TORONTO -- Researchers at the University of British Columbia are compiling CT scans and chest X-rays from around the world to create a global dataset aimed at helping physicians determine the best treatment courses for people with COVID-19.
Thanks to a partnership with Amazon Web Services, the UBC team is sharing its data online for free, with the goal of helping in the battle against the novel coronavirus by using predictive modelling to better diagnose the severity of the disease and improve treatment.
Radiology resident Dr. William Parker and his research partner Dr. Savvas Nicolaou, a professor of radiology at UBC and the director of emergency and trauma radiology at Vancouver General Hospital, began collecting CT images from colleagues in multiple countries in January. They developed an artificial intelligence algorithm to better identify the percentage of lung tissue involvement and the subtle patterns of infection documented in the CT scans and what that indicates about how a patient may fare in the long run.
Developing a better understanding of how the virus presents in CT images will help doctors identify which patients “will do better to go home and self-isolate and which ones may need more support, like ventilation and ICU admission,” Parker told CTV’s Your Morning on Friday.
“And unfortunately, some patients go on to death.”
The effects of COVID-19 on the lungs show up on CT scans in a white haziness where healthy lungs would appear dark. Parker says the disease first presents as a “subtle ground glass” in the lungs. If the patient doesn’t improve, the lung tissue appears snowy and opaque.
The UBC researchers believe they are compiling the largest dataset of COVID-19 positive images from around the world, including North America, Europe, Asia and Australia. The UBC researchers have also partnered with three artificial intelligence coding teams that are developing models.
They hope to release the first version of their cloud-based AI software on May 10, so that other researchers and companies can start using it.
CT scans are believed to hold promise as a diagnostic tool. A recent study of more than 1,000 COVID-19 patients in China found CT images better detected the disease than commonly used polymerase chain reaction diagnostic tests.
The idea is that, with enough data and modelling, the system could produce “virtual twins,” which would allow doctors to predict how a patient would react to treatment by comparing them to a model of themselves created through the compiled data from similar patients.
UBC was among the first recipients of funding through AWS’s global Diagnostic Development Initiative announced in March, which offers an initial US$20 million in cloud credits and technical support to accelerate research and development of coronavirus diagnostic tools.
