Biomechanics and Modeling in Mechanobiology

https://doi.org/10.1007/s10237-019-01124-6

ORIGINAL PAPER

Homogenization of heterogeneous brain tissue under quasi‑static

loading: a visco‑hyperelastic model of a 3D RVE
Morteza Kazempour1 · Majid Baniassadi1 · Hamid Shahsavari1 · Yves Remond2 · Mostafa Baghani1 

Received: 27 August 2018 / Accepted: 4 February 2019

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Researches, in the recent years, reveal the utmost importance of brain tissue assessment regarding its mechanical properties,
especially for automatic robotic tools, surgical robots and helmet producing. For this reason, experimental and computational
investigation of the brain behavior under different conditions seems crucial. However, experiments do not normally show the
distribution of stress and injury in microscopic scale, and due to various factors are costly. Development of micromechanical
methods, which could predict the brain behavior more appropriately, could highly be helpful in reducing these costs. This
study presents computational analysis of heterogeneous part of the brain tissue under quasi-static loading. Heterogeneity is
created by irregular distribution of neurons in a representative volume element (RVE). Considering time-dependent behavior
of the tissue, a visco-hyperelastic constitutive model is developed to predict the RVE behavior more realistically. The RVE
is studied in different loads and load rates; 1, 2, 3, 10 and 15% strain load are applied at 0.03 and 0.2 s on the RVE as tensile
and shear loads. Due to complexity in geometry, self-consistent approximation method is employed to increase the volume
fraction of neurons and analyze RVE behavior in various NVFs. The results show increasing the load rate leads to a raise in
the maximum stress that indicates the tissue is more vulnerable at higher rates. Moreover, stiffness of the tissue is enhanced
in higher NVFs. Additionally, it is found that axons undergo higher stresses; hence, they are more sensitive in accidents
which lead to axonal death and would cause TBI and DAI.

Keywords  Visco-hyperelastic · Traumatic brain injury (TBI) · Diffuse axonal injury (DAI) · FEM

1 Introduction damage originating from brain impact or TBI (Peter and

An external force, causing an alteration in brain function, or
other brain pathologies, is defined as traumatic brain injury
(TBI) (Menon et al. 2010). TBI is one of the well-known
public health problems. In the United States, for example,
about 1.7 million TBIs occur per year; falls are the most
important external cause, resulting in 35.2% of TBIs. Next
major causes are motor vehicle accidents, struck by/against
injuries and assaults leading to 17.3%, 16.5% and 10% of
TBIs, respectively (Faul et al. 2010). Unfortunately, symp-
toms of TBI are often invisible to the eye (Tanielian et al.
2008). Axonal death and disorientation can result from axon
* Mostafa Baghani
baghani@ut.ac.ir
1
School of Mechanical Engineering, College of Engineering,
University of Tehran, Tehran 11155‑4563, Iran
2
IMFS/CNRS, University of Strasbourg, 2 Rue Boussingault,
67000 Strasbourg, France
Mofrad 2012). Damage and impacts are one of the topics
widely studied in mechanics. From a biomechanics point
of view, brains can also be simulated and regarded as an
object/material; hence, it is very important for them to be
characterized. With the advances made in finite element
(FE) modeling of brain tissue in recent years, studies on
TBI have increasingly made use of computer simulations.
However, the precision of material properties used in these
models affects the results dramatically (Laksari et al. 2012;
Karami et al. 2009).
In recent years, many different experimental test setups
have been developed to study brain matter, especially inves-
tigating mechanical properties of the brain tissue. Atomic
force microscopy (AFM) (Bernick et al. 2011), micropipette
aspiration (Schmid-Schönbein et al. 1981), scanning force
microscopy (SFM) (Christ et al. 2010), transmission elec-
tron microscopy (TEM) (Labus and Puttlitz 2016), magnetic
resonance elastography (MRE) (Hiscox et al. 2016; Vap-
pou et al. 2007), custom-designed devices (Feng et al. 2013;

