It’s a contradiction that’s long slowed the forward march of artificial intelligence in medicine: Machine learning models need to be trained on lots of diverse data from hospitals around the world — but those hospitals are often reluctant to ship out their data due to privacy concerns, legal issues, and a cautious culture.
One promising way to get around that problem is a technique known as federated learning, which allows models to be trained without having to share data to a central server or in the cloud. Now, the approach is being put to the test in an ambitious project to build an AI system from thousands of brain tumor scans from several dozen hospitals and research institutions around the globe.
About the Author Reprints
