Generative AI for Rare Disease Data Augmentation
Keywords:
Generative AI Models, Data Sources, Generative Model Types, Data Generation Metrics, Biases, and Capability, Generative AI Model Evaluation in Healthcare.Abstract
Shallow Learning agents require abundant labelled data to excel. In rare diseases, where patient population is often limited to a few hundred patients, such a data source is not readily available. Generative AI models can synthesize new, similar data by learning an approximate distribution of the original data. Their application in rare disease cohorts and clinical data is detailed, with supervisory signals derived from the data labelling mechanisms. The response from Generative AI models is further validated by comparing model predictions with real-world outcomes.
Rare diseases are conditions that affect a small percentage of the population — definitions suggest a prevalence of below 1:2000 but this varies among nations. Globally, a collection of more than 7000 rare diseases, affecting 25 million patients, has been enumerated, with the rapid emergence of new entities identified. Early diagnosis, clinical management, and developing novel therapeutics for rare diseases offer challenging problems with limited labelled data-sets available to Deep Learning agents. Synthetic Data Generation, by augmentation or providing an alternative data source, enables Shallow Learning agents to attempt these tasks across phenotype, genotype, and narrative data.
References
[1] Achterberg, J. L., Koops, R., & de Bruijne, M. (2024). On the evaluation of synthetic longitudinal electronic health records. Journal of the American Medical Informatics Association.
[2]Alabdulwahab, S., Alotaibi, B., & Alotaibi, S. (2024). Privacy-preserving synthetic data generation method using conditional tabular generative adversarial networks (CTGAN). Sensors, 24(22).
[3]Apellániz, P. A., Hummel, B., & García, A. (2025). Improving synthetic data generation through federated learning: Sharing synthetic patients across nodes. Journal of Personalized Medicine, 9(2).
[4]Baowaly, M. K., Lin, C.-C., Liu, K.-T., & Chen, K.-C. (2018). Synthesizing electronic health records using improved generative adversarial networks. Journal of the American Medical Informatics Association.
[5]Brophy, E., Wang, R., & Ward, T. (2023). Generative adversarial networks in time series: A survey and taxonomy. ACM Computing Surveys.
[6]Budu, E., & Nascimento, C. (2024). Evaluation of synthetic electronic health records: A review of fidelity, utility, and privacy metrics. Neurocomputing.
[7]Cao, Z., Li, Y., & Zhang, H. (2025). Diffusion model-based medical image generation as a potential data augmentation strategy for AI applications. Medical Image Analysis.
[8]Chadalawada, P. (2025). Balancing privacy and utility in electronic health records: A literature review. European Society of Medicine.
[9]Chen, J., Wang, X., & Li, J. (2025). Source-free cross-modality medical image synthesis with diffusion priors. Journal of Imaging Informatics in Medicine.
[10]Choi, E., Biswal, S., Malin, B., Duke, J., Stewart, W. F., & Sun, J. (2017). Generating multi-label discrete patient records using generative adversarial networks. In Proceedings of Machine Learning Research (Vol. 68).
[11]Finetti, R., & colleagues. (2025). Data augmentation and synthetic data generation in rare disease research: Methods, applications, and challenges. Journal of Clinical Medicine.
[12]Fang, M. L., & colleagues. (2022). Differentially private medical data generation using CTGANs (DP-CTGAN). In Proceedings of the Machine Learning Research community (workshop proceedings).
[13]Giuffrè, M., & colleagues. (2023). Harnessing the power of synthetic data in healthcare. npj Digital Medicine, 6.
[14]Grigoraș, A., & colleagues. (2024). Synthetic time series generation for decision intelligence: Transformer-based approaches and applications. Mathematics, 12(16).
[15]Hernadez, M., & colleagues. (2023). Synthetic tabular data evaluation in the health domain: Dimensions, metrics, and orchestration strategy. BMC Medical Informatics and Decision Making.
[16]Hu, C., & colleagues. (2023). Survey of time series data generation in IoT: Classical to state-of-the-art methods. Sensors.
[17]Hyrup, T., & colleagues. (2025). A systematic review of privacy-preserving techniques for synthetic tabular health data generation. AI and Ethics.
[18]Islam, T., & colleagues. (2024). A systematic review of deep learning data augmentation in medical imaging. Artificial Intelligence in Medicine.
