mlsanityData to 7 Oct 2026

The Devil is in the Prompts: De-Identification Traces Enhance Memorization Risks in Synthetic Chest X-Ray Generation

arXiv 2502.07516 ↗First posted 11 Feb 2025eess.IV cs.AI cs.CVcited by 2

On the map: Diffusion models → Medical imaging and segmentation

Abstract (arXiv)

Generative models, particularly text-to-image (T2I) diffusion models, play a crucial role in medical image analysis. However, these models are prone to training data memorization, posing significant risks to patient privacy. Synthetic chest X-ray generation is one of the most common applications in medical image analysis with the MIMIC-CXR dataset serving as the primary data repository for this task. This study presents the first systematic attempt to identify prompts and text tokens in MIMIC-CXR that contribute the most to training data memorization. Our analysis reveals two unexpected findings: (1) prompts containing traces of de-identification procedures (markers introduced to hide Protected Health Information) are the most memorized, and (2) among all tokens, de-identification markers contribute the most towards memorization. This highlights a broader issue with the standard anonymization practices and T2I synthesis with MIMIC-CXR. To exacerbate, existing inference-time memorization mitigation strategies are ineffective and fail to sufficiently reduce the model's reliance on memorized text tokens. On this front, we propose actionable strategies for different stakeholders to enhance privacy and improve the reliability of generative models in medical imaging. Finally, our results provide a foundation for future work on developing and benchmarking memorization mitigation techniques for synthetic chest X-ray generation using the MIMIC-CXR dataset. The anonymized code is available at https://anonymous.4open.science/r/diffusion_memorization-8011/

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Medical imaging and segmentation, one of 14 directions in Diffusion models. Builds on 20 of its 28 references that are in the map. First posted on arXiv 11 Feb 2025.

Builds on

20 of 28 references.

Closest ancestors · 3

Same direction, newest first

  1. 2023Using Diffusion Models to Generate Synthetic Labelled Data for Medical Image Segmentation25 Oct
  2. 2023Investigating Data Memorization in 3D Latent Diffusion Models for Medical Image Synthesis3 Jul
  3. 2023Beware of diffusion models for synthesizing medical images -- A comparison with GANs in terms of memorizing brain MRI and chest x-ray images12 May

Other references · 17

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2020Denoising Diffusion Implicit Modelscited by 9,655
  3. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,290
  4. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
  5. 2023Extracting Training Data from Diffusion Modelscited by 723
  6. 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 418
  7. 2023Learning to Exploit Temporal Structure for Biomedical Vision-Language Processingcited by 210
  8. 2023Understanding and Mitigating Copying in Diffusion Modelscited by 205
  9. 2024Detecting, Explaining, and Mitigating Memorization in Diffusion Modelscited by 94
  10. 2023Generative models improve fairness of medical classifiers under distribution shiftscited by 90
  11. 2024Unveiling and Mitigating Memorization in Text-to-image Diffusion Models through Cross Attentioncited by 57
  12. 2023Parameter-Efficient Fine-Tuning for Medical Image Analysis: The Missed Opportunitycited by 54
5 more
  1. 2023On the De-duplication of LAION-2Bcited by 47
  2. 2023RadEdit: stress-testing biomedical vision models via diffusion image editingcited by 28
  3. 2023Privacy Distillation: Reducing Re-identification Risk of Multimodal Diffusion Modelscited by 18
  4. 2024From Majority to Minority: A Diffusion-based Augmentation for Underrepresented Groups in Skin Lesion Analysiscited by 9
  5. 2024Capacity Control is an Effective Memorization Mitigation Mechanism in Text-Conditional Diffusion Modelscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 1; 2026: 1.

2 papers.

  1. 2025CheXGenBench: A Unified Benchmark For Fidelity, Privacy and Utility of Synthetic Chest Radiographscited by 5
  2. 2026Diffusion Models Memorize in Training -- and Generalize in Inferencecited by 2

Similar, not cited

Close in wording, no citation either way

  1. 2024Unconditional Latent Diffusion Models Memorize Patient Imaging Data: Implications for Openly Sharing Synthetic Data
  2. 2026A Utility-preserving De-identification Pipeline for Cross-hospital Radiology Data Sharing
  3. 2024Towards Predicting Temporal Changes in a Patient's Chest X-ray Images based on Electronic Health Records
  4. 2026When Retrieval Hurts: Measuring and Explaining Retrieval-Induced Hallucination in Chest X-ray Report Generation
  5. 2024TiBiX: Leveraging Temporal Information for Bidirectional X-ray and Report Generation
  6. 2026What Makes a Good Medical Image Tokenizer? Rethinking Reconstruction and Generation in Medical Image Tokenization
  7. 2023Synthetically Enhanced: Unveiling Synthetic Data's Potential in Medical Imaging Research
  8. 2022Generation of Anonymous Chest Radiographs Using Latent Diffusion Models for Training Thoracic Abnormality Classification Systems