Preprint
Machine Learning

Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data

Micah Sheller(Intel (United States)), Brandon Edwards(Intel (United States)), G. Anthony Reina(Intel (United States)), Jason Martin(Intel (United States)), Sarthak Pati(University of Pennsylvania), Aikaterini Kotrotsou(The University of Texas MD Anderson Cancer Center), Mikhail Milchenko(Washington University in St. Louis), Weilin Xu(Intel (United States)), Daniel C. Marcus(Washington University in St. Louis), Rivka R. Colen(The University of Texas MD Anderson Cancer Center), Spyridon Bakas(Laboratoire d’Imagerie Biomédicale)
July 28, 2020Scientific Reports1,500 citations
PDFDOI

1.5k

Citations

38

Influential Citations

Scientific Reports

Venue

2020

Year

Abstract

Several studies underscore the potential of deep learning in identifying complex patterns, leading to diagnostic and prognostic biomarkers. Identifying sufficiently large and diverse datasets, required for training, is a significant challenge in medicine and can rarely be found in individual institutions. Multi-institutional collaborations based on centrally-shared patient data face privacy and ownership challenges. Federated learning is a novel paradigm for data-private multi-institutional collaborations, where model-learning leverages all available data without sharing data between institutions, by distributing the model-training to the data-owners and aggregating their results. We show that federated learning among 10 institutions results in models reaching 99% of the model quality achieved with centralized data, and evaluate generalizability on data from institutions outside the federation. We further investigate the effects of data distribution across collaborating institutions on model quality and learning patterns, indicating that increased access to data through data private multi-institutional collaborations can benefit model quality more than the errors introduced by the collaborative method. Finally, we compare with other collaborative-learning approaches demonstrating the superiority of federated learning, and discuss practical implementation considerations. Clinical adoption of federated learning is expected to lead to models trained on datasets of unprecedented size, hence have a catalytic impact towards precision/personalized medicine.

Analysis

Why This Paper Matters

This paper addresses a critical bottleneck in medical AI: the need for large, diverse datasets while respecting patient privacy and institutional data governance. By demonstrating that federated learning can achieve 99% of centralized model quality across 10 institutions, it provides a viable path for multi-institutional collaborations that were previously hindered by privacy concerns. The work is particularly significant because it validates the approach on real-world medical data and shows that the benefits of increased data access outweigh the errors introduced by the federated learning method.

Technical Contributions

  • Federated learning framework: The authors implement a data-private collaborative learning paradigm where model training occurs at each institution and only model updates (gradients or weights) are shared with a central server.
  • Quality preservation: They show that federated models retain 99% of the performance of models trained on centrally pooled data, a critical finding for clinical adoption.
  • Generalizability testing: The study evaluates models on data from institutions not part of the federation, demonstrating robustness to unseen data distributions.
  • Data distribution analysis: They systematically investigate how non-IID data distributions across collaborators affect model quality and learning dynamics.
  • Comparative analysis: Federated learning is shown to outperform other collaborative approaches (e.g., ensemble learning, transfer learning) in terms of model quality and privacy preservation.

Results

  • Federated learning across 10 institutions achieves 99% of the model quality of centralized training.
  • Models generalize well to data from institutions outside the federation.
  • Increased data access through collaboration benefits model quality more than the errors introduced by the federated method.
  • Federated learning outperforms alternative collaborative-learning approaches.

Significance

This paper has catalyzed the adoption of federated learning in medical imaging and beyond. By providing empirical evidence that privacy-preserving collaboration can match centralized performance, it has opened the door for large-scale multi-institutional studies in precision medicine. The work also highlights practical considerations for deployment, such as communication efficiency and handling heterogeneous data distributions, which are essential for real-world clinical systems.

federated-learningdeep-learningmedical-imagingprivacymulti-institutionalcollaborative-learningprecision-medicine

Related Papers