GTC 2020 S21536
Presenters: Yuhong Wen,NVIDIA; Nicola Rieke, NVIDIA
Abstract
While deep neural networks have shown promising results in various medical applications, they highly depend on the amount and diversity of the training data. In the context of medical imaging, this poses a major challenge because patient data needs to be protected and cannot easily be shared. The training data that is required to train a reliable and robust algorithm may not be available in a single institution due to the low incidence rate of some pathologies and limited numbers of patients. At the same time, it is often not feasible to collect and share patient data in a centralized data lake due to patient privacy concerns and regulations. Federated learning — as a collaborative machine learning paradigm — combined with an advanced privacy-preserving mechanism has the potential of solving this issue: models can be trained across several institutions without explicitly sharing patient data. When implementing and deploying a federated learning system into the real-world medical imaging ecosystem, participants can authenticate and communicate securely, and exchange model weights efficiently, enabling model training to be successful. In this talk, we present an introduction to the core concepts of federated learning and discuss the benefits as well as the unique considerations and challenges of implementing a federated-learning system in the context of health care.
Watch this session
Join in the conversation below.