This is the associated repository for our manuscript titled "Methods and Impact for using Federated Learning to collaborate on clinical research". The manuscript and repository should serve as a practical guide to conceptualizing, building, and executing a federated learning project to train a decentralized model across multiple medical centers. Please fork this repository and edit it as needed for your own projects using nvflare, as well as copy as examples the documentation we provided our various cites and the on-prem and cloud hardware that was utilized as part of this project. The ich-fl folder contains the scripts necessary to train a classifier for ICH detection through a federated paradigm, while the src folder is merely for comparison to help illustrate how to convert "traditional", centralized machine learning code to be compatible with nvflare.
As an end-user, we highly recommend inspecting and familiarizing yourself with the base repository hosted by nVidia: https://github.com/NVIDIA/NVFlare .
Please see the associated document "Federated Learning in Healthcare: A roadmap and resource for clinicians" for more details on how to set up your own federated network.