Researchers: Christian Poellabauer, Sandra Schneider, Bryan Xia, Louis Daudet, and John Templeton
Concussions are a form of mild Traumatic Brain Injuries (mTBI) that are widespread in contact sports. Individuals with mTBI display a range of somatic, affective, and cognitive symptoms, such as headaches, depression, loss of memory, and brain function. These symptoms are collectively known as Post Concussion Syndrome (PCS). To make treatment more effective and to prevent the risk of further injury, concussions need to be detected as soon as possible. However, diagnosis is difficult as mTBI symptoms could take several hours, or worse, days, to manifest themselves. Existing concussion assessment tests rely on establishing a baseline pre-injury test score against which subsequent results are compared to identify mTBI. In this project, we developed an iPad-based "reading test" that captures the speech responses of youth athletes to assess the likelihood of a concussion. Each recording undergoes a series of speech feature extraction steps, and machine learning techniques are used to evaluate changes of these acoustic features compared to the baseline recording. The tool has so far been used on over 2,500 college and high school athletes (including over 100 concussed individuals), leading to over 10,000 recordings, which are currently studied to identify the optimal feature combination to maximize concussion detection probability.