Generative models improve fairness of medical classifiers under distribution shifts
Open accessPublished version of a preprint
The DOI matches its Crossref record; 85 of the 85 references it cites were found in a registry or an index.
Abstract
Domain generalization is a ubiquitous challenge for machine learning in healthcare. Model performance in real-world conditions might be lower than expected because of discrepancies between the data encountered during deployment and development. Underrepresentation of some groups or conditions during model development is a common cause of this phenomenon. This challenge is often not readily addressed by targeted data acquisition and ‘labeling’ by expert clinicians, which can be prohibitively expensive or practically impossible because of the rarity of conditions or the available clinical expertise. We hypothesize that advances in generative artificial intelligence can help mitigate this unmet need in a steerable fashion, enriching our training dataset with synthetic examples that address shortfalls of underrepresented conditions or subgroups. We show that diffusion models can automatically learn realistic augmentations from data in a label-efficient manner. We demonstrate that learned augmentations make models more robust and statistically fair in-distribution and out of distribution. To evaluate the generality of our approach, we studied three distinct medical imaging contexts of varying difficulty: (1) histopathology, (2) chest X-ray and (3) dermatology images. Complementing real samples with synthetic ones improved the robustness of models in all three medical tasks and increased fairness by improving the accuracy of clinical diagnosis within underrepresented groups, especially out of distribution.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures











Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“LLM-based oncological tools exhibit racial and gender biases that could widen, rather than narrow, existing cancer care disparities”
From A generative AI multi-agent framework with integrated XAI governance for cancer diagnostics: from multi-omics interpretation to lifestyle risk stratification (2026) · cites this paper as Ktena et al., 2024
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Generative models have shown promise in partially mitigating these disparities by generating synthetic data for underrepresented subgroups, improving classifier fairness under distribution shift”
From A generative AI multi-agent framework with integrated XAI governance for cancer diagnostics: from multi-omics interpretation to lifestyle risk stratification (2026) · cites this paper as Ktena et al., 2024
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“lifestyle data from epidemiological studies systematically overrepresents high-income Western populations (Ktena et al., 2024; Marchesi et al., 2025).”
From A generative AI multi-agent framework with integrated XAI governance for cancer diagnostics: from multi-omics interpretation to lifestyle risk stratification (2026) · cites this paper as Ktena et al., 2024
Experimental — accuracy not yet measuredIn the citing paper
“Generative AI models, such as large language models or vision-language models capable of synthesizing images, additionally allow for tailored data augmentation by creating new examples that meet a set of targeted criteria.71, 72, 73”
From Bias in artificial intelligence for medical imaging: fundamentals, detection, avoidance, mitigation, challenges, ethics, and prospects (2024) · cites this paper as [73]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 2×
“, 2017) and ImageNet21K (Deng et al., 2009) for pretraining to explore how much different pretraining datasets impact final results.”
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
Learning Transferable Visual Models From Natural Language Supervision
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.00020 · Cited by 15,293
Dermatologist-level classification of skin cancer with deep neural networks
Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017) · Nature · doi:10.1038/nature21056 · Cited by 14,362
Abdi, H., & Williams, L. J. (2010) · WIREs Computational Statistics · doi:10.1002/wics.101 · Cited by 11,104
Focal Loss for Dense Object Detection
Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollar, P. (2020) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2018.2858826 · Cited by 9,556
Denoising Diffusion Probabilistic Models
Ho, J., Jain, A., & Abbeel, P. (2020) · arXiv · doi:10.48550/arxiv.2006.11239 · Cited by 8,712 · cited 5×
“19,20,43 for the generation and the upsampler (only the radiology and dermatology datasets required higher-resolution images).”
InfluentialWang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.369 · Cited by 4,616
A generalization of sampling without replacement from a finite universe
Horvitz, D. G., & Thompson, D. J. (1952) · Journal of the American Statistical Association · doi:10.1080/01621459.1952.10483446 · Cited by 4,293 · cited 1×
“…examples of chest X-rays from the CheXpert dataset, we evaluated on a held-out CheXpert test set (containing 13,332 images), which we considered in-distribution, and the test set of ChestX-ray14 (containing 17,723 images), which we considered OOD because of demographic and acquisition shifts.”
(2012) · MPG.PuRe (Max Planck Society) · doi:10.5555/2188385.2188410 · Cited by 3,603
A theory of learning from different domains
Ben-David, S., Blitzer, J., Crammer, K., Kulesza, A., Pereira, F., & Vaughan, J. W. (2010) · Machine Learning · doi:10.1007/s10994-009-5152-4 · Cited by 3,515 · cited 1×
“Previous work on OOD generalization 65–67 pointed out that several factors can affect the performance of a model on samples from domains beyond the training data.”
