Generative Medical Image Anonymization Based on Latent Code Projection and Optimization

2025-01-15

AI Summary

Balancing medical image privacy protection with data utility remains challenging. To address this, we propose a two-stage generative anonymization method: first, a lightweight encoder maps images into a latent space with co-optimized representations; second, a dual deep loss function—comprising identity confusion loss and medical semantic fidelity loss—jointly optimizes latent codes to explicitly trade off irreversible de-identification against diagnostic accuracy. Our approach introduces the first latent-space joint projection and dual-objective co-optimization framework tailored for medical imaging, and constitutes the first verifiable anonymization scheme for medical images. Evaluated on the MIMIC-CXR dataset, it generates high-fidelity anonymized chest radiographs, achieving an AUC of 0.892 for pulmonary pathology detection—3.7% higher than the state of the art—and demonstrates zero identity leakage under third-party face recognition attacks.

Technology Category

• Machine Learning: Privacy
• Computer Vision: Medical and Biological Imaging
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

Medical image anonymization aims to protect patient privacy by removing identifying information, while preserving the data utility to solve downstream tasks. In this paper, we address the medical image anonymization problem with a two-stage solution: latent code projection and optimization. In the projection stage, we design a streamlined encoder to project input images into a latent space and propose a co-training scheme to enhance the projection process. In the optimization stage, we refine the latent code using two deep loss functions designed to address the trade-off between identity protection and data utility dedicated to medical images. Through a comprehensive set of qualitative and quantitative experiments, we showcase the effectiveness of our approach on the MIMIC-CXR chest X-ray dataset by generating anonymized synthetic images that can serve as training set for detecting lung pathologies. Source codes are available at https://github.com/Huiyu-Li/GMIA.

Problem

• Privacy Protection
• Medical Imaging
• Research Utilization

Innovation

• Medical Image Anonymization
• Improved Cryptographic Technique
• Privacy Protection and Usability