Lyndia C. Wu

A head impact detection system using SVM classification and proximity sensing in an instrumented mouthguard

Injury from blunt head impacts causes acute neurological deficits and may lead to chronic neurodegeneration. A head impact detection device can serve both as a research tool for studying head injury mechanisms and a clinical tool for real-time trauma screening. The simplest approach is an acceleration thresholding algorithm, which may falsely detect high-acceleration spurious events such as manual manipulation of the device. We designed a head impact detection system that distinguishes head impacts from nonimpacts through two subsystems. First, we use infrared proximity sensing to determine if the mouthguard is worn on the teeth to filter out all off-teeth events. Second, on-teeth, nonimpact events are rejected using a support vector machine classifier trained on frequency domain features of linear acceleration and rotational velocity. The remaining events are classified as head impacts. In a controlled laboratory evaluation, the present system performed substantially better than a 10-g acceleration threshold in head impact detection (98% sensitivity, 99.99% specificity, 99% accuracy, and 99.98% precision, compared to 92% sensitivity, 58% specificity, 65% accuracy, and 37% precision). Once adapted for field deployment by training and validation with field data, this system has the potential to effectively detect head trauma in sports, military service, and other high-risk activities.

Effect of the mandible on mouthguard measurements of head kinematics.

Wearable sensors are becoming increasingly popular for measuring head motions and detecting head impacts. Many sensors are worn on the skin or in headgear and can suffer from motion artifacts introduced by the compliance of soft tissue or decoupling of headgear from the skull. The instrumented mouthguard is designed to couple directly to the upper dentition, which is made of hard enamel and anchored in a bony socket by stiff ligaments. This gives the mouthguard superior coupling to the skull compared with other systems. However, multiple validation studies have yielded conflicting results with respect to the mouthguard׳s head kinematics measurement accuracy. Here, we demonstrate that imposing different constraints on the mandible (lower jaw) can alter mouthguard kinematic accuracy in dummy headform testing. In addition, post mortem human surrogate tests utilizing the worst-case unconstrained mandible condition yield 40% and 80% normalized root mean square error in angular velocity and angular acceleration respectively. These errors can be modeled using a simple spring-mass system in which the soft mouthguard material near the sensors acts as a spring and the mandible as a mass. However, the mouthguard can be designed to mitigate these disturbances by isolating sensors from mandible loads, improving accuracy to below 15% normalized root mean square error in all kinematic measures. Thus, while current mouthguards would suffer from measurement errors in the worst-case unconstrained mandible condition, future mouthguards should be designed to account for these disturbances and future validation testing should include unconstrained mandibles to ensure proper accuracy.

Bandwidth and sample rate requirements for wearable head impact sensors

Wearable inertial sensors measure human head impact kinematics important to the on-going development and validation of head injury criteria. However, sensor specifications have not been scientifically justified in the context of the anticipated field impact dynamics. The objective of our study is to determine the minimum bandwidth and sample rate required to capture the impact frequency response relevant to injury. We used high-bandwidth head impact data as ground-truth measurements, and investigated the attenuation of various injury criteria at lower bandwidths. Given a 10% attenuation threshold, we determined the minimum bandwidths required to study injury criteria based on skull kinematics and brain deformation in three different model systems: helmeted cadaver (no neck), unhelmeted cadaver (no neck), and helmeted dummy impacts (with neck). We found that higher bandwidths are required for unhelmeted impacts in general and for studying strain rate injury criteria. Minimum gyroscope bandwidths of 300Hz in helmeted sports and 500Hz in unhelmeted sports are necessary to study strain rate based injury criteria. A minimum accelerometer bandwidth of 500Hz in unhelmeted sports is necessary to study most injury criteria. Current devices typically sample at 1000Hz, with gyroscope bandwidths below 200Hz, which are not always sufficient according to these requirements. With hard contact test conditions, the identified requirements may be higher than most soft contacts on the field, but should be satisfied to capture the worst contact, and often higher risk, scenarios relative to the specific sport or activity. Our findings will help establish standard guidelines for sensor choice and design in traumatic brain injury research.

Resonance of human brain under head acceleration

Although safety standards have reduced fatal head trauma due to single severe head impacts, mild trauma from repeated head exposures may carry risks of long-term chronic changes in the brains function and structure. To study the physical sensitivities of the brain to mild head impacts, we developed the first dynamic model of the skull–brain based on in vivo MRI data. We showed that the motion of the brain can be described by a rigid-body with constrained kinematics. We further demonstrated that skull–brain dynamics can be approximated by an under-damped system with a low-frequency resonance at around 15 Hz. Furthermore, from our previous field measurements, we found that head motions in a variety of activities, including contact sports, show a primary frequency of less than 20 Hz. This implies that typical head exposures may drive the brain dangerously close to its mechanical resonance and lead to amplified brain–skull relative motions. Our results suggest a possible cause for mild brain trauma, which could occur due to repetitive low-acceleration head oscillations in a variety of recreational and occupational activities.

Detection of American Football Head Impacts Using Biomechanical Features and Support Vector Machine Classification

Accumulation of head impacts may contribute to acute and long-term brain trauma. Wearable sensors can measure impact exposure, yet current sensors do not have validated impact detection methods for accurate exposure monitoring. Here we demonstrate a head impact detection method that can be implemented on a wearable sensor for detecting field football head impacts. Our method incorporates a support vector machine classifier that uses biomechanical features from the time domain and frequency domain, as well as model predictions of head-neck motions. The classifier was trained and validated using instrumented mouthguard data from collegiate football games and practices, with ground truth data labels established from video review. We found that low frequency power spectral density and wavelet transform features (10~30 Hz) were the best performing features. From forward feature selection, fewer than ten features optimized classifier performance, achieving 87.2% sensitivity and 93.2% precision in cross-validation on the collegiate dataset (n = 387), and over 90% sensitivity and precision on an independent youth dataset (n = 32). Accurate head impact detection is essential for studying and monitoring head impact exposure on the field, and the approach in the current paper may help to improve impact detection performance on wearable sensors.

Propagation of errors from skull kinematic measurements to finite element tissue responses

Real-time quantification of head impacts using wearable sensors is an appealing approach to assess concussion risk. Traditionally, sensors were evaluated for accurately measuring peak resultant skull accelerations and velocities. With growing interest in utilizing model-estimated tissue responses for injury prediction, it is important to evaluate sensor accuracy in estimating tissue response as well. Here, we quantify how sensor kinematic measurement errors can propagate into tissue response errors. Using previous instrumented mouthguard validation datasets, we found that skull kinematic measurement errors in both magnitude and direction lead to errors in tissue response magnitude and distribution. For molar design instrumented mouthguards susceptible to mandible disturbances, 150–400% error in skull kinematic measurements resulted in 100% error in regional peak tissue response. With an improved incisor design mitigating mandible disturbances, errors in skull kinematics were reduced to <50%, and several tissue response errors were reduced to <10%. Applying 30

Comparing In Vivo Head Impact Kinematics From American Football With Laboratory Drop and Linear Impactors

^{\circ }

Comparison of video-based and sensor-based head impact exposure

∘ rotations to reference kinematic signals to emulate sensor transformation errors yielded below 10% error in regional peak tissue response; however, up to 20% error was observed in peak tissue response for individual finite elements. These findings demonstrate that kinematic resultant errors result in regional peak tissue response errors, while kinematic directionality errors result in tissue response distribution errors. This highlights the need to account for both kinematic magnitude and direction errors and accurately determine transformations between sensors and the skull.