Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Inclusive, Differentially Private Federated Learning for Clinical Data

Santhosh Parampottupadam, Melih Coşğun, Sarthak Pati, Maximilian Zenk, Saikat Roy, Dimitrios Bounias, Benjamin Hamm, Sinem Sav, Ralf Floca, Klaus Maier-Hein

TL;DR

This paper tackles privacy-preserving federated learning for clinical data under regulatory constraints by introducing a compliance-aware differential privacy framework that tunes per-client noise using a computed compliance score $S_c$, with a noise multiplier $N_m=(1.0-S_c)+\text{Min Noise Multiplier}$ and $\text{Min Noise Multiplier}=1e-10$; it also provides a web-based compliance scoring tool aligned with healthcare standards. The authors implement an adaptive server-side DP approach, integrate 12 compliance factors, and deploy a DP-enabled FL system using Opacus over a 50-round protocol with ResNet-18, demonstrating up to 1%–15% accuracy gains when mixing under-resourced clinics with highly regulated institutions. Experimental results on PneumoniaMNIST and BreastMNIST show that different aggregators (e.g., FedYogi, FedAdam, FedAvg) perform best under various compliance distributions, and a data-quality scenario confirms robustness of the approach, with dp_FedAvg achieving 72.68% accuracy among DP configurations. Overall, the work offers a practical path toward inclusive, privacy-preserving FL in real-world clinical workflows by balancing privacy, compliance, and utility without requiring specialized hardware.

Abstract

Federated Learning (FL) offers a promising approach for training clinical AI models without centralizing sensitive patient data. However, its real-world adoption is hindered by challenges related to privacy, resource constraints, and compliance. Existing Differential Privacy (DP) approaches often apply uniform noise, which disproportionately degrades model performance, even among well-compliant institutions. In this work, we propose a novel compliance-aware FL framework that enhances DP by adaptively adjusting noise based on quantifiable client compliance scores. Additionally, we introduce a compliance scoring tool based on key healthcare and security standards to promote secure, inclusive, and equitable participation across diverse clinical settings. Extensive experiments on public datasets demonstrate that integrating under-resourced, less compliant clinics with highly regulated institutions yields accuracy improvements of up to 15% over traditional FL. This work advances FL by balancing privacy, compliance, and performance, making it a viable solution for real-world clinical workflows in global healthcare.

Inclusive, Differentially Private Federated Learning for Clinical Data

TL;DR

This paper tackles privacy-preserving federated learning for clinical data under regulatory constraints by introducing a compliance-aware differential privacy framework that tunes per-client noise using a computed compliance score , with a noise multiplier and ; it also provides a web-based compliance scoring tool aligned with healthcare standards. The authors implement an adaptive server-side DP approach, integrate 12 compliance factors, and deploy a DP-enabled FL system using Opacus over a 50-round protocol with ResNet-18, demonstrating up to 1%–15% accuracy gains when mixing under-resourced clinics with highly regulated institutions. Experimental results on PneumoniaMNIST and BreastMNIST show that different aggregators (e.g., FedYogi, FedAdam, FedAvg) perform best under various compliance distributions, and a data-quality scenario confirms robustness of the approach, with dp_FedAvg achieving 72.68% accuracy among DP configurations. Overall, the work offers a practical path toward inclusive, privacy-preserving FL in real-world clinical workflows by balancing privacy, compliance, and utility without requiring specialized hardware.

Abstract

Federated Learning (FL) offers a promising approach for training clinical AI models without centralizing sensitive patient data. However, its real-world adoption is hindered by challenges related to privacy, resource constraints, and compliance. Existing Differential Privacy (DP) approaches often apply uniform noise, which disproportionately degrades model performance, even among well-compliant institutions. In this work, we propose a novel compliance-aware FL framework that enhances DP by adaptively adjusting noise based on quantifiable client compliance scores. Additionally, we introduce a compliance scoring tool based on key healthcare and security standards to promote secure, inclusive, and equitable participation across diverse clinical settings. Extensive experiments on public datasets demonstrate that integrating under-resourced, less compliant clinics with highly regulated institutions yields accuracy improvements of up to 15% over traditional FL. This work advances FL by balancing privacy, compliance, and performance, making it a viable solution for real-world clinical workflows in global healthcare.

Paper Structure

This paper contains 6 sections, 1 equation, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methods
  3. Results
  4. Discussion
  5. Limitations and Future Works
  6. Acknowledgments

Figures (1)

  • Figure 1: (a) Existing FL with client-side DP uses uniform noise, requiring DP-compliant hardware, limiting less compliant, resource-constrained clinics. (b) Server-side DP adds uniform noise post-aggregation, reducing privacy-utility efficiency and further excluding less compliant clinics. (c) Our compliance-aware adaptive DP applies per-client noise before aggregation, enabling participation from low-resource, less compliant clinics while optimizing privacy and performance.