You are on page 1of 7

Artificial Organs

35(4):404410, Wiley Periodicals, Inc.

2011, Copyright the Authors
Artificial Organs 2011, International Center for Artificial Organs and Transplantation and Wiley Periodicals, Inc.

Use of a Genetic Algorithm for Multiobjective Design

Optimization of the Femoral Stem of a Cemented Total
Hip Arthroplasty

*Toshimasa Ishida, *Ikuya Nishimura, Hiromasa Tanino, Masaru Higa, Hiroshi Ito, and
Yoshinori Mitamura
*Graduate School of Information Schience and Technology, Hokkaido University, Sapporo; Department of Orthopaedic
Surgery, Asahikawa Medical College, Asahikawa; Graduate School of Engineering, University of Hyogo, Himeji; and
School of Biological Science and Engineering, Tokai University, Sapporo, Japan

Abstract: There are many designs of the femoral stem of a introduced to minimize these objective functions. The
cemented total hip arthroplasty, and mechanical failure of results showed that the geometry that leads to a decrease in
the stem is caused by several factors related to the cement, the proximal cement stress and the geometry that leads to
such as failure of the cement. Optimization of the shape of a decrease in the distal cement stress were not the same.
the stem, especially multiobjective optimization, is required However, the results of the walking and the stair climbing
to solve these design problems because a cement fracture is conditions matched. Five dominant stem designs were con-
caused by multiple factors. The objective of this study was sidered to be the Pareto solution, and one design was iden-
to determine a stem geometry considering multiple factors tified as the better design for all objective functions. It
at the same time. A three-dimensional finite element model was shown that multiobjective optimization using a genetic
of the proximal femur was developed from a composite algorithm may be used for optimizing the shape of the
femur. A total of four objective functionstwo objective femoral stem in order to avoid cement fracture. Key
functions, the largest maximum principal stress of proximal Words: Computer analysisFinite element analysis
and distal sections in the cement mantle, for each of the two OptimizationTotal hip replacementFemoral stem
boundary conditions, walking and stair climbingwere Cement failure.
used. The neighborhood cultivation genetic algorithm was

Failure of the femoral prosthesis component of a aided engineering, may help to reduce the frequency
cemented total hip arthroplasty (THA) is often of implant loosening. For example, finite element
attributed to failure of cement mantle associated with analysis (FEA), design sensitivity analysis (7), and
many stress patterns in the cement (1). shape optimization have been used to evaluate and
There have been many changes in the design of the create stem designs (811).
femoral stem of THA, with many of these changes Although optimization is a suitable technique for
being implicated in failure of the cement layer (2,3). developing femoral stems, there were many limita-
Many design and surgical factors may affect the tions in optimization procedure. Huiskes and Boek-
stress distribution related with aseptic loosening in lagen (8) and Yoon et al. (9) reported one of the
cemented femoral hip components (4,5). Therefore, earliest optimization studies. However, simple two-
proper preclinical testing (6), including computer- dimensional analysis models were used in both
reports because of lack of computer resources in the
1980s. A three-dimensional analysis cannot obtain
doi:10.1111/j.1525-1594.2010.01117.x dramatic parameter changes; however, these analyses
Received June 2009; revised July 2010. are important because of the asymmetric femoral
Address correspondence and reprint requests to Dr. Ikuya Nish- geometry. Katoozian et al. (10) reported the rela-
imura, Graduate School of Information Science and Technology,
Hokkaido University, Kita 14 Nishi 9, Kita-ku, Sapporo, Hokkaido tionship between objective functions, boundary con-
060-0814, Japan. E-mail: ditions, and optimum geometries. These authors


aor_1117 404..410

concluded that different optimization conditions led

to different results. Thus, the use of multiple condi-
tions and objective functions, called multiobjective
optimization, is one of the best methods for solving
complex optimization problems. Fernandes et al. (11)
analyzed the multiobjective optimization of a two-
dimensional cementless stem with a weighting
method. This method was useful in easily obtaining
the optimization direction. However, when there is a
large number of parameters or when detailed solu-
tions are sought, the calculation time increased
greatly because of the increase in the number of
parameter combinations.
One way of overcoming all of these limitations is to
use a genetic algorithm (GA). GA is an algorithm for
solving large-scale optimization problems. Most of
the optimization algorithms consider only one solu-
tion, whereas GA considers many solutions as one
set and renews the solution sets during the
calculation. Therefore, a sufficient number of first-
solution parameters and renewal of solution sets can
easily solve large and complex problems.
The objective of this study was to optimize the
design of a cemented stem considering the several
simultaneous causes for cement failure. For this
purpose, GA was introduced in the optimization FIG. 1. Finite element model of the proximal femur.


