Deep learning models for COVID-19 chest X-ray classification: preventing shortcut learning using feature Disentanglement
Open accessPublished version of a preprint
Abstract
In response to the COVID-19 global pandemic, recent research has proposed creating deep learning based models that use chest radiographs (CXRs) in a variety of clinical tasks to help manage the crisis. However, the size of existing datasets of CXRs from COVID-19+ patients are relatively small, and researchers often pool CXR data from multiple sources, for example, using different x-ray machines in various patient populations under different clinical scenarios. Deep learning models trained on such datasets have been shown to overfit to erroneous features instead of learning pulmonary characteristics -- a phenomenon known as shortcut learning. We propose adding feature disentanglement to the training process, forcing the models to identify pulmonary features from the images while penalizing them for learning features that can discriminate between the original datasets that the images come from. We find that models trained in this way indeed have better generalization performance on unseen data; in the best case we found that it improved AUC by 0.13 on held out data. We further find that this outperforms masking out non-lung parts of the CXRs and performing histogram equalization, both of which are recently proposed methods for removing biases in CXR datasets.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
AI models learn the underlying data set distribution instead of disease characteristics, leading to model failure.
“AI models learn the underlying data set distribution instead of disease characteristics leading to model failure. 51”
From AI pitfalls and what not to do: mitigating bias in AI (2023) · cites this paper as [51]
Shortcut learning can be mitigated using techniques including feature disentanglement and pre-processing techniques like histogram equalization and lung masking.
“This phenomenon, also called shortcut learning, can be mitigated using various techniques including feature disentanglement (where you reward a model when it learns disease characteristics and penalize the model when it learns features that can discriminate between the original data sets where the images are obtained) 51 ; applying pre-processing techniques like histogram equalization to correct”
From AI pitfalls and what not to do: mitigating bias in AI (2023) · cites this paper as [51]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Squeeze-and-Excitation Networks
Hu, J., Shen, L., & Sun, G. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00745 · Cited by 35,667 · see citing works
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. (2019) · arXiv · doi:10.48550/arxiv.1912.01703 · Cited by 31,473 · see citing works
Squeeze-and-Excitation Networks
Hu, J., Shen, L., Albanie, S., Sun, G., & Wu, E. (2020) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2019.2913372 · Cited by 12,157 · see citing works
Bradski, G. R. (2000) · Medical Entomology and Zoology · Cited by 5,994 · see citing works
Unsupervised Domain Adaptation by BackpropagationPreprint
Ganin, Y., & Lempitsky, V. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1409.7495 · Cited by 5,360 · see citing works
Wang, L., Lin, Z. Q., & Wong, A. M. (2020) · Scientific Reports · doi:10.1038/s41598-020-76550-z · Cited by 3,151 · see citing works
Wynants, L., Van Calster, B., Bonten, M. M., Collins, G. S., Debray, T. P., De Vos, M., Haller, M. C., Heinze, G., Moons, K. G., Riley, R. D., et al. (2020) · doi:10.1101/2020.03.24.20041020 · Cited by 3,070 · see citing works
Automated detection of COVID-19 cases using deep neural networks with X-ray images
Öztürk, T., Talo, M., Yildirim, E. A., Baloğlu, U. B., Yıldırım, Ö., & Acharya, U. R. (2020) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2020.103792 · Cited by 2,607 · see citing works
Apostolopoulos, I. D., & Mpesiana, T. A. (2020) · Physical and Engineering Sciences in Medicine · doi:10.1007/s13246-020-00865-4 · Cited by 2,338 · see citing works
Shortcut learning in deep neural networks
Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., & Wichmann, F. A. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-00257-z · Cited by 2,224 · see citing works
Artificial Intelligence (AI) applications for COVID-19 pandemic
Vaishya, R., Javaid, M., Khan, I. H., & Haleem, A. (2020) · Diabetes & Metabolic Syndrome Clinical Research & Reviews · doi:10.1016/j.dsx.2020.04.012 · Cited by 1,502 · see citing works
Classification of COVID-19 in chest X-ray images using DeTraC deep convolutional neural network
Abbas, A., Abdelsamea, M. M., & Gaber, M. M. (2020) · Applied Intelligence · doi:10.1007/s10489-020-01829-7 · Cited by 1,142 · see citing works
Deep-COVID: Predicting COVID-19 from chest X-ray images using deep transfer learning
Minaee, S., Kafieh, R., Milan Sonka, A., Yazdani, S., & Soufi, G. J. (2020) · Medical Image Analysis · doi:10.1016/j.media.2020.101794 · Cited by 1,026 · see citing works
Zhang, K., Liu, X., Shen, J., Li, Z., Sang, Y., Wu, X., Zha, Y., Liang, W., Wang, C., Wang, K., et al. (2020) · Cell · doi:10.1016/j.cell.2020.04.045 · Cited by 917 · see citing works
Detection of Coronavirus Disease (COVID-19) Based on Deep FeaturesPreprint
Sethy, P. K., & Behera, S. K. (2020) · Preprints.org · doi:10.20944/preprints202003.0300.v1 · Cited by 723 · see citing works
COVID-19 Image Data CollectionPreprint
Cohen, J., Morrison, P., & Dao, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.11597 · Cited by 722 · see citing works
AI for radiographic COVID-19 detection selects shortcuts over signal
DeGrave, A. J., Janizek, J. D., & Lee, S.-I. (2021) · Nature Machine Intelligence · doi:10.1038/s42256-021-00338-7 · Cited by 631 · see citing works
COVID-19 identification in chest X-ray images on flat and hierarchical classification scenarios
Pereira, R. M., Bertolini, D. S., Teixeira, L., Junior, C. N. S., & Costa, Y. M. G. (2020) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2020.105532 · Cited by 511 · see citing works
COVID-ResNet: A Deep Learning Framework for Screening of COVID19 from RadiographsPreprint
Farooq, M. A., & Hafeez, A. B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.14395 · Cited by 409 · see citing works
Deep Learning-Based Decision-Tree Classifier for COVID-19 Diagnosis From Chest X-ray Imaging
Yoo, S. H., Geng, H., Chiu, T. L., Yu, S. K., Cho, D. C., Heo, J. U., Choi, M., Choi, I. H., Văn, C. C., Nhung, N. V., et al. (2020) · Frontiers in Medicine · doi:10.3389/fmed.2020.00427 · Cited by 289 · see citing works
Wang, L., & Wong, A. M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.09871 · Cited by 248 · see citing works
BIMCV COVID-19+: a large annotated dataset of RX and CT images from COVID-19 patientsPreprint
MdlI, V., Saborit-Torres, J. M., Ja, M., Pertusa, A., Bustos, A., Cazorla, M., Galant, J., Barber, X., Orozco‐Beltrán, D., García-García, F., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.01174 · Cited by 246 · see citing works
Unveiling COVID-19 from CHEST X-ray with deep learning: A hurdles race with small data
Tartaglione, E., Barbano, C. A., Berzovini, C., Calandri, M., & Grangetto, M. (2020) · International Journal of Environmental Research and Public Health · doi:10.3390/ijerph17186933 · Cited by 211 · see citing works
Bukhari, S. U. K., Bukhari, S. S. K., Syed, A., & Shah, S. S. H. (2020) · medRxiv · doi:10.1101/2020.03.26.20044610 · Cited by 96 · see citing works
Peng, Y., Tang, Y., Lee, S., Zhu, Y., Summers, R. M., & Lu, Z. (2020) · IEEE Transactions on Big Data · doi:10.1109/tbdata.2020.3035935 · Cited by 90 · see citing works
TorchXRayVision: A library of chest X-ray datasets and models
Cohen, J. P., Viviano, J. D., Bertin, P., Morrison, P., Torabian, P., Guarrera, M., Lungren, M. P., Chaudhari, A., Brooks, R., Hashir, M., et al. (2021) · arXiv · doi:10.48550/arxiv.2111.00595 · Cited by 82 · see citing works
On the limits of cross-domain generalization in automated X-ray predictionPreprint
Cohen, J., Hashir, M., Brooks, R., & Bertrand, H. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.02497 · Cited by 78 · see citing works
Using Artificial Intelligence for COVID-19 Chest X-ray Diagnosis
