A deep learning-based pipeline for developing multi-rib shape generative model with populational percentiles or anthropometrics as predictors.

Rib cross-sectional shapes (characterized by the outer contour and cortical bone thickness) affect the rib mechanical response under impact loading, thereby influence the rib injury pattern and risk. A statistical description of the rib shapes or their correlations to anthropometrics is a prerequisite to the development of numerical human body models representing target demographics. Variational autoencoders (VAE) as anatomical shape generators remain to be explored in terms of utilizing the latent vectors to control or interpret the representativeness of the generated results. In this paper, we propose a pipeline for developing a multi-rib cross-sectional shape generative model from CT images, which consists of the achievement of rib cross-sectional shape data from CT images using an anatomical indexing system and regular grids, and a unified framework to fit shape distributions and associate shapes to anthropometrics for different rib categories. Specifically, we collected CT images including 3193 ribs, surface regular grid is generated for each rib based on anatomical coordinates, the rib cross-sectional shapes are characterized by nodal coordinates and cortical bone thickness. The tensor structure of shape data based on regular grids enable the implementation of CNNs in the conditional variational autoencoder (CVAE). The CVAE is trained against an auxiliary classifier to decouple the low-dimensional representations of the inter- and intra- variations and fit each intra-variation by a Gaussian distribution simultaneously. Random tree regressors are further leveraged to associate each continuous intra-class space with the corresponding anthropometrics of the subjects, i.e., age, height and weight. As a result, with the rib class labels and the latent vectors sampled from Gaussian distributions or predicted from anthropometrics as the inputs, the decoder can generate valid rib cross-sectional shapes of given class labels (male/female, 2nd to 11th ribs) for arbitrary populational percentiles or specific age, height and weight, which paves the road for future biomedical and biomechanical studies considering the diversity of rib shapes across the population.

Material Properties of Fiber Bundles of the Superficial Medial Collateral Ligament of the Knee Joint.

The superficial medial collateral ligament (sMCL) of the human knee joint has functionally separate anterior and posterior fiber bundles. The two bundles are alternatively loaded as the knee flexion angle changes during walking. To date, the two bundles are usually not distinguished in knee ligament simulations because there has been little information about their material properties. In this study, we conducted quasi-static tensile tests on the sMCL of matured porcine stifle joints and obtained the material properties of the anterior bundle (AB), posterior bundle (PB), and whole ligament (WL). AB and PB have similar failure stress but different threshold strain, modulus, and failure strain. As a result, we recommend assigning different material properties (i.e., modulus and failure strain) to the two fiber bundles to realize biofidelic ligament responses in human body models. However, it is often inconvenient to perform tensile tests on AB and PB. Hence, we proposed a microstructural model-based approach to predict the material properties of AB and PB from the test results of WL. Such obtained modulus values of AB and PB had an error of 2% and 0.3%, respectively, compared with those measured from the tests. This approach can reduce the experimental cost for acquiring the needed mechanical property data for simulations.

Kinematic responses of child as second rider of electric-two-wheelers under lateral impact with vehicle.

Electric-two-wheeler (E2W) related accidents have become a major safety concern on road due to the growing prevalence and the high casualty rate. Most existing studies focus on drivers of the E2W, while ignore the second rider (usually a child) as passenger. This study aims at investigating the kinematic response of the child rider upon vehicle impact and analyzing how motion patterns are influenced by the geometric parameters of the vehicle and E2W. A computational framework was established for the intended task. We modeled the E2W-rider system in Madymo, including an E2W with parametric geometry and two riders, one adult and one child respectively. This study focuses on lateral impact in terms of the accident scenarios, as the case dominates in the field data reports. Vehicle types, seating height of the E2W and sitting position of the child rider were considered as variables in the simulation matrix. Results show that the relative height between child's sitting and vehicle hood front-edge, and the sitting position (back-seated or front-seated) are two main influencing parameters on kinematic responses of child rider. The child rider tends to bounce higher on hood upon impact when sitting above the hood front-edge, while might be laterally pushed away by the car-front when sitting below the hood front-edge. Meanwhile, back-seated child rider is more likely to rise higher and rotate faster upon impact compared to a front-seated one. These findings may guide safe riding and safety countermeasure development for child riders of E2W.

Can we reposition finite element human body model like dummies?

