Annals of Biomedical Engineering, Vol. 39, No. 10, October 2011 ( 2011) pp. 2568–2583 DOI: 10.1007/s10439-011-0359-5
Development of a Full Body CAD Dataset for Computational Modeling: A Multi-modality Approach F. S. GAYZIK,1,2 D. P. MORENO,1,2 C. P. GEER,2 S. D. WUERTZER,2 R. S. MARTIN,2 and J. D. STITZEL1,2 1
Virginia Tech, Wake Forest University Center for Injury Biomechanics, Winston-Salem, NC 27157, USA; and 2Wake Forest University School of Medicine, Winston-Salem, NC 27157, USA (Received 3 May 2011; accepted 13 July 2011; published online 23 July 2011) Associate Editor Stefan M. Duma oversaw the review of this article.
anatomy for such models. External anthropometry measurements have been studied for the development of both physical and virtual human surrogates.11,26,53,58 Recent datasets for model development are commonly based on medical imaging data. Perhaps the most well known of these datasets is the Visible Human Male (VHM).61 While it is a landmark dataset, there are drawbacks to using VHM in the development of average-sized full body models (FBMs). The subject was quite large, (90 kg, 180 cm) which requires models derived from this set to be scaled to the population of interest. VHM data is also derived from scans of a formalin-perfused cadaver in the supine position. This provides an extra challenge for model developers in translating the data to seated or other postures. Nonetheless, the VHM dataset is valuable and has been used in the development of a number of well-known FEA models.55,56 Recent FBMs demonstrate that imaging techniques are not the only method for development. The HUMOS I model was generated from a male cadaver frozen and sectioned in the driving position.54 However, medical images of this individual were not acquired before sectioning, and the anatomy was from an individual of an advanced age. Yet another set of data commonly used by the modeling community was generated by ViewPoint Datalabs/Digimation, (St. Rose, LA, USA). The bones of this model were digitally recorded from physical specimens, and the set also employed VHM data. It has been used in past modeling efforts as well, including the development of the Total HUman Model for Safety (THUMS) developed by Toyota Central R&D labs.31,59 More recent iterations of the THUMS model employ data from supine CT medical images of an average male subject.60 While these models have made great progress in human body modeling, the data used as their foundation were based on either cadaveric specimens whose
Abstract—The objective of this study was to develop full body CAD geometry of a seated 50th percentile male. Model development was based on medical image data acquired for this study, in conjunction with extensive data from the open literature. An individual (height, 174.9 cm, weight, 78.6 ± 0.77 kg, and age 26 years) was enrolled in the study for a period of 4 months. 72 scans across three imaging modalities (CT, MRI, and upright MRI) were collected. The whole-body dataset contains 15,622 images. Over 300 individual components representing human anatomy were generated through segmentation. While the enrolled individual served as a template, segmented data were verified against, or augmented with, data from over 75 literature sources on the average morphology of the human body. Non-Uniform Rational B-Spline (NURBS) surfaces with tangential (G1) continuity were constructed over all the segmented data. The sagittally symmetric model consists of 418 individual components representing bones, muscles, organs, blood vessels, ligaments, tendons, cartilaginous structures, and skin. Length, surface area, and volumes of components germane to crash injury prediction are presented. The total volume (75.7 9 103 cm3) and surface area (1.86 9 102 cm2) of the model closely agree with the literature data. The geometry is intended for subsequent use in nonlinear dynamics solvers, and serves as the foundation of a global effort to develop the next-generation computational human body model for injury prediction and prevention. Keywords—CAD, Full body model, Seated, Occupant, Injury, Biomechanics, Simulation, Modeling, FEA.
INTRODUCTION Computational models are a cost-effective means of evaluating product and safety system designs in dynamic impact environments.69 Developers have relied on a variety of sources to describe human
Address correspondence to J. D. Stitzel, Virginia Tech, Wake Forest University Center for Injury Biomechanics, Winston-Salem, NC 27157, USA. Electronic mail:
[email protected]
2568 0090-6964/11/1000-2568/0
2011 Biomedical Engineering Society
Development of a Full Body CAD Dataset for Computational Modeling
size and age did not closely align with an average anthropometry, or where of a single modality and in a supine position. The objectives of this study are twofold. The first is to present an approach for full body CAD development based on a living subject, scanned in multiple modalities and postures. The second is to present morphological data on the resulting CAD model including its physical dimensions (volumes, surface areas, and thicknesses). The CAD model presented is the foundation for the development of the Global Human Body Models Consortium 50th percentile male model (M50).22 The consortium’s mission is to create and maintain the world’s most biofidelic computational human body models. The percentile is based on standing stature and weight standards that have been used in the development of regulatory crash test dummies.58 These data are based on the 1974 National Health and Nutrition Examination Survey (NHANES). The motivation for this choice of dataset for overall anthropometric dimensions is examined in greater detail in the ‘‘Discussion’’ section. The recruited subject met these standards and is considered as the mid-sized male for the purposes of this study. The M50 data presented is intended to serve as a reference for biomedical scientists and engineers regarding anatomical data on the average male human body. METHODS Subject Recruitment The subject selection and imaging protocol was approved by the Wake Forest University School of Medicine Institutional Review Board (IRB, #5705). The targeted height and weight of the individual sought for the study were 175.3 ± 2.54 cm and 77.1 ± 3.9 kg, respectively. In addition, extensive anthropometry data were acquired from the applicants.26 Complete details of the selection protocol can be found in a previous study.22 A single individual meeting all criteria was selected for the study (M50). Image Data Protocol Supine MRI data were collected on a 1.5 Tesla Twin Speed scanner (GE, Milwaukee, WI). A 3D Fast Spoiled Gradient Recalled pulse sequence was used. TE and TR were selected such that the fat and water signals were out of phase (Fig. 1). This resulted in a darkened outline around viscera and muscles that aided segmentation. Breath-held scans were acquired in the chest and abdomen so that acquisition time for these was correspondingly short (~30 s.). All the other scans were non-breath held.
2569
Parameters of a typical acquisition were, TR = 5.26 ms, TE = 1.80 ms, Flip angle = 10, Bandwidth = 62.5 MHz, Matrix = 256 9 192 pixels. The field of view (FOV) and slice thickness varied depending on the anatomy being imaged: head and neck (280 mm, and 1 mm thick), thorax (480 mm, and 2 mm thick), shoulder and upper extremity (260 mm, and 2 mm thick), abdomen, pelvis, and lower extremity (400 mm, and 2 mm thick), and foot (300 mm, and 2 mm thick). An eight-channel-phased-array body coil was employed to collect the majority of data. An eightchannel neurovascular coil was employed to collect data from the head and neck. Images were predominantly acquired in the transverse plane, although coronal images of selected anatomy were acquired in head and abdomen. All images were reformatted to a matrix size of 512 9 512. The slice thickness was interpolated in the frequency domain to the value reported above via the GE platform software (14.0 HDx M5). Fifty image sets were collected over three scanning sessions, covering the full body. The upright MRI protocol utilized a 0.6 T Fonar Upright MRI (Fonar Inc., Melville, NY). 3D gradient echo pulse sequences of similar nature were used (Fig. 1). Parameters of a typical axial skeleton acquisition were TR = 14.7 ms, TE = 5.6 ms, Flip angle = 30, and Matrix = 200 9 200 pixels. The FOV and slice thickness varied depending on the anatomy being scanned, head and neck (320 mm, and 1.6 mm thick), thorax through pelvis (430 mm, and 2 mm thick), flexed knee (250 mm, and 1.6 mm thick), and standing shoulder breadth (475 mm, and 2 mm thick). The shoulder scan required a matrix size of 240 9 240 mm. A quadrature head coil and a set of spine and body coils provided by the manufacturer were employed to acquire the images. The slice thickness at acquisition varied between 1.6 and 2 mm. Images were acquired in both the seated (head, neck, thorax, abdomen, and flexed knee) and standing (shoulder, thorax, abdomen, and load-bearing knee) positions. The seat back angle was set to 23 for the seated scans. Images were acquired using Sympulse v. 7.0 software. A total of 17 image sets were collected. Computed tomography (CT) scans were acquired using a GE LightSpeed, 16-slice scanner, software revision 07MW11.10, service pack 2. Images were acquired in helical mode, with the subject in the supine and an approximately seated position. Scanning parameters were set to allow for the lowest practical dose. The approximately seated position was accomplished in two scans and foam and acrylic inserts to accommodate this posture within the restrictive bore size of 72 cm.22 The FOV and slice thickness of the final CT images varied depending on the anatomy being scanned, head (250 mm, and 0.65 mm thick),
2570
GAYZIK et al.