Borkowski, A., Viswanadham, N. A., Thomas, L. B., Guzman, R. D., Deland, L. A., & Mastorides, S. (2020) · Federal Practitioner · doi:10.12788/fp.0045 · Cited by 74 · see citing works
Lung Segmentation from Chest X-rays using Variational Data ImputationPreprint
Selvan, R., Dam, E. B., Detlefsen, N. S., Rischel, S., Sheng, K., Nielsen, M., & Pai, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.10052 · Cited by 54 · see citing works
Peng, Y., Tang, Y., Lee, S., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.06177 · Cited by 15 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Interpretable and Explainable Machine Learning for Materials Science and Chemistry
Oviedo, F., Ferres, J. L., Buonassisi, T., & Butler, K. T. (2022) · Accounts of Materials Research · doi:10.1021/accountsmr.1c00244 · Cited by 435 · see citing works
AI pitfalls and what not to do: mitigating bias in AI
Gichoya, J. W., Thomas, K., Celi, L. A., Safdar, N., Banerjee, I., Banja, J. D., Seyyed-Kalantari, L., Trivedi, H., & Purkayastha, S. (2023) · The British Journal of Radiology · doi:10.1259/bjr.20230023 · Cited by 277 · see citing works
“…characteristics and penalize the model when it learns features that can discriminate between the original data sets where the images are obtained) 51 ; applying pre-processing techniques like histogram equalization to correct for differences in the images and lung masking to exclude areas where…”
Lo, J., Yu, T. T., Ma, D., Zang, P., Owen, J. P., Zhang, Q., Wang, R. K., Beg, M. F., Lee, A. Y., Jia, Y., et al. (2021) · Ophthalmology Science · doi:10.1016/j.xops.2021.100069 · Cited by 105 · see citing works
Self-citationGonzález-Gonzalo, C., Thee, E. F., Klaver, C. C. W., Lee, A. Y., Schlingemann, R. O., Tufail, A., Verbraak, F., & Sánchez, C. I. (2022) · Progress in Retinal and Eye Research · doi:10.1016/j.preteyeres.2021.101034 · Cited by 103 · see citing works
Self-citationGarcia Santa Cruz, B., Bossa, M. N., Sölter, J., & Husch, A. D. (2021) · Medical Image Analysis · doi:10.1016/j.media.2021.102225 · Cited by 89 · see citing works
TorchXRayVision: A library of chest X-ray datasets and models
Cohen, J. P., Viviano, J. D., Bertin, P., Morrison, P., Torabian, P., Guarrera, M., Lungren, M. P., Chaudhari, A., Brooks, R., Hashir, M., et al. (2021) · arXiv · doi:10.48550/arxiv.2111.00595 · Cited by 82 · see citing works
Ding, X., Shang, B., Xie, C., Xin, J., & Yu, F. (2025) · Humanities and Social Sciences Communications · doi:10.1057/s41599-025-04564-x · Cited by 40 · see citing works
Detecting and Mitigating the Clever Hans Effect in Medical Imaging: A Scoping Review
Vásquez-Venegas, C., Wu, C., Sundar, S., Prôa, R., Beloy, F. J., Medina, J. R., McNichol, M. C., Parvataneni, K., Kurtzman, N., Mirshawka, F., et al. (2024) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-024-01335-z · Cited by 26 · see citing works
Horry, M. J., Chakraborty, S., Pradhan, B., Paul, M., Zhu, J. F., Loh, H. W., Barua, P. D., & Acharya, U. R. (2023) · Sensors · doi:10.3390/s23146585 · Cited by 23 · see citing works
Asif, M., Rehman, F. U., Rashid, Z., Hussain, A., Mirza, A., & Qureshi, W. S. (2025) · IEEE Access · doi:10.1109/access.2025.3583647 · Cited by 21 · see citing works
Nakayama, L. F., de Matos, J. C. R. G., Quion, J., Novaes, F. D. C., Mitchell, W. G., Mwavu, R., Hung, C. J.-Y. J., Santiago, A. P. D., Phanphruk, W., Cardoso, J. S., et al. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000618 · Cited by 19 · see citing works
Rectify ViT Shortcut Learning by Visual Saliency
Ma, C., Zhao, L., Chen, Y., Guo, L., Zhang, T., Hu, X., Shen, D. G., Jiang, X., & Liu, T. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3310531 · Cited by 19 · see citing works
Dhont, J., Wolfs, C. J. A., & Verhaegen, F. (2021) · Medical Physics · doi:10.1002/mp.15419 · Cited by 15 · see citing works
Fragemann, J., Ardizzone, L., Egger, J., Kleesiek, J., & Workshop, M. (2022) · doi:10.36227/techrxiv.19364897 · Cited by 10 · see citing works
Tuberculosis Lesion Segmentation Improvement in X-Ray Images Using Contextual Background Label
Khumang, S., Kansomkeat, S., Tanomkiat, W., & Intajag, S. (2025) · IEEE Access · doi:10.1109/access.2025.3532631 · Cited by 9 · see citing works
