Real-World Federated Learning in Radiology: Hurdles to overcome and Benefits to gain
Authors:
Markus R. Bujotzek,
Ünal Akünal,
Stefan Denner,
Peter Neher,
Maximilian Zenk,
Eric Frodl,
Astha Jaiswal,
Moon Kim,
Nicolai R. Krekiehn,
Manuel Nickel,
Richard Ruppel,
Marcus Both,
Felix Döllinger,
Marcel Opitz,
Thorsten Persigehl,
Jens Kleesiek,
Tobias Penzkofer,
Klaus Maier-Hein,
Rickmer Braren,
Andreas Bucher
Abstract:
Objective: Federated Learning (FL) enables collaborative model training while kee** data locally. Currently, most FL studies in radiology are conducted in simulated environments due to numerous hurdles impeding its translation into practice. The few existing real-world FL initiatives rarely communicate specific measures taken to overcome these hurdles, leaving behind a significant knowledge gap.…
▽ More
Objective: Federated Learning (FL) enables collaborative model training while kee** data locally. Currently, most FL studies in radiology are conducted in simulated environments due to numerous hurdles impeding its translation into practice. The few existing real-world FL initiatives rarely communicate specific measures taken to overcome these hurdles, leaving behind a significant knowledge gap. Minding efforts to implement real-world FL, there is a notable lack of comprehensive assessment comparing FL to less complex alternatives. Materials & Methods: We extensively reviewed FL literature, categorizing insights along with our findings according to their nature and phase while establishing a FL initiative, summarized to a comprehensive guide. We developed our own FL infrastructure within the German Radiological Cooperative Network (RACOON) and demonstrated its functionality by training FL models on lung pathology segmentation tasks across six university hospitals. We extensively evaluated FL against less complex alternatives in three distinct evaluation scenarios. Results: The proposed guide outlines essential steps, identified hurdles, and proposed solutions for establishing successful FL initiatives conducting real-world experiments. Our experimental results show that FL outperforms less complex alternatives in all evaluation scenarios, justifying the effort required to translate FL into real-world applications. Discussion & Conclusion: Our proposed guide aims to aid future FL researchers in circumventing pitfalls and accelerating translation of FL into radiological applications. Our results underscore the value of efforts needed to translate FL into real-world applications by demonstrating advantageous performance over alternatives, and emphasize the importance of strategic organization, robust management of distributed data and infrastructure in real-world settings.
△ Less
Submitted 15 May, 2024;
originally announced May 2024.