Randaugment: Practical automated data augmentation with a reduced search space
Cubuk, E. D., Zoph, B., Shlens, J., & Le, Q. V. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw50498.2020.00359 · Cited by 3,223 · cited 9×
“We investigate using RandAugment (Cubuk et al., 2020), ImageNet Augmentations as described above, and RandAugment + ImageNet Augmentations to determine how much performance we can gain by utilizing heuristic augmentations.”
InfluentialRandAugment: Practical automated data augmentation with a reduced search space
Cubuk, E. D., Zoph, B., Shlens, J., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1909.13719 · Cited by 3,223 · cited 9×
“We investigate using RandAugment (Cubuk et al., 2020), ImageNet Augmentations as described above, and RandAugment + ImageNet Augmentations to determine how much performance we can gain by utilizing heuristic augmentations.”
InfluentialClinically applicable deep learning for diagnosis and referral in retinal disease
De Fauw, J., Ledsam, J. R., Romera-Paredes, B., Nikolov, S., Tomasev, N., Blackwell, S., Askham, H., Glorot, X., O’Donoghue, B., Visentin, D., et al. (2018) · Nature Medicine · doi:10.1038/s41591-018-0107-6 · Cited by 2,639 · cited 1×
“We found that using synthetic data outperformed both in-distribution baselines in the less skewed (with 1,000 labeled samples from hospitals 3 and 4) and more skewed setting (with only 100 labeled samples) while closing the fairness gap between hospitals.”
Self-citationRevisiting Unreasonable Effectiveness of Data in Deep Learning Era
Sun, C., Shrivastava, A., Singh, S., & Gupta, A. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.97 · Cited by 2,484 · cited 6×
“ImageNet-21K is publicly available at the ImageNet website (https://www.image-net.org/), but the JFT-300M dataset is not publicly available due to restrictions in the data sharing agreement.”
InfluentialArdila, D., Kiraly, A. P., Bharadwaj, S., Choi, B., Reicher, J. J., Peng, L., Tse, D., Etemadi, M., Ye, W., Corrado, G., et al. (2019) · Nature Medicine · doi:10.1038/s41591-019-0447-x · Cited by 2,167 · cited 1×
“The advent of machine learning (ML) in healthcare has led to many advances in various facets of care and in a wide range of applications (Ardila et al., 2019; De Fauw et al., 2018; Esteva et al., 2017).”
Frid-Adar, M., Diamant, I., Klang, E., Amitai, M., Goldberger, J., & Greenspan, H. (2018) · Neurocomputing · doi:10.1016/j.neucom.2018.09.013 · Cited by 1,938 · cited 3×
“Such models have been used to augment medical images for liver lesion classification 30 , classification of diabetic retinopathy from fundus images 31 and breast mass diagnosis in mammography 32 .”
Liu, X., Faes, L., Kale, A. U., Wagner, S. K., Fu, D. J., Bruynseels, A., Mahendiran, T., Moraes, G., Shamdas, M., Kern, C., et al. (2019) · The Lancet Digital Health · doi:10.1016/s2589-7500(19)30123-2 · Cited by 1,883 · cited 1×
“Despite these rapid methodological developments and the promise of transformative impact in different areas of healthcare (Liu et al., 2019), few of these approaches (if any) have achieved the ambitious goal of fostering clinical progress (Varoquaux and Cheplygina, 2022).”
Hay, R. J., Johns, N. E., Williams, H. C., Bolliger, I. W., Dellavalle, R. P., Margolis, D. J., Marks, R., Naldi, L., Weinstock, M. A., Wulf, S. K., et al. (2014) · Journal of Investigative Dermatology · doi:10.1038/jid.2013.446 · Cited by 1,706 · cited 1×
“For example, in dermatology, skin disease impacts at least 30% of people worldwide (Hay et al., 2014), leading to 85 million doctor visits a year and a cost of $75 billion (Lim et al.”
Improving predictive inference under covariate shift by weighting the log-likelihood function
Shimodaira, H. (2000) · Journal of Statistical Planning and Inference · doi:10.1016/s0378-3758(00)00115-4 · Cited by 1,678
Improved Denoising Diffusion Probabilistic Models
Nichol, A., & Dhariwal, P. (2021) · arXiv · doi:10.48550/arxiv.2102.09672 · Cited by 1,550 · cited 2×
“19,20,43 for the generation and the upsampler (only the radiology and dermatology datasets required higher-resolution images).”
Potential Biases in Machine Learning Algorithms Using Electronic Health Record Data
Gianfrancesco, M. A., Tamang, S., Yazdany, J., & Schmajuk, G. (2018) · JAMA Internal Medicine · doi:10.1001/jamainternmed.2018.3763 · Cited by 1,395 · cited 1×
“Similar findings were reported by Puyol-Antón et al. 53 in a cardiac segmentation task with respect to sex and racial biases, and by Gianfrancesco et al. 54 in a different modality (electronic health records) for patients with low socioeconomic status.”
GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models
Nichol, A., Dhariwal, P., Ramesh, A., Shyam, P., Mishkin, P., McGrew, B., Sutskever, I., & Chen, M. (2021) · arXiv · doi:10.48550/arxiv.2112.10741 · Cited by 1,357
Ensuring fairness in machine learning to advance health equity.
Rajkomar, A., Hardt, M., Howell, M. D., Corrado, G., & Chin, M. H. (2018) · Annals of Internal Medicine · doi:10.7326/m18-1990 · Cited by 1,341 · cited 1×
“Many scholars recently scrutinized ML systems and surfaced different types of biases that emerge through the ML pipeline, including problems due to data acquisition protocols, flawed human decision-making, missing features and label scarcity 52 .”
CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning
Rajpurkar, P., Irvin, J., Zhu, K., Yang, B., Mehta, H., Duan, T., Ding, D. Y., Bagul, A., Langlotz, C. P., Shpanskaya, K. S., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.05225 · Cited by 1,219 · cited 2×
“These datasets have been widely studied by the community (Larrazabal et al., 2020; Rajpurkar et al., 2017; Seyyed-Kalantari et al., 2021) for model development and fairness analyses.”
Classifier-Free Diffusion Guidance
Ho, J., & Salimans, T. (2022) · arXiv · doi:10.48550/arxiv.2207.12598 · Cited by 1,126 · cited 4×
“This allowed us to use classifier-free guidance 20 to make images more ‘canonical’ with respect to a given label or property.”
InfluentialBig Transfer (BiT): General Visual Representation Learning
Kolesnikov, A., Beyer, L., Zhai, X., Puigcerver, J., Yung, J., Gelly, S., & Houlsby, N. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-58558-7_29 · Cited by 1,039 · cited 2×
“(Kolesnikov et al., 2020) We consider baselines that use (1) different pretraining, (2) different heuristic augmentations, (3) resample the dataset and (4) use the focal loss.”
Big Transfer (BiT): General Visual Representation Learning
Kolesnikov, A., Beyer, L., Zhai, X., Puigcerver, J., Yung, J., Gelly, S., & Houlsby, N. (2019) · arXiv · doi:10.48550/arxiv.1912.11370 · Cited by 1,039 · cited 2×
“(Kolesnikov et al., 2020) We consider baselines that use (1) different pretraining, (2) different heuristic augmentations, (3) resample the dataset and (4) use the focal loss.”
Seyyed-Kalantari, L., Zhang, H., McDermott, M. B. A., Chen, I. Y., & Ghassemi, M. (2021) · Nature Medicine · doi:10.1038/s41591-021-01595-0 · Cited by 928 · cited 3×
“In a similar vein, Seyyed-Kalantari et al. 13 further explored how differences in age, race or ethnicity, and insurance type (as a proxy of socioeconomic status) are manifested in the performance of a classifier operating on chest radiographs.”
A deep learning system for differential diagnosis of skin diseases
Liu, Y., Jain, A., Eng, C., Way, D. H., Lee, K., Bui, P., Kanada, K., de Oliveira Marinho, G., Gallegos, J., Gabriele, S., et al. (2020) · Nature Medicine · doi:10.1038/s41591-020-0842-3 · Cited by 853 · cited 1×
“We found that using synthetic data outperformed both in-distribution baselines in the less skewed (with 1,000 labeled samples from hospitals 3 and 4) and more skewed setting (with only 100 labeled samples) while closing the fairness gap between hospitals.”
Domain Generalization via Invariant Feature Representation
Muandet, K., Balduzzi, D., & Schölkopf, B. (2013) · arXiv · doi:10.48550/arxiv.1301.2115 · Cited by 771 · cited 1×
“Previous work on out-of-distribution generalization (Albuquerque et al., 2019; Ben-David et al., 2010; Muandet et al., 2013) has pointed out that several factors can affect the performance of a model on samples from domains beyond the training data.”
Synthetic data in machine learning for medicine and healthcare
Chen, R. J., Lu, M. Y., Chen, T. Y., Williamson, D. F. K., & Mahmood, F. (2021) · Nature Biomedical Engineering · doi:10.1038/s41551-021-00751-8 · Cited by 728 · cited 1×
“Given the ethical questions around the use of synthetic images in medicine and healthcare 46,49 , it is important to make a distinction between using generative models to augment the original training dataset and replacing real images with synthetic ones, especially in the absence of privacy…”
Principles of Risk Minimization for Learning Theory
Vapnik, V. N. (1991) · neural information processing systems · Cited by 692
Larrazabal, A. J., Nieto, N., Peterson, V., Milone, D. H., & Ferrante, E. (2020) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1919012117 · Cited by 685 · cited 3×
“One of the first studies to dive into the effect of training data composition on model performance across the sexes when using chest X-rays to diagnose thoracic diseases was the one led by Larrazabal et al. 12 .”
