Pith. sign in

REVIEW 3 major objections 5 minor 66 references

Mask of truth: model sensitivity to unexpected regions of medical images

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Masking the lungs does not stop chest X-ray classifiers from scoring above chance, and full-image models score higher without the lungs than with only the lungs.

desk verdict Solid, honest empirical study; central 'shortcut' claim is overstrong for chest X-rays because 'without lungs' still contains clinically relevant anatomy. read the letter →

arxiv 2412.04030 v3 pith:NGHH6TW3 submitted 2024-12-05 cs.CV

Théo Sourget , Michelle Hestbek-M{o}ller , Amelia Jiménez-Sánchez , Jack Junchi Xu , Veronika Cheplygina This is my paper · ORCID
classification cs.CV
keywords shortcutlearningspuriouscorrelationchestX-rayclassificationglaucomamedicalimagemaskingmodelrobustnessexplainabilityout-of-distributionevaluation
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

Masking experiments on chest X-rays and eye fundus images show that a standard CNN keeps classifying above chance even when the clinically relevant anatomy is removed, and that full-image chest models actually score higher on images without the lungs than on images with only the lungs. The paper argues this is evidence of shortcut learning: models exploit non-clinical cues in the periphery rather than the anatomy clinicians use. On a glaucoma dataset, performance looks unbiased at first, but a mask-dilation experiment reveals that optic disc size alone drives predictions. A radiology resident could rarely diagnose the same masked chest images, and external validation shows poor transfer of all masked models. The work matters because benchmark AUC can overstate the true clinical capability of medical imaging models.

What carries the argument

The controlling mechanism is a systematic occlusion protocol built from anatomical segmentation masks. For lungs, the CheXmask masks define the region of interest; for fundus images, the Cháks.u optic disc and cup masks do. Each image is presented in five versions—full, outside the mask, outside the bounding box, inside the mask, inside the bounding box—and one DenseNet-121 model is trained per version, then cross-evaluated by AUC with DeLong significance tests, mask-dilation sweeps, SHAP attributions, t-SNE and cosine embedding comparisons, and a radiology-resident reading study. The bounding-box variants are the key control: they remove shape and location information of the ROI without removing nearby structures, which is what reveals reliance on peripulmonary or boundary cues.

What would settle it

Evaluate the no-lungs and no-lungs-bounding-box models on images in which everything outside the lung mask is replaced by class-matched noise while the lung region is blackened; if the above-chance AUC persists, non-clinical peripheral cues drive the predictions, but if it falls to 0.5, the apparent shortcut is leakage of lung pixels through imperfect masks.

Watch

Extended reading notes

Core claim

The central claim is that convolutional classifiers trained on PadChest chest X-rays rely on spurious correlations: every model, no matter which masking strategy it was trained on, achieves an AUC above random in matched evaluation, and models trained on full images perform better on images without the lungs than on images containing only the lungs. For atelectasis, effusion, and pneumonia, models evaluated on only-lung images fall near or below chance, while bounding-box versions recover performance, pointing to usage of peripulmonary structures. On the Cháks.u glaucoma dataset, full-image models reach 0.68 AUC while no-disc models fall to near chance, but the dilation experiment shows optic disc size alone drives predictions, which is a shortcut masquerading as clinically sensible focus.

Load-bearing premise

The central claim rests on the assumption that the segmentation masks accurately identify all clinically relevant parts of the image; if the masks are systematically wrong, the 'without-ROI' images still contain the relevant anatomy or the 'only-ROI' images exclude it, and the shortcut conclusion collapses.

Editorial extensions

If this is right

  • Benchmark AUCs on PadChest overstate real capability: models can score well without ever seeing the lungs.
  • Masking the ROI is not a sufficient robustness audit; only cross-evaluation across masking conditions reveals shortcut reliance.
  • Peripheral and boundary cues such as pacemakers, image edges, and scanner-specific artifacts are learnable shortcuts that survive targeted anatomical removal.
  • Models trained only on the ROI, or only outside it, transfer poorly to external datasets, so shortcut reliance limits deployment across hospitals.
  • For glaucoma, models can fixate on a clinically plausible region while using the wrong feature, so explainability maps alone cannot certify correct reasoning.