FIGURE 1. Overview of data collected for full body CAD development, showing examples of thorax data. Top Left, CT, supine, arms at sides, Top Right, Conventional MRI, supine, arms at side. Bottom Left, Upright MRI, Bottom Right, External body laser scan in seat buck (raw point data).
neck through thigh (500 mm, and 1.25 mm thick), lower extremity (400 mm, and 0.65 mm thick). The matrix size was 512 9 512 pixels for all scans (Fig. 1). External Anthropometry Model construction relied on external anthropometry data collected from the participant. Bony landmark data and 3D surface scans were collected with a 7-axis 3D digitizer (Faro, Platinum Model arm, 8 ft. (2.4 m), Lake Mary, FL). Surface data from the individual’s body were collected in the form of a point cloud with 0.254-mm spacing between points using the Faro Laser Scan Arm attachment (model, Titanium). Landmark data represented a single point in Cartesian space. A seat buck was designed to acquire landmark data and external body contours in the seated position. Seat buck parameters were adjusted to meet seating models from previous studies.21 The SAE J1733 sign convention for vehicle crash testing coordinate system was determined relative to the seat buck and was used as the model coordinate system. Seat back and pan angles were 23 from vertical and 14.5 from horizontal, respectively.39 After the subject was seated on the buck, 54 landmark locations (21 left, 21 right, and 12 along the midsagittal plane) were identified (Table A1) using Studio software (Studio, v. 11, Geomagic, Research Triangle
Park, NC). All off-axis landmarks were acquired on the right and left sides. Landmarks along the spine were preliminarily marked with small circular stickers before the subject donning the compression fit clothing in an effort to minimize the time spent on the buck. This marking was used as a guide, and the landmarks were re-palpated during the actual data collection. The laser scanner attachment was employed to record the complete body shape of the M50 subject in the seat buck. Figure 1d shows raw, pre-conditioned data. The right (dominant) side of the subject’s body was scanned to reduce scan time. Portions of the left side were also scanned to validate symmetry of the model. During scanning, the subject wore commercially available compression fit athletic clothing. Pre-test trials showed that white fabric allowed for the fastest data collection. A number of sizes were acquired and the subject was allowed to choose the size that best fit his body. The compression fit clothing also circumvented artifacts that would be generated by scanning body hair. Preparation of Image and External Anthropometry Data All images were reviewed by a faculty radiologist at Wake Forest University Baptist Medical Center before use (D.S.W.). Adjacent image sets from the same
Development of a Full Body CAD Dataset for Computational Modeling
modality were merged into a continuous set of images using Amira software (Visage Imaging, San Diego, CA). Merged and aligned image sets were used in segmentation for all body regions. Six sets of transverse images depicting the neck, thorax, abdomen, and pelvis from the upright data were used as a framework for model development in the seated posture.
2571
included noise reduction smoothing, regional decimation (removal of vertices) or refinement (increase of vertices), hole removing, and small feature deletion. The development of CAD data was limited to anatomy pertinent to injury prediction and mitigation. The model is not intended to be a comprehensive reconstruction of the human body, rather a resource for computational model generation for finite element modeling or related simulation approaches.
Model Development Approach A schematic of the overall model development process is shown in Fig. 2. Each component is described in detail below. As many structural data as possible were extracted from the image set. While this subject was carefully chosen to represent an average male occupant, the objective was not to precisely replicate this individual for the model. Instead, the subject was used as a template. During the course of model development, there were instances where it was more appropriate to defer to the literature values for the size or shape of specific components. These instances were limited to finer structures that were not possible to precisely segment from the scan data. In general, these included thin cortical bone structures, thin-walled vessels, and cartilaginous, ligamentous, or tendinous tissues. Segmentation, Conditioning, and Assembly by Tissue Type Mimics software was used for segmentation (v. 13.0, Materialise, Leuven, Belgium). Remaining steps were completed using a variety of commercially available CAD packages (Studio v.12, Geomagic, Research Triangle Park, NC and Rhinoceros v. 4.0, McNeel and Associates, Seattle, WA). Conditioning processes applied to the polygon data using Studio software
FIGURE 2. Model development phases for full body CAD.
Bone All bones of the body were individually segmented from the CT data. Fused bones were not separated. Applicable bone segmentation and image processing methods were tested to facilitate segmentation.57 However, such automated algorithms have generally been developed for specific bones (such as the distal femur in the above reference) and did not apply to a whole body model generation task such as the one used in this study. In addition, specific algorithms for segmentation could not be readily imported into the commercial code used for segmentation, which was the preferred tool, given its flexibility and a large set of pre-programmed tools. Therefore, a semi-automated procedure was utilized to complete the bony segmentation. Bone segmentation began by selecting pixels exceeding 226 HU. To assist in determining cortex edges, a gradient magnitude filter was applied to the images in Mimics. Bones with small articular spaces (such as cervical and thoracic vertebra) were manually separated. Standard segmentation operations, such as region growing, morphological operations, and multi-slice linear interpolation were used. For the majority of bones, only the periosteal surface was segmented. These bones were exported as a 3D polygon shells.
2572
GAYZIK et al.
FIGURE 3. Bone segmentation and conditioning process. (a) Bone (C5) from segmentation, (b) C5, with symmetry plane fit, (c) C5, mirrored, showing deviations from original segmentation (Units: mm).
