Deep learning methodology for predicting time history of head angular kinematics from simulated crash videos.

Head kinematics information is important as it is used to measure brain injury risk. Currently, head kinematics are measured using wearable devices or instrumentation mounted on the head. This paper evaluates the deep learning approach in predicting time history of head angular kinematics directly from videos without any instrumentation. To prove the concept, a deep learning model was developed for predicting time history of head angular velocities using finite element (FE) based crash simulation videos. This FE dataset was split into training, validation, and test datasets. A combined convolutional neural network and recurrent neural network 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.

Development of brain injury criteria (BrIC).

Rotational motion of the head as a mechanism for brain injury was proposed back in the 1940s. Since then a multitude of research studies by various institutions were conducted to confirm/reject this hypothesis. Most of the studies were conducted on animals and concluded that rotational kinematics experienced by the animal's head may cause axonal deformations large enough to induce their functional deficit. Other studies utilized physical and mathematical models of human and animal heads to derive brain injury criteria based on deformation/pressure histories computed from their models. This study differs from the previous research in the following ways: first, it uses two different detailed mathematical models of human head (SIMon and GHBMC), each validated against various human brain response datasets; then establishes physical (strain and stress based) injury criteria for various types of brain injury based on scaled animal injury data; and finally, uses Anthropomorphic Test Devices (ATDs) (Hybrid III 50th Male, Hybrid III 5th Female, THOR 50th Male, ES-2re, SID-IIs, WorldSID 50th Male, and WorldSID 5th Female) test data (NCAP, pendulum, and frontal offset tests) to establish a kinematically based brain injury criterion (BrIC) for all ATDs. Similar procedures were applied to college football data where thousands of head impacts were recorded using a six degrees of freedom (6 DOF) instrumented helmet system. Since animal injury data used in derivation of BrIC were predominantly for diffuse axonal injury (DAI) type, which is currently an AIS 4+ injury, cumulative strain damage measure (CSDM) and maximum principal strain (MPS) were used to derive risk curves for AIS 4+ anatomic brain injuries. The AIS 1+, 2+, 3+, and 5+ risk curves for CSDM and MPS were then computed using the ratios between corresponding risk curves for head injury criterion (HIC) at a 50% risk. The risk curves for BrIC were then obtained from CSDM and MPS risk curves using the linear relationship between CSDM - BrIC and MPS - BrIC respectively. AIS 3+, 4+ and 5+ field risk of anatomic brain injuries was also estimated using the National Automotive Sampling System - Crashworthiness Data System (NASS-CDS) database for crash conditions similar to the frontal NCAP and side impact conditions that the ATDs were tested in. This was done to assess the risk curve ratios derived from HIC risk curves. The results of the study indicated that: (1) the two available human head models - SIMon and GHBMC - were found to be highly correlated when CSDMs and max principal strains were compared; (2) BrIC correlates best to both - CSDM and MPS, and rotational velocity (not rotational acceleration) is the mechanism for brain injuries; and (3) the critical values for angular velocity are directionally dependent, and are independent of the ATD used for measuring them. The newly developed brain injury criterion is a complement to the existing HIC, which is based on translational accelerations. Together, the two criteria may be able to capture most brain injuries and skull fractures occurring in automotive or any other impact environment. One of the main limitations for any brain injury criterion, including BrIC, is the lack of human injury data to validate the criteria against, although some approximation for AIS 2+ injury is given based on the angular velocities calculated at 50% probability of concussion in college football players instrumented with 5 DOF helmet system. Despite the limitations, a new kinematic rotational brain injury criterion - BrIC - may offer a way to capture brain injuries in situations when using translational accelerations based HIC alone may not be sufficient.

Association of contact loading in diffuse axonal injuries from motor vehicle crashes.

BACKGROUND: Although studies have been conducted to analyze brain injuries from motor vehicle crashes, the association of head contact has not been fully established. This study examined the association in occupants sustaining diffuse axonal injuries (DAIs). METHODS: The 1997 to 2006 motor vehicle Crash Injury Research Engineering Network database was used. All crash modes and all changes in velocity were included; ejections and rollovers were excluded; injuries to front and rear seat occupants with and without restraint use were considered. DAI were coded in the database using Abbreviated Injury Scale 1990. Loss of consciousness was included and head contact was based on medical- and crash-related data. RESULTS: Sixty-seven occupants with varying ages were coded with DAI. Forty-one adult occupants (mean, 33 years of age, 171-cm tall, 71-kg weight; 30 drivers, 11 passengers) were analyzed. Mean change in velocity was 41.2 km/h and Glasgow Coma Scale score was 4. There were 33 lateral, 6 frontal, and 2 rear crashes with 32 survivors and 9 were fatalities. Two occupants in the same crash did not sustain DAI. Although skull fractures and scalp injuries occurred in some impacts, head contact was identified in all frontal, rear, and far side, and all but one nearside crashes. CONCLUSIONS: Using a large sample size of occupants sustaining DAI in 1991 to 2006 model year vehicles, DAI occurred more frequently in side than frontal crashes, is most commonly associated with impact load transfer, and is not always accompanied by skull fractures. The association of head contact in >95% of cases underscores the importance of evaluating crash-related variables and medical information for trauma analysis. It would be prudent to include contact loading in addition to angular kinematics in the analysis and characterization of DAI.