13
Vol.:(0123456789)

Budday et al. 2015; Shulyakov et al. 2009) as well as digital
image correlation (DIC) technique (Libertiaux et al. 2011;
Coudrillier et al. 2013) are the most important tools used
to evaluate the mechanical properties of the brain tissue.
Rashid et al. (2014) used a custom-designed device to per-
form dynamic tensile loading in various strain rates. Bernick
et al. (2011), implemented an indentation test on a neuron
cell using AFM. In their work, the soma is examined in vari-
ous loading and relaxation conditions using AFM, and the
experimental outcomes were employed to derive material
properties. Libertiaux et al. (2011) used DIC technique to
verify incompressibility of the brain tissue under different
loading rates. They showed that the brain tissue is incom-
pressible regardless of the sample preparation method and
mechanical properties of the tissue are independent of the
sample preparation method.
Macro scale FE simulation could also be a powerful way
to investigate material properties of the brain tissue (Forte
et al. 2018; Ganpule et al. 2013; Laksari et al. 2015; Ji
et al. 2014). Maltese and Margulies (2016) constructed a
FE model of the brain tissue using images of CT and MRI
scans and found that some regions of the white matter are
stiffer than those of the gray matter. Kleiven and von Holst
(2002) employed 2D images of CT and MRI scans of male
and female bodies to create a 3D FE model of the brain.
They investigated the effect of the head size on the impact
and inertial loads.
Representative volume element (RVE) is a realized model
that can be used in FE simulations to determine the effec-
tive properties of a macroscopic homogeneous material.
A suitable size selection for a RVE would yield accurate
information on the microstructure (Hashin 1983). RVE
concept has been vastly employed to model microstructures
(Yousefi et al. 2017) and biologics materials (Sheidaei et al.
2013; Abdel Rahman et al. 2012; Cloots et al. 2013). Some
researchers have been using the repeating unit cell (RUC)
(Karami et al. 2009; Abolfathi et al. 2009) instead of RVE, in
order to simplify simulations in the study of the brain tissue.
Karami et al. (2010) simulated axons as cylinders within an
extra cellular matrix (ECM), considering a RUC with dif-
ferent volume fractions.
Optimization functions are employed as useful tools to
determine all coefficients of the utilized constitutive model.
Javid et al. (2014) performed an experimental test on a pig
brainstem and derived its viscoelastic model coefficients
using genetic algorithm (GA). RVE -including unidirectional
axons and a matrix was simulated as a geometrical model.
It was found that the axon is three times stiffer than ECM.
Sotudeh-Chafi et al. (2008) also employed GA method to
identify parameters of a viscoelastic model for the brain
tissue. Pan et al. (2013) used a pseudo-3D RVE in which
axons were embedded into the matrix. Ogden’s energy func-
tion was employed into a FE model, and the squared error
13
M. Kazempour et al.
between the force–stretch experimental curve and the simu-
lated model was considered as the objective function.
Despite the importance of macro scale analysis, it does
not reveal how the macro scale stresses and deformations
of the tissue are transferred into the microscale. Microscale
modeling and simulation of the brain tissue can expand our
understanding of micro-mechanisms and help in gaining
a better prediction of the brain tissue behavior in a larger
scale. Karami et al. (2009) conducted a micromechanical
analysis on the brain white matter. They modeled axons as
undulant hyperelastic cylinders in a RUC. Their results show
an increasing trend of stresses in both axons and ECM due
to an increase in volume fraction of axons. The model did
not account for time-dependent behavior of the brain tis-
sue, and the neuron cell was not considered in the simu-
lation. Considering time-dependent behavior of the brain
tissue, viscoelastic models are popular when predicting
the tissue properties. Laksari et al. (2012) implemented a
visco-hyperelastic model to characterize dynamic material
properties of the brain tissue. In their work, hyperelastic
material parameters were determined through fitting the
isochronous curves derived from the model to the experi-
mental results in various strain levels. The viscoelastic
coefficients were determined by fitting convolution model
outcomes to the relaxation experimental results. Although
the material properties in tissue level were identified, this
study demonstrated neither the stress distributions in neuron
and ECM separately nor their material properties. Miller
(1999) developed a linear viscoelastic model and evaluated
the brain tissue behavior under compressive loads at low
strain rates. While the single-phase linear viscoelastic model
detected stress–strain curves, it did not show the distribu-
tion of stress on components of the brain. Abolfathi et al.
(2009) assumed the axons and the matrix as a linear viscoe-
lastic material in RUC without considering the dendrites or
the soma. They validated simulation data with experimen-
tal observations; the results revealed that the axon volume
fraction is more effective than the axon’s undulation on the
brain matter properties. Moreover, viscoelastic parameters of
some heterogeneous regions of the rat brain were measured
using an optical coherence tomography (OCT) by Lee et al.
(2014). They showed that various regions of the brain have
different mechanical behaviors.
In this study, the brain tissue is micromechanically stud-
ied using a RVE. Brain tissue includes neurons, and ECM
is considered as isotropic visco-hyperelastic materials. A
VUMAT ABAQUS subroutine with two terms of viscoelas-
tic and one term of hyperelastic parameters is developed to
define the material properties of the RVE. Neurons include
neurites and a soma, distributed randomly in a realistic 3D
RVE, which would be analyzed under five strain loads in two
various strain rates. Assuming perfect bonding between the
neurons and the matrix, homogenization of the brain tissue is
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…

studied. This model is found to be suitable for homogeniza-

tion of the heterogeneous part of the isotropic brain tissue.
With respect to the complexity in the geometry of the model,
the homogenization of the study was carried out using the
self-consistent method (Rémond et al. 2016). Taking advan-
tage of the abovementioned techniques has increased the
accuracy of RVE simulation significantly, leading to a more
realistic representation of stress distribution between neu-
rons and matrix.

2 Constitutive model and simulation

2.1 RVE construction

Considering the fact that the macroscopic dimensions of the

brain compared to its components are so large, the use of FE
simulation to study the tissue involves high computational
costs. That is one of the reasons why homogenization pro- Fig. 1  The distribution of neurons in ECM
cedures were developed. All of these procedures use a com-
mon definition of composite materials being reconstructed
by repeating a volume element. RVE is the smallest volume
assumed that RVE follows the orthotropic model, incorporat-
element of a heterogeneous material that is large enough
ing nine independent coefficients (Eq. 1).
to be statistically similar to macroscale composite (Kanit
et al. 2003). A 3D RVE consisting of neurons and ECM is 𝜀ij = Sijkl 𝜎kl (1)
constructed for studying the brain tissue. The optimum size
of RVE is chosen by comparing the results of different sizes ⎡ 1 0 ⎤
−𝜈21 −𝜈31
0 0
of RVEs. Moreover, the simulated neurons are constituted ⎢ −𝜈E1 E2
1
E3
−𝜈32
⎥
of a spherical neuron cell, five undulate dendrites and an ⎢ E 12 0 0 0 ⎥
⎢ −𝜈113 E2 E3
⎥
undulate axon. It is noteworthy that the neurites have sinu- ⎢
−𝜈23 1
0 0 0 ⎥
soidal geometry. S = ⎢ E1 E2 E3
1 ⎥
A computational code has been developed as a MATLAB ⎢ 0 0 0 G23
0 0 ⎥
⎢ 0 0 0 0 1
0 ⎥
script to randomly distribute neurons and check the distance ⎢ G31 ⎥
between the neurons to avoid any intersection of them into ⎢ 0 0 0 0 0 1
⎥
⎣ G12 ⎦
RVE. Therefore, a neuron is created as the reference one;
then, other neurons are distributed by random translation
and rotation matrixes. Neurons volume fraction (NVF), RVE
size, reference neuron coordinates and minimum distance where Ei , Gij and 𝜈ij are Young’s modulus, shear modulus
between neuron cells are given as input values to the code, and Poisson’s ratios in different directions, separately. There-
and reference neuron coordinates, translation and rotation fore, six tests including three tensile and three shear tests in
matrixes are saved as output values. In Fig. 1, the distribu- different directions are conducted. Through a comparison of
tion of neurons in RVE is shown. the coefficients of orthotropic model, the isotropy is studied.
Elastic constants for the neurons and ECM are summarized
in Table 2.
2.2 Constitutive model After isotropy check and determining the optimum size
of RVE, in order to predict a more realistic behavior of the
Considering the brain structure, in some regions of the brain, a visco-hyperelastic model is implemented for ECM and
brain, neurons are distributed in ECM irregularly; hence, it neurons. Assuming an isotropic nature (Wang and Wineman
is expected that these regions behave in an isotropic way. In 1972) for neuron and ECM, strain energy We would be a func-
order to lower the computational costs, initially neurons and tion of invariants of left or right Cauchy–Green deformation
ECM are individually assumed as elastic and isotropic materi- tensor (Holzapfel 2000).
als. Utilizing the elastic model, the optimum size of RVE is ( )
extracted; then, its isotropy is studied. For this purpose, it is We = We I1 , I2 , I3 (2)

