- About XNAT
- News & Events
- XNAT Marketplace
- Contact Us
For the last several months, we have been working on adding support for Machine Learning and model training workflows to XNAT, in collaboration with NVIDIA, Radiologics, and the ICR Imaging Informatics group. We first demonstrated a proof of concept using models and APIs from the NVIDIA Clara™ Imaging framework with accelerated GPU computing at the 2019 RSNA conference, and have since been working to standardize these features and also add functionality.
What Can You Do With the XNAT Machine Learning Suite?
With XNAT ML (Beta), you can assemble training-specific collections of your imaging data files in a given project, draw new segmentations and annotations on that data, install and configure a training model, then train that model on your imaging data to enable an AI-assisted annotation workflow. Future releases in the XNAT ML line will enable model sharing and inference.
The XNAT ML (Beta) distribution is built on a pre-release version of XNAT 1.8, which is built on Java 8, PostgreSQL 12, and includes a number of enhancements, plugins, and custom components to support the model training workflow. All of these components are wrapped up in a docker-compose package, to give you a single-step installation process. This package includes documented workflows with known limitations. We are looking for your feedback on how we can improve this toward a full production release.
NVIDIA Clara™ Imaging is an application framework that accelerates the development and deployment of AI in medical imaging. Built for data scientists and researchers, Clara Imaging offers easy-to-use, domain-optimized tools to create high-quality, labeled datasets, collaborative techniques to train robust AI models, and end-to-end software for scalable and modular AI deployments.