An Anatomically Representative Atlas-Based Finite Element Model … · 2017. 8. 24. · An...
Transcript of An Anatomically Representative Atlas-Based Finite Element Model … · 2017. 8. 24. · An...
An Anatomically Representative Atlas-Based Finite Element Model of
the Human Brain for Studying Brian Injury Biomechanics:
Development, Validation and Comparison to Existing Models
By
Logan E. Miller
A Thesis Submitted to the Graduate Faculty of
VIRGINIA TECH – WAKE FOREST UNIVERSITY
SCHOOL OF BIOMEDICAL ENGINEERING & SCIENCES
In Partial Fulfillment of the Requirements
for the Degree of
MASTER OF SCIENCE
Biomedical Engineering
August 2015
Winston-Salem, North Carolina
Approved by:
Joel D. Stitzel, PhD, Advisor, Chair
Examining Committee:
F. Scott Gayzik, PhD
Ashley A. Weaver, PhD
ii
ACKNOWLEDGEMENTS
First, I would like to thank my advisor, Dr. Joel Stitzel for your guidance and support
throughout my graduate school experience. Thank you for your motivation, encouragement and
for creating an environment that fosters academic growth and exploration. I value your example
as a mentor, teacher, and researcher and I feel fortunate for the opportunity to learn from you.
I would also like to thank Jill Urban. Your feedback and advice have truly enriched my
experience at Wake over the past two years. I attribute much of my academic development to
your guidance and am extremely grateful. Also, I want to thank Dr. Ashley Weaver and Dr. Scott
Gayzik for serving on my committee.
I am thankful for all those who have inspired me throughout my academic career. I thank
Dr. Jake Socha, who introduced me to research as an undergraduate student. I am grateful for
your encouragement, advice and enthusiasm for science. Your example as a scientist and mentor
was instrumental in my decision to pursue a career in research. Thank you to Pat McGrath, my
high school calculus teacher, who shared my love for math and stimulated my love for learning
and pursuit of knowledge. Also, thank you to Adam Golman for being an excellent mentor when I
was a summer intern at Wake Forest. Your willingness to teach and your advice throughout that
summer were crucial to the satisfaction I felt after that internship and to my desire to return to the
lab as a graduate student.
I also would like to thank my peers in the Center for Injury Biomechanics for your
support and collaboration. I am grateful for the opportunity to work with and learn from everyone
in the CIB.
Finally, I’d like to thank my family for their endless support. Thank you to my parents for
inspiring a love of science and providing me with a wonderful education. And thank you to my
brother for being my first friend and remaining a wonderful one today.
iii
TABLE OF CONTENTS
Acknowledgements ........................................................................................................... ii
Table of Contents ............................................................................................................. iii
List of Tables .................................................................................................................... vi
List of Figures .................................................................................................................. vii
CHAPTER I: INTRODUCTION & BACKGROUND ................................................. 1
Background and Significance .......................................................................................... 1
Anatomy of the Human Head .......................................................................................... 3 Scalp ........................................................................................................................................... 3 Skull ........................................................................................................................................... 3 Meninges .................................................................................................................................... 4 Cerebrospinal fluid (CSF) ........................................................................................................ 5 Brain .......................................................................................................................................... 6 Cerebral Vasculature ............................................................................................................... 8
Mechanical Properties of Brain Tissue ........................................................................... 8 Creep Tests ................................................................................................................................ 9 Relaxation Tests ........................................................................................................................ 9 Shear Relaxation Tests ........................................................................................................... 10 Dynamic Mechanical Analysis ............................................................................................... 11
Mechanical Response to Impact .................................................................................... 12 Cadaver data ........................................................................................................................... 13 Animal data ............................................................................................................................. 14 Volunteer data ......................................................................................................................... 14
Biomechanics of Brain Injury ........................................................................................ 16 Positive Pressure ..................................................................................................................... 16 Negative Pressure.................................................................................................................... 16 Pressure Gradients ................................................................................................................. 17 Rotation ................................................................................................................................... 17
Head Injury Criteria....................................................................................................... 18 Gadd Severity Index (GSI) .................................................................................................... 18 Head Injury Criterion (HIC) ................................................................................................. 19 Head Impact Power (HIP) ..................................................................................................... 19 Brain Rotational Injury Criterion (BRIC) ........................................................................... 20
Finite Element Modeling of the Human Head ............................................................. 20 Model Geometry ..................................................................................................................... 20 Constitutive Models for Brain Material ............................................................................... 22
iv
Brain-Skull Interface .............................................................................................................. 23 Validation ................................................................................................................................ 24 Review of Finite Element Models of the Human Head ....................................................... 25
Chapter Summaries ........................................................................................................ 33 Chapter II: Development and Validation of an Atlas-Based Finite Element Brain Model
.................................................................................................................................................. 33 Chapter III: Comparison of Validation Data for Finite Element Models of the Human
Head ......................................................................................................................................... 33 Chapter IV: Summary of Research ...................................................................................... 33
References ........................................................................................................................ 34
CHAPTER II: DEVELOPMENT AND VALIDATION OF AN ATLAS-BASED
FINITE ELEMENT BRAIN MODEL .......................................................................... 41
Abstract ............................................................................................................................ 42
1 Introduction .................................................................................................................. 43
2 Methods ......................................................................................................................... 45 2.1 Brain Material Optimization ........................................................................................... 48 2.2 Experimental Tests ........................................................................................................... 49 2.3 Model Performance .......................................................................................................... 51 2.4 Falx Material Optimization ............................................................................................. 52 2.5 Data Analysis ..................................................................................................................... 52
3 Results ........................................................................................................................... 53 3.1 C755-T2 ............................................................................................................................. 55 3.2 C383-T1 ............................................................................................................................. 55 3.3 C291-T1 ............................................................................................................................. 56
4 Discussion...................................................................................................................... 58 4.1 Limitations ........................................................................................................................ 64 4.2 Future Work...................................................................................................................... 65
5 Conclusion .................................................................................................................... 66
6 Acknowledgements ...................................................................................................... 68
References ........................................................................................................................ 69
CHAPTER III: COMPARISON OF VALIDATION DATA FOR FINITE
ELEMENT MODELS OF THE HUMAN HEAD ....................................................... 72
Abstract ............................................................................................................................ 73
v
1 Introduction .................................................................................................................. 74
2 Methods ......................................................................................................................... 78 2.1 FEM Description and Comparison ................................................................................. 79 2.2 Validation against localized brain displacements .......................................................... 85 2.3 Model Performance .......................................................................................................... 88
3 Results ........................................................................................................................... 89
4 Discussion...................................................................................................................... 98 4.1 C755-T2 Occipital Impact ................................................................................................ 99 4.2 C383-T1 Frontal Impact ................................................................................................ 101 4.3 C291-T1 Parietal Impact ................................................................................................ 102 4.4 Overall Performance ...................................................................................................... 105
5 Conclusion .................................................................................................................. 106
6 Acknowledgements .................................................................................................... 108
References ...................................................................................................................... 109
Appendix. Model Development................................................................................ 10913
CHAPTER IV: SUMMARY OF RESEARCH .......................................................... 115
Scholastic Vita ............................................................................................................... 116
vi
LIST OF TABLES
Table 1. Material Properties of the ABM ......................................................................... 47
Table 2. Parameter Ranges for Latin Hypercube Sample ................................................. 48
Table 3. Summary of Experimental Test Conditions........................................................ 50
Table 4. Material Properties from each experimental configuration optimization ........... 54
Table 5. Comparison of FEMs .......................................................................................... 85
Table 6. CORA Scores for each model for all experimental configurations .................... 98
Table 7. Publication plan for research outlined in this thesis ......................................... 115
vii
LIST OF FIGURES
Figure 1. Percentage of annual TBI (combined emergency department visits,
hospitalizations, and deaths) by external cause [2]. ............................................................ 1
Figure 2. (A) External view of the human skull [9] and (B) the base of the cranial cavity
[10]. ..................................................................................................................................... 4
Figure 3. View inside the skull showing the protruding portions of the dura mater - falx
cerebri and tentorium cerebelli [9]. ..................................................................................... 5
Figure 4. The meningeal membranes in a coronal cross section ........................................ 5
Figure 5. T2-weighted MRI scan of a normal head. CSF is depicted as light gray, gray
matter is darker gray, and white matter is darkest. ............................................................. 6
Figure 6. (A) Internal structures of the human brain and (B) the four lobes of the
cerebrum. ............................................................................................................................ 7
Figure 7. Corpus callosum [9]. ........................................................................................... 7
Figure 8. Distribution of cerebral arteries on the base of the brain [9]. .............................. 8
Figure 9. In vitro relaxation moduli versus time from the literature [21]. ........................ 10
Figure 10. (A) Storage and (B) loss moduli of brain tissue from the literature of in vitro
dynamic frequency sweep tests in shear [21]. .................................................................. 11
Figure 11. Comparison of thresholds determined from various animal data [39, 40, 41]. 14
Figure 12. Combined probability of concussion contours relating concussion risk to linear
and rotational head acceleration [42]. ............................................................................... 15
Figure 13. Equivalent stress fields for models with varying geometries under the same
boundary conditions showing stress concentrations in the heterogeneous models [61]. .. 21
viii
Figure 14. (A) Direction of impact in experiment conducted by Nahum et al. [31] and
(B) locations of pressure measurements. .......................................................................... 24
Figure 15. (A) Skull model, (B) brain model, (C) section view of the KTH head and neck
model [65], and (D) cross section showing FE mesh and anatomical structures
represented in the KTH model [65]. ................................................................................. 26
Figure 16. Various versions of the WSUHIM. (A) The original model developed by Ruan
et al. [67], (B) the improved model with gray and white matter differentiation [70], and
(C) the current version of the model [69]. ........................................................................ 27
Figure 17. Overview of the head model developed by TUE [72]. .................................... 28
Figure 18. (A) Isometric view of the THUMS v4.01 head model with the brain exposed
and (B) midsagittal view of the FEM. .............................................................................. 29
Figure 19. (A) Isometric view of the GHBMC head model with the brain exposed and (B)
midsagittal view of the FEM [74]. .................................................................................... 30
Figure 20. Overview of the ULP head model [77]. .......................................................... 30
Figure 21. (A) Skull, (B), brain, (C), and sagittal cross section of the UCDBTM. .......... 31
Figure 22. (A) Isometric view of the SIMon FEM with exposed brain and (B) isometric
view of the FEM showing various anatomical features represented in the model. .......... 32
Figure 23. ABM showing the falx, tentorium, and ventricles .......................................... 47
Figure 24. Cross sections of ABM. (a) axial, (b) sagittal, (c) coronal, (d) axial isometric
........................................................................................................................................... 47
Figure 25. Approximate cross-sectional planes where NDT columns are located for each
experimental test (a), locations of NDTs in the XZ plane on the right side of the brain (b),
and locations of NDTs in the XZ plane on the left side of the brain (c) ........................... 49
ix
Figure 26. Relationships between CORA and brain material parameters from brain
parameter variations in three experimental configurations (using the original falx model
in all variations). The simulation with the highest CORA value for each individual impact
condition is highlighted..................................................................................................... 54
Figure 27. CORA relationships with falx pretension (a) and falx thickness (b) for the
C291-T1 impact ................................................................................................................ 55
Figure 28. Relationship between CORA and shear modulus, G0, for material
optimizations in occipital (a), frontal (b), and parietal (using optimized falx model) (c)
impact configurations. The simulation representing the optimal material model for each
condition optimized individually is shown (C755*, C383*, C291*), as well as optimized
concurrently (ALL*) ......................................................................................................... 56
Figure 29. Displacement results and individual CORA scores for the optimized material
properties for Hardy C755-T2 occipital impact for anterior (A) and posterior (P) NDT
columns ............................................................................................................................. 57
Figure 30. Displacement results and individual CORA scores for the optimized material
properties for Hardy C383-T1 frontal impact for anterior (A) and posterior (P) NDT
columns ............................................................................................................................. 57
Figure 31. Displacement results and individual CORA scores for the optimized material
properties for Hardy C291-T1 lateral impact for the anterior (A) and posterior (P) right
(R) and left (L) columns ................................................................................................... 58
Figure 32. (A) Sagittal and (B) coronal cross sections of the FEM developed by Hosey
and Liu [9] in 1981. .......................................................................................................... 75
Figure 33. Relaxation moduli versus time from the literature [15]. ................................. 76
x
Figure 34. (A) ABM showing the falx, tentorium, and ventricles, and (B) axial, (C)
sagittal, (E) coronal, (D) isometric cross sections. ........................................................... 79
Figure 35. (A) Isometric view of the SIMon FEM with exposed brain and (B) isometric
view of the FEM showing various anatomical features represented in the model. .......... 80
Figure 36. (A) Isometric view of the GHBMC head model with the brain exposed and (B)
midsagittal view of the FEM [36]. .................................................................................... 81
Figure 37. (A) Isometric view of the THUMS v4.01 head model with the brain exposed
and (B) midsagittal view of the FEM. .............................................................................. 82
Figure 38. (A) Section view of the KTH head and neck model [28], and (B) cross section
showing FE mesh and anatomical structures represented in the model [12]. ................... 83
Figure 39. (A) Isometric view of the DHIM FEM and (B) isometric view of the FEM
showing various anatomical features represented in the model [48]. ............................... 84
Figure 40. Velocity load curves for the three experimental configurations. .................... 87
Figure 41. Approximate NDT locations in ABM, SIMon, GHBMC, and THUMS models
for C755-T2 occipital impact (A), C383-T1 frontal impact (B) and C291-T1 parietal
impact (C). ........................................................................................................................ 88
Figure 42. Linear velocity at head CG for ABM, SIMon, GHBMC, and THUMS
simulations. ....................................................................................................................... 90
Figure 43. Local displacement results for Hardy C755-T2 occipital impact for the anterior
NDT column. .................................................................................................................... 91
Figure 44. Local displacement results for Hardy C755-T2 occipital impact for the
posterior NDT column. ..................................................................................................... 92
xi
Figure 45. Local displacement results for Hardy C383-T1 frontal impact for the anterior
NDT column. .................................................................................................................... 93
Figure 46. Local displacement results for Hardy C383-T1 frontal impact for the posterior
NDT column. .................................................................................................................... 94
Figure 47. Local displacement results for Hardy C291-T1 parietal impact for the right
posterior NDT column. ..................................................................................................... 95
Figure 48. Local displacement results for Hardy C291-T1 parietal impact for the left
posterior NDT column. ..................................................................................................... 96
Figure 49. Local displacement results for Hardy C291-T1 parietal impact for the left
anterior NDT column. ....................................................................................................... 97
1
Chapter I: Introduction & Background
BACKGROUND AND SIGNIFICANCE
Traumatic brain injury (TBI) is the most severe and most common trauma-related
injury [1]. According to the Centers for Disease Control and Prevention (CDC), it is a
major public health concern in the United States, with approximately 1.7 million people
in the US suffering a TBI annually [2]. TBI has a mortality rate ranging from 6% to
23.1% and is also a leading cause of disability, which introduces significant societal and
economic costs [3]. The leading causes of TBI are falls, motor vehicle crashes (MVCs)
and struck by or against events (Figure 1) [2].
Figure 1. Percentage of annual TBI (combined emergency department visits,
hospitalizations, and deaths) by external cause [2].
2
Mild traumatic brain injury (mTBI), also called concussion, minor brain injury or
minor TBI, is an important subset of TBI that is of increasing research interest –
particularly sports-related concussion. A study conducted by the CDC estimates that
approximately 300,000 sports-related mTBIs occur each year, although the study
included only those injuries involving loss of consciousness [4]. This estimate of mTBI
incidence is likely severely underestimated due to underreporting and unrecognized
injuries. Other studies suggest that loss of consciousness may only occur for between 8%
and 19.2% of sports-related mTBIs [5];[6]. Combining this rate of occurrence with the
CDC estimate of mTBI incidence, it may be more accurate to assume that 1.6 to 3.8
million sports-related mTBIs occur each year, including those for which medical
attention is not sought.
Additionally, there have been growing concerns about the long-term effects of
exposure to repetitive, sub-concussive head impacts. Chronic traumatic encephalopathy
(CTE) is a neurodegenerative disease characterized by the deposition of tau proteins at
the depths of the sulci thought to be associated with repetitive head impacts, such as those
sustained in contact sports [7].
Despite the high incidence of TBI and mTBI and the growing concerns about
long-term consequences of such injuries, the underlying injury mechanisms are not well
understood. The fundamental biomechanics and injury thresholds associated with TBI
must be well-characterized and understood to allow us to prevent and treat these injuries.
While there are many theories about the underlying injury mechanisms of various brain
injuries, we still do not fully understand the biomechanical basis for many of these
injuries. Various methods have been used over the years to investigate and gain a deeper
3
understanding of brain injury mechanisms, including experimental testing of animal
models, post-mortem human subjects (PMHS), anthropomorphic test devices (ATDs), as
well as computational modeling of the brain [8].
ANATOMY OF THE HUMAN HEAD
The human head is made up of the scalp, skull, meningeal membranes,
cerebrospinal fluid (CSF), brain and cerebral vasculature. The shape and size of the
different tissues varies greatly between individuals, but they share many common
features.
Scalp
The scalp is defined as soft tissue layers of dermal skin and subcutaneous
connective tissues covering the top of the human head. It is 5 to 7 mm thick and consists
of the skin, subcutaneous connective tissue layer, galea aponeurotica, loose areolar
connective tissue layer and pericranium.
Skull
The skull rests at the superior end of the vertebral column and forms the cavity for
the brain. It is formed by two sets of bones; the 8 cranial bones and the 14 facial bones
(Figure 2A). The thickness of the skull varies between 4 and 7 mm and consists of three
4
Figure 2. (A) External view of the human skull [9] and (B) the base of the cranial cavity
[10].
layers: inner table, diploë, and outer table. The inner and outer tables are made up of
compact bone, while the diploë layer is spongy bone. The base of the cranial cavity is an
irregular surface of bone containing depressions and ridges as well as holes (foramen):
small holes for arteries, veins and nerves; and a large hole (foramen magnum), which is
the area where the brainstem transitions to the spinal cord. The skull base is made up of
the anterior cranial fossa, which houses the frontal lobes; the middle cranial fossa, which
houses the temporal lobes; and the posterior cranial fossa, which contains the cerebellum
(Figure 2B).
Meninges
Three membranes called the meninges surround the brain. The most superficial
membrane is the dura mater, which is a tough fibrous membrane that is adhered to the
inner surface of the skull. This membrane extends into the falx cerebri, which separates
the cerebral hemispheres and the tentorium cerebelli, which separates the cerebrum from
the cerebellum (Figure 3). Beneath the dura is the arachnoid, a thin and fibrous layer that
encloses the the CSF and blood vessels. The CSF is located in the subarachnoid space,
5
Figure 3. View inside the skull showing the protruding portions of the dura mater - falx
cerebri and tentorium cerebelli [9].
which is the area between the arachnoid and the third meningeal layer, the pia mater
(Figure 4). The pia mater is the innermost layer and is a thin, delicate membrane that is
closely adhered to the surface of the brain and follows the sulci (depressions in the
cerebral cortex) and gyri (ridges on the cerebral cortex).
Figure 4. The meningeal membranes in a coronal cross section
Cerebrospinal fluid (CSF)
CSF is a liquid that surrounds the brain between the pia and arachnoid meningeal
layers and maintains a uniform pressure within the cranium (Figure 5). It contains
6
sodium, chloride, glucose, proteins, lipids, and leukocytes, and allows the circulation of
nutrients and waste. Additionally, CSF serves as mechanical protection to the brain,
which would otherwise be exposed to hitting the bony walls of the cranium, while still
allowing relative motion between the brain and skull [11].
Figure 5. T2-weighted MRI scan of a normal head. CSF is depicted as light gray, gray
matter is darker gray, and white matter is darkest.
Brain
The average adult human head weights approximately 4.50–5.00 kg – one-third of
this mass is made up of the brain (1.60 kg in mean, 1.45 kg in women) [12]. The brain
consists of the cerebrum, cerebellum and brain stem (Figure 6A). The cerebrum is largest
part of the brain and consists of two hemispheres, which are each broken up into four
lobes: the frontal, temporal, parietal and occipital (Figure 6B). The outer layer of the
cerebrum is made up of gray matter (cerebral cortex), which consists of neuronal cell
bodies, synapses and glia. Immediately beneath the cerebral cortex, is white matter,
which consists of glial cells and myelinated axons. Due to the myelinated fiber, which
matter demonstrates directional dependence in its material properties. An important
7
Figure 6. (A) Internal structures of the human brain and (B) the four lobes of the cerebrum.
portion of the white matter is the corpus callosum, which connects the two hemispheres
of the cerebrum (Figure 7). The cerebellum is located inferior to the cerebrum in the
posterior cranial fossa. The cortex of the cerebellum is covered in gyri and sulci similar to
the cerebrum, but they are much smaller (Figure 6A). The brain stem, which originates at
the posterior part of the brain, is the lower extension of the brain and structurally joins the
brain to the spinal cord.