13
 M. Kazempour et al.

where Ii’s are invariants of either the left or right Green stress, as mentioned earlier, two relaxation time constants
deformation tensor (B or C). Hence, are assumed to consider the fast and slow loading velocities.

(3) ( ( )) ( t−𝜏 )
𝐁 = J −2∕3 𝐁, 𝐂 = J −2∕3 𝐂 t 𝜕Wv1
∫0
𝜕 −1∕2 −
𝛔v1 (t) = 2I3 𝐁 e 𝜃1 d𝜏
( ) ( ) 𝜕𝜏 𝜕𝐁
We = Weiso I 1 , I 2 + Wevol I3 (4) t ( (
𝜕Wv2
)) ( t−𝜏 ) (9)

∫0
v2 𝜕 −1∕2 −
𝛔 (t) = 2I3 𝐁 e 𝜃2 d𝜏
𝜕𝜏 𝜕𝐁
where B̄  , C̄  , J  , Weiso and Wevol are deviatoric parts of left and
right Cauchy–Green deformation tensor, the root of I3 , devi-
atoric and volumetric parts of the strain energy, respectively. where 𝜃i is relaxation time constant. Moreover, the strain
In order to predict brain response in different conditions, energy function related to the volumetric part of the energy
several material constitutive models have been presented is expressed with a well-known relation as (Horgan and
including elastic, hyperelastic, viscoelastic and visco-hyper- Murphy 2009; Whitford et al. 2018; Zhang et al. 2014).
elastic models (Miller 1999; Wang and Wineman 1972;
Kyriacou et al. 2002; Couper and Albermani 2008). Since 1
Wevol = K(J − 1)2
the responses vary in different physical conditions such as 2
1
variation of the strain rate, temperature, strain levels and Wvvol = K � (J − 1)2 (10)
1 2
other parameters, a visco-hyperelastic constitutive model is 1 ��
developed to consider the time-dependent behavior of the Wvvol = K (J − 1)2
2 2
brain tissue. Moreover, the stress is divided into two time-
dependent and elastic parts (Holzapfel 2000).
where K  , K ′ and K ′′ are the volumetric modulus of elastic
𝛔 = 𝛔e + 𝛔v (t) (5) and viscous parts which are considered to be equal in this
study. Finally, this model is numerically implemented as a
𝛔v (t) = 𝛔v1 (t) + 𝛔v2 (t) (6)
VUMAT subroutine for studying the brain tissue.
where 𝛔e and 𝛔vi are quasi-static and viscous stress, respec-
tively. In addition, the viscous stress is further divided into
two parts including deviatoric stress and volumetric stress. 2.3 Finite element simulation
Therefore, the strain energy for time-dependent part contains
a deviatoric and a volumetric energy. In order to generate the geometry of the micromechanical
model (RVE) in ABAQUS software, a PYTHON code is
( ) ( ) developed. For this end, the code reads reference neuron
Wv = Wviso I 1 , I 2 + Wvvol I3 (7)
coordinates, rotation and translation matrixes from a MAT-
Based on Bergström and Boyce (1998) studies, quasi- LAB script output. This code creates the neurons and ECM
static stress equations could be used for deviatoric part of as solid parts and assigns material properties to the RVE.
the viscous response. Additionally, it can identify contacts between the neurons
( ) and ECM and assign perfect bounding between them. More-
Weiso = ce I1 − 3 over, C3D4 elements are employed to geometrical discretiza-
( )
Wviso = cv1 I1 − 3 tion of both the neurons and ECM.
(8)
1
iso
( ) The loading method presented by Shaoning (2014) is
Wv = cv2 I1 − 3 used to apply load on the RVE. Symmetry boundary condi-
2

tion is employed in tension. In this way, while one of the

As for the strain energy function, Neo–Hookean strain
energy model (Eq. 8) is employed to include deviatoric parts
of both the elastic and viscous stresses. It is noteworthy that
Neo–Hookean coefficients have different values for each of
the elastic and viscous parts. Also, for the viscous part of the
faces of the RVE is subjected to uniaxial displacement, the
opposite side is fixed in the loading direction. Two other per-
pendicular faces from four remained faces are fixed in their
normal direction. The boundary conditions are demonstrated
in Table 1 using notations given in Fig. 2. Subsequently,

Table 1  Zero boundary U1 = 0 U2 = 0 U3 = 0 UR1 = 0 UR2 = 0 UR3 = 0

conditions in tension (see
Fig. 2) Tension test Back Up Right Up Back Back
Right Right Up
Shear test Back Back Back Back Back Back