The polygon surfaces were conditioned to remove artifacts and any structural abnormalities that were found in the resulting segmentations using Studio. The segmentation and conditioning process for a sample bone is shown in Fig. 3. For all the bones that lie on the axial skeleton, symmetry was enforced. A least squares fit plane through the polygon data was utilized to mirror the object. Following the mirroring process, deviations from the original segmented polygon were calculated to verify that significant structural changes had not been introduced in this process. For bones that do not cross the sagittal midplane, the right side was mirrored, so that entire skeleton exhibits sagittal symmetry. For a number of bones, two concentric bone surfaces were modeled: One to represent the periosteal surface of the bone, and a second, within the first, to represent the endosteal surface. Periosteal bone surfaces were determined using the method above. For endosteal cortical surfaces, thickness samples within the image were taken at diaphyseal and epiphyseal regions (in the case of long bones), or in various substructures of the bone. Transverse slices or oblique slices normal to the main axis of the bone were used. Thicknesses were estimated using 50% of the full width of a profile line through the bone.4 These thickness values were utilized to offset the periosteal surface inward (opposite direction of the surface normal) to create a separate surface describing the interior cortical layer. The resulting exterior and interior layers were then manually refined against the original CT images. Previous research has shown that segmentation accuracy becomes unreliable when the thickness of a given structure falls below 1.5 times the full width half max (FWHM) of the point spread function (PSF) of the CT scanner.52 Based on the FOV used in acquiring the scans,16 a limit threshold for thickness was set at 2.75 mm. Thus, when the segmented cortical thickness fell below this value, the interior surface was not used. Instead, the literature data on the cortical thickness in these regions was prescribed by offsetting the exterior surface by the appropriate value. Transitions between
the image- and literature-based endosteal surfaces were made to be smooth and continuously differentiable. The midshaft portion thickness values from the image sets were validated against the literature sources. Bones were assembled by relocating each from the CT image set to the model coordinate system derived from the external anthropometry seat buck. Bone locations were verified against the landmark locations acquired from the external anthropometry portion of the study. Table A1 indicates the landmarks that were used in this process. The acceptable average deviation from a bone surface to its landmarks was 2 cm, accounting for the fact that landmarks were taken on the skin surface. Joint placement was also verified with image data where possible, such as the use of the flexed knee MRI scan or positioned CT scans. Organ Supine MRI data were employed to segment all organs in the CAD model. The majority of organ segmentation was conducted manually using standard techniques, including dynamic region growing, multi-slice editing, and morphological operations. One notable exception to this was the white matter of the brain, which was segmented using statistical parametric mapping software (SPM5, Functional Imaging Laboratory, University College London). This is an atlasbased approach for segmentation. Rather than relying on user segmentation, the software aligns the image set with an atlas (a template brain with label maps for given structures) and uses probability maps to select voxels corresponding to the white matter.3 The segmentation is then transformed back to the original scan space. The mask was verified against the images and manually edited if needed. After segmentation in supine MRI, 3D polygon models were transformed to the upright MRI scan data coordinate system using linear transformation tools in Mimics. This method took advantage of the strengths of both modalities, the field strength and resolution of the conventional, closed-bore MRI, and the image data in the seated position from the upright
Development of a Full Body CAD Dataset for Computational Modeling
MRI. It was applied to CAD development in the head, thorax, and abdomen to capture relative motion between organs that occur with changes in posture. Within the head, this technique was employed to match the distribution of Cerebrospinal Fluid (CSF) between the skull and the cortex of the brain. The shape of the skull itself was assumed to not vary between postures. As per the bone segmentation methods outlined above, the CT-segmented skull was imported into the space of the upright MRI scan. It was aligned to appropriate location in the MRI scan using linear transformation tools in Mimics and visually checked against the location of the bone in the scan. Once aligned, the inner surface of this bone mask was considered the superficial bound of the subarachnoid space (space containing of CSF). A similar process was employed to determine the deep border of the subarachnoid space. The outer cortex of the brain was manually segmented in the upright MRI scan (Fig. 4). As mentioned above, a detailed segmentation of the brain from the supine MRI was conducted. This segmentation was then imported into the upright scan space and linearly transformed to align with the outer cortex segmentation of the brain from the upright MRI scan. The aligned skull and brain segmentations therefore delineated the CSF space. Regarding assembly in the thorax and abdomen, the assembly process took advantage of the body’s natural compartmentalization. The mediastinum and abdominal cavity (peritoneum, retroperitoneum, and perineum) were segmented in the upright scan (Fig. 5). These physical compartments were then used as a
FIGURE 4. Rendering of upright MRI exterior cerebral cortex (red) with aligned CT skull (blue). Slice shows pixels from a sagittal rendering of the image. Direction of gravity is superior to inferior.
2573
bounding region into which organ segmentations from the supine scans were made to conform. Manual morphing of the thoracic and abdominal organs was conducted if necessary, using the upright image data as a means to verify changes that were made. The upright MRI data did not yield organ segmentations of the same quality as the supine MRI data because of its lower resolution. Therefore, segmentations from the latter were imported into the upright scan coordinate system and linearly transformed to overlap with the location of the same organ in the upright scan. Once aligned, both the upright MRI segmentations and the MRI segmentations were imported into Studio software in STL (polygon) file format. The detailed MRI segmentations were edited and morphed using the conditioning tools described previously, as well as the following morphing tools: additive and subtractive sculpting, surface offsets, and defeaturing. Edited STL files were re-imported back to Mimics software to verify that good agreement with the scan data was reached, and that changes did not violate the bounds of the compartment in which the objects were located. Volume was conserved within a tolerance of 5%.
FIGURE 5. Compartmentalization showing skeleton and major cavities used in CAD assembly (mediastinum, red, abdominal, blue).
2574
GAYZIK et al.
Vessels Where possible, vasculature was segmented from the MRI and CT scans of M50. Based on the need for contrast-enhanced CT, an anonymous scan from the medical image database at Wake Forest University of male with similar stature and weight as the M50 subject (185 cm, 79 kg) was employed to supplement the M50 image data. Geometry data acquired from the supplemental scan were limited to vascular connections of the liver, kidneys, and spleen and were verified against the literature studies before use in the model. An approach based on the literature data and the centerline of the vessel trajectory was employed to develop CAD geometry of the vessels within the thorax and abdomen. The approach is shown graphically in Fig. 6. Following segmentation, a center line feature within Mimics software was employed to reduce the vessel to only its basic trajectory. The literature data on the average values of the diameter of the vessel of interest were then employed to generate a circular cross section radially outward from the centerline. Segmentation data were used at the organ attachment location whenever possible to allow for smooth transitions between the geometry obtained from direct segmentation and the one generated via the literature data. The aorta and the vena cava were directly segmented as they were readily visible in the scans.
morphed to match attachment locations. Muscle volumes were conserved during the morphing process. Fifty-two muscles of the neck were explicitly segmented. This included 26 on each side, with only the sternothyroid and thyrohyoid grouped together as a single structure. Most tendinous muscle attachments in this region were not segmentable due to their thin cross section and low water content. In these cases, classical anatomical texts were employed to determine muscle attachment locations and 3D polygon models were modified accordingly in the model coordinate system.27,48 In some cases, neck muscles segmentation was not possible using the 1.5 T scans of the neck. In these cases, high-resolution digitized photographs of anatomical cross sections of the Visible Human Project were used.61
Muscle Muscles were segmented for the model development effort as were germane to modeling blunt injuries. In each body region, muscles were directly segmented from the supine MRI image data using techniques described above (dynamic region growing, multi-slice editing, and morphological operations). The MRI pulse sequence yielded images with fat and water out of phase and proved well suited for segmenting large muscle groups since the muscle boundaries were readily visible. Muscle segmentations were imported directly into the model coordinate space and manually
External Skin Surface An outer skin surface was developed from the external anthropometry scan data. The raw data in Fig. 1d underwent conditioning using Studio software (mirroring, noise reduction, regional decimation or refinement, hole removing, and small feature deletion) since slight motion by the subject created an uneven surface. The resulting polygon skin was verified against nine external measurements acquired from the study subject. These measurements were acquired as part of the screening process and per the methods described in Gordon et al.19 The measurements were, head circumference, head breadth, neck circumference, bideltoid breadth, chest circumference, waist circumference, hip breadth, buttock to knee length, and knee height. Distance measurements on the skin surface model were taken point to point using Studio software. Planes positioned at the appropriate anatomical level were utilized to cut ellipses in the surface shell for the various circumferential measures. The lengths of the ellipses were determined in Studio software as well.
FIGURE 6. Center line approach to developing vasculature in M50. Segmented vasculature (splenic artery and vein) is transparent showing underlying reduced centerline (Left). CAD data are shown at the right with prescribed diameters (Right).
Cartilage, Fibrocartilage, Ligament, and Tendon A small number of cartilage, fibrocartilage, ligament, and tendon anatomy were included in the model. These were segmented manually, and included the costal cartilage, articular cartilage of the knee (femoral, tibial, and patellar), ligaments of the knee, the menisci, and the calcaneus tendon. For structures on the surfaces of bone, such as articular cartilage, Boolean operations in the polygon conditioning phase were employed.