Augmenting existing deterioration indices with chest radiographs to predict clinical deterioration
Mu, E., Jabbour, S., Dalca, A. V., Guttag, J. V., Wiens, J., & Sjoding, M. W. (2022) · PLoS ONE · doi:10.1371/journal.pone.0263922 · Cited by 7 · see citing works
Pathak, A. K., Gupta, M., & Jain, G. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2025.1692454 · Cited by 6 · see citing works
Fragemann, J., Ardizzone, L., Egger, J., Kleesiek, J., & Workshop, M. (2022) · doi:10.36227/techrxiv.19364897.v1 · Cited by 6 · see citing works
Feature Selection-driven Bias Deduction in Histopathology Images: Tackling Site-Specific Influences
Kheiri, F., Bidgoli, A. A., Makrehchi, M., & Rahnamayan, S. (2024) · IEEE Congress on Evolutionary Computation (CEC) · doi:10.1109/cec60901.2024.10611837 · Cited by 3 · see citing works
An Efficient One-Dimensional Texture Representation Approach for Lung Disease Diagnosis
Alabdulwahab, A., Park, H.-C., Jeong, H., & Lee, S.-W. (2024) · Applied Sciences · doi:10.3390/app142210661 · Cited by 3 · see citing works
Müller, N. M., Jacobs, J., Williams, J. R., & Böttinger, K. (2023) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw59228.2023.00382 · Cited by 3 · see citing works
Rai, N., Boyd, N. S., Vallad, G. E., & Schumann, A. W. (2026) · AgriEngineering · doi:10.3390/agriengineering8030096 · Cited by 2 · see citing works
American Society of Retina Specialists Artificial Intelligence Task Force Report
Talcott, K. E., Baxter, S. L., Chen, D. K., Korot, E., Lee, A., Kim, J. E., Modi, Y. S., Moshfeghi, D. M., & Singh, R. P. (2024) · Journal of VitreoRetinal Diseases · doi:10.1177/24741264241247602 · Cited by 2 · see citing works
Self-citationArtificial Intelligence Research as a Continuous Clinical Service
Xu, Y., Dodhia, R., Ferres, J. M. L., & Weeks, W. B. (2025) · Mayo Clinic Proceedings Digital Health · doi:10.1016/j.mcpdig.2025.100330 · Cited by 1 · see citing works
Self-citationworkflow., T. P. P. D. A. E. T. I. O. C. D. S. (. I. A. A. Q. H. O. R. (2024) · Journal of Medical Imaging and Radiation Oncology · doi:10.1111/1754-9485.13777
Disentanglement and Assessment of Shortcuts in Ophthalmological Retinal Imaging Exams
Fernandes, L., Gonçalves, T., de Matos, J. C. R. G., Nakayama, L. F., & Cardoso, J. S. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-05870-6_21
Srinivasulu, A., Agrawal, A., Kamma, P., Ponnada, V. T., & Vedaiyan, R. (2026) · IntechOpen eBooks · doi:10.5772/intechopen.1016401
Preprocessing-Induced Ranking Reversal is Architecture-Dependent in Cross-Domain Pneumonia Detection
Sanjaya, U. P., Mujiyono, S., Pratama, A., & Wibisono, I. S. (2026) · Journal of Information Systems and Informatics · doi:10.63158/journalisi.v8i5.1944
Sekhar, R. V., & Thangakumar, J. (2026) · doi:10.1109/click68749.2026.11741745
- Authors
- Trivedi, A., Robinson, C., Blazes, M., Ortiz, A., Desbiens, J., Gupta, S., Dodhia, R., Bhatraju, P. K., Liles, W. C., Kalpathy-Cramer, J., Lee, A. Y., & Lavista Ferres, J. M.
- Affiliations
- Microsoft (United States) (US)
- University of Washington (US)
- Corresponding author
- Lee, Aaron Y.
- ORCID
- Venue
- PLOS ONE
- Publisher
- Public Library of Science (PLoS)
- Published
- 2022-10-06
- Volume
- 17
- Issue
- 10
- Pages
- e0274098
- ISSN
- 1932-6203 (ISSN-L)
- PMC
- PMC7885941
- OpenAlex
- W4302760599
- Cited by
- 28
- References
- 29
- Citations by year
- 2026: 4 · 2025: 5 · 2024: 5 · 2023: 5 · 2022: 4 · 2021: 5
- FWCI
- 2.70
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- PLoS ONE (publishedVersion, cc-by)
- medRxiv (acceptedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- national institutes of health (K23DK116967)
- national eye institute (K23EY029246)
- research to prevent blindness
- lantham vision innovation award
- C. Dan and Irene Hunter Endowed Professorship
- Topics
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › COVID-19 diagnosis using AI
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- COVID-19 pandemic, chest X-ray classification, shortcut learning, feature disentanglement, domain generalization, cross-dataset generalization, data bias, histogram equalization, small medical image datasets
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
