Improving Accurate Detection of Head Impact Exposure through Auxiliary Sensor Input and Post-Processing Algorithms

About This Grant

ABSTRACT Mild traumatic brain injury (mTBI) or “concussive” injuries are a major societal issue and are associated with activities such as sports, motor vehicle crashes, and falls. Sports-related concussions in children and adolescents (5-18 years) account for between 30-60% of all pediatric concussions. There currently exists no available technology to provide accurate measurement of head impacts in concussive or sub-concussive environments. While a number of head impact exposure devices are on the market or are in development, few of these devices have undergone baseline validation studies to assess the accuracy of their results in biofidelic environments. When events are detected using these systems, there is little support as to whether the data is of a relevant impact or spurious in nature when device deployment is unsupervised. Most existing systems rely on simple acceleration thresholds as a trigger to begin data collection. While such methods are easy to implement and interpret, the trade-off is collection of spurious events above the threshold, and loss of data for events below the threshold. We propose utilizing the Data Acquisition System for Head Response (DASHR) to better understand patterns in head kinematic behaviors leading to and characteristic of head impact in youth football as a mechanism by which relevant and spurious wearable sensor data can be distinguished. We propose leveraging existing DASHR head impact exposure data and video data from prior seasons in concert with prospective data acquired from the DASHR in two upcoming seasons alongside newly acquired high-definition video data. This study benefits from extensive leveraging of existing work stemming from an on-going longitudinal study that the investigators and community collaborators are already engaged in, allowing for additional resources to be devoted to high-definition video acquisition coupled with an overall reduction in the total-price point for comparable studies due to the existing and parallel data sources. These data will be utilized to develop and assess post-processing algorithms using machine learning methods to develop an improved data acquisition pipeline for the DASHR and similar systems more accurately distinguishing relevant impacts from non-relevant events. DASHR data and auxiliary sensor input from the prior four years and the first prospective year will be utilized to train (learning dataset) for the algorithm. Year two of the prospective study will be utilized as a validation dataset to assess the effectiveness of the post-processing algorithm on correctly identifying relevant and non-relevant head impact exposure. The proposed study would be the first to design and develop an algorithm capable of effectively identifying relevant impacts from continuous time-series head kinematic data, using pre-impact behavioral patterns and impact characteristics, with video data as ground truth for validation and assessment.