Figure 7. Corpus callosum [9].
8
Cerebral Vasculature
The cerebral vasculature consists of arteries that provide the brain with
oxygenated blood and veins that drain the circulated blood (Figure 8). Major arteries in
the cranial cavity include the vertebral, internal carotid, anterior cerebral, posterior
cerebral and inferior cerebellar arteries; major veins include the superior sagittal, inferior
sagittal, straight, cavernous, transverse, and sigmoid sinuses. The superior cerebral veins
communicate with the superior sagittal sinus through intermediate collection reservoirs
called the lateral venous lacunae. In the parasagittal region, these bridging veins are
thought to be associated with acute subdural hematoma (SDH) when compromised.
Figure 8. Distribution of cerebral arteries on the base of the brain [9].
MECHANICAL PROPERTIES OF BRAIN TISSUE
The material properties of brain tissue play a vital role in understanding the
mechanical response of the human head to impact. Researchers have been conducting in
9
vitro experimental testing of brain tissue since the 1960s. Even with over 50 years of
research and experimental data, material property values of the brain remain somewhat
controversial due to the large variation in reported values throughout the literature. In
some cases, the discrepancies can be attributed to methodological flaws, such as post
mortem testing time, tissue storage and hydration conditions, tissue preparation/excision
methods, and interfacial testing conditions [13]. Variation also results from the designed
testing conditions; specifically, many investigators have focused their efforts on
characterizing the tissue response under a specific type of loading, such as compression,
tension or shear, or a specific loading condition, such as ramp, hold or oscillation. Most
in vitro tests described in the literature were based on compression creep tests and
relaxation tests in compression and shear, or on dynamic frequency or strain sweep
experiments and are summarized in the following section.
Creep Tests
The first in vitro mechanical testing conducted on brain tissue considered
compression creep tests and used animal brain samples [14, 15]. Dodgson [14] performed
creep tests on mouse brain tissue and identified the quasi-linear relation between
logarithmic scaled time and compression strain. This quasi-linear behavior was later
confirmed by Koeneman [15] in 1966 who performed compression creep tests on
porcine, calf and rabbit brain samples.
Relaxation Tests
Compression stress relaxation tests were first conducted by Galford and
McElhaney [16] in 1970 on monkey and human brain samples. Their results showed a
quick relaxation for less than 100 s with an initial relaxation modulus of 6.65 kPa for
10
human and 10.3 kPa for monkey. This decrease in stress relaxation modulus has been
confirmed by more recent relaxation tests [36, 38, 39].
Shear Relaxation Tests
Shear relaxation tests were performed on porcine and bovine brain samples at
varying strain levels [40, 41]. The goal of these early studies was to estimate the limit of
linear viscoelastic response of brain tissue. A significant increase in shear relaxation
modulus was found below the 0.1% strain level, with a long time (t > 100 s) relaxation
modulus was determined to be less than 100 Pa. Since then, many additional shear
relaxation tests have been performed with a large range in results for the shear modulus
value. Figure 9 summarizes the relaxation moduli obtained through shear relaxation
testing, and superposes relaxation moduli determined from compressive relaxation testing
by assuming incompressibility. From Figure 9, it can be seen that relaxation was found to
be slower at higher strains – a behavior that was observed by all investigators.
Figure 9. In vitro relaxation moduli versus time from the literature [21].
11
Dynamic Mechanical Analysis
The first in vitro shear dynamic tests were performed by Fallenstein et al.
[22] at frequencies ranging from 9 to 10 Hz. Shortly after, in vitro studies of shear under
dynamic torsion were conducted by Shuck and Advani [23], who investigated frequencies
Figure 10. (A) Storage and (B) loss moduli of brain tissue from the literature of in vitro
dynamic frequency sweep tests in shear [21].
ranging from 5 to 350 Hz. The high frequency range is of particular importance when
characterizing brain tissue - it has been suggested that brain tissue should be
characterized over a range up to 1000 Hz [21]. To this end, a large range of frequencies
has been investigated by various researchers through oscillatory testing and the results are
presented as storage and loss moduli in Figure 10A and B, respectively. Although there is
a disparity in the results, the trend of shear moduli increasing with increasing frequency
12
exists for all sources. An interesting aspect of these results is the inflection point
occurring around 150 Hz that was observed by Shuck and Advani [23] and Thibault and
Margulies [24] and at more than 1000 Hz by Nicolle et al. [25]. This inflection point
characterizes an inversion of the elastic and viscous components of the response. In
general, the relationship between modulus and frequency demonstrates two major trends.
The first trend encompasses the majority of the studies reported and results in modulus
values ranging from 100 to 1000 Pa. Nicolle et al. [26] and Shuck and Advani [23],
however, measured values about 10 times higher.
MECHANICAL RESPONSE TO IMPACT
Various methods have been used over the years to investigate and characterize the
mechanical response of the human head to impact, including experimental testing of
animal models, cadavers, and human volunteers. Cadaver tests allow for well controlled
tests and have the same anatomical features as a living person, but are limited by the lack
of muscular and physiologic response. To account for the physiologic response, animal
models can be used; however, these models require scaling of shape and size to fit the
animal model to humans. Although human volunteers present an opportunity to capture
the physiologic response without the need to scale or extrapolate the data, there are
limitations to this method; that is, volunteer data are generally either obtained in
uncontrolled conditions (such as real-world athletic events) or at impact severities well
below injury level.
13
Cadaver data
Human response to impact was investigated using cadavers by Kallieries et al.
[27] by conducting frontal, lateral and occipital impacts using both a rigid and a padded
impactor. Subarachnoid hematomas were observed at the contrecoup site in 40% of the
experiments. In 1997, Hardy et al. [28] proposed a new technique for studying relative
motion of the brain with respect to the skull using neutral density accelerometers (NDA-
3’s) and neutral density targets (NDTs). Additional experiments using this method were
conducted by Hardy et al. studying a range of impact directions and severities [27, 28]. In
situ tests were conducted by Nahum et al. [31] and Trosseille et al. [32], but this time
looking at the pressure response. Nahum et al. [31] reported experimental data for a
frontal blow to the head for several seated cadavers. Various masses and velocities were
investigated, as well as different materials for padding. Trosseille et al. [32] later
conducted several impacts to the face of cadavers and measured pressure response
similarly to Nahum et al. [31].
Following these early cadaver tests, in situ strain and pressure experimental tests
were conducted [28, 29]. To determine local strain, these experiments used a high-speed
x-ray system to track the relative motion of radio-opaque neutral density targets (NDTs)
implanted in the brain and implanted cranial pressure transducers (CPTs) to measure the
intracranial pressure following head impacts at concussive severities. These experiments
observed an increase in brain pressure and strain as linear and rotational acceleration
increase.
14
Animal data
There were over 200 primate tests performed from 1966-1983, which primarily
looked at rotational acceleration [32, 33, 34]. For example, Gennarelli et al. [33]
produced traumatic coma in 45 monkeys by accelerating the head in one of three
directions without impact. The results of these experiments related the magnitude of head
motion to the duration of coma, degree of impairment, and amount of DAI. Various other
animal models have also been employed over the years, including pigs [36], ferrets [37],
and rats [38]. Experimental data derived from animal experiments can be used to
establish tolerance to various brain injury (Figure 11).
Figure 11. Comparison of thresholds determined from various animal data [39, 40, 41].
Volunteer data
Football and other contact sports present a unique opportunity to ethically study
real-world head impacts by providing a large sample of human volunteers.
0 2500 5000 7500 10000 12500 15000 17500 200000
10
20
30
40
50
60
Rotational Acceleration, rad/s2
Rota
tional V
elo
city,
rad/s
Concussion, Ommaya (1985)
DAI, Davidsson (2009)
DAI, Margulies (1992)
15
Instrumentation of football players with accelerometers has allowed the compilation of a
large biomechanical data set of linear and rotational acceleration. Biomechanical impact
data collected from NFL players, along with injury information (concussion incidence)
has been used to reconstruct impacts using anthropomorphic test dummies (ATDs) to
learn more about how head kinematics are correlated with risk of concussion.
Additionally, this data has been used to investigate head impact exposure as well to
assess concussion risk based on head kinematics. Rowson and Duma [42] proposed a
combined probability risk function for concussion that was developed from a large
dataset of real-world collegiate football impacts and accounted for both linear and
rotational accelerations. Figure 12 shows risk contours of the combined probability risk
function.
Figure 12. Combined probability of concussion contours relating concussion risk to linear
and rotational head acceleration [42].
16
BIOMECHANICS OF BRAIN INJURY
TBI is damage to the brain caused by high-energy mechanical loading, such as
impact, blast loading, or sudden acceleration or deceleration. TBI can be categorized
based on the affected area: diffuse (over widespread area) and focal (localized) injuries.
Diffuse injuries include cerebral concussion, diffuse axonal injury (DAI), and cerebral
edema. Concussion is the most common diffuse injury and results in temporary loss or
alteration in brain function. Focal injuries include hematomas and contusions. Numerous
theories exist about the underlying mechanisms of various brain injuries, including the
development of positive and negative pressures, pressure gradients, and rotational effects.
Positive Pressure
The theory of positive pressure is based on the theory that compressive stresses
develop in the brain at the site of impact due to the skull moving more quickly than the
brain, which causes damage to the brain. This accounts for coup injuries (injury
experienced at site of impact) when the head is impacted and contrecoup injuries (injury
experienced at contralateral site) during a fall.
Negative Pressure
Negative pressure develops in the contrecoup site upon impact as a result of
tension generated by the relative motion between the skull and the response-delayed
brain. Another possibility for the development of negative pressure at the contrecoup site
is due to a tensile wave formed by the reflection of the original compressive stress wave
from the impact site. In contrast, at the coup site, negative pressure develops as a result of
the deformed skull rapidly returning to its original geometry. Denny-Brown and Russell
17
[43] described this mechanism of injury, but state that contrecoup injury is not caused by
negative pressure alone.
Pressure Gradients
The formation of pressure gradients is proposed as another mechanism of brain
injury. It is thought that these gradients create shear stresses that result in local
deformation of brain tissue. Gurdjian and Lissner [44] and Thomas et al. [45] found that
impact-induced acceleration and pressure are responsible for the development of pressure
gradients. These pressure gradients cause injury as the brain attempts to ‘flow’ from areas
of high pressure to low pressure, and the brain experiences regional deformation in
response to the generated shear stresses. Additionally, these gradients can induce
damaging shear effects at the junctions between gray and white matter leading to DAI
[45].
Rotation
The hypothesis that brain injury is based on angular acceleration was introduced
by Holbourn [46] in 1943 states that injury is induced by shear strains generated by
angular acceleration. Major proponents of this theory are Gennarelli et al. [32, 42, 47,
48], whose experimental work focused on the rotational acceleration of subhuman
primates. Gennarelli et al. [47] related levels of concussion to angular acceleration and
showed good agreement with clinical findings. Generalli [48] stated that “virtually all
types of primary head injury can be produced by angular acceleration applied to the head
in the appropriate manner.” In contrast, Ommaya et al. [49] and Ommaya and Hirsch
[50], contend that rotation alone is not able to produce head injury. Additionally, Ono et
18
al. [51] found that the occurrence of concussion in monkeys did not correlate with
angular acceleration of the head.
HEAD INJURY CRITERIA
Mechanically relevant brain responses or parameters, such as accelerations,
intracranial pressure, stresses and strains, can be used as predictors of various brain
injuries. The current standards used by the sports and automotive industries are the Gadd
Severity Index (GSI) [52] and the Head Injury Criterion (HIC) [53], which were both
developed from the Wayne State Tolerance Curve (WSTC). There are opponents of these
indices for various reasons, such as the fact that rotational acceleration is not taken into
account [54], and that they were developed from WSTC data, which only considered
frontal impacts [55]. Taking into account the contribution of rotational acceleration of the
head, later studies proposed additional injury criteria such as Head Impact Power (HIP)
[56] and Brain Injury Criteria (BRIC) [57].
Gadd Severity Index (GSI)
The Gadd Severity Index (GSI), described by Gadd [52], is an extension of the
WSTC, whose basic finding was that high acceleration can be withstood for short
durations, while lower accelerations can be tolerated for longer intervals. The GSI is a
weighted integral of the form:
𝐺𝑆𝐼 = ∫[𝑎(𝑡)]2.5𝑑𝑡
(1)
19
where a is acceleration and t is time. A value of 1000 was suggested as an injury
threshold from analysis of data from Wayne State, the FAA and NASA. This index is not
valid for long durations.
Head Injury Criterion (HIC)
The Head Injury Criterion (HIC) was introduced by Versace [53] and was
eventually adopted by the National Highway Traffic Safety Administration (NHTSA) as
the head injury criterion for Federal Motor Vehicle Safety Standard (FMVSS) 208 [58]. It
has the form:
𝐻𝐼𝐶 = (𝑡2 − 𝑡1)[
1
𝑡2 − 𝑡1∫ 𝑎(𝑡)𝑑𝑡𝑡2
𝑡1
]𝑀𝑎𝑥2.5
(2)
where a is acceleration and t is time. This index is calculated by choosing the range of
integration (t1 – t2) such that the function is maximized. The duration of integration is
either 15 ms, recommended by the International Organization for Standardization (ISO),
or 36 ms, recommended by NHTSA. A tolerance limit of 1000 has been adopted by
NHTSA.
Head Impact Power (HIP)
The Head Impact Power (HIP) was proposed by Newman et al. [56]. It combines
translational and rotational head accelerations and has the form:
𝐻𝐼𝑃 = 4.50𝑎𝑥∫𝑎𝑥𝑑𝑡 + 4.50𝑎𝑦∫𝑎𝑦𝑑𝑡 + 4.50𝑎𝑧∫𝑎𝑧𝑑𝑡
+ 0.016𝛼𝑥∫𝛼𝑥𝑑𝑡 + 0.024𝛼𝑦∫𝛼𝑦𝑑𝑡 + 0.022𝛼𝑧∫𝛼𝑧𝑑𝑡
(3)
where a is acceleration, t is time and the coefficients were set to equal the mass and
appropriate mass moments of inertia for the human head.
20
Brain Rotational Injury Criterion (BRIC)
The Brain Injury Criterion (BRIC) was originally proposed by Takounts et al.
(2011) [59], but the formulation was revised in 2013 [57] to the current form:
𝐵𝑅𝐼𝐶 = √(
𝜔𝑥
𝜔𝑥𝑐)2 + (
𝜔𝑦
𝜔𝑦𝑐)2 + (
𝜔𝑧
𝜔𝑧𝑐)2
(4)
where ωx, ωy, and ωz are maximum angular velocities in the x-, y- and z-axes,
respectively, and ωxc, ωyc, and ωzc are the critical angular velocities in their respective
directions.
FINITE ELEMENT MODELING OF THE HUMAN HEAD
To overcome the limitations presented by other means of studying brain injury,
finite element modeling can be employed. FE modeling offers a cost-effective alternative
to many experimental methods and has the additional advantage of providing
exponentially more data through spatial and temporal distributions of stress and strains
and other biomechanical parameters of interest. The quality of a model’s predictions,
however, is dependent on the accuracy of the modeled geometry, the model’s ability to
describe complex material behavior, and the accuracy of the modeled interface
conditions.
Model Geometry
The FE model development process typically involves image segmentation of
medical images, such as computed tomography (CT) and magnetic resonance imaging
(MRI) to define three-dimensional (3D) surfaces of desired structures, which is then
followed by a surface-based mesh generation process. During this process, complex
21
anatomical features are frequently simplified and ‘smoothed’ to allow faster development
and simple contact interaction at model interfaces. One feature of particular importance
that is over-simplified in many brain FEMs is the gyral folds on the surface of the brain.
The brain surface is typically modeled with a smooth surface surrounded by a uniform
layer of cerebrospinal fluid (CSF), which ignores the intricate gyral folds throughout the
brain. This simplification and reduction in anatomic accuracy may significantly impact
the model’s local brain response and subsequent prediction of injury. Indeed, studies have
indicated that the intricate folding patterns on the surface of the brain greatly affect both
the distribution and magnitude of stress and strain throughout the brain [60]. Cloots et al.
[61] subjected four plane strain models of different geometries (a homogeneous model
and three heterogeneous models) to the same boundary conditions and examined the
equivalent stress and maximum principal strain to quantify the influence of morphologic
heterogeneities (Figure 13).
Figure 13. Equivalent stress fields for models with varying geometries under the same
boundary conditions showing stress concentrations in the heterogeneous models [61].
22
Constitutive Models for Brain Material
Some of the earliest brain FE models characterized brain tissue as linear elastic. A
linear viscoelastic constitutive model was adopted shortly after and better represents the
response of the brain by accounting for time-dependent properties. The shear relaxation
response of the linear viscoelastic model is described by the following expression:
𝐺(𝑡) = 𝐺∞ + (𝐺0 − 𝐺∞)𝑒−𝛽𝑡
(5)
where G∞ is the long term shear modulus, G0 is the short term shear modulus, and β is the
decay constant.
Another common model for brain tissue employed in FE modeling is a
hyperelastic formulation because it is capable of large elastic deformations. In a
hyperelastic material model, the stress is derived from a strain energy potential (W),
which can be written as a function of the principal invariants (I1, I2, I3) of the right
Cauchy-Green deformation tensor. Two hyperelastic constitutive models that are
frequently used are the Mooney-Rivlin model and the Ogden model. To model the brain
as an unconstrained material, a hydrostatic work term, WH(J), can be added to the strain
energy function, which is a function of the relative volume, J. The standard strain energy
function for an incompressible Mooney-Rivlin hyperelastic material is:
𝑊(𝐼1, 𝐼2) = 𝐶10(𝐼1 − 3) + 𝐶01(𝐼2 − 3) +𝑊𝐻(𝐽) (6)
where C10 and C01 are the Mooney-Rivlin material constants, I1 and I2 are principal
invariants, and J is relative volume [62]. Using an Ogden hyperelastic model, the strain
energy function takes the form:
23
𝑊(𝐼1, 𝐼2, 𝐼3) =∑𝜇𝑖𝛼𝑖(𝐼1
𝛼𝑖 + 𝐼2𝛼𝑖 + 𝐼3
𝛼𝑖 − 3) +1
2𝐾(𝐽 − 1)2
3
𝑖
(7)
where αi and μi are Ogden constants, I1, I2, and I3 are principal invariants, K is bulk
modulus and J is relative volume. For these hyperelastic models, the stress tensor is then
derived using:
𝑆𝑖𝑗 =
1
2(𝜕𝑊
𝜕𝐸𝑖𝑗+𝜕𝑊
𝜕𝐸𝑗𝑖)
(8)
in terms of the second Piola-Kirchhoff stress, Sij, and Green’s strain tensor Eij.
Additionally, hyperelastic constitutive models can be employed in combination
with viscoelasticity incorporating rate effects using a convolution integral of the form:
𝑆𝑖𝑗 = ∫𝐺𝑖𝑗𝑘𝑙(𝑡 − 𝜏)𝜕𝐸𝑖𝑗
𝜕𝜏𝑑𝜏
𝑡
0
(9)
where Sij is the second Piola-Kirchhoff stress, Eij is Green’s strain tensor, and Gijkl is the
stress relaxation function represented by a Prony series:
𝑔(𝑡) =∑𝐺𝑖𝑒−𝛽𝑖𝑡
𝑁
𝑖=1
(10)
where Gi are the shear moduli and βi are the decay constants.
Brain-Skull Interface
The boundary conditions at the brain-skull interface have been an important
aspect of FE modeling. Preservation of accurate relative motion between the brain and
skull is important in the simulation of head impact. In general, there are three accepted
methods of modeling the brain-skull interface. The first models the brain-skull interface
with a common set of nodes, which directly couples the brain to the skull. The second
24
method, which has been most common throughout FEMs models the CSF as a ‘soft’ solid
(solid with a low shear modulus). The third method employs various contact algorithms,
such as tied interface, sliding interface with separation, and sliding interface without
separation.
