Abstract
Instrumented head acceleration measurement devices are commonly used in research studies to determine head acceleration exposure in certain populations. Instrumented mouthguards pair directly to the user's teeth and offer six-degree-of-freedom measurements. Though many studies have recently used these devices, post-processing techniques vary by study. Other studies have attempted to label impact quality or coupling status, also with varying methods. This study sought to compare the effect of post-processing and labeling methods on reported exposure distribution characteristics in instrumented mouthguard data from ice hockey players. We collected data from 18 female adolescent ice hockey players on two teams for an entire season. We then post-processed the measured signals using five different techniques: (1) the instrumented mouthguard manufacturer's data output, (2) a 500 Hz linear acceleration filter and a 300 Hz angular velocity filter, (3) HEADSport, (4) a 100 Hz linear acceleration filter and a 175 Hz angular velocity filter, and (5) a salvaging process to detect and remove decoupling based on signal frequency content. The post-processing techniques affected the reported exposure distributions by changing the mean, median, and 95th percentile values of peak linear and angular kinematics. We also compared labeling techniques by measuring agreement and inter-rater reliability between three labeling techniques: the instrumented mouthguard manufacturer's label, Luke et al.'s coupling label, and our classification learner that detects and labels decoupling. We found that the labeling techniques had low agreement about which acceleration events were the best to keep. Labeling technique also influenced the reported distributions' descriptive statistics. Post-processing and event labeling are crucial components of head acceleration event exposure studies. Methods should be described by researchers, and standardization should be sought to allow for better cross-study comparison. Published and publicly available techniques can help move the field toward this ideal. Researchers should be aware of the potential effect post-processing can have on a population's final reported exposure metrics.