NVIDIA recently released Clara Train 3.1 for healthcare developers to collaborate on secure, enterprise-grade AI models.
Building robust AI can be a challenge for healthcare organizations due to the massive amounts of data required to produce reliable algorithms. With Clara Train, organizations can share and combine their local knowledge to create global models without compromising privacy.
Clara Train 3.1 introduces a number of features to help healthcare developers scale federated learning securely and boost research productivity. Provisioning a client site is easier than ever with a provisioning tool that simplifies and streamlines deployment. A new flexible authorization framework enhances security to ensure sensitive data is protected. And last but not least, the new administration tool enables a 10x increase in algorithm experimentation to boost researcher productivity.
Last month, NVIDIA and collaborating institutions used Clara Federated Learning to train an AI model that predicts the oxygen needs of patients presenting to emergency rooms with COVID-19 symptoms. The initiative, called EXAM (EMR CXR AI Model), was the largest real-time clinical federated learning initiative to date, with contributions from 20 hospitals around the world. Researchers at individual hospitals used chest x-rays, patient vitals, and lab values to train a local model and share only a subset of model weights with the federated server to train a global model using NVIDIA’s Clara Federated Learning Framework. The global model achieved an AUC of 0.94.
Various pre-trained AI models for medical imaging are available for developers using Clara Train 3.1. These include 3D segmentation models for spleen, brain, prostate, and liver tumor and 2D models for chest x-ray classification. Developers can also leverage models to fight COVID-19, including COVID classification and COVID lesions segmentation, both trained on CT images.
To learn more about Federated Learning and the new features of Clara Train 3.1, watch the webinar, Collaborating on Global Healthcare AI Models with Clara Train 3.1.
Download Clara Train 3.1 from NGC, and try out the example Jupyter notebooks on GitHub.