[19]Isasa, I., & colleagues. (2025). Synthetic tabular data generation under horizontal federated learning for a rare hematological disorder (acute myeloid leukemia). Frontiers in Artificial Intelligence.
[20]Jeon, J., & colleagues. (2022). GT-GAN: General purpose time series synthesis with generative adversarial networks. Advances in Neural Information Processing Systems.
[21]Jordon, J., Yoon, J., & van der Schaar, M. (2019). PATE-GAN: Generating synthetic data with differential privacy guarantees. In International Conference on Learning Representations.
[22]Kabir, M. F., & colleagues. (2025). Synthetic data generation for emergency medical systems: Comparative evaluation of GAN- and DP-based approaches. In Proceedings of the International Conference on Data Science, Technology and Applications (DATA).
[23]Khosravi, B., & colleagues. (2025). Exploring the potential of generative artificial intelligence in medical imaging: Synthetic data opportunities and challenges. The Lancet Digital Health.
[24]Kita, F. J., & colleagues. (2025). Innovative synthetic EHR data generation: Diffusion models and evaluation considerations. Connection Science.
[25]Kurakova, A., & colleagues. (2025). A comprehensive evaluation framework for synthetic medical tabular data generation. Journal of Biomedical Informatics.
[26]Lange, L., & colleagues. (2024). Generating synthetic health sensor data for privacy-preserving stress detection: Utility–privacy trade-offs with DP-GANs. Sensors.
[27]Li, J., & colleagues. (2023). Generating synthetic mixed-type longitudinal electronic health records using EHR-M-GAN. npj Digital Medicine, 6.
[28]Lin, Z., Jain, A., & colleagues. (2019). Generating high-fidelity synthetic time series datasets with DoppelGANger. Proceedings of the ACM on Measurement and Analysis of Computing Systems.
[29]Machado, J., & colleagues. (2025). Data augmentation with generative methods for inherited retinal diseases. Applied Sciences, 15(6).
[30]Mallardi, G., & colleagues. (2025). Diffusion models for neuroimaging data augmentation with relevance to rare neurodegenerative disease research. Frontiers in Neuroscience.
[31]Naseer, A. A., & colleagues. (2023). ScoEHR: Generating synthetic electronic health records using continuous-time diffusion models. In Proceedings of Machine Learning Research.
[32]Qian, Z., & colleagues. (2024). Synthetic data for privacy-preserving clinical risk prediction. Scientific Reports, 14.
[33]Radford, A., Metz, L., & Chintala, S. (2016). Unsupervised representation learning with deep convolutional generative adversarial networks. In International Conference on Learning Representations.
[34]Rujas, M., & colleagues. (2024). Synthetic data generation in healthcare: A scoping review. International Journal of Medical Informatics.
[35]Shi, Y., & colleagues. (2025). Diffusion models for medical image computing: A survey. Tsinghua Science and Technology.
[36]Sun, C., & colleagues. (2023). Generating synthetic personal health data using a differentially private conditional GAN (DP-CGANS). Journal of Biomedical Informatics, 141.
[37]Sun, C., & colleagues. (2025). Generating unseen diseases’ patient data using ontology-guided conditional GANs (Onto-CGAN). npj Digital Medicine.
[38]Tian, M., & colleagues. (2024). Reliable generation of privacy-preserving synthetic electronic health record time series using diffusion models. Journal of the American Medical Informatics Association, 31(11).
[39]Torfi, A., & colleagues. (2022). Differentially private synthetic medical data generation using Rényi differential privacy. Information Sciences, 599.
[40]Walonoski, J., Kramer, M., Nichols, J., Quina, A., Moesel, C., Hall, D., Duffett, C., Dube, K., Gallagher, T., McLachlan, S., & others. (2018). Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health record. Journal of the American Medical Informatics Association, 25(3).
[41]Waseem, H. M., & colleagues. (2025). Review of generative AI for synthetic data: Methods, evaluation, privacy, and applications. Artificial Intelligence Review.
[42]Williams, M. C., & colleagues. (2025). Researchers’ perspectives on federated learning and synthetic data for cardiovascular imaging research. European Heart Journal – Imaging Methods and Practice.
[43]Xu, L., Skoularidou, M., Cuesta-Infante, A., & Veeramachaneni, K. (2019). Modeling tabular data using conditional GAN. Advances in Neural Information Processing Systems.
[44]Yale, A., Dash, S., Dutta, R., Guyon, I., Pavao, A., & Bennett, K. (2020). Generation and evaluation of privacy preserving synthetic health data. Neurocomputing, 416.
