NVIDIA Clara AI model offers AI services to hospitals while upholding patient data privacy

NVIDIA has announced Clara Federated Learning, its latest edge computing AI platform that delivers AI services to medical institutions without compromising the privacy of patients' data.

NVIDIA Clara Federated Learning, or Clara FL, leverage on "a distributed, collaborative learning technique", which ensures patient data remains on-premise of the healthcare service provider. In terms of hardware, Clara FL runs on NVIDIA EGX intelligent edge computing platform.

In this particular distributing computing environment, the Clara FL application is "distributed as a Helms chart", a sort of package manager for applications deployment on Kubernetes infrastructure.

The NVIDIA EGX platform is managed by a point healthcare provider, who delivers the application containers, as well as the initial AI model to participating clients, in order to start the federated learning project.

[embed]https://www.youtube.com/watch?v=2NQ8otZvNyo&feature=emb_title[/embed]

The participating clients, i.e., hospitals and other healthcare service providers, will start labelling their own patient data by making use of pre-trained models and deep learning techniques like transfer learning. This process is sped up with Clara AI-assisted annotation.

In turn, the NVIDIA EGX servers at the premises make use of the labelled data to train the global data model, and the local training results are shared with the federated learning server. No patient data is exposed as the global model is enhanced through federated averaging, i.e., only training results is shared, not patient data.

The other advantage of federated learning is increased efficiency as little data is sent over the shared network, and FL also reduces the need for real-time processing of data. The training of the AI model is performed until the training results are deemed accurate.

NVIDIA is also announcing Clara AGX, "an embedded AI developer kit that can handle image and video processing at high data rates." This will bring AI services to healthcare service providers. The Clara AGX runs on NVIDIA Xavier SoCs, they are low-powered and suitable for integration into existing medical equipment or operating as a parallel system.

NVIDIA is also proud to showcase Hyperfine, a portable MRI system that's based on Clara AGX. The Clara AGX SDK will also be made available soon, in order to grow the number of applications that are ready to operate off Clara AGX.

This article was first published in Hardware Zone.