Journal of Biomechanics 45 (2012) 2241–2247

Contents lists available at SciVerse ScienceDirect

Journal of Biomechanics
journal homepage: www.elsevier.com/locate/jbiomech
www.JBiomech.com

A paradigm for the development and evaluation of novel implant topologies

for bone fixation: Implant design and fabrication
Heesuk Kang a, Jason P. Long b, Gary D. Urbiel Goldner a,b, Steven A. Goldstein a,b,c, Scott J. Hollister a,c,d,n
a
Department of Mechanical Engineering, the University of Michigan, USA
b
Department of Orthopaedic Surgery, the University of Michigan, USA
c
Department of Biomedical Engineering, the University of Michigan, USA
d
Department of Surgery, the University of Michigan, USA

a r t i c l e i n f o abstract

Article history: The future development of bio-integrated devices will improve the functionality of robotic prosthetic
Accepted 9 June 2012 limbs. A critical step in the advancement of bio-integrated prostheses will be establishing long-term,
secure fixation to the remnant bone. To overcome limitations associated with contemporary bone-
Keywords: anchored prosthetic limbs, we established a paradigm for developing and fabricating novel orthopedic
Orthopedic implant design implants undergoing specified loading. A topology optimization scheme was utilized to generate
Topology optimization optimal implant macrostructures that minimize deformations near the bone-implant interface.
Solid freeform fabrication Variations in implant characteristics and interfacial connectivity were investigated to examine how
Microstructure these variables influence the layout of the optimized implant. For enhanced tissue integration, the
optimally designed macroscopic geometry of a titanium (Ti)-alloy implant was further modified by
introducing optimized microstructures. The complex geometries of selected implants were successfully
fabricated using selective laser sintering (SLS) technology. Fabrication accuracy was assessed by
comparing volumes and cross-sectional areas of fabricated implants to CAD data. The error of fabricated
volume to CAD design volume was less than 8% and differences in cross sectional areas between SEM
images of fabricated implants and corresponding cross sections from CAD design were on average less
than 9%. We have demonstrated that this computational design method, combined with solid freeform
fabrication techniques, provides a versatile way to develop novel orthopedic implants.
& 2012 Published by Elsevier Ltd.

1. Introduction The concept of bone-anchored prosthetic limbs is not novel.

Contemporary robotic prostheses are capable of high-levels of
articulation and complex movements, but the usability of these
devices are limited by inadequate exchange of control and
sensory data between the user and prosthetic limb (Loeb, 2009).
In the future, major advancements in functionality will likely
require permanent integration with the user’s body. A critical step
in the development of a bio-integrated prosthesis will be estab-
lishing technologies that enable long-term, secure fixation of the
devices to the bone in the limb remnant. This would allow for the
transfer of multi-axial and multi-directional loads generated
during normal daily activity, and establish a structurally-secure
interface and anchor for the acquisition and transmission of
neural or muscular data.
n
Corresponding author at: Department of Biomedical Engineering, University of
Michigan, Lurie Biomedical Engineering Bldg. Rm 2208, 1101 Beal Ave. Ann Arbor,
MI 48109-2110, USA. Tel.: þ 1 734 647 9962; fax: þ1 734 647 0003.
E-mail address: scottho@umich.edu (S.J. Hollister).
Currently, these devices are used by a limited number of patients
with most having transfemoral amputations (Branemark et al.,
2001; Hagberg and Branemark, 2009). Complications with these
implant systems, including bone resorption around the anchor
due to infection or due to stress-shielding of the surrounding
bone (Tillander et al., 2010; Tomaszewski et al., 2010; Xu and
Robinson, 2008), have yet to be adequately addressed. Addition-
ally, current bone-anchored devices generally require a substan-
tial amount of remnant bone to structurally secure the prosthesis
(Branemark et al., 2001; Hagberg and Branemark, 2009), which
limits these systems to a narrow range of anatomic sites and
clinical circumstances. These concerns highlight the need for
further development of bone-secured prosthetic limbs and also
offer insight into the complex design requirements for an anchor
system that provides both robust mechanical function and
permits long-term preservation of the surrounding, structurally-
critical bone.
One approach to address these design complexities utilizes
optimization strategies. For example, topology optimization in
conjunction with the finite element (FE) method could be used to

0021-9290/$ - see front matter & 2012 Published by Elsevier Ltd.