[45]Yan, C., & colleagues. (2022). A multifaceted benchmarking of synthetic electronic health record generation models. Nature Communications, 13.
[46]Yan, C., & colleagues. (2024). Generating synthetic electronic health record data using generative adversarial networks: A reproducible tutorial. JMIR AI, 3.
[47]Yang, Z., & colleagues. (2023). TransformEHR: A transformer-based generative encoder–decoder model for electronic health records. Nature Communications, 14.
[48]Yoon, J., Jarrett, D., & van der Schaar, M. (2019). Time-series generative adversarial networks. Advances in Neural Information Processing Systems.
[49]Zhong, Y., & colleagues. (2024). Synthesizing multimodal electronic health records via predictive diffusion models (EHRPD). Proceedings of the ACM Conference on Health, Inference, and Learning (CHIL).
[50]Fan, L., & colleagues. (2025). GADM: Data augmentation using a diffusion model–GAN hybrid for physiological signals. Biomedical Signal Processing and Control.
[51]Ryspayeva, M., & colleagues. (2025). Multi-domain synthetic medical image generation and augmentation for imbalanced datasets. Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization.
[52]Tian, M., & colleagues. (2023). Reliable generation of privacy-preserving synthetic electronic health record time series with denoising diffusion probabilistic models. arXiv preprint.
[53]Alatrista-Salas, H., & colleagues. (2022). Differentially private generative modeling for mobility trajectory data. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies.
[54]Harder, N., & colleagues. (2021). Differential privacy for deep generative models: Empirical analysis and best practices. Proceedings of the Privacy in Machine Learning workshops.
[55]Wang, Y., & colleagues. (2021). Differentially private image generation with deep generative models. Proceedings of machine learning privacy workshops.
[56]Bie, A., & colleagues. (2022). Privacy-preserving image synthesis with differentially private generative models. Proceedings of ML privacy workshops.
[57]Xie, L., Lin, K., Wang, S., Wang, F., & Zhou, J. (2018). Differentially private generative adversarial network. arXiv preprint.
[58]Brophy, E., & colleagues. (2021). Generation of synthetic electronic health records using a federated GAN. Proceedings of health data privacy workshops.
[59]Chen, R. J., Lu, M. Y., Chen, T. Y., Williamson, D. F. K., & Mahmood, F. (2022). Synthetic data in machine learning for medicine and healthcare. Nature Biomedical Engineering.
[60]Goncalves, A., Ray, P., Soper, B., Stevens, J., Coyle, L., & Sales, A. P. (2020). Generation and evaluation of synthetic patient data. BMC Medical Research Methodology, 20.
[61]El Emam, K., Mosquera, L., & Hoptroff, R. (2020). Practical synthetic data generation: Balancing privacy and utility in health data. Frontiers in Public Health, 8.
[62]Stadler, T., Oprisanu, B., & Troncoso, C. (2021). Synthetic data—Anonymisation groundhog day. Proceedings on Privacy Enhancing Technologies.
[63]Uhler, C., & colleagues. (2023). Privacy risks and auditing methods for synthetic health data. Patterns.
[64]Rajabi, A., & Garibay, I. (2021). TabFairGAN: Fairness-aware synthetic tabular data generation. Proceedings of AAAI/ACM AI ethics venues.
[65]Zhao, Z., & colleagues. (2022). CTabGAN+: Improving tabular GANs for mixed-type data synthesis. Proceedings of machine learning data synthesis venues.
[66]Rustad, R. (2022). Copula-based generative models for tabular synthetic data generation. Proceedings of data science venues.
[67]Rosenblatt, L., & colleagues. (2020). PATE-CTGAN: Private aggregation for tabular GAN-based synthesis. Proceedings of privacy-preserving ML venues.
[68]Choi, E., Bahadori, M. T., Schuetz, A., Stewart, W. F., & Sun, J. (2017). Doctor AI: Predicting clinical events via recurrent neural networks. Proceedings of Machine Learning for Healthcare.
[69]Miotto, R., Li, L., Kidd, B. A., & Dudley, J. T. (2016). Deep patient: An unsupervised representation to predict the future of patients from the electronic health records. Scientific Reports, 6.
[70]Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative adversarial nets. In Advances in Neural Information Processing Systems.
Additional Files
Published
Data Availability Statement
None
