Follow Datanami:
March 30, 2020

DarwinAI Unleashes COVID-Net

COVID-19 has gripped the world for the past month as it brought global industries to a screeching halt and overwhelmed healthcare systems. From AI and big data to supercomputing, tech has been fighting back by leveraging complex data analysis to aid in diagnosis, epidemiology and treatment. Now, DarwinAI, a Canadian AI startup, has announced a new tool to fight back against the pandemic: COVID-Net.

The news was announced in a blog post by Sheldon Fernandez, CEO of DarwinAI

“The global crisis brought on by COVID-19 has affected us all,” Fernandez wrote. “Like many businesses, we’ve been grappling with how to best deploy our skills in service of the present crisis. To this end, we have collaborated with researchers at the University of Waterloo’s VIP Lab to develop COVID-Net: a convolutional neural network for COVID-19 detection via chest radiography.”

Differentiating between COVID-19 and non-COVID-19 pneumonia on x-rays or CT scans is key to quickly diagnosing the disease in critically ill patients, but differentiation can be difficult for humans, leaving ample room for AI to improve the process. COVID-Net isn’t the first tool to apply AI for COVID-19 diagnosis: early in the pandemic, Chinese hospitals deployed an AI-powered detection tool by Beijing startup Infervision at 34 hospitals, helping to examine more than 32,000 patients, and another Chinese AI-based model was trained by China’s first petascale supercomputer to perform a similar function.

Example chest radiography images of COVID-19 cases from two patients and their associated critical factors. Image courtesy of DarwinAI.

DarwinAI knows that it isn’t the first. “However,” the authors of the tool’s GitHub page wrote, “to the best of [our] knowledge, these developed AI systems have been closed source and unavailable to the research community for deeper understanding and extension, and unavailable for public access and use.”

COVID-Net, they explain, is open source and available to the general public. While they caution that the tool is “by no means a production-ready solution” and is not intended for direct clinical diagnosis, the authors hope that the tool “will be leveraged and [built] upon … to accelerate development of highly accurate yet practical deep learning solutions for detecting COVID-19 cases and accelerate treatment of those who need it the most.”

DarwinAI also announced that it was making the underlying dataset available. The dataset, called COVIDx, comprises nearly 16,756 chest radiography images from nearly 14,000 patients. The team is also accepting submissions to add to its dataset via e-mail (addresses are listed on the GitHub page). The dataset (which started at less than half that size) has since received major data contributions from organizations including the Radiological Society of North America and Mila.

Finally, the COVID-Net team announced that they are currently working on another AI-powered tool: COVID-RiskNet, “a deep neural network tailored for COVID-19 risk stratification” that would be able to help hospitals manage the huge influxes of COVID-19 patients based on risk. On that front, the team said, “stay tuned.”