Impact response of restrained PMHS in frontal sled tests: skeletal deformation patterns under seat belt loading.

This study evaluated the response of restrained post-mortem human subjects (PMHS) in 40 km/h frontal sled tests. Eight male PMHS were restrained on a rigid planar seat by a custom 3-point shoulder and lap belt. A video motion tracking system measured three-dimensional trajectories of multiple skeletal sites on the torso allowing quantification of ribcage deformation. Anterior and superior displacement of the lower ribcage may have contributed to sternal fractures occurring early in the event, at displacement levels below those typically considered injurious, suggesting that fracture risk is not fully described by traditional definitions of chest deformation. The methodology presented here produced novel kinematic data that will be useful in developing biofidelic human models. Additional analysis of the data produced by the reported tests as well as additional tests with a variety of loading conditions are required to fully characterize torso response including ribcage fracture tolerance.

Investigation of traumatic brain injuries using the next generation of simulated injury monitor (SIMon) finite element head model.

The objective of this study was to investigate potential for traumatic brain injuries (TBI) using a newly developed, geometrically detailed, finite element head model (FEHM) within the concept of a simulated injury monitor (SIMon). The new FEHM is comprised of several parts: cerebrum, cerebellum, falx, tentorium, combined pia-arachnoid complex (PAC) with cerebro-spinal fluid (CSF), ventricles, brainstem, and parasagittal blood vessels. The model's topology was derived from human computer tomography (CT) scans and then uniformly scaled such that the mass of the brain represents the mass of a 50th percentile male's brain (1.5 kg) with the total head mass of 4.5 kg. The topology of the model was then compared to the preliminary data on the average topology derived from Procrustes shape analysis of 59 individuals. Material properties of the various parts were assigned based on the latest experimental data. After rigorous validation of the model using neutral density targets (NDT) and pressure data, the stability of FEHM was tested by loading it simultaneously with translational (up to 400 g) combined with rotational (up to 24,000 rad/s2) acceleration pulses in both sagittal and coronal planes. Injury criteria were established in the manner shown in Takhounts et al. (2003a). After thorough validation and injury criteria establishment (cumulative strain damage measure--CSDM for diffuse axonal injuries (DAI), relative motion damage measure--RMDM for acute subdural hematoma (ASDH), and dilatational damage measure--DDM for contusions and focal lesions), the model was used in investigation of mild TBI cases in living humans based on a set of head impact data taken from American football players at the collegiate level. It was found that CSDM and especially RMDM correlated well with angular acceleration and angular velocity. DDM was close to zero for most impacts due to their mild severity implying that cavitational pressure anywhere in the brain was not reached. Maximum principal strain was found to correlate well with RMDM and angular head kinematic measures. Maximum principal stress didn't correlate with any kinematic measure or injury metric. The model was then used in the investigation of brain injury potential in NHTSA conducted side impact tests. It was also used in parametric investigations of various "what if" scenarios, such as side versus frontal impact, to establish a potential link between head kinematics and injury outcomes. The new SIMon FEHM offers an advantage over the previous version because it is geometrically more representative of the human head. This advantage, however, is made possible at the expense of additional computational time.

Age and gender based biomechanical shape and size analysis of the pediatric brain.

