The application of machine learning models in the concussion diagnosis process

“Concussions represent a growing health concern and are challenging to diagnose and manage. Roughly four million concussions are diagnosed every year in the United States. Although research into the application of advanced metrics such as neuroimages and blood biomarkers has shown promise, they are yet to be implemented at a clinical level due to cost and reliability concerns. Therefore, concussion diagnosis is still reliant on clinical evaluations of symptoms, balance, and neurocognitive status and function. The lack of a universal threshold on these assessments makes the diagnosis process entirely reliant on a physician’s interpretation of these assessment scores. This study aims to show that the implementation of machine learning models can be beneficial to the concussion diagnosis process. While studies on machine learning applications for traumatic brain injuries are gaining traction, previous studies have primarily relied on neuroimaging metrics. The few that used clinical assessment tests have employed only univariate models. This study explores the use of multiple assessment scores in the models and evaluates the importance of each assessment score from the clinical tests. A comprehensive predictive modeling approach was conducted with a number of candidate models and subsampling techniques being evaluated. The findings in this research demonstrate the potential benefits of machine learning models to identify concussed and non-concussed subjects at a 24-48-hour post-injury time point. The results also suggest that not all clinical assessment test scores are of equal importance”--Abstract, page iv