Researchers at the University of Waterloo have partnered with an artificial intelligence (AI) startup on a project that uses AI to improve COVID-19 screening using X-rays. Although not production ready yet, it already can differentiate with high precison lung infections from bacteria, non-COVID-19 virus and COVID-19 virus.

The Waterloo research team publicly released AI software that can better detect infections from chest x-rays and is looking to enlist expertise from around the world to aid in the project.

“This software has had promising initial results,” said Alexander Wong, a systems design engineering professor and director of the Vision and Image Processing (VIP) Lab at Waterloo. “We hope that by making this software open, we can attract clinicians and scientists far and wide to improve upon the technology.”

Wong launched the COVID-Net project in conjunction with DarwinAI, a Waterloo startup, this week.

The new, AI-assisted, x-ray screening method is meant to augment the polymerase chain reaction (PCR) swab tests now in short supply in many areas of the world on the front lines of the coronavirus pandemic.

“Chest radiography can be conducted quickly and is relatively low-cost and widely available. It is already used by several countries to complement PCR tests,” said Wong. “Augmenting it with AI to improve screening accuracy could have a lot of value.”

In addition to the deep-learning AI software researchers developed to detect COVID-19 from chest x-ray images, they have made a dataset and a scientific paper on their work publicly available on GitHub at https://github.com/lindawangg/COVID-Net

“Our hope is that COVID-Net will be used and built upon by clinicians, researchers, and citizen data scientists,” said Wong, who is also a founding member of the Waterloo Artificial Intelligence Institute. “Ideally, this tool will allow us to accelerate the global use and development of effective, radiography-based COVID-19 screening solutions and treatment of those who need it the most.”

To develop their screening tool, the researchers used public sources to compile a dataset of almost 6,000 chest radiography images from nearly 3,000 patients.  It was found in early studies that patients present abnormalities in chest radiography images that are characteristic of those infected with COVID-19.

The predictive results are impressive and will improve with additional cases being learned by the AI system:

Positive Predictive Value (%)
Normal lung	Bacterial	Non-COVID19 Viral	COVID-19 Viral
95.1	87.1	67.0	80.0

By no means a production-ready solution, the hope is that the open access COVID-Net, along with the description on constructing the open source COVIDx dataset, will be leveraged and build upon by both researchers and citizen data scientists alike to accelerate the development of highly accuracy yet practical deep learning solutions for detecting COVID-19 cases and accelerate treatment of those who need it the most.