Validation
There are several sets of experimental data that are commonly used to validate FE
head models. These include the intracranial pressure (ICP) experiments conducted by
Nahum et al. [31] and Trosseille et al. [32], as well as the localized brain motion
experiments conducted by Hardy et al. [31, 32, 59]. Nahum et al. [31] impacted a seated
cadavers at the frontal bone (Figure 14A) using impactors with various padding material
and thicknesses. Pressure histories were measured during the impact using pressure
transducers at 5 locations within the brain (Figure 14B). Of this group of tests,
experiment 37 is frequently used for validation. Trosseille et al. [32] conducted six
experiments on seated cadavers and impacted with a 23.4 kg impactor at velocities
Figure 14. (A) Direction of impact in experiment conducted by Nahum et al. [31] and (B)
locations of pressure measurements.
25
varying from 5 to 7 m/s, resulting in longer impact durations than were observed by
Nahum et al. [31]. In the long-duration impacts of Trosseille et al. [32], pressure was
measured at five locations within the brain: frontal lobe, occipital lobe, parietal lobe,
lateral ventricle and third ventricle. Validation is frequently reported against experiment
MS 428-2 from the group of tests conducted by Trosseille et al. [32]. Hardy et al. [63]
conducted a total of sixty-two experiments using seventeen cadavers looking at relative
displacement between the brain and skull using high-speed biplanar x-ray to track the
location of implanted neutral density targets (NDTs) during impact. These tests varied in
setup configuration (impact surface, location and severity), as well the configuration of
NDT implantation (columns or clusters). Two of these experiments that are frequently
used in the validation of head FE models are C755-T2 and C383-T1, which are occipital
and frontal impacts, respectively.
Review of Finite Element Models of the Human Head
Many FE brain models with varying degrees of anatomical accuracy, element size
and material properties have been developed to study brain injury since the first 3D head
model described by Ward and Thompson [64] in 1975. A brief review of relevant 3D FE
head models is presented in the following section.
Kungliga Tekniska Högskolan (KTH)
The original Kungliga Tekniska Högskolan (KTH) model was a detailed and
parameterized FEM proposed by Kleiven and von Holst in 2002 [65]. It included the
scalp, skull, brain, meninges, CSF, bridging veins and a simplified neck. The current
version of the model consists of approximately 18,400 elements and 19,000 nodes
(Figure 15). The brain-skull interface is modeled with tied nodes between the skull and
26
dura and a sliding contact with no separation between the CSF and its contact with the
meningeal membranes and the brain. Brain tissue is modeled with a hyperelastic
Mooney-Rivlin model (6). The model was validated by Kleiven [66] in 2006 against two
pressure experiments: Experiment 37 conducted by Nahum et al. [31] and the long-
duration impact conducted by Trosseille et al. [32], as well as the following local brain
displacement experiments conducted by Hardy et al. [31, 32]: C755-T2, C383-T1, and
C291-T1.
Figure 15. (A) Skull,(B) brain, (C) section view of the KTH head and neck model [65], and
(D) cross section showing FE mesh and anatomical structures represented in the KTH
model [65].
Wayne State University Head Injury Model (WSUHIM)
The first version of the Wayne State University Head Injury Model (WSUHIM)
was described by Ruan et al. [67] in 1993 and consisted of a scalp, skull, dura, falx, brain,
27
and CSF. This model was later improved by Zhou et al. [68] in 1995 by adding ventricles
and differentiating between white and gray matter in the cerebrum. The model was again
revised in 2001 by Zhang et al. [69] to increase the mesh density and introduce a sliding
interface between the exterior surface of the arachnoid membrane and the interior surface
of the dura (Figure 16). The brain response is modeled with a linear viscoelastic model
with differentiation between gray matter, white matter, and the brainstem. This model
was validated by Zhang et al. [69] against the pressure data of Nahum et al. [31] and
Trosseille et al. [32], as well as the C755-T2 occipital impact from the brain displacement
experiments of Hardy et al. [29].
Figure 16. Various versions of the WSUHIM. (A) The original model developed by Ruan et
al. [67], (B) the improved model with gray and white matter differentiation [70], and (C) the
current version of the model [69].
Eindhoven University of Technology (TUE)
The FE model developed at Eindhoven University of Technology (TUE) was
developed by Claessens et al. [71] in 1997 based on CT and MRI scans from the Visible
Human Database and included the skull, brain and facial bone. Model geometry was
improved by Brands et al. [72] in 2002 along with the implementation of the CSF, dura,
28
falx and tentorium (Figure 17). The brain-skull interface was modeled with shared nodes
and brain response was modeled with a 4-term Mooney-Rivlin hyperelastic model. This
model was validated against Nahum’s pressure data [31].
Figure 17. Overview of the head model developed by TUE [72].
Total Human Model for Safety (THUMS)
The head model from the Total Human Model for Safety (THUMS) Version 4.01
(THUMS AM50 Ver4.01, Toyota TCRDL, Japan) was developed from the geometry of a
data set available in the Visible Human Project (NIH, USA) and was created according to
anatomical references. The THUMS head FEM consists of the skull and facial bones,
cerebrum (distinct white and gray matter), cerebellum, brainstem, CSF, meningeal
membranes, falx, tentorium, and the sagittal sinus (Figure 18). To represent contact with
the CSF, tiebreak interfaces were used to simulate frictionless sliding behavior. This
29
contact formulation was assumed between the dura and arachnoid layers, as well as
between the falx and surrounding soft tissue elements. A linear viscoelastic material
model was used to model the response of the gray and white matter. A previous version
of the THUMS head model was validated against pressure and brain displacement
experiments by Kimpara et al. [73], but the current version has not yet been validated.
Figure 18. (A) Isometric view of the THUMS v4.01 head model with the brain exposed and
(B) midsagittal view of the FEM.
Global Human Body Models Consortium (GHBMC)
The Global Human Body Models Consortium (GHBMC) head model was meshed
from a Computer Aided Design (CAD) dataset developed from MRI and CT scans of an
average adult male, and includes scalp, skull, facial bones, cerebrum, cerebellum, lateral
ventricles, corpus callosum, sinuses, thalamus, brainstem, meningeal layers, falx,
tentorium, and bridging veins (Figure 19). A tie-break contact was defined between the
30
dura and arachnoid layers to model the relative motion at the brain-skull interface and
Kelvin-Maxwell viscoelastic model was used to model the gray and white matter.
Figure 19. (A) Isometric view of the GHBMC head model with the brain exposed and (B)
midsagittal view of the FEM [74].
Université Louis Pasteur of Strasbourg (ULP)
The Université Louis Pasteur of Strasbourg (ULP) model was developed by Kang
et al. [75] in 1997 and included the skull, falx, tentorium, CSF, scalp, cerebellum,
cerebrum, and brainstem. The model was later improved by Deck et al. [76] to include
the falx and tentorium and model subarachnoid space (Figure 20). Additionally a
Figure 20. Overview of the ULP head model [77].
31
Lagrangian formulation for CSF was implemented and an elastic-brittle constitutive law
for the brain-skull interface was employed. A linear viscoelastic constitutive model was
used to represent brain tissue. The ULP model has been validated against both sets of
pressure data [53, 54], as well as the C755-T2 impact conducted by Hardy et al. [29].
University College Dublin Brain Trauma Model (UCDBTM)
The University College Dublin Brain Trauma Model (UCDBTM) was developed
in 2003 by Horgan and Gilchrist [78] from skull geometries extracted from CT scans
from the Visible Human Database. The model includes a skull, scalp, dura, CSF, pia,
falx, tentorium, cerebrum, cerebellum and brainstem Figure 21. The brain-skull interface
Figure 21. (A) Skull, (B), brain, (C), and sagittal cross section of the UCDBTM.
is modeled with solid CSF elements with a very low shear modulus and brain tissue is
modeled with a hyperelastic constitutive model combined with a viscoelastic model in
shear. The model has been validated against the pressure data of Nahum et al. [31] and
Trosseille et al. [32], as well as the C383-T4 impact from the brain displacement
experiments of Hardy et al. [29].
32
Simulated Injury Monitor (SIMon)
The Simulated Injury Monitor (SIMon) model was developed by Takhounts and
Eppinger in 2003 and consists of a rigid skull, CSF, cerebrum, cerebellum, brain stem,
ventricles, bridging veins and a falx and tentorium (Figure 22). All part interfaces are
modeled with merged nodes. A Kelvin-Maxwell viscoelastic formulation is used to
model brain material. The model was validated by Takhounts et al. [79] in 2008 against
two pressure experiments: Experiment 37 conducted by Nahum et al. [31] and the long-
duration impact conducted by Trosseille et al. [32], as well as the following local brain
displacement experiments conducted by Hardy et al. [31, 32]: C755-T2, C383-T1, and
C291-T1.
Figure 22. (A) Isometric view of the SIMon FEM with exposed brain and (B) isometric view
of the FEM showing various anatomical features represented in the model.
33
CHAPTER SUMMARIES
Chapter II: Development and Validation of an Atlas-Based Finite Element Brain
Model
The development of the ABM FE model and the optimization of the brain material
properties using a Latin Hypercube Sample (LHS) of the five brain parameters is
described. Additionally, multiobjective optimization was performed against localized
brain motion data from three distinct cadaver impacts, using CORA as the optimization
parameter.
Chapter III: Comparison of Validation Data for Finite Element Models of the
Human Head
Validation of the ABM against the localized brain displacements of three cadaver impacts
is described. Model performance is quantified using CORA and compared to the
performance of other currently used FE models of the human head.
Chapter IV: Summary of Research
A brief overview of work presented in this thesis.
34
REFERENCES
[1] T. Mayer, M. L. Walker, D. G. Johnson, and M. E. Matlak, “Causes of morbidity
and mortality in severe pediatric trauma,” JAMA, vol. 245, no. 7, pp. 719–721,
1981.
[2] T. R. Frieden, R. Ikeda, R. C. Hunt, M. Faul, L. Xu, M. M. Wald, and V. G.
Coronado, “Traumatic Brain Injury in the United States: Emergency Department
Visits, Hospitalizations and Deaths 2002-2006,” Natl. Cent. Inj. Prev. Control Cent.
Dis. Control Prev. US Dep. Health Hum. Serv., 2010.
[3] V. G. Coronado, L. Xu, S. V. Basavaraju, L. C. McGuire, M. M. Wald, M. D. Faul,
B. R. Guzman, and J. D. Hemphill, “Surveillance for Traumatic Brain Injury-
Related Deaths - United States, 1997–2007,” Morb. Mortal. Wkly. Rep., vol. 60, no.
5, May 2011.
[4] D. J. Thunnan, C. M. Branche, and J. E. Sniezek, “The epidemiology of sports-
related traumatic brain injuries in the United States: recent developments.,” J. Head
Trauma Rehabil., vol. 13, no. 2, pp. 1–8, 1998.
[5] M. R. Schulz, S. W. Marshall, F. O. Mueller, J. Yang, N. L. Weaver, W. D.
Kalsbeek, and J. M. Bowling, “Incidence and risk factors for concussion in high
school athletes, North Carolina, 1996–1999,” Am. J. Epidemiol., vol. 160, no. 10,
pp. 937–944, 2004.
[6] M. W. Collins, G. L. Iverson, M. R. Lovell, D. B. McKeag, J. Norwig, and J.
Maroon, “On-field predictors of neuropsychological and symptom deficit following
sports-related concussion,” Clin. J. Sport Med., vol. 13, no. 4, pp. 222–229, 2003.
[7] C. M. Baugh, C. A. Robbins, R. A. Stern, and A. C. McKee, “Current understanding
of chronic traumatic encephalopathy,” Curr. Treat. Options Neurol., vol. 16, no. 9,
pp. 1–13, 2014.
[8] E. G. Takhounts, R. H. Eppinger, J. Q. Campbell, R. E. Tannous, E. D. Power, and
L. S. Shook, “On the Development of the SIMon Finite Element Head Model,”
Stapp Car Crash J., vol. 47, pp. 107–133, Oct. 2003.
[9] H. Gray, “Anatomy of the human body. Philadelphia: Lea & Febiger, 1918;
Bartleby. com, 2000,” 2001.
[10] “Cranium (skull) with calvarium removed - showing cranial fossas,” 21-Jul-2015.
[Online]. Available: http://www.anatomyexpert.com/app/structure/1619/1121/.
35
[11] G. J. Tortora and B. H. Derrickson, Principles of anatomy and physiology. Wiley,
2011.
[12] K. S. Saladin, “The brain and cranial nerves,” in Human Anatomy, Internation
Edition., New York: McGraw Hill Higher Education, 2007, pp. 414–451.
[13] T. P. Prevost, A. Balakrishnan, S. Suresh, and S. Socrate, “Biomechanics of brain
tissue,” Acta Biomater., vol. 7, no. 1, pp. 83–95, Jan. 2011.
[14] M. Dodgson, “Colloidal structure of brain,” Biorheology, vol. 1, no. 1, pp. 21–30,
1962.
[15] J. B. Koeneman, “Viscoelastic properties of brain tissue,” 1966.
[16] J. E. Galford and J. H. McElhaney, “A viscoelastic study of scalp, brain, and dura,”
J. Biomech., vol. 3, no. 2, pp. 211–221, 1970.
[17] S. Cheng and L. E. Bilston, “Unconfined compression of white matter,” J.
Biomech., vol. 40, no. 1, pp. 117–124, 2007.
[18] M. T. Prange and S. S. Margulies, “Regional, directional, and age-dependent
properties of the brain undergoing large deformation,” J. Biomech. Eng., vol. 124,
no. 2, pp. 244–252, 2002.
[19] K. B. Arbogast, D. F. Meaney, and L. E. Thibault, “Biomechanical characterization
of the constitutive relationship for the brainstem,” SAE Technical Paper, 0148-
7191, 1995.
[20] L. E. Bilston, Z. Liu, and et al., “Linear viscoelastic properties of bovine brain tissue
in shear,” Biorheology, vol. 34, no. 6, pp. 377–385, 1997.
[21] S. Chatelin, A. Constantinesco, and R. Willinger, “Fifty years of brain tissue
mechanical testing: from in vitro to in vivo investigations,” Biorheology, vol. 47,
no. 5, pp. 255–276, 2010.
[22] G. T. Fallenstein, V. D. Hulce, and J. W. Melvin, “Dynamic mechanical properties
of human brain tissue,” J. Biomech., vol. 2, no. 3, pp. 217–226, Jul. 1969.
[23] L. Shuck and S. Advani, “Rheological response of human brain tissue in shear,” J.
Fluids Eng., vol. 94, no. 4, pp. 905–911, 1972.
[24] K. L. Thibault and S. S. Margulies, “Age-dependent material properties of the
porcine cerebrum: effect on pediatric inertial head injury criteria,” J. Biomech., vol.
31, no. 12, pp. 1119–1126, 1998.
36
[25] S. Nicolle, M. Lounis, and R. Willinger, “Shear Properties of Brain Tissue over a
Frequency Range Relevant for Automotive Impact Situations: New Experimental
Results.,” Stapp Car Crash J., vol. 48, pp. 239–258, 2004.
[26] S. Nicolle, M. Lounis, R. Willinger, and J.-F. Palierne, “Shear linear behavior of
brain tissue over a large frequency range,” Biorheology, vol. 42, no. 3, pp. 209–223,
2005.
[27] D. Kallieris, A. Rizzetti, and R. Mattern, “Some Observations to the Skull-Brain
Trauma,” NASA, no. 19980003880, 1997.
[28] W. Hardy, C. Foster, A. King, and S. Tashman, “Investigation of brain injury
kinematics: introduction of a new technique,” ASME Appl. Mech. Div.-Publ.-AMD,
vol. 225, pp. 241–254, 1997.
[29] W. N. Hardy, C. D. Foster, M. J. Mason, K. H. Yang, A. I. King, and S. Tashman,
“Investigation of Head Injury Mechanisms Using Neutral Density Technology and
High-Speed Biplanar X-ray,” Stapp Car Crash J., vol. 45, pp. 337–368, Nov. 2001.
[30] W. N. Hardy, M. J. Mason, C. D. Foster, C. S. Shah, J. M. Kopacz, K. H. Yang, A.
I. King, J. Bishop, M. Bey, W. Anderst, and S. Tashman, “A study of the response
of the human cadaver head to impact,” Stapp Car Crash J., vol. 51, pp. 17–80, Oct.
2007.
[31] A. M. Nahum, R. Smith, and C. C. Ward, “Intracranial pressure dynamics during
head impact,” SAE Technical Paper, 1977.
[32] X., Domont, A., Guillon, F., Lavaste, F., Tarriere, C., Trosseille, Development of a
F.E.M. of the human head according to a specific test protocol. 1992.
[33] T. A. Gennarelli, L. E. Thibault, J. H. Adams, D. I. Graham, C. J. Thompson, and R.
P. Marcincin, “Diffuse axonal injury and traumatic coma in the primate,” Ann.
Neurol., vol. 12, no. 6, pp. 564–574, Dec. 1982.
[34] Abel, J., Gennarelli, T., and Segawa, H., “Incidence and Severity of Cerebral
Concussion in the Rhesus Monkey Following Sagittal Plane Angular Acceleration,”
SAE Tech. Pap., 1978.
[35] A. K. Ommaya, “Mechanical properties of tissues of the nervous system,” J.
Biomech., vol. 1, no. 2, pp. 127–138, 1968.
[36] D. T. Ross, D. F. Meaney, M. K. Sabol, D. H. Smith, and T. A. Gennarelli,
“Distribution of forebrain diffuse axonal injury following inertial closed head injury
in miniature swine,” Exp. Neurol., vol. 126, no. 2, pp. 291–299, 1994.
37
[37] J. W. Lighthall, H. G. Goshgarian, and C. R. Pinderski, “Characterization of axonal
injury produced by controlled cortical impact,” J. Neurotrauma, vol. 7, no. 2, pp.
65–76, 1990.
[38] A. Marmarou, M. A. A.-E. Foda, W. van den Brink, J. Campbell, H. Kita, and K.
Demetriadou, “A new model of diffuse brain injury in rats: Part I: Pathophysiology
and biomechanics,” J. Neurosurg., vol. 80, no. 2, pp. 291–300, 1994.
[39] A. K. Ommaya, “Biomechanics of Trauma,” Appleton-Century-Crofts East Norwalk
CT, 1985.
[40] J. Davidsson, M. Angeria, and M. Risling, “Injury threshold for sagittal plane
rotational induced diffuse axonal injuries,” presented at the Proceedings of the
International Research Council on the Biomechanics of Injury conference, 2009,
vol. 37, pp. 43–56.
[41] S. S. Margulies and L. E. Thibault, “A proposed tolerance criterion for diffuse
axonal injury in man,” J. Biomech., vol. 25, no. 8, pp. 917–923, 1992.
[42] S. Rowson and S. M. Duma, “Brain injury prediction: assessing the combined
probability of concussion using linear and rotational head acceleration,” Ann.
Biomed. Eng., vol. 41, no. 5, pp. 873–882, 2013.
[43] D. Denny-Brown and W. R. Russell, “Experimental cerebral concussion,” Brain,
vol. 64, no. 2–3, pp. 93–164, 1941.
[44] E. Gurdjian and H. Lissner, “Photoelastic Confirmation of the Presence of Shear
Strains at the Craniospinal Junction in Closed Head Injury*,” J. Neurosurg., vol. 18,
no. 1, pp. 58–60, 1961.
[45] L. Thomas, V. Roberts, and E. Gurdjian, “Impact-induced pressure gradients along
three orthogonal axes in the human skull.,” J. Neurosurg., vol. 26, no. 3, p. 316,
1967.
[46] A. Holbourn, “Mechanics of head injuries,” The Lancet, vol. 242, no. 6267, pp.
438–441, 1943.
[47] T. Gennarelli, J. Adams, and D. Graham, “Acceleration induced head injury in the
monkey. I. The model, its mechanical and physiological correlates,” in
Experimental and Clinical Neuropathology, Springer, 1981, pp. 23–25.
[48] T. A. Gennarelli, “Head injury in man and experimental animals: clinical aspects,”
in Trauma and Regeneration, Springer, 1983, pp. 1–13.
38
[49] A. K. Ommaya, R. L. Grubb, and R. A. Naumann, “Coup and contre-coup injury:
observations on the mechanics of visible brain injuries in the rhesus monkey,” J.
Neurosurg., vol. 35, no. 5, pp. 503–516, Nov. 1971.
