PaperlayerView exact revision on arXiv

Source-linked AI summary

Validation and Comparison of Instrumented Mouthguards for Measuring Head Kinematics and Assessing Brain Deformation in Football Impacts

Yuzhe Liu, August G. Domel, Seyed Abdolmajid Yousefsani, Jovana Kondic, Gerald Grant, Michael Zeineh, David B. Camarillo

arXiv:2008.01903v2physics.bio-pheess.SP

TL;DR

Conflicting reports of instrumented mouthguard accuracy make systematic comparison important for interpreting head-impact measurements and derived brain measures. This study tested five devices against ATD reference sensors under controlled impacts; derived brain-strain and injury-criteria errors were no greater than or close to peak-measurement error, supporting further analysis with the CNN-based model.

  • Problem

    Prior evaluations reported conflicting mouthguard accuracy, while device differences and inconsistent mandible treatment complicated comparison under a common protocol.

    Source: §1 · p. 2

  • Method

    Five instrumented mouthguards were tested on a Hybrid III headform using a pneumatic impactor, with their kinematics compared against reference ATD sensors.

    Source: §1 · p. 3, §2 · p. 3

  • Results

    Brain-strain and brain-injury-criteria relative errors were lower than or close to peak-measurement relative error, and angular-velocity trace correlation exceeded 0.92.

    Source: §4 · pp. 11–12

  • Takeaways & Limitations

    The tested mouthguards’ accuracy was judged adequate for further analysis of kinematic measurements with the CNN-based brain model.

    Source: §4 · p. 12

    The study did not consider mandible strikes, instead constraining the mandible to prevent direct strikes against the mouthguard.

    Source: §4 · pp. 15–16

Abstract

from arXiv · show

Because of the relatively rigid coupling between the upper dentition and the skull, instrumented mouthguards have been shown to be a viable way of measuring head impact kinematics for assisting in understanding the underlying biomechanics of concussions. This has led various companies and institutions to further develop instrumented mouthguards. However, their use as a research tool for understanding concussive impacts makes quantification of their accuracy critical, especially given the conflicting results from various recent studies. Here we present a study that uses a pneumatic impactor to deliver impacts characteristic to football to a Hybrid III headform, in order to validate and compare five of the most commonly used instrumented mouthguards. We found that all tested mouthguards gave accurate measurements for the peak angular acceleration (mean relative error, MRE < 13%), the peak angular velocity (MRE < 8%), brain injury criteria values (MRE < 13%) and brain deformation (described as maximum principal strain and fiber strain, calculated by a convolutional neural network based brain model, MRE < 9%). Finally, we found that the accuracy of the measurement varies with the impact locations yet is not sensitive to the impact velocity for the most part.

1. Introduction

The paper addresses conflicting reports about instrumented mouthguard accuracy and the need to compare devices under a common protocol. It evaluates five commonly available mouthguards against reference sensors, including how measurement differences affect brain-strain and injury-criteria calculations.

1. Introduction

  • Prior mouthguard studies reported accuracy ranging from excellent to poor, with inconsistent mandible treatment and differences among devices proposed as explanations (1, 3, 30).The devices also differ in sensors, fit, and other design parameters.Source: §1 · p. 2
  • Systematic assessment is needed to understand how device variations affect kinematics, brain-deformation and injury-criteria calculations, and interpretation of measurements.The paper links this need to interpreting data that could guide athlete removal and return to play in the future.Source: §1 · pp. 2–3
  • The study compares five mouthguards using football-characteristic impacts on a Hybrid III headform and reference ATD sensors, assessing kinematics and derived brain strain and injury criteria.It reports linear acceleration, angular velocity, and angular acceleration comparisons.Source: §1 · p. 3

2. Materials and Methods

The methods compare mouthguard measurements with reference ATD sensors under controlled football-characteristic impacts. Testing covers five devices, multiple impact locations and velocities, and metrics spanning kinematics, brain deformation, and injury criteria.

