Nvidia’s Clara AI platform made its official debut this week at the 19th GPU Technology Conference in San Jose, California.

Equipped with software tools and 13-state-of-the-art classification and segmentation AI algorithms, the toolkit is designed to create and deploy AI in clinical workflows, and to integrate the latest advancements in AI, high-performing computing, and visualization to create intelligent imaging modalities.

“The goal of Clara AI is to enable more participation in developing AI by domain experts that might not have expertise in data science or modern software methodologies,” Chris Scotto DiVetta, product management at Nvidia, told HCB News. “We will continue to create tools to remove the barriers of adoption in areas like creating data sets to feed deep learning training, sharing knowledge between collaborators, and integrating into the clinic.”

Among the main functions of the toolkit is the integration of AI-assisted Annotation SDK into any medical viewer. The addition of this feature provides radiologists with AI tools that can annotate images 10 times faster for greater deep learning training, speeding up the creation of structured data sets, and enabling annotations in minutes instead of hours.

It also provides transfer learning, which allows users to rely on less data to create highly accurate deep learning models on the basis of other preexisting ones. In adapting existing models to fit local variables, the function can create deep learning algorithms to data comprising local demographics and imaging devices, without moving or sharing patient information, and enables physicians to create models of their own patients with 10 time less data they would need if creating a model from scratch. The toolkit facilitates the integration of these models into existing radiology workflows, using industry standards such as DICOM.

The 13 incorporated AI algorithms further assist the objective of the Clara AI platform by simplifying the segmentation of organs in 3D space. This allows automatic measurements such as size and growth of tumors to take place faster and more accurately.

The system is currently in use at a number of institutions, including Ohio State University where radiologists used it to incorporate and validate a model from another institution, and annotate a local data set to adapt the model for its patients and speed up AI development of effective algorithms for clinical care.

Others are the National Institutes of Health Clinical Center, for developing a domain generalization method for the segmentation of the prostate from surrounding tissue on MR scans; and the University of California, San Francisco for the creation, testing and deployment of multiple algorithms across radiology that can serve as guides to help future clinicians adopt the system.

“This is a toolkit for developers in medical imaging to commercialize. We expect them to use these tools and build on them, commercializing it for their customers," said Scotto DiVetta. "We are trying to broaden the definition of developer to include more domain experts.”

Development of the platform was announced at last year's GPU Technology Conference, with Nvidia saying it would be a "single virtual supercomputer".

Included in the toolkit are two software development toolkits, Clara Train SDK and Clara Deploy SDK. Both can be accessed from NGC, and deployed on the NVIDIA T4 server and NVIDIA DGX POD. Clara Deploy SDK open beta is available today on NGC. Clara Train SDK is also available for early access.