[50] A. Ommaya and A. Hirsch, “Tolerances for cerebral concussion from head impact
and whiplash in primates,” J. Biomech., vol. 4, no. 1, pp. 13–21, 1971.
[51] K. Ono, A. Kikuchi, M. Nakamura, H. Kobayashi, and N. Nakamura, “Human head
tolerance to sagittal impact reliable estimation deduced from experimental head
injury using subhuman primates and human cadaver skulls,” SAE Technical Paper,
1980.
[52] Gadd, C., “Use of a Weighted-Impulse Criterion for Estimating Injury Hazard,”
SAE Tech. Pap., 1966.
[53] J. Versace, “A review of the severity index,” SAE Technical Paper, 0148-7191,
1971.
[54] L. Zhang, K. H. Yang, and A. I. King, “A proposed injury threshold for mild
traumatic brain injury,” J. Biomech. Eng., vol. 126, no. 2, pp. 226–236, 2004.
[55] C. Simms and D. Wood, “Injury mechanisms and injury criteria,” Pedestr. Cycl.
Impact Biomech. Perspect., pp. 75–97, 2009.
[56] J. A. Newman, N. Shewchenko, and E. Welbourne, “A proposed new biomechanical
head injury assessment function - the maximum power index,” Stapp Car Crash J.,
vol. 44, pp. 215–247, Nov. 2000.
[57] E. G. Takhounts, M. J. Craig, K. Moorhouse, J. McFadden, and V. Hasija,
“Development of brain injury criteria (BrIC),” Stapp Car Crash J., vol. 57, pp. 243–
266, Nov. 2013.
[58] NHTSA, “Final Rule, FMVSS 208, Occupant Crash Protection,” Fed. Regist., 1984.
[59] Takhounts, Erik G., Hasija, Vikas, Ridella, Stephen A., Rowson, Steven, and Duma,
Stefan M., “Kinematic Rotational Brain Injury Criterion (BRIC),” presented at the
The 22nd International Technical Conference on the Enhanced Safety of Vehicles
(ESV), 2011.
[60] C. Lauret, M. Hrapko, J. A. W. van Dommelen, G. W. M. Peters, and J. S. H. M.
Wismans, “Optical characterization of acceleration-induced strain fields in
inhomogeneous brain slices,” Med. Eng. Phys., vol. 31, no. 3, pp. 392–399, Apr.
2009.
39
[61] R. Cloots, H. Gervaise, J. van Dommelen, and M. Geers, “Biomechanics of
Traumatic Brain Injury: Influences of the Morphologic Heterogeneities of the
Cerebral Cortex,” Ann. Biomed. Eng., vol. 36, no. 7, pp. 1203–1215, Apr. 2008.
[62] R. W. Ogden, “Non-linear elastic deformations,” Eng. Anal., vol. 1, no. 2, p. 119,
Jun. 1984.
[63] W. N. Hardy, “Response of the Human Cadaver Head to Impact,” Wayne State
University, Detroit, MI, 2007.
[64] C. C. Ward and R. B. Thompson, “The development of a detailed finite element
brain model,” SAE Technical Paper, 1975.
[65] S. Kleiven and H. von Holst, “Consequences of head size following trauma to the
human head,” J. Biomech., vol. 35, no. 2, pp. 153–160, Feb. 2002.
[66] S Kleiven, “Evaluation of head injury criteria using a finite element model validated
against experiments on localized brain motion, intracerebral acceleration, and
intracranial pressure,” Int. J. Crashworthiness, vol. 11, no. 1, pp. 65–79, 2006.
[67] J. S. Ruan, T. B. Khalil, and A. I. King, “Finite element modeling of direct head
impact,” SAE Technical Paper, 1993.
[68] C. Zhou, T. B. Khalil, and A. I. King, “A new model comparing impact responses of
the homogeneous and inhomogeneous human brain,” SAE Technical Paper, 1995.
[69] L. Zhang, K. H. Yang, R. Dwarampudi, K. Omori, T. Li, K. Chang, W. N. Hardy, T.
B. Khalil, and A. I. King, “Recent advances in brain injury research: a new human
head model development and validation,” Stapp Car Crash J., vol. 45, pp. 369–394,
Nov. 2001.
[70] A. S. Al-Bsharat, W. N. Hardy, K. H. Yang, T. B. Khalil, S. Tashman, and A. I.
King, “Brain/skull relative displacement magnitude due to blunt head impacts: new
experimental data and model.,” Stapp Car Crash J., vol. 43, no. 2, pp. 321–332,
1999.
[71] Mauric Claessens, Fons Sauren, and Jac Wismans, “Modeling of the Human Head
Under Impact Conditions: A Parametric Study,” 1997.
[72] D. W. A. Brands, P. H. M. Bovendeerd, and J. S. H. M. Wismans, “On the potential
importance of non-linear viscoelastic material modelling for numerical prediction of
brain tissue response: test and application,” Stapp Car Crash J., vol. 46, pp. 103–
121, Nov. 2002.
40
[73] H. Kimpara, Y. Nakahira, M. Iwamoto, K. Miki, K. Ichihara, S. Kawano, and T.
Taguchi, “Investigation of anteroposterior head-neck responses during severe
frontal impacts using a brain-spinal cord complex FE model.,” Stapp Car Crash J.,
vol. 50, pp. 509–544, 2006.
[74] H. Mao, L. Zhang, B. Jiang, V. V. Genthikatti, X. Jin, F. Zhu, R. Makwana, A. Gill,
G. Jandir, A. Singh, and K. H. Yang, “Development of a finite element human head
model partially validated with thirty five experimental cases,” J. Biomech. Eng., vol.
135, no. 11, p. 111002, Nov. 2013.
[75] H.-S. Kang, R. Willinger, B. M. Diaw, and B. Chinn, “Validation of a 3D anatomic
human head model and replication of head impact in motorcycle accident by finite
element modeling,” SAE Technical Paper, 1997.
[76] C. Deck, S. Nicolle, and R. Willinger, “Human head FE modelling: improvement of
skull geometry and brain constitutive laws,” presented at the Proceedings of the
International Research Council on the Biomechanics of Injury conference, 2004,
vol. 32.
[77] D. Marjoux, D. Baumgartner, C. Deck, and R. Willinger, “Head injury prediction
capability of the HIC, HIP, SIMon and ULP criteria,” Accid. Anal. Prev., vol. 40,
no. 3, pp. 1135–1148, May 2008.
[78] T. Horgan and M. D. Gilchrist, “The creation of three-dimensional finite element
models for simulating head impact biomechanics,” Int. J. Crashworthiness, vol. 8,
no. 4, pp. 353–366, 2003.
[79] E. G. Takhounts, S. A. Ridella, V. Hasija, R. E. Tannous, J. Q. Campbell, D.
Malone, K. Danelson, J. Stitzel, S. Rowson, and S. Duma, “Investigation of
traumatic brain injuries using the next generation of simulated injury monitor
(SIMon) finite element head model,” Stapp Car Crash J., vol. 52, pp. 1–31, Nov.
2008.
41
Chapter II: Development and Validation of an Atlas-Based
Finite Element Brain Model
Logan E. Miller, Jillian E. Urban, Joel D. Stitzel
Virginia Tech – Wake Forest University Center for Injury Biomechanics,
Winston-Salem, NC
The following manuscript has been submitted to Biomechanics and Modeling in
Mechanobiology.
42
ABSTRACT
Traumatic brain injury (TBI) is a leading cause of disability and injury-related death. To
enhance our ability to prevent such injuries, we can study brain response using validated
finite element (FE) models. In the current study, a high-resolution, anatomically accurate
FE model was developed from the International Consortium for Brain Mapping (ICBM)
brain atlas. Due to the wide variation in published brain material parameters, a technique
called Latin Hypercube Sampling (LHS) was employed to optimize the brain material
properties against three experimental cadaver tests to achieve ideal biomechanics.
Additionally, pretension and thickness of the falx were varied in a lateral impact
configuration. The ABM was subjected to the boundary conditions from three high-rate
experimental cadaver tests with different parameter combinations for each test. Brain
displacements at the experimental neutral density target (NDT) locations in the model
were compared to the displacements observed experimentally and error was quantified
using CORA (CORrelation and Analysis), a method to evaluate the correlation of two
curves. An average CORA score was computed for each variation and maximized to
identify the optimal combination of parameters. The strongest relationships between
CORA and material parameters were observed for the shear parameters. Using brain
properties obtained through multi-objective optimization, the ABM was validated in three
impact configurations and shows good agreement with experimental data. The final
model developed in this study consists of optimized brain material properties and is
validated in three different, multidirectional cadaver impacts against local brain
displacement data.
43
1 INTRODUCTION
Each year, approximately 1.7 million people in the United States suffer from
traumatic brain injury (TBI) [1]. TBI is a major public health concern as it is a leading
cause of disability and injury-related death - accounting for nearly one third of all injury-
related deaths [2]. To prevent and treat these types of injuries, the fundamental injury
mechanisms need to be well-characterized and understood. While there are many theories
about the underlying injury mechanisms of various brain injuries, we still do not fully
understand the biomechanical basis for many of these injuries. Various methods have
been used over the years to investigate and gain a deeper understanding of brain injury
mechanisms, including experimental testing of animal models, post-mortem human
subjects (PMHS), anthropomorphic test devices (ATDs), as well as computational
modeling of the brain [3]. Finite element (FE) models are powerful tools for investigating
injury mechanisms because they can provide spatial and temporal distributions of stresses
and strains over the problem domain. The quality of a model’s predictions, however, is
dependent on the accuracy of the modeled geometry and the model’s ability to describe
complex mechanical behavior and material response.
Many FE brain models with varying degrees of anatomical accuracy, element size
and material properties have been developed to study brain injury over the last few
decades. Early models represented the brain-skull structure with a simple fluid-filled
spherical shell; subsequently, model complexity increased, and the most advanced
models now include major anatomical brain structures and different material properties
for each component in the brain. Among these are the Royal Institute of Technology
KTH (Kungliga Tekniska Högskolan) brain model, the Simulated Injury Monitor
44
(SIMon), Dartmouth Head Injury Model (DHIM), Wayne State University Brain Injury
Model (WSUBIM) as well as the head models of full-body FE models such as the Total
Human Model for Safety (THUMS) and Global Human Body Models Consortium
(GHBMC) [4, 3, 5, 6, 7, 8]. The geometry of various structures within these head models
were derived from medical images, such as computed tomography (CT) and magnetic
resonance imaging (MRI). The model development process typically involves image
segmentation to define three-dimensional (3D) surfaces of desired structures followed by
a surface-based mesh generation process. During this process, complex anatomical
features are frequently simplified and ‘smoothed’ to allow faster development and simple
contact definition at model interfaces. One feature of particular importance that is over-
simplified in many brain FEMs is the gyral folds on the surface of the brain. The brain
surface is typically modeled as a smooth surface surrounded by a uniform layer of
cerebrospinal fluid (CSF), which ignores the intricate gyral folds throughout the brain.
This simplification and reduction in anatomic accuracy may significantly impact the
model’s local brain response and subsequent prediction of injury. Indeed, studies have
indicated that the intricate folding patterns on the surface of the brain greatly affect both
the distribution and magnitude of stress and strain throughout the brain [9]. Another
limitation of previous models is the substantial amount of time required for the surface-
based mesh generation process, which hinders production of patient-specific models.
Patient-specific and population-specific models are desirable for studying head injury due
to significant variations in brain morphology between individuals [4, 10, 11, 12, 13].
45
Correct anatomical geometry is particularly important for modelling injuries like
concussion, whose occurrence and severity is dependent on deformation of specific areas
of the brain [14]. Another area of growing interest that is not well understood and could
benefit from FE modelling is chronic traumatic encephalopathy (CTE). The pathology of
this disease has been extensively studied and is characterized by the deposition of tau
proteins at the depths of the sulci [15]. Although the focal nature of CTE pathology has
been well characterized, there is relatively little known about the underlying injury
mechanisms. If finite element models are to be employed in studying CTE, the inclusion
of gyral folds in brain FE models is crucial to gaining a better understanding of the
disease, given the focal nature of CTE pathology.
The objective of this study is two-fold. The first objective was to create a high
resolution, anatomically accurate brain finite element model from the International
Consortium for Brain Mapping (ICBM) brain atlas using a voxel-based mesh generation
approach. The second objective was to optimize the brain and falx material properties
using cadaveric experimental data in three distinct impact configurations. This atlas-
based brain model (ABM) is proposed as a new tool for studying brain injury that can
accurately predict local stresses and strains in brain tissue by providing a high level of
anatomic accuracy, particularly at the brain-skull interface.
2 METHODS
The ABM was developed from the geometry of the ICBM adult brain atlas, a
probabilistic atlas derived from a large dataset of 7000 subjects (ages 18-90) that
represents the average adult brain [16]. The ICBM atlas contains 58 independently
46
labeled regions of the brain including cortical gyri, white matter, gray matter, and CSF.
An FE model was created from this image set by converting each 1mm isotropic voxel
into a single element of the same size using a custom code developed in MATLAB (The
MathWorks, Natick, MA). While the high level of detail provided by the atlas allows the
opportunity to represent many detailed brain structures, initial model development
included only four distinct parts from the atlas geometry: cerebrum (combined white and
gray matter), cerebellum, CSF, and ventricles. Because the model was developed from a
skull-stripped image set, there are places on the outer surface of the brain that are not
completely covered by CSF. Since this is not a realistic representation of the brain-skull
interface, a single, uniform layer of CSF was added around the entire model to ensure
that there was at least one CSF element between the brain and skull at all points. This
added layer of CSF was 74 mL, and increased the CSF volume from 192.3 mL to 266.3
mL, which is reasonable when compared to real world CSF volumes [17]. The falx
cerebri and tentorium cerebelli are important structures in the cranium which were not
represented in the ICBM atlas, so they were manually implemented into the model. The
falx was defined as a layer of shell elements along the midsagittal plane, and the
tentorium as a layer of shell elements on the superior surface of the cerebellum,
separating the cerebrum and cerebellum (Figure 23). Lastly, a layer of rigid shells
surrounding the external surface of the CSF was generated in LS-PrePost to completely
enclose the model. The current model, shown in Figure 24A-D, has approximately 2
million nodes and elements.
47
Figure 23. ABM showing the falx, tentorium,
and ventricles
Figure 24. Cross sections of ABM. (a) axial,
(b) sagittal, (c) coronal, (d) axial isometric
Initial material property values for all parts, excluding the brain material model, were
similar to the SIMon model (Table 1), with the exception of the CSF which was modeled
with the same material properties as the ventricles, as opposed to the Kelvin-Maxwell
fluid used in the SIMon model [3]. This implementation of the CSF was motivated by the
results of Chafi et al. (2009) which investigated brain response using an FEM that
modeled the CSF using three different formulations: elastic fluid, viscoelastic, and nearly
incompressible elastic material. Chafi et al. (2009) found that the most comparable results
to experimental data resulted from modelling the CSF as a fluid-like material. Contact at
all part interfaces was modeled with merged nodes.
Table 1. Material Properties of the ABM
Part LS-DYNA Material Model Material Properties
CSF/Ventricles Elastic fluid ρ=1000 kg/m
3, E=0 MPa, ν=0.5,
K=2100 MPa
Falx*/Tentorium Elastic ρ=1130 kg/m
3, E=31.5 MPa, ν=0.45,
K=2100 MPa
Skull Rigid ρ=35200 kg/m3, E=6900 MPa, ν=0.3
Brain Kelvin-Maxwell viscoelastic Optimized
48
2.1 Brain Material Optimization
The values of brain material parameters vary greatly throughout the literature,
which motivates the current optimization study (Table 2). Brain tissue is often modeled
using a linear viscoelastic material formulation [3, 7, 8, 18, 19, 20, 21]. The shear
relaxation behavior is described by:
𝐺(𝑡) = 𝐺∞ + (𝐺0 − 𝐺∞)𝑒−𝛽𝑡
(1)
where G∞ is the infinite shear modulus, G0 is the initial shear modulus, and β is the decay
constant. Therefore, this material model was selected for material optimization
(*MAT_061 in LS-DYNA). Brain density (ρ), bulk modulus (K), and the three shear
response parameters (G∞, G0, β) were varied using Latin Hypercube Sampling (LHS) to
generate 100 combinations of material parameters, or 100 distinct brain material models.
Because of the relationship between G∞ and G0 in defining shear relaxation, G∞ must be
less than G0 for every material model generated by the LHD. To accomplish this, instead
of varying G∞ and G0 independently, G0 and the ratio G∞/G0 were varied. In LHS each
Table 2. Parameter Ranges for Latin Hypercube Sample
Min/Max Reference
ρ
kg/m3
1000 Kimpara et al. (2006)
1140 Willinger et al. (1999)
K
GPa
0.002 Ji et al. (2014), (Ji et al. 2014)
2.19
Al-Bsharat et al. (1999), Willinger et al. (1999), Zhang et al. (2001),
Kimpara et al. (2006), Chafi et al. (2009), McAllister et al. (2012),
Kangarlou (2013), Mao et al. (2013), Post et al. (2014)
G0
kPa
0.772 Kleiven and Hardy (2002)
49 Pinnoji et al. (2010), Tinard et al. (2012)
G∞
kPa
0.185 Jirousek et al. (2005)
16.7 Pinnoji et al. (2010), Tinard et al. (2012)
G∞/G0 0.1 Ji et al. (2014)
0.5584 Takhounts et al. (2003)
β
s-1
0.06 Kimpara et al. (2006)
400 Kangarlou (2013)
49
parameter is varied over a predefined range (Table 2) independent of the values of other
variables and orthogonal sampling is employed over the multidimensional sample space
to ensure the generated samples are representative of the total variability. The parameter
ranges displayed in Table 2 were obtained from literature values and other published
finite element models. Because the literature-reported ranges for bulk modulus, shear
modulus and the decay constant span orders of magnitude, logarithmic sampling was
employed for these variables in order to appropriately sample the lower range of values.
To evaluate the effect of each parameter and determine optimal values for each, the
performance of the ABM with each material model is compared to experimental data.
2.2 Experimental Tests
Three experimental tests were investigated in this study: the C755-T2 (occipital
impact ), C383-T1 (frontal impact), and C291-T1 (parietal impact) [22, 23]. These three
Figure 25. Approximate cross-sectional planes where NDT columns are located for each
experimental test (a), locations of NDTs in the XZ plane on the right side of the brain (b),
and locations of NDTs in the XZ plane on the left side of the brain (c)
50
tests were chosen for optimization purposes because they cover a range of impact
velocities and directions. In the cadaveric tests, local displacements at various locations
within the brain were analyzed following impacts of varying severity and direction [23].
In each experiment, vertical columns of 5-6 radio-opaque neutral density targets (NDTs)
were implanted in the brain, and a high-speed biplanar X-ray system was used to track
their relative motion during the impact. Three-dimensional skull kinematics were
evaluated with an accelerometer array affixed to the cadaver skull to determine linear and
angular velocities at the head center of gravity (CG), which were then used as boundary
conditions to drive each ABM simulation [23]. To examine comparable displacements in
the ABM, nodes closest to the physical location of each NDT were identified with respect
to the Frankfurt plane (Figure 25) and their displacements were calculated throughout
each simulation. Summary characteristics for the three tests are shown in Table 3.
Table 3. Summary of Experimental Test Conditions
51
2.3 Model Performance
To evaluate the effect of varying each material parameter and identify the optimal
material properties, the accuracy of the ABM’s response for each combination of material
parameters was quantified. The metric used to quantify the difference between model
response and experimental data was CORA (CORrelation and Analysis) [24]. The CORA
method evaluates the correlation of two curves by calculating a rating using two methods:
the corridor method and the cross-correlation method. These two ratings range from 0 to
1 and are averaged to determine a total CORA score (1 indicates a perfect match). The
corridor method computes a rating based on where the comparison curve falls in relation
to corridors around the reference curve. The cross-correlation method is based on ratings
for the phase shift, size and progression of time-shifted curves.
The ABM was subjected to the boundary conditions from the three experimental
tests with each of the 100 brain material models implemented into the FEM, resulting in a
total of 300 simulations. Average CORA values computed from all NDTs were
calculated for all 300 simulations to identify the combination of properties that yielded
the best correlation to experimental data. It was hypothesized that this would result in
similar material parameters for each experimental condition. During our optimization
process, however, we observed ideal shear properties approximately five times larger for
the lateral condition than for the frontal and occipital cases (Table 4). Given the lateral
impact condition, one would anticipate the falx to play a major role in brain response.
