Partners HealthCare has launched a “federated learning” initiative, using the NVIDIA Clara application framework that will help clinicians and researchers to develop better AI models that are used at the point of care, particularly in medical imaging. Federated learning makes it possible for AI algorithms to learn from a vast range of data located at different sites and enables organizations to collaborate on the development of models without needing to share sensitive clinical data with each other.

The initiative will be spearheaded by the Massachusetts General Hospital (MGH) and Brigham and Women’s Hospital’s (BWH) Center for Clinical Data Science (CCDS).

Keith Dreyer, DO, PhD, FACR, FSIIM, Chief Data Science Officer, Partners HealthCare, stated, “We believe that making federated learning widespread will unlock value to our patients while maintaining their data privacy. The NVIDIA and CCDS collaboration provides us with the cutting-edge technology needed to drive AI innovation to the next step – large-scale product development efforts that ensure diversity of data without compromising safety and privacy.”

Using the NVIDIA Clara AI Toolkit, the CCDS is improving a pre-trained AI model, which was developed at Partners HealthCare by accumulating individual contributions from models trained on diverse datasets to generate a global model. Mona Flores, MD, MBA, FACS, Global Lead for Hospitals and Clinical Partnerships, NVIDIA, said, “Broadening our collaboration with Partners HealthCare in this area is an exciting step forward for AI in healthcare. This federated learning initiative will combine technical and clinical expertise with privacy preservation to deliver more robust AI algorithms for healthcare.”

Privacy, security and bias have been fundamental challenges of developing and deploying AI across institutions. This initiative will leverage the extensive data assets and clinical expertise of the Partners HealthCare system, as well as the NVIDIA DGX computational infrastructure and new privacy-preserving federated learning feature of the NVIDIA Clara application framework to address these challenges.

“Federated learning enables collaborative, decentralized training of AI models without the need to share patient data,” said Ittai Dayan, MD, Executive Director, CCDS. “Aggregating knowledge from various institutions creates a more robust, accurate and generalizable AI product. This approach has the potential to improve ‘model resiliency’ and reduce bias.”