Statistical and engineering approaches to federated learning: Comprehensive benchmarking for healthcare applications

Statistical and Engineering Approaches to Federated Learning: Comprehensive Benchmarking for Healthcare Applications

A groundbreaking study conducted by Duke-NUS Medical School evaluates federated learning (FL) methods to guide healthcare researchers in choosing privacy-preserving algorithms tailored to their clinical goals. This comprehensive benchmark compared statistical and engineering FL frameworks, offering actionable insights to balance predictive accuracy and interpretability in medical research.

Federated learning (FL) has emerged as a powerful tool in healthcare, enabling collaboration across institutions without compromising patient privacy. With stringent data privacy regulations like GDPR, FL frameworks have gained traction. However, their varying methodologies—statistical and engineering-based—pose challenges in selecting the right approach for specific research needs.

In the first comprehensive comparison of its kind, Duke-NUS researchers evaluated seven FL frameworks—three statistical and four engineering-based—using both simulated data and real-world emergency department datasets. Statistical FL methods were found to be more reliable for interpreting relationships between factors and clinical outcomes, making them ideal for non-predictive tasks. Engineering-based FL algorithms, on the other hand, demonstrated superior predictive performance, excelling in outcome prediction tasks.

"Our study bridges the gap between federated learning approaches and their practical application in healthcare," said lead author Siqi Li, a PhD candidate at Duke-NUS Medical School. "By providing clear guidelines, we empower researchers to choose FL frameworks that align with their specific priorities, whether it’s interpretability or predictive power."

Senior Research Assistant Di Miao added, "These findings highlight the unique strengths of each method, paving the way for more effective and ethical collaborations in clinical research."

This research not only underscores the potential of federated learning in advancing privacy-preserving AI but also lays the groundwork for future innovations in clinical decision-making and patient care.

The team aims to further refine FL algorithms for broader applications, fostering privacy-preserving collaborations across institutions and enhancing healthcare outcomes through innovative AI solutions.

END