Local Differential Privacy Image Generation Using Flow-based Deep Generative Models

TL;DR

This paper introduces DP-GLOW, a novel method combining local differential privacy with flow-based generative models to protect medical image privacy while maintaining diagnostic features.

Contribution

We propose DP-GLOW, a hybrid model that applies LDP to latent vectors of a flow-based model, enabling privacy-preserving medical image generation without losing critical pathologies.

Findings

Successfully generated privacy-preserving chest X-ray images.

Preserved key pathological features in LDP images.

Demonstrated effective privacy protection in medical imaging.

Abstract

Diagnostic radiologists need artificial intelligence (AI) for medical imaging, but access to medical images required for training in AI has become increasingly restrictive. To release and use medical images, we need an algorithm that can simultaneously protect privacy and preserve pathologies in medical images. To develop such an algorithm, here, we propose DP-GLOW, a hybrid of a local differential privacy (LDP) algorithm and one of the flow-based deep generative models (GLOW). By applying a GLOW model, we disentangle the pixelwise correlation of images, which makes it difficult to protect privacy with straightforward LDP algorithms for images. Specifically, we map images onto the latent vector of the GLOW model, each element of which follows an independent normal distribution, and we apply the Laplace mechanism to the latent vector. Moreover, we applied DP-GLOW to chest X-ray images to generate LDP images while preserving pathologies.