TORONTO – Within a week of completing clinical trials the chest radiography AI tool developed by Vancouver, B.C.-based 1Qbit Inc. has been given the all-clear from Health Canada for deployment across the country. The XrAI was originally developed to better identify patients with respiratory illness including SARS, pneumonia and tuberculosis (TB), but then in February was tested on a publicly available data set of COVID-19 X-ray images.
“What had been most important for us was TB which is still one of the deadliest diseases in the world,” 1Qbit Chief Medical Officer Deepak Kaura explained. “What we then saw on a chest X-ray of a patient with COVID-19 infection were common abnormalities that our algorithm was trained to identify.”
Single blinded, randomized controls
COVID-19, SARS and TB belong to what is called “community acquired pneumonia,” and are associated with various lung abnormalities, Kaura told BioWorld. These include partial collapse of the lung, abnormal density in lung tissue and in more serious cases pleural effusion, i.e., excess fluid build-up in membranes lining the lungs and the inside the chest cavity.
The XrAI algorithm doesn’t diagnose COVID-19, Kaura was quick to stress. Rather it provides clinicians with high or low numerical values of confidence indicating the presence or absence in lung X-rays of abnormalities associated with COVID-19. Get a rating of 98% any where in the lung and you can be pretty sure abnormalities associated with disease are present.
“We recognized that by using the algorithm the frontline physicians who are supposed to be `actioning’ the abnormality of the chest radiograph are significantly helped by having this AI co-pilot beside them,” said Kaura. “And our study showed that beautifully.”
That clinical study, scheduled to begin later this spring, was brought forward three months, 1Qbit President Landon Downs told BioWorld, with 28 clinicians working overtime to expedite results. Prior to the study, a 1Qbit researcher also ran COVID-19 chest X-ray images contained in a New England Journal of Medicine article through the algorithm. “The algorithm performed flawlessly detecting abnormalities,” said Downs.
The clinical trial itself was a gold standard, randomized control study and single blinded so that 1Qbit had no idea who the participating clinicians were. Funded by the Saskatchewan Health Authority, five separate groups of radiology, pulmonary, ER, family physicians and radiology residents were each given 500 radiographs to review.
“The outcomes I think are really wonderful in that the algorithm improves the performance and accuracy of the diagnoses rendered by those physician groups compared to those that did not have it,” said Kaura. “The percentages of improvement were quite significant and will be published at a later date.”
The world comes knocking
1Qbit has been in conversation or signed letters of partnership with no fewer than eight provincial heath authorities across Canada, plus governments in Nigeria, Ethiopian and Kenya. In fact, Kaura has traveled extensively for more than seven years to places where radiologists are few and far between, “so that we can provide a tool like this that assists doctors and nurses making this diagnosis.”
“Nothing is cemented yet, but we can build this tool in Canada, working and designing it in conjunction with health authorities and physicians and then use it as a tool in other parts of the world,” Downs added. Granted broader global approvals, the XrAI would complement the other respiratory imaging tools these countries use to identify COVID-19, but also more traditional diseases such as TB.
“TB still kills upward of 1.5 million people a year,” said Downs, “and even though it’s not something we think about as a disease in Canada, it’s something that’s actually very pervasive in Canada’s north as well as in indigenous communities.”