CAD Development The final stage of model development was the CAD development. In this stage, a mathematically defined NURBS patchwork was constructed over conditioned
Development of a Full Body CAD Dataset for Computational Modeling
polygon data. Border continuity was enforced such that the patchwork was G1 continuous, or tangentially continuous with neighboring patches. If sagittal symmetry was required for a given part, then the patch work was completed on one half of the part and then mirrored. Given the amount of parts in the complete model, care was taken to use the appropriate resolution for the surface patches. The patchwork was refined to balance between computational efficiency and structural detail. For all the components, the lowest number of splines per patch, which still captured the contours of the underlying polygon model was used. RESULTS A 26-year-old male was selected for the study. His height, weight, and BMI were 174.9 cm, 78.6 ± 0.77 kg, and 25.7 ± 0.25, respectively. Standard deviations represent variation over the enrollment period. The subject passed all exclusion criteria, had a clean medical history, and was in excellent health. The subject was deemed to be typical, exhibiting no major anatomical abnormality or pathologic condition. CAD Summary The M50 model comprises the elements for biomechanical modeling efforts of the mid-sized male in the seated position. The NURBS surfaces of the CAD matched the segmented polygon data closely, with a typical differences in volume and surface areas of the components of less than or equal to 1%. There are 418 individual CAD parts in the M50 model, including 179
2575
individual bones (described with 216 parts), 46 organs and components thereof, 96 muscles, 37 vessels and 26 ligaments, tendons, and other cartilaginous structures. Vasculature, ligamentous, and tendinous structures were also included in the model development, and these account for less than 1% of the total model volume. The NURBS surfaces that compose the model are water tight to a tolerance of 0.01 mm. Various aspects of the CAD data are shown in Fig. 7.
Skeletal System The skeletal system of the human body is represented in the model. As shown in Fig. 3, deviations between asymmetric segmentations of axial bones and the same bones with enforced symmetry were on the order of 2 mm. Figure 7 shows the bony assembly of CAD model. Geometrical features are included that are germane to biomechanical modeling efforts. The skull includes the frontal sinuses, diploe¨, and tables. The transitions from cortical bone surface to costal cartilage are denoted by concave surfaces in the ribs. Relevant protuberances in the lower extremity for muscle attachment are also included. For a total of 24 bones (left and right sides), both periosteal and endosteal surfaces are modeled (Table 1). Epiphyseal thicknesses are accurately modeled. For a number of commonly injured bones, the cortical thickness of nearly the entire structure was below the scanner cut off, but variable thickness would be required for accurate modeling. Since the study is intended for subsequent Finite Element modeling, and 2D shell elements are likely to be used for these bones, the literature data
FIGURE 7. Summary of M50 CAD model. (a) Exterior skin and skeletal structure. (b) Muscle and ligamentous structures within the knee. (c) Organ components of the M50 model with detailed image of brain structures. (d) Posterior view of organ structures.
2576
GAYZIK et al. TABLE 1. Bone CAD cortical thicknesses.
Bone Clavicle Scapula Sternum Humerus
Radius Ulna Femur Femoral neck Sacrum Patella Tibia Fibula Talus Calcaneus
CAD Region average thickness (mm) from image segmentation
Literature Thickness at location, reference (mm)
MS: LB: FS: P: MS: D: P: MS: P: MS: MS: FS: FS: FS: MS: MS: FS: FS:
2.01 3.8 ± 0.9547 1.520 3.05–6.7564 – 3.35–9.064 1.7–3.5564 – 2.85–5.6564 – 6.81 ± 0.0729 1.5–3.56,7 1.050 0.619 4.1–8.264 2.3–4.7564 0.4568 0.4568
2.0 2.7 1.5 5.1 6.1 6.5 3.5 4.3 4.5 4.8 6.2 1.5–3.5 1.0 0.6 6.7 4.2 0.45 0.45
CAD Region of thickness below CT threshold Entire structure Glenoid fossa Entire structure Distal and Proximal Epiphyses
Prescribed thickness, epiphysis (mm) 1.01 GF: 0.641 – 1.015
1.0 1.0
Entire structure Entire structure Entire structure Distal and Proximal Epiphyses Entire structure Entire structure
1.08 – – – 1.0 1.0 – –
The regions below CT scanner reconstruction threshold are noted, see ‘‘Methods’’ section. Locations: FS full structure, LB lateral border, P 1/3 length of bone from proximal end, D 1/3 length of bone from distal end, MS Midshaft, GF Glenoid fossa.
can be utilized to prescribe cortical thickness2,33,36 on the outer shell. With regard to the assembly process, the overall average deviation between landmarks and corresponding bones was 1.6 cm. Exterior Skin A single surface representing the external skin was developed from the external anthropometry laser scanning. In some regions, external measurements on the M50 participant indicated that breathing or motion artifacts affected the accuracy of the skin layer. In these regions, data from segmentations of the CT images were substituted. The surface is shown superimposed over the skeleton in Fig. 7. The exterior surface of the model has a total surface area of 1.86 m2, which is within 5% of estimates in the literature for a male of the same height and body weight.9 The total enclosed volume is 75.7 9 103 cm3, and this volume also closely matches the literature data.42 The average deviation of the nine measurements taken to verify the external skin CAD geometry against the study subject was 2.2%. The single largest percentage difference, 4.7% (1.7 mm), was found in the neck circumference. Organ Systems of M50 Nervous system modeling focuses on the brain and spinal cord. The following 16 structures are repre-
TABLE 2. Selected brain structures from the M50 model with corresponding volumes.
Structure Basal ganglia (L + R) Cerebellum Cerebrum (L + R) Corpus callosum Ventricles (lateral, 3rd, 4th)a Thalamus and brainstema Fornix Dural venous sinuses (superior + transverse only)
M50 volume (cm3) 19.4 141.8 1021.1 20.1 22.9 34.5 0.43 9.8 9 1023
a
Volumes summed.
sented by the CAD model of the brain: The left and right cerebral hemispheres, venous sinuses (transverse, superior), ventricles (lateral, 3rd, and 4th), brainstem, basal ganglia (left and right), corpus callosum, fornix, thalamus, cerebellum, falx, and tentorium. Volumes of these components are provided in Table 2. All the structures in the M50 brain are the result of segmentation, with the exception of the falx cerebri and tentorium, which were constructed in CAD since reliable segmentation of thin structures was not possible. The M50 model includes detailed CAD of the thoracic and abdominal viscera (Fig. 7, Table 3). Major components of the circulatory system are modeled, with the heart and great vessels being the most prominent. Primary and secondary branches off of major
Development of a Full Body CAD Dataset for Computational Modeling TABLE 3. Thoracic and abdominal organ structures with corresponding volumes. M50 value: volume (V, cm3)Surface area (SA, cm2), or diameter (D, cm)
Structure Heart Lung (L) Lung (R) Sum L and R lungsa Diaphragm Stomach Liver Gallbladder Pancreas Duodenum Jejunum (proximal) Colon Left kidney Right kidney Sum kidneys Bladder Ureters Spleen
V V V V SA V V V V D D D V V V V D V
681.0 1729 1801 3530 1025 653.3 1435 15.3 77.7 2.2 2.1 2.0–4.0 145.2 146.4 291.6 99.6 0.4 188.7
a
Volume at functional residual capacity (FRC).
arteries with diameters greater than 4 mm were included, so that their role as organ tethers could be modeled in dynamic crash events. Vessel diameters were validated against literature sources (Table 4). Major components of the digestive system were modeled, including the stomach, liver, duodenum, proximal jejunum, distal ileum, and all portions of the colon. The small bowel was included as a volume region within the abdominal cavity, but not modeled in detail. Components of the urinary system were modeled including the kidneys, ureters, and bladder. The pancreas and spleen were also represented.
Muscular Components of M50 Ninety-six muscles are explicitly modeled, many of which are symmetric about the sagittal midplane, including the muscles of the neck (26 on each side), and 22 muscles throughout the thorax, abdomen, and extremities (Fig. 7). Specific muscles are detailed in Table 5. The volume of the upper extremity muscles matched literature sources closely.30 Volumes of the 26 muscles within the Neck region compared well with a previous study.35 Origin and insertion points were modeled by trimming the muscle geometry to the appropriate bones. Given the anatomically accurate models of the muscles, they can be included in subsequent dynamic FEA analysis as either passive or active structures.
