Privacy-Net: An Adversarial Approach for Identity-Obfuscated Segmentation of Medical Images

TL;DR

Privacy-Net introduces an adversarial framework that obfuscates patient identity in medical images while maintaining high segmentation accuracy, enhancing privacy in multicentric medical image analysis.

Contribution

It proposes a novel adversarial architecture with an encoder, discriminator, and analysis network to remove identity features while preserving task-relevant information.

Findings

Discriminator effectively distorts images to hide identity.

High segmentation accuracy maintained despite obfuscation.

Applicable to large-scale Parkinson's MRI dataset.

Abstract

This paper presents a client/server privacy-preserving network in the context of multicentric medical image analysis. Our approach is based on adversarial learning which encodes images to obfuscate the patient identity while preserving enough information for a target task. Our novel architecture is composed of three components: 1) an encoder network which removes identity-specific features from input medical images, 2) a discriminator network that attempts to identify the subject from the encoded images, 3) a medical image analysis network which analyzes the content of the encoded images (segmentation in our case). By simultaneously fooling the discriminator and optimizing the medical analysis network, the encoder learns to remove privacy-specific features while keeping those essentials for the target task. Our approach is illustrated on the problem of segmenting brain MRI from the large-scale Parkinson Progression Marker Initiative (PPMI) dataset. Using longitudinal data from PPMI, we show that the discriminator learns to heavily distort input images while allowing for highly accurate segmentation results.