Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Jul 14, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

Security Analysis of a Federated Learning Framework for Medical Image-to-Image Translation.

Ciro Benito Raggio1, Lina Bucher2, Oliver Blanck3

  • 1Institute of Biomedical Engineering, Karlsruhe Institute of Technology, Fritz-Haber-Weg 1, 76131, Karlsruhe, Baden-Württemberg, Germany. ciro.raggio@kit.edu.

Journal of Medical Systems
|July 4, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

A privacy-preserving federated learning framework for generalizable CBCT to synthetic CT translation in head and neck.

Frontiers in digital health·2026
Same author

CT-based radiomics as a non-invasive virtual biopsy for high PD-L1 expression prediction in non-small cell lung cancer.

Cancer imaging : the official publication of the International Cancer Imaging Society·2026
Same author

Electrical storm treated with radiation therapy in a patient with ischaemic cardiomyopathy and left ventricular thrombus: a case report.

European heart journal. Case reports·2026
Same author

Standardized processing of explanted hearts after stereotactic arrhythmia radiotherapy (Bio-STAR): a STOPSTORM framework and first human findings.

European heart journal open·2026
Same author

Real-world outcomes of [<sup>177</sup>Lu]Lu-PSMA-I&T in [<sup>18</sup>F]FDG-positive metastatic castration-resistant prostate cancer: factors related to response and survival.

European journal of nuclear medicine and molecular imaging·2026
Same author

Correction: Biochemical failure after radical prostatectomy with PSA ≤ 1 ng/mL: prediction of PSMA-positive metastatic disease.

Annals of nuclear medicine·2026

Federated learning for medical image translation is not inherently secure. This study found critical privacy risks from membership inference attacks and data poisoning, highlighting the need for robust security safeguards.

Area of Science:

  • Artificial Intelligence
  • Medical Imaging
  • Cybersecurity

Background:

  • Federated Learning (FL) enables collaborative deep learning model training without data sharing, applied to medical image-to-image (I2I) translation.
  • Existing FL frameworks for I2I tasks often lack explicit privacy validation, leaving them vulnerable to adversarial threats.

Purpose of the Study:

  • To evaluate the security vulnerabilities of a federated MRI-to-synthetic CT (sCT) translation framework (FedSynthCT-Brain).
  • To assess the effectiveness of defense mechanisms like Secure Aggregation (SecAgg) and Byzantine-robust median aggregation (FedMedian) against identified threats.

Main Methods:

  • Assessed Deep Leakage from Gradients (DLG), Federated Membership Inference Attack (FedMIA), and data poisoning.
  • Evaluated SecAgg and FedMedian defenses against these attacks.
Keywords:
Federated learningImage-to-image translationSecurity attacksSynthetic computed tomography

Related Experiment Videos

Last Updated: Jul 14, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

  • Quantified synthesis quality using SSIM and PSNR, and attack efficacy using AUC.

    • Main Results:

    • DLG attacks yielded only coarse anatomical structures, with limited clinical detail.
    • FedMIA demonstrated significant privacy breaches (AUC 0.92-0.99), which SecAgg effectively mitigated (AUC 0.23-0.56) without compromising synthesis quality.
    • Data poisoning with FedAvg rendered the model inoperative; FedMedian largely restored performance, though stealthy degradation was observed at low noise levels.

    Conclusions:

    • Federated I2I translation frameworks require explicit, multi-layered security evaluations, as they are not inherently secure.
    • Integrating cryptographic, algorithmic, and infrastructural safeguards is crucial for the secure deployment of FL in clinical settings.