2577
TABLE 4. Typical diameter of selected vessels included in M50 model.
Arteries Ascending aorta Descending aorta Pulmonary Common carotid Subclavian Celiac trunk Splenic Hepatic Renal Superior mesenteric Inferior mesenteric Common iliac
M50 diameter (mm) 28.4 16.0 24.0 6.6 8.4 8.5 6.3 6.1 6.5 7.3 3.3 10.5
Veins
M50 diameter (mm)
Vena cavaa Pulmonary Subclavian Internal jugular Portal Splenic Superior mesenteric Renal Common iliac
20a 16.0 11.0 13.0 12.8 6.7 10.0 9.0 12.0
a
Superior and inferior.
Ligamentous, and Cartilaginous Components of the M50 Model CAD A limited number of ligamentous, tendinous, and cartilaginous components were included in the CAD model. Since the CAD data will be used predominantly in an FEA-modeling effort, many of these structures will be modeled directly as 1D beam or 2D shell elements. The approach is made possible by the discretization of 3D CAD structures into finite elements. The discretized shape is therefore composed of a finite number of points in space called nodes. 1D or 2D structures are readily attached to 3D structures via node-to-node attachments.66 Instead of replicating many of these structures in CAD, development study was focused on a limited number articular cartilage components, ligaments, and tendons which could be reliably segmented, and have been shown to be of importance in vehicular crash. These structures were found predominantly in the knee (Fig. 7). The volume and average thickness of each of these structures closely matched the data of previous research.13
DISCUSSION This study presented methods to generate a fullbody CAD model of the mid-sized male in the seated posture, leveraging modern imaging techniques, and the significant quantity of data that is available in the open literature. There are a number of advantages in the chosen approach, beginning with subject selection. All bony data and nearly all soft tissue data in the M50 model were derived from scans of a living subject who was thoroughly pre-screened. The subject met a large set of acceptance criteria ranging from his anthropometry to MRI compliance.
2578
GAYZIK et al. TABLE 5. Selected musculature included in M50 model, with volumes.
Region Upper extremity
Thoracic
Abdominal
Lower extremity
Muscle
M50 volume (cm3)
Deltoid Biceps brachii Triceps brachii Pectoralis major Rhomboid major Trapezius (lower)a Latissimus dorsi Diaphragmb (cm2) Quadratus lumborum Transverse abdominisc Internal obliquec External obliquec Rectus abdominis Erector spinaed Iliacus Psoas (minor and major)c Levator ani Piriformis Obturator internus Quadriceps femoris Gluteus maximus Gastrocnemius (medial and lateral) Plantaris Soleus (medial and lateral)
460.5 186.2 460.3 368.7 25.6 39.6 362.4 997.3 65.6 419.4
251.2 465.2 141.7 210.1 67.1 42.0 57.2 1897.2 1298.6 412.0 6.9 463.2
Region Neck
Muscle
M50 volume (cm3)
Splenius capitis Splenius cervicis Semispinalis capitis Semispinalis cervicis Trapezius (upper/middle) Longissimus capitis Longissimus cervicis Iliocostalis cervicis Oblique capitis superior Oblique capitis inferior Multifidus Levator scapulae Rectus capitis posterior maj. Rectus capitis posterior min. Rhomboid minor Rectus capitis anterior Rectus capitis lateralis Sternocleidomastoid Longus colli Longus capitis Scalene anterior Scalene medius Scalene posterior Omohyoid Sternohyoid Sternothyroid/thyrohyoid
30.8 19.7 47.5 15.1 113.1 9.4 6.5 6.5 3.7 4.2 8.6 29.2 5.6 1.8 9.3 0.6 0.6 45.5 8.9 5.5 7.9 6.9 6.5 9.4 6.8 9.8
a
Middle and upper portions included in neck muscle group. Surface area at FRC. c Grouped as a single component of M50. d CAD of this muscle represents the structure inferior to the 8th rib. b
Subject selection was guided by the ultimate intended use of this dataset, vehicle safety system design simulations. Therefore, the anthropometry of the individual selected was selected to match existing ATDs (Anthropomorphic Test Devices, commonly known as crash test dummies).43,46,58 The height and weight targets for the individual selected for this study originate from the NHANES study in 1974, which today is known to underestimate the weight of the average sized U.S. male because of increases in corpulence over the last 3 and a half decades.49 Despite this limitation, there were two main motivations for using the original NHANES data. First, the resulting computational models will be directly comparable to the ATDs, since they are based on the same data. ATDs are an integral part of regulatory and consumeroriented crash tests and are standardized across all manufacturers. Matching the ATD anthropometry will facilitate comparisons of injury assessment values from ATDs to more localized injury measures that can be derived from FEA models based on this CAD data. Second, as manufacturers design vehicles for a global market, the 1974 NHANES data may be a better predictor for a global mid-sized male than more recent
data from a U.S. citizen only sample, since such increases in weight may not be uniform across all cultures. Anthropometry data in addition to height and weight measures were utilized to select the study participant. The acquisition methods and target values for these 15 measures22 were taken from a study by Gordon et al. on the U.S. armed services members, known as ANSUR (the U.S. Army Anthropometry SURvey).26 While it is clear that the population in Gordon’s study is dated and not equivalent to a world average population, the ANSUR data were ultimately selected for use in this study. The sheer size and comprehensive nature of the dataset make it one of the best sources available. The ANSUR study screened over 25,000 potential subjects and ultimately reported measurements on 1774 males. To compensate for the differences between the population considered in the present study (the general driving population) and the reference population reported by Gordon, a ±5% deviation from the reported value of the 50th percentile male in Gordon was permitted on all measures. In addition being that the target individual was near the mean values in the set, the data from ANSUR were likely not
Development of a Full Body CAD Dataset for Computational Modeling
skewed by a small number of individuals, as might be the case if the target were near the 5th or 95th percentiles. The image dataset used in the development of the CAD model is distinct from many of the other sets of data that have been used in the development of full body CAD or FEA models, many of which contain cadaveric data, from individuals of an advanced age, or of a dramatically different size than the mid-sized male. While the study involved a single individual, the components of the M50 model agree with broader studies within the literature. No previous datasets found in the literature contain significant scan data in the upright position. The upright MRI and quasi-seated CT data acquired are the first of their kind to be collected for the expressed purpose of developing a full-body CAD model for the injury or broader biomechanics research communities. The upright MRI data was critical for the assembly process. While the scan quality did not match the closed-bore MRI due to the reduced field strength of this scanner (0.6 T vs. 1.5 T), the data proved valuable in positioning of soft tissues. These scans were employed to more accurately depict the CSF layer within the skull when the subject is seated in the upright position. Cerebrospinal fluid (CSF) provides a buoyant effect on the brain, and is therefore thought to provide some mitigation of external forces. Posture may affect the distribution of CSF, in turn affecting the biomechanics of head impacts, yet most models are developed from supine medical images that do not take postural changes into account. Furthermore, recent studies have highlighted the affect of posture on the location of abdominal contents. Using the same upright scanner model employed in the present study, Beillas et al. found that the movement of organs within the abdomen is consistent with the effects of gravity and most pronounced when shifting from the supine to an upright position.5 While majority of variance in organ location was a result of differences between subjects, it is clear that the location of these organs is affected by posture. This is of paramount importance for human body simulations aimed at identifying the interaction that various countermeasures (air bags, belts, etc.) have on the internal organs of the body. Structures have been developed in such a way that they are amenable to FEA or computational fluid dynamics (CFD) meshing. In any biomechanical study, the scale in which an object is being studied will be an important factor. For example, investigators researching pulmonary pathology may be interested in what occurs at the alveolar level of the lung, or may study on a more macroscopic scale, by considering the lung a continuum structure.23,24 In developing the
2579
CAD geometry, this effort was focused on a model that would be useful at the continuum level. Much of the microstructure of the human body has been omitted since validation of any subsequent FEA models will be conducted using experimental data collected at the organ or full body scale. By working additionally on a macroscopic level, the CAD is not a mere replication the individual scanned in this study, but a model that was made to match literature where necessary. The intended application of this study for FEA modeling at the continuum level was also an important consideration when undertaking the development of the model’s musculature. The 96 muscles explicitly represented in the CAD dataset are a subset of the total number of skeletal muscles in the human body. CAD representations of these particular muscles were developed for a number of reasons. Muscles that play a large role in the biomechanics of frequently occurring crash induced injuries were targeted, such as the 52 muscles of the neck. Secondly, many of the selected muscles are located near regions that are loaded during vehicular crash. Seat belt loading, for example, has clear interaction with muscles of the neck, pectoral muscles of the chest, and rectus abdominis, and oblique muscles in the trunk. Modeling the morphology, insertion, and origin points of these structures will ultimately provide more information in terms of a biofidelic response and mass distribution in simulations. In injury biomechanics studies, the typical approach for modeling the role of musculature is to model the passive components of the muscle with a hyperelastic and a viscoelastic rubber behavior such as that developed by Ogden12 and to model the active component of muscle resistance with 1D beam elements following a hill-type material model.38 The CAD representation of the muscles is well suited for the passive component. 1D muscle components can easily be added following discretization of the musculature through the meshing process.69 This approach has been previously used in modeling the response of the neck28 as well as the lower extremities.10 In recent full body FEA models, the approach employed to model musculature varies. Ruan et al. do not explicitly include skeletal muscle,55 whereas Shigeta models the active portion of muscles with the 1D components described above.60 The volumes of the muscles included in the model are provided in Table 5. Physiological aspects of the muscles such as physiological cross-sectional area and pennation angle are well described in the literature, for the neck,32 lower extremity37 and upper extremity,30 and are considered beyond the scope of the present study. Such data from the literature, in conjunction with the muscle volumes presented here, may be useful in the implementation of
2580
GAYZIK et al.
