    2023

  24. [32]

    Peruch, F., Bogo, F., Bonazza, M., Cappelleri, V.M., Peserico, E.: Simpler, faster, moreaccuratemelanocyticlesionsegmentationthroughMEDS.IEEETransactions on Biomedical Engineering61(2), 557–565 (2013)

    2013

  25. [33]

    Machine Learning for Biomedical Imaging1, 1–26 (Dec 2021)

    Raumanns, R., Schouten, G., Joosten, M., Pluim, J.P., Cheplygina, V.: ENHANCE (enriching health data by annotations of crowd and experts): A case study for skin lesion classification. Machine Learning for Biomedical Imaging1, 1–26 (Dec 2021)

    2021

  26. [34]

    arXiv preprint arXiv:1906.02415 (2019)

    Ribeiro, V., Avila, S., Valle, E.: Handling inter-annotator agreement for automated skin lesion segmentation. arXiv preprint arXiv:1906.02415 (2019)

    1906 arXiv

  27. [35]

    Ribeiro, V., Avila, S., Valle, E.: Less is more: Sample selection and label condition- ingimproveskinlesionsegmentation.In:ProceedingsoftheIEEE/CVFConference on Computer Vision and Pattern Recognition (CVPR) ISIC Skin Image Analysis Workshop (ISIC). pp. 738–739 (2020)

    2020

  28. [36]

    In: 2006 International Conference on Image Process- ing

    Sampat, M.P., Wang, Z., Markey, M.K., Whitman, G.J., Stephens, T.W., Bovik, A.C.: Measuring intra-and inter-observer agreement in identifying and localizing structures in medical images. In: 2006 International Conference on Image Process- ing. pp. 81–84. IEEE (2006)

    2006

  29. [37]

    European Radiology21(12), 2475–2483 (2011)

    Sohns, C., Scherrer, M., Staab, W., Obenauer, S.: Value of the BI-RADS classifi- cation in MR-mammography for diagnosis of benign and malignant breast tumors. European Radiology21(12), 2475–2483 (2011)

    2011

  30. [38]

    IEEE Journal of Biomedical and Health Informatics 24(10), 2912–2921 (2020) 12 Abhishek et al

    Song, L., Lin, J., Wang, Z.J., Wang, H.: An end-to-end multi-task deep learn- ing framework for skin lesion analysis. IEEE Journal of Biomedical and Health Informatics 24(10), 2912–2921 (2020) 12 Abhishek et al

    2020

  31. [39]

    Scientific Data 5(1), 1–9 (2018)

    Tschandl, P., Rosendahl, C., Kittler, H.: The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific Data 5(1), 1–9 (2018)

    2018

  32. [40]

    Computers in Biology and Medicine104, 111–116 (2019)

    Tschandl, P., Sinz, C., Kittler, H.: Domain-specific classification-pretrained fully convolutional network encoders for skin lesion segmentation. Computers in Biology and Medicine104, 111–116 (2019)

    2019

  33. [41]

    IEEE Transactions on Medical Imaging (2025)

    Wang, J., Zhou, C., Huang, Y.: Contour-aware multi-expert model for ambiguous medical image segmentation. IEEE Transactions on Medical Imaging (2025)

    2025

  34. [42]

    arXiv preprint arXiv:2311.10380 (2023)

    Wang, S., Weng, T., Wang, J., Shen, Y., Zhao, Z., Liu, Y., Jiao, P., Cheng, Z., Wang, Y.: MSE-Nets: Multi-annotated semi-supervised ensemble networks for im- proving segmentation of medical image with ambiguous boundaries. arXiv preprint arXiv:2311.10380 (2023)

    2023 arXiv

  35. [43]

    Cambridge University Press (2019)

    Whang, Y.J.: Econometric analysis of stochastic dominance: Concepts, methods, tools, and applications. Cambridge University Press (2019)

    2019

  36. [44]

    JAMA Dermatology 157(9), 1078–1088 (2021)

    Williams,N.M.,Rojas,K.D.,Reynolds,J.M.,Kwon,D.,Shum-Tien,J.,Jaimes,N.: Assessment of diagnostic accuracy of dermoscopic structures and patterns used in melanoma detection: A systematic review and meta-analysis. JAMA Dermatology 157(9), 1078–1088 (2021)

    2021

  37. [45]

    IEEE Transactions on Medical Imaging 39(7), 2482–2493 (2020)

    Xie, Y., Zhang, J., Xia, Y., Shen, C.: A mutual bootstrapping model for automated skin lesion segmentation and classification. IEEE Transactions on Medical Imaging 39(7), 2482–2493 (2020)

    2020

  38. [46]

    arXiv preprint arXiv:1703.01025 (2017)

    Yang, X., Zeng, Z., Yeo, S.Y., Tan, C., Tey, H.L., Su, Y.: A novel multi-task deep learning model for skin lesion segmentation and classification. arXiv preprint arXiv:1703.01025 (2017)

    2017 arXiv

  39. [47]

    arXiv preprint arXiv:2303.15850 (2023)

    Zepf, K., Petersen, E., Frellsen, J., Feragen, A.: That label’s got style: Handling label style bias for uncertain image segmentation. arXiv preprint arXiv:2303.15850 (2023)

    2023 arXiv

  40. [48]

    Computers in Biology and Medicine153, 106496 (2023)

    Zhao, Y., Wang, X., Che, T., Bao, G., Li, S.: Multi-task deep learning for medical image computing and analysis: A review. Computers in Biology and Medicine153, 106496 (2023)

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.