13
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…
Fig. 2  Notation of the faces for identifying load cases
each of RVEs is subjected to five quasi-static loads in two
different strain rates. In order to implement the shear test,
shear displacement load is applied to one of the faces, while
the opposite face is fixed in a fully constrained boundary
condition.
Homogenized variables at macroscopic scale are deter-
mined using mean volume of variables at the microscopic
scale in the RVE. The mean macroscopic stress 𝛔 and strain
𝛆 could be calculated from Eq. (11).
�𝛺m � ∫ m
⟨𝛔⟩ =
1
𝛔 d𝛺
� �𝛺
m
(11)
�𝛺m � ∫ m
1 the computational cost significantly (Bergström and Boyce
⟨𝛆⟩ = 𝛆 d𝛺
� � 𝛺m
where 𝛺m , 𝛔m and 𝛆m are the total volume of RVE, local
microscopic stress and strain, respectively.
Considering the geometrical complexity in the RVE,
increasing neuron volume fraction increases the computa-
tional cost. On the other hand, an increase in NVF highly
reduces the distance between neurons, which could result
in different problems including meshing and assigning ele-
ments to the RVE. Therefore, a self-consistent procedure
is necessary to be employed to resolve these problems.
According to the self-consistent method, the heterogene-
ous composite material could be simulated as a homogene-
ous material, having average mechanical constants. In this
procedure, homogenized coefficients of the heterogeneous
composite will be assigned to the ECM, while the main rea-
son for the raise in NVF would be the redistribution of neu-
rons (see Rémond et al. 2016; Mura 2013; Torquato 2013
for more details). In order to implement self-consistent
approximation, particle swarm optimization (PSO) function
(Dréo et al. 2006; Poli 2007) is implemented in MATLAB
programming. Homogenized normal and shear stresses ver-
sus time curves of the heterogeneous composite are con-
sidered as input data. Utilizing the optimization function,
a homogeneous matrix is fitted on these curves and a new
RVE is generated with a higher volume fraction. Hence,
MATLAB script runs a PYTHON code, PYTHON pro-
gram initiates ABAQUS software analysis, and VUMAT
subroutine is also employed by ABAQUS software as the
constitutive model. Finally, PYTHON program extracts the
stress–time curves when the analysis is completed. MAT-
LAB program, then, compares the homogeneous matrix
curves with heterogeneous composite ones and keeps the
loop running until the area between these curves reaches a
desirable tolerance. Considering an isotropic state for the
used RVE, the consistency between these two curves would
be enough for computing the mechanical properties. Fig-
ure 3 demonstrates the procedure of the analysis in brief.
3 Results and discussion
3.1 Employing elastic model to investigate isotropy,
RVE size and irregular distribution
It is known that RVE size is one of the important parameters
in predicting effective properties of heterogeneous materials.
Moreover, mechanical properties must be independent of
the RVE size. Considering boundary effects on the results,
the calculated outcomes from a small size of RVE could not
be valid. Larger sizes of RVE, on the other hand, increase
1998). Therefore, five RVEs in different sizes (300, 400,
700, 900, 1100 µm) are created with the same NVF (0.3%).
Material properties of the components of RVEs are sum-
marized in Table 2. RVEs are subsequently analyzed by
the elastic model, and an optimum size for these RVEs is
found considering their elastic modulus. In Fig. 4, the elas-
tic modulus of different RVE sizes is plotted for the same
volume fraction.
Figure 4 depicts that RVE size with dimensions of 900 µm
would be a suitable choice, since no significant variation in
the results is observed for the larger sizes. Considering an
elastic model for RVEs, it is obvious that the results must be
independent of the RVE size. Therefore, the computational
cost is optimum and the results are valid when RVE dimen-
sions are 900 µm.
Since random distribution of the neurons is assumed for
the creation of RVEs, a thorough investigation of RVEs’
behavior seems essential. Therefore, the elastic orthotropic
behavior is initially assumed for the RVEs with dimensions
of 900 µm. The results are summarized in Table 3.
13
 M. Kazempour et al.

Start

Assigning RVE length, reference Guess coefficients of ECM using

Assigning mechanical properties
neuron coordinates, minimum PSO optimization function to fit
distance between neurons and ECM parameters on homogenization
number of neurons data
VUMAT
subroutine VUMAT
subroutine
Generating randomly neurons in the
1 Determine interaction between
RVE Submit ABAQUS
neurons and ECM

Remove and change Python code

the neurons
Read stress, strain and element
Apply boundary conditions, volume
Yes Check overlap or
intersection between neurons displacement and generate mesh
occurs?
Is stresses
and strains curves fit on No
No homogenization
Submit ABAQUS curves?
Save reference neuron coordinates, Yes
rotation and transfer matrixes
Use neuron coefficients and new
Python code coefficients of ECM increase VF of
Read reference neuron coordinates, neurons
rotation and transfer matrixes

Read stress, strain and element No

1 Is VF of neurons enough?
volume during loading
Python code
Yes

Create reference neuron part by

reference neuron coordinates and Use homogenization equations and
Drawing curves
create ECM considering RVE size save dataa

End

Fig. 3  Flowchart of modeling and analysis of brain tissue

Table 2  Material properties of ECM and neurons Table 3  Material coefficients of orthotropic assumption for the RVE
Elastic modulus (Pa) Poisson’s ratio Direction Elastic modulus Shear modulus Poisson’s ratio
of proper- (Pa) (Pa)
ECM 2040 0.5 ties
Neuron 6160 0.5
X 2050.084 – –
Y 2050.079 – –
Z 2050.117 – –
2.0500 VF=0.3% XY – 687.956 0.489982
XZ – 687.960 0.489972
YZ – 687.959 0.489973
Ex (kPa)