muscle models in dynamic FEA software packages like LS-Dyna’s *mat_muscle material card.38 There are limitations to the chosen approach. The use of a live subject imposes certain limitations relative to scan time, since participant comfort and safety were a major consideration. A live subject also introduces the possibility of motion artifacts due to gross motion of the subject, which was controllable through communication or from random muscle contracts such as peristalsis, which was not. In all the cases, the data required to construct the model was acquired. The cases where motion was not controllable (i.e., small bowel) were generally not targeted for model inclusion. While it is true that a single subject was used for the scanning process, many components that are critical to injury prediction, such as cortical thickness, vessel diameter, and vessel thickness were constructed from population-based literature sources. There are a number of promising future directions for the methods presented here. The advanced imaging and scanning protocols described can be applied to develop models of varying sizes (e.g., the 95th percentile male), and models of the opposite gender
(e.g., the 5th percentile female). For example, a recent 5th percentile female model in the literature was scaled from an average male model using anthropometric data.34,69 A more holistic approach would be to begin with image data of the targeted population as per the methods in the article. Furthermore, these techniques could be applied to models of vulnerable populations (e.g., the pregnant female) to advance their use and accuracy.44,45 The methods described can be used for other body regions to refine and add detail to specific anatomical models of the eye,17,51,63,65–67 neck,40 and thorax18,25,62 for example. Improved modeling techniques will enable better design and development of physical models of the human body.14,62 Implementation of this imaging and CAD development protocol could also begin to fill a need in the literature for population-based data on the relative position of organs in postures of interest. While the CAD data will certainly be used in injury biomechanics studies, there are myriad other potential applications for this data, including fluid dynamics, ergonomics studies, anatomical instruction, or preoperative planning simulations.
APPENDIX TABLE A1. Landmarks acquired for the external anthropometry study used in model construction.
Landmark
Bone(s) location determined via landmark
Top of head Back of head Tragion (R&L) Glabella Infraorbitale (R&L) Corner of eye (R&L) Anterior superior iliac spine (R&L) Posterior superior iliac spine (R&L) Pubic symphysis Suprasternale Substernale Med. clavicle (R&L) Lat. clavicle (R&L) Lat. humeral condyle (R&L) Med. humeral condyle (R&L) Ulnar styloid (R&L)
Skull Skull Skull Skull Skull Skull Pelvis Pelvis Pelvis Sternum Sternum Clavicle Clavicle, scapula Humerus, radius Humerus, ulna Ulna, triquetral, (HWC)
Landmark
Bone(s) location determined via landmark
Radial styloid (R&L) Second metacarpal (R&L) Fifth metacarpal (R&L) Lat. femoral condyle (R&L) Med. femoral condyle (R&L) Supra-patella (R&L) Lat. malleolus (R&L) Med. malleolus (R&L) Ball of foot (R&L) Fifth metatarsal (R&L) Heel (R&L) 7th cervical vertebrae (C7) 4th thoracic vertebrae (T4) 8th thoracic vertebrae (T8) 12th thoracic vertebrae (T12) 3rd lumbar vertebrae (L3) 5th lumbar vertebrae (L5)
Radius, trapezium, (HWC) 2nd metacarpal, 2nd prox. phalange (HWC) 5th metacarpal, 5th prox. phalange (HWC) Femur, fibula Femur, tibia Patella Fibula, calcaneus, FC Tibia, talus, FC 1st metatarsal, 1st prox. phalange, FC 5th metatarsal, 5th prox. phalange, FC Calcaneus, FC C7, CS T4, TS T8, TS T12, TS L3, LS L5, LS
R and L, respectively, indicate that landmark was acquired on the right and left sides. HWC Hand and Wrist Complex, treated as single unit for placement into final model space, FC Foot complex, treated as single unit for placement into final model space, CS Cervical spine, also verified with upright MRI scan, TS Thoracic spine, verified using positioned CT scan, LS Lumbar spine, verified using positioned CT scan.
Development of a Full Body CAD Dataset for Computational Modeling
ACKNOWLEDGMENTS Funding for this study was provided by the Global Human Body Models Consortium, LLC (GHBMC) through grant WFU: FBM-001. CAD development study was supported by Zygote Media Group, Inc. (American Fork, UT).
CONFLICT OF INTEREST The authors have no conflict of interest to report.
