Canadian startup DarwinAI and researchers from the University of Waterloo are open-sourcing COVID-Net, a convolutional neural network that aims to detect COVID-19 in X-ray imagery. In response to the pandemic, a global community of health care and AI researchers have produced a number of AI systems for identifying COVID-19 in CT scans.
Companies like Alibaba and AI startups RadLogics and Lunit claim they’ve created systems capable of recognizing COVID-19 in X-ray or CT scans with more than 90% accuracy. Early work from Chinese medical researchers and a system published in the journal Radiology last week demonstrated similar results.
Like other companies making AI to detect COVID-19 from chest X-rays, DarwinAI said it’s creating COVID-Net and the accompanying COVIDx data set to give doctors a way to quickly triage and screen potential cases. But DarwinAI says that unlike other projects, its work is being open-sourced to make the neural net available to radiologists and researchers around the world.
“[Though it is 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 data set, will be leveraged and built upon by both researchers and citizen data scientists alike to accelerate the development of highly accurate yet practical deep learning solutions for detecting COVID-19 cases and accelerate treatment of those who need it the most,” the paper reads.
Dr. Alexander Wong is an associate professor at the Waterloo AI Institute, codirector of the Vision and Image Processing Group at the University of Waterloo, and lead researcher at DarwinAI.
He said the goal is to revolve COVID-Net around the open source open access movement to enable innovation and access for others to build upon.
“If you remember back to things like ImageNet or AlexNet, the open aspect completely changed the way deep learning works. My hope is that this will change the way people actually work together to build a solution for our common problem right now,” he said.
COVID-Net is trained using COVIDx, a data set comprising nearly 6,000 X-ray images of 2,800 patients from a Kaggle challenge, as well as the COVID chest X-ray data set. COVIDx contains only 68 X-ray images from 19 confirmed COVID-19 cases, according to an arXiv paper released this week. The data set also includes hundreds of non-COVID-19 viral infection images, like SARS, MERS, and influenza. COVID-Net also uses DarwinAI’s explainability tools to highlight areas the model uses to justify its decision-making.
According to the paper, in initial results COVID-Net was able to detect COVID-19 in 83.5% of cases. COVID-Net is designed to differentiate between COVID-19 and influenza, SARS, and MERS, though at launch today it can return a high number of false positives.
While AI startups and researchers move forward with efforts to create computer vision that recognizes COVID-19 in medical imagery, the Centers for Disease Control and Prevention (CDC) in the United States currently does not recommend the use of CT scans or X-rays for COVID-19 diagnosis.
In recent weeks, the American College of Radiology (ACR) and similar radiological organizations in Canada, New Zealand, and Australia also released statements telling radiologists that they do not currently recommend the use of CT scans for COVID-19 detection.
The problem, said American College of Radiology (ACR) Thoracic Imaging Panel chair Dr. Ella Kazerooni, is we’re in the midst of influenza season, and it’s hard to tell the difference between COVID-19 and common lung infections like bacterial or viral pneumonia. She said even if a chest X-ray exhibited signs of COVID-19, a lab test is still required for confirmation.
“If you suspect a patient has COVID, you’re going to test them. If they have mild respiratory symptoms, you’re going to test them and send them home to quarantine. If you think they have COVID and they’re really sick and need to be admitted to the hospital, you’re going to admit them, take care of them, and perform the viral test,” she said. “I think the ACR is supportive, but if there’s not an indication to do the test to begin with, then what’s the application?”
Kazerooni criticized research that makes no attempt to distinguish between other illnesses that appear in CT or X-ray imagery.
Despite its stance that radiologists should avoid using chest X-rays or CT scans to diagnose COVID-19, the ACR is not against putting AI to use. Earlier this month, the ACR’s Data Science Institute released an open project call for AI that can detect COVID-19 from CT scans. ACR DSI Thoracic chair Dr. Eric Stern called AI for COVID-19 detection from CT scans a potentially useful tool in health care settings where there are few human radiologists available.
Wong said he believes the ACR’s current recommendation is based on the fact that it’s hard for people to differentiate between COVID-19 and other illnesses that appear in medical imagery.
“By being able to train a deep neural network to capture the fine nuances, we’re able to show at least preliminary results are quite promising,” Wong said. “Our goal here is if we’re able to build an AI that could assist a radiologist or clinician to be able to differentiate that, it breaks the underlying barrier.”
Partner organizations interested in sharing chest X-ray imagery to grow the COVIDx data set will be asked to fulfill certain privacy and ethical guidelines. DarwinAI may add a web login for radiologists to scan images, while the use of federated learning to preserve privacy may be considered for COVID-Net in the future, Wong said.
VentureBeatVentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:
- up-to-date information on the subjects of interest to you
- our newsletters
- gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
- networking features, and more