http://dx.doi.org/10.1016/j.jbiomech.2012.06.011
2242 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247
optimize the layout of limited implant material in order to meet
specific performance criteria (Lin et al., 2004, 2007). This
approach could lead to the development of novel implant geo-
metries and topologies predicted to have robust mechanical
fixation under pre-determined loading conditions while limiting
the amount of implant material. As part of this design strategy, it
is also critical that the implants can be feasibly manufactured and
fabricated from material with high mechanical strength and
toughness to withstand repetitive loading over decades of use.
To address these issues, the aims of this study were: (1) to use
topology optimization in conjunction with a FE model to develop
implant designs for a favorable bone fixation response and (2) to
fabricate these optimized designs using available technologies.
2. Materials and methods
Topology optimization is an approach to optimize the material layout within a
specified design domain, such that the resulting structure meets specific perfor-
mance criteria (Bendsøe and Kikuchi, 1988). In general, this optimization method
is executed in conjunction with FE methods, on which a density value is assigned
to each element that represents the structural topology. The density value ranges
from 0 to 1, with 0 indicating no material (void) and 1 indicating a solid.
Introduction of a material interpolation law such as the homogenization method
or solid isotropic material penalization (SIMP) relates the density values to the
element material properties (Sigmund, 1994). In this way, the intermediate values
can indicate partial material that requires post-processing to establish a clear
solid-void boundary. The optimization algorithm evolves the material density
distribution such that the objective of the optimization problem is minimized
under specified boundary conditions and volume constraints.
The objective function in this particular optimization scheme was to minimize
the compliance of the bone-implant system, which was achieved by altering the
distribution of implant material. In doing so, deformations near the bone-implant
interface were minimized and the stiffness at this interface region was maximized.
As an initial investigation towards minimizing the amount of implant material, we
constrained the volume of the optimized macrostructure to compose no more
than 50% of the design domain volume.
Our FE model had a cylindrical implant design domain surrounded by a bed of
homogeneous trabecular bone (Fig. 1), which represents a generic metaphyseal
region rather than a specific anatomic site. With this approach, we sought to
establish a design paradigm rather than create an implant for clinical application.
Nevertheless, for potential translation, our model could represent a single
structural projection that would work in concert with similar projections to meet
the mechanical requirements of an implant system (Goldstein et al., 1991;
Matthews and Goldstein, 1992). This concept would be particularly attractive in
cases where a patient has a limited length of remnant bone for attachment of a
prosthetic anchor.
Other aspects of our FE model included an interfacial zone between the
implant and bone with a loading post attached to the top of the implant. The
design domain was defined as a hollow cylinder to maintain implant material near
the loading post. To represent the entire cylindrical model, an FE mesh was
Fig. 1. The finite element model of the bone-implant system for topology
optimization was cylindrical and consisted of a hollow design domain with a
central loading post surrounded by trabecular bone. An interfacial tissue zone was
also included between design domain and surrounding trabecular bone.
Table 1
Linear elastic, isotropic material properties for finite element model.
Component Young’s modulus (MPa) Poisson’s ratio
Trabecular bone 500 0.4
Fibrous tissue 1 0.45
Titanium 110,000 0.33
75% Porous tantalum 3300 0.31
created for a 451 wedge and then duplicated around the central vertical axis. The
mesh was generated using HyperMesh (HyperWorks 8.0, Altair Engineering Inc.,
Troy, MI) and consisted of 8-noded hexahedral and 6-noded pentahedral ele-
ments. All elements were assigned linear elastic, isotropic material properties
(Table 1).
To develop a range of implant designs, we investigated variations in implant
characteristics and interfacial connectivity. Changes to implant characteristics
included different metals, Ti-alloy or 70% porous tantalum (Table 1), that are
commonly used in contemporary orthopedic implants, and variations in design
domain size (implant dimensions were constant, but the design domain made up
67%, 42% or 25% of the overall radial dimension). Interfacial connectivity was
varied to simulate potential differences in bone-implant integration that may
develop during the early post-operative period. Three interface variations were
examined, including fibrous tissue or trabecular bone material properties assigned
to the entire interface zone. For the third variation, we assigned fibrous tissue to
the circumferential implant interface zone and trabecular bone to the bottom
implant interface zone. For all models, implant-tissue surfaces were displacement
compatible. In total, we analyzed 18 different FE models.
