Instrumented Mouthguards: Measurement & TBI Research

• Instrumented mouthguards are custom intraoral devices embedding triaxial accelerometers and gyroscopes to measure six-degree-of-freedom head kinematics during sports impacts.
• They employ rigid dental coupling and high-frequency data sampling to reliably capture peak linear and angular accelerations validated in both laboratory and on-field studies.
• Advanced signal processing—including adaptive filtering and neural-network denoising—enhances accuracy, paving the way for improved concussion risk modeling and personalized interventions.

Instrumented mouthguards (iMGs) are intraoral devices incorporating triaxial inertial measurement units, custom-fitted to the upper dentition, designed to quantify six-degree-of-freedom head kinematics during sports impacts with high fidelity. By leveraging rigid dental coupling and advanced microelectronics, iMGs serve as direct proxies for skull motion, supporting on-field measurement of peak linear and angular accelerations, as well as derived metrics relevant to traumatic brain injury (TBI) research. They represent the current gold standard for wearable head-impact sensing in contact sports and have been validated for both laboratory and field scenarios.

1. Hardware Architecture and Sensor Coupling

Instrumented mouthguards universally employ triaxial MEMS accelerometers and gyroscopes, embedded within dental-grade ethylene-vinyl acetate or similar polymer shells, vacuum-formed or 3D printed to match an athlete's dental scan (Rotundo et al., 20 Feb 2025, Quigley et al., 2023, Wu et al., 2015, Liu et al., 2020). Devices such as Prevent Biometrics iMG and Stanford MiG2.0 achieve skull coupling error in vivo within 0.5 ± 0.2 mm, outperforming skin-patch (3 ± 0.7 mm) and skull-cap sensors (5 ± 3 mm) (Wu et al., 2015). The rigid upper dentition fit ensures minimal artifact from soft tissue or lower-jaw interference. Sampling rates of 1–3.2 kHz (accelerometers) and up to 8 kHz (gyroscopes) are standard to adequately capture impact transients up to several hundred g and tens of krad/s² (Rotundo et al., 20 Feb 2025, Raymond et al., 2021).

Signal triggering is typically initiated by a single-axis threshold crossing (≥8–10 g), storing pre- and post-trigger windows on local memory. Proximity sensing and seat-detection algorithms further guard against off-teeth recordings (Rotundo et al., 20 Feb 2025, Domel et al., 2020). Calibration is performed via factory bias correction, video verification against multi-angle high-speed cameras, and/or inter-laboratory cross-checks. Electronics placement within the dental arch minimizes resonance and extraneous vibrations.

2. Kinematic Measurement and Signal Processing

Upon triggering, iMGs record raw accelerometer (a_x, a_y, a_z) and gyroscope (ω_x, ω_y, ω_z) signals, which are digitally filtered using a fourth-order Butterworth low-pass filter (typical cutoff 100–200 Hz) (Rotundo et al., 20 Feb 2025, Liu et al., 2020, Tierney et al., 2023). Proprietary noise classification may assign stricter filtering to artefactual signals; filtering cutoffs directly modulate signal-to-noise ratio (SNR) and outlier prevalence.

Advanced methodologies use power spectral density (PSD) analysis to adapt filter selection post hoc, restricting the maximum cutoff to the physics-derived credible frequencies observed in laboratory (≤312 Hz) (Tierney et al., 2023). The artefact attenuation method, which adapts the cutoff frequency to the 95th percentile of PSD or the first PSD local minimum, outperforms both fixed Butterworth and CFC180 filtering by yielding kinematic magnitudes consistent with reference headform impacts.

Angular acceleration $\alpha(t) = d\omega(t)/dt$ is extracted by numerical differentiation of filtered angular velocity, followed by vector magnitude computation. Transformation from the mouthguard's local frame to the estimated head center of gravity is achieved via rigid-body kinematic formulas.

Device/System	Sampling Rate (Accel/Gyro)	Filtering (Hz)	Skull Coupling Error
Prevent Biometrics	3.2 kHz / 3.2 kHz	200, 100, 50	0.5 mm
Stanford MiG2.0	1 kHz / 8 kHz	160	0.5 mm
SWA-C	1 kHz / 952 Hz	None	32.4% error (linear @ CoG)
ATD Reference	100 kHz / 100 kHz	300	N/A

3. Accuracy, Validity, and Error Characterization

Laboratory validations against anthropomorphic test device (ATD) headform references consistently show that custom-molded iMGs achieve mean relative errors (MRE) of <13% for peak angular acceleration, <8% for peak angular velocity, <9% for maximum principal strain, and <13% for brain-injury criteria (BrIC, PCS, BAM) (Liu et al., 2020). In vivo, in high-speed stereo video studies, iMGs measured true head kinematics to <30% normalized RMS error, with mounting displacement below 1 mm (Wu et al., 2015).