Rapidly repositioning finite element human body models (FE-HBMs) with high biofidelity is an important but notorious problem in vehicle safety and injury biomechanics. We propose to reposition the FE-HBM in a dummy-like manner, i.e., through pose parameters prescribing joint configurations. Skeletons are reconfigured along the trajectories inferred from model-specific bone geometries. We leverage differential geometry to steer equidistant moves along the congruent articulated bone surfaces. Soft tissues are subsequently adapted to reconfigured skeletons through a series of operations. The morph-contact algorithm allows the joint capsule to slide and wrap around the repositioned skeletons. Nodes on the deformed capsule are redistributed following an optimization-based approach to enhance element regularity. The soft tissues are transformed accordingly via thin plate spline. The proposed toolbox can reposition the Total Human Body Model for Safety (THUMS) in a few minutes on a whole-body level. The repositioned models are simulation-ready, with mesh quality maintained on a comparable level to the baseline. Simulations of car-to-pedestrian impact with repositioned models exhibiting active collision-avoidance maneuvers are demonstrated to illustrate the efficacy of our method. This study offers an intuitive, effective, and efficient way to reposition FE-HBMs. It benefits all posture-sensitive works, e.g., out-of-position occupant safety and adaptive pedestrian protection. Pose parameters, as an intermediate representation, join our method with recently prosperous perception and reconstruction techniques of the human body. In the future, it is promising to build a high-fidelity digital twin of real-world accidents using the proposed method and investigate human biomechanics therein, which is of profound significance in reshaping transportation safety studies in the future.

Influence of walking on knee ligament response in car-to-pedestrian collisions.

Pedestrians are likely to experience walking before accidents. The walking process imposes cyclic loading on knee ligaments and increases knee joint temperature. Both cyclic loading and temperature affect the material properties of ligaments, which further influence the risk of ligament injury. However, the effect of such walking-induced material property changes on pedestrian ligament response has not been considered. Therefore, in this study, we investigated the influence of walking on ligament response in car-to-pedestrian collisions. Using Total Human Model for Safety (THUMS) model, knee ligament responses (i.e., cross-sectional force and local strain) were evaluated under several crash scenarios (i.e., two impact speeds, two knee contact heights, and three pedestrian postures). In worst case scenarios, walking-induced changes in ligament material properties led to a 10% difference in maximum local strain and a 6% difference in maximum cross-sectional force. Further considering the material uncertainty caused by experimental dispersion, the ligament material property changes due to walking resulted in a 28% difference in maximum local strain and a 26% difference in maximum cross-sectional force. This study demonstrates the importance of accounting for walking-induced material property changes for the reliability of safety assessments and injury analysis.

Self-Supervised Learning for Non-Rigid Registration Between Near-Isometric 3D Surfaces in Medical Imaging.

Non-rigid registration between 3D surfaces is an important but notorious problem in medical imaging, because finding correspondences between non-isometric instances is mathematically non-trivial. We propose a novel self-supervised method to learn shape correspondences directly from a group of bone surfaces segmented from CT scans, without any supervision from time-consuming and error-prone manual annotations. Relying on a Siamese architecture, DiffusionNet as the feature extractor is jointly trained with a pair of randomly rotated and scaled copies of the same shape. The learned embeddings are aligned in spectral domain using eigenfunctions of the Laplace-Beltrami Operator. Additional normalization and regularization losses are incorporated to guide the learned embeddings towards a similar uniform representation over spectrum, which promotes the embeddings to encode multiscale features and advocates sparsity and diagonality of the inferred functional maps. Our method achieves state-of-the-art results among the unsupervised methods on several benchmarks, and presents greater robustness and efficacy in registering moderately deformed shapes. A hybrid refinement strategy is proposed to retrieve smooth and close-to-conformal point-to-point correspondences from the inferred functional map. Our method is orientation and discretization-invariant. Given a pair of near-isometric surfaces, our method automatically computes registration in high accuracy, and outputs anatomically meaningful correspondences. In this study, we show that it is possible to use neural networks to learn general embeddings from 3D shapes in a self-supervised way. The learned features are multiscale, informative, and discriminative, which might potentially benefit almost all types of morphology-related downstream tasks, such as diagnostics, data screening and statistical shape analysis in future.

Assessment of brain injury characterization and influence of modeling approaches.