Reading between the lines

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

  • A similar train-and-evaluation masking audit could be run cheaply on other pathologies and architectures as a pre-deployment shortcut screen, since the released code supports swapping models.
  • The disc-size result predicts measurable subgroup performance gaps across populations with different optic disc size distributions; this is directly testable by stratifying AUC by disc size in external data.
  • The embedding analysis shows that penultimate-feature cosine similarity and t-SNE can rank masking conditions differently from AUC, so representations before the classification head are weak evidence about shortcut use without interventional masking.
  • Perturbing mask boundaries and measuring the AUC slope would quantify how much of the no-lungs performance is an artifact of imperfect masks versus genuine peripheral shortcut use.
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 / 5 minor

Summary. The paper studies shortcut learning in medical image classifiers by training DenseNet-121 models on chest X-rays (PadChest) and eye fundus images (Cháks.u) under five masking strategies: full images, images with the clinically relevant region (lungs or optic disc) removed using precise masks, the same with bounding boxes, and images retaining only the ROI via masks or bounding boxes. The main claims are that all chest X-ray models achieve above-random AUC regardless of masking, that models trained on full images perform better on images without the ROI than on images with only the ROI, that out-of-distribution performance is generally poor, and that a radiology resident found it nearly impossible to diagnose without-ROI images. The paper also reports embedding similarities, SHAP visualizations, and a disc-size dilation experiment for glaucoma that suggests reliance on optic disc size rather than cup-to-disc ratio. The authors conclude that models are affected by spurious correlations, with chest X-rays showing stronger evidence than fundus images.

Significance. If the conclusions hold, the paper provides a useful cautionary demonstration that CNNs can achieve non-trivial AUC even when clinically relevant structures are removed, and it offers a multi-pronged methodology (masking variants, OOD evaluation, embeddings, SHAP, expert study) for auditing shortcut behavior. The glaucoma disc-size experiment is a particularly nice falsifiable test, and the release of code supports reproducibility. However, the central chest X-ray claim is currently overstated: the 'without lungs' condition does not remove all clinically relevant anatomy for cardiomegaly (heart) and effusion (costophrenic angles), so the headline comparison between 'No lungs' and 'Only lungs' does not cleanly separate clinical features from non-clinical shortcuts. The paper's value lies in its methodology and per-class observations, but the interpretation needs substantial refinement before the general conclusion about non-relevant cues is justified.

major comments (3)
  1. [§4.1, Fig. 5] The headline finding that full-image models score higher on 'No lungs' than on 'Only lungs' images is presented as evidence of reliance on non-clinical shortcuts, but this comparison is confounded for at least two of the five PadChest classes. Cardiomegaly is diagnosed from heart size (cardiothoracic ratio), and the heart lies outside the CheXmask lung segmentation, so the 'No lungs' condition preserves the clinically decisive organ while 'Only lungs' removes it; for pleural effusion, the costophrenic angles may be partially outside the lung mask. The paper itself acknowledges this in §3.2 ('include or exclude relevant parts like the heart for the cardiomegaly condition') and in §4.5 reports that the radiology resident correctly identified a cardiomegaly case in a no-lungs image. The claim that high without-ROI performance reflects spurious correlation should be restricted to classes whose relevant anatomy is fully contained in the lung mask (e.g., pneumonia, atelectasis) or re-analyzed per class with the bounding-box conditions separated from the mask conditions.
  2. [§4.1, Fig. 5] The dilation experiment does not isolate the shortcut mechanism. Applying a model trained on full images to heavily blacked-out 'Only lungs' images induces a large distribution shift, because most pixels are replaced by a constant value; the observed drop in AUC may reflect the model's sensitivity to the masking operation itself rather than reliance on non-lung features. Without a control model trained on images with the same distribution of masks (e.g., random masking or train-time dilation), the conclusion that the model 'uses elements near the lungs' is not uniquely supported. The authors should either add such a control or temper the interpretation of Fig. 5.
  3. [§4.5, §3.5] The radiologist study is used to conclude that models rely on non-relevant features because the resident found only 2 of 35 conditions in without-ROI images. This inference does not follow: the resident was deprived of projection information (as the paper notes) and saw only a single image per case, while the model may exploit subtle but clinically valid cues such as heart size for cardiomegaly. The study is also underpowered (one expert, 75 images) and the image selection by model probability may bias the difficulty. The conclusion should be tempered to state that the models' decision cues are not readily usable by a human expert under the same information restrictions, rather than claiming that the resident's difficulty proves the model used non-clinical features.
minor comments (5)
  1. [Fig. 4] The AUC heatmaps show only mean values across the five folds; adding per-fold values or standard deviations would help assess variability, especially since the significance conclusions rely on a 3-of-5 folds rule.
  2. [§3.2] The sentence about the bounding box 'include or exclude relevant parts like the heart for the cardiomegaly condition' is stated in the methods but its implications for interpreting the mask-based results are not carried through in the discussion; consider moving or expanding it in the results section.
  3. [Abstract and §3.1] The name of the eye fundus dataset is spelled inconsistently ('Cháks.u' vs 'Cháks.u'); please standardize the spelling throughout.
  4. [Table 2] The OOD evaluation is on full images; clarify in the caption or text that the 'No ROI' rows refer to models trained on masked images and evaluated on full images, to avoid confusion with the in-distribution masked evaluation.
  5. [§4.3] The cosine similarity results are acknowledged as limited and not aligning with AUC; consider reporting them in the appendix or adding a short interpretation of why shared background pixels dominate the embeddings.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper reports held-out empirical AUC comparisons with external OOD datasets; no fitted parameter is renamed as a prediction and no result is defined into existence.