2.0495

2.0490
400 500 600 700
RVE size (µm)
Fig. 4  Elastic modulus variation with RVE size for VF = 0.3%
Regarding the observed similar behavior of the RVE in
different perpendicular directions, it is concluded that the
RVE behaves isotropically. The same result has been dem-
onstrated by Koser et al. (2015) for the white matter under
infinitesimal tension. Due to the isotropic behavior of the
RVE, the characterization and mechanical properties could
800 900 1000 1100 be easily predicted by performing only two sets of tests. This
significantly lowers the computational costs.
However, there are still two concerns remaining to con-
firm isotropy, the effect of various irregular distributions
and different NVFs. As for the effects of various irregular
distributions of neurons on the mechanical properties, three
different RVEs are created with NVF of 0.3% exhibiting
a random distribution. The results show that different dis-
tributions yield approximately equal mechanical properties
(see Table 4). Hence, it is deduced that the effective general

13
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…

Table 4  Elastic modulus for different distributions of neurons in some regions of the brain, increasing the axon volume
Elastic modu- Distribution 1 Distribution 2 Distribution 3 fraction results in the enhancement of the elastic modulus.
lus (Pa) In addition, these regions show an anisotropic behavior.
However, elastic modulus of different volume fractions is
2050.084 2050.084 2050.080
Ex
approximately equal in perpendicular directions, once more
2050.079 2050.092 2050.087
Ey
revealing an isotropic behavior in the RVEs.
2050.117 2050.111 2050.100
Ez
In Fig. 5, 1% strain load as a tension static load is applied
on the RVE. It is found that the stresses in neurons are
higher than those in the matrix. Moreover, Fig. 5 shows that
Table 5  Homogenized elastic modulus in different VFs for ortho- in neurons, the stress experience in dendrites and axon is
tropic model higher than soma. Therefore, the neuron plays a key role
Elastic 0.3 0.6 0.9 1.2 1.5 in investigating diffuse axonal injury (DAI) compared to
modulus ECM. Distribution of the stress is demonstrated in Fig. 5.
(Pa) As one may observe, the micromechanical investigation on
Ex 2050.084 2060.080 2070.075 2080.071 2090.066 the stress distribution is one of the most important matters in
Ey 2050.079 2060.075 2070.071 2080.066 2090.062 TBI study. Under elastic assumption in this study, the impor-
Ez 2050.117 2060.113 2070.108 2080.104 2090.099 tance of micromechanical analysis is shown. Therefore, to
arrive at more realistic results, a visco-hyperelastic model
is developed in this study.
mechanical properties are independent from distribution of
the neurons in the regions of the brain where distribution of 3.2 Developing visco‑hyperelastic model to have
the neurons are irregularly. a more realistic study on brain tissue
In order to investigate the impact of NVF variation on the
mechanical properties, five RVEs are generated with 0.3, Javid et al. (2014) conducted an experiment on the brain
0.6, 0.9, 1.2 and 1.5% neurons and 1% tension strain load is tissue and presented the mechanical properties of axon and
applied on them. They are analyzed under static loads, while ECM separately. In this study, the outcomes of that experi-
the self-consistent approximation is employed to increase the ment are employed to find the necessary visco-hyperelastic
volume fraction. It is found that the elastic modulus rises at coefficients. To evaluate the employed constitutive model,
higher values of NVF. However, due to the low increase in a RVE including 52.7% axons was created. The RVE was
neuron volume fraction, the observed variation of the elastic studied under 3% strain, and the obtained results were
modulus is not significant. In Table 5, elastic modulus of compared with experimental tests reported by Javid et al.
different NVFs is presented. Abolfathi et al. (2009) simu- (2014). The results reveal a good agreement. Table 6 indi-
lated only axons of the brain stem. The study showed that cates these coefficients for the RVE components, and Fig. 6

Fig. 5  Tension stress distribution in neurons for elastic modeling (the results should be multiplied by 1e12 Pa)

13
 M. Kazempour et al.

Table 6  Visco-hyperelastic Ce (Pa) Cv1 (Pa) Cv2 (Pa) K(kPa) 𝜃1 (s) 𝜃2 (s)

coefficients of the ECM and
axon ECM 170.2 110.8 167.7 199.9 0.20 0.93
Axon 615.5 72.6 1543.4 70.91 1e−7 0.57

250 (a)
3
200 1% strain at 0.03s
Employed Model 1% strain at 0.2s

Strain (%)
Stress (Pa)

150 Javid study 2 2% strain at 0.03s

2% strain at 0.2s
100 3% strain at 0.03s
1 3% strain at 0.2s
50

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 2 4
Time (s) Time (s)

(b) 15
Fig. 6  The mean volumetric stress of the RVE in axons direction
with 52.7% axons volume fraction under 3% strain and experiments
reported by Javid et al. (2014) Strain (%)
10
10% strain at 0.03s

10% strain at 0.2s

demonstrates outcomes of study by Javid et al. (2014) and 5 15% strain at 0.03s
the mean volumetric stress of the RVE for the present work. 15% strain at 0.2s
The RVE size obtained in the elastic region (900 µm) is 0
used for the rest of this analysis. Various loads in differ- 0 1 2 3 4
ent rates are applied, and the tissue behavior is evaluated Time (s)
with respect to time. RVEs are subjected to various loads
including 1, 2, 3, 10 and 15% tension and shear strains in Fig. 7  Different strain sizes in various rates: a 1,2 and 3% logarithmic
a quasi-static manner; each of them are applied at 0.2 and strain and b 10 and 15% logarithmic strain
0.03 s. In Fig. 7, the strain variations versus time are plotted
for different strains in various rates. 450
It should be considered that micromechanical analysis 400
is time-consuming, especially, in quasi-static and dynamic 350 15% strain at 0.2s
loadings in which the employed constitutive model has a 300 15% strain at 0.03s
time-dependent behavior. It is noteworthy that in this study,
Stress (Pa)