Machine learning for medical imaging: methodological failures and recommendations for the future
Varoquaux, G., & Cheplygina, V. (2022) · npj Digital Medicine · doi:10.1038/s41746-022-00592-y · Cited by 656 · cited 2×
“For example, as Varoquaux et al. 10 high-light, clinician-level performance is often overstated without validating models OOD.”
Tellez, D., Litjens, G., Bándi, P., Bulten, W., Bokhorst, J.-M., Ciompi, F., & van der Laak, J. (2019) · Medical Image Analysis · doi:10.1016/j.media.2019.101544 · Cited by 649 · cited 2×
“Hue and saturation were sufficient augmentations to achieve the highest-quality results by Tellez et al. 56 ; hence, we did not evaluate other heuristic augmentations.”
Bandi, P., Geessink, O., Manson, Q., Van Dijk, M., Balkenhol, M., Hermsen, M., Ehteshami Bejnordi, B., Lee, B., Paeng, K., Zhong, A., et al. (2019) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2018.2867350 · Cited by 569 · cited 2×
“The cancer metastases in lymph nodes challenge (CAMELYON17) by Bandi et al. 21 aims to improve generalization capabilities of automated solutions and reduce the workload on pathologists who have to manually label those cases.”
Artificial intelligence and computational pathology
Cui, M., & Zhang, D. Y. (2021) · Laboratory Investigation · doi:10.1038/s41374-020-00514-0 · Cited by 544 · cited 1×
“classification, prediction, and prognostication of diseases (Cui and Zhang, 2021).”
The burden of skin disease in the United States
Lim, H. W., Collins, S. A. B., Resneck, J. S., Bolognia, J. L., Hodge, J. A., Rohrer, T. A., Van Beek, M. J., Margolis, D. J., Sober, A. J., Weinstock, M. A., et al. (2017) · Journal of the American Academy of Dermatology · doi:10.1016/j.jaad.2016.12.043 · Cited by 520 · cited 1×
“, 2014), leading to 85 million doctor visits a year and a cost of $75 billion (Lim et al., 2017) in the US alone.”
WILDS: A Benchmark of in-the-Wild Distribution Shifts
Koh, P. W., Sagawa, S., Marklund, H., Xie, S. M., Zhang, M., Balsubramani, A., Hu, W., Yasunaga, M., Phillips, R. L., Gao, I., et al. (2020) · arXiv · doi:10.48550/arxiv.2012.07421 · Cited by 518 · cited 3×
“Both experimental design decisions, that is, the image resolution and the classification setup, were made to align with the WILDS challenge 22 and the wealth of literature that evaluates ML methods on in-the-wild distribution shifts https://doi.org/10.1038/s41591-024-02838-6 using the same setting…”
InfluentialCascaded Diffusion Models for High Fidelity Image Generation
Ho, J., Saharia, C., Chan, W., Fleet, D. J., Norouzi, M., & Salimans, T. (2021) · arXiv · doi:10.48550/arxiv.2106.15282 · Cited by 514
End-to-end privacy preserving deep learning on multi-institutional medical imaging
Kaissis, G., Ziller, A., Passerat-Palmbach, J., Ryffel, T., Usynin, D., Trask, A., Lima, I., Mancuso, J., Jungmann, F., Steinborn, M.-M., et al. (2021) · Nature Machine Intelligence · doi:10.1038/s42256-021-00337-8 · Cited by 513 · cited 1×
“There is preliminary evidence that federated learning can be used to learn classification models from multiple institutions 26 ; if it were possible to generate private synthetic data, these synthetic data could be used for data augmentation along with a smaller, public dataset to improve…”
Shifting machine learning for healthcare from development to deployment and from models to data
Zhang, A., Xing, L., Zou, J., & Wu, J. C. (2022) · Nature Biomedical Engineering · doi:10.1038/s41551-022-00898-y · Cited by 420 · cited 2×
“None of these works claimed that the latter would be preferable, but rather came to the rescue when obtaining more abundant real data is either expensive or not feasible (for example, in the case of rare conditions), even if this solution is not a panacea 42 .”
Generative Adversarial Networks in Medical Image augmentation: A review
Chen, Y., Yang, X.-H., Wei, Z., Heidari, A. A., Zheng, N., Li, Z., Chen, H., Hu, H., Zhou, Q., & Guan, Q. (2022) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2022.105382 · Cited by 414 · cited 2×
“Given that pixel-wise downstream tasks were not within the scope of our study, we refer the reader to a more thorough review of GAN-based methods in medical https://doi.org/10.1038/s41591-024-02838-6 image augmentation by Chen et al. 40 ; Bissoto et al. 41 , in turn, provide an overview of GAN-based augmentation techn…”
InfluentialCheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison
Irvin, J., Rajpurkar, P., Ko, M., Yu, Y., Ciurea-Ilcus, S., Chute, C., Marklund, H., Haghgoo, B., Ball, R., Shpanskaya, K., et al. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.3301590 · Cited by 392
CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison
Irvin, J., Rajpurkar, P., Ko, M., Yu, Y., Ciurea-Ilcus, S., Chute, C., Marklund, H., Haghgoo, B., Ball, R., Shpanskaya, K., et al. (2019) · arXiv · doi:10.48550/arxiv.1901.07031 · Cited by 392 · cited 9×
“The original CheXpert training set contained positive, negative, uncertain and unmentioned labels.”