full rationale

This is an empirical benchmarking study rather than a derivation. The central claims—that models trained on each masking strategy obtain above-random AUC, and that Full-image models score higher on No-lungs than on Only-lungs test images—are measured on a held-out PadChest test split and on external ChestX-ray14 and AIROGS data, with 5-fold cross-validation and DeLong significance tests. No parameter is fitted to the test set and then called a prediction; the masking strategies are pre-defined image transformations, and the AUC values are evaluated, not derived from the mask definitions. The paper's terminology equating 'outside the CheXmask lung segmentation' with 'non-clinically relevant' is an operational assumption, and the authors explicitly acknowledge its limits for cardiomegaly and effusion, and even report a radiology resident finding a cardiomegaly case in a No-lungs image. Such anatomical validity concerns are correctness risks, not circularity: the conclusion does not reduce to the assumption by construction. Self-citations appear only as background context or methodological support (e.g., shortcut detection, tabular baselines) and are not load-bearing for the empirical findings. No equations are used to define a quantity in terms of itself, and no uniqueness theorem is invoked. The honest finding is therefore no significant circularity, score 0.

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

The paper introduces no new theoretical entities; its claims rest on empirical assumptions about mask quality, model representativeness, and the validity of explanation and expert evaluation. No fitted free parameters are used in a derivation sense.

assumptions (4)
  • domain assumption CheXmask segmentation masks are accurate enough to define the clinically relevant ROI.
    Section 3.1 selects masks with Dice RCA > 0.7, but Fig. 11b shows a non-accurate mask example, so this assumption is acknowledged but load-bearing.
  • domain assumption DenseNet-121 is representative of CNN behavior for the phenomena being studied.
    Section 3.2 uses a single architecture; the discussion notes other architectures may differ.
  • domain assumption SHAP attributions provide meaningful localization of the model's evidence.
    Section 3.4 cites prior work supporting SHAP over gradient methods, and Section 4.4 discusses limitations of SHAP.
  • domain assumption A single radiology resident's assessments represent a valid clinical benchmark for the expert study.
    Section 3.5 describes one resident with five years of experience; the discussion notes the limitation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mask of truth: model sensitivity to unexpected regions of medical images." pith.science (2026). https://pith.science/paper/NGHH6TW3

@misc{pith2026241204030,
  author       = {Pith},
  title        = {Pith review of: Mask of truth: model sensitivity to unexpected regions of medical images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGHH6TW3}},
  note         = {Machine review of arXiv:2412.04030}
}
read the original abstract

The development of larger models for medical image analysis has led to increased performance. However, it also affected our ability to explain and validate model decisions. Models can use non-relevant parts of images, also called spurious correlations or shortcuts, to obtain high performance on benchmark datasets but fail in real-world scenarios. In this work, we challenge the capacity of convolutional neural networks (CNN) to classify chest X-rays and eye fundus images while masking out clinically relevant parts of the image. We show that all models trained on the PadChest dataset, irrespective of the masking strategy, are able to obtain an Area Under the Curve (AUC) above random. Moreover, the models trained on full images obtain good performance on images without the region of interest (ROI), even superior to the one obtained on images only containing the ROI. We also reveal a possible spurious correlation in the Chaksu dataset while the performances are more aligned with the expectation of an unbiased model. We go beyond the performance analysis with the usage of the explainability method SHAP and the analysis of embeddings. We asked a radiology resident to interpret chest X-rays under different masking to complement our findings with clinical knowledge. Our code is available at https://github.com/TheoSourget/MMC_Masking and https://github.com/TheoSourget/MMC_Masking_EyeFundus

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 57 canonical work pages

  1. [1]

    Characterizing the clinical adoption of medical ai devices through u.s

    Kevin Wu, Eric Wu, Brandon Theodorou, Weixin Liang, Christina Mack, Lucas Glass, Jimeng Sun, and James Zou. Characterizing the clinical adoption of medical ai devices through u.s. insurance claims.NEJM AI, 1(1):AIoa2300030, 2023. Mask of truth: model sensitivity to unexpected regions of medical images 19

    work page 2023

  2. [2]

    Artificial intelligence versus clinicians in disease diagnosis: systematic review.JMIR medical informatics, 7(3):e10010, 2019

    Jiayi Shen, Casper JP Zhang, Bangsheng Jiang, Jiebin Chen, Jian Song, Zherui Liu, Zonglin He, Sum Yi Wong, Po-Han Fang, Wai-Kit Ming, et al. Artificial intelligence versus clinicians in disease diagnosis: systematic review.JMIR medical informatics, 7(3):e10010, 2019

    work page 2019

  3. [3]