2. Materials and Methods

  • The study assumes rigid coupling between upper dentition and skull, and compares five mouthguards’ measurements with high-accuracy ATD sensors used as the reference.The sensors measured linear acceleration and angular velocity at the ATD center of gravity.Source: §2 · pp. 3–4

2.1. Laboratory Setup

  • The setup constrained the movable mandible to prevent strikes against the mouthguard and used biofidelic dentition fitted to each device.The protocol also modified the ATD’s upper lip to reduce vibration transferred to the mouthguard.Source: §2.1 · p. 4

2.2. Mouthguards

  • Five devices from three developers included customized and boil-and-bite models, tested as MiG-C, MiG-B, PRE-C, PRE-B, and SWA-C.All were fitted to the study’s representative dentition.Source: §2.2 · p. 5

2.3. Testing Protocol

  • Testing used five impact locations and four target velocities, with three repetitions per condition for 54 total impacts; facemask impacts used only the two lower velocities.The velocity levels were 3.6, 5.5, 7.4, and 9.3 m/s.Source: §2.3 · pp. 5–6

2.4. Data Processing and Analysis

  • Mouthguard data were aligned to ATD data and transformed to its coordinate system before performance analysis.MG and ATD denote brain-strain or brain-injury-criteria values calculated from mouthguard and ATD kinematics, respectively.Source: §2.4 · pp. 6–7
  • Peak-magnitude relative error, trace correlation, and instantaneous-axis error assessed kinematic peaks, traces, and directions.Trace correlation reflects magnitude-trace variation; instantaneous-axis error assesses direction at peak magnitude.Source: §2.4 · p. 6
  • Relative errors in brain strain and injury criteria assessed how kinematic measurement differences carried into derived measures.Brain-strain errors covered 95% maximum principal strain and 95% fiber strain measures; criteria included BAM, BrIC, and PCS.Source: §2.4 · pp. 6–7
  • The analysis used regression, Kruskal-Wallis tests, pairwise p-values, and violin plots to examine device differences and effects of impact velocity and location.Violin plots displayed density, a box-and-whisker summary, mean, and median.Source: §2.4 · p. 7

3. Results

Across the reported comparisons, mouthguards generally measured peak angular kinematics and downstream brain-strain and injury-criteria values with relatively low errors, though PRE mouthguards were excluded from strain analysis. Accuracy also varied by impact location, while most measures were not highly related to impact velocity, with device- and metric-specific exceptions.

3.1. Typical Kinematic Traces

  • Peak magnitudes and whole kinematic traces generally matched the ATD reference, but PRE missed part of angular-velocity deceleration and longer-window MiG and SWA captured a later back-impact peak.Differentiation amplifies noise in angular acceleration; translating mouthguard acceleration to the headform CoG also amplifies noise.Source: §3.1 · pp. 7–8

3.3.1. Relative Error in Peak of Magnitude (REPM)

  • Mean REPM was below 13% for angular acceleration and below 8% for angular velocity; linear acceleration error was below 14% for MiG, below 4% for PRE, and 32.4% for SWA-C.SWA-C had the lowest angular-velocity error, MiG-C and PRE had the lowest angular-acceleration error, and PRE had the lowest linear-acceleration error at CoG.Source: §3.3.1 · pp. 8–9

3.3.2. Correlation Coefficients of Magnitude (CCM)

  • Mean CCM exceeded 0.9 for MiG and 0.95 for PRE; SWA-C exceeded 0.9 for angular velocity, while its linear and angular acceleration CCMs ranged from 0.8 to 0.9.CCM values were similar within developers but statistically significantly different across developers.Source: §3.3.2 · p. 9

3.3.3. Instantaneous Axis Error (IAE)

  • Mean instantaneous-axis errors for angular velocity, angular acceleration, and linear acceleration were below or around 10°, with MiG-C performing best overall.Errors were lower for angular velocity than for angular acceleration and linear acceleration.Source: §3.3.3 · p. 9

3.3.4. Relative Error in Brain Strain (REBS)

  • Mean and median REBS were below 10% for MiG and SWA across the three reported strain measures; PRE was excluded because its short sampling window makes the CNN overestimate strain.The measures were 95% MPS in the whole brain, 95% MPS at the corpus callosum, and 95% fiber strain at the corpus callosum.Source: §3.3.4 · p. 9

