Artificial Brains Helping Human Brains

Machine learning is helping researchers detect past concussions, with the hope of finding a way to diagnose acute concussions.


Most of us will experience some form of cognitive decline as we get older, but how do we know whether it’s just a normal part of ageing or if there’s something more to it? For many retired athletes, answering this question can mean the difference between winning and losing a court case.

Now, a group from Montreal has used machine learning to identify patients with a history of concussions, years after the fact.

Learning about concussions

Although concussion awareness has increased in recent years, there is still no diagnostic test for concussions and no cure except rest. Often, players return to action much too soon, increasing the risk for further brain injury. Some of those players are now demanding compensation for long-term effects caused by concussions sustained in their youth, but it often boils down to their word against someone else’s.

“Machine learning is revolutionizing medicine. Its ability to make predictions based on a very high number of factors largely surpasses that of humans,” says Sébastien Tremblay, post-doctoral fellow in Dr. Julio C. Martinez-Trujillo’s lab at the University of McGill.

So Tremblay and his colleagues from the Montreal Neurological Institute and the Ludmer Center for Neuroinformatics and Mental Health decided to harness machine learning capabilities for concussion detection.

They recruited 30 former university athletes, 15 of whom reported suffering prior concussions, and gathered a range of data, including MRI brain scans, genotyping, and memory tests. This data was then fed into a computer so that it could learn the typical profile of someone who had previously sustained a concussion.

Equipped with this new knowledge, the program correctly identified prior concussions in patients with a 90% success rate. The next step is to validate these results with a larger sample size, to make sure it is an effective diagnostic tool for chronic concussion effects.

Eventually, Tremblay hopes this technology could be used to diagnose acute concussions as well, but this depends on finding the right data to feed into the machine learning algorithms.

“As algorithms get better, we will also see physicians making treatment recommendations based on these predictions. All that pertains to image analysis, like radiology and pathology, will significantly benefit from these tools,” says Tremblay.

A big investment

This study is just one example of the power of machine learning and artificial intelligence, and Canada is a hotbed of activity in this area. The federal and provincial governments recently invested $100 million in Toronto’s Vector Institute for Artificial Intelligence, and Google invested $4.5 million in Montreal’s Institute for Learning Algorithms (MILA) last year.

As Yoshua Bengio, head of MILA, told the Canadian Press “(AI) will affect pretty much every economic sector; right now is just the tip of the iceberg.”

‹ Previous post
Next post ›

Malgosia Pakulska is a freelance science writer, speaker, and blogger. She completed her PhD in Professor Molly Shoichet’s lab studying drug delivery systems for spinal cord regeneration after injury. She is still passionate about research and wants to share that excitement with the public. When she is not in the lab, she is experimenting in the kitchen and blogging about it at Smart Cookie Bakes.