The optimized implant structure will be highly dependent upon its loading
environment. A bone anchor for a bio-integrated prosthetic limb will undergo
complex loading during daily use, and the loading environment will depend upon
a number of factors, including anatomic site and patient activity level. As an initial
iteration for establishing our design paradigm, loading was simplified to uni-axial
forces (100 N tension/compression) applied to the top of the loading post. The
model was fully constrained by assigning a zero displacement boundary condition
to the outer surfaces (i.e. sides and base) of the bone.
The optimization problems were solved using OptiStruct (HyperWorks 8.0,
Altair Engineering Inc., Troy, MI), a commercial topology optimization solver.
Following optimization convergence, a threshold value was applied to the design
domain’s density distribution to define the structural boundary of each implant.
Regions with density values higher than the threshold were considered as solid;
regions below the threshold were considered void. Then, the material volume
fraction or porosity of the designed implant was evaluated to determine if it met
the volume fraction constraints of the optimization scheme.
Following the solution of the various optimization problems, implant designs
were selected for fabrication and future in vivo evaluation. For our initial selection,
we chose an optimized design made of Ti-alloy. As a subsequent design process,
this selected design was modified to have a porous architecture by replacing
material layout with appropriate microstructures to further enhance the integra-
tion of bone and implant. Specifically, the density distribution was segmented into
high and low density regions, which served as a map to place specific micro-
structures (Hollister and Lin, 2007; Lin et al., 2004). The low and high porosity
microstructures used for this mapping were designed to be optimally stiff and
permeable, as per isotropic cross-property bounds (Gibiansky and Torquato,
1996). To do this, a multi-objective topology optimization code was used, and
the microstructures were optimized based upon bulk modulus and diffusivity
properties (Kang et al., 2010). The low and high density microstructures had
porosities of 63% and 33%, respectively (Fig. 2). The overall porosity of the scaffold
was constrained to be 50%.
For fabrication, optimized implant dimensions were set to 6.4 mm diameter
by 6.4 mm height. The shape of the loading post was modified, such that a narrow
post diameter was maintained at the junction with the optimized structure but
flared away from the implant to a diameter of 6.4 mm. The additional post
material allowed for the future placement of a drilled and tapped hole for in vivo
evaluation. Overall implant height was set to 12.7 mm (including the loading
post). Implants were fabricated from extra low interstitial (ELI)-grade Ti–6Al–4V
alloy by Synergeering Group (Farmington Hills, MI) using a selective laser
sintering (SLS) system (EOSINT M 270, EOS of North America Inc., Novi, MI).
Implants were not annealed, which could refine structure and alter material
properties (Hollander et al., 2006).
The manufacturing accuracy of the solid freeform fabricated implants was
assessed by two methods. First, micro-computed tomography (microCT) was used
to measure the volume fraction of implant material with respect to the cylindrical
design domain. To do this, implants were scanned with an eXplore Locus SP
microCT system (GE Healthcare Preclinical Imaging, London, Ontario) using
80 kVp X-ray potential, 6000 ms exposure time and a 29 mm isotropic voxel size.
Metallic artifacts were minimized by aligning the axisymmetric implants with the
specimen holder’s rotation axis. To measure volume fraction, MicroView (v 2.2, GE
Healthcare Preclinical Imaging) was used to generate a cylindrical region of
 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247
Fig. 2. Optimized microstructures were used for generating a hierarchical porous
implant by mapping the density distribution from the macrostructure topology
optimization. The effective mechanical and mass transport properties of the
microstructures are close to theoretical upper bounds, indicating that these
microstructures simultaneously provide maximized mechanical and mass trans-
port properties for a given porosity. In this figure, K refers to the bulk modulus of
the optimized microstructure and K1 refers to the bulk modulus of a solid
microstructure (i.e. zero porosity). Similarly, D refers to the diffusivity of the
optimized microstructure and D2 refers to the diffusivity of void space (i.e. 100%
porosity).
