arXiv:2609.20650v1 Announce Type: cross Abstract: Federated learning enables collaborative training without sharing patient-level data, but most studies remain simulations. Based on five requirements derived from the literature, we analyzed 14 FL frameworks and found that none fully satisfied these requirements. We present FL-Net, a novel federated clinical research framework to fulfill all requi
FL-Net is a novel federated clinical research framework introduced in arXiv:2609.20650 to address the gap between simulated and real-world federated learning by fulfilling five key requirements derived from existing literature. Unlike previous studies that remain simulations, FL-Net integrates modular data harmonization, data discovery, disclosure control, and containerized federated workflow execution into a persistent network.
The framework enables the re-use of harmonized data and workflows across studies, demonstrating end-to-end capabilities through cross-study patient discovery and reproducible, audited federated workflows with up to 50 concurrent clients. It is being developed within the dAIbetes and Microb-AI-ome EU projects to cover over 800,000 patients across 10 hospitals in 9 countries, providing a practical foundation for interoperable, privacy-preserving multicenter clinical research.
Multi-center Medical Data Mining with FL-Net is an arXiv preprint that addresses the practical gap between federated learning research and its deployment in real multi-institutional clinical settings. The authors argue that federated learning can enable collaborative model training without centralizing patient-level data, but much of the existing literature remains confined to simulated benchmarks. To make the field more actionable, they distill five requirements for a clinically usable federated learning framework from prior work and use them to evaluate 14 existing FL systems. Their comparative analysis finds that none of the surveyed frameworks fully satisfies the complete set of requirements, suggesting that current tooling is fragmented and often oriented toward research prototypes rather than operational clinical research workflows.
The paper’s central contribution is FL-Net, presented as a novel federated clinical research framework intended to serve as a “one-stop shop” for multi-center medical data mining. Rather than focusing narrowly on a single aggregation algorithm or privacy mechanism, FL-Net is positioned as an integrated platform for federated clinical studies, addressing end-to-end needs such as coordination across data sources, model development, and practical research workflow support. The key insight is that the barrier to meaningful medical federated learning is not only algorithmic but also architectural and operational: a usable system must combine privacy-preserving collaboration with the reproducibility, usability, and clinical compatibility required by real research teams.
This matters because medical data are inherently distributed across hospitals, clinics, and health systems, and privacy constraints make centralized data pooling difficult. A well-designed federated learning framework can enable multi-site model training, benchmarking, and deployment while keeping raw patient data local, which is especially important in regulated healthcare environments. By benchmarking existing frameworks against explicit requirements and proposing FL-Net as a candidate that fulfills them, the paper provides both a diagnostic of the current federated learning ecosystem and a practical path toward more scalable, reproducible, and clinically grounded medical data mining.