NVIDIA's AI Models Joined With NIH Might Help Curb COVID-19 Fall 2020

New AI models from NVIDIA and NIH may see rapid availability to differentiate COVID-19 pneumonia.
Brad Bergan

New AI models from NVIDIA and the National Institutes of Health (NIH) might help researchers analyze COVID-19 in chest-intensive CT scans to develop novel tools in the fight to measure and detect COVID-19 infections this Fall, according to a new paper published in Nature Communications.


New AI models from NVIDIA might help curb COVID-19

In the paper, the NIH and NVIDIA showed how they developed AI models to help scientists learn how COVID-19 progresses via CT scans in the chest. Publicly viewable on NVIDIA NGC, this research will help researchers in the global fight to curb the COVID-19 crisis.

The duo's new classification model for differentiating pneumonia from COVID-19 from other causes of pneumonia could become extremely useful come Fall cold and flu season — to help distinguish COVID-19 from other viral, fungal, or bacterial types of pneumonia contracted from a community.

The new model is strong because it was trained with a diverse and multinational data set comprised of more than 1,200 patients. COVID-19 CT scans were acquired from four hospitals — located in Italy, China, and Japan, where a wide spectrum of clinical timing and practice for CT use in outbreak applications were available.

This model showed potential for scaled-up generalizability once it was tested on an independent and previously-unseen demographic, according to a blog post on NVIDIA's website.

NVIDIA 3D Classification Workflow
NVIDIA's workflow of 3D classification. Source: NVIDIA

Two AI classification models to combat COVID-19

In all, 2,724 scans taken from 2,619 patients were used in the new study — consisting of 1,029 scans of 922 patients with RT-CPR confirmed COVID-19 and lung lesions associated with COVID-19-inflicted pneumonia. Of the scans, 1,387 from 1,280 patients helped progress algorithm development, with 1,337 patients used for algorithm testing, analysis, and evaluation.

Two models fell under the purview of this work and both were used in a series to develop the COVID-19 final classification model. The first was segmented — used to define lung regions that were later employed in the classification model.

At first, they developed two classification models — on to utilize the entire lung region with fixed input size (full 3D), and another using the average score of multiple regions inside each lung with a fixed image resolution.

CT scans may help differentiate COVID-19 pneumonia from other illnesses, immediate availability

Developed with the NVIDIA Clara application framework used for medical imaging — which includes domain-specific AI training and deployment workflow tools that helped NIH and NVIDIA create the models in less than three weeks.

AI may take up a new role for CT-enhanced diagnosis as a result of these models' ability to differentiate between COVID-19 pneumonia and pneumonia associated with other causes. Notably, it could be available immediately.

Further models might include point-of-care detection for the isolation of asymptomatic patients, resource allocation, prognostication, or monitoring the development of infections in response to clinical trials for medical countermeasures — like anti-viral drugs, monoclonal antibodies, or the severity of the disease amid COVID-19 vaccine trials.

We have created an interactive page to demonstrate engineers’ noble efforts against COVID-19 across the world. If you are working on a new technology or producing any equipment in the fight against COVID-19, please send your project to us to be featured.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board