AI centres of excellence and companies collaborate on apps

Artificial intelligence and machine learning are playing an increasingly important role in virtually every sector of the economy, including healthcare. AI algorithms and machine learning models are now used to optimize resource scheduling, flag abnormal chest X-rays and predict the deterioration of patients in intensive care units.

Together, the Montreal Institute for Learning Algorithms (Mila), the Vector Institute in Toronto, the Alberta Machine Intelligence Institute (AMII) and the Vancouver-based Digital Technology Supercluster are positioning Canada as an AI superpower. All four organizations are partnering with hospitals and private-sector businesses to create innovative solutions that improve efficiency and patient care.

Progress is being made, “but there are a number of challenges we’re dealing with in healthcare,” said Azra Dhalla, the Vector Institute’s director of health AI implementation. “We work in a lot of data silos, so the quality and interoperability of the data is an issue. We also need to ensure there is a balance between privacy protection and making sure we can create solutions that have public benefit.”

Then there’s the issue of dataset shift and its impact on mission critical applications. “Just last summer, “said Dhalla, “we were in discussions with some of our healthcare partners about some of the models they’re developing, and it dawned on us to ask what mechanisms are in place for monitoring and evaluating machine learning prediction models because once a model goes into production its performance has the ability to change and, in some cases, even deteriorate.”

It was this realization that led to the Vector Institute’s CyclOps project, which has as its goal the development of a framework for “rigorous evaluation of machine learning models across time, hospital sites and diverse patient cohorts.”

“If I build a machine learning model that shows me a pair of socks and I don’t buy it, nothing happens,” said Deval Pandya, the Vector Institute’s director of AI engineering. In healthcare, however, an incorrect treatment recommendation from a machine learning model can have fatal consequences due to dataset shift.

“What happens a month after you deploy the model?” asks Pandya. “What happens when you take the model from hospital A to hospital B? And what happens when there’s a pandemic like COVID? To establish trust in the model, you have to do very rigorous evaluation, not just from the mathematical perspective, but also by bringing in the domain expertise of physicians and front-line workers who use the model.”

Pandya is aiming for a first release of the CyclOps framework on an open-source basis in the next few months and believes it will go a long way toward realizing the potential for machine learning in healthcare.

In addition to research projects like CyclOps, the Vector Institute and other AI-focused institutions across the country are engaged in the development of AI talent and the deployment of AI solutions. “Research isn’t enough,” remarked Dhalla. “We have to be a bridge to translate it into something tangible that can be used in a clinical setting.”

One recent application of AI co-funded by the Vancouver-based Digital Technology Supercluster and Synthesis Health is making a huge impact on access to healthcare in several remote Indigenous communities in Saskatchewan.

Just this summer, Dr. Deepak Kaura, a pediatric interventional radiologist and chief medical officer of Synthesis Health, delivered an integrated AI-powered diagnostic imaging solution to several Peter Ballantyne Cree Nation communities whose residents were previously required to travel up to six hours to Prince Albert for a simple chest X-ray. The solution includes an algorithm trained to flag abnormal chest X-rays, a cloud-based clinical management platform called Synth.OS and ultra-portable X-ray machines from Fuji that can be used in a nursing station or easily carried on a snowmobile to someone’s home.

The algorithm developed by Synthesis Health was trained on a million cases from the Saskatchewan Health Authority, and other sources, and has been approved by Health Canada.

The Synth.OS images are accessed in the cloud and read by radiologists within 24 hours, but nursing station staff in the remote communities receive an AI-generated report within minutes. “The thing that’s pretty amazing about what we’re doing is that it changes the economics of healthcare delivery as well as access,” said Dr. Kaura.

Using so-called portable X-ray machines on wheels that are popular and well-suited for hospitals, together with conventional PACS software, would cost hundreds of thousands of dollars, noted Dr. Kaura. “We can do the X-ray machine, the AI algorithm and a cloud-based PACS system with a reporting solution all for $2,100 per month. That’s in stark contrast to the cost of sending a few patients away every month for X-rays. We are now at the point where we have a lot of interest expressed by health systems from across the country and overseas.”

Synthesis Health is also leading the Digital Technology Supercluster-funded Iris project to develop an AI-based platform or “co-pilot for diagnostic imaging analysis and course-of-treatment planning and monitoring.” Announced in July, the $13.5 million project is being carried out in partnership with GE Healthcare, Konica Minolta Healthcare, the University of British Columbia, the Vancouver Coastal Health Research Institute and BC Cancer.

Iris is being developed to increase the speed and accuracy of front-line course-of-treatment decisions and improve the ability of radiologists to consistently find abnormalities in diagnostic images. The algorithms will be trained on more than 10 years of imaging data from Canadian health authority partners with the objective of seamlessly integrating them into PACS and radiology information systems.

Funding for the Iris project will also go toward the establishment of a National Advisory Council on AI in Healthcare. A healthcare specific council was thought necessary to bring together stakeholders from across the country. “There is a huge opportunity for Canada to lead the rest of the world in artificial intelligence, but we need to have a more intelligent and wholesome dialogue around how that happens and work toward overcoming the challenges we face,” said Dr. Kaura.

The Digital Technology Supercluster has also co-invested in several other AI-related projects, one of which uses AI technology to detect fragments of DNA shed by cancer tumors, said supercluster CEO Sue Paish. Another project uses AI to expedite the generation and validation of novel compounds with desirable properties for the development of new drugs to combat the SARS-CoV-2 virus.

The Alberta Machine Intelligence Institute (AMII) is also helping healthcare organizations harness the power of AI. In 2019, AMII received funding from PrairiesCan, the federal government’s economic diversification agency for the three Prairie provinces, to work with nine healthcare organizations on AI adoption, said Mara Cairo, product owner of AMII’s advanced technology team.

The Cross Cancer Institute in Edmonton and CardiAI, a Calgary-based medical device R&D company, were among the organizations with the most impressive results, said Mehadi Sayed, president and CEO of Clinisys, one of several companies hired to provide the nine companies with AI consulting services.

CardiAI, for example, applied AI technology to more efficiently and accurately analyze Holter monitor data for abnormal heartbeats. Current methods of analyzing Holter monitor data can be time consuming and error prone, according to Sayed, creating an opportunity for CardiAI to patent and license an AI-based solution. CliniSys used the University of Alberta’s new AI supercomputing hub to process and find patterns in the data.

AMII’s work with these and other organizations is consistent with its focus on “taking machine learning out of academia and applying it to real world problems,” said Cairo. “We’re producing more and more data and there’s so much we can do with it.”

There’s no doubt that AI is having a positive impact on healthcare in Canada, and no better example of that is the availability of AI-powered diagnostic imaging in some of the country’s most remote Indigenous communities. All too often, residents of remote Indigenous communities in need of diagnostic imaging won’t bother making the journey to a distant hospital.

“Many Indigenous people in remote communities end up getting treated with a presumptive diagnosis or no diagnosis at all,” said Dr. Kaura. “As a physician, I’m deeply moved by the capacity to actually help change lives beyond what I’m able to touch with my own hands.”