May 29, 2020
Listen to an audio version of this story.

What would the coronavirus crisis look like if you could walk into an X-ray clinic, get a scan and receive a tentative positive or negative assessment instantly?

That’s a potential scenario offered by COVID-Net, an artificial intelligence developed by Waterloo startup DarwinAI and University of Waterloo researchers — and released freely to the public.

COVID-Net is a neural network — a brain-like AI — that can look at an x-ray of your chest and predict whether you have COVID-19 in less than 10 milliseconds, according to DarwinAI CEO Sheldon Fernandez. Its documentation says it catches the infection eight or nine times out of 10.

If you get a standard chemical-based PCR test, you might have a swab stuck up your nose, the sample will be taken to a lab in, say, Hamilton and you’ll get your result back up to four days later, according to the Public Health Ontario website. The test consumes precious physical and chemical supplies.

“COVID-Net, which uses X-rays, is a complementary tool,” Fernandez said in an interview Thursday, April 2. “It will not give you a(n) answer with 100 per cent certainty, but it’ll give you a statistical answer so that you can rapidly screen a lot of people and say, ‘Look, there’s a (for example) 93 per cent chance you have corona. Go home and self isolate.’”

Fernandez cautions the AI’s prediction can’t replace a true chemical test’s verdict. But the faster screening could mean people take steps not to spread the virus earlier and make Ontario’s (former) backlog of PCR tests less significant, Wilfrid Laurier University epidemiology expert Todd Coleman said in an interview April 1.

The COVID-Net team released the AI — light- and heavyweight models, with a more capable network coming — to the public in hopes of prompting collaboration.

“Researchers from Turkey, from India, from Singapore, Italy, Mexico, (the) UK have all reached out,” Fernandez said, in a show of solidarity he called “overwhelming.”

But the AI isn’t ready for use by the medical community yet. First, it needs an application with a user interface so non-computer-scientists can interact with it.

“That’s what we’re hoping to get some (government) funding to actually create.”

Fernandez first announced the project’s release March 22.

Leave a Reply