Additionally, the high shear properties that resulted from the C291-T1 optimization
indicated that there was too much lateral motion; or in other words, that the falx was not
constraining lateral motion sufficiently.
52
2.4 Falx Material Optimization
The observation that brain shear stiffness were unusually high to satisfy the lateral
impact case motivated investigation and optimization of the falx model. An additional
LHS was conducted that varied pretension from 0-10% and falx shell thickness from 0.41
mm to 7 mm to optimize the falx model in the ABM. These ranges were adapted from
Golman et al. (2013). Pretension was implemented using an optional material card that
pre-strains the material by a desired factor. The resulting values for pretension and
thickness of the falx were then implemented into the ABM with each of the original 100
brain material models, and the C291-T1 variation was rerun. The frontal and occipital
cases were not rerun because the motion was in the sagittal plane for both cases, and not
greatly affected by the falx properties.
2.5 Data Analysis
A CORA score was computed for the ABM’s performance in all 300 simulations
– the three impact conditions with each combination of material parameters incorporated
into the model. The single CORA score for each simulation was calculated by first
averaging the X- and Z-CORA (or Y- and Z-CORA for the C291-T1 case) scores for
each NDT to get individual NDT CORA scores, and then averaging all NDT CORA
scores to determine an overall average CORA score for a given simulation. The
relationship between CORA scores and material parameters was evaluated by computing
the strength of the piecewise linear regression for each parameter. Material parameters
were optimized under each impact condition using CORA as the optimization parameter
and the combination of material properties that produced the best result was identified for
the three impact configurations. Additionally, we can concurrently optimize the material
53
properties against the three impact conditions simultaneously. We do this by averaging
the CORA scores for the C755-T2, C383-T1, and C291-T1 cases for the ABM with each
combination of material parameters. Finally, the model with the highest average CORA
between all three impact conditions (using the optimized falx model) was identified and
its material parameters were determined to be the optimal material properties.
3 RESULTS
The ABM has been developed from the ICBM brain atlas and currently contains 7
parts and about 2 million nodes and elements. The brain and CSF volumes are 1,639.4
mL and 277.1 mL, respectively, and the total model mass is 6.8 kg.
ABM response was compared with experimental data for all 100 variations for
each of the three test configurations and average CORA values for each simulation were
computed. Relationships between CORA scores and material parameters for each of the
test conditions are shown in Figure 26. The results for the lateral case shown in Figure 26
represent the ABM before the falx model was optimized. Also displayed in these plots
are piecewise linear regressions for CORA regressed against each material parameter.
The material properties for the variation with the highest CORA score from each
experimental configuration are shown in Table 4.
Figure 27 shows the relationships between CORA and the falx pretension and
thickness varied in the falx optimization using the lateral experimental configuration. The
results from this variation indicate that the optimal falx pretension is 6.52% and the
optimal thickness is 2.86 mm. Using these values in the falx model, the brain material
optimization in the C291-T1 parietal impact was rerun. The results of this updated
54
Figure 26. Relationships between CORA and brain material parameters from brain
parameter variations in three experimental configurations (using the original falx model in
all variations). The simulation with the highest CORA value for each individual impact
condition is highlighted.
optimization are included in Table 4 as well as the material properties that result from
concurrently optimizing the CORA values from the frontal, occipital, and updated lateral
conditions. We note, from Figure 26, that the shear parameter, G0, has the largest
influence on CORA. Considering the results from the frontal, occipital, and updated
lateral (with optimized falx model) optimizations, we examine the relationship between
G0 and CORA for each experimental configuration (Figure 28).
Table 4. Material Properties from each experimental configuration optimization
Optimization Condition ρ (kg/m3) K (GPa) G0 (kPa) G∞ (kPa) β (s
-1) CORA
C755* 1061 0.2539 4.121 2.160 33.48 0.453
C383* 1036 0.5326 4.634 1.321 64.75 0.424
C291* (original) 1021 0.1066 19.20 10.80 283.2 0.491
C291* (optimized falx) 1113 0.2706 17.26 5.099 247.4 0.473
ALL* 1123 0.1069 5.161 1.859 67.58 0.426
55
Figure 27. CORA relationships with falx pretension (a) and falx thickness (b) for the C291-
T1 impact
3.1 C755-T2
The local displacements of each NDT for each experimental impact are presented
in Figure 29 - Figure 31 using the material model obtained through concurrent
optimization of all three experimental conditions. Relative displacement results for the
NDTs in the C755-T2 are shown in Figure 29. The average CORA score for all NDTs (in
x- and z-directions) is 0.445. The average CORA score in the x-direction is 0.383 and in
the z-direction is 0.508. The NDT with the highest CORA rating in the x-direction was
A2 and in the z-direction was P3 with ratings 0.653 and 0.713, respectively. When
considering combined x- and z-direction CORA scores, P2 had the highest score of 0.539
and A4 had the lowest with 0.275.
3.2 C383-T1
The results for the NDT relative displacements for the C383-T1 impact are shown
in Figure 30. The average CORA score for all NDTs for the frontal case is 0.421. The
average CORA score in the x-direction is 0.391 and in the z-direction is 0.452. The NDT
with the highest CORA rating in the x-direction and the z-direction was P6 with ratings
0.564 and 0.675, respectively. The NDT with the highest average x- and z-direction
56
CORA rating was P6 with a score of 0.620 and the NDT with the lowest average CORA
score was P1 with a score of 0.261.
3.3 C291-T1
The results for the relative displacements of the NDTs in the C291-T1 case are
shown in Figure 31. The average CORA score for all NDTs for this case is 0.412. The
average CORA score in the y-direction is 0.416 and in the z-direction is 0.408. The NDT
Figure 28. Relationship between CORA and shear modulus, G0, for material optimizations
in occipital (a), frontal (b), and parietal (using optimized falx model) (c) impact
configurations. The simulation representing the optimal material model for each condition
optimized individually is shown (C755*, C383*, C291*), as well as optimized concurrently
(ALL*)
with the highest CORA rating in the y-direction and the z-direction was PL4 with ratings
0.519 and 0.689, respectively. The NDT with the highest average y- and z-direction
CORA rating was PL4 with a score of 0.604 and the NDT with the lowest average CORA
score was A4 with a score of 0.207. Refer to Figure 25 for anatomical reference of NDT
locations.
57
Figure 29. Displacement results and individual CORA scores for the optimized material
properties for Hardy C755-T2 occipital impact for anterior (A) and posterior (P) NDT
columns
Figure 30. Displacement results and individual CORA scores for the optimized material
properties for Hardy C383-T1 frontal impact for anterior (A) and posterior (P) NDT
columns
58
Figure 31. Displacement results and individual CORA scores for the optimized material
properties for Hardy C291-T1 lateral impact for the anterior (A) and posterior (P) right (R)
and left (L) columns
4 DISCUSSION
A finite element model of the human brain was developed in the current study
using a voxel-based mesh generation approach. A novel method to determine the material
properties of the brain was employed using Latin Hypercube Sampling to vary material
parameter values and CORA to evaluate response and quantify error. During this process,
59
brain and falx material properties were concurrently optimized against localized
displacement data from three cadaver experiments. The three experiments considered
(C755-T2, C383-T1, and C291-T1) were chosen because they are dynamic impacts,
varying in direction of impact as well as severity of impact which allows for evaluation
of the ABM in a wide range of loading conditions.
An important observation from the relationships displayed in Figure 26 is the
regression strengths between CORA and each material parameter. The parameter with the
strongest relationship to CORA is, by far, the shear modulus, G0 with piecewise
regression coefficients of 0.755, 0.686, and 0.887 for the occipital, frontal and lateral
impacts, respectively. The next largest coefficients result from regressing the shear ratio,
G∞/G0, and CORA, and the strongest of these achieves only 0.101, which signifies a
much weaker relationship than was observed for G0. This reinforces that G0 stands out as
the most important parameter in determining the response of the brain. The general trend
between G0 and CORA, is that increasing the shear stiffness improves the CORA rating
up until a critical value – after which continued increase in stiffness causes the response
to deviate from the experimental data and the CORA score to decrease. For the occipital
and frontal cases, this critical shear modulus value occurs around 4.5 kPa (log(4.5 kPa) =
1.50). The critical value for the lateral case, however, is almost 20 kPa (log(20 kPa) =
3.00). In other words, during this original material optimization process the frontal and
occipital cases optimized to similar material properties, while the optimal shear properties
for the lateral case were much stiffer. This can be deduced from the values in Table 4, as
well as by examining the plots in Figure 28 that relate CORA and the shear modulus, G0.
60
The relationship between CORA and the shear parameters is comparable for the
C755-T2 and C383-T1 variations. The CORA score reaches a maximum at very similar
values of G0 for these two configurations, but the CORA score clearly peaks at a larger
G0 in the C291-T1 variation. Because the two cases with motion in the sagittal plane
(C755-T2 and C383-T1), which is parallel to the falx, gave similar results, while the
lateral case differed, we conducted an additional LHS varying falx pretension and
thickness. Examining the results of this variation (Figure 27), the general trend observed
in pretension, is an increase in model performance with increasing pretension up to about
4%, and then a plateau in CORA score to around 8%, followed by a slight decrease when
pretension is increased beyond 8%. We see a stronger relationship between CORA and
falx thickness than was observed between CORA and pretension. With increasing falx
thickness, there is an increase in the CORA score up to the peak thickness of 2.86 mm
followed by a relatively constant CORA score following this maximum. This plateau in
CORA scores between 2 and 3 mm supports falx thickness measurements reported in
literature, as opposed to the larger values investigated which were not derived from
physiological measurements. This plateau behavior indicates that optimized material
properties may be achieved without the need to use an unrealistically large falx thickness.
The upper value for the thickness range was derived from Takhounts et al. (2003), who
stated that this thickness value is larger than physiologically observed, but was used
anyway in an effort to avoid sharp contact along edges of the falx and to ease the
numerical contact interaction [3].
At first glance, there does not appear to be much improvement in the lateral
impact results after incorporating the optimized falx model (Table 4), considering the
61
shear modulus only decreases from 19.20 kPa to 17.26 kPa, which is still much greater
than 4.121 kPa and 4.634 kPa, the values for the occipital and frontal cases, respectively.
Looking closer at the relationships between CORA and G0 (Figure 28), however, there is
an interesting characteristic of the relationship in the lateral variation (Figure 28C) – a
local maximum, which was not present when using the original falx model. While the
global maximum value of CORA for this variation occurs at 17.26 kPa, there is a local
maximum at 5.6 kPa, which is much more comparable to the optimal shear modulus
values obtained from the occipital and frontal variations. Furthermore, the result of
concurrently optimizing the three impact conditions produces a shear modulus value very
near this local maximum.
The results of the brain parameter variation study in the lateral case using the
optimized falx model (Table 4), along with the results from the occipital and frontal
variations suggest that the best model response is achieved by using the brain material
model determined by a concurrent optimization of the C755-T2, C383-T1, and C291-T1
impact conditions. Examining the experimental NDT motion following the C755-T2
occipital impact, the response is generally characterized by a figure-eight or looping
pattern. Each NDT displays a small (~1 mm) initial displacement followed by a larger
displacement and rotates either counterclockwise or clockwise, depending on the z
location of the NDT with respect to the CG. Counterclockwise rotation is observed in the
NDTs closer to the CG of the head, while the rotation is clockwise nearing the top of the
head. In general, ABM validation against C755-T2 local displacements shows good
agreement with experimental data for nearly all NDTs. ABM response in the x-direction,
in general, displays very good agreement in magnitude and shape for all NDTs, excluding
62
A4, P4, and P5. We observe a magnitude discrepancy in A4 and P5, and in P4, there
appears to be a phase error. In the z-direction, there is very good shape and magnitude
agreement overall, however, there is a slight magnitude difference in A1 and a minor
shape difference in P1.
The ABM validation results from the frontal impact condition, C383-T1, show
good agreement with experimental data for many of the NDTs. As seen in Figure 30,
C383-T1 is a longer impact (118 ms) with a more complex response than the occipital
case (59 ms). In the C755-T2 case, the displacement responses, in general, are simple
curves with a single peak. For C383-T1, however, the responses have more complex
characteristics and usually contain multiple peaks. However, even with this added
complexity, the ABM predicts local brain displacements with comparable shape to the
experimental data. In the x direction, the ABM predicts response with very similar shape
to the experimental data for A1, A2, A5, A6, P1, P2, P5 and P6. While the overall shape
is similar, the ABM under-predicts displacement in the inferior NDTs, and over-predicts
displacement for more superior NDTs. For some NDTs, such as A4 and P3, the ABM
response generally agrees with the experimental data at some phases of the simulation,
while deviating at others. In the z direction, the ABM follows the overall trend of the
experimental data in most cases (A1-A5, P2-P6). When the predominant experimental
response is increasing displacement, the ABM also predicts increasing displacement
throughout the simulation, although in the case of the ABM we see more pronounced
local minimums and maximums when compared to the experimental curves, where the
response is steadier. Comparing the overall CORA score of the C755-T2 case (0.445) to
the C383-T1 case (0.421), we see that the ABM does not perform as well in the C383-T1
63
frontal configuration. This is not unexpected since the NDT displacement responses are
more complex in the longer, frontal case, and therefore are more difficult to match.
Despite this challenge, however, the ABM still predicts overall displacement magnitude
with acceptable accuracy.
The experimental NDT displacements, as well as the local displacements
predicted by the ABM, for the C291-T1 parietal impact are shown in Figure 31. For this
experimental condition, there were three columns of five NDTs; one column in the
hemisphere of the brain which was impacted (right), and two columns in the hemisphere
opposite to the location of impact (left). For the right NDT column, displacement results
were reported for only 3 of the 5 NDTs (1, 4 and 5), and for the posterior column on the
left side, results were reported for 4 of the 5 NDTs (2 - 5). Examining the results of the
right NDT column, the ABM predicts response very well in overall shape as well as
magnitude in the y-direction. The predicted displacement in the z-direction does not
agree quite as well with experimental data over the whole simulation, but does display
reasonably good agreement at the beginning (~50 ms) of the response. Considering the
two NDT columns in the left cerebral hemisphere, the predicted y response generally
shows good agreement with experimental displacement, with the exception of A4 and
A5. The ABM considerably underestimated the magnitude of the y-displacement for both
of these NDTs. In the z-direction we see a similar trend – very good agreement for all
NDT’s except A4 and A5. Even though A4 and A5 show the worst agreement of all the
z-direction responses of the left NDTs, the ABM did estimate the z-displacements for
these two NDTs much better in the z-direction than the y-direction.
64
The selection of material properties for the brain model and parameters is a vital
aspect of studying and understanding brain injury. There are several different material
models that are frequently used in current brain FEMs including linear elastic,
viscoelastic, and hyperelastic. Additionally, material parameter values differ significantly
throughout the literature due to the difficulty in characterizing the brain material
response. The range of values reported for shear properties in particular is quite large –
spanning two orders of magnitude (Table 2). Examining the values of shear moduli
reported throughout literature, we do see a large number of references to values between
5 and 15 kPa, despite the much larger overall range. The shear material properties
identified in this study are similar to those values consistently reported in literature.
4.1 Limitations
Limitations of this study include various simplifications and assumptions in the
model itself, as well as limitations of the experimental methods considered and their
application to FE models. Geometric simplifications of the ABM include the combined
gray and while matter in the cerebrum, as well as the omission of some other functionally
important brain structures, such as the corpus callosum. Also, while the cerebellum is
separate from the cerebrum, it does not have distinct material properties. Even with such
simplified geometry, however, the ABM was still able to achieve good overall response
and perform well in validation tests. Additional limitations are introduced by the
assumption that the brain is isotropic. We know from the examination of its complex
structure that brain matter is anisotropic - particularly white matter, which consists of
bundles of axons that are highly oriented [25]. The brain-skull interface is also simplified
in the ABM. It is modeled with shared nodes, which does not allow lateral motion at the
65
interface of the CSF and skull. While this is a simplification of the brain-skull interface,
we investigated other boundary conditions, such as surface to surface and sliding
contacts, and determined its formulation had little effect on the results of interest in the
current investigation. We concluded that the separation of the brain and skull with the
CSF material at all boundary locations was the most important and ensuring this resulted
in a good response. Another characteristic of the ABM that may introduce error is the
‘stair-stepped’ nature of the elements. Since the ABM was developed using a voxel-based
approach, all interfaces in the model possess cubic elements interacting with one another.
While this feature of the ABM is not ideal, we did not observe any unusual strains in the
edge elements at the boundary between the brain and CSF, probably owing to small
element size and the compliant boundary condition created by surrounding CSF.
Limitations of the application of the experimental conditions include errors introduced by
the difference in head geometry between the ABM and cadaveric specimens. The three
experimental configurations considered all used different cadavers with slightly different
head size and shape, which were all compared to the ABM, which represents the average
brain. However, the average brain dimensions for the three cadavers tested
experimentally did not differ greatly from the ABM. Additionally, there is error
associated with the location of the head CG since it was estimated with respect to the
Frankfurt plane using medical images.
4.2 Future Work
Future development of this model will involve a distinction between gray and
white matter, which will allow more differentiated and directional material models for
both. Additionally, there may be additional brain structures distinguished in the model,
66
such as the corpus callosum, brainstem, and meningeal layers. Additional improvements
to these material models may include the incorporation of directional anisotropy (white
matter), regional differences, if any, and age dependence. Also, adding a skull presents
great room for expanded capability. A more detailed skull model is of interest to be able
to model a wider variety of impacts, such as blunt impact which may involve skull
deformation. Improvement to the model may also include enhancement of the brain-skull
interface, which could involve smoothing of the elements resulting from the voxel-based
mesh generation method. Additionally, the voxel-based method presented in the current
study allows the production of patient or population-specific models in the future. The
detailed comparison of the model performance with other computational models from the
literature was beyond the scope and space limitations of the current study but will be
explored separately. Overall, the ABM CORA analysis demonstrated improvement over
the capabilities of existing, validated brain FE models, and is the first FE model to
determine brain material parameters through multi-objective optimization.
5 CONCLUSION
The model developed in the current study from the ICBM atlas is an anatomically
accurate brain model developed through voxel-based mesh generation. It is the first
dynamic biomechanical finite element brain model to include detailed three-dimensional
gyral folds for studying detailed brain deformation that has been validated in multiple
cadaver experiments against brain displacement data. It is also the first to utilize
multiobjective, multidirectional optimization with quantitative comparison between brain
deformations and experimental findings. A great deal of attention has been paid to the
67
boundary conditions of the model and many mechanical parameters which influence its
biofidelity. The high image resolution of the atlas used in this approach allows detailed
modeling of the intricate gyral folds on the surface of the brain, which enables the study
of local tissue response, particularly at the depths of the sulci. Since there is substantial
variation in reported values of brain material properties throughout the literature, and
direct quantitative comparison to experimental results is lacking in the literature, the
parameters governing the viscoelastic response of the brain and the displacements of
multiple locations were examined in great detail in this study. Material optimization was
performed using a novel method which simultaneously optimized model response in
three distinct experimental configurations to establish appropriate values for brain
material parameters. The use of multiple experimental conditions in this optimization
process allows researchers to verify that the identified properties are actually approaching
the true properties of the brain, as opposed to creating a model that is highly optimized to
one impact configuration of a single severity and direction or developed based on
inconsistent material testing methods. Additionally, CORA is applied to the results of
each simulation and used to quantify the error between experimental and predicted
displacements. CORA is demonstrated to be a powerful metric for evaluating model
performance, as it is able to capture multiple aspects of curve similarities, such as phase
and progression.
The CORA ratings reveal that the ABM performs well in all dynamic
experimental configurations investigated- including frontal, lateral, and occipital. Even
with known simplifications, the ABM is able to predict local brain displacements with
considerable accuracy when compared to experimental cadaver data, which supports the
68
model’s use in the study and simulation of brain injury. Initial validation of the ABM
shows competitive results when compared with existing FE models, and the anatomic
accuracy permits enhanced representations of local stress and strain. Investigation of
stress and strain distributions throughout the brain from FE models may help characterize
injury mechanisms, and the enhanced detail and accuracy provided by the ABM may
allow greater insight into region-dependent injuries, such as concussion and CTE. Further
development of the ABM will lead to improved ability to model brain response and
predict injury, which aid in a deeper understanding of the biomechanics of head injury.
Ultimately, the investigation of injury mechanisms and thresholds through the use of
finite element models can be used to predict and prevent traumatic brain injuries.