REFERENCES 1
Andermahr, J., A. Jubel, A. Elsner, J. Johann, A. Prokop, K. E. Rehm, and J. Koebke. Anatomy of the clavicle and the intramedullary nailing of midclavicular fractures. Clin. Anat. 20(1):48–56, 2007. 2 Anderson, A. E., C. L. Peters, B. D. Tuttle, and J. A. Weiss. Subject-specific finite element model of the pelvis: development, validation and sensitivity studies. J. Biomech. Eng. 127(3):364–373, 2005. 3 Ashburner, J., and K. J. Friston. Unified segmentation. NeuroImage 26(3):839–851, 2005. 4 Au, A. G., D. Palathinkal, A. B. Liggins, V. J. Raso, J. Carey, R. G. Lambert, and A. Amirfazli. A NURBSbased technique for subject-specific construction of knee bone geometry. Comput Methods Progr. Biomed. 92(1):20– 34, 2008. 5 Beillas, P., Y. Lafon, and F. W. Smith. The effects of posture and subject-to-subject variations on the position, shape and volume of abdominal and thoracic organs. Stapp Car Crash J. 53:127–154, 2009. 6 Bell, K. L., N. Garrahan, M. Kneissel, N. Loveridge, E. Grau, M. Stanton, and J. Reeve. Cortical and cancellous bone in the human femoral neck: evaluation of an interactive image analysis system. Bone 19(5):541–548, 1996. 7 Blain, H., P. Chavassieux, N. Portero-Muzy, F. Bonnel, F. Canovas, M. Chammas, P. Maury, and P. D. Delmas. Cortical and trabecular bone distribution in the femoral neck in osteoporosis and osteoarthritis. Bone 43(5):862– 868, 2008. 8 Braun, M. J., M. D. Meta, P. Schneider, and C. Reiners. Clinical evaluation of a high-resolution new peripheral quantitative computerized tomography (pQCT) scanner for the bone densitometry at the lower limbs. Phys. Med. Biol. 43(8):2279–2294, 1998. 9 Burmaster, D. Lognormal distributions for skin area as a function of body weight. Risk Anal. 18(1):27, 1998. 10 Chang, C. Y., J. D. Rupp, N. Kikuchi, and L. W. Schneider. Development of a finite element model to study the effects of muscle forces on knee-thigh-hip injuries in frontal crashes. Stapp Car Crash J. 52:475–504, 2008. 11 Cheng, H., L. Obergefell, and A. Rizer. Generator of body data (GEBOD) manual report, AL/CF-TR-1994-00511. Armstrong Laboratory, Wright Patterson Air Force Base, 1994. 12 Christensen, R. A nonlinear theory of viscoelasticity for application to elastomers. J. Appl. Mech. 47:762–768, 1980. 13 Cohen, Z. A., D. M. McCarthy, S. D. Kwak, P. Legrand, F. Fogarasi, E. J. Ciaccio, and G. A. Ateshian. Knee car-
2581
tilage topography, thickness, and contact areas from MRI: in vitro calibration and in vivo measurements. Osteoarthr. Cartilage/OARS, Osteoarthritis Research Society 7(1):95– 109, 1999. 14 Crowley, J. S., F. T. Brozoski, S. M. Duma, and E. A. Kennedy. Development of the Facial and Ocular Countermeasures Safety (FOCUS) headform. Aviat. Space Environ. Med. 80(9):831, 2009. 15 Diederichs, G., A. S. Issever, S. Greiner, B. Linke, and J. Korner. Three-dimensional distribution of trabecular bone density and cortical thickness in the distal humerus. J. Should. Elb. Surg. 8(3):399–407, 2009. 16 Dougherty, G., and D. Newman. Measurement of thickness and density of thin structures by computed tomography: a simulation study. Med. Phys. 26(7):1341–1348, 1999. 17 Duma, S. M., T. P. Ng, E. A. Kennedy, J. D. Stitzel, I. P. Herring, and F. Kuhn. Determination of significant parameters for eye injury risk from projectiles. J. Trauma 59(4):960–964, 2005. 18 Duma, S. M., P. H. Schreiber, J. D. McMaster, J. R. Crandall, C. R. Bass, and W. D. Pilkey. Dynamic injury tolerances for long bones of the female upper extremity. J. Anat. 194(3):463–471, 1999. 19 Eckstein, F., S. Milz, H. Anetzberger, and R. Putz. Thickness of the subchondral mineralised tissue zone (SMZ) in normal male and female and pathological human patellae. J. Anat. 192(1):81–90, 1998. 20 Fawzy, H., N. Alhodaib, C. D. Mazer, A. Harrington, D. Latter, D. Bonneau, L. Errett, and J. Mahoney. Sternal plating for primary and secondary sternal closure, can it improve sternal stability? J. Cardiothorac. Surg. 4:19, 2009. 21 Flannagan, C., M. Manary, L. Schneider, and M. Reed. An improved seating accommodation model with application to different user populations. Proceedings of the SAE World Congress & Exposition, Detroit, MI, Feb. 23–26, 1998. Vol. 1358, pp. 43–50, SAE, Warrendale, PA, USA. 22 Gayzik, F. S., C. A. Hamilton, J. C. Tan, C. McNally, S. M. Duma, K. D. Klinich, and J. D. Stitzel. A multimodality image data collection protocol for full body finite element model development. SAE Technical Paper 200901-2261. doi:10.4271/2009-01-2261. 23 Gayzik, F. S., J. J. Hoth, M. Daly, J. W. Meredith, and J. D. Stitzel. A finite element-based injury metric for pulmonary contusion: investigation of candidate metrics through correlation with computed tomography. Stapp Car Crash J. 51:189–209, 2007. 24 Gayzik, F. S., J. J. Hoth, and J. D. Stitzel. Finite elementbased injury metrics for pulmonary contusion via concurrent model optimization. Biomech. Model. Mechanobiol. 10(4):505–520, 2011. 25 Gayzik, F. S., R. S. Martin, H. C. Gabler, J. J. Hoth, S. M. Duma, J. W. Meredith, and J. D. Stitzel. Characterization of crash-induced thoracic loading resulting in pulmonary contusion. J. Trauma 66(3):840–849, 2009. 26 Gordon, C. C., T. Churchill, C. E. Clauser, B. Bradtmiller, J. T. McConville, I. Tebbetts, and R. A. Walker. 1988 Anthropometric survey of US Army personnel: methods and summary statistics. Technical Report NATICK/TR89/044. US Army Natick Research, Development, and Engineering Center, Natick, MA, 1989. 27 Gray, H., and D. C. Carmine. Anatomy of the Human Body. Philadelphia: Lea & Febiger, 1985. 28 Hedenstierna, S., and P. Halldin. How does a threedimensional continuum muscle model affect the kinematics and muscle strains of a finite element neck model compared
2582
GAYZIK et al.
to a discrete muscle model in rear-end, frontal, and lateral impacts. Spine 33(8):E236–E245, 2008. 29 Ho¨gler, W., C. J. R. Blimkie, C. T. Cowell, A. F. Kemp, J. Briody, P. Wiebe, N. Farpour-Lambert, C. S. Duncan, and H. J. Woodhead. A comparison of bone geometry and cortical density at the mid-femur between prepuberty and young adulthood using magnetic resonance imaging. Bone 33(5):771–778, 2003. 30 Holzbaur, K. R., W. M. Murray, G. E. Gold, and S. L. Delp. Upper limb muscle volumes in adult subjects. J. Biomech. 40(4):742–749, 2007. 31 Iwamoto, M., Y. Kisanuki, I. Wantanabe, K. Furusu, K. Miki, and J. Hasegawa. Development of a finite element model of the Total Human Model for Safety (THUMS) and application to injury reconstruction. Proceedings of the International Research Council on the Biomechanics of Injury (IRCOBI). Sept. 18–20, 2002. Munich, Germany, pp. 1–12. 32 Kamibayashi, L. K., and F. J. Richmond. Morphometry of human neck muscles. Spine 23(12):1314–1323, 1998. 33 Kemper, A. R., C. McNally, E. A. Kennedy, S. J. Manoogian, A. L. Rath, T. P. Ng, J. D. Stitzel, E. P. Smith, S. M. Duma, and F. Matsuoka. Material properties of human rib cortical bone from dynamic tension coupon testing. Stapp Car Crash J. 49:199–230, 2005. 34 Kimpara, H., J. B. Lee, K. H. Yang, A. I. King, M. Iwamoto, I. Watanabe, and K. Miki. Development of a three-dimensional finite element chest model for the 5(th) percentile female. Stapp Car Crash J. 49:251–269, 2005. 35 Knaub, K., C. Van Ee, C. Cheng, B. Poon, C. Spritzer, and B. S. Myers. Measurement of Human Neck Muscle Volume Geometry and Physiologic Cross-Sectional Area in 5th, 50th, and 95th Percentile Subjects Using Cadaveric Dissection and MRI. National Highway Traffic Safety Administration, NHTSA-98-3588-34, 1999. 36 Li, Z., M. W. Kindig, J. R. Kerrigan, C. D. Untaroiu, D. Subit, J. R. Crandall, and R. W. Kent. Rib fractures under anterior-posterior dynamic loads: experimental and finite-element study. J. Biomech. 43:228–234, 2010. 37 Lieber, R. L. Skeletal Muscle Structure, Function, & Plasticity: The Physiological Basis of Rehabilitation. Baltimore: Lippincott Williams and Wilkins, 2002. 38 Gladman, B. LS-Dyna Keyword User’s Manual v. 971. Livermore, CA: Livermore Software Technology Corporation, 1997. 39 Manary, M., C. Flannagan, M. Reed, and L. Schneider. Human Subject Testing in Support of ASPECT, SAE Technical Paper 1999-01-0960. doi:10.4271/1999-01-0960, 1999. 40 Manoogian, S. J., E. A. Kennedy, K. A. Wilson, S. M. Duma, and N. M. Alem. Predicting neck injuries due to head-supported mass. Aviat. Space Environ. Med. 77(5):509–514, 2006. 41 Maurel, N., A. Diop, and J. Grimberg. A 3D finite element model of an implanted scapula: importance of a multiparametric validation using experimental data. J. Biomech. 38(9):1865–1872, 2005. 42 McConville, J. T., T. D. Churchill, I. Kaleps, C. E. Clauser, and J. Cuzzi. Anthropometric relationships of body and body segment moments of inertia. Technical Report AFAMRL-TR-80-119, Aerospace Medical Research Laboratory, Wright–Patterson Air Force Base, Dayton, Ohio, 1980. 43 Mertz, H. Anthropomorphic test devices. In: Accidental Injury, Biomechanics and Prevention, edited by A. Nahum, and J. Melvin. New York: Springer, 1993, pp. 66–84.
