Denoising instrumented mouthguard measurements of head impact kinematics with a convolutional neural network

Zhan, Xianghao; Liu, Yuzhe; Cecchi, Nicholas J.; Callan, Ashlyn A.; Flao, Enora Le; Gevaert, Olivier; Zeineh, Michael; Grant, Gerald A.; Camarillo, David B.

doi:10.48550/arxiv.2212.09832

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…When coupling cannot be improved, algorithms and processing methods show potential. Rooks et al have demonstrated that better algorithms can be developed to improve the identi cation of "bad" recordings [19,25], and models are being proposed to denoise kinematic signals [30]. Improvements can be expected from both manufacturers and research teams in the future.…”

Section: Discussionmentioning

confidence: 99%

Quality issues in kinematic traces from three head impact sensors in boxing: prevalence, effects, and implications for exposure assessment

Flao,

Lenetsky,

Siegmund

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Purpose: When used in-vivo or in biofidelic environments, many head impact sensors have shown limitations related to the quality and validity of the kinematics measured. The objectives were to assess the quality of kinematic traces from three head impact sensors, determine the effects of signal quality on peak accelerations, and compare measurements across sensors. Methods: Head impacts were collected with instrumented mouthguards, skin patches, and headgear patches during boxing sparring. The quality of the raw kinematic traces for 442 events was visually assessed for each sensor. The proportion of high-quality recordings was analyzed by participant, type of impact, and impact location. The associations between signal quality and peak kinematics, and peak kinematics between sensors were assessed. Results: High-quality criteria were met in 53%, 20%, and 26% of events for the mouthguard, skin patch and headgear patch, respectively. High-quality recordings were less frequent for impacts occurring close to the sensor (e.g., 30% vs. 61% for the mouthguard) and showed lower peak kinematics than low-quality recordings (p < 0.001). Despite careful selection of high-quality simultaneous recordings, there was little to no association between the sensors’ measurements (Spearman’s p ≥ 0.043). Conclusions: The kinematic data often reflected the motion of the sensor itself rather than the motion of the head, overestimating head impact exposure. Researchers should evaluate data quality prior to analyzing kinematics or injury severity metrics. Comparison of data across studies or in relation to injury risk functions needs to be done with caution when data were acquired from different sensors.

show abstract

Section: Discussionmentioning

confidence: 99%