Head injury data reliability
Since the assessment of the predictive performance of different injury criteria for severe brain injury and the establishment of the WIC4 injury criteria were mainly based on the results of head injury reconstruction of 60 in-depth accident cases in this study. Therefore, ensuring the reliability of the whole accident reconstruction chain is an important guarantee of the high credibility of the results. Traditional kinematic reconstruction was mainly based on accident sketches and photographs to compare the final position of the vehicle with the VRUs, as well as the head impact point and the damaged part of the vehicle to determine the validity of the kinematic reconstruction10,46-48. However, this reconstruction method is difficult to ensure the consistency of the impact angle and the precise impact location of the VRUs head with the real accident, and it is also difficult to determine whether the head injury was mainly caused by the vehicle or the ground. In this study, the kinematic reconstruction models were established in MADYMO software using the Chalmers Pedestrian Model49 (CPM) multi-body model that has been verified by cadavers, and the kinematic response of VRUs during the whole collision process was compared with the real accident video to complete the kinematic reconstruction29,30. This approach not only provides more information about the head impact boundary conditions of the VRUs but also further improves the reliability of the kinematic reconstruction. Reliable kinematic reconstruction results are essential for VRUs’ head injury reconstruction. Traditional head injury reconstruction was mostly based on a single criterion to evaluate the severity of head injury50-52. In this study, the VRUs’ head impact boundary conditions were assigned to the THUMS (ver. 4.02) finite element model with high bio-fidelity, and several head injury criteria, including global kinematic-based injury measures and brain tissue level-based criteria were used to reconstruct the VRUs head injury, thus ensuring high reliability of the VRUs head injury reconstruction data.
Sensitivity analysis of single injury criterion to severe brain injury
In this study, the predictive effectiveness of different head injury criteria for severe brain injury was evaluated based on the results of 50 accident reconstructions, and nine injury criteria with better predictive effectiveness were obtained, namely angular acceleration, linear acceleration, HIC15, BRIC, GAMBIT, C.P., MPS, CSDM 0.15 and 0.25, as shown in Fig. 3.
For global kinematic-based injury measures, Chinn et al37. defined an angular acceleration value of 10,000 rad/s2 as the threshold for severe brain injury by statistically analyzing 253 accident cases from the COST 327 database. For cases in which severe brain injury (AIS 4+) occurred (approximately two-thirds of all accident cases), the angular acceleration exceeded 10,000 rad/s2; in addition, for non-severe brain injury cases, there were also 12 cases in which the angular acceleration exceeded 10,000 rad/s2, indicating that the threshold for angular acceleration was set relatively conservatively29. Previous studies have shown that linear head acceleration and HIC15 are good predictors of skull fracture12,13,22, which is usually accompanied by extradural hematoma38 because the protective capacity of the brain is significantly reduced after skull fracture, which explains why linear head acceleration and HIC15 are also good predictors of the risk of severe brain injury. Pure angular velocity is not sufficient to predict brain injury, and the effects of angular acceleration as well as duration need to be considered8. Takhounts et al.15 established a kinematically based brain injury criterion (BrIC) that considers only head rotation velocity based on ATDs test data. However, the THUMS finite element head model produced more deformation than ATDs under the same impact conditions, resulting in part of the impact power being converted into head deformation, which in turn resulted in lower angular velocities based on the THUMS head model, which explains the poor performance of the angular velocity and BrIC in predicting severe brain injury in this study. Both HIP and GAMBIT consider the combined effects of head linear and rotational movements, and HIP was considered to perform more satisfactorily in predicting moderate brain injury, skull fracture, and SDH, and less well than HIC15 in predicting severe brain injury, because linear acceleration causes more severe head injury than angular acceleration when violent collisions were involved38.
For brain tissue response-based injury measures. Previous authors obtained the von Mises stress threshold on concussion risk by analyzing NFL cases18,53,54. Willinger considered von Mises stress as a better predictor of moderate or severe neurological brain injury and obtained a threshold of 38 kPa for severe brain injury risk20. However, the results of the 50 accident reconstructions based on von Mises stress in this study were not evaluated more satisfactorily, which may be due to the different material properties of the brainstem and midbrain of the UPL-FEM and THUMS finite element head models55,56. Takhounts24 used the SIMon finite element head model to simulate the most severe 24 of 1712 impacts in soccer players and obtained that MPS and CSDM were better predictors of DAI, while DDM was weakly correlated with contusion and focal lesion, this result was supported by subsequent studies10,29,57. In this study, the correlations between severe brain injury and MPS, CSDM, and DDM obtained based on 50 in-depth accident reconstructions were similar to the results of Takhounts et al24.
