Nvidia Clara Federated Learning allows participating hospitals to train a global model on their own data.

The results – in the form of partial model weights – are fed back to the federated learning server via a secure link.

This approach means deep learning training can be carried out across multiple hospitals without having to share the clinical data.

Clara Federated Learning makes use of the Nvidia Clara AI-Assisted Annotation SDK which is integrated into medical viewers including 3D slicer, MITK, Fovia and Philips Intellispace Discovery to make it easier for radiologists to label their data.

Organisations piloting or adopting Clara Federated Learning include the American College of Radiology (which is building a national platform for medical imaging) and the UK's National Health Service (Nvidia, King’s College London and Owkin are creating a platform that is initially being used by four leading teaching hospitals in London and will spread to at least 12 hospitals around the country during 2020).

Clara Federated Learning runs on the Nvidia EGX edge computing platform.

In related news, Nvidia has announced Clara AGX, an embedded AI developer kit for handling image and video processing at high data rates.

It runs on Nvidia Xavier SoCs (systems on a chip), which can be embedded in medical instruments or delivered as small add-on systems.

An early example is Hyperfine's portable MRI system.

Nvidia's Clara platform for applying machine learning and graphics to medical instruments was announced just over a year ago.