CryptoUNets: Applying Convolutional Networks to Encrypted Data for Biomedical Image Segmentation
John Chiang
TL;DR
This work demonstrates the feasibility of performing biomedical image segmentation purely in the encrypted domain by conducting U-Net inference with Fully Homomorphic Encryption. It introduces a practical encoding approach, Double Volley Revolver, to handle large matrices and supports skip connections and upsampling within an encrypted pipeline, using an $HEAAN$-based implementation. The authors provide an HE-friendly U-Net design (square activations, mean pooling, ConvTranspose2d) trained in plaintext and deployed on encrypted data, and validate the approach on the ISBI 2012 EM Segmentation dataset with detailed hardware and parameter settings, achieving encryped-inference capabilities with tangible latency and memory usage. The work advances privacy-preserving medical imaging by expanding encrypted-inference applicability to more complex architectures and highlights directions for performance optimization and encrypted training in future research.
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.