Sensitivity analysis of weighted injury criterion to severe brain injury
Generally, brain injuries caused by traffic accidents are mainly related to a variety of factors such as translational and rotational motion, collision duration, and impact direction, so it’s difficult to use a single injury criterion to evaluate multiple brain injury types15,19,21. Based on 289916 head impact data collected from football players, Greenwald et al. developed a criterion of weighted Principal Component Score (wPCS), which considers both translational and rotational motion, and has been proved to be superior in predicting concussion injuries58. However, the criteria of wPCS did not consider the effects of different impact directions. Afshari et al. analyzed the sensitivity of von Mises stress, coup/contrecoup pressure to impact direction based on 150 finite element simulations, and proposed three new injury criteria, which were SVFEM_index, PCFE_index M, and PCCFEM_index21. However, these criteria were mainly developed based on HIC, and the capability to predict brain injury caused by rotational motion is inadequate. Liu et al. investigated the correlation between kinematic parameters and strain-based measures by using the SIMon model reconstructed 218 accidents from National Highway Traffic Safety Administration and proposed brain injury index (BII) and simple brain injury index (SBII)28. The novel criterion of BII and SBII were superior and more reliable than the single injury criteria for assessing brain injury, but the criterion of SBII ignored the effects of translational velocity and acceleration on brain injury. In addition, He/She established a new head injury criterion ASDHI and ASDHs for the assessment and prediction of ASDH, and although this criterion had better results for ASDH52, its effectiveness in assessing other head injury types was still to be proven.
In this study, based on the reconstruction results of 50 in-depth accidents, four injury criteria that are better for evaluating severe brain injury were selected to establish WIC4 (see equation 5), namely HIC15, angular acceleration, coup pressure, and MPS, which not only considered the advantages of each criterion for the evaluation of different types of brain injury8,13,19,23,43 but also adequately combined the effect of the rotational and translational factors, impact duration and directions12,21. The results showed that the criterion of WIC4 was a better classifier for severe brain injury evaluation than a single injury criterion, as it had a prediction accuracy of over 93% for both severe and non-severe brain injury cases (Fig. 5), with an AUC value of over 0.98 (Fig. 6). The reconstructed results of the other 10 cases also validated the strong correlation between WIC4 and severe brain injury (Fig. 7). Furthermore, there are two main methods for selecting the weight parameters, one is to determine the proportion of injuries in each body part according to the accident statistics, and then assign the weight parameters32; the other is to obtain the prediction model according to the related algorithm and calculate each weight parameter directly21,22,28. In this work, the weight parameters were selected based on the proportion of different head injury types in 104 cases with detailed injury records from the VRU-TRAVi database (Table 1), which was similar to the first method32. The advantage of this approach is that more in-depth accidents could be involved in the process of developing the criteria of WIC4. Admittedly, more accidents need to be counted if the weight parameters are to be closer to real accidents, and this is a deficiency that needs to be improved in future work. In addition, this study obtained WIC 1, 2, 3, and 5 injury risk curves via WIC4 curves based on the scaling method used by Takhounts et al.15 as shown in Fig 8. The scaling factors were determined based on the ratio of HIC15 value in each AIS level at 50% injury risk59, which implies that the effects of WIC and HIC15 are consistent for different head injury AIS levels. However, whether the scale factors of WIC and HIC15 are consistent for different injury levels needs further discussion. Therefore, more cases should be subsequently used to establish the injury risk curves of WIC regarding other levels of AIS, respectively.
Limitations and future works
In this study, there were 34 severe brain injuries and 16 non-severe brain injuries among the 50 accident cases used to establish WIC4. The imbalance of accident data leads to the establishment of injury risk curves of different criteria that are not ideal, because for an ideal injury risk curve when the injury value is zero, the corresponding injury risk should also be zero (Fig 4). Therefore, it is necessary to increase the number of non-severe brain injury cases in the future to equalize the accident data18,19. In addition, due to the high screening requirements of head injury cases, only 104 in-depth accident cases were used to determine the weight factors of each injury criterion (Table 1); however, the number of 104 accident cases was not enough and insufficient to reflect the proportion of each injury type in real traffic accidents. Another limitation of this study is that the THUMS finite element head model used for injury reconstruction was only representative of the 50th percentile adult male and did not consider the differential effects of individual age, height, and weight on the material properties of brain tissue10,30.