    Autonomous chest radiograph reporting using ai: estimation of clinical impact

    Louis L Plesner, Felix C Müller, Janus D Nybing, Lene C Laustrup, Finn Ras- mussen, Olav W Nielsen, Mikael Boesen, and Michael B Andersen. Autonomous chest radiograph reporting using ai: estimation of clinical impact. Radiology, 307(3):e222268, 2023

    work page 2023

  4. [4]

    Ho, and James Zou

    Eric Wu, Kevin Wu, Roxana Daneshjou, David Ouyang, Daniel E. Ho, and James Zou. How medical ai devices are evaluated: limitations and recommendations from an analysis of fda approvals.Nature Medicine, 27(4):582–584, 2021

    work page 2021

  5. [5]

    Detecting shortcuts in medical images-a case study in chest x-rays

    Amelia Jiménez-Sánchez, Dovile Juodelyte, Bethany Chamberlain, and Veronika Cheplygina. Detecting shortcuts in medical images-a case study in chest x-rays. In 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), pages 1–5, 2023

    work page 2023

  6. [6]

    Hidden stratification causes clinically meaningful failures in machine learning for medical imaging

    Lauren Oakden-Rayner, Jared Dunnmon, Gustavo Carneiro, and Christopher Ré. Hidden stratification causes clinically meaningful failures in machine learning for medical imaging. In ACM Conference on Health, Inference, and Learning, pages 151–159, 2020

    work page 2020

  7. [7]

    Counterfactual contrastive learning: robust representations via causal image synthesis

    Mélanie Roschewitz, Fabio de Sousa Ribeiro, Tian Xia, Galvin Khara, and Ben Glocker. Counterfactual contrastive learning: robust representations via causal image synthesis. InMICCAI Workshop on Data Engineering in Medical Imaging, pages 22–32, 2024

    work page 2024

  8. [8]

    All you need is a guiding hand: Mitigating short- cut bias in deep learning models for medical imaging

    Christopher Boland, Owen Anderson, Keith A Goatman, John Hipwell, Sotirios A Tsaftaris, and Sonia Dahdouh. All you need is a guiding hand: Mitigating short- cut bias in deep learning models for medical imaging. InMICCAI Workshop on Fairness of AI in Medical Imaging, pages 67–77, 2024

    work page 2024

Show all 66 references
  1. [9]

    Deep learning for understanding multilabel imbalanced chest x-ray datasets

    Helena Liz, Javier Huertas-Tato, Manuel Sánchez-Montañés, Javier Del Ser, and David Camacho. Deep learning for understanding multilabel imbalanced chest x-ray datasets. Future Generation Computer Systems, 144:291–306, 2023

    2023

  2. [10]

    Airogs: artificial intelligence for robust glaucoma screening chal- lenge

    Coen De Vente, Koenraad A Vermeer, Nicolas Jaccard, He Wang, Hongyi Sun, Firas Khader, Daniel Truhn, Temirgali Aimyshev, Yerkebulan Zhanibekuly, Tien- Dung Le, et al. Airogs: artificial intelligence for robust glaucoma screening chal- lenge. IEEE transactions on medical imaging, 2023

    2023

  3. [11]

    Machine learning and deep learning methods for skin lesion classification and diagnosis: a systematic review.Diagnostics, 11(8):1390, 2021

    Mohamed A Kassem, Khalid M Hosny, Robertas Damaševičius, and Mo- hamed Meselhy Eltoukhy. Machine learning and deep learning methods for skin lesion classification and diagnosis: a systematic review.Diagnostics, 11(8):1390, 2021

    2021

  4. [12]

    Dynamic routing between capsules

    Sara Sabour, Nicholas Frosst, and Geoffrey E Hinton. Dynamic routing between capsules. Advances in neural information processing systems, 30, 2017

    2017

  5. [13]

    Capsule networks against medical imaging data challenges

    Amelia Jiménez-Sánchez, Shadi Albarqouni, and Diana Mateus. Capsule networks against medical imaging data challenges. InIntravascular Imaging and Computer Assisted Stenting and Large-Scale Annotation of Biomedical Data and Expert Label Synthesis, pages 150–160, 2018

    2018

  6. [14]

    A capsule network-based for identification of glaucoma in retinal images

    Patrick Ryan Sales Dos Santos, Vitória de Carvalho Brito, Antonio Oseas de Car- valho Filho, Flávio Henrique Duarte de Araújo, Ricardo de Andrade Lira Rabêlo, and Mano Joseph Mathew. A capsule network-based for identification of glaucoma in retinal images. In2020 IEEE Symposiu...