InfluentialWang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2019) · Advances in Computer Vision and Pattern Recognition · doi:10.1007/978-3-030-13969-8_18 · Cited by 371
Causality matters in medical imaging
Castro, D. C., Walker, I., & Glocker, B. (2020) · Nature Communications · doi:10.1038/s41467-020-17478-w · Cited by 364 · cited 1×
“This is because of data scarcity 16 , challenges in the acquisition strategies of evaluation datasets (for example, different imaging or screening protocols 10,17,18 ) and the limitations of evaluation metrics 10 .”
Denoising diffusion probabilistic models for 3D medical image generation
Khader, F., Müller-Franzes, G., Tayebi Arasteh, S., Han, T., Haarburger, C., Schulze-Hagen, M., Schad, P., Engelhardt, S., Baeßler, B., Foersch, S., et al. (2023) · Scientific Reports · doi:10.1038/s41598-023-34341-2 · Cited by 338
Causality matters in medical imaging
Castro, D. C., Walker, I. D., & Glocker, B. (2020) · RePEc: Research Papers in Economics · Cited by 288
Robust and data-efficient generalization of self-supervised machine learning for diagnostic imaging
Azizi, S., Culp, L., Freyberg, J., Mustafa, B., Baur, S., Kornblith, S., Chen, T., Tomasev, N., Mitrović, J., Strachan, P., et al. (2023) · Nature Biomedical Engineering · doi:10.1038/s41551-023-01049-7 · Cited by 250 · cited 1×
“25.”
Self-citationTime to reality check the promises of machine learning-powered precision medicine.
Wilkinson, J., Arnold, K. F., Murray, E. J., van Smeden, M., Carr, K., Sippy, R., de Kamps, M., Beam, A., Konigorski, S., Lippert, C., et al. (2020) · The Lancet Digital Health · doi:10.1016/s2589-7500(20)30200-4 · Cited by 233
A clarification of the nuances in the fairness metrics landscape
Castelnovo, A., Crupi, R., Greco, G., Regoli, D., Penco, I. G., & Cosentini, A. C. (2022) · Scientific Reports · doi:10.1038/s41598-022-07939-1 · Cited by 218 · cited 1×
“For radiology, we considered AUC parity, namely the parity of the area under the ROC for different demographic subgroups identified by the sensitive attribute A , which can be seen as the analog of equality of accuracy 64 .”
Synthesizing retinal and neuronal images with generative adversarial nets
Zhao, H., Li, H., Maurer-Stroh, S., & Cheng, L. (2018) · Medical Image Analysis · doi:10.1016/j.media.2018.07.001 · Cited by 210 · cited 1×
“Apart from whole-image downstream tasks, GAN-based augmentation techniques have been used to improve performance on pixel-wise classification tasks, for example, vessel contour segmentation on fundus images 38 and brain lesion segmentation 39 .”
Addressing fairness in artificial intelligence for medical imaging
Ricci Lara, M. A., Echeveste, R., & Ferrante, E. (2022) · Nature Communications · doi:10.1038/s41467-022-32186-3 · Cited by 204 · cited 3×
“The number of individuals associated with each racial label was as follows: white, 6,047; other, 1,623; white, non-Hispanic, 1,359; Asian, 1,254; unknown, 1,019; Black or African American, 557; race and ethnicity unknown, 513; other, Hispanic, 239; native Hawaiian or other Pacific Islander, 177; Asian, non-Hispanic, 1…”
A Fine-Grained Analysis on Distribution ShiftPreprint
Wiles, O., Gowal, S., Stimberg, F., Rebuffi, S.-A., Ktena, S. I., Dvijotham, K., & Cemgil, A. T. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.11328 · Cited by 196 · cited 2×
“For the sake of our experiments, we used empirical risk minimization 57 because there is no single method that consistently outperforms it under distribution shifts 55 .”
Self-citationRadiologist shortage leaves patient care at risk, warns royal college
Rimmer, A. (2017) · BMJ · doi:10.1136/bmj.j4683 · Cited by 193 · cited 1×
“, in the case of radiologists (Rimmer, 2017), and demonstrate that machine learning models can facilitate detection of conditions (Rajpurkar et al.”
Jain, A., Way, D., Gupta, V., Gao, Y., de Oliveira Marinho, G., Hartford, J., Sayres, R., Kanada, K., Eng, C., Nagpal, K., et al. (2021) · JAMA Network Open · doi:10.1001/jamanetworkopen.2021.7249 · Cited by 164 · cited 1×
“, 2020) have the potential to allow patients to assess their conditions better and improve diagnostic accuracy (Jain et al., 2021).”