3.3.5. Relative Error in Brain Injury Criteria (REBIC)

  • Mean and median relative errors for BAM, BrIC, and PCS were below 13% for all mouthguards; MiG-C and PRE-C had statistically identical, best overall REBIC performance.BrIC had the lowest error, consistent with its angular-velocity inputs being the most accurate kinematic measurements; BAM and PCS use inputs with higher errors.Source: §3.3.5 · p. 10

3.4. Assessment at Different Impact Locations

  • Accuracy depended on impact location, especially for boil-and-bite mouthguards; most PRE-C metrics were not highly location-related, while most PRE-B metrics were.For MiG, most metrics were highly related to location; REBIC depended less on location for MiG-C than MiG-B, and SWA-C showed less location dependence than MiG-B and PRE-B.Source: §3.4 · p. 10

3.5. Assessment at Different Impact Velocities

  • Most MiG and SWA metrics were not highly related to impact speed, and most REBS and REBIC measures also showed little velocity dependence; MiG-C linear-acceleration CCM and several PRE metrics were exceptions.The velocity analysis used four speeds per location except facemask impacts, which were excluded to avoid bias because only two lower speeds were tested there.Source: §3.5 · pp. 10–11

4. Discussion

Across the tested mouthguards, downstream brain-strain and injury-criterion calculations did not amplify kinematic measurement errors, while accuracy varied by kinematic type and impact location. Device-specific results identify trade-offs in measurement range, sampling window, and fit; the assessment also has laboratory and mandible-strike boundaries.

4. Discussion

  • Angular velocity was more accurate than angular acceleration and CoG linear acceleration because differentiation amplifies noise, which then propagates into the linear-acceleration estimate.The discussion notes that angular velocity may be more relevant to mTBI prediction and correlates with MPS in cited studies.Source: §4 · p. 11
  • Angular-velocity traces were accurate across mouthguards (CCM > 0.92), while mean angular-velocity IAE ranged from 3° to 10° and may matter for regional deformation.The authors recommend investigating its influence before correlating pathology with regions of high MPS.Source: §4 · pp. 11–12
  • REBS and REBIC errors were lower than or close to REPM, indicating the CNN-based brain model and injury-criteria calculation did not further amplify kinematic errors.The authors judged all tested mouthguards adequate for further analysis with the CNN-based model.Source: §4 · p. 12
  • Facemask impacts were generally less accurate, and most accuracy metrics showed no significant velocity dependence; impact location also altered acceleration-error propagation and customized guards showed fewer location-dependent metrics.Relative motion between the mouthguard and upper dentition can add noise; a more sophisticated customized fit may explain the location-dependence difference.Source: §4 · pp. 12–13
  • Most performance metrics did not differ significantly between customized and boil-and-bite guards, so the latter may suit studies needing quicker fitting.Customized guards generally appeared more accurate and had fewer metrics dependent on impact location, possibly because of their tighter fit.Source: §4 · p. 13
  • MiG guards had higher REPM but similar CCM and IAE to PRE and SWA, while their wide sensor ranges make them suited to severe impacts, including some above 35 rad/s.Side impacts were most accurate for MiG; facemask impacts and MiG-B front impacts had relatively larger errors.Source: §4 · pp. 13–14
  • PRE tracked CoG linear acceleration well, whereas SWA overestimated it despite accurate rotational measurements.PRE-C gyroscope saturation above 35 rad/s in one direction and short sampling windows that missed angular-velocity deceleration may affect strain estimates; SWA’s molar sensor position may increase susceptibility to mandible strikes.Source: §4 · pp. 14–15
  • The assessment fixed the mandible and omitted ground-head impacts; laboratory REBS may differ on field, and only five guard types from three developers were tested.Prior work reported higher mouthguard readings than ATD data when an unconstrained mandible could strike the guard.Source: §4 · pp. 11, 16

Figures and Tables for Manuscript Figures:

The figures document the impact setup and compare mouthguard measurements with the ATD across head kinematics, brain strain, and brain injury criteria. The tables identify abbreviations and mouthguard and reference-ATD specifications.

