From Challenges and Pitfalls to Recommendations and Opportunities: Implementing Federated Learning in Healthcare

TL;DR

This review analyzes recent federated learning applications in healthcare (up to May 2024), revealing widespread methodological flaws that impede clinical utility. It synthesizes findings across data, topology, models, optimization, privacy, fairness, and evaluation, highlighting overreliance on simulation, insufficient data standardization, and limited open science practices. The authors offer concrete recommendations, including standardized data harmonization, more realistic open-domain evaluations, personalized initialization, robust privacy-security strategies, and comprehensive benchmarking. Collectively, the work underscores the need for rigorous methodological standards and community-driven, open frameworks to realize FL's potential in clinical care and research.

Abstract

Federated learning holds great potential for enabling large-scale healthcare research and collaboration across multiple centres while ensuring data privacy and security are not compromised. Although numerous recent studies suggest or utilize federated learning based methods in healthcare, it remains unclear which ones have potential clinical utility. This review paper considers and analyzes the most recent studies up to May 2024 that describe federated learning based methods in healthcare. After a thorough review, we find that the vast majority are not appropriate for clinical use due to their methodological flaws and/or underlying biases which include but are not limited to privacy concerns, generalization issues, and communication costs. As a result, the effectiveness of federated learning in healthcare is significantly compromised. To overcome these challenges, we provide recommendations and promising opportunities that might be implemented to resolve these problems and improve the quality of model development in federated learning with healthcare.

Figures (8)

• Figure 1: Difference between centralized FL paradigm and decentralized FL paradigm. Centralized FL relies on a central server to manage the training. While decentralized FL eliminates the need for a central server. Instead, clients can directly communicate with connected ones.
• Figure 2: Visual representation of three categories of FL, illustrating their distribution across feature and sample spaces.
• Figure 3: PRISMA flow diagram for our review, highlighting the inclusion and exclusion of studies at each stage.
• Figure 4: Taxonomy of challenges and pitfalls (red blocks) as well as recommended solutions and opportunities (green blocks).
• Figure 5: Simulation scenario VS. real-world distributed scenario.