RoentGen: Vision-Language Foundation Model for Chest X-ray GenerationPreprint
Chambon, P. J., Blüthgen, C., Delbrouck, J.-B., van der Sluijs, R., Połacin, M., Chaves, J., Abraham, T., Purohit, S., Langlotz, C., & Chaudhari, A. (2022) · arXiv.org · doi:10.48550/arxiv.2211.12737 · Cited by 157 · cited 2×
“, 2022) and demonstrated that they can be conditioned on text prompts for chest X-ray generation (Chambon et al., 2022).”
Skin Lesion Classification Using GAN based Data Augmentation
Rashid, H., Tanveer, M. A., & Aqeel Khan, H. (2019) · 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc.2019.8857905 · Cited by 153 · cited 1×
“, 2014), have been employed by various studies to improve performance in different medical imaging tasks (Baur et al., 2018; Frid-Adar et al., 2018; Ju et al., 2021; Li et al., 2019; Rashid et al., 2019) and, in particular, for underrepresented conditions (Han et al.”
GAN-Based Data Augmentation and Anonymization for Skin-Lesion Analysis: A Critical Review
Bissoto, A. E., Junior, E. A. D. V., & Avila, S. (2021) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw53098.2021.00204 · Cited by 137
Generating Highly Realistic Images of Skin Lesions with GANs
Baur, C., Albarqouni, S., & Navab, N. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-01201-4_28 · Cited by 118
The Pitfalls of Simplicity Bias in Neural Networks
Shah, H., Tamuly, K., Raghunathan, A., Jain, P., & Netrapalli, P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07710 · Cited by 93 · cited 2×
“This phenomenon is linked to the underlying mechanisms that lead models models to rely on spurious correlations to make predictions [1].”
BackgroundMethodMedical Diffusion: Denoising Diffusion Probabilistic Models for 3D Medical Image GenerationPreprint
Khader, F., Mueller-Franzes, G., Arasteh, S. T., Han, T., Haarburger, C., Schulze-Hagen, M., Schad, P., Engelhardt, S., Baessler, B., Foersch, S., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2211.03364 · Cited by 89 · cited 1×
“Other works extended diffusion models to 3D MR and CT images (Khader et al., 2022) and demonstrated that they can be conditioned on text prompts for chest X-ray generation (Chambon et al.”
Preventing dataset shift from breaking machine-learning biomarkers
Dockès, J., Varoquaux, G., & Poline, J.-B. (2021) · GigaScience · doi:10.1093/gigascience/giab055 · Cited by 86 · cited 1×
“This is because of data scarcity 16 , challenges in the acquisition strategies of evaluation datasets (for example, different imaging or screening protocols 10,17,18 ) and the limitations of evaluation metrics 10 .”
Puyol-Antón, E., Ruijsink, B., Mariscal Harana, J., Piechnik, S. K., Neubauer, S., Petersen, S. E., Razavi, R., Chowienczyk, P., & King, A. P. (2022) · Frontiers in Cardiovascular Medicine · doi:10.3389/fcvm.2022.859310 · Cited by 85 · cited 1×
“Similar findings were reported by Puyol-Antón et al. 53 in a cardiac segmentation task with respect to sex and racial biases, and by Gianfrancesco et al. 54 in a different modality (electronic health records) for patients with low socioeconomic status.”
Generalizing to unseen domains via distribution matching
Albuquerque, I., Monteiro, J., Darvishi-Bayazi, M.-J., Falk, T. H., & Mitliagkas, I. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.00804 · Cited by 81 · cited 1×
“Previous work on OOD generalization 65–67 pointed out that several factors can affect the performance of a model on samples from domains beyond the training data.”
Self-citationRisk of Training Diagnostic Algorithms on Data with Demographic Bias
Abbasi-Sureshjani, S., Raumanns, R., Michels, B. E. J., Schouten, G., & Cheplygina, V. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-61166-8_20 · Cited by 71
“Fairness for unobserved characteristics: insights from technological impacts on queer communities”.
Tomasev, N., McKee, K. R., Kay, J., & Mohamed, S. (2021) · Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3461702.3462540 · Cited by 70 · cited 1×
“Sensitive attributes are not always observed or explicitly tracked and reported (Tomasev et al., 2021), often to protect people’s privacy.”
Ju, L., Wang, X., Zhao, X., Bonnington, C. P., Drummond, T. W., & Ge, Z. (2021) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2021.3056395 · Cited by 64
Poulain, R., Bin Tarek, M. F., & Beheshti, R. (2023) · 2023 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3593013.3594102 · Cited by 63
Breaking medical data sharing boundaries by using synthesized radiographs
Han, T., Nebelung, S., Haarburger, C., Horst, N., Reinartz, S., Merhof, D., Kiessling, F., Schulz, V., & Truhn, D. (2020) · Science Advances · doi:10.1126/sciadv.abb7973 · Cited by 61 · cited 1×
“In dermoscopic imaging 33 , a progressive generative model was introduced to produce realistic high-resolution synthetic images, while 34 focused on improving balanced multiclass accuracy and, in particular, sensitivity for high-risk underrepresented diagnostic labels like melanoma 37 .”
Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?
Bommasani, R., Creel, K. A., Kumar, A., Jurafsky, D., & Liang, P. (2022) · arXiv · doi:10.48550/arxiv.2211.13972 · Cited by 60 · cited 1×
“Inspired by a recent study by Bommasani et al. 70 that looked at how often the same individuals are underserved by ML models that have been trained on the same data, we investigated whether the same individuals with high-risk conditions were consistently misclassified.”
Medical domain knowledge in domain-agnostic generative AI
Kather, J. N., Ghaffari Laleh, N., Foersch, S., & Truhn, D. (2022) · npj Digital Medicine · doi:10.1038/s41746-022-00634-5 · Cited by 57 · cited 2×
“Given the ethical questions around the use of synthetic images in medicine and healthcare 46,49 , it is important to make a distinction between using generative models to augment the original training dataset and replacing real images with synthetic ones, especially in the absence of privacy…”
Achieving Robustness in the Wild via Adversarial Mixing With Disentangled Representations
Gowal, S., Qin, C., Huang, P.-S., Cemgil, T., Dvijotham, K., Mann, T., & Kohli, P. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.00129 · Cited by 46
Self-citationConditional generation of medical images via disentangled adversarial inference
Havaei, M., Mao, X., Wang, Y., & Lao, Q. (2021) · Medical Image Analysis · doi:10.1016/j.media.2021.102106 · Cited by 37 · cited 2×
“Generative models, especially generative adversarial networks (GANs) 29 , have been used by several studies to improve performance in different medical imaging tasks 30–34 and, in particular, for underrepresented conditions 35 .”
Diagnosing Failures of Fairness Transfer Across Distribution Shift in Real-World Medical Settings
Schrouff, J., Harris, N., Koyejo, S., Alabdulmohsin, I., Schnider, E., Opsahl-Ong, K., Brown, A. P. Y., Roy, S., Mincu, D., Chen, C., et al. (2022) · neural information processing systems · doi:10.52202/068431-1403 · Cited by 35 · cited 2×
“This is impressive as Schrouff et al. 18 demonstrated that improving fairness on in-distribution datasets does not guarantee performance improvements on OOD datasets.”
Self-citationDiagnosing failures of fairness transfer across distribution shift in real-world medical settingsPreprint
Schrouff, J., Harris, N., Koyejo, O. O., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.01034 · Cited by 32
Signed Laplacian Deep Learning with Adversarial Augmentation for Improved Mammography Diagnosis
Li, H., Chen, D., Nailon, W. H., Davies, M. E., & Laurenson, D. I. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-32226-7_54 · Cited by 18
Uzunova, H., Ehrhardt, J., & Heinz Handels, A. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-59719-1_49 · Cited by 15 · cited 2×
“Apart from whole-image downstream tasks, GAN-based augmentation techniques have been used to improve performance on pixel-wise classification tasks, for example, vessel contour segmentation on fundus images 38 and brain lesion segmentation 39 .”
Classify epithelium‐stroma in histopathological images based on deep transferable network
Yu, X., ZHENG, H., Liu, C., Huang, Y., & Ding, X. (2018) · Journal of Microscopy · doi:10.1111/jmi.12705 · Cited by 15 · cited 1×
“, 2021), histopathology (Yu et al., 2018) and dermatology (Abbasi-Sureshjani et al.”
Anton, E. P., Ruijsink, B., Harana, J. M., et al. (2021) · medRxiv · doi:10.1101/2021.07.19.21260749 · Cited by 13
Online Graph Completion: Multivariate Signal Recovery in Computer Vision
Kim, W. H., Jalal, M., Hwang, S., Johnson, S. C., & Singh, V. (2017) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2017.533 · Cited by 6
Conditional Generation of Medical Images via Disentangled Adversarial Inference
Havaei, M., Mao, X., Wang, Y., & Lao, Q. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88210-5_5 · Cited by 4
Extracting Training Data from Diffusion Models
Carlini, N., Hayes, J., Nasr, M., Jagielski, M., Sehwag, V., Tramèr, F., Balle, B., Ippolito, D., & Wallace, E. (2023) · arXiv · doi:10.48550/arxiv.2301.13188 · cited 2×
“While some studies view generative models as a means of replacing real data with ‘anonymized’ synthetic data, we abstain from such claims because greater care needs to be taken to ensure that generative models are trained with privacy guarantees, as shown by Carlini et al. 50 and Somepalli et al. 51 .”
Research works printed but not linked: 11
- Irvin, J. et al. Proc. AAAI Conference on Artificial Intelligence (AAAI, 2019).
