Deep Learning Model for Predicting Head Kinematics from Crash Videos

Head kinematics information is very valuable as it is used to measure brain injury risk. Currently, head kinematics are measured using wearable devices or instrumentation mounted on the head. These instrumentation and wearable devices can have errors due to faulty sensors and due to relative motion between the wearable device and the respective body region. This paper proposes a novel method to predict the head kinematics directly from videos without any instrumentation using a deep learning approach. To prove the concept, a deep learning model was developed for predicting time history of head angular velocities and their respective peaks using Finite Element (FE) based crash simulation data. This FE dataset was split into training, validation, and test datasets. A combined Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) based deep learning model was developed using the training and validations sets. The test (unseen) dataset was used to evaluate the predictive capability of the deep learning model. On the test dataset, correlation coefficient obtained between the actual and predicted peak angular velocities was 0.73, 0.85, and 0.92 for X, Y, and Z components respectively.

Sex, Age and Stature Affects Neck Biomechanical Responses in Frontal and Rear Impacts Assessed Using Finite Element Head and Neck Models

The increased incidence of injury demonstrated in epidemiological data for the elderly population, and females compared to males, has not been fully understood in the context of the biomechanical response to impact. A contributing factor to these differences in injury risk could be the variation in geometry between young and aged persons and between males and females. In this study, a new methodology, coupling a CAD and a repositioning software, was developed to reposture an existing Finite element neck while retaining a high level of mesh quality. A 5th percentile female aged neck model (F0575YO) and a 50th percentile male aged neck model (M5075YO) were developed from existing young (F0526YO and M5026YO) neck models (Global Human Body Models Consortium v5.1). The aged neck models included an increased cervical lordosis and an increase in the facet joint angles, as reported in the literature. The young and the aged models were simulated in frontal (2, 8, and 15 g) and rear (3, 7, and 10 g) impacts. The responses were compared using head and relative facet joint kinematics, and nominal intervertebral disc shear strain. In general, the aged models predicted higher tissue deformations, although the head kinematics were similar for all models. In the frontal impact, only the M5075YO model predicted hard tissue failure, attributed to the combined effect of the more anteriorly located head with age, when compared to the M5026YO, and greater neck length relative to the female models. In the rear impacts, the F0575YO model predicted higher relative facet joint shear compared to the F0526YO, and higher relative facet joint rotation and nominal intervertebral disc strain compared to the M5075YO. When comparing the male models, the relative facet joint kinematics predicted by the M5026YO and M5075YO were similar. The contrast in response between the male and female models in the rear impacts was attributed to the higher lordosis and facet angle in females compared to males. Epidemiological data reported that females were more likely to sustain Whiplash Associated Disorders in rear impacts compared to males, and that injury risk increases with age, in agreement with the findings in the present study. This study demonstrated that, although the increased lordosis and facet angle did not affect the head kinematics, changes at the tissue level were considerable (e.g., 26% higher relative facet shear in the female neck compared to the male, for rear impact) and relatable to the epidemiology. Future work will investigate tissue damage and failure through the incorporation of aged material properties and muscle activation.

Reproducibility and Characterization of Head Kinematics During a Large Animal Acceleration Model of Traumatic Brain Injury