interest (ROI) that matched the cylindrical design domain of the optimization
scheme (6.4 mm diameter by 6.4 mm height). Implant material was thresholded
to calculate the volume fraction of the ROI. For the second method of assessing
manufacturing accuracy, cross-sections from porous optimized implants were
imaged with a backscatter scanning electron microscope (SEM). To do this, two
implants were embedded in methylmethacrylate and cut with a diamond saw
(IsoMet Low Speed Saw, Buehler LTD, Lake Bluff, IL) generating roughly 1000 mm
thick sections; one implant was sectioned orthogonal to the long axis (transverse
sections) and the other parallel (longitudinal sections). Sections were wet polished
to a final thickness of 800–900 mm, carbon-coated and imaged at 10  magnifica-
tion with an Amray SEM (model 1810, KLA-Tencor Corp., Milpitas, CA). Using
Matlab (v 7.11, The MathWorks Inc., Natick, MA), digital SEM cross-sections were
then quantitatively compared to matched CAD cross-sections by overlaying
images at a pixel size of 18 mm. Each implant pixel was then categorized as
overlap (both SEM and CAD), SEM only or CAD only. For each cross-section, two
error percentages were calculated: (1) excess material was indicated by the pixel
number of SEM only and (2) deficient material by the pixel number of CAD only.
Both error percentages were normalized by the number of overall CAD pixels
within the cross-section.
3. Results
To minimize total deformation under uni-axial loading, the
optimization algorithm assigned higher density values to angled
elements that generally formed a bell-shaped topology (Fig. 3).
These higher density regions corresponded to finite elements
under high von Mises stresses. The stress distribution revealed
that load could be transmitted to the bone along the side and
bottom of the design domain.
Applying the FE topology optimization to the permutations of
implant characteristics and interface connectivity resulted in the
development of 18 macrostructural implant designs (Fig. 4). In
general, many of the model permutations resulted in subtle
changes to the implant’s topology. The most substantial effect
on design topology was associated with a fibrous or bone tissue
interface. Without a fibrous tissue zone (i.e. direct bone integra-
tion), the topology optimization tended to develop a bell-shaped
implant design with a hollow region near the base. Upon
2243
Fig. 3. The topology optimization distributed more material in high stress regions
such that the stiffness of the system was maximized for a given material volume
fraction. In this example, the entire interfacial zone was assumed to be
trabecular bone.
Fig. 4. Implant material, design domain size and interfacial connectivity were
varied to explore the effects of these variables upon the topology optimization
results. Each row of implants represents a different design domain size (67%, 42%
or 25% of the overall radial dimension). (A) and (D) are with fibrous tissue along
the entire interface, (B) and (E) are with fibrous tissue interface along the implant
sides and a bony interface along the base, and (C) and (F) are with bony tissue
along the entire interface.
introducing a fibrous tissue zone around the design domain, the
hollowed space filled with more implant material to increase the
overall stiffness of the bone-implant system. Varying implant
material only affected the topology when there was direct bone
apposition along the entire interface (i.e. both sides and bottom).
Specifically, the titanium implant had a large hollow base without
any implant material, while the porous tantalum implant had a
central column within this hollow region that connected the
implant and bone along the base of the design domain. Variations
in the size of the design domain had limited effects on topology.
For fabrication, a candidate design was chosen that repre-
sented a titanium implant with a large design domain and
trabecular bone interface (no fibrous tissue). As an additional
design step, this solid macrostructure was transformed into a
hierarchical scaffold by mapping microstructure sub-units to the
design domain’s density distribution (Fig. 5). This added process
created a porous implant structure designed to facilitate bone
infiltration and improve implant fixation. The solid, macrostruc-
ture and porous scaffold designs were then converted to CAD
format for fabrication (Fig. 6).
Both optimized implant designs were successfully fabricated
using SLS technology (Fig. 7). In general, the topology of the
fabricated implants matched the CAD data; however, some of the
smaller pores associated with the porous design were filled with
excess Ti-alloy. Volume fractions of the manufactured implants,
2244 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247
Fig. 5. Integration of the optimized macroscopic topology with the optimized
microstructural unit cells created a hierarchical porous implant. Microstructures
with 63% and 33% porosities replaced low and high density regions, respectively.
Fig. 6. Designs selected for fabrication were a solid design derived directly from
the macroscopic topology optimization and a hierarchical scaffold design gener-
ated from optimal microstructures.
as measured from microCT, were similar to CAD volume fractions.
More specifically, the manufactured solid design had a lower
volume fraction than the CAD image (34.4% versus 37.4%), while
volume fraction of the manufactured porous design was higher
than the CAD image (51.7% versus 50.3%). Seven cross-sections
(five transverse and two longitudinal) of the porous implant were
imaged with SEM, which showed that the hierarchical scaffold
was successfully fabricated (Fig. 8). However, dimensional devia-
tions, due to excess and absent material, were identified at all
cross-sections (Table 2). Error percentages for excess material
ranged from 5.5% to 11.8%, and absent material ranged from 3.3%
to 12.8%. Of note, many internal pore structures had excess
material resulting in smaller pores than designed.
