Federated Learning in Medical Imaging: Framework, Use case, and Research
(Ziyue Xu, PhD, Nvidia; Jayashree Kalpathy-Cramer, PhD, MGH)

The accuracy and robustness of AI algorithms rely heavily on the quantity, quality, and diversity of the training dataset. For medical imaging applications, the challenge of constructing such a dataset is particularly significant, mainly due to the privacy concerns in data sharing across multiple institutions. Federated learning (FL) has emerged as a potential solution due to its capability in training models without sharing data. To enable effective FL in real applications, a robust communication framework is crucial. In this talk, we will cover the open-source NVIDIA FLARE infrastructure for orchestrating an FL study. Together with Project MONAI, a medical imaging use case will be discussed, recent research towards better performing FL pipelines, and a current MICCAI challenge on Breast Density FL, will be introduced.  TBD