CryptoUNets: Applying Convolutional Networks to Encrypted Data for Biomedical Image Segmentation

2025-04-30

AI Summary

This work addresses the challenge of simultaneously ensuring privacy preservation and maintaining model functionality in biomedical image segmentation. We propose the first end-to-end homomorphic encryption (HE)-based inference framework for U-Net, enabling secure inference directly on encrypted data. To overcome key HE bottlenecks—specifically, modeling skip connections and upsampling operations—we introduce the Double Volley Revolver, a novel two-level dynamic encoding scheme, and design an HE-friendly U-Net variant featuring square activation functions, average pooling, and transposed convolution. Implemented atop the HEAAN library, our framework performs full segmentation inference entirely in the ciphertext domain. Experimental results demonstrate that it preserves the accuracy of the original U-Net while enabling verifiable, privacy-preserving deployment of medical AI. To the best of our knowledge, this is the first practical, fully HE-based deep learning inference solution for privacy-sensitive healthcare applications.

Technology Category

• Machine Learning: Privacy
• Computer Vision: Segmentation
• Philosophy and Ethics of AI: Privacy & Security

Application Category

• Security and Privacy: Privacy-enhancing technologies
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

In this manuscript, we demonstrate the feasibility of a privacy-preserving U-Net deep learning inference framework, namely, homomorphic encryption-based U-Net inference. That is, U-Net inference can be performed solely using homomorphic encryption techniques. To our knowledge, this is the first work to achieve support perform implement enable U-Net inference entirely based on homomorphic encryption ?. The primary technical challenge lies in data encoding. To address this, we employ a flexible encoding scheme, termed Double Volley Revolver, which enables effective support for skip connections and upsampling operations within the U-Net architecture. We adopt a tailored HE-friendly U-Net design incorporating square activation functions, mean pooling layers, and transposed convolution layers (implemented as ConvTranspose2d in PyTorch) with a kernel size of 2 and stride of 2. After training the model in plaintext, we deploy the resulting parameters using the HEAAN homomorphic encryption library to perform encrypted U-Net inference.

Problem

• Applying convolutional networks to encrypted biomedical image segmentation
• Enabling U-Net inference using homomorphic encryption techniques
• Addressing data encoding challenges for privacy-preserving deep learning

Innovation

• Homomorphic encryption-based U-Net inference
• Double Volley Revolver encoding scheme
• HE-friendly U-Net with square activations