- Tomasev, N., McKee, K. R., Kay, J. & Mohamed, S. Proc. 2021 AAAI/ACM Conference on AI, Ethics, and Society (Association for Computing Machinery, 2021).
- Goodfellow, I. et al. Generative adversarial nets. In Advances in Neural Information Processing Systems (eds. Ghahramani, Z. et al.) 2672–2680 (2014).
- Rashid, H., Tanveer, M. A. & Khan, H. A. Proc. 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (IEEE, 2019).
- Gowal, S. et al. Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (IEEE, 2020).
- Bissoto, A., Valle, E. & Avila, S. Proc. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (IEEE, 2021).
- Somepalli, G., Singla, V., Goldblum, M., Geiping, J., & Goldstein, T. Proc. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (IEEE, 2023).
- Sun, C., Shrivastava, A., Singh, S. & Gupta, A. Proc. 2017 IEEE International Conference on Computer Vision (ICCV) (IEEE, 2017).
- Deng, J. et al. Proc. 2009 IEEE Conference on Computer Vision and Pattern Recognition (IEEE, 2009).
- Cubuk, E. D., Zoph, B., Shlens, J. & Le, Q. V. Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops (CVPRW) (IEEE, 2020).
- Lin, T.-Y., Goyal, P., Girshick, R., He, K. & Dollár, P. Proc. IEEE International Conference on Computer Vision (ICCV) (IEEE, 2017).
Cited by
Ibrahim, M., Khalil, Y. A., Amirrajab, S., Sun, C., Breeuwer, M. M., Pluim, J. P. W., Elen, B., Ertaylan, G., & Dumontier, M. (2025) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2025.109834 · Cited by 223
Wang, J., Wang, K., Yu, Y., Lu, Y., Xiao, W., Sun, Z., Liu, F., Zou, Z., Gao, Y., Yang, L., et al. (2024) · Nature Medicine · doi:10.1038/s41591-024-03359-y · Cited by 231
“…AI can synthesize high-quality chest X-ray images 5–7 , morphologically preserving three-dimensional (3D) images of the brain 8 and two-dimensional (2D) images of histopathology 9,10 and dermatology 10,11 , and consequently enhancing medical image understanding and improving downstream analysis.”
Koçak, B., Ponsiglione, A., Stanzione, A., Bluethgen, C., Santinha, J., Ugga, L., Huisman, M., Klontzas, M. E., Cannella, R., & Cuocolo, R. (2024) · Diagnostic and Interventional Radiology · doi:10.4274/dir.2024.242854 · Cited by 168
“68 71-73 An overview of bias avoidance strategies at the data processing phase is presented in Figure 10.”
Generative AI for Medical Imaging: extending the MONAI FrameworkPreprint
Pinaya, W. H. L., Graham, M. S., Kerfoot, E., Tudosiu, P.-D., Dafflon, J., Fernández, V., Sanchez, P., Wolleb, J., da Costa, P. F., Patel, A., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2307.15208 · Cited by 214
“Furthermore, machine learning models trained to perform a downstream task using datasets that also include synthetic data have achieved similar, or even superior performance, compared to models trained with only real data [2,12,48,30].”
ResultA vision–language foundation model for the generation of realistic chest X-ray images
Bluethgen, C., Chambon, P., Delbrouck, J.-B., van der Sluijs, R., Połacin, M., Zambrano Chaves, J. M., Abraham, T. M., Purohit, S., Langlotz, C. P., & Chaudhari, A. S. (2024) · Nature Biomedical Engineering · doi:10.1038/s41551-024-01246-y · Cited by 108
Metadata
- Authors
- Ktena, I., Wiles, O., Albuquerque, I., Rebuffi, S.-A., Tanno, R., Roy, A. G., Azizi, S., Belgrave, D., Kohli, P., Cemgil, T., Karthikesalingam, A., & Gowal, S.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Google (United Kingdom) (GB)
- GlaxoSmithKline (United Kingdom) (GB)
- Corresponding author
- Wiles, Olivia
- ORCID
- Venue
- Nature Medicine
- Published
- 2024-04
- Volume
- 30
- Issue
- 4
- Pages
- 1166-1173
- ISSN
- 1078-8956 (ISSN-L)
- 1546-170X
- PMC
- PMC11031395
- OpenAlex
- W4394683903
- Cited by
- 165
- References
- 73
- Citations by year
- 2026: 58 · 2025: 78 · 2024: 28
- FWCI
- 15.36
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor, 9 d)
- Locations
- Nature Medicine (publishedVersion, cc-by)
- Research Square (acceptedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › COVID-19 diagnosis using AI
- Keywords
- generative models, diffusion models, domain generalization, distribution shift, algorithmic fairness, synthetic data augmentation, underrepresented groups, medical image classification, histopathology, chest radiography, dermatological images, model robustness, label efficiency
- MeSH terms
- Machine Learning, Artificial Intelligence
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.