Photo by Ben Hershey
Once viewed as a short-term injury, today concussions are understood as a chronic health issue. They can leave individuals with long-lasting effects on the brain’s electrical signals that can persist for years after the initial concussion. Yet doctors, who rely on their own observations and anecdotal evidence of patients, are generally unable to determine the degree of deterioration.
Now, a new paper, “From Group-Level Statistics to Single-Subject Prediction: Machine Learning Detection of Concussion in Retired Athletes,” by a team of McMaster University researchers, including Vector Institute Postgraduate Affiliates Rober Boshra, Kiret Dhindsa, and Omar Boursalie; Vector Faculty Affiliates John Connolly, Jim Reilly, Ranil Sonnadara, Thomas Doyle, and Reza Samavi; and collaborator Kyle Ruiter, offers the potential of identifying these effects even decades after an injury, with the help of machine learning (ML) .
The paper expands on a previous study the group did in collaboration with The Hamilton Spectator. That study demonstrated the persistence of deficits in brain signal responses amongst a group of retired Canadian Football League (CFL) players. Using the same dataset, the team developed an ML algorithm that can detect the effects of concussions in individual players (as opposed to the group as a whole) as long as three decades after the fact. The method’s 81 per cent accuracy rate surpasses all currently available clinical tools and has the potential to help individuals who were misdiagnosed or unaware of the severity of their injury.
Many ML outputs are the product of a “black box,” meaning researchers are unable to explain how the algorithm arrived at its conclusions. However, the study was also able to show not just whether someone had been concussed, but what specific brain responses were the cause of the concussion. Doing so not only helped validate the clinically oriented ML application, but it also identified a previously undocumented sign of concussion in the brain.