Acceleration parameters have been utilized for the last six decades to investigate pathology in both human and animal models of traumatic brain injury (TBI), design safety equipment, and develop injury thresholds. Previous large animal models have quantified acceleration from impulsive loading forces (i.e., machine/object kinematics) rather than directly measuring head kinematics. No study has evaluated the reproducibility of head kinematics in large animal models. Nine (five males) sexually mature Yucatan swine were exposed to head rotation at a targeted peak angular velocity of 250 rad/s in the coronal plane. The results indicated that the measured peak angular velocity of the skull was 51% of the impulsive load, was experienced over 91% longer duration, and was multi- rather than uni-planar. These findings were replicated in a second experiment with a smaller cohort (N = 4). The reproducibility of skull kinematics data was mostly within acceptable ranges based on published industry standards, although the coefficients of variation (8.9% for peak angular velocity or 12.3% for duration) were higher than the impulsive loading parameters produced by the machine (1.1 vs. 2.5%, respectively). Immunohistochemical markers of diffuse axonal injury and blood–brain barrier breach were not associated with variation in either skull or machine kinematics, suggesting that the observed levels of variance in skull kinematics may not be biologically meaningful with the current sample sizes. The findings highlight the reproducibility of a large animal acceleration model of TBI and the importance of direct measurements of skull kinematics to determine the magnitude of angular velocity, refine injury criteria, and determine critical thresholds.

Postural and Head Control Given Different Environmental Contexts

Virtual reality allows for testing of multisensory integration for balance using portable Head Mounted Displays (HMDs). HMDs provide head kinematics data while showing a moving scene when participants are not. Are HMDs useful to investigate postural control? We used an HMD to investigate postural sway and head kinematics changes in response to auditory and visual perturbations and whether this response varies by context. We tested 25 healthy adults, and a small sample of people with diverse monaural hearing (n = 7), or unilateral vestibular dysfunction (n = 7). Participants stood naturally on a stable force-plate and looked at 2 environments via the Oculus Rift (abstract “stars;” busy “street”) with 3 visual and auditory levels (static, “low,” “high”). We quantified medio-lateral (ML) and anterior-posterior (AP) postural sway path from the center-of-pressure data and ML, AP, pitch, yaw and roll head path from the headset. We found no difference between the different combinations of “low” and “high” visuals and sounds. We then combined all perturbations data into “dynamic” and compared it to the static level. The increase in path between “static” and “dynamic” was significantly larger in the city environment for: Postural sway ML, Head ML, AP, pitch and roll. The majority of the vestibular group moved more than controls, particularly around the head, when the scenes, especially the city, were dynamic. Several patients with monaural hearing performed similar to controls whereas others, particularly older participants, performed worse. In conclusion, responses to sensory perturbations are magnified around the head. Significant differences in performance between environments support the importance of context in sensory integration. Future studies should further investigate the sensitivity of head kinematics to diagnose vestibular disorders and the implications of aging with hearing loss to postural control. Balance assessment and rehabilitation should be conducted in different environmental contexts.

Evaluation of the Kinematic Biofidelity and Inter-Test Repeatability of Global Accelerations and Brain Parenchyma Pressure for a Head-Brain Physical Model

Head surrogates are used in biomechanical research and headgear assessment. They are designed to approximate the properties of the head and are instrumented to measure global head kinematics. Due to the recent interest in studying disruption to the brain, some head models include internal fluid layers and brain tissue, and instrumentation to measure head intracranial biomechanics. However, it is unknown whether such models exhibit realistic human responses. Therefore, this study aims to assess the biofidelity and repeatability of a head model, the Blast Injury Protection Evaluation Device (BIPED), that can measure both global head kinematics and intraparenchymal pressure (IPP) for application in blunt impact, a common loading scenario in civilian life. Drop tests were conducted with the BIPED and the widely used Hybrid III headform. BIPED measures were compared to the Hybrid III data and published cadaveric data, and the biofidelity level of the global linear acceleration was quantified using CORrelation and Analysis (CORA) ratings. The repeatability of the acceleration and IPP measurements in multiple impact scenarios was evaluated via the coefficient of variation (COV) of the magnitudes and pulse durations. BIPED acceleration peaks were generally not significantly different from cadaver and Hybrid III data. The CORA ratings for the BIPED and Hybrid III accelerations ranged from 0.50 to 0.61 and 0.51 to 0.77, respectively. The COVs of acceleration and IPP were generally below 10%. This study is an important step toward a biofidelic head surrogate measuring both global kinematics and IPP in blunt impact.