250
since the stress variations are not sensitive after 4 s, in order
200
to avoid unnecessary computational costs, the analysis is ter-
150
minated at 4 s. In Fig. 8, variations of the mean volumetric
100
stress in RVE in loading direction are demonstrated under
15% tension loading for 3.9% NVF. 50

In Fig. 9a, b, variations of mean volumetric stress ver- 0

0 0.5 1 1.5 2 2.5 3 3.5 4
sus time are plotted with an NVF of 1.2%. As shown, Time (s)
the stress rises until 0.2 and 0.03  s for different strain
rates. Considering viscous behavior of RVE, when the Fig. 8  Stress variation in 3.9% VF under 15% strain
strain becomes constant, a reduction in the stress could
be observed. Due to rate-dependent behavior of the brain
tissue (Pervin and Chen 2009), increasing the strain rate more vulnerable. These figures show that the developed
would lead to a stiffer response in the tissue and outcomes model is suitable to study the rate dependency of the brain
of Rashid et al. (2014) confirm the stiffer behavior of the tissue. Additionally, Fig. 10 demonstrates that enlarging
tissue at higher rates. Thus, considering the fact that in NVF would lead to an increase in the stress throughout
accidents strain rate is usually high, the brain tissue is the analysis.

13
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…

(a) 30 90
80
25 1% strain at 0.03s 70
3%strain at 0.03s
1% strain at 0.2s 60
20 3% strain at 0.2s

Stress (Pa)
Stress(Pa)

50
15
40

10 30
20
5
10
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4
Time(s) Time (s)
(b) 60
Fig. 11  Mean volumetric stress for 2.1% VF of neurons under 3%
50
2% strain at 0.2s
strain load in various rates

40 2% strain at 0.03s
Stress (Pa)

30 30

20 1% strain at 0.2s for 3% VF

25
1% strain at 0.2s for 3.9% VF
1% straon at 0.03s for 3% VF

Stress (Pa)
10 20
1% strain at 0.03s for 3.9% VF

0 15
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Time (s) 10

5
Fig. 9  The mean volumetric stress of the RVE with 1.2% VF of neu-
rons under different quasi-static loads in various rates: a 1% strain 0
0 0.5 1 1.5 2 2.5 3 3.5 4
and b 2% strain
Time (s)

250 Fig. 12  Mean volumetric stresses for different volume fractions under

VF=3.9% equal load
200 VF=3%
Stress (Pa)

150
100
50
0
0 0.5 1 1.5 2 2.5
Time (s)
Fig. 10  The mean volumetric stress in 10% strain applied 0.2 s in 3%
VF and 3.9% VF of the RVE
One of the important points in time-dependent materials
is the loading rate; in Fig. 11, variations of the mean volu-
metric stress for 2.1% NVF are plotted in two strain rates
under 3% tension. Although the same values are expected at
0.2 s, the presence of load history results in a non-similarity
in the extracted values for different strain rates. The differ-
ence in the peaks of the two curves results from the strain
rate dependency of RVE. Since NVFs and loads magnitude
are equal, the mean volumetric stresses reach an equal value
in both strain rates.
In Fig. 12, variations of mean volumetric stress are shown
in two strain rates and various NVFs. This figure demon-
strates that the stress grows with the rise of NVF; confirm-
ing that although the increase is not significant, the tissue
3 3.5 4 has become stiffer (Karami et al. 2009; Karami et al. 2010).
It is found that in higher rates, NVF has a lower impact on
the maximum stress, while it has a more severe effect in
lower strain rates. Comparing the results obtained for the
NVF, load and strain rates effects on the brain tissue, Fig. 12
show that the brain tissue is more sensitive to the strain rate
regarding NVF effects, and according to Figs. 10, 12 and 13,
it is also more sensitive to the magnitude of the load with
respect to the other two parameters. As shown, enhancing
the strain rate up to about seven times of its initial value,
the maximum stress becomes about 1.18 times for both the
strains. Meanwhile, tripling the strain leads to a rise of stress
by 2.87 times for two different rates. As shown in Fig. 14,
the enhancement of mean volumetric shear stress could be

13
 M. Kazempour et al.

90 stress in a neuron is experienced by the dendrites and axon.

80 1% strain at 0.2s with 3.9% VF
Therefore, micromechanical analysis shows dendrites and
1% strain at 0.03s with 3.9% VF
70
3% strain at 0.03s with 3.9% VF axon having a more sensitive role in brain injury.
60
Figures 16 and 17 depict the distribution of von Mises
Stress (pa)

3% strain at 0.2s with 3.9% VF

50
stress in the neurons of RVE at 0.2 s and homogenized stress
40
of the RVE having a NVF of 3.9%, respectively. RVE is
30
subjected to 3% strain, which is applied at 0.2 s. The maxi-
20
10
mum mean volume of the stress in Fig. 17 is about 71.89 Pa;
0
meanwhile according to Fig. 16, the experienced stress by
0 0.5 1 1.5 2 2.5 3 3.5 4 neurons is between 126.2 and 443.6 Pa at the same time. It is
Time (s) obvious that the stresses in ECM are smaller, confirming the
previous outcome that in the case of loading on the brain tis-
Fig. 13  RVE with equal VF under different loads in various rates sue, the neurons experience higher stresses compared to the
ECM and are more vulnerable in brain injuries. Javid et al.
(2014) and Karami et al. (2009) also draw the same conclu-
sion. Moreover in all experiments, the experienced stress in
20 neurons is higher than that in ECM (Abolfathi et al. 2009),
1%strain at 0.2s
1% strain at 0.03s
while the experienced strain by neurons is lower than that in
Shear Stress (Pa)

