Nvidia Corp. is stepping up its efforts in healthcare with today’s launch of its Medical Open Network for AI, or Monai, an open-source framework that’s used to train artificial intelligence-powered models for medical imaging.

Monai is based on the open-source PyTorch framework that was first created by Facebook Inc.’s research group. PyTorch is a framework for building deep learning models that attempt to emulate the way the human brain functions, and has led to major breakthroughs in areas such as language translation, voice recognition and image classification.

The Monai framework is essentially a version of PyTorch that has been streamlined to create AI models for medical imaging. Features include “industry-specific data handling, high-performance training workflows and reproducible reference implementations,” Nvidia said. It also provides more than 20 pre-trained models to work with, including some for COVID-19, and is optimized to run on Nvidia’s most powerful A100 graphics processing units to ensure rapid training.

Medical imaging is one of the most promising use cases for AI technology. AI models can assist medical professionals by helping them analyze computerized tomography or CT scans and other images to make diagnoses faster and with greater accuracy than is possible when doing so manually.

Kimberly Powell, Nvidia’s vice president and general manager of healthcare, said in a press briefing that PyTorch can be thought of as a training engine for AI models, and that “Monai is a wrapper around that” specifically for medical imaging training that can train models up to six times faster.

“This year and next year are defining moments in healthcare,” Powell said.

Monai is being released as part of the latest Nvidia Clara application framework for AI-powered healthcare. Nvidia Clara is the company’s computational platform for healthcare, and the latest release adds new 3D AI-assisted data labeling capabilities that enable radiologists to label complex 3D cat scan data using 10 times fewer clicks, thanks to a new model called “DeepGrow 3D.”

Instead of the traditional time-consuming method of segmenting an organ or lesion image by image or slice by slice, which can be up to 250 clicks for a large organ such as the liver, the company said, users can segment them with far fewer clicks.

Nvidia said the latest Clara annotation tools were recently used to label the public COVID-19 data set published by The Cancer Imaging Archive at the U.S. National Institutes of Health. That labeled data set was then used to develop a generalized AI model called EXAM that helps to predict the oxygen requirements for the most serious COVID-19 cases.

In addition, Clara is being expanded to pathology for the first time, Nvidia said. This is an important step because images used in pathology are often so detailed that most AI tools simply can’t handle them. The new Clara for pathology module comes with reference pipelines for both training and deploying digital pathology applications.

Alongside today’s software updates, Nvidia said it’s expanding the Inception Alliance for Healthcare, an accelerator program for AI startups that provides technology assistance, go-to-market support and discounts on Nvidia hardware, among other benefits.

Nvidia Inception Alliance for Healthcare members are now being invited to join the GE Healthcare Edison Developer Program run by General Electric Co. The program provides access to GE Healthcare’s global network, which has a global install base of more than 4 million imaging, mobile diagnostics and monitoring units across 160 countries. Moreover, startups with United States Food and Drug Administration clearance will also be able to join the Nuance AI Marketplace for Diagnostic Imaging, where they can sell their products directly to healthcare providers, Nvidia said.

Image: Nvidia