Screening for COVID-19 with AI
Open-source project takes next step with release of new models, largest datasets of CT scans
Open-source project takes next step with release of new models, largest datasets of CT scansBy Brian Caldwell Faculty of Engineering
Researchers at Waterloo Engineering have taken another step forward in their open-source COVID-19 project with the release of new datasets and artificial intelligence (AI) models that were built using them.
Launched in March, the initiative involves the use of deep-learning AI software to screen chest x-rays and CT scans for evidence of infection and degree of disease progression.
The new datasets, released as part of the COVID-Net project, include CT scans from over 4,500 patients in more than 15 countries, making them the largest and most diverse resources of their kind in the world.
The new AI models, dubbed COVID-Net CT-2, have been found via explainable AI to make their decisions on COVID-19 screening with CT scans using some of the same types of visual cues as expert radiologists.
“Taking this path enables building AI with much greater transparency and trust,” Alexander Wong, a professor of systems design engineering, Canada Research Chair in Artificial Intelligence and Medical Imaging, and a director of the Vision and Image Processing (VIP) Research Group, wrote in a post announcing the new developments.
“We hope making the AI models and datasets available will drive global innovation in healthcare.”
Photo: CT scans of lungs show areas of concern highlighted in red by new AI models developed by Waterloo Engineering researchers.