6 ACKNOWLEDGEMENTS
Funding for this project is provided by the National Institutes of Health (R01 NS082453).
All simulations were run on the DEAC Cluster at Wake Forest University. The authors
would like to thank the ANSIR lab for providing the ICBM labelmaps and Elizabeth
Lillie for her work on the MATLAB code to produce the ABM from the labelmaps.
69
REFERENCES
[1] T. R. Frieden, R. Ikeda, R. C. Hunt, M. Faul, L. Xu, M. M. Wald, and V. G.
Coronado, “Traumatic Brain Injury in the United States: Emergency Department
Visits, Hospitalizations and Deaths 2002-2006,” Natl. Cent. Inj. Prev. Control Cent.
Dis. Control Prev. US Dep. Health Hum. Serv., 2010.
[2] V. G. Coronado, L. Xu, S. V. Basavaraju, L. C. McGuire, M. M. Wald, M. D. Faul,
B. R. Guzman, and J. D. Hemphill, “Surveillance for Traumatic Brain Injury-
Related Deaths - United States, 1997–2007,” Morb. Mortal. Wkly. Rep., vol. 60, no.
5, May 2011.
[3] E. G. Takhounts, R. H. Eppinger, J. Q. Campbell, R. E. Tannous, E. D. Power, and
L. S. Shook, “On the Development of the SIMon Finite Element Head Model,”
Stapp Car Crash J., vol. 47, pp. 107–133, Oct. 2003.
[4] S. Kleiven and H. von Holst, “Consequences of head size following trauma to the
human head,” J. Biomech., vol. 35, no. 2, pp. 153–160, Feb. 2002.
[5] S. Ji, W. Zhao, Z. Li, and T. W. McAllister, “Head impact accelerations for brain
strain-related responses in contact sports: a model-based investigation,” Biomech.
Model. Mechanobiol., vol. 13, no. 5, pp. 1121–1136, Oct. 2014.
[6] A. I. King, K. H. Yang, L. Zhang, W. Hardy, and D. C. Viano, “Is head injury
caused by linear or angular acceleration,” presented at the IRCOBI conference,
2003, pp. 1–12.
[7] H. Kimpara, Y. Nakahira, M. Iwamoto, K. Miki, K. Ichihara, S. Kawano, and T.
Taguchi, “Investigation of anteroposterior head-neck responses during severe
frontal impacts using a brain-spinal cord complex FE model.,” Stapp Car Crash J.,
vol. 50, pp. 509–544, 2006.
[8] H. Mao, L. Zhang, B. Jiang, V. V. Genthikatti, X. Jin, F. Zhu, R. Makwana, A. Gill,
G. Jandir, A. Singh, and K. H. Yang, “Development of a finite element human head
model partially validated with thirty five experimental cases,” J. Biomech. Eng., vol.
135, no. 11, p. 111002, Nov. 2013.
[9] C. Lauret, M. Hrapko, J. A. W. van Dommelen, G. W. M. Peters, and J. S. H. M.
Wismans, “Optical characterization of acceleration-induced strain fields in
inhomogeneous brain slices,” Med. Eng. Phys., vol. 31, no. 3, pp. 392–399, Apr.
2009.
[10] J. E. Urban, J. A. Maldjian, C. T. Whitlow, and J. D. Stitzel, “A method to
investigate the size and shape variation of the lateral ventricles with age.,” Biomed.
70
Sci. Instrum., vol. 48, pp. 447–453, 2012.
[11] Pfefferbaum A, Mathalon DH, Sullivan EV, Rawles JM, Zipursky RB, and Lim KO,
“A quantitative magnetic resonance imaging study of changes in brain morphology
from infancy to late adulthood,” Arch. Neurol., vol. 51, no. 9, pp. 874–887, Sep.
1994.
[12] C. D. Good, I. Johnsrude, J. Ashburner, R. N. A. Henson, K. J. Friston, and R. S. J.
Frackowiak, “Cerebral Asymmetry and the Effects of Sex and Handedness on Brain
Structure: A Voxel-Based Morphometric Analysis of 465 Normal Adult Human
Brains,” NeuroImage, vol. 14, no. 3, pp. 685–700, Sep. 2001.
[13] C. D. Good, I. S. Johnsrude, J. Ashburner, R. N. Henson, K. Fristen, and R. S.
Frackowiak, “A voxel-based morphometric study of ageing in 465 normal adult
human brains,” NeuroImage, vol. 14, pp. 21–36, 2001.
[14] D. F. Meaney and D. H. Smith, “Biomechanics of concussion,” Clin. Sports Med.,
vol. 30, no. 1, pp. 19–31, 2011.
[15] C. M. Baugh, C. A. Robbins, R. A. Stern, and A. C. McKee, “Current understanding
of chronic traumatic encephalopathy,” Curr. Treat. Options Neurol., vol. 16, no. 9,
pp. 1–13, 2014.
[16] J. Mazziotta, A. Toga, A. Evans, P. Fox, J. Lancaster, K. Zilles, R. Woods, T. Paus,
G. Simpson, B. Pike, C. Holmes, L. Collins, P. Thompson, D. MacDonald, M.
Iacoboni, T. Schormann, K. Amunts, N. Palomero-Gallagher, S. Geyer, L. Parsons,
K. Narr, N. Kabani, G. L. Goualher, D. Boomsma, T. Cannon, R. Kawashima, and
B. Mazoyer, “A probabilistic atlas and reference system for the human brain:
International Consortium for Brain Mapping (ICBM),” Philos. Trans. R. Soc. Lond.
B Biol. Sci., vol. 356, no. 1412, pp. 1293–1322, Aug. 2001.
[17] M. Matsumae, R. Kikinis, I. A. Morocz, A. V. Lorenzo, T. Sandor, M. S. Albert, P.
M. Black, and F. A. Jolesz, “Age-related changes in intracranial compartment
volumes in normal adults assessed by magnetic resonance imaging,” J. Neurosurg.,
vol. 84, pp. 982–991, 1996.
[18] L. Zhang, K. H. Yang, and A. I. King, “Comparison of Brain Responses Between
Frontal and Lateral Impacts by Finite Element Modeling,” J. Neurotrauma, vol. 18,
no. 1, pp. 21–30, 2001.
[19] A. S. Al-Bsharat, W. N. Hardy, K. H. Yang, T. B. Khalil, S. Tashman, and A. I.
King, “Brain/skull relative displacement magnitude due to blunt head impacts: new
experimental data and model.,” Stapp Car Crash J., vol. 43, no. 2, pp. 321–332,
1999.
71
[20] T. J. Horgan and M. D. Gilchrist, “Influence of FE model variability in predicting
brain motion and intracranial pressure changes in head impact simulations,” Int. J.
Crashworthiness, vol. 9, no. 4, pp. 401–418, 2004.
[21] O. Jiroušek, J. Jíra, J. Jírová, and M. Micka, “Finite element model of human skull
used for head injury criteria assessment,” presented at the IUTAM Symposium on
Impact Biomechanics: From Fundamental Insights to Applications, 2005, pp. 459–
467.
[22] W. N. Hardy, C. D. Foster, M. J. Mason, K. H. Yang, A. I. King, and S. Tashman,
“Investigation of Head Injury Mechanisms Using Neutral Density Technology and
High-Speed Biplanar X-ray,” Stapp Car Crash J., vol. 45, pp. 337–368, Nov. 2001.
[23] W. N. Hardy, “Response of the Human Cadaver Head to Impact,” Wayne State
University, Detroit, MI, 2007.
[24] C. Gehre, H. Gades, and P. Wernicke, “Objective Rating of Signals Using Test and
Simulation Responses,” Pap. Present. 21st ESV Conf., Jun. 2009.
[25] M. T. Prange and S. S. Margulies, “Regional, directional, and age-dependent
properties of the brain undergoing large deformation,” J. Biomech. Eng., vol. 124,
no. 2, pp. 244–252, 2002.
72
Chapter III: Comparison of Validation Data for Finite Element
Models of the Human Head
Logan E. Miller, Jillian E. Urban, Joel D. Stitzel
Virginia Tech – Wake Forest University Center for Injury Biomechanics,
Winston-Salem, NC
73
ABSTRACT
Traumatic brain injury (TBI) is a leading cause of disability and injury-related death. To
prevent these injuries, we must develop a better understanding of the underlying injury
mechanisms. One way we can study brain injury is through finite element (FE) models.
In the current study, a high-resolution, anatomically accurate FE model, the ABM is
proposed as a new tool for studying brain injury with the ability to capture accurate local
stresses and strains throughout the sulci and gyri, which have been shown to be important
in injury prediction. Validation against local brain deformations for three experimental
cadaver tests (C755-T2, C383-T1, and C291-T1) is presented for the ABM and
quantitatively compared to other validated FEMs using a comparison metric called
CORA (CORrelation and Analysis). The model developed and presented in this study
will be used to better understand the fundamental mechanisms behind TBI and serve as a
tool to predict and prevent brain injury.
Keywords: Brain model, head injury, validation, optimization, finite element model,
CORA
74
1 INTRODUCTION
Each year, approximately 1.7 million people in the United States suffer from
traumatic brain injury (TBI) [1]. TBI is a major public health concern as it is a leading
cause of disability and injury-related death - accounting for nearly one third of all injury-
related deaths [2]. To prevent and treat these types of injuries, the fundamental injury
mechanisms need to be well-characterized and understood. There are various theories
about what causes brain injury, such as the development of positive and negative pressure
in coup and contrecoup injuries, rotational effects, and relative motion between the brain
and skull [3, 4, 5, 6]. Although each of these mechanisms has been shown to induce
injury, there is still a great deal that we do not know about the fundamentals of brain
injury, as well as injury thresholds. Various methods have been used over the years to
investigate and gain a deeper understanding of brain injury mechanisms, including animal
tests, cadaver studies, anthropomorphic test devices (ATDs), and computational models
[7]. Finite element (FE) models are powerful tools because they can provide spatial and
temporal distributions of stresses and strains over the problem domain. The quality of a
model’s predictions, however, is dependent on the accuracy of the modeled geometry and
the model’s ability to describe complex mechanical behavior and material response.
The earliest efforts to study the human brain through FE modeling was conducted
in 1943 by Hardy and Marcal [8], who constructed a two-dimensional (2D) FE model
consisting of only the skull. As computation power improved, three-dimensional (3D)
models became more achievable. The earliest 3D models generally used very simplified
(spherical or ellipsoidal) geometry and generally considered detailed material properties
75
of only either the brain or skull. An example of an early 3D FEM developed by Hosey
and Liu [9] in 1981 is shown in (Figure 32). It was not until the 1990s that we began to
Figure 32. (A) Sagittal and (B) coronal cross sections of the FEM developed by Hosey and
Liu [9] in 1981.
see models developed based on the true geometry of the human head. The first of its kind
was described by Ruan et al. [10] in 1993. Since then, FEMs with varying degrees of
anatomical accuracy, element size and material properties, have been developed. The
process typically employed in current FEM development begins with image segmentation
of medical images, such as computed tomography (CT) and magnetic resonance imaging
(MRI), to define 3D surfaces of desired structures, followed by a surface-based mesh
generation process. During the mesh generation process, complex anatomical features are
frequently simplified and ‘smoothed’ to allow faster development and simple contact
definition at model interfaces. One feature of particular importance that is over-simplified
in many brain FEMs is the gyral folds on the surface of the brain. The brain surface is
76
typically modeled as a smooth surface surrounded by a uniform layer of cerebrospinal
fluid (CSF), which ignores the intricate gyral folds throughout the brain. This
simplification and reduction in geometrical accuracy could significantly impact the
model’s local brain response and prediction of injury. Indeed, studies have indicated that
the intricate folding patterns on the surface of the brain greatly affect both the distribution
and magnitude of stress and strain throughout the brain [11]. Another limitation of
previous models is the substantial amount of time required for the surface-based mesh
generation process, which hinders production of patient-specific models. Patient-specific
models are desirable to study head injury due to significant variations in brain
morphology between individuals [12, 13, 14].
Another challenge in FE modeling of the human head is the determination of
accurate material properties for various anatomic structures. Despite over 50 years of
biomechanical testing and data, there are large variations in reported mechanical
properties of brain tissue. Reported values for shear relaxation moduli, for example, span
Figure 33. Relaxation moduli versus time from the literature [15].
77
orders of magnitude (Figure 33). Additionally, constitutive models for brain tissue vary in
the literature. These models include linear elastic models [5, 15, 16], linear and quasi-
linear viscoelastic models [17, 18, 19], hyperelastic [21], and fully nonlinear Green-
Rivlin models [21, 22].
Before numerical models are applied to injury prediction and simulation, they
must be validated against experimental data inputs to ensure that the model behaves the
way we expect. Brain FEMs are frequently validated against brain pressure and/or brain
motion data. The studies typically used to validate the pressure response are the cadaver
experiments conducted by Nahum et al. (1977) [24] and the long-duration impact
conducted by Troisseille et al. (1992) [25]. The localized brain motion data typically used
for validation purposes is the set of cadaver experiments conducted by Hardy et al. [13,
14]. Many currently used FEMs have only been validated intracranial pressure (ICP)
experiments, but it has been shown that the correct pressure response does not necessarily
mean the model is predicting the correct strain [27, 28]. Although the pressure data is
important to be able to replicate in a brain FEM, there is some controversy about whether
validating against only pressure data is sufficient, since some brain injuries (e.g., diffuse
axonal injury) are found to be dependent on strain [30]. This motivates validation using
the more robust displacement data set that captures the strain field that develops in the
brain, as opposed to a single value of pressure.
The objective of the current study is to present local brain motion validation
results for three experimental configurations for a high resolution, anatomically accurate
FEM, the ABM (atlas-based brain model), and compare results to other validated FEMs.
The three cadaver impact tests considered in the current study are: C755-T2, C383-T1,
78
and C291-T1 [13, 19]. The additional FEM’s considered are the Simulated Injury
Monitor (SIMon), Global Human Body Models Consortium (GHBMC) head model,
Total Human Model for Safety (THUMS) head model, Kungliga Tekniska Högskolan
(KTH) model, and the Dartmouth Head Injury Model (DHIM). Results for three of these
models, SIMon, GHBMC, and THUMS, were obtained through simulation of the
experimental conditions, whereas results for the remaining models were obtained from
published literature.
2 METHODS
The ABM FE model [32] was developed from the geometry of the ICBM adult
brain atlas, which represents the average adult brain [33] (see Appendix for model
development summary). The ICBM atlas contains 58 independently labeled regions of
the brain including cortical gyri, white matter, gray matter, and CSF. An FE model was
created from this image set by converting each 1mm isotropic voxel into a single element
of the same size using a custom code developed in MATLAB (The MathWorks, Natick,
MA) (Appendix . The ABM consists of a rigid skull, cerebrum, cerebellum, CSF,
ventricles, falx and tentorium (Figure 23). The brain-skull interface was modeled with a
soft CSF layer between the brain surface and the skull to allow relative motion between
the two through CSF deformation (all part interfaces are modeled with shared nodes).
Brain tissue in the ABM is modeled using a linear viscoelastic material formulation
(*MAT_061 in LS-DYNA). The shear relaxation behavior is characterized by:
𝐺(𝑡) = 𝐺∞ + (𝐺0 − 𝐺∞) ∗ 𝑒−𝛽𝑡 (11)
79
where G∞ is the infinite shear modulus, G0 is the initial shear modulus, and β is the decay
constant. The parameters for this material model were optimized in Miller et al. [32] and
are listed in Table 5.
Figure 34. (A) ABM showing the falx, tentorium, and ventricles, and (B) axial, (C) sagittal,
(E) coronal, (D) isometric cross sections.
2.1 FEM Description and Comparison
In the current study, the ABM response will be compared to five brain FEMs,
including two head models from full body FEMs. Brief descriptions and a comparison of
the models are provided in the following section and summarized in Table 5.
Simulated Injury Monitor (SIMon)
The SIMon model was developed by Takhounts and Eppinger in 2003 and
consists of relatively simple geometry with the goal of very short run time. The model
consists of a rigid skull, CSF, cerebrum, cerebellum, brain stem, ventricles, bridging
veins and a falx and tentorium (Figure 35). All part interfaces are modeled with merged
nodes and a Kelvin-Maxwell viscoelastic formulation is used to model brain material.
80
Figure 35. (A) Isometric view of the SIMon FEM with exposed brain and (B) isometric view
of the FEM showing various anatomical features represented in the model.
The model was improved upon and validated by Takhouts et al. [35] in 2008 against two
pressure experiments: Experiment 37 conducted by Nahum et al. [24] and the long-
duration impact conducted by Trosseille et al. [25], as well as the following local brain
displacement experiments conducted by Hardy et al. [17, 18]: C755-T2, C383-T1, and
C291-T1.
Global Human Body Models Consortium (GHBMC)
The GHBMC head model was meshed from a Computer Aided Design (CAD)
dataset developed from MRI and CT scans of an average adult male. The set included
geometry representing skin surface, skull and facial bones, sinuses, cerebrum,
cerebellum, lateral ventricles, corpus callosum, thalamus, and brainstem (Figure 36).
Geometry for cerebral white matter was also used to develop white matter meshes.
Aspects of the anatomy not included in the CAD dataset but that were implemented into
the model include the falx and tentorium, bridging veins, and the meningeal layers (pia,
arachnoid, dura). A tie-break contact was defined between the dura and arachnoid layers
to model the relative motion at the brain-skull interface. A Kelvin-Maxwell viscoelastic
81
Figure 36. (A) Isometric view of the GHBMC head model with the brain exposed and (B)
midsagittal view of the FEM [36].
model was used to model the gray and white matter. The pressure response of the
GHBMC head model was validated against six cases from the study conducted by
Nahum et al. [24] and one case from Trosseille et al. [25]. Additionally, Mao et al. [36]
stated that eight impact cases from the series of cadaver impact tests conducted by Hardy
et al. [17, 18] were selected for validation, although only results for the C383-T3 case
were presented in [36].
Total Human Model for Safety (THUMS)
The head model from the Total Human Model for Safety (THUMS) Version 4.01
(THUMS AM50 Ver 4.01, Toyota TCRDL, Japan) was used in the current study. Model
development for previous versions of the head model are described in Kimpara et al. [37].
The basic anthropometry of the skull was partially obtained from a commercial data
package (Viewpoint Datalabs, USA) and then modified based on anatomical references
[38]. The basic geometry of the brain model was obtained from a male data set available
82
in the Visible Human Project (NIH, USA) and was created according to anatomical
references. The THUMS head FEM consists of the skull and facial bones, cerebrum
(distinct white and gray matter), cerebellum, brainstem, CSF, meningeal membranes, falx
cerebri, tentorium cerebelli, and the sagittal sinus (Figure 37). To represent contact with
Figure 37. (A) Isometric view of the THUMS v4.01 head model with the brain exposed and
(B) midsagittal view of the FEM.
the CSF, tiebreak interfaces were used to simulate frictionless sliding behavior. This
contact formulation was assumed between the dura and arachnoid layers, as well as
between the falx and surrounding soft tissue elements. A linear viscoelastic material
model was used to model the response of the gray and white matter. A previous version
of the THUMS head model was validated against pressure and brain displacement
experiments by Kimpara et al. [37], but the current version has not yet been validated.
83
Kungliga Tekniska Högskolan (KTH)
Kleiven and Hardy proposed an FEM of the human head in 2002 known as the
Kungliga Tekniska Högskolan (KTH) FEM [12]. This model includes the scalp, skull,
cerebrum, cerebellum, meninges, CSF, bridging veins, and a simplified neck (Figure 38).
Figure 38. (A) Section view of the KTH head and neck model [28], and (B) cross section
showing FE mesh and anatomical structures represented in the model [12].
The brain-skull interface is modeled with tied nodes between the skull and dura with a
sliding contact with no separation between the CSF and the meningeal membranes and
brain. Brain tissue is modeled with a hyperelastic Mooney-Rivlin constitutive model
combined with a linear viscoelastic model to account for rate effects. The Mooney-Rivlin
and shear constants used in this model (Table 5) are based on those derived by Mendis et
al. [39], but scaled corresponding to an effective (long-term) modulus of 520 Pa, while
the decay constants were not altered [40]. The model was validated by Kleiven [40] in
2006 against two pressure experiments: Experiment 37 conducted by Nahum et al. [24]
and the long-duration impact conducted by Trosseille et al. [25], as well as the following
84
local brain displacement experiments conducted by Hardy et al. [17, 18]: C755-T2,
C383-T1, and C291-T1.