4. Discussion
To improve functionality of prosthetic limbs, there are efforts
to develop bio-integrated devices. An important step in advancing
this technology will be establishing long-term, secure fixation of
the prosthesis to remnant bone. To overcome limitations asso-
ciated with contemporary bone-anchor systems, we established a
paradigm for designing, fabricating and evaluating novel ortho-
pedic implant topologies that will maintain long-term, robust
fixation in bone while undergoing loading. In the present study,
we used topology optimization to explore potential implant
geometries that minimize deformation at the bone-implant inter-
face with limited implant material. The versatility of the FE
topology optimization design scheme was demonstrated by
examining the effects of variations in implant material, design
domain size and interfacial connectivity upon the optimized
results. As an additional design step, optimal microstructures
were utilized to replace the global material density distribution
determined by the topology optimization. This created a macro–
micro implant structure that was designed to encourage deeper
bone infiltration and establish better implant-bone integration.
Selected implants were then fabricated by SLS of medical-grade
Ti-alloy with overall dimensions of 6.4  12.7 mm. Given the
small implant size, this fabrication technology proved to be
accurate with errors assessed by volume and cross-sectional
shape generally within710%.
An advantage of topology optimization is that it can be used
directly in conjunction with bone mechano-regulatory models for
developing orthopedic implants. In the present study, the opti-
mization objective was to minimize the total compliance of the
bone-implant system, which was intended to minimize tissue
strains near the bone-implant interface. This objective function
was motivated by previous research that demonstrated bone-
implant integration is related to the local mechanical environ-
ment, which in turn can be affected by implant surface geometry
(Simmons et al., 2001a, 2001b). For example, Simmons et al.
(2001a) suggested 3% distortional strain as a threshold level,
above which local bone formation is inhibited, leading to poor
integration within an early healing period. Optimal implant
geometry could be designed by limiting the strain magnitudes
at the bone-implant interface. Although we did not assign con-
straints directly on the strain levels, all of our optimal implant
designs showed less than 1% strain when minimizing compliance.
Thus, we expect early interfacial bone formation during in vivo
evaluation of our optimally designed implants.
The integration of macro–micro topology optimized structures
can bring additional versatility in the design of implant struc-
tures. Microstructural topology optimization has been utilized to
find optimal microstructures by defining the design domain
within the representative microstructure space (Sigmund, 1994).
The macroscopic or average properties are then evaluated
using the homogenization method by solving characteristic
equations within the microstructural domain (Gründemann,
1987). At each optimization iteration, material density at the
microstructure is updated to minimize the error between target
properties and current properties at the macroscopic level.
Optimal microstructures can then be designed such that the
equivalent macroscopic properties match those given as design
targets. Furthermore, multiple properties, such as mechanical
and mass transport, can be considered simultaneously in
the microstructure design (Challis et al., 2008; Kang et al.,
2010). In addition, several microstructures can be tailored to
have different properties at the same porosity (Kang et al., 2010).
This allows us to design porous implants with varying mechanical
and mass transport properties for a given macroscopic material
distribution.
 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247 2245

Fig. 7. Scanning electron microscopy (SEM) images show implants fabricated with selective laser sintered (SLS) technology.

Fig. 8. For assessing fabrication accuracy, implant cross-sections were imaged with SEM and then overlayed with CAD cross-sections.
2246 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247
Table 2
Comparison of implant cross-sections (Fabricated versus CAD).
Section
1 (Long.) 2 (Long.) 3 (Trans.)
% Excess Material (SEM only) 11.7 7.0 8.8
% Deficient Material (CAD only) 4.5 6.2 4.8
During the optimization process, implant material was removed in
low von Mises stress regions and assumed to be replaced with void.
This would be equivalent to the initial post-operative period where
bone ingrowth would have not yet occurred. Alternatively, one
could modify the material interpolation such that lower modulus
(trabecular bone) and higher modulus (implant metal) materials are
incorporated into the design domain. This would assume full integra-
tion of the host bone to the implant surface. However, since the
modulus of bone is less than that of the implant materials, the effect
of introducing low stiffness material into the design domain may not
significantly impact the optimization results.
