Secure-by-Disguise: A Systematic Evaluation of Image Disguising for Confidential Medical Image Modeling

2026-07-09

AI Summary

This study addresses the privacy risks associated with outsourcing medical images in cloud environments by establishing a unified evaluation framework to systematically analyze the predictive utility, computational efficiency, and resilience against reconstruction attacks of image disguising techniques in medical image classification and semantic segmentation tasks. The experimental evaluation encompasses representative methods such as DisguisedNets and NeuraCrypt, including Randomized Multi-dimensional Transform (RMT) and AES encryption, with a novel regression-based reconstruction attack introduced for security validation. Results demonstrate that RMT achieves the best trade-off between privacy preservation and model utility, whereas AES severely degrades performance. While disguising techniques remain effective for classification, they incur substantial performance drops in dense prediction tasks like semantic segmentation. The findings also reveal that medical images exhibit strong robustness against existing reconstruction attacks, offering critical empirical evidence for privacy-preserving medical AI.

Abstract

Cloud-based deep learning enables large-scale medical image analysis but raises significant privacy concerns when sensitive patient images are outsourced for model development. Image disguising has recently emerged as a promising privacy-enhancing technology (PET) that transforms images into visually unintelligible representations while preserving information for downstream learning. We established a unified framework to evaluate representative methods, DisguisedNets and NeuraCrypt, across four datasets involving classification and semantic segmentation tasks. Our analysis assessed predictive utility, efficiency, and robustness against reconstruction attacks. Results showed that image disguising performance varies significantly between tasks; while methods preserved utility for medical image classification, they caused substantial degradation in dense semantic segmentation. Specifically, Randomized Multidimensional Transformation (RMT) offered the optimal balance of performance and security, whereas AES-based disguising severely impacted utility. Furthermore, regression-based reconstruction attacks effective on natural images proved considerably less successful on realistic medical images. These findings provide a systematic assessment of PET suitability for confidential medical AI applications.

Problem

• medical image privacy
• image disguising
• privacy-enhancing technology
• cloud-based deep learning
• confidential medical AI

Innovation

• image disguising
• privacy-enhancing technology
• medical image analysis
• reconstruction attack
• Randomized Multidimensional Transformation