Nvidia uses federated learning to enable AI in hospitals
Nvidia Corp. wants to make artificial intelligence a staple of the healthcare industry with a new distributed learning technique that’s able to train machine learning models while protecting patient privacy.
AI holds great promise, but for industries such as healthcare where data privacy is of paramount importance, tapping into that potential is big a challenge. The problem is that any data which might be useful to train models is almost always confidential, which means it can’t be shared with technology partners.
Nvidia reckons it can solve this problem with its new Clara Federated Learning technique , which ensures that patient data remains within healthcare providers’ systems at all times.
Clara FL is a reference application for distributed AI training that’s designed to run on Nvidia’s recently announced EGX intelligent edge computing platform . Those systems are capable of performing deep learning training locally at the “network edge”, where the data resides, without moving it. Clara FL is also collaborative, which means multiple systems can work together at different locations to create more accurate, global models, Nvidia Vice President of Healthcare Kimberly Powell said in a blog post .
Clara FL has already found a practical application among radiologists at several top healthcare providers, including the American College of Radiology, the BWH Center for Clinical Data Science, and UCLA Health. First, the radiologists label their data using the NVIDIA Clara AI-Assisted Annotation software development kit, which is integrated with medical viewers such as 3D slicer, MITK, Fovia and Philips Intellispace Discovery.
“Using pre-trained models and transfer learning techniques, NVIDIA AI assists radiologists in labeling, reducing the time for complex 3D studies from hours to minutes,” Powell said.
The models are then trained on Nvidia’s servers located on site, with the results being shared over a secure link back to the federated learning center. Only the results of the training are shared, with the patient data remaining where it is, ensuring it remains secure.
“In the U.K., NVIDIA is partnering with King’s College London and Owkin to create a federated learning platform for the National Health Service,” Powell wrote.