    2020

  7. [15]

    Transformers in medical imaging: A survey.Medical Image Analysis, 88:102802, 2023

    Fahad Shamshad, Salman Khan, Syed Waqas Zamir, Muhammad Haris Khan, Munawar Hayat, Fahad Shahbaz Khan, and Huazhu Fu. Transformers in medical imaging: A survey.Medical Image Analysis, 88:102802, 2023

    2023

  8. [16]

    Lt-vit: A vision transformer for multi-label chest x-ray classification

    Umar Marikkar, Sara Atito, Muhammad Awais, and Adam Mahdi. Lt-vit: A vision transformer for multi-label chest x-ray classification. In2023 IEEE International Conference on Image Processing (ICIP), pages 2565–2569, 2023

    2023

  9. [17]

    Detecting glaucoma from fundus photographs using deep learning without convolutions: transformer for improved generalization.Ophthal- mology science, 3(1):100233, 2023

    Rui Fan, Kamran Alipour, Christopher Bowd, Mark Christopher, Nicole Brye, James A Proudfoot, Michael H Goldbaum, Akram Belghith, Christopher A Girkin, Massimo A Fazio, et al. Detecting glaucoma from fundus photographs using deep learning without convolutions: transformer for i...

    2023

  10. [18]

    Advancing human-centric ai for robust x-ray analysis through holistic self-supervised learning.arXiv preprint arXiv:2405.01469, 2024

    Théo Moutakanni, Piotr Bojanowski, Guillaume Chassagnon, Céline Hudelot, Ar- mand Joulin, Yann LeCun, Matthew Muckley, Maxime Oquab, Marie-Pierre Revel, and Maria Vakalopoulou. Advancing human-centric ai for robust x-ray analysis through holistic self-supervised learning.arXiv...

    2024 arXiv

  11. [19]

    Towards generalist foundation model for radiology.arXiv preprint arXiv:2308.02463, 2023

    Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, and Weidi Xie. Towards generalist foundation model for radiology.arXiv preprint arXiv:2308.02463, 2023

    2023 arXiv

  12. [20]

    Visual–language foundation models in medicine.The Visual Computer, pages 1–20, 2024

    Chunyu Liu, Yixiao Jin, Zhouyu Guan, Tingyao Li, Yiming Qin, Bo Qian, Ze- hua Jiang, Yilan Wu, Xiangning Wang, Ying Feng Zheng, et al. Visual–language foundation models in medicine.The Visual Computer, pages 1–20, 2024

    2024

  13. [21]

    Visionunite: A vision-language foundation model for oph- thalmology enhanced with clinical knowledge

    Zihan Li, Diping Song, Zefeng Yang, Deming Wang, Fei Li, Xiulan Zhang, Paul E Kinahan, and Yu Qiao. Visionunite: A vision-language foundation model for oph- thalmology enhanced with clinical knowledge. arXiv preprint arXiv:2408.02865, 2024

    2024 arXiv

  14. [22]

    Shortcutlearningindeepneural networks

    Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel,MatthiasBethge,andFelixAWichmann. Shortcutlearningindeepneural networks. Nature Machine Intelligence, 2(11):665–673, 2020

    2020

  15. [23]

    shortcuts

    Imon Banerjee, Kamanasish Bhattacharjee, John L Burns, Hari Trivedi, Saptarshi Purkayastha,LalehSeyyed-Kalantari,BhavikNPatel,RakeshShiradkar,andJudy Gichoya. “shortcuts” causing bias in radiology artificial intelligence: causes, evalu- ation and mitigation.Journal of the Amer...

    2023

  16. [24]

    Detecting and mitigating the clever hans effect in medical imaging: A scoping review.Journal of Imaging Informatics in Medicine, pages 1–17, 2024

    Constanza Vásquez-Venegas, Chenwei Wu, Saketh Sundar, Renata Prôa, Fran- cis Joshua Beloy, Jillian Reeze Medina, Megan McNichol, Krishnaveni Parvataneni, Nicholas Kurtzman, Felipe Mirshawka, et al. Detecting and mitigating the clever hans effect in medical imaging: A scoping r...