There are several limitations associated with the implant
design scheme, topology optimization and SLS fabrication tech-
nology. Our FE model was a generic representation of a small
bone-implant system exposed to simplified boundary conditions.
Only uni-axial loads were applied to the implant for this initial
iteration of our model. While this is not a complex loading
condition, our goal at this stage was to establish a design
paradigm for generating and evaluating novel implant structures;
we did not seek to create an implant for clinical use. Focusing on
uni-axial loads also enabled us to evaluate our fabricated
implants in a companion experiment that incorporated in vivo
application of cyclic, tension-compression loads (Long et al.,
under review). Another limitation of our FE model was using
the 451 wedge to represent the entire 3D structure. This was done
to improve the computational efficiency of the optimization
algorithm; however, the wedge mesh (and the associated bound-
ary conditions) created stress artifacts near the edges. In some
cases, these artifacts affected the topology of the optimized
implant. For example, the small radial projections near the top
of the solid design fabricated from Ti-alloy are artifacts
(Figs. 6 and 7). In all cases, however, these artifacts were small
relative to the global optimized structure. While there were
minor issues with our design scheme, it is important to note that
the FE model could be modified and refined to develop other
implant structures, in which case the framework of our design
paradigm is still applicable. Regarding limitations of topology
optimization methods, future iterations of our optimization
scheme could benefit from next generation software that would
permit modeling of more complex physical phenomena. For
example, topology optimization typically requires linear elasti-
city; nonlinear material properties, nonlinear interface conditions
and large deformations may not be correctly modeled. In this
study, we applied relatively small magnitude loads, which led to
small displacements. In this situation bone is generally consid-
ered as linear elastic material (Keaveny et al., 1994). Another
limitation of current topology optimization algorithms is the
inability to account for time-dependent history of the mechanical
environment. In general, mechano-regulatory models relate
mechanical loading history to bone maintenance, so developing
algorithms that account for time-dependent mechanical stimuli
may improve the utility for orthopedic implant design. Overall, it
is important to note that topology optimization is primarily useful
for developing structures during the early stages of the design
process. In other words, results from topology optimization
should serve as initial designs, and fine tuning can be executed
Average
4 (Trans.) 5 (Trans.) 6 (Trans.) 7 (Trans.)
11.4 5.3 8.7 9.5 8.9
2.6 12.6 10.7 6.6 6.9
afterwards by using additional design and analysis tools. There
are also limitations associated with the SLS technology we used to
fabricate two optimized implant designs. In general, the small
pores of the hierarchical scaffold were smaller than designed due
to excess implant material. These pores were designed to be
350 mm, which is near the fabrication resolution of contemporary
SLS systems. Additionally, material properties of the implant will
be dependent upon SLS fabrication parameters (e.g. powder size
and laser temperature). Post-processing, like annealing, can be
used to alter material properties, such as toughness and fatigue
life (Hollander et al., 2006).
5. Conclusions
We successfully demonstrated the development of implant
designs for favorable bone fixation using topology optimization.
This approach allows for designing novel orthopedic implants for
specific loading environments that have intricate porous archi-
tectures for tissue integration. Candidate implants were then
successfully fabricated using SLS technology with medical-grade
Ti-alloy. The fabricated implants closely matched CAD data,
generally within 7 10%. The next step will be testing our implants
in an in vivo environment to identify key aspects of the designs
that promote osseointegration.
Conflicts of interest
The authors have no conflicts of interest to disclose.
Acknowledgments
This work was supported by a Multi-disciplinary University
Research Initiative (MURI) from the Army Research Office (Pro-
posal 50376-LS-MUR, Grant W911NF-06-0218) as well as the
National Institutes of Health (Grants 5T90-DK070071, 5RO1-
AR051504 and 1R01AR053379).
References
Bendsøe, M.P., Kikuchi, N., 1988. Generating optimal topologies in structural
design using a homogenization method. Computer Methods in Applied
Mechanics 71, 197–224.
Branemark, R., Branemark, P.I., Rydevik, B., Myers, R.R., 2001. Osseointegration in
skeletal reconstruction and rehabilitation: a review. Journal of Rehabilitation
Research and Development 38, 175–181.
Challis, V.J., Roberts, A.P., Wilkins, A.H., 2008. Design of three dimensional
isotropic microstructures for maximized stiffness and conductivity. Inter-
national Journal of Solids and Structures 45, 4130–4146.
