“The process repeats until the AI model reaches its desired accuracy.”
This week at the annual conference of the Radiological Society of North America, machine learning and AI developers Nvidia unveiled its federated learning platform Clara. A system designed to protect patient privacy while still enabling medical centres to collaboratively train models and process patient data.
Due to the sensitive nature of medical data hospitals and medical centers, for both legal and privacy concerns, don’t share images or data. Unfortunately, keeping all of this data in a silo means that no AI or ML models can be trained on it. Federated learning involves creating a central global server that sends a training algorithm to each medical centre taking part in the model training. Each institution trains the model on their private dataset, before sending it back to be aggregated by the central server. At no point does the sensitive data leave the medical centre.
Hospitals or medical centres using the Clara Federated Learning (Clara FL) system will label all of their data using an AI-assisted annotation SDK that is currently integrated in medical viewers such as 3D slicer, MITK and Philips Intellispace Discovery. This data is then trained on in-house servers before it is sent to the global server.
Nvidia commented in a release that: “Clara FL is a reference application for distributed, collaborative AI model training that preserves patient privacy. Running on NVIDIA NGC-Ready for Edge servers from global system manufacturers, these distributed client systems can perform deep learning training locally and collaborate to train a more accurate global model.”
Clara Federated Learning
Currently the University of California, Los Angeles is using Clara FL to introduce AI technology into its radiology department. Radiology departments produce a wealth of medical images captured from patients with a host of medical conditions, these include X-Rays, Computed tomography (CT scans) and magnetic resonance imaging (MRI) images.
Training an AI model on these images in order for it to spot patterns and help identify potential illness earlier in the diagnostic process has in the past been difficult due to patient privacy concerns, as such these medical images rarely leave the radiology department. With Clara AI and ML models can be trained without compromising patient privacy.
Clara FL operates using Nvidia’s EGX Edge platform, a high-performance platform that has been created to tackle the massive amounts of data created by modern technology. The EGX stack includes a driver, Kubernetes plug-in, container runtime plug-in and GPU monitoring software. Telcos can install all required Nvida software as containers that run on Kubernetes, giving flexibility. (The stack architecture is supported by Canonical, Cisco, Nutanix, Red Hat and VMware.)
Nvidia have also teased the release of Clara AGX an AI developer kit that aims to process high-data rate video and images that are flooding in from sensors embedded in medical devices.
Nvidia states that the: “Clara AGX is powered by NVIDIA Xavier SoCs, the same processors that controls self-driving cars. They consume as little as 10W, making them suitable for embedding inside a medical instrument or running in a small adjacent system.”