NVIDIA Clara and XNAT Join Forces to Easily Deliver AI-in-a-Box to Hospitals on NVIDIA Edge Computing Platform

NVIDIA today introduced the integration of XNAT, the most widely-used informatics platform for imaging research, and NVIDIA Clara, a collection of developer toolkits built on the NVIDIA compute platform aimed at accelerating compute, artificial intelligence, and advanced visualization.

NVIDIA Clara and XNAT will jointly be served on the NVIDIA EGX Platform, a high-performance, cloud native edge computing platform that brings AI to medical imaging devices and other tools at the point of care. This framework is aimed at providing informaticists, data scientists, and ultimately radiologists, with access to valuable AI workflows in real time.

The core of any AI solution consists of three key components: compute execution infrastructure, dataset management, and model management. These three components allow the core workflows of AI: annotation, training, validation, and inference, to be realized. Accelerated by GPU infrastructure, the Clara application framework includes developer tools that provide these services. XNAT, a web-based dataset, and model management makes it easy for those performing and supporting data science to focus their time on processing data and delivering the vision of artificial intelligence in medical imaging, and not on needlessly staging custom unmanaged file repositories of data and models. 

XNAT provides capabilities such as connectivity into the DICOM ecosystem, anonymization services, secure access and permission control, integrated search and reporting, pipeline processing, and modular extensibility. For example, XNAT provides interfaces to create a dataset based on certain criteria such as, “all CT studies of the abdomen with slice thickness thinner than 1mm ordered from the oncology department.” Studies can then be annotated using XNAT’s web-based viewer, leveraging NVIDIA Clara’s AI-assisted annotation tool to rapidly annotate structures within the imaging studies. These datasets can then be used to train appropriate AI models for the automated detection of structures in new studies – and ultimately, to validate such models and conduct inference for research and clinical analytics.

These two technologies come together to form the next advancement in AI, federated learning. Federated learning occurs where datasets are managed in on-premise repositories and do not leave the institution. Deployed through NVIDIA EGX, each participating hospital is able to rapidly deploy an on-premise environment where they can create and annotate their datasets. A group of hospitals can initiate federated learning from there, each contributing model weights to the federated model without exposing patient-confidential information, ultimately contributing to a model using a broader set of data than what could be used on just one institution’s dataset. XNAT empowers each hospital in the creation of their local datasets, and Clara provides the accelerated deep learning techniques to contribute to the trained model. 

“We built XNAT to enable the building of datasets securely, at a massive scale, and to easily annotate and compute on these data” say Daniel Marcus, Director of the XNAT program and Professor of Radiology at Washington University School of Medicine. “With its ability to accelerate image annotation and model development, Clara integration takes XNAT to the next level.  These are capabilities we get a lot of requests for, so I expect we’ll see rapid adoption of Clara across the global XNAT community.”

“It takes a village to build an AI platform,” says Brad Genereaux, Medical Imaging Alliance Manager with NVIDIA. “Building upon what the medical imaging community has been asking for in 2019 – a solution that takes into account dataset management, model management, and AI computation, with a smart interface for administration and privacy access controls – this is a means to meet those needs. What excites me most about this collaboration is how well they complement one another in providing hospitals and data scientists with the ecosystem they need to make this real. Powered by NVIDIA Clara and fed data by XNAT, we collectively unlock and enable a rapid, open, on-premise, packaged-in-a-box, web-based swiss-army-knife of AI tooling.”

NVIDIA Clara and XNAT deployed on the NVIDIA EGX platform will simplify the management of distributed fleets of AI-enabled medical systems to bring real-time AI analysis to medical imaging workflows, and prepare for the next wave of exciting medical breakthroughs.

Visit NVIDIA’s booth at RSNA #10939, or http://ngc.nvidia.com to get started.


XNAT is an open source software supported by the National Institute for Biomedical Imaging and Bioengineering, the National Cancer Institute, the National Institute of Neurological Disorders and Stroke, and the Mallinckrodt Institute of Radiology.  Contact the XNAT team at info@xnat.org to schedule a demonstration and no-cost consultation.