RSNA 2021 Presentation: Accelerating Healthcare AI with Federated Learning

Artificial intelligence (AI)-based diagnostics and treatment pathways are redefining radiology. Yet the impact often falls short of expectations when new tools are introduced in-clinic and used across today’s increasingly diverse patient populations. This is because many AI solutions are trained using a small dataset that does not accurately reflect the real-world reality.

To overcome this challenge, forward-thinking medical researchers and AI developers are embracing a new approach to training AI models: federated learning (FL). With federated learning, data remains at the site where it was created - so privacy is always protected. Copies of the AI model are sent to each site, and training is performed locally. No moving massive amounts of data and creating redundancies. Aggregate learnings inform an optimized model. This approach to utilizing larger, more diverse datasets enables AI-based solutions to scale globally at an unprecedented pace.

Moderated by Ittai Dayan, cofounder and CEO of Rhino Health, this session presented at the RSNA 2021 conference focuses on how federated learning is changing the way radiology research and practice is conducted today and what that means for the profession moving forward. Watch now to gain a better understanding of how to:

* Converge AI models created on disparate datasets, making it possible to analyze data across hospitals without ever moving data or risking patient privacy

* Support the full lifecycle of healthcare AI solutions - creation, training, validation, measurement, improvement

* Identify clinically relevant datasets necessary to solve some of the most complex health problems facing today’s increasingly diverse patient populations

This split-panel discussion centers on two examples of federated learning in action: 1) Learnings from the EXAM study, which was published in Nature Medicine in September 2021; and 2) Insights into the NIH's Early Detection Research Network's use of federated learning to accelerate detection of pancreatic cancer.

Participants include:
* Elliot Fishman, MD - Johns Hopkins Medicine
* Eugene Koay, MD PhD - MD Anderson Cancer Network
* Fiona Gilbert, MD - University of Cambridge School of Medicine
* Holger Roth, PhD - NVIDIA
* Marius Linguraru, PhD - Children's National Health System
* Michael Rosenthal, MD PhD - Dana-Farber Cancer Institute
* Michal Guindy, MD - Assuta Medical Centers
* Mona Flores, MD - NVIDIA