Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MedHE: Communication-Efficient Privacy-Preserving Federated Learning with Adaptive Gradient Sparsification for Healthcare

Farjana Yesmin

TL;DR

MedHE addresses the dual challenge of privacy and communication efficiency in healthcare federated learning by co-designing adaptive gradient sparsification with CKKS homomorphic encryption. The framework introduces an error-feedback-based sparsification scheme and optimized ciphertext packing to achieve a 97.5% reduction in per-round communication while preserving model utility, demonstrated by accuracy of $89.5\% \pm 0.8\%$ across five trials and a minimal $p$-value difference ($p=0.32$) versus standard FL. Formal security analysis provides IND-CPA guarantees for CKKS under RLWE, differential privacy with $\epsilon$ bounded (e.g., $\epsilon\leq 1$ in experiments), and convergence assurances under adaptive sparsity with error compensation. Deployment analyses confirm HIPAA-aligned privacy, scalability to 100+ institutions, and practical cost reductions, making MedHE a deployment-ready approach for secure, large-scale, privacy-preserving healthcare collaboration.

Abstract

Healthcare federated learning requires strong privacy guarantees while maintaining computational efficiency across resource-constrained medical institutions. This paper presents MedHE, a novel framework combining adaptive gradient sparsification with CKKS homomorphic encryption to enable privacy-preserving collaborative learning on sensitive medical data. Our approach introduces a dynamic threshold mechanism with error compensation for top-k gradient selection, achieving 97.5 percent communication reduction while preserving model utility. We provide formal security analysis under Ring Learning with Errors assumptions and demonstrate differential privacy guarantees with epsilon less than or equal to 1.0. Statistical testing across 5 independent trials shows MedHE achieves 89.5 percent plus or minus 0.8 percent accuracy, maintaining comparable performance to standard federated learning (p=0.32) while reducing communication from 1277 MB to 32 MB per training round. Comprehensive evaluation demonstrates practical feasibility for real-world medical deployments with HIPAA compliance and scalability to 100 plus institutions.

MedHE: Communication-Efficient Privacy-Preserving Federated Learning with Adaptive Gradient Sparsification for Healthcare

TL;DR

MedHE addresses the dual challenge of privacy and communication efficiency in healthcare federated learning by co-designing adaptive gradient sparsification with CKKS homomorphic encryption. The framework introduces an error-feedback-based sparsification scheme and optimized ciphertext packing to achieve a 97.5% reduction in per-round communication while preserving model utility, demonstrated by accuracy of across five trials and a minimal -value difference () versus standard FL. Formal security analysis provides IND-CPA guarantees for CKKS under RLWE, differential privacy with bounded (e.g., in experiments), and convergence assurances under adaptive sparsity with error compensation. Deployment analyses confirm HIPAA-aligned privacy, scalability to 100+ institutions, and practical cost reductions, making MedHE a deployment-ready approach for secure, large-scale, privacy-preserving healthcare collaboration.

Abstract

Healthcare federated learning requires strong privacy guarantees while maintaining computational efficiency across resource-constrained medical institutions. This paper presents MedHE, a novel framework combining adaptive gradient sparsification with CKKS homomorphic encryption to enable privacy-preserving collaborative learning on sensitive medical data. Our approach introduces a dynamic threshold mechanism with error compensation for top-k gradient selection, achieving 97.5 percent communication reduction while preserving model utility. We provide formal security analysis under Ring Learning with Errors assumptions and demonstrate differential privacy guarantees with epsilon less than or equal to 1.0. Statistical testing across 5 independent trials shows MedHE achieves 89.5 percent plus or minus 0.8 percent accuracy, maintaining comparable performance to standard federated learning (p=0.32) while reducing communication from 1277 MB to 32 MB per training round. Comprehensive evaluation demonstrates practical feasibility for real-world medical deployments with HIPAA compliance and scalability to 100 plus institutions.

Paper Structure

This paper contains 36 sections, 4 theorems, 15 equations, 4 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Why This is Not a Simple Combination
  3. Contributions
  4. Related Work
  5. Privacy-Preserving Healthcare FL
  6. Homomorphic Encryption for FL
  7. Gradient Compression for FL
  8. Threat Model and Security Requirements
  9. Adversary Model
  10. Attack Vectors
  11. Security Requirements
  12. Notation and System Model
  13. MedHE Framework Design
  14. Adaptive Gradient Sparsification with Error Compensation
  15. Optimized CKKS Integration
  16. ...and 21 more sections

Key Result

Theorem 1

Under the Ring Learning with Errors (RLWE) assumption with parameters $(N=8192, q=240\text{ bits}, \chi)$ where $\chi$ is a discrete Gaussian error distribution, the CKKS encryption scheme provides IND-CPA semantic security with 128-bit security level against polynomial-time adversaries.

Figures (4)

  • Figure 1: Communication overhead comparison showing MedHE achieves 97.5% reduction compared to standard FL while maintaining superior accuracy.
  • Figure 2: Sparsity sensitivity: 90% sparsity provides best accuracy-communication trade-off.
  • Figure 3: Convergence comparison: MedHE maintains similar convergence rate to standard FL with error feedback mechanism.
  • Figure 4: Scalability analysis: MedHE scales linearly to 100+ clients with consistent communication savings.

Theorems & Definitions (7)

  • Theorem 1: CKKS Semantic Security
  • proof
  • Theorem 2: Differential Privacy with Advanced Composition
  • proof
  • Theorem 3: Convergence with Error Feedback
  • proof
  • Corollary 4: Privacy-Utility Trade-off