Nova Scotia

How machine learning is revolutionizing medical research in Nova Scotia and beyond

Advanced computer programs that use machine learning are transforming the way medical research is done in Nova Scotia and around the world.

"With all this big data … it just surpasses humans' ability to comprehend what is going on'

Ghulam Jalani runs a test on one of the Prospector diagnostic devices in the lab. (Submitted by Alan Fine)

Advanced computer programs that use machine learning are transforming the way medical research is done in Nova Scotia and around the world.

Work that might have taken years to complete, or would have been astronomically expensive, can now be done faster and at lower cost.

It has allowed teams in this province to develop better ways to identify and treat cancer, discover new drugs to help blind children see, and speed up medical tests.  

A computer program learns from data and identifies patterns with little human intervention. A more advanced form of machine learning is often referred to as neural networks. 

For example, a program can be shown millions of pictures of cars and, eventually, it will identify a particular car, says Thomas Trappenberg, a Dalhousie University computer science professor.

Medical researchers have turned that learning power inward, setting up programs to recognize cancer cells and proteins. 

Brendan Leung is an assistant professor in applied oral sciences at Dalhousie University. (Submitted by Brendan Leung)

One group is trying to figure out how to better identify the differences between cancer cells and healthy cells, and, in doing so, what drug treatments will work best for an individual. 

"Target discovery is very important right now," said Brendan Leung, an assistant professor in applied oral sciences at Dalhousie. "So knowing what to hit is just as important as designing the weapon to hit it."

The research team he's part of also includes a tumour biologist and computer scientist. 

Leung hopes the technology will eventually allow scientists to design drugs to better target cancer cells without harming healthy cells.

He said the research would be almost impossible without a computer capable of machine learning.

"With all this big data … it just surpasses humans' ability to comprehend what is going on," he said.

"Not to mention human beings are notoriously biased.… If you've been working with a particular gene for the past 20 years, you know, it's your favourite thing to look at, you will find what you want to see. So the way I see it, it's a great way to take away that bias."     

Leung said software can be biased as well, but perhaps not as biased as a person. 

The research is aimed at helping patients, whether it be with specialized treatments or returning test results faster. (Evan Mitsui/CBC)

Machine learning has already helped develop new drugs that treat a rare hereditary disease that can cause children to go blind. 

The disease is called Familial Exudative Vitreoretinopathy, or FEVR. It prevents the proper amount of blood from reaching the eye.

Depending on severity, it can result in poor vision or blindness, said Christopher McMaster, a Dalhousie professor of pharmacology. 

Christopher McMaster is a professor of pharmacology at Dalhousie University. (Submitted by Christopher McMaster)

McMaster's goal is to turn off a protein that prevents the arteries and veins in the eye from growing properly. The computer uses all available information to create a three-dimensional model of what that protein could look like. 

"Once you have this three-dimensional picture you can then use the AI to say, 'OK, I need to stick a drug-like molecule essentially into the gears of this protein to turn it off. Give me a list of drugs, not known drugs but anything you could synthesize in a lab that we could stick into this spot that could turn it on or off,'" said McMaster. 

The system has worked. McMaster and his team have created a drug that treats FEVR. 

"If you were a mouse with FEVR right now we could restore your vision quite well," he said.      

It will be a year or longer before McMaster files the documents to start human trials. 

Leung says his research is multi-disciplinary, requiring co-operation from experts from many different fields at the university. (CBC)

Doing this work without computers capable of machine learning would have been challenging as thousands, even millions, of drugs would need to be tested in a lab, as opposed to the computer running virtual tests, said McMaster. 

"Diseases like this one that don't affect a lot of children, they'd not have any shot at a therapy whatsoever," he said. "So this has really opened up the avenue for a lot of different diseases that would never see the light of day."

That's not the only success story in the province.

Another professor at Dalhousie has helped develop a device that can quickly perform a blood test without a technician or doctor present.

Alan Fine is a professor with the faculty of medicine in the school of biomedical engineering. He's also founder of the company Alentic Microscience. 

Fine says the Prospector could be set up to do a multitude of other tests besides a complete blood count.  (Submitted by Alan Fine)

Fine developed a device, called the Prospector, that is about the size of a debit machine and can take images of blood cells with a sensor. The machine's neural network has been taught to recognize different parts of the blood and perform a complete blood count. 

That test can tell the number of red blood cells, the number of white cells, platelets and gather other information. 

"It's a sort of snapshot image of the overall health of an individual and it provides clues to many different kinds of illnesses," said Fine. 

In the best-case scenario, the traditional test for a complete blood count would take 20 minutes. More often it can take hours or a day to get results back, said Fine. 

His device takes five minutes and is portable.

It could be years before these new treatments and technologies find their way into local hospitals. (Robert Short/CBC)

Right now, the machine is in its testing form and hasn't yet been approved for diagnostic use by Health Canada or other regulatory agencies. Fine hopes those approvals will come later this year.

"These neural network approaches, they have proven so massively effective," said Fine.

"We were very early beneficiaries of this novel computing technology. It's totally transformed the way that we do this and as I think you can see it's not just our little application, it's spreading throughout medicine."



To encourage thoughtful and respectful conversations, first and last names will appear with each submission to CBC/Radio-Canada's online communities (except in children and youth-oriented communities). Pseudonyms will no longer be permitted.

By submitting a comment, you accept that CBC has the right to reproduce and publish that comment in whole or in part, in any manner CBC chooses. Please note that CBC does not endorse the opinions expressed in comments. Comments on this story are moderated according to our Submission Guidelines. Comments are welcome while open. We reserve the right to close comments at any time.

Become a CBC Member

Join the conversation  Create account

Already have an account?