Skip to content

Secure federated learning applied to medical imaging with fully homomorphic encryption

Authors: Xavier Lessage; Leandro Collier; Charles-Henry Bertrand Van Ouytsel; Axel Legay; Saïd Mahmoudi; Philippe Massonet
Year: 2024
Venue: 2024 IEEE 3rd International Conference on AI in Cybersecurity (ICAIC)

type: application collection: ppml status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

This study combines federated learning with fully homomorphic encryption to protect the exchange of model updates between participants and a central server, applied to a confidential dataset of Belgian mammograms for mass and microcalcification detection. The authors assess the feasibility of FHE in the FL setting with a focus on memory constraints, finding that FHE keeps ROC performance comparable to the unencrypted approach while imposing notable memory-usage limitations that call for further optimization.

Abstract

This study explores the convergence of Federated Learning (FL) and Fully Homomorphic Encryption (FHE) through an innovative approach applied to a confidential dataset composed of mammograms from Belgian medical records. Our goal is to clarify the feasibility and challenges associated with integrating FHE into the context of Federated Learning, with a particular focus on evaluating the memory constraints inherent in FHE when using sensitive medical data. The results highlight notable limitations in terms of memory usage, underscoring the need for ongoing research to optimize FHE in real-world applications. Despite these challenges, our research demonstrates that FHE maintains comparable performance in terms of Receiver Operating Characteristic (ROC) curves, affirming the robustness of our approach in secure machine learning applications, especially in sectors where data confidentiality, such as medical data management, is imperative. The conclusions not only shed light on the technical limitations of FHE but also emphasize its potential for practical applications. By combining Federated Learning with FHE, our model preserves data confidentiality while ensuring the security of exchanges between participants and the central server
Source: pdf

Extracted details

Field Value Source Confidence
contribution Integration of FHE into a federated learning pipeline (PyTorch, Flower, SEAL) for medical imaging llm 0.60
contribution Feasibility study on a confidential Belgian mammogram dataset llm 0.60
contribution Evaluation of FHE memory constraints and ROC performance compared to plain FL llm 0.60
schemes FHE llm 0.60
libraries PyTorch llm 0.60
libraries Flower llm 0.60
libraries SEAL llm 0.60
models CNN llm 0.60
datasets Belgian mammogram dataset (confidential medical records) llm 0.60
tasks federated learning llm 0.60
tasks breast cancer detection (masses and microcalcifications) in mammograms llm 0.60
security level 128-bit (given as example security level) llm 0.60
threat model FL participants exchanging encrypted model updates with a central aggregation server llm 0.60
interaction non-interactive FHE llm 0.60
code url https://github.com/microsoft/SEAL/blob/master/native/src/seal/util/hestdparms.h (URL found in the full text) regex 0.50
limitations High memory usage inherent to FHE llm 0.60
limitations Need for further FHE optimization for real-world applications llm 0.60
key result ROC performance: comparable to unencrypted FL (FL + FHE on mammogram dataset) llm 0.60
key result memory usage: notable limitations reported (FHE on sensitive medical data) llm 0.60

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: CKKS 13 [p.3], Differential privacy 7 [p.2], BFV 5 [p.3], TFHE 2 [p.6], MPC 2 [p.2], Partially homomorphic encryption 2 [p.3]
  • library: Microsoft SEAL 12 [p.4], TenSEAL 10 [p.4], PyTorch 5 [p.5]
  • model: MobileNet 5 [p.5], AlexNet 1 [p.12], CNN 1 [p.7], SVM 1 [p.4], k-NN 1 [p.4]
  • dataset: Breast cancer (Wisconsin) 17 [p.1], ImageNet 1 [p.7]
  • hardware: GPU 2 [p.5], CUDA 1 [p.5]
  • technique: Federated learning 39 [p.1], Convolution 6 [p.6], Secure aggregation 3 [p.2], Bootstrapping 2 [p.10], Softmax 2 [p.7], Model inversion 2 [p.2]
  • assumption: Lattice-based 2 [p.11]
  • task: Medical / healthcare 24 [p.1], Image classification 3 [p.12], Language modeling / generation 2 [p.12]
  • topic: Homomorphic encryption 63 [p.1], Open source 3 [p.5]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/federated-learning/lessage-2024_secure-federated-learning-applied-to-medical-imaging-with-fhe.pdf ppml/federated-learning 12 2.8 MB cf7638de39d3 was Lessage et al. - 2024 - Secure federated learning applied to medical imaging with fully homomorphic encryption.pdf

Database key: secure-federated-learning-applied-to-medical-imaging-with-fully · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set secure-federated-learning-applied-to-medical-imaging-with-fully COLUMN VALUE