    2024

  17. [25]

    An unex- pected confounder: how brain shape can be used to classify mri scans? InMedical Imaging with Deep Learning, 2024

    Valentine Wargnier-Dauchelle, Thomas Grenier, and Michaël Sdika. An unex- pected confounder: how brain shape can be used to classify mri scans? InMedical Imaging with Deep Learning, 2024

    2024

  18. [26]

    There are no shortcuts to anywhere worth going: Identifying shortcuts in deep learningmodelsformedicalimageanalysis

    Christopher Boland, Keith A Goatman, Sotirios A Tsaftaris, and Sonia Dahdouh. There are no shortcuts to anywhere worth going: Identifying shortcuts in deep learningmodelsformedicalimageanalysis. In Medical Imaging with Deep Learning, 2024

    2024

  19. [27]

    Fast diffusion-based counterfactuals for shortcut removal and generation

    Nina Weng, Paraskevas Pegios, Eike Petersen, Aasa Feragen, and Siavash Bigdeli. Fast diffusion-based counterfactuals for shortcut removal and generation. InEu- ropean Conference on Computer Vision, pages 338–357, 2025

    2025

  20. [28]

    Radedit: stress-testing biomedical vision models via diffusion image editing

    Fernando Pérez-García, Sam Bond-Taylor, Pedro P Sanchez, Boris van Breugel, Daniel C Castro, Harshita Sharma, Valentina Salvatelli, Maria TA Wetscherek, Hannah Richardson, Matthew P Lungren, et al. Radedit: stress-testing biomedical vision models via diffusion image editing. I...

    2024

  21. [29]

    (de)constructing bias on skin lesion datasets

    Alceu Bissoto, Michel Fornaciali, Eduardo Valle, and Sandra Avila. (de)constructing bias on skin lesion datasets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019

    2019

  22. [30]

    Deep learning on fundus images detects glau- coma beyond the optic disc.Scientific Reports, 11(1):20313, 2021

    Ruben Hemelings, Bart Elen, João Barbosa-Breda, Matthew B Blaschko, Patrick De Boever, and Ingeborg Stalmans. Deep learning on fundus images detects glau- coma beyond the optic disc.Scientific Reports, 11(1):20313, 2021

    2021

  23. [31]

    Generalisation chal- lenges in deep learning models for medical imagery: insights from external valida- tionofcovid-19classifiers

    Sophie Crawford Haynes, Pamela Johnston, and Eyad Elyan. Generalisation chal- lenges in deep learning models for medical imagery: insights from external valida- tionofcovid-19classifiers. Multimedia Tools and Applications,83(31):76753–76772, 2024

    2024

  24. [32]

    Optimising chest x-rays for image analysis by identi- fying and removing confounding factors

    Shahab Aslani, Watjana Lilaonitkul, Vaishnavi Gnanananthan, Divya Raj, Boji- dar Rangelov, Alexandra L Young, Yipeng Hu, Paul Taylor, Daniel C Alexander, NCCID Collaborative, et al. Optimising chest x-rays for image analysis by identi- fying and removing confounding factors. I...

    2022

  25. [33]

    Shortcut learning in medical image segmentation

    Manxi Lin, Nina Weng, Kamil Mikolaj, Zahra Bashir, Morten BS Svendsen, Mar- tin G Tolsgaard, Anders N Christensen, and Aasa Feragen. Shortcut learning in medical image segmentation. InInternational Conference on Medical Image Com- puting and Computer-Assisted Intervention, pag...

    2024

  26. [34]

    Source matters: Source dataset impact on model robustness in medical imaging

    Dovile Juodelyte, Yucheng Lu, Amelia Jiménez-Sánchez, Sabrina Bottazzi, Enzo Ferrante, and Veronika Cheplygina. Source matters: Source dataset impact on model robustness in medical imaging. InInternational Workshop on Applications of Medical AI, pages 105–115, 2024

    2024

  27. [35]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. InComputer Vision and Pattern Recog- nition, 2009. CVPR 2009. IEEE Conference on, pages 248–255, 2009

    2009

  28. [36]

    Radimagenet: an open radiologic deep learning research dataset for effective trans- fer learning

    Xueyan Mei, Zelong Liu, Philip M Robson, Brett Marinelli, Mingqian Huang, Amish Doshi, Adam Jacobi, Chendi Cao, Katherine E Link, Thomas Yang, et al. Radimagenet: an open radiologic deep learning research dataset for effective trans- fer learning. Radiology: Artificial Intelli...

    2022

  29. [37]

    Susu Sun, Lisa M Koch, and Christian F Baumgartner. Right for the wrong reason: Caninterpretablemltechniquesdetectspuriouscorrelations? In International Con- ference on Medical Image Computing and Computer-Assisted Intervention, pages 425–434. Springer, 2023

    2023

  30. [38]

    Transparent medical image ai via an image–text foundation model grounded in medical literature

    Chanwoo Kim, Soham U Gadgil, Alex J DeGrave, Jesutofunmi A Omiye, Zhuo Ran Cai, Roxana Daneshjou, and Su-In Lee. Transparent medical image ai via an image–text foundation model grounded in medical literature. Nature Medicine, pages 1–12, 2024

    2024

  31. [39]

    Concept bottleneck models

    Pang Wei Koh, Thao Nguyen, Yew Siang Tang, Stephen Mussmann, Emma Pier- son, Been Kim, and Percy Liang. Concept bottleneck models. In International conference on machine learning, pages 5338–5348, 2020

    2020

  32. [40]

    Padchest: A large chest x-ray image dataset with multi-label annotated reports

    Aurelia Bustos, Antonio Pertusa, Jose-Maria Salinas, and Maria de la Iglesia-Vayá. Padchest: A large chest x-ray image dataset with multi-label annotated reports. Medical Image Analysis, 66:101797, 2020

    2020

  33. [41]

    Chexpert: A large chest radiograph dataset with uncertainty labels and ex- pert comparison

    Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, et al. Chexpert: A large chest radiograph dataset with uncertainty labels and ex- pert comparison. InAAAI Conference on Arti...

