DarwinAI Open-Sources COVID-Net as Medical Imaging in COVID-19 Diagnosis Debate Continues
As the world struggles with the COVID-19 pandemic, the global scientific community is exploring all options in an effort to develop new ways to fight back. Effective screening of infected patients plays a critical role on the front lines, and the gold standard for this is Polymerase Chain Reaction (PCR) testing.
A recently proposed alternative COVID-19 screening alternative is AI-powered diagnosis based on chest radiography images such as X-rays or computed tomography (CT) scans.
Over the last few years AI researchers have demonstrated numerous ways that computer vision diagnostic systems can help radiologists interpret medical images with greater speed and accuracy.
Researchers Linda Wang and Alexander Wong from the University of Waterloo Vision and Image Processing Lab and the Canada Waterloo Artificial Intelligence Institute recently developed and open-sourced COVID-Net, a convolutional neural network for detecting COVID-19 through chest radiography.
Wong is the Canada Research Chair in AI and Medical Imaging and Co-founder and Chief Scientist of Canadian startup DarwinAI. He told Synced, “We made model and data all available open source and open access on GitHub… This is the first time where an AI explainability strategy is leveraged to give deep insights into the visual indicators that COVID-Net leverages to make COVID-19 decisions, which will hopefully help clinicians in better screening and trust in the system.”
Researchers trained and tested COVID-Net using the COVIDx dataset, which consists of almost 6,000 chest X-ray images from 2,839 patients from the COVID chest X-ray dataset; and the Kaggel chest X-ray images (pneumonia) dataset which covers bacterial pneumonia, non-COVID19 viral pneumonia, and no pneumonia classes. DarwinAI this week updated the COVIDx dataset, which now has 16,756 chest X-Rays across 13,645 patient cases.
The team also used DarwinAI’s generative synthesis based explainability method GSInquire which the company says can investigate how COVID-Net makes predictions and where critical factors in the input chest X-rays are determined.
Experiment results showed that COVID-Net can detect COVID-19 infection with a positive predictive value (PPV) of 88.9 percent.
There are however ongoing discussions in the scientific community regarding the appropriateness of radiography scans for COVID-19 diagnosis, as these approaches require hospital hardware while PCR testing — the most widely performed diagnostic method — is reasonably fast and can accurately detect SARS-Cov-2 RNA through simple nasal swabs.
The American College of Radiology (ACR) recommends CT scans be used sparingly in COVID-19 detection, reserved for hospitalized, symptomatic patients with specific clinical indications for CT. Appropriate infection control procedures should be followed before scanning subsequent patients, and CT should not be used to screen for or as a first-line test to diagnose COVID-19. The USA CDC and many international radiological organizations also remain unconvinced regarding the use of CT scans for COVID-19 diagnosis.
But the COVID-19 pandemic has thrown the world into an unfamiliar and uncertain place, with strategies and tools in constant evolution. While concerns continue, the plan for DarwinAI right now is to work with different clinical sites across Canada to collaborate on data collection and validation in order to improve COVID-Net prediction quality.
Globally, some hospitals have already deployed AI tools to detect COVID-19 using chest scans. On February 21, Chinese tech giant Alibaba announced a new AI algorithm that can diagnose suspected cases in 20 seconds with 96 percent accuracy. The algorithm was trained with data and CT scans from more than5,000 confirmed COVID-19 cases and first deployed at Wuhan №6 Hospital in the centre of the outbreak. Alibaba says as of mid-March the AI system had been used in 26 hospitals and helped diagnose over 30,000 cases.
On March 12 researchers from RADLogics, Tel-Aviv University, New York Mount Sinai Hospital and University of Maryland School of Medicine introduced an AI-based deep learning image analysis system based on CT chest scans that can assist clinicians by accurately monitoring disease progression or regression in patients.
On March 30 a team of University of San Diego researchers open-sourced their COVID-CT-Dataset, which contains 275 CT scans collected from 143 confirmed cases and is one of the largest such collections. The team noted that although studies have show the potential of CT scans in screening and testing COVID-19, little data has been shared with the public due to privacy concerns: “This greatly hinders the research and development of more advanced AI methods for more accurate testing of COVID-19 based on CT.”
The DarwinAI researchers say that the scarcity of COVID-19 radiography images in the public domain was a main motivation for their open-sourcing the COVID-Net project, as training data is key to boosting the speed and accuracy of AI-powered CT lung scan based diagnosis.
In a recent blog post, DarwinAI CEO Sheldon Fernandez said the company plans to add at least 500 chest X-rays of COVID-positive patients to improve COVID-Net performance.
The paper COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest Radiography Images is on arXiv. The open source project is on GitHub.
Journalist: Fangyu Cai | Editor: Michael Sarazen
Thinking of contributing to Synced Review? Synced’s new column Share My Research welcomes scholars to share their own research breakthroughs with global AI enthusiasts.
We know you don’t want to miss any story. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.
Need a comprehensive review of the past, present and future of modern AI research development? Trends of AI Technology Development Report is out!
2018 Fortune Global 500 Public Company AI Adaptivity Report is out!
Purchase a Kindle-formatted report on Amazon.
Apply for Insight Partner Program to get a complimentary full PDF report.