Gibiansky, L.V., Torquato, S., 1996. Connection between the conductivity and bulk
modulus of isotropic composite materials. Proceedings of the Royal Socety of
London A: Mathematical, Physical and Engineering Sciences 452, 253–283.
Goldstein, S.A., Matthews, L.S., Kuhn, J.L., Hollister, S.J., 1991. Trabecular bone
remodeling: an experimental model. Journal of Biomechanics 24 (1), 135–150.
Gründemann, H., 1987. Homogenization techniques for composite media. In:
Sanchez-Palencia, E., Zaoui, A., (Eds.), Homogenization techniques for compo-
site media, Lectures Delivered at the Proceedings of the International Center
for Mechanical Sciences, CISM, Udine, Italy, 1–5 July, 1985. Springer-Verlag,
Berlin.
 H. Kang et al. / Journal of Biomechanics 45 (2012) 2241–2247
Hagberg, K., Branemark, R., 2009. One hundred patients treated with osseointe-
grated transfemoral amputation prostheses—rehabilitation perspective. Jour-
nal of Rehabilitation Research and Development 46, 331–344.
Hollander, D.A., von Walter, M., Wirtz, T., Sellei, R., Schmidt-Rohlfing, B., Paar, O.,
Erli, H.J., 2006. Structural, mechanical and in vitro characterization of indivi-
dually structured Ti–6Al–4V produced by direct laser forming. Biomaterials
27, 955–963.
Hollister, S.J., Lin, C.Y., 2007. Computational design of tissue engineering scaffolds.
Computer Methods in Applied Mechanics 196, 2991–2998.
Kang, H., Lin, C.Y., Hollister, S.J., 2010. Topology optimization of three dimensional
tissue engineering scaffold architectures for prescribed bulk modulus and
diffusivity. Structural and Multidisciplinary Optimization 42, 633–644.
Keaveny, T.M., Guo, X.E., Wachtel, E.F., McMahon, T.A., Hayes, W.C., 1994.
Trabecular bone exhibits fully linear elastic behavior and yields at low strains.
Journal of Biomechanics 27, 1127–1136.
Lin, C.Y., Hsiao, C.C., Chen, P.Q., Hollister, S.J., 2004. Interbody fusion cage design
using integrated global layout and local microstructure topology optimization.
Spine 29, 1747–1754.
Lin, C.Y., Wirtz, T., LaMarca, F., Hollister, S.J., 2007. Structural and mechanical
evaluations of a topology optimized titanium interbody fusion cage fabricated by
selective laser melting process. Journal of Biomedical Materials Research A 83,
272–279.
Loeb, G.E., 2009. Taking control of prosthetic arms. JAMA 301, 670–671.
2247
Long, J.P., Hollister, S.J., Goldstein, S.A., A paradigm for the development and
evaluation of novel implant topologies for bone fixation: in vivo evaluation.
Journal of Biomechanics, under review.
Matthews, L.S., Goldstein, S.A., 1992. The prosthesis-bone interface in total knee
arthroplasty. Clinical Orthopaedics and Related Research, 50–55.
Sigmund, O., 1994. Materials with prescribed constitutive parameters—An inverse
homogenization problem. International Journal of Solids and Structures 31,
2313–2329.
Simmons, C.A., Meguid, S.A., Pilliar, R.M., 2001a. Mechanical regulation of localized
and appositional bone formation around bone-interfacing implants. Journal of
Biomedical Materials Research 55, 63–71.
Simmons, C.A., Meguid, S.A., Pilliar, R.M., 2001b. Differences in osseointegration
rate due to implant surface geometry can be explained by local tissue strains.
Journal of Orthopedic Research 19, 187–194.
Tillander, J., Hagberg, K., Hagberg, L., Branemark, R., 2010. Osseointegrated
titanium implants for limb prostheses attachments: infectious complications.
Clinical Orthopaedics and Related Research 468, 2781–2788.
Tomaszewski, P.K., Verdonschot, N., Bulstra, S.K., Verkerke, G.J., 2010. A compara-
tive finite-element analysis of bone failure and load transfer of osseointe-
grated prostheses fixations. Annals of Biomedical Engineering 38, 2418–2427.
Xu, W., Robinson, K., 2008. X-ray image review of the bone remodeling around an
osseointegrated trans-femoral implant and a finite element simulation case
study. Annals of Biomedical Engineering 36, 435–443.