Figures and Tables for Manuscript Figures:

  • Figure 1 depicts the impactor, supporting table, helmeted ATD, titanium dentition, and facemask, front, oblique, side, and back impact setups.Table 2 lists parameters and specifications for the MiG, PRE, and SWA mouthguards and the reference ATD; Table 1 lists abbreviations.Source: Figure 1 · p. 20, Table 1 · p. 28, Table 2 · p. 29
  • Figure 3 compares mouthguard and ATD peak angular acceleration, angular velocity, and linear acceleration at CoG across five mouthguards.Rows identify the mouthguards, columns identify kinematic components, and plots show regression and identity lines with corresponding R-squared values.Source: Figure 3 · p. 22
  • Figures 4–6 present mouthguard comparisons for peak-magnitude error, trace correlation, and instantaneous axis error across the three kinematic components.Their panels include pairwise Kruskal-Wallis test p-values and each mouthguard’s mean and median for the relevant measure.Source: Figure 4 · p. 23, Figure 5 · p. 24, Figure 6 · p. 25
  • Figure 7 compares relative error for whole-brain and corpus-callosum strain measures, while excluding PRE mouthguards because their time windows are too short.The strain measures are 95% maximum principal strain, 95% maximum principal strain at the corpus callosum, and 95% fiber strain at the corpus callosum.Source: Figure 7 · p. 26
  • Figure 8 compares relative error for BAM, BrIC, and PCS across mouthguards, with pairwise test p-values and mean and median errors shown.The caption identifies the three brain injury criteria but gives no numerical error values.Source: Figure 8 · p. 27

Output time

The supplementary material lists output timing and coordinate information, then describes comparisons of accuracy across impact directions and velocities. Some listed timing values are shown before and after alignment processing.

Output time

  • The output-time entries include [-49,150] ms, [0,50] ms, [1,103] ms, and [-200,800] ms; post-alignment entries include [-48,151] ms, [-10,40] ms, [-1,101] ms, and [-200,800] ms.The passages provide these sequences without identifying which value belongs to each device.Source: Output time · p. 29, Time windows after alignment processing · p. 29

Output; coordinate axes direction

Time windows after alignment processing

  • Table 3 compares accuracy metrics across impact directions; PRE-mouthguard REBS is omitted because the time windows are too short for the CNN model.The compared measures are REPM, CCM, IAE, REBS, and REBIC.Source: Table 3 · p. 30
  • Table 4 compares accuracy metrics across impact velocities, excluding facemask impacts and PRE-mouthguard REBS for stated testing and time-window constraints.Facemask impacts are excluded because the two higher velocities were not performed; PRE REBS is omitted because the time windows are too short for the CNN model.Source: Table 4 · p. 31

S1. Instrumented Mouthguard Information

The three developers’ mouthguards differ in sensor placement, data-collection triggers, filtering, and angular-acceleration processing. The two models from each of the MiG and PRE lines share a PCB.

S1. Instrumented Mouthguard Information

  • MiG-B and MiG-C share a PCB with triaxial accelerometer and gyroscope sensors positioned in front of the incisors.They trigger at 10 g, filter raw data at 160 Hz, downsample angular velocity to the linear-acceleration sequence, and derive angular acceleration using a five-point stencil.Source: S1. Instrumented Mouthguard Information · p. 32
  • PRE-B and PRE-C share a PCB with triaxial accelerometer and gyroscope sensors near the first left lateral incisor.They trigger at 5 g and filter data at 200 Hz.Source: S1. Instrumented Mouthguard Information · p. 32
  • SWA-C places its accelerometer and gyroscope sensors near the left molar and applies no filter to raw data.It triggers at 10 g and derives angular acceleration from angular velocity using a five-point stencil; it was initially developed for rugby.Source: S1. Instrumented Mouthguard Information · p. 32

S2. Data Processing and Analysis S2.1. Reference Data

The data-processing passages describe reference-sensor filtering and transform mouthguard outputs into the ATD coordinate system for direct comparison.