15 3% strain at 0.2s
3% strain at 0.03s ECM (Abolfathi et al. 2009). Figure 18 shows the variation
10 of the ECM and neurons’ strain, demonstrating that the mean
volume strain of the ECM is higher than that of the neurons.
5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4 Summaries and conclusion
Time (s)

In the present contribution, the behavior of the brain tissue

Fig. 14  Variations of mean volumetric shear stress during the analy-
sis having a 3% NVF under two different strains in two various rates
reported by almost 1.2 and 2.9 times for increasing strain
rates and the strain, respectively.
Considering a NVF of 3.9% for the RVE, Fig. 15 illus-
trates the stress contour in loading direction while 3% strain
at 0.2 s is applied. According to the contour, the maximum
was investigated under different loads with various rates and
volume fraction of neurons. To this aim, RVEs with neu-
rons were generated using MATLAB and PYTHON scripts
in ABAQUS. Neurons distributed randomly in the RVEs
and the interaction between neurons and ECM were con-
sidered as perfect bonding. The orthotropic elastic model
was employed to study the isotropic behavior and neurons
distribution effect on the mechanical properties. The results

Fig. 15  The stress distribu-

tion in loading direction of a
neuron for 3% strain applied on
the RVE (the results should be
multiplied by 1e12 Pa)

13
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…

Fig. 16  Distribution of the stress in neurons of a RVE that 3% strain is applied at 0.2 s on the RVE having 3.9% volume fraction of neuron (the
results should be multiplied by 1e12 Pa)

70
showed that in different distributions, RVEs have the same
response; also RVEs showed isotropic behavior in all volume
60
fractions. Additionally, the optimum RVE size was evaluated
50
using the elastic constitutive model to eliminate the bound-
Stress (Pa)

40 ary conditions effects.

30 Furthermore, the visco-hyperelastic constitutive model
20 was used to study loads, load rates and NVFs impacts on the
10 mechanical response of RVEs. It was found that the maxi-
0
mum stress was raised with increasing the load rate that
0 0.5 1 1.5 2 2.5 3 3.5 4 shows the tissue was more vulnerable in high load rates.
Time (s) Moreover, the load rate had lower impact than load itself on
the maximum stress. According to the proposed result, in
Fig. 17  The mean volumetric stress of a RVE with 3.9% volume frac- the similar loading conditions, axons were more vulnerable
tion of neuron and 3% strain is applied at 0.2 s than soma that is a reason for DAI. These results could be
beneficial for production of helmet, surgical robots and all
tools that are employed to prevent or treat the brain injuries.

10

8 References
Strain %

6 ECM Strain Abdel Rahman R et al (2012) An asymptotic method for the predic-
tion of the anisotropic effective elastic properties of the cortical
4 vein: superior sagittal sinus junction embedded within a homog-
enized cell element. Journal of Mechanics of Materials and
2
Structures 7(6):593–611
0 Abolfathi N et al (2009) A micromechanical procedure for model-
0 0.5 1 1.5 2 2.5 3 3.5 4 ling the anisotropic mechanical properties of brain white matter.
Time (s) Computer Methods in Biomechanics and Biomedical Engineering
12(3):249–262
Bergström J, Boyce M (1998) Constitutive modeling of the large
Fig. 18  Variations of computed strain for neurons and ECM sepa- strain time-dependent behavior of elastomers. J Mech Phys Sol-
rately ids 46(5):931–954