Injuries caused by motor vehicle crashes (MVCs) are the leading cause of head injury and death for children in the United States. This study aims to describe the shape and size (morphologic) changes of the cerebrum, cerebellum, brainstem, and ventricles of the pediatric occupant to better predict injury and assess how these changes affect finite element model (FEM) response. To quantify morphologic differences in the brain, a Generalized Procrustes Analysis (GPA) with a sliding landmark method was conducted to isolate morphologic changes using magnetic resonance images of 63 normal subjects. This type of geometric morphometric analysis was selected for its ability to identify homologous landmarks on structures with few true landmarks and isolate the shape and size of the individuals studied. From the resulting landmark coordinates, the shape and size changes were regressed against age to develop a model describing morphologic changes in the pediatric brain as a function of age. The most statistically significant shape change was in the cerebrum with p-values of 0.00346 for males and 0.00829 for females. The age-based model explains over 80% of the variation in size in the cerebrum. Using size and shape models, affine transformations were applied to the SIMon FEM to determine differences in response given differences in size and size plus shape. The geometric centroid of the elements exceeding 15% strain was calculated and compared to the geometric centroid of the entire structure. Given the same Haversine pulse, the centroid location, a metric for the spatial distribution of the elements exceeding an injury threshold, varied based on which transformation was applied to the model. To assess the overall response of the model, three injury metrics were examined to determine the magnitude of the metrics each element sustained and the overall volume of elements that experienced that value. These results suggested that the overall response of the model was driven by the variation in size, with little variation due to changes in shape. This study demonstrates a new methodology to quantify the shape and size variation of the brain from infancy to adulthood. The use of the changes in shape and size when applied to a FEM suggests that there are differences in the spatial distribution of the elements that exceed a specific threshold based on shape but the overall volume of elements experiencing the specified magnitude was more dependent on the changes in the size of the model with little change due to shape.

On the importance of nonlinearity of brain tissue under large deformations.

Linear shear properties of human and bovine brain tissue were determined from transient stress-relaxation experiments and their material functions were compared. Quasi-linear viscoelastic theory was then utilized to determine material constants for bovine brain tissue subjected to large deformations. The range of applicability for linear and quasi-linear constitutive models of brain tissue was determined. A nonlinear Green-Rivlin constitutive model was subsequently applied to characterize temporal nonlinearity of bovine brain tissue in shear. Overall, 10 brain specimens from 5 fresh human cadavers and 156 brain specimens from 26 fresh bovine cadaver brains were used to quantify and compare shear brain responses under various loading conditions. The assumptions of homogeneity, isotropy, and incompressibility of brain material were made in order to reduce the required number of experiments. A series of single-, two-, and three-step strain inputs was applied to one end of a cylindrical brain specimen and the stress-time histories were measured at the other end. The time delays between the applied strain step inputs were altered in order to determine the temporal nonlinearity of brain tissue. The study resulted in linear constitutive models for human and bovine brain tissue, and quasi-linear and nonlinear constitutive equations for bovine brain tissue in shear. It was found that human brain is somewhat stiffer than bovine brain; the difference, however, was not statistically significant and bovine brain may be a good substitute in studying nonlinear human brain response. A linear constitutive model was found to be sufficient to characterize brain tissue response when Lagrangian shear strains do not exceed 0.2 (advised to limit the range to the shear strain of 0.175), a quasi-linear constitutive model can then be used for loading conditions of up to 0.5 of Lagrangian shear strains (advised to limit the range to the shear strain of 0.325). For any shear strain magnitudes and histories a fully nonlinear Green-Rivlin viscoelastic constitutive model may be utilized.

On the Development of the SIMon Finite Element Head Model.

The SIMon (Simulated Injury Monitor) software package is being developed to advance the interpretation of injury mechanisms based on kinematic and kinetic data measured in the advanced anthropomorphic test dummy (AATD) and applying the measured dummy response to the human mathematical models imbedded in SIMon. The human finite element head model (FEHM) within the SIMon environment is presented in this paper. Three-dimensional head kinematic data in the form of either a nine accelerometer array or three linear CG head accelerations combined with three angular velocities serves as an input to the model. Three injury metrics are calculated: Cumulative strain damage measure (CSDM) - a correlate for diffuse axonal injury (DAI); Dilatational damage measure (DDM) - to estimate the potential for contusions; and Relative motion damage measure (RMDM) - a correlate for acute subdural hematoma (ASDH). During the development, the SIMon FEHM was tuned using cadaveric neutral density targets (NDT) data and further validated against the other available cadaveric NDT data and animal brain injury experiments. The hourglass control methods, integration schemes, mesh density, and contact stiffness penalty coefficient were parametrically altered to investigate their effect on the model's response. A set of numerical and physical parameters was established that allowed a satisfactory prediction of the motion of the brain with respect to the skull, when compared with the NDT data, and a proper separation of injury/no injury cases, when compared with the brain injury data. Critical limits for each brain injury metric were also established. Finally, the SIMon FEHM performance was compared against HIC15 through the use of NHTSA frontal and side impact crash test data. It was found that the injury metrics in the current SIMon model predicted injury in all cases where HIC15 was greater than 700 and several cases from the side impact test data where HIC15 was relatively small. Side impact was found to be potentially more injurious to the human brain than frontal impact due to the more severe rotational kinematics.