    2019

  34. [42]

    Chestx-ray8: Hospital-scale chest x-ray database and bench- marks on weakly-supervised classification and localization of common thorax dis- eases

    Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mohammadhadi Bagheri, and Ronald M Summers. Chestx-ray8: Hospital-scale chest x-ray database and bench- marks on weakly-supervised classification and localization of common thorax dis- eases. In Computer Vision and Pattern Recognit...

    2017

  35. [43]

    Chexmask: a large-scale dataset of anatomical segmentation masks for multi-center chest x-ray images

    Nicolás Gaggion, Candelaria Mosquera, Lucas Mansilla, Julia Mariel Saidman, Martina Aineseder, Diego H Milone, and Enzo Ferrante. Chexmask: a large-scale dataset of anatomical segmentation masks for multi-center chest x-ray images. Scientific Data, 11(1):511, 2024

    2024

  36. [44]

    Reverse classi- fication accuracy: predicting segmentation performance in the absence of ground truth

    Vanya V Valindria, Ioannis Lavdas, Wenjia Bai, Konstantinos Kamnitsas, Eric O Aboagye, Andrea G Rockall, Daniel Rueckert, and Ben Glocker. Reverse classi- fication accuracy: predicting segmentation performance in the absence of ground truth. IEEE transactions on medical imagin...

    2017

  37. [45]

    Cháks.u: A glaucoma specific fundus image database.Scientific data, 10(1):70, 2023

    JR Harish Kumar, Chandra Sekhar Seelamantula, JH Gagan, Yogish S Kamath, Neetha IR Kuzhuppilly, U Vivekanand, Preeti Gupta, and Shilpa Patil. Cháks.u: A glaucoma specific fundus image database.Scientific data, 10(1):70, 2023

    2023

  38. [46]

    Simultaneous truth and performance level estimation (STAPLE): an algorithm for the validation of image segmentation

    Simon K Warfield, Kelly H Zou, and William M Wells. Simultaneous truth and performance level estimation (STAPLE): an algorithm for the validation of image segmentation. IEEE Transactions on Medical Imaging, 23(7):903–921, 2004

    2004

  39. [47]

    Densely connected convolutional networks

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4700–4708, 2017

    2017

  40. [48]

    In the picture: Medical imaging datasets, artifacts, and their living review.arXiv preprint arXiv:2501.10727, 2025

    Amelia Jiménez-Sánchez, Natalia-Rozalia Avlona, Sarah de Boer, Víctor M Campello, Aasa Feragen, Enzo Ferrante, Melanie Ganz, Judy Wawira Gichoya, Camila González, Steff Groefsema, et al. In the picture: Medical imaging datasets, artifacts, and their living review.arXiv preprin...

    2025

  41. [49]

    Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach

    Elizabeth R DeLong, David M DeLong, and Daniel L Clarke-Pearson. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics, pages 837–845, 1988

    1988

  42. [50]

    Fast implementation of delong’s algorithm for comparing the areas under correlated receiver operating characteristic curves.IEEE Signal Processing Letters, 21(11):1389–1393, 2014

    Xu Sun and Weichao Xu. Fast implementation of delong’s algorithm for comparing the areas under correlated receiver operating characteristic curves.IEEE Signal Processing Letters, 21(11):1389–1393, 2014

    2014

  43. [51]

    Classification of copd with multiple in- stance learning

    Veronika Cheplygina, Lauge Sørensen, David MJ Tax, Jesper Holst Pedersen, Marco Loog, and Marleen De Bruijne. Classification of copd with multiple in- stance learning. In 2014 22nd International Conference on pattern recognition, pages 1508–1513, 2014

    2014

  44. [52]

    A new method using deep learning to predict the response to cardiac resyn- chronization therapy

    Kristoffer Larsen, Zhuo He, Fernando de A Fernandes, Xinwei Zhang, Chen Zhao, Qiuying Sha, Claudio T Mesquita, Diana Paez, Ernest V Garcia, Jiangang Zou, et al. A new method using deep learning to predict the response to cardiac resyn- chronization therapy. Journal of Imaging ...

