Challenges in Implementing the Local Node Infrastructure for a National Federated Machine Learning Network in...

Area of Science:

• Medical research
• Machine learning
• Radiology

Background:

• Data-driven machine learning requires large, diverse datasets for medical research and diagnostics.
• Generating large datasets is resource-intensive and time-consuming.
• Federated learning (FL) offers a decentralized approach to train models on local data, enhancing data privacy and reducing effort.

Purpose of the Study:

• To propose basic infrastructure requirements for implementing federated learning in clinical and scientific environments.
• To detail data governance, data science workflows, and local node setup for FL networks.
• To report on the advantages and challenges of implementing FL infrastructure using the German Radiological Cooperative Network as a use case.

Main Methods:

• Development of a standard framework for federated learning networks, comprising fundamental components.
• Proposal of infrastructure requirements for data governance, data science workflows, and local node setup.
• Analysis of the deployment process in various settings, with a focus on integration into existing clinical IT infrastructure.

Main Results:

• Federated learning allows for local data maintenance, avoiding data safety issues by not sharing patient data.
• The proposed infrastructure can be built upon base components to meet FL network needs, considering both local and global requirements.
• Integrating local FL nodes into existing clinical IT infrastructure offers benefits in maintenance and deployment effort.

Conclusions:

• The proposed infrastructure requirements provide a guideline for future federated learning applications in clinical and scientific settings.
• Integrating FL nodes into existing clinical IT infrastructure is recommended for efficient maintenance and deployment.
• Federated learning presents a viable solution for data-driven medical research while preserving patient data privacy.