13

Bernick KB et al (2011) Biomechanics of single cortical neurons. Acta
Biomater 7(3):1210–1219
Budday S et al (2015) Mechanical properties of gray and white mat-
ter brain tissue by indentation. J Mech Behav Biomed Mater
46:318–330
Christ AF et al (2010) Mechanical difference between white and gray
matter in the rat cerebellum measured by scanning force micros-
copy. J Biomech 43(15):2986–2992
Cloots RJ et al (2013) Multi-scale mechanics of traumatic brain injury:
predicting axonal strains from head loads. Biomech Model Mech-
anobiol 12(1):137–150
Coudrillier B et al (2013) Scleral anisotropy and its effects on the
mechanical response of the optic nerve head. Biomech Model
Mechanobiol 12(5):941–963
Couper Z, Albermani F (2008) Infant brain subjected to oscillatory
loading: material differentiation, properties, and interface condi-
tions. Biomech Model Mechanobiol 7(2):105
Dréo J et al (2006) Metaheuristics for hard optimization: methods and
case studies. Springer, Berlin
Faul M et  al (2010) Traumatic brain injury in the United States:
national estimates of prevalence and incidence, 2002–2006. Injury
Prevention 16(Suppl 1):A268
Feng Y et al (2013) Measurements of mechanical anisotropy in brain
tissue and implications for transversely isotropic material models
of white matter. J Mech Behav Biomed Mater 23:117–132
Forte AE, Galvan S, Dini D (2018) Models and tissue mimics for brain
shift simulations. Biomech Model Mechanobiol 17(1):249–261
Ganpule S et al (2013) Mechanics of blast loading on the head mod-
els in the study of traumatic brain injury using experimental
and computational approaches. Biomech Model Mechanobiol
12(3):511–531
Hashin Z (1983) Analysis of composite materials—a survey. J Appl
Mech 50(3):481–505
Hiscox LV et al (2016) Magnetic resonance elastography (MRE) of the
human brain: technique, findings and clinical applications. Phys
Med Biol 61(24):R401
Holzapfel GA (2000) Nonlinear solid mechanics: a continuum
approach for engineering, 1st edn. Wiley, New York
Horgan CO, Murphy JG (2009) On the volumetric part of strain-energy
functions used in the constitutive modeling of slightly compress-
ible solid rubbers. Int J Solids Struct 46(16):3078–3085
Javid S, Rezaei A, Karami G (2014) A micromechanical procedure for
viscoelastic characterization of the axons and ECM of the brain-
stem. J Mech Behav Biomed Mater 30:290–299
Ji S et al (2014) Head impact accelerations for brain strain-related
responses in contact sports: a model-based investigation. Biomech
Model Mechanobiol 13(5):1121–1136
Kanit T et al (2003) Determination of the size of the representative
volume element for random composites: statistical and numerical
approach. Int J Solids Struct 40(13):3647–3679
Karami G et  al (2009) A micromechanical hyperelastic modeling
of brain white matter under large deformation. J Mech Behav
Biomed Mater 2(3):243–254
Karami G, Shankar S, Ziejewski M, Azarmi F (2010) Micromechanical
hyperelastic modeling of bi-directional oriented axons in brain
white matter. In: Proc. ASME. biomedical and biotechnology
engineering, pp 619–625
Kleiven S, von Holst H (2002) Consequences of head size following
trauma to the human head. J Biomech 35(2):153–160
Koser DE et al (2015) CNS cell distribution and axon orientation
determine local spinal cord mechanical properties. Biophys J
108(9):2137–2147
Kyriacou SK et al (2002) Brain mechanics for neurosurgery: modeling
issues. Biomech Model Mechanobiol 1(2):151–164
Labus KM, Puttlitz CM (2016) An anisotropic hyperelastic con-
stitutive model of brain white matter in biaxial tension and
13
M. Kazempour et al.
structural–mechanical relationships. J Mech Behav Biomed
Mater 62:195–208
Laksari K, Shafieian M, Darvish K (2012) Constitutive model for
brain tissue under finite compression. J Biomech 45(4):642–646
Laksari K et al (2015) Computational simulation of the mechanical
response of brain tissue under blast loading. Biomech Model
Mechanobiol 14(3):459–472
Lee S et al (2014) Measurement of viscoelastic properties in multiple
anatomical regions of acute rat brain tissue slices. J Mech Behav
Biomed Mater 29:213–224
Libertiaux V, Pascon F, Cescotto S (2011) Experimental verification
of brain tissue incompressibility using digital image correlation.
J Mech Behav Biomed Mater 4(7):1177–1185
Maltese MR, Margulies SS (2016) Biofidelic white matter heteroge-
neity decreases computational model predictions of white mat-
ter strains during rapid head rotations. Computer methods in
biomechanics and biomedical engineering 19:1–12
Menon DK et al (2010) Position statement: definition of traumatic
brain injury. Arch Phys Med Rehabil 91(11):1637–1640
Miller K (1999) Constitutive model of brain tissue suitable for
finite element analysis of surgical procedures. J Biomech
32(5):531–537
Mura T (2013) Micromechanics of defects in solids. Springer, Berlin
Pan Y et al (2013) Finite element modeling of CNS white matter kin-
ematics: use of a 3D RVE to determine material properties. Fron-
tiers in bioengineering and biotechnology 1:19
Pervin F, Chen WW (2009) Dynamic mechanical response of bovine
gray matter and white matter brain tissues under compression. J
Biomech 42(6):731–735
Peter SJ, Mofrad MR (2012) Computational modeling of axonal micro-
tubule bundles under tension. Biophys J 102(4):749–757
Poli R (2007) An analysis of publications on particle swarm optimiza-
tion applications. Department of Computer Science, University
of Essex, Essex
Rashid B, Destrade M, Gilchrist MD (2014) Mechanical characteri-
zation of brain tissue in tension at dynamic strain rates. J Mech
Behav Biomed Mater 33:43–54
Rémond Y et al (2016) Homogenization of reconstructed RVE. In:
Rémond Y, Ahzi S, Baniassadi M, Garmestani H (eds) Applied
RVE reconstruction and homogenization of heterogeneous materi-
als. Wiley, Hoboken
Schmid-Schönbein G et al (1981) Passive mechanical properties of
human leukocytes. Biophys J 36(1):243–256
Shaoning S (2014) Mechanical characterization and modeling of poly-
mer/clay nanocomposites
Sheidaei A et al (2013) 3-D microstructure reconstruction of polymer
nano-composite using FIB–SEM and statistical correlation func-
tion. Composites Science and Technology 80:47–54
Shulyakov AV et al (2009) Simultaneous determination of mechanical
properties and physiologic parameters in living rat brain. Biomech
Model Mechanobiol 8(5):415–425
Sotudeh-Chafi M et al (2008) A multi-scale finite element model for
shock wave-induced axonal brain injury. In: ASME 2008 sum-
mer bioengineering conference. American Society of Mechanical
Engineers
Tanielian T et al (2008) Invisible wounds of war. Summary and recom-
mendations for addressing psychological and cognitive injuries.
RAND Corp, Santa Monica
Torquato S (2013) Random heterogeneous materials: microstructure
and macroscopic properties, vol 16. Springer, Berlin
Vappou J et al (2007) Magnetic resonance elastography compared with
rotational rheometry for in vitro brain tissue viscoelasticity meas-
urement. Magn Reson Mater Phys, Biol Med 20(5–6):273
Wang HC, Wineman AS (1972) A mathematical model for the deter-
mination of viscoelastic behavior of brain in vivo—I Oscillatory
response. J Biomech 5(5):431–446
Homogenization of heterogeneous brain tissue under quasi-static loading: a visco-hyperelastic…

Whitford C et al (2018) A viscoelastic anisotropic hyperelastic consti-
tutive model of the human cornea. Biomech Model Mechanobiol
17(1):19–29
Yousefi E et al (2017) Effect of nanofiller geometry on the energy
absorption capability of coiled carbon nanotube composite mate-
rial. Composites Science and Technology 153:222–231
Zhang M-G et al (2014) Spherical indentation method for determining
the constitutive parameters of hyperelastic soft materials. Bio-
mech Model Mechanobiol 13(1):1–11
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

13