Futuristic handheld scanners that can instantly diagnose all that ails us are on the verge of becoming a reality, thanks to a University of Alberta researcher whose use of artificial intelligence (AI) to pinpoint a growing host of disorders landed him an unprecedented research chair.
Jacob Jaremko, a pediatric and musculoskeletal radiologist and researcher in the Faculty of Medicine & Dentistry, has become the only practising physician to be awarded a Canadian Institute for Advanced Research (CIFAR) AI Chair. He has been granted this chair at the Alberta Machine Intelligence Institute (Amii) for his application of artificial intelligence to medical problems.
“As a clinician scientist I kind of straddle that divide,” said Jaremko, who completed a PhD in biomedical engineering and developed expertise in the earliest form of AI, before attending medical school at the U of A. “I hold a unique position in that way.”
Jaremko’s team is particularly interested in the use of portable ultrasound machines alongside AI to play a larger role in diagnosing injury and illness.
Using AI to interpret ultrasound scans
In his most recent study, his team looked at whether a quick scan from an ultrasound machine could be as effective as an X-ray in diagnosing broken arms in children visiting an emergency room.
The current process of waiting to see a physician, going for X-rays and getting a diagnosis takes several hours. As well, roughly half of the time there is no fracture.
Ultrasound can be instantaneous, but the black-and-white snowstorm images can be quite confusing for the uninitiated to interpret, noted Jaremko, who holds the Alberta Health Services Endowed Chair in Diagnostic Imaging at the U of A and is a member of the Women and Children's Health Research Institute.
“However, an artificial intelligence network can recognize patterns in the ultrasound, much the same way that your iPhone can recognize your face and unlock itself.”
For the study, a cohort of ultrasound experts reviewed both ultrasound and X-ray images of bone breaks. They found that, with the right expertise, ultrasounds were as good as X-rays in identifying fractures.
Those same images were then fed into a neural network that researchers in Jaremko’s lab trained to search for fractures. The study showed the neural network accuracy was the same as the human expert accuracy.
“In the future, a child with a sore arm comes in and instead of sending them for this several-hours-long procedure, (the clinician) scans the patient for a few seconds, looks at the pictures and gets an opinion from the computer as to whether a fracture is detected,” he said.
“In at least half of the kids there is no fracture, so you're done—they go home and an emergency room bed opens up. Similarly, you could expand that out to walk-in clinics, ambulances and remote communities.”
A versatile “21st-century stethoscope”
Jaremko said this “21st-century stethoscope” doesn’t just stop at fractures, but can also be used to look into the body and detect many kinds of pathology.
In 2017, Jaremko, former U of A post-doctoral fellow Dornoosh Zonoobi and Jeevesh Kapur, a radiologist from Singapore, with support from Alberta Innovates, formed U of A spinoff MEDO.ai, a startup based on the artificial intelligence analysis of images.
MEDO is also commercializing a prenatal diagnosis of hip dysplasia, which occurs when the ball and socket joint is poorly formed.
Though this affliction is easy to correct at an earliest stages of life, it often goes undiagnosed, which dooms those who live with it to a life of pain, suffering and limited mobility.
Using AI, MEDO trained a neural network using ultrasound images of hip dysplasias to reliably detect the disorder. It is currently being piloted in some clinics in the Edmonton region, as well as the U.K. and Brazil.
Another technology will help physicians identify types of thyroid nodules.
“Here's something you could do in the clinic, scan more patients and reduce wait times,” he said.
“Especially in followup, if you're always doing the scan with the same computer system, then you get a very consistent approach and you can say whether the nodules changed. The computer network really supports thyroid ultrasound and makes it a lot more robust and a lot easier to use.”
Three new CIFAR AI Chairs
Joining Jaremko in this latest round of U of A CIFAR AI Chair recipients at Amii are Faculty of Science adjunct professors Neil Burch, who was part of the U of A teams that solved checkers in 2007 and heads-up limit Texas hold-’em poker in 2016, and Marlos Machado, also a U of A graduate and Deepmind researcher, who worked on Atari learning with Michael Bowling, a machine learning expert in the Department of Computing Science. All three will join Amii as fellows in addition to their appointment as Canada CIFAR AI Chairs.
“Canada is a wellspring of AI talent, and we have an enormous opportunity for global leadership with our pan-Canadian strategy,” said Bill Flanagan, U of A president. “AI researchers at the U of A are revolutionizing sectors like health, agriculture and transportation, and working together with industry to drive growth in Alberta and around the world.”
CIFAR is a Canadian-based global research organization that brings together teams of top researchers from around the world to address complex questions and identify new areas of scholarship where Canada has the potential to lead.
Amii CEO Cam Linke said these newest Canada CIFAR AI Chairs will further strengthen our ties with one another while also advancing Canada's AI excellence.
“Together, and with long-term provincial support, we've established Alberta as a global leader in machine learning that attracts and trains the world's brightest minds and drives economic growth for the province,” said Linke. “We look forward to continuing our work together and leveraging scientific advancement to accelerate industry adoption of AI."
CIFAR works in close collaboration with Canada’s three national AI Institutes—Mila in Montreal, the Vector Institute in Toronto and the Edmonton-based Amii, a non-profit institute that supports world-leading research in AI and machine learning and translates scientific advancement into industry adoption.