In this study, using computational biomechanics models, we investigated influence of the skull-brain interface modeling approach and the material property of cerebrum on the kinetic, kinematic and injury outputs. Live animal head impact tests of different severities were reconstructed in finite element simulations and DAI and ASDH injury results were compared. We used the head/brain models of Total HUman Model for Safety (THUMS) and Global Human Body Models Consortium (GHBMC), which had been validated under several loading conditions. Four modeling approaches of the skull-brain interface in the head/brain models were evaluated. They were the original models from THUMS and GHBMC, the THUMS model with skull-brain interface changed to sliding contact, and the THUMS model with increased shear modulus of cerebrum, respectively. The results have shown that the definition of skull-brain interface would significantly influence the magnitude and distribution of the load transmitted to the brain. With sliding brain-skull interface, the brain had lower maximum principal stress compared to that with strong connected interface, while the maximum principal strain slightly increased. In addition, greater shear modulus resulted in slightly higher the maximum principal stress and significantly lower the maximum principal strain. This study has revealed that using models with different modeling approaches, the same value of injury metric may correspond to different injury severity.

A GAN based approach for inferring progression trajectories of costal cartilage calcification from cross-sectional data at image level.

BACKGROUND: Costal cartilage calcification (CCC) increases with age and presents differently for men and women. In individuals, however, the cross-sectional studies that show such trends do not reveal the geometric trajectories through which calcification might accumulate across a lifetime. Generative adversarial networks have the potential to reveal such trajectories from cross-sectional data by learning population trends and synthesizing individualized images at progressive levels of calcification. METHODS: Chest wall mid-surface CT images with normalized cartilage morphologies were produced for 379 subjects aged 6 to 90, and labeled by sex and calcification severity. A conditional GAN with added loss terms to favor one-way accumulation of CCC was trained using organized image batches. GAN performance was assessed by comparing the distributions of images between the training and synthetic groups. RESULTS: Synthetic images generated from a common seed for a given sex and at successive calcification severity levels showed incremental and regional growth of calcification sites. CCC patterns for synthetic male and female images matched known sex-based differences, and individual CCC growth in synthetic images was consistent with previously observed population trends. These trends in the synthetic images were also quantified by structural similarity scores. Synthetic images generated from different input seeds further showed individual variance in specific regions and trajectories of CCC accumulation. CONCLUSION: This study inferred individual progression of CCC accumulation from uncalcified to severely calcified using cross-sectional image data. This information can inform computational models of the changing chest wall biomechanics with age, and the GAN-based technique shows potential for inferring longitudinal data from population trends in other clinical areas.

Obesity effects on pedestrian lower extremity injuries in vehicle-to-pedestrian impacts: A numerical investigation using human body models.

OBJECTIVE: The objectives of this study were to develop a method for modeling obese pedestrians and to investigate effects of obesity on pedestrian impact responses and injury outcomes. METHODS: The GHBMC (Global Human Body Model Consortium) 50th percentile male pedestrian model was morphed into geometries with 4 body mass index (BMI) levels (25/30/35/40 kg/m2) predicted by statistical body shape models. Each of the 4 morphed models was further morphed from a standing posture into 2 other gaits (toe-off and mid-swing), which resulted in a total of 12 (4 BMIs × 3 postures) models. Each model was used to simulate vehicle-to-pedestrian impact under 2 impact velocities. Pedestrian kinematics and injury measures were analyzed focusing on lower extremities. Statistical analyses were performed to examine significance of obesity on concerned injury measures. RESULTS: Peak values of the bending moment at tibia, force at medial collateral ligament (MCL), bending angle at knee joint, and contact force between vehicle and pedestrian increased significantly (P < .05) with increased BMI. By analyzing kinematics of the lower extremity, the overall vehicle-to-pedestrian impact was divided into 2 phases: "initial contact" and "tibia rebound." For obese pedestrians, the added mass caused a higher tibia bending moment in the initial contact phase, and the greater moment of inertia led to greater bending angle and MCL force in the tibia rebound phase. Statistical analyses also revealed that pre-impact posture and impact velocity had significant effects on all injury measures. CONCLUSIONS: Obesity could significantly increase the risk of pedestrian lower extremity injuries due to the inertial effect from the added mass. Pre-impact posture and impact velocity also significantly affect pedestrian injury measures. Future vehicle designs for pedestrian protection should consider populations with obesity. This study demonstrated the feasibility of using parametric human modeling to account for population diversity in injury prediction.