Dartmouth Head Injury Model (DHIM)
The Dartmouth Head Injury Model (DHIM) was created from a high-resolution
T1-weighted MRI of an athlete clinically diagnosed with concussion. The model features
a skull, facial bones, cerebrum (combined white and gray matter), cerebellum, brainstem,
corpus callosum, meningeal layers, CSF, ventricles, falx cerebri and tentorium cerebelli
(Figure 39). The brain-skull interface was modeled with a soft CSF layer between the
Figure 39. (A) Isometric view of the DHIM FEM and (B) isometric view of the FEM
showing various anatomical features represented in the model [48].
brain surface and the surrounding structures to allow brain interfacial sliding through
CSF deformation (nodes between all anatomical interfaces were shared) [41]. An Ogden
hyperelastic material model identical to the ‘average’ model reported by Kleiven [42] in
2007 was employed for brain tissue, in addition to a six-term Prony series characterizing
viscoelasticity. The model was validated by Ji et al. [43] against the following local brain
85
displacement experiments conducted by Hardy et al. [25, 26]: C755-T2, C383-T1, and
C393-T4.
Table 5. Comparison of FEMs
#elements/
#nodes
Mass
(kg)
Material
Model Brain Shear Parameters
ABM 2,000,000/
2,000,000 1.74
† Viscoelastic G0 = 5.16 kPa, G∞ = 1.86 kPa, β = 67.58 s
-1
SIMon 46,000/
42,500 1.10
† Viscoelastic G0 = 1.66 kPa, G∞ = 0.928 kPa, β = 16.95 s
-1
GHBMC 230,000/
190,000 1.19
† Viscoelastic
G0 = 6 kPa, G∞ = 1.2 kPa, β = 12.5 s-1
(gray)
G0 = 7.5 kPa, G∞ = 1.5 kPa, β = 12.5 s-1
(white)
THUMS 62,000/
38,000 1.08
† Viscoelastic G0 = 6 kPa, G∞ = 1.2 kPa, β = 0.06 s
-1
KTH 18,000/
19,000 4.44* Hyperelastic
G1 = 1628 Pa, G2 = 930 Pa, β1 = 125 s-1
,
β2 = 6.67 s-1
DHIM 115,000/
101,000 4.56* Hyperelastic
g1 = 7.69e-1, τ1 = 1e-6 s
g2 = 1.86e-1, τ2 = 1e-5 s
g3 = 1.48e-2, τ3 = 1e-4 s
g4 = 1.90e-2, τ4 = 1e-3 s
g5 = 2.56e-3, τ5 = 1e-2 s
g6 = 7.04e-3, τ6 = 1e-1 s †brain mass
*total model mass
2.2 Validation against localized brain displacements
The validation experiments considered in the current study are the following
cadaveric tests: C755-T2, C383-T1, and C291-T1 [13, 19]. Summary characteristics for
the three tests are shown in Table 3. In these experiments, local displacements at various
locations within the brain were analyzed† using implanted radio-opaque neutral density
targets (NDTS) and a high-speed biplanar X-ray system to track their relative motion.
Three-dimensional skull kinematics were evaluated with an accelerometer array affixed
to the cadaver skull and used to determine linear and angular velocities at the head center
of gravity (CG), which were then used as boundary conditions to drive each FEM
86
simulation [31]. The linear and rotational velocity curves for each experimental
configuration are shown in Figure 40. To compare local displacements between the ABM
Table 3. Summary of Experimental Test Conditions
Hardy C755-T2 Hardy C383-T1 Hardy C291-T1
Impact location occipital frontal parietal (R)
Impact type acceleration deceleration deceleration
Relative severity low medium high
Peak G 22 63 162
Delta V (m/s) 1.90 3.91 4.47
NDTs 2 columns of 5 2 columns of 6 3 columns of 5* *results for all NDTs in this test were not reported
and the experimental data, nodes closest to the physical location of each NDT were
identified for each test configuration and local displacements at these nodes were
computed throughout simulation. For each NDT, the element within which the initial
coordinate NDT location fell was located, and the 8 nodes used to define that cubic
element were tracked - the mean displacement of these 8 nodes was taken as the
displacement of the simulated NDT. Approximate NDT locations for the three impacts
are shown in Figure 41 in the plane of primary motion – for the occipital and frontal
impacts, the NDTs are displayed in the sagittal plane and for the lateral impact they are
displayed in the coronal plane. For the occipital and frontal cases, the locations of the
cross-sections are at the approximate y-location of the NDT columns – approximately 35
87
mm in the negative y direction for both cases. For the lateral case, Table 3 shows that
there are three columns of NDTs: anterior and posterior columns on the left side of the
Figure 40. Velocity load curves for the three experimental configurations.
brain and a second posterior column on the right side. The coronal cross sections to
visualize the NDT locations for the C291-T1 impact are taken near the CG of the head, or
between the anterior and posterior columns. In Figure 41C, the two posterior columns are
displayed in white and the anterior column in gray.
The results for the SIMon, GHBMC, and THUMS models were obtained through
simulation, in a similar process as described above for the ABM. For each model, the
velocity curves were applied to the rigid skull with respect to the head center of gravity
(CG). For NDT tracking, the element that encompassed the coordinate NDT location was
identified, and the 8 nodes defining that element were tracked and the mean displacement
was calculated. The average element size for each of these three models is larger than in
the ABM, so the average distance from the true NDT coordinate to each of the 8 nodes
88
was larger. Figure 41 shows the initial NDT locations for all three experimental
configurations for the ABM, SIMon, GHBMC and THUMS models.
Figure 41. Approximate NDT locations in ABM, SIMon, GHBMC, and THUMS models for
C755-T2 occipital impact (A), C383-T1 frontal impact (B) and C291-T1 parietal impact (C).
For the remaining models, validation results available in the literature were digitized for
comparison to results from the other FEMs. Results for all three tests for the KTH model
were reported in Kleiven (2006) [40]; not all NDTs were reported for the C291-T1 case,
however. For the DHIM model, Ji et al. (2014) [34] reported validation results for the
frontal and occipital cases only. Additionally, only the ‘corner’ NDTs were reported –
that is, the highest and lowest NDT in the anterior and posterior columns.
2.3 Model Performance
89
To evaluate model performance for each impact condition, the error between
experimental and predicted displacements for each NDT was quantified using a metric
called CORA (CORrelation and Analysis) [44]. CORA evaluates the similarity of two
curves by calculating a rating using two methods: the corridor method and the cross-
correlation method. Both ratings range from 0 to 1 and are averaged to determine a total
CORA score (1 corresponds to a perfect match). The corridor method computes a rating
based on where the comparison curve falls in relation to corridors around the reference
curve. The cross-correlation method is based on ratings for the phase shift, size and
progression of time-shifted curves.
3 RESULTS
The boundary conditions for each experimental configuration were applied to the
four simulated models using an identical method, but since the geometry of the four
models differs, there were slight variations in the linear kinematics of the CG of the head.
The rotational velocity of the CG was identical for all models since the rotational
boundary conditions were prescribed about this point, but we see small differences in the
linear velocity of the CG between the four models (Figure 42).
90
Figure 42. Linear velocity at head CG for ABM, SIMon, GHBMC, and THUMS
simulations.
The displacement results for the Hardy occipital impact (C755-T2) for the
anterior and posterior NDT columns are displayed in Figure 43 and Figure 44,
respectively. The displacement results for the frontal (C383-T1) impact for the anterior
and posterior NDT columns are displayed in Figure 45 and Figure 46, respectively.
Finally, the results from the parietal (C291-T1) impact for the right posterior column are
shown in Figure 47, and the left posterior and anterior columns are shown in Figure 48
and Figure 49, respectively. CORA scores for each displacement response for each model
are also displayed in Figure 43-Figure 49. Additionally, average CORA scores were
computed for each model under each of the three experimental configurations and are
displayed in Table 6.
91
Figure 43. Local displacement results for Hardy C755-T2 occipital impact for the anterior
NDT column.
92
Figure 44. Local displacement results for Hardy C755-T2 occipital impact for the posterior
NDT column.
93
Figure 45. Local displacement results for Hardy C383-T1 frontal impact for the anterior
NDT column.
94
Figure 46. Local displacement results for Hardy C383-T1 frontal impact for the posterior
NDT column.
95
Figure 47. Local displacement results for Hardy C291-T1 parietal impact for the right
posterior NDT column.
96
Figure 48. Local displacement results for Hardy C291-T1 parietal impact for the left
posterior NDT column.
97
Figure 49. Local displacement results for Hardy C291-T1 parietal impact for the left
anterior NDT column.
98
Table 6. CORA Scores for each model for all experimental configurations
C755-T2 C383-T1 C291-T1 Average
ABM 0.445 0.421 0.412 0.426
SIMon 0.342 0.411 0.393 0.382
GHBMC 0.433 0.386 0.241 0.353
THUMS 0.418 0.302 0.343 0.354
KTH 0.476 0.419 0.355 0.417
DHIM* 0.316 0.432 - 0.374 *only includes results for 4 ‘corner’ NDTs
4 DISCUSSION
To compare the performance of existing FEMs to experimental data and to each
other, the response under three distinct experimental configurations was examined. To
obtain results for each configuration, either the impact was simulated (for the ABM,
SIMon, GHBMC, THUMS) or results were digitized from existing literature (KTH,
DHIM). Validation data generated through simulation were verified with published data
when available. To verify the application of the boundary conditions for the simulated
models, kinematics of the head CG were tracked and compared to experimental data for
each model. The time history of angular velocity at the CG matched the applied angular
boundary conditions for each model since the boundary conditions were prescribed about
the coordinate system originating at the CG node; the linear velocity time history at the
CG, however, differed slightly between models (Figure 42) and from the applied
boundary curves since the size and geometry of the skull varied between the models. We
see similar trends for the three experimental impacts between the models. Overall, the
models track the experimental velocity curves well and verify that the boundary
conditions have been applied comparably between the four models.
99
4.1 C755-T2 Occipital Impact
Experimental Response
Examining the experimental x-displacement of the NDTs following the C755-T2
occipital impact, we observe that each NDT displays a small (~1 mm) initial
displacement in the negative x-direction. The initial negative x-displacement is followed
by a larger displacement in the positive x-direction that peaks at approximately 26 ms
with maximum displacements of larger magnitude for the lowest NDTs (A1-A3, P1-P3).
The initial motion in the z-direction is positive for all NDTs. This motion is most
pronounced in the lowest three NDTs in the anterior column (A1-A3) – for the remaining
NDTs, displacement remains very close to zero for approximately the first 10 ms. Each of
the anterior NDTs experiences a local minimum around 17 ms, followed by the
maximum positive z-displacement around 30 ms. The z- responses then approach a peak
in negative displacement that varies in timing between different NDTs – the minimum
generally occurs later in the posterior NDTs. In general, the magnitude of NDT
displacement was greater in the x-direction than in the z-direction for this impact.
Comparison of FEM Response
The models examined in the current study display similar response when
compared to each other, as well as to the experimental data. We see, in most cases, the
models predict NDT motion in the direction that it was experimentally measured. There
is, however, significant variation between the different models in the predicted
displacement magnitudes for each NDT. Looking at the response of the anterior NDTs
(Figure 43) the ABM, KTH, and GHBMC models display very similar responses to one
another and the best performance overall when compared to the experimental data. The
100
response deviates most drastically at the A1 location in the z-direction, where it appears
there may be a phase discrepancy, as well as over prediction by the models. The THUMS
model generally predicts motion of comparable magnitude and pulse width to
experimental displacements, but displays phase discrepancies at many of the NDT
locations (A3X and A4X). For numerous NDTs in the z-direction, the THUMS model
largely underestimates the displacement – specifically A1, A2 and A3. The SIMon model
agrees with experimental data very well in the x-direction for NDT A3, and relatively
well in the z-direction for NDTs A4 and A5, but the remaining responses do not perform
well. There is very little displacement predicted by the SIMon model at A1 and A2 in the
x direction, and at A1, A2, and A3 in the z direction, whereas the experimental data
display displacements between 2 and 3 mm. At NDTs A4 and A5 in the x-direction, a
significant phase error is observed. DHIM results were only reported for the ‘corner’
NDTs (A1, A5, P1, P5). The DHIM performs very well in the two anterior NDTs
reported except in the z direction for NDT A1, where there is a large under prediction in
displacement magnitude. In the posterior column, we again see similar responses for the
ABM, KTH, and GHBMC models for many of the NDTs, although we see diverging
behavior for P4 and P5 in the x-direction and P1 in the z-direction (Figure 44). The
remaining models display responses similar to their behavior in the anterior column. The
THUMS model generally performs better in the posterior column. The SIMon model
again predicts very small displacements at the most inferior NDTs (P1, P2, P3), and the
DHIM performs well at the corner NDTs.
101
4.2 C383-T1 Frontal Impact
The frontal impact experiment was longer in duration (118 ms) than the occipital
impact (64 ms). This increase in response time resulted in more complex displacement
signals than were observed in the shorter, C755-T2 impact.
Experimental Response
Similar to the C755-T2 experimental NDT response, there is an initial
displacement in the x-direction of approximately 1 mm, although for the frontal impact,
the initial displacements are in the positive x direction for most NDT locations, as
expected. Looking more closely at the experimental NDT response following the initial
displacement, we observe negative x- and z-motion in the inferior portion of the anterior
NDT column (A1-A3) until a peak occurs at approximately 25 ms; following this
minimum, the x- and z-displacements increase to reach maximum displacements around
90 ms ranging from 2 to 4 mm. The same pattern is observed, although slightly less
pronounced for NDTs A4, A5 and A6 in the z-direction. The x-response for these three
NDTs generally displays less total displacement, although all NDTs in the anterior
column do appear to reach a peak at approximately 90 ms. Looking at the posterior
column, for NDTs P1-P4, we again see behavior of the same shape as the response for
many of the anterior NDTs. The x-responses for the remaining NDTs (P5, P6), as well as
the z-responses for all posterior NDTs excluding P1, exhibit the behavior just described,
but of opposite polarity; there is initial negative displacement, a maximum around 25 ms,
followed by a minimum around 90 ms. The maximum displacement at these locations is
generally smaller than the x-displacement of inferior NDTs, but the trend still exists.
102
Comparison of FEM Response
Several models are able to capture the experimental displacement behavior for
many of the anterior NDTs. The ABM, KTH, and GHBMC models consistently
displayed this similar behavior to the experimental NDT motion. Both the ABM and
KTH models predict displacements that match the experimental shape very closely, but
there are magnitude discrepancies of varying size at different locations throughout the
brain. The GHBMC model displays good behavior for most anterior NDTs until about 90
ms, after which the predicted displacements deviate rather significantly from both the
experimental and other simulated data. SIMon predicts a displacement response with
comparable shape at a few anterior NDT locations (A2X, A3X, A3Z), but does not
reliably perform well when compared to experimental results. Examining the results in
the posterior column, all simulated responses appear to have captured the experimental
behavior well, with the exception of SIMon in the case of NDT P1. The overall shape is
accurately predicted by many models, as well as comparable values of displacement. The
largest discrepancies occur in the x-response of P4 and the z-responses of P1 and P3.
4.3 C291-T1 Parietal Impact
The final impact simulated in the current study is the C291-T1 parietal impact. It
is the most severe of the three considered, with a maximum head acceleration of 162 g’s.
This experiment was initialized with three columns of NDTs - an anterior and posterior
column in the left hemisphere of the brain (opposite to impact site), and a posterior
column in the right hemisphere (at impact site). Additionally, this is the only experiment
of the three considered where all data were not able to be collected. Two NDTs in the
right column and 1 NDT in the left posterior column were omitted from the published
103
experimental data. This could be attributed to the severity of the impact, especially since
two of the omitted NDTs were located on the side of the head that was impacted, and
therefore sustained more severe boundary conditions. For this reason, the results for the
right NDT column are frequently omitted in model validation studies, so we are only able
to compare results from this column for the four models specifically simulated in the
current study.
Experimental Response
Looking at the experimental response in the y-direction, or the direction of
impact, we see initial relative displacement in the negative y direction for NDTs PR4 and
PR5, which is expected as the brain lags behind the skull motion. The positive y-
displacement observed in the inferior-most NDT (PR1) likely occurs because this NDT is
closest to the skull, and therefore its motion does not lag behind skull movement like
locations deeper within the brain. The z-displacement for all the right NDTs oscillates
throughout the response and stays relatively close to zero, with a maximum z-
displacement of 2.2 mm. Considering the response of the other posterior NDT column
(left), we see an initial negative y-displacement, which is capturing the brain motion
lagging behind the skull. Similar to the right NDT column, the z-response displays less
overall motion than the y-direction. In the z-direction, there is a slight initial increasing
displacement, followed by a decreasing trend throughout the rest of the impact. For the
anterior column, the observed trend in y-response is an initial increase and peak around
15 ms, with subsequent decreasing y-displacement to a local minimum at approximately
20-30 ms. This overall trend is consistent in the y-response of the anterior NDTs, with the
exception of A5, whose response is initially in the negative y-direction. The remaining
104
response of A5, however, agrees with the rest of the anterior NDTs. The z-response of the
anterior NDTs is similar, in that there is a peak in positive displacement occurring around
15 ms, although less prominent in the z-direction than in the y-direction. The initial z-
response for all anterior NDTs, however, is in the negative direction. Overall, the
displacement magnitudes are much smaller in the z-direction when compared to the y-
direction.
Comparison of FEM Response
Looking at the response in the right posterior column, the ABM and THUMS
models generally track the experimental data, particularly in the y-direction, although the
ABM over predicts the initial displacement for the most inferior NDT (PR1). Both
models were able to capture the timing of local peaks throughout the response, as well as
overall magnitude for the NDT response in the y direction. In the z-direction, the ABM
and THUMS predictions deviate from the observed displacement, although both models
track relatively well at the beginning of the simulation, up to about 40 ms. The
displacement response predicted by the GHBMC model is much larger in magnitude than
was experimentally observed. This over prediction by GHBMC does not emerge until
about 30 ms into the simulation; the GHBMC estimation matches the actual response
fairly well up to that point. In the left posterior NDT column, all the models considered
capture the timing and magnitude of the initial y-response and peak and the behavior
immediately following the maximum, but we see some deviation approximately 30 ms
into the impact, specifically for NDTs PL2 and PL3. The GHBMC response deviates in
the y-response after 30 ms for all NDTs in this NDT column. A similar trend is observed
in the KTH and GHBMC z-response for this NDT column. The initial z maximum is
105
estimated by all the models reasonably well, but after about 30 ms, the KTH and GHMC
responses diverge from the experimental data. The GHBMC response diverges rather
significantly in the z-response for all NDTs. In contrast, the ABM, SIMon and THUMS
models display acceptable agreement throughout the entire simulation. Considering the
response of the anterior NDT column, the KTH and GHBMC models, in general, are able
to capture the initial y-response, but the simulated displacements diverge from the
experimental data as the simulation progresses. In contrast, the ABM, SIMon and
THUMS models under predict the initial peak, but they better match the experimental
response later in the simulation. Although the KTH and GHBMC models were able to
predict the large displacements seen experimentally in the y-direction, these models also
predicted large displacements in the z-direction, which were not observed experimentally.
The models that did not match the large initial y-displacement well (ABM, SIMon,
THUMS) because of under predicted displacement, performed comparatively better in
the z-direction for this NDT column.
4.4 Overall Performance
Looking at model response in the C755-T2 impact, we see that all models perform
relatively well, with the KTH moel scoring the highest average CORA rating of 0.476.
Most of the models considered score the highest CORA rating under this impact
configuration. This is likely because it is the lowest severity impact with the simplest
experimental curves (i.e. few local maxima and minima). In the C383-T1 test, the DHIM
has the highest CORA score. It should be noted, however, that this CORA rating is based
on the response of only 4 of the 12 total NDTs, so it may not be representative of the
response of the DHIM in general. Considering the remaining models, whose performance
106
is based on the response of all 12 NDTs, the ABM performs best with a CORA score of
0.421. Considering the response in the C291-T1 parietal configuration, which was the
most severe impact simulated, the ABM scores the highest CORA rating (0.412).
Additionally, looking at each model’s average CORA score between the three impacts,
the ABM again scores the best CORA rating. This result indicates that of the models
considered, the ABM demonstrates the strongest ability to predict local brain
deformations under a range of impact severities and directions.