44
Moorcroft, D. M., J. D. Stitzel, G. G. Duma, and S. M. Duma. Computational model of the pregnant occupant: predicting the risk of injury in automobile crashes. Am. J. Obstet. Gynecol. 189(2):540–544, 2003. 45 Moorcroft, D. M., J. D. Stitzel, S. M. Duma, and G. G. Duma. The effects of uterine ligaments on the fetal injury risk in frontal automobile crashes. Proc. Inst. Mech. Eng. D J. Automob. Eng. 217(D):1049–1055, 2003. 46 Moss, S., Z. Wang, M. Salloum, M. Reed, M. van Ratingen, D. Cesari, R. Scherer, T. Uchimura, and M. Beusenberg. Anthropometry for WorldSID a world-harmonized midsize male side impact crash dummy. Proceedings of the SAE Government Industry Meeting. Washington, DC, June 19th–21st, pp. 1–11, 2000. 47 Myoung, H., Y. Y. Kim, M. S. Heo, S. S. Lee, S. C. Choi, and M. J. Kim. Comparative radiologic study of bone density and cortical thickness of donor bone used in mandibular reconstruction. Oral Surg. Oral Med. Oral Pathol. Oral Radiol. Endod. 92(1):23–29, 2001. 48 Netter, F. Atlas of Human Anatomy, 4th ed. Philadelphia: Saunders/Elsevier, 2006. 49 Ogden, C. L., C. D. Fryar, M. D. Carroll, and K. M. Flegal. Mean body weight, height, and body mass index, United States 1960–2002. Advance data from vital and health statistics; no. 347, 1–18. Hyattsville, Maryland: National Center for Health Statistics, 2004. 50 Peretz, A. M., J. A. Hipp, and M. H. Heggeness. The internal bony architecture of the sacrum. Spine 23(9):971– 974, 1998. 51 Power, E. D., S. M. Duma, J. D. Stitzel, I. P. Herring, R. L. West, C. R. Bass, J. S. Crowley, and F. T. Brozoski. Computer modeling of airbag-induced ocular injury in pilots wearing night vision goggles. Aviat. Space Environ. Med. 73(10):1000–1006, 2002. 52 Prevrhal, S., K. Engelke, and W. A. Kalender. Accuracy limits for the determination of cortical width and density: the influence of object size and CT imaging parameters. Phys. Med. Biol. 44(3):751–764, 1999. 53 Robbins, D. Anthropometric specifications for mid-sized male dummy. In: Anthropometry of Motor Vehicle Occupants, Vol. 2. US Department of Transportation, National Highway Traffic Safety Administration, Washington, DC, UMTRI-83-53-2, 1983. 54 Robin, S. Human Model for Safety—A joint effort towards the development of redefined human-like car-occupant models. In: Proceedings of the 17th International Technical Conference for the Enhanced Safety of Vehicles, 4–7 June 2001, Amsterdam, pp. 297–306, 2001. 55 Ruan, J., R. El-Jawahri, L. Chai, S. Barbat, and P. Prasad. Prediction and analysis of human thoracic impact responses and injuries in cadaver impacts using a full human body finite element model. Stapp Car Crash J. 47:299–321, 2003. 56 Ruan, J. S., R. El-Jawahri, S. Barbat, and P. Prasad. Biomechanical analysis of human abdominal impact responses and injuries through finite element simulations of a full human body model. Stapp Car Crash J. 49:343–366, 2005. 57 Scherf, H., and R. Tilgner. A new high-resolution computed tomography (CT) segmentation method for trabecular bone architectural analysis. Am. J. Phys. Anthropol. 140(1):39–51, 2009. 58 Schneider, L., D. Robbins, M. Pflug, and R. Snyder. Development of anthropometrically based design specifications for an advanced adult anthropometric dummy
Development of a Full Body CAD Dataset for Computational Modeling family. In: Anthropometry of Motor Vehicle Occupants, Vol. 1. US Department of Transportation, National Highway Traffic Safety Administration, Washington, DC. UMTRI-83-53-1, 1983. 59 Seki, S., and H. Iwamoto. Disruptive forces for swine heart, liver, and spleen: their breaking stresses. J. Trauma 45(6):1079–1083, 1998. 60 Shigeta, K., Y. Kitagawa, and T. Yasuki. Development of next generation human FE model capable of organ injury prediction. Procedings of the 21st International Technical Conference on the Enhanced Safety of Vehicles. Stuttgart, Germany, June 15th–18th, 09-0111, 2009. 61 Spitzer, V., M. Ackerman, A. Scherzinger, and D. Whitlock. The visible human male: a technical report. J. Am. Med. Inform. Assoc. 3:118–130, 1996. 62 Stitzel, J. D., J. T. Barretta, and S. M. Duma. Predicting fractures due to blunt impact: a sensitivity analysis of the effects of altering failure strain of human rib cortical bone. Int. J. Crashworthiness 9(6):633–642, 2004. 63 Stitzel, J. D., G. A. Hansen, I. P. Herring, and S. M. Duma. Blunt trauma of the aging eye: injury mechanisms and increasing lens stiffness. Arch. Ophthalmol. 123(6):789–794, 2005.
64
2583
Virtama, P., and T. Helela. Radiographic measurements of cortical bone: Variations in a normal population between 1 and 90 years of age. Acta Radiol Suplementum 293:1–268, 1969. 65 Weaver, A. A., E. A. Kennedy, S. M. Duma, and J. D. Stitzel. Evaluation of different projectiles in matched experimental eye impact simulations. J. Biomech. Eng. 133(3):031002, 2011. 66 Weaver, A. A., K. L. Loftis, S. M. Duma, and J. D. Stitzel. Biomechanical modeling of eye trauma for different orbit anthropometries. J. Biomech. 44(7):1296–1303, 2011. 67 Weaver, A. A., K. L. Loftis, J. C. Tan, S. M. Duma, and J. D. Stitzel. CT based three-dimensional measurement of orbit and eye anthropometry. Invest. Ophthalmol. Vis. Sci. 51(10):4892–4897, 2010. 68 Xia, Y., W. Lin, and Y. X. Qin. The influence of cortical endplate on broadband ultrasound attenuation measurements at the human calcaneus using scanning confocal ultrasound. J. Acoust. Soc. Am. 118(3 Pt 1):1801–1807, 2005. 69 Yang, K. H., J. Hu, N. A. White, A. I. King, C. C. Chou, and P. Prasad. Development of numerical models for injury biomechanics research: a review of 50 years of publications in the Stapp Car Crash Conference. Stapp Car Crash J. 50:429–490, 2006.