On-field validation, involving video-confirmed impact time-matching, established sensitivity at 0.89 (TP/[TP+FN]) and positive predictive values between 0.76–0.98 depending on false-positive definitions (Rotundo et al., 20 Feb 2025). Denoising via 1D convolutional neural networks further lowered peak absolute errors and improved the coefficient of determination for kinematic peaks up to 0.99, with substantial reductions in tissue-level strain error (Zhan et al., 2022).

Impact location affects relative error, with facemask and front hits producing the largest deviations due to helmet and shell transmission paths, while side and oblique impacts are most accurately measured (Liu et al., 2020).

4. Real-World Measurement and Head Acceleration Event (HAE) Incidence

iMGs are deployed in NCAA football, rugby, soccer, and laboratory studies to quantify on-field head acceleration event (HAE) incidence. Standard metrics include peak linear acceleration (PLA), peak angular acceleration (PAA), and derived strain/injury values. In NCAA American football, HAE incidence rates per player-match were similar across positions: PLA > 10 g ≈ 11.2 (defense) and 11.3 (offense); PLA > 30 g ≈ 1.6–2.6; PAA > 1.0 krad/s² ≈ 5.5–6.9; PAA > 2.0 krad/s² ≈ 0.9–1.4 (Rotundo et al., 20 Feb 2025). Statistical comparisons revealed no significant difference in overall rates between Offense and Defense.

Hybrid approaches employing deep learning (MiGNet) for impact detection yield sensitivity/accuracy >90% and integrate with open-source platforms (FITBIR) for standardized data archiving (Domel et al., 2020). Such platforms facilitate multi-institutional meta-analysis and harmonize kinematic data, crucial for studying links between impact severity and concussion risk.

Direct and indirect impacts (e.g., torso or shoulder loads) produce kinematic profiles and brain strain indistinguishable from true head contact except for PLA, underscoring the necessity to monitor all forms of impact exposure in TBI research (Raymond et al., 2021).

5. Advanced Signal Post-processing and Denoising

Artefact and noise attenuation are central to improving iMG-derived kinematics. PSD-driven adaptive filtering reliably mitigates impulse artefacts, bringing SNR and kinematic distributions within laboratory reference ranges (Tierney et al., 2023). Deep learning denoising (1D-CNN) trained on ATD reference waveforms further reduces error in kinematic peaks and downstream brain finite element modeling metrics, with reductions in peak error up to 56% and SNR increases of up to +12 dB (Zhan et al., 2022).

Practical workflow recommendations include triple-stage filtering (raw, then differentiated, then frame-transformed signals) and rigorous synchronization/alignment procedures. Denoised signals enhance the accuracy of injury metrics (HIC, BrIC) and tissue-level strain, improving concussion risk modeling and clinical decision-making.

6. Limitations, Design Recommendations, and Future Directions

Key limitations include proprietary signal processing opacity, modest sample sizes, exclusion of impact directionality and duration, and sensitivity to impact location. Mandible-induced artefacts and on-field jaw dynamics remain areas for further characterization; sampling windows ≥100 ms are recommended for accurate strain modeling (Liu et al., 2020, Zhan et al., 2022). Behavioral adaptations under protective gear (e.g., Guardian Caps) have not been systematically quantified (Quigley et al., 2023).

Recommendations include standardizing filter parameters, coordinating multi-device data fusion (e.g., combining mouthguard and helmet IMUs), statistically robust cohort expansion, and unifying reporting via SAE J211 coordinate conventions. Expanding brain-injury modeling to subject-specific finite element geometries (MRI, DTI-derived) and real-time feedback are recognized future trends.

Recent findings indicate no significant reductions in head kinematics or impact incidence with Guardian Cap soft-shell add-ons during practice, confirming that real-world attenuation remains limited and continued hardware and algorithmic innovation is required (Quigley et al., 2023).

7. Implications for TBI and Sports Biomechanics Research

Instrumented mouthguards offer precise measurement of cranial kinematics, supporting mechanistic studies of concussion and mTBI, benchmarking head injury exposure, and validating helmet and intervention efficacy. Open-access benchmarking via platforms such as FITBIR is enabling cross-study comparison and algorithmic reproducibility, advancing collaborative biomechanics research (Domel et al., 2020). Adaptive thresholds, PSD-driven filter selection, and neural-network denoising establish the contemporary iMG pipeline as robust against artefact and suited for large-scale, high-intensity sports environments.

A plausible implication is that future iMGs, incorporating real-time classification and device-agnostic, physics-derived post processing, will enable granular athlete risk profiling and personalized intervention strategies. Continued multidisciplinary research spanning hardware innovation, computational modeling, and standardized field protocols is necessary for realizing the potential of iMGs in the prevention and clinical management of brain injury.