    2025

  45. [53]

    Visualizing data using t-sne.Jour- nal of machine learning research, 9(11), 2008

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne.Jour- nal of machine learning research, 9(11), 2008

    2008

  46. [54]

    A unified approach to interpreting model pre- dictions

    Scott M Lundberg and Su-In Lee. A unified approach to interpreting model pre- dictions. Advances in neural information processing systems, 30, 2017

    2017

  47. [55]

    Navigating the maze of explainable ai: A systematic approach to evaluating methods and metrics

    Lukas Klein, Carsten Lüth, Udo Schlegel, Till Bungert, Mennatallah El-Assady, and Paul Jaeger. Navigating the maze of explainable ai: A systematic approach to evaluating methods and metrics. InAdvances in Neural Information Processing Systems, volume 37, pages 67106–67146, 2024

    2024

  48. [56]

    Impossibility theorems for feature attribution.Proceedings of the National Academy of Sciences, 121(2):e2304406120, 2024

    Blair Bilodeau, Natasha Jaques, Pang Wei Koh, and Been Kim. Impossibility theorems for feature attribution.Proceedings of the National Academy of Sciences, 121(2):e2304406120, 2024. Mask of truth: model sensitivity to unexpected regions of medical images 23

    2024

  49. [57]

    Grad-cam: Visual explanations from deep networks via gradient-based localization

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedan- tam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. InProceedings of the IEEE international conference on computer vision, pages 618–626, 2017

    2017

  50. [58]

    Automatic detection of glaucoma via fundus imaging and artificial intelligence: A review.Survey of ophthalmology, 68(1):17–41, 2023

    Lauren J Coan, Bryan M Williams, Venkatesh Krishna Adithya, Swati Upadhyaya, Ala Alkafri, Silvester Czanner, Rengaraj Venkatesh, Colin E Willoughby, Srini- vasan Kavitha, and Gabriela Czanner. Automatic detection of glaucoma via fundus imaging and artificial intelligence: A re...

    2023

  51. [59]

    Optic disc diameter influences the ability to detect glaucomatous disc damage.Acta ophthalmologica, 71(1):122–129, 1993

    Anders Heijl and Harras Mölder. Optic disc diameter influences the ability to detect glaucomatous disc damage.Acta ophthalmologica, 71(1):122–129, 1993

    1993

  52. [60]

    Optic disc size, an important consideration in the glaucoma evaluation

    Michael D Hancox OD. Optic disc size, an important consideration in the glaucoma evaluation. Clinical Eye and Vision Care, 11(2):59–62, 1999

    1999

  53. [61]

    Model-based cleaning of the quilt-1m pathology dataset for text-conditional image synthesis

    Marc Aubreville, Jonathan Ganz, Jonas Ammeling, Christopher Kaltenecker, and Christof Bertram. Model-based cleaning of the quilt-1m pathology dataset for text-conditional image synthesis. InMedical Imaging with Deep Learning

  54. [62]

    Investigating the quality of dermamnist and fitzpatrick17k dermatological image datasets.Scientific Data, 12(1):196, 2025

    Kumar Abhishek, Aditi Jain, and Ghassan Hamarneh. Investigating the quality of dermamnist and fitzpatrick17k dermatological image datasets.Scientific Data, 12(1):196, 2025

    2025

  55. [63]

    Navigating the landscape of multimodal ai in medicine: a scoping review on technical challenges and clinical applications

    Daan Schouten, Giulia Nicoletti, Bas Dille, Catherine Chia, Pierpaolo Vendittelli, Megan Schuurmans, Geert Litjens, and Nadieh Khalili. Navigating the landscape of multimodal ai in medicine: a scoping review on technical challenges and clinical applications. arXiv preprint arX...

    2024 arXiv

  56. [64]

    The risk of shortcut- ting in deep learning algorithms for medical imaging research.Scientific Reports, 14(1):29224, 2024

    Brandon G Hill, Frances L Koback, and Peter L Schilling. The risk of shortcut- ting in deep learning algorithms for medical imaging research.Scientific Reports, 14(1):29224, 2024

    2024

  57. [65]

    Are vision transformers robust to spurious correlations? International Journal of Computer Vision, 132(3):689–709, 2024

    Soumya Suvra Ghosal and Yixuan Li. Are vision transformers robust to spurious correlations? International Journal of Computer Vision, 132(3):689–709, 2024

    2024

  58. [66]

    Metrics reloaded: recommendations for image analysis validation

    Lena Maier-Hein, Annika Reinke, Patrick Godau, Minu D Tizabi, Florian Buet- tner, Evangelia Christodoulou, Ben Glocker, Fabian Isensee, Jens Kleesiek, Michal Kozubek, et al. Metrics reloaded: recommendations for image analysis validation. Nature methods, pages 1–18, 2024. 24 T...

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.