5 CONCLUSION
The model presented in the current study, the ABM, is a high-resolution,
anatomically accurate brain model developed through voxel-based mesh generation. It is
the first dynamic biomechanical finite element brain model to include detailed three-
dimensional gyral folds for studying localized brain deformations that has been validated
in multiple cadaver experiments against brain displacement data. The high image
resolution of the atlas used in this approach allows detailed modeling of the intricate
gyral folds on the surface of the brain, which enables the study of local tissue response,
particularly at the depths of the sulci. The ABM is also the first to utilize multiobjective,
multidirectional optimization of material properties, which was motivated by the
substantial variation in reported values of brain material properties throughout the
literature. The material parameters of the ABM were optimized using a novel method that
simultaneously optimized model response in three distinct experimental configurations
and used CORA as the optimization parameter. The use of multiple experimental
conditions in this optimization process allows researchers to verify that the identified
107
properties are actually approaching the true properties of the brain, as opposed to creating
a model that is highly optimized to one impact configuration of a single severity and
direction or developed based on inconsistent material testing methods. Additionally, a
great deal of attention has been paid to the boundary conditions of the model and many
mechanical parameters that influence its biofidelity.
In the current study, validation is presented against local brain displacements.
This is considered a more robust validation metric than frequently used pressure data
because it represents the strain field throughout the brain. Additionally, it has been shown
that a correct pressure response does not necessarily mean the model is predicting the
correct strain [27, 28].
The validation results for three experimental configurations are presented for the
ABM and compared to the following existing FEMs of the human head: SIMon,
GHBMC, THUMS, KTH and DHIM. CORA was applied to the results for each model
and used to quantify the error between experimental and predicted displacements. CORA
is demonstrated to be a powerful metric for evaluating model performance, as it is able to
capture multiple aspects of curve similarities, such as phase and progression. The CORA
ratings revealed that the ABM performs well in all dynamic experimental configurations
investigated, which supports the model’s use in the study and simulation of brain injury.
Additionally, when compared to existing FEMs, the ABM performs the best in the range
of impacts considered.
Investigation of stress and strain distributions throughout the brain from FE
models may help characterize injury mechanisms, and the enhanced detail and accuracy
provided by the ABM may allow greater insight into region-dependent injuries. Further
108
development of the ABM will lead to improved ability to model brain response and
predict injury, which aid in a deeper understanding of the biomechanics of head injury.
Ultimately, the investigation of injury mechanisms and thresholds through the use of
finite element models can be used to predict and prevent traumatic brain injuries.
6 ACKNOWLEDGEMENTS
Funding for this project is provided by the National Institutes of Health (R01 NS082453).
All simulations were run on the DEAC Cluster at Wake Forest University. The authors
would like to thank the ANSIR lab for providing the ICBM labelmaps, the THUMS and
GHBMC teams at Wake Forest University for providing the respective head models, and
Elizabeth Lillie for her work on the MATLAB code to produce the ABM from the
labelmaps.
109
REFERENCES
[1] T. R. Frieden, R. Ikeda, R. C. Hunt, M. Faul, L. Xu, M. M. Wald, and V. G.
Coronado, “Traumatic Brain Injury in the United States: Emergency Department
Visits, Hospitalizations and Deaths 2002-2006,” Natl. Cent. Inj. Prev. Control Cent.
Dis. Control Prev. US Dep. Health Hum. Serv., 2010.
[2] V. G. Coronado, L. Xu, S. V. Basavaraju, L. C. McGuire, M. M. Wald, M. D. Faul,
B. R. Guzman, and J. D. Hemphill, “Surveillance for Traumatic Brain Injury-
Related Deaths - United States, 1997–2007,” Morb. Mortal. Wkly. Rep., vol. 60, no.
5, May 2011.
[3] V. Hodgson, L. M. Thomas, E. Gurdjian, O. Fernando, S. Greenberg, and J. Chason,
“Advances in understanding of experimental concussion mechanisms,” SAE
Technical Paper, 1969.
[4] A. G. Gross, “Impact thresholds of brain concussion.,” J. Aviat. Med., vol. 29, no.
10, pp. 725–732, 1958.
[5] A. Holbourn, “Mechanics of head injuries,” The Lancet, vol. 242, no. 6267, pp.
438–441, 1943.
[6] R. H. Pudenz and C. H. Shelden, “The lucite calvarium; a method for direct
observation of the brain; cranial trauma and brain movement,” J. Neurosurg., vol. 3,
no. 6, pp. 487–505, Nov. 1946.
[7] E. G. Takhounts, R. H. Eppinger, J. Q. Campbell, R. E. Tannous, E. D. Power, and
L. S. Shook, “On the Development of the SIMon Finite Element Head Model,”
Stapp Car Crash J., vol. 47, pp. 107–133, Oct. 2003.
[8] C. H. Hardy and P. V. Marcal, “Elastic Analysis of a Skull,” J. Appl. Mech., vol. 40,
no. 4, pp. 838–842, Dec. 1973.
[9] R. R. Hosey and Y. K. Liu, “A homeomorphic finite element model of the human
head and neck,” Finite Elem. Biomech., pp. 379–401, 1982.
[10] J. S. Ruan, T. B. Khalil, and A. I. King, “Finite element modeling of direct head
impact,” SAE Technical Paper, 1993.
[11] C. Lauret, M. Hrapko, J. A. W. van Dommelen, G. W. M. Peters, and J. S. H. M.
Wismans, “Optical characterization of acceleration-induced strain fields in
inhomogeneous brain slices,” Med. Eng. Phys., vol. 31, no. 3, pp. 392–399, Apr.
2009.
110
[12] S. Kleiven and H. von Holst, “Consequences of head size following trauma to the
human head,” J. Biomech., vol. 35, no. 2, pp. 153–160, Feb. 2002.
[13] J. E. Urban, J. A. Maldjian, C. T. Whitlow, and J. D. Stitzel, “A method to
investigate the size and shape variation of the lateral ventricles with age.,” Biomed.
Sci. Instrum., vol. 48, pp. 447–453, 2012.
[14] Pfefferbaum A, Mathalon DH, Sullivan EV, Rawles JM, Zipursky RB, and Lim KO,
“A quantitative magnetic resonance imaging study of changes in brain morphology
from infancy to late adulthood,” Arch. Neurol., vol. 51, no. 9, pp. 874–887, Sep.
1994.
[15] S. Chatelin, A. Constantinesco, and R. Willinger, “Fifty years of brain tissue
mechanical testing: from in vitro to in vivo investigations,” Biorheology, vol. 47,
no. 5, pp. 255–276, 2010.
[16] J. E. Galford and J. H. McElhaney, “A viscoelastic study of scalp, brain, and dura,”
J. Biomech., vol. 3, no. 2, pp. 211–221, 1970.
[17] G. T. Fallenstein, V. D. Hulce, and J. W. Melvin, “Dynamic mechanical properties
of human brain tissue,” J. Biomech., vol. 2, no. 3, pp. 217–226, Jul. 1969.
[18] L. Shuck and S. Advani, “Rheological response of human brain tissue in shear,” J.
Fluids Eng., vol. 94, no. 4, pp. 905–911, 1972.
[19] L. E. Bilston, Z. Liu, and et al., “Linear viscoelastic properties of bovine brain tissue
in shear,” Biorheology, vol. 34, no. 6, pp. 377–385, 1997.
[20] M. S. Estes and J. H. McElhaney, “Response of brain tissue to compressive
loading,” ASME No 70-BHF-13, vol. 92, p. 58, 1970.
[21] S. Nicolle, M. Lounis, and R. Willinger, “Shear Properties of Brain Tissue over a
Frequency Range Relevant for Automotive Impact Situations: New Experimental
Results.,” Stapp Car Crash J., vol. 48, pp. 239–258, 2004.
[22] K. Darvish and J. Crandall, “Nonlinear viscoelastic effects in oscillatory shear
deformation of brain tissue,” Med. Eng. Phys., vol. 23, no. 9, pp. 633–645, 2001.
[23] E. G. Takhounts, J. R. Crandall, and K. Darvish, “On the importance of nonlinearity
of brain tissue under large deformations,” Stapp Car Crash J., vol. 47, p. 79, 2003.
[24] A. M. Nahum, R. Smith, and C. C. Ward, “Intracranial pressure dynamics during
head impact,” SAE Technical Paper, 1977.
111
[25] X., Domont, A., Guillon, F., Lavaste, F., Tarriere, C., Trosseille, Development of a
F.E.M. of the human head according to a specific test protocol. 1992.
[26] W. N. Hardy, C. D. Foster, M. J. Mason, K. H. Yang, A. I. King, and S. Tashman,
“Investigation of Head Injury Mechanisms Using Neutral Density Technology and
High-Speed Biplanar X-ray,” Stapp Car Crash J., vol. 45, pp. 337–368, Nov. 2001.
[27] W. N. Hardy, M. J. Mason, C. D. Foster, C. S. Shah, J. M. Kopacz, K. H. Yang, A.
I. King, J. Bishop, M. Bey, W. Anderst, and S. Tashman, “A study of the response
of the human cadaver head to impact,” Stapp Car Crash J., vol. 51, pp. 17–80, Oct.
2007.
[28] S. Kleiven and W. N. Hardy, “Correlation of an FE Model of the Human Head with
Local Brain Motion--Consequences for Injury Prediction,” Stapp Car Crash J., vol.
46, pp. 123–144, Nov. 2002.
[29] S.-C. Huang, “Numerical simulations of human head-neck dynamics.,” Biomed.
Mater. Eng., vol. 9, no. 1, pp. 61–71, 1998.
[30] K. M. Tse, L. B. Tan, S. J. Lee, S. P. Lim, and H. P. Lee, “Development and
validation of two subject‐specific finite element models of human head against three
cadaveric experiments,” Int. J. Numer. Methods Biomed. Eng., vol. 30, no. 3, pp.
397–415, 2014.
[31] W. N. Hardy, “Response of the Human Cadaver Head to Impact,” Wayne State
University, Detroit, MI, 2007.
[32] L. E. Miller, “Development and Validation of an Atlas-Based Finite Element Brain
Model,” Biomech Model Mechan, In Review 2015.
[33] J. Mazziotta, A. Toga, A. Evans, P. Fox, J. Lancaster, K. Zilles, R. Woods, T. Paus,
G. Simpson, B. Pike, C. Holmes, L. Collins, P. Thompson, D. MacDonald, M.
Iacoboni, T. Schormann, K. Amunts, N. Palomero-Gallagher, S. Geyer, L. Parsons,
K. Narr, N. Kabani, G. L. Goualher, D. Boomsma, T. Cannon, R. Kawashima, and
B. Mazoyer, “A probabilistic atlas and reference system for the human brain:
International Consortium for Brain Mapping (ICBM),” Philos. Trans. R. Soc. Lond.
B Biol. Sci., vol. 356, no. 1412, pp. 1293–1322, Aug. 2001.
[34] S. Ji, W. Zhao, Z. Li, and T. W. McAllister, “Head impact accelerations for brain
strain-related responses in contact sports: a model-based investigation,” Biomech.
Model. Mechanobiol., vol. 13, no. 5, pp. 1121–1136, Oct. 2014.
[35] E. G. Takhounts, S. A. Ridella, V. Hasija, R. E. Tannous, J. Q. Campbell, D.
Malone, K. Danelson, J. Stitzel, S. Rowson, and S. Duma, “Investigation of
traumatic brain injuries using the next generation of simulated injury monitor
112
(SIMon) finite element head model,” Stapp Car Crash J., vol. 52, pp. 1–31, Nov.
2008.
[36] H. Mao, L. Zhang, B. Jiang, V. V. Genthikatti, X. Jin, F. Zhu, R. Makwana, A. Gill,
G. Jandir, A. Singh, and K. H. Yang, “Development of a finite element human head
model partially validated with thirty five experimental cases,” J. Biomech. Eng., vol.
135, no. 11, p. 111002, Nov. 2013.
[37] H. Kimpara, Y. Nakahira, M. Iwamoto, K. Miki, K. Ichihara, S. Kawano, and T.
Taguchi, “Investigation of anteroposterior head-neck responses during severe
frontal impacts using a brain-spinal cord complex FE model.,” Stapp Car Crash J.,
vol. 50, pp. 509–544, 2006.
[38] C. D. Clemente, Gray’s Anatomy, 30th American Edition of the Anatomy of the
Human Body by Henry Gray, 1825-1861, Ed. (1985). PA: Lea & Febiger, 1985.
[39] K. Mendis, R. Stalnaker, and S. Advani, “A constitutive relationship for large
deformation finite element modeling of brain tissue,” J. Biomech. Eng., vol. 117, no.
3, pp. 279–285, 1995.
[40] S Kleiven, “Evaluation of head injury criteria using a finite element model validated
against experiments on localized brain motion, intracerebral acceleration, and
intracranial pressure,” Int. J. Crashworthiness, vol. 11, no. 1, pp. 65–79, 2006.
[41] S. Ji, W. Zhao, J. C. Ford, J. G. Beckwith, R. P. Bolander, R. M. Greenwald, L. A.
Flashman, K. D. Paulsen, and T. McAllister, “Group-wise evaluation and
comparison of white matter fiber strain and maximum principal strain in sports-
related concussion,” J. Neurotrauma, no. ja, 2014.
[42] S. Kleiven, “Predictors for traumatic brain injuries evaluated through accident
reconstructions,” Stapp Car Crash J., vol. 51, pp. 81–114, Oct. 2007.
[43] S. Ji, H. Ghadyani, R. P. Bolander, J. G. Beckwith, J. C. Ford, T. W. McAllister, L.
A. Flashman, K. D. Paulsen, K. Ernstrom, S. Jain, R. Raman, L. Zhang, and R. M.
Greenwald, “Parametric comparisons of intracranial mechanical responses from
three validated finite element models of the human head,” Ann. Biomed. Eng., vol.
42, no. 1, pp. 11–24, Jan. 2014.
[44] C. Gehre, H. Gades, and P. Wernicke, “Objective Rating of Signals Using Test and
Simulation Responses,” Pap. Present. 21st ESV Conf., Jun. 2009.
113
APPENDIX. MODEL DEVELOPMENT
114
115
Chapter IV: Summary of Research
The work presented in this thesis represents an important contribution to the study and
simulation of head injury. The ABM is proposed as a new method for simulating brain
response that possesses the novel qualities:
1) High-resolution representation of gyral folds on the surface of the brain.
2) Multi-objective parameter optimization of the brain material model against the
local brain displacements of three cadaver experiments.
3) The best performance of currently used FE models of the human head in a range
of experimental configurations of varying severity and direction, quantified using
CORA.
Investigation of stress and strain distributions throughout the brain from FE models may
help characterize injury mechanisms, and the enhanced detail and accuracy provided by
the ABM may allow greater insight into region-dependent injuries, such as concussion
and CTE. Ultimately, the investigation of injury mechanisms and thresholds through the
use of finite element models can be used to predict and prevent traumatic brain injuries.
Research presented in Chapter II and Chapter III is expected to be published in scientific
journals as listed in Table 7.
Table 7. Publication plan for research outlined in this thesis
Chapter Topic Journal
II
Development and Validation of an
Atlas-Based Finite Element Brain
Model
Biomechanics and Modeling in
Mechanobiology†
III
Comparison of Validation Data for
Finite Element Models of the Human
Head
Journal of Neurotrauma*
†Submitted
*Manuscript in preparation
116
Scholastic Vita
Logan E. Miller Graduate Student and Research Engineer Virginia Tech-Wake Forest University Center for Injury Biomechanics 575 N. Patterson Ave, Suite 120, Winston-Salem, NC 27101
Work: 336-716-0946 Cell: 540-908-1279
Email: [email protected]
EDUCATION Virginia Tech – Wake Forest University, Winston-Salem, NC Master of Science, Biomedical Engineering 2013 – Present Thesis: Development and Validation of an Atlas-Based Brain Finite Element Model Research Lab: Virginia Tech – Wake Forest University Center for Injury Biomechanics Advisor: Dr. Joel Stitzel Virginia Tech, Blacksburg, VA Bachelor of Science, Engineering Science and Mechanics 2009 – 2013 Minor: Mathematics
PROFESSIONAL EXPERIENCE Graduate Research Engineer 2013 – Present Virginia Tech- Wake Forest University Center for Injury Biomechanics Wake Forest School of Medicine Winston-Salem, NC - Toyota’s Collaborative Safety Research Center - Development of injury metrics using the Total Human Model for Safety
Undergraduate Research Intern Summer 2012 Virginia Tech-Wake Forest University Center for Injury Biomechanics Winston-Salem, NC Undergraduate Research Intern 2012-2013 Virginia Tech Comparative Biomechanics Lab Blacksburg, VA
LEADERSHIP
Session Tutor 2011 –2013
Academic tutor at Virginia Tech
Led tutoring sessions for individual classes
Social Chair, Engineering in Medicine and Biology Society 2013 – 2014 COMPUTING SKILLS - LS-DYNA
- LS-PrePost
- Matlab
- Microsoft Office
- Geomagic Studios
- Hyperworks
- R
- JMP
- Avizo
PROFESSIONAL AND STUDENT MEMBERSHIPS
Biomedical Engineering Society, VT-WFU Chapter 2013 - Present
Engineering in Medicine and Biology Society, VT-WFU Chapter 2013 - Present
117
PROFESSIONAL CERTIFICATION AND WORKSHOPS
LS-DYNA Fracture Training, Winston-Salem, NC March 2015
LS-DYNA Material Models Training, Winston-Salem, NC February 2014
LS-DYNA Contacts Training, Winston-Salem, NC January 2014
Hyperworks Training, Winston-Salem, NC October 2013
LS-PrePost and LS-DYNA Training, Troy, MI August 2013
BIBLIOGRAPHY
Papers in Refereed Publications
1. Golman AJ, Danelson KA, Miller LE, Stitzel JD. “Injury Prediction in a Side Impact Crash
Using Human Body Model Simulation.” Accident Analysis & Prevention, October 2013;
64:1-8.
2. Miller LE, Urban JE, Stitzel JD. “Development and Validation of an Atlas-Based Brain
Finite Element Model.” Biomechanics and Modeling in Mechanobiology, July 2015. (In
Review)
Papers in Refereed Conference Publications
1. Miller LE, Urban JE, Lillie EM, Stitzel JD. “Preliminary Development and Validation of an
Atlas-Based Finite Element Brain Model.” Biomed Sci Instrum. April 2015; 51:253-259.
2. Kelley, ME, Miller LE, Urban JE, Stitzel JD. “Mesh Smoothing Algorithm Applied to a
Finite Element Model of the Brain for Improved Brain-Skull Interface.” Biomed Sci
Instrum. April 2015; 51:253-259.
Other Conference Papers (Abstract Style or Non-Refereed)
1. Miller LE, Urban JE, Stitzel JD. “Development and Validation of an Atlas-Based Brain
Finite Element Model.” Summer Biomechanics, Bioengineering and Biotransport
Conference, Snowbird Resort, UT, June 2015.
2. Miller LE, Urban JE, Stitzel JD. “Development and Validation of an Atlas-Based Brain
Finite Element Model.” North American Brain Injury Society Annual Meeting, San
Antonio, TX, May 2015.
3. Miller LE, Urban JE, Stitzel JD. “Development and Validation of an Atlas-Based Brain
Finite Element Model.” VT-WFU SBES Graduate Research Symposium Oral
Presentation, Virginia Tech, Blacksburg, VA, May 2015.
4. Miller LE, Urban JE, Lillie EM, Stitzel JD. “Development of an Atlas-Based Finite Element
Head Model.” Biomedical Engineering Society Annual Fall Scientific Meeting, San
Antonio, TX, October 2014.
5. Miller LE, Urban JE, Lillie EM, Stitzel JD. “Development of a Voxel-Based Finite Element
Head Model.” VT-WFU SBES Graduate Research Symposium Poster Session, Wake
Forest, Winston-Salem, NC, May 2014.
Technical Reports
1. Gaewsky JG, Marcus IP, Jones DA, Kelley ME, Miller LE, Koya B, Weaver AA, Stitzel JD.
“Crash Simulation using the Total Human Model for Safety.” Year 4, Quarter 2, Report to
Toyota’s Collaborative Safety Research Center, Ann Arbor, MI, March, 2015.
118
2. Stitzel, JD, Danelson KA, Golman AJ, Gaewsky JG, Miller LE, Weaver CM. “Crash
Simulation using the Total Human Model for Safety.” Year 1 Annual Report to Toyota’s
Collaborative Safety Research Center, Ann Arbor, MI, September, 2012.