S2. Data Processing and Analysis S2.1. Reference Data

  • Reference raw data is filtered at 300 Hz with a fourth-order Butterworth filter, following the Hybrid III ATD standard.Reference angular acceleration is calculated by applying a five-point stencil derivative to filtered angular velocity.Source: S2. Data Processing and Analysis S2.1.… · pp. 32–33
  • Mouthguard outputs are transformed to the standard ATD coordinate system, with the origin at the ATD’s CoG.The axes are +X forward, +Y right, and +Z downward with respect to the ATD headform.Source: S2. Data Processing and Analysis S2.1.… · p. 33

S2.2. Kinematic Assessment

The assessment aligns mouthguard kinematics with reference ATD measurements despite differences in trigger times, sampling rates, and recording windows. It evaluates both magnitude accuracy and impact direction.

S2.2. Kinematic Assessment

  • ATD data are interpolated to each mouthguard’s time resolution, aligned by maximizing pairwise magnitude correlation, and cut to the mouthguard’s recording window.This produces matching time resolution, time window, and trigger time point for the comparison.Source: S2.2. Kinematic Assessment · p. 33
  • Peak relative error (REPM) and magnitude correlation coefficient (CCM) assess the accuracy of mouthguard-measured kinematic magnitudes against ATD references.Source: S2.2. Kinematic Assessment · p. 33
  • Instantaneous axis error (IAE) measures the angle between mouthguard and ATD kinematic vectors at the ATD peak-magnitude time point.Measurements are transformed from sensor coordinates to the global coordinate frame before calculating IAE.Source: S2.2. Kinematic Assessment · p. 33

S2.3. Brain Deformation Assessment

Brain deformation is estimated from mouthguard or ATD angular-velocity data using a validated CNN-based head model, then compared using three strain measures. PRE mouthguards’ short recordings make their strain estimates unsuitable for assessment.

S2.3. Brain Deformation Assessment

  • The validated CNN-based head model estimates 95% maximum principal strain, 95% maximum principal strain at the corpus callosum, and 95% fiber strain at the corpus callosum.The model was trained on an on-field dataset and validated against the WHIM V1.0 model.Source: S2.3. Brain Deformation Assessment · p. 34
  • The model requires angular velocity sampled at 1 ms resolution over 100 ms; PRE recordings are shorter, so constant angular velocity is appended to meet this input window.Source: S2.3. Brain Deformation Assessment · p. 34
  • For PRE-C, appending constant angular velocity yielded REBS of 22.7% for 95% MPS, 18.5% for 95% MPS at CC, and 31.5% for 95% FS at CC.A matched-padding test using ATD data still substantially overestimated strain; PRE mouthguards were therefore excluded from brain-strain analysis.Source: S2.3. Brain Deformation Assessment · p. 34

S2.4. Brain Injury Criteria Assessment

The study assesses mouthguard-derived brain injury criteria by comparing their predicted values with those from reference ATD data. It uses three mTBI-related criteria with different modeling bases.

S2.4. Brain Injury Criteria Assessment

S3. Component traces of angular acceleration, angular velocity and linear acceleration (corresponding to Fig.2)

The supplementary figures show component traces from five mouthguards for a roughly 9.3 m/s side impact. Each presents X, Y, and Z directions plus magnitude.

S3. Component traces of angular acceleration, angular velocity and linear acceleration (corresponding to Fig.2)

  • Figure S1 presents angular-acceleration component traces for MiG-C, MiG-B, PRE-C, PRE-B, and SWA-C.Panels identify X (back to front), Y (left to right), Z (top to bottom), and magnitude.Source: Figure S1 · p. 36
  • Figure S2 presents angular-velocity component traces for the five mouthguards.The traces show X, Y, and Z directions and magnitude.Source: Figure S2 · p. 37
  • Figure S3 presents linear-acceleration-at-CoG component traces for the five mouthguards.The traces show X, Y, and Z directions and magnitude.Source: Figure S3 · p. 38

S4. Transformation of linear acceleration from the sensor to CoG

The supplementary figures show linear-acceleration components at the CoG and mouthguard location, alongside mouthguard performance measures across impact locations and velocities.

S4. Transformation of linear acceleration from the sensor to CoG

Loading 2008.01903v2…