Sensitivity-Aware Differential Privacy for Federated Medical Imaging

Federated learning (FL) enables collaborative model training across multiple institutions without the sharing of raw patient data, making it particularly suitable for smart healthcare applications. However, recent studies revealed that merely sharing gradients provides a false sense of security, as...

Full description

Saved in:
Bibliographic Details
Main Authors: Lele Zheng, Yang Cao, Masatoshi Yoshikawa, Yulong Shen, Essam A. Rashed, Kenjiro Taura, Shouhei Hanaoka, Tao Zhang
Format: Article
Language:English
Published: MDPI AG 2025-04-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/25/9/2847
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850279070523195392
author Lele Zheng
Yang Cao
Masatoshi Yoshikawa
Yulong Shen
Essam A. Rashed
Kenjiro Taura
Shouhei Hanaoka
Tao Zhang
author_facet Lele Zheng
Yang Cao
Masatoshi Yoshikawa
Yulong Shen
Essam A. Rashed
Kenjiro Taura
Shouhei Hanaoka
Tao Zhang
author_sort Lele Zheng
collection DOAJ
description Federated learning (FL) enables collaborative model training across multiple institutions without the sharing of raw patient data, making it particularly suitable for smart healthcare applications. However, recent studies revealed that merely sharing gradients provides a false sense of security, as private information can still be inferred through gradient inversion attacks (GIAs). While differential privacy (DP) provides provable privacy guarantees, traditional DP methods apply uniform protection, leading to excessive protection for low-sensitivity data and insufficient protection for high-sensitivity data, which degrades model performance and increases privacy risks. This paper proposes a new privacy notion, sensitivity-aware differential privacy, to better balance model performance and privacy protection. Our idea is that the sensitivity of each data sample can be objectively measured using real-world attacks. To implement this new notion, we develop the corresponding defense mechanism that adjusts privacy protection levels based on the variation in the privacy leakage risks of gradient inversion attacks. Furthermore, the method extends naturally to multi-attack scenarios. Extensive experiments on real-world medical imaging datasets demonstrate that, under equivalent privacy risk, our method achieves an average performance improvement of 13.5% over state-of-the-art methods.
format Article
id doaj-art-003eff9a4d2a4cfea1fb891ae5c7fd3d
institution OA Journals
issn 1424-8220
language English
publishDate 2025-04-01
publisher MDPI AG
record_format Article
series Sensors
spelling doaj-art-003eff9a4d2a4cfea1fb891ae5c7fd3d2025-08-20T01:49:14ZengMDPI AGSensors1424-82202025-04-01259284710.3390/s25092847Sensitivity-Aware Differential Privacy for Federated Medical ImagingLele Zheng0Yang Cao1Masatoshi Yoshikawa2Yulong Shen3Essam A. Rashed4Kenjiro Taura5Shouhei Hanaoka6Tao Zhang7School of Computer Science and Technology, Xidian University, Xi’an 710126, ChinaDepartment of Computer Science, Institute of Science Tokyo, Tokyo 152-8550, JapanFaculty of Data Science, Osaka Seikei University, Osaka 533-0007, JapanSchool of Computer Science and Technology, Xidian University, Xi’an 710126, ChinaGraduate School of Information Science, University of Hyogo, Hyogo 670-0092, JapanGraduate School of Information Science and Technology, University of Tokyo, Tokyo 113-0033, JapanGraduate School of Medicine, University of Tokyo, Tokyo 113-0033, JapanSchool of Computer Science and Technology, Xidian University, Xi’an 710126, ChinaFederated learning (FL) enables collaborative model training across multiple institutions without the sharing of raw patient data, making it particularly suitable for smart healthcare applications. However, recent studies revealed that merely sharing gradients provides a false sense of security, as private information can still be inferred through gradient inversion attacks (GIAs). While differential privacy (DP) provides provable privacy guarantees, traditional DP methods apply uniform protection, leading to excessive protection for low-sensitivity data and insufficient protection for high-sensitivity data, which degrades model performance and increases privacy risks. This paper proposes a new privacy notion, sensitivity-aware differential privacy, to better balance model performance and privacy protection. Our idea is that the sensitivity of each data sample can be objectively measured using real-world attacks. To implement this new notion, we develop the corresponding defense mechanism that adjusts privacy protection levels based on the variation in the privacy leakage risks of gradient inversion attacks. Furthermore, the method extends naturally to multi-attack scenarios. Extensive experiments on real-world medical imaging datasets demonstrate that, under equivalent privacy risk, our method achieves an average performance improvement of 13.5% over state-of-the-art methods.https://www.mdpi.com/1424-8220/25/9/2847smart healthcaredifferential privacygradient inversion attacksfederated learning
spellingShingle Lele Zheng
Yang Cao
Masatoshi Yoshikawa
Yulong Shen
Essam A. Rashed
Kenjiro Taura
Shouhei Hanaoka
Tao Zhang
Sensitivity-Aware Differential Privacy for Federated Medical Imaging
Sensors
smart healthcare
differential privacy
gradient inversion attacks
federated learning
title Sensitivity-Aware Differential Privacy for Federated Medical Imaging
title_full Sensitivity-Aware Differential Privacy for Federated Medical Imaging
title_fullStr Sensitivity-Aware Differential Privacy for Federated Medical Imaging
title_full_unstemmed Sensitivity-Aware Differential Privacy for Federated Medical Imaging
title_short Sensitivity-Aware Differential Privacy for Federated Medical Imaging
title_sort sensitivity aware differential privacy for federated medical imaging
topic smart healthcare
differential privacy
gradient inversion attacks
federated learning
url https://www.mdpi.com/1424-8220/25/9/2847
work_keys_str_mv AT lelezheng sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT yangcao sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT masatoshiyoshikawa sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT yulongshen sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT essamarashed sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT kenjirotaura sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT shouheihanaoka sensitivityawaredifferentialprivacyforfederatedmedicalimaging
AT taozhang sensitivityawaredifferentialprivacyforfederatedmedicalimaging