Measuring head impacts in sports can further our understanding of brain injury biomechanics and, hopefully, advance concussion diagnostics and prevention. Although there are many head impact sensors available, skepticism on their utility exists over concerns related to measurement error. Previous studies report mixed reliability in head impact sensor measurements, but there is no uniform approach to assessing accuracy, making comparisons between sensors and studies difficult. The objective of this paper is to introduce a two-phased approach to evaluating head impact sensor accuracy. The first phase consists of in-lab impact testing on a dummy headform at varying impact severities under loading conditions representative of each sensor’s intended use. We quantify in-lab accuracy by calculating the concordance correlation coefficient (CCC) between a sensor’s kinematic measurements and headform reference measurements. For sensors that performed reasonably well in the lab (CCC ≥ 0.80), we completed a second phase of evaluation on-field. Through video validation of impacts measured by sensors on athletes, we classified each sensor measurement as either true-positive and false-positive impact events and computed positive predictive value (PPV) to summarize real-world accuracy. Eight sensors were tested in phase one, but only four sensors were assessed in phase two. Sensor accuracy varied greatly. CCC from phase one ranged from 0.13 to 0.97, with an average value of 0.72. Overall, the four devices that were implemented on-field had PPV that ranged from 16.3 to 91.2%, with an average value of 60.8%. Performance in-lab was not always indicative of the device’s performance on-field. The methods proposed in this paper aim to establish a comprehensive approach to the evaluation of sensors so that users can better interpret data collected from athletes.

测量运动中的头部撞击可以进一步了解脑损伤生物力学，并有望促进脑震荡的诊断和预防。尽管有许多头部撞击传感器可用，但由于担心与测量误差相关，对其效用存在怀疑。先前的研究报告头部撞击传感器测量的可靠性不一，但没有统一的方法来评估准确性，这使得传感器和研究之间的比较变得困难。本文的目的是介绍一种评估头部碰撞传感器精度的两阶段方法。第一阶段包括在代表每个传感器预期用途的负载条件下，在不同冲击强度下对假人头型进行实验室内冲击测试。我们通过计算传感器运动学测量值和头型参考测量值之间的一致性相关系数 (CCC) 来量化实验室内的准确性。对于在实验室中表现良好（CCC ≥ 0.80）的传感器，我们完成了现场评估的第二阶段。通过对运动员测量的传感器影响的视频验证，我们将每个传感器测量值分类为真阳性和假阳性影响事件，并计算出阳性预测值 (PPV) 以总结真实世界的准确性。在第一阶段测试了八个传感器，但在第二阶段仅评估了四个传感器。传感器精度差异很大。第一阶段的 CCC 范围从 0.13 到 0.97，平均值为 0.72。总体而言，现场实施的四款设备的 PPV 范围为 16.3% 至 91.2%，平均值为 60.8%。实验室性能并不总是代表设备在现场的性能。本文提出的方法旨在建立一种评估传感器的综合方法，以便用户可以更好地解释从运动员那里收集的数据。