Modeling and design parameters

A three-dimensional finite element model of
the proximal femur was developed (Fig. 1). The
femoral geometry was determined on the basis of
the CAD data of the composite femur. Next, the
parametric stem model was developed using a CAD
program (Pro/Engineer, PTC, Inc.). The initial stem
shape was based on the results of our previous
study (12), and the 10 design parameters considered
were defined on the basis of the stem cross-sections
(Fig. 2). Later on, this initial model was used to
define the objective function. The design parameters
were defined as medial-lateral (ML) width and
anterior-posterior (AP) width of the stem proximal,
middle, and distal cross-sections. In particular, in the
case of the proximal sections, the ML width was
divided into two parameters considering the medial
and lateral sides along the stem neck axis. The stem
length was fixed at 140 mm and the head center
offset, at 38 mm.
All design parameters were labeled L1 to L10,
and the range of parameters is listed in Table 1.
These parameters are subjected to a set of constraints FIG. 2. Cross-section of the stem geometry.

Artif Organs, Vol. 35, No. 4, 2011


TABLE 1. Default values and the range of design ler and the CAD and FEA programs. The FE models
parameters used in stem CAD model comprised 11 792 8-noded brick elements and 12 667
Initial Min Max nodes. The number of elements was the same for
all analysis models. The elastic modulii of the stem
Proximal ML L1 8 6 21
L2 14 12 15 material, bone cement (polymenthyl methacrylate
AP L3 6 4 6.5 [PMMA]), cortical bone, and cancellous bone were
L4 6 4 6.5 210, 2.2, 17, and 1 GPa, respectively (12). For each of
Middle ML L5 12.6 10 13.6
AP L6 6 4 6.5 these materials, Poissons ratio was taken to be 0.3.
L7 6 4 6.5 The cementbone interface was assumed to be fully
Distal ML L8 8.8 6 9 bonded. The cementstem interface was considered
AP L9 3.5 2.5 4.5
L10 3.5 2.5 4.5 to be a Coulomb frictional interface, with a coeffi-
cient of friction of 0.3 (16). We considered two
All dimensions are in millimeters.
loading cases to simulate a physiological condition.
One corresponded to walking whereas the other cor-
responded to stair climbing. For both cases, the mag-
in order to ensure that the cross-section size
nitudes and directions of the applied loads were
decreased from the middle cross-section to the distal
taken from the work of Stolk et al. (17).

g 1 = L5 L8 < 0 Objective functions

g 2 = L6 L9 < 0
A schematic presentation of the complete optimi-
g 3 = L7 L10 < 0. zation procedure is shown in Fig. 3. Four objective
After that, a cement mantle, with a constant thick- functions are used: the largest maximum principal
ness of 2 mm was added around the stem and a stresses of the distal (walk-distal; F1, stair-distal; F2)
mantle, 30-mm thick (1315), was added below the tip and proximal (walk-proximal; F3, stair-proximal; F4)
of the stem. This model was fed into the FEA part in the cement mantle. These objective functions
program (ANSYS6.0, ANSYS, Inc., Cybernet were defined on the basis of the results of the static
Systems Co. Ltd., Tokyo, Japan) in the IGES format. stress analysis of the initial model. Therefore, all opti-
mization problems can be formulated as follows:
Static stress analysis
Minimize Fi ( x) i = 1 . . . 4
The values of all objective functions were calcu-
lated using a static stress analysis. All analysis models such that
and design parameters were automatically con-
structed or determined by the optimization control- (Lj )min < x j < (Lj )max g k (x) < 0 k = 1 . . . 3

FIG. 3. Procedure for design optimization.

Artif Organs, Vol. 35, No. 4, 2011


where Fi(x) is the objective function and (Lj)min and

(Lj)max are the upper and lower limits of the design
parameter xj.

GA parameter settings
The neighborhood cultivation genetic algorithm
(NCGA) (18) was introduced to minimize these
objective functions. NCGA is an example of a multi-
objective GA and can optimize several objective
functions while maintaining the diversity of solutions.
The GA requires the encoding of design parameters
to bit strings in order to obtain the gene parameters.
In this study, it was mounted as follows:
1 Minimum value of a design parameter is defined as
000 . . . 0 on a gene parameter.
2 Maximum value is defined as 111 . . . 1.
3 A design parameter is encoded to the nearest bit- FIG. 4. Maximum principal stress in the cement mantle, load
string area that is defined in (1) and (2). case of the walking (a) and stair-climbing (b) conditions.

A 10-bit string was used for defining the design

parameter used in this calculation. The population 1 solution number;
size was set to 50 in a single generation, and the 2 value of all design parameters L1 . . . L10, and
number of generations was set to 20. Therefore, a value of objective functions F1 . . . F4;
total of 1000 calculations were performed. During 3 where the solution number contains the generation
each successive generation, 50% of the existing popu- data.
lation was selected to determine the next generation Therefore, these charts provide the relationship
according to the fitness function. Double crossover between the objective functions.
points on both parents organism strings were
selected. In NCGA, individuals who were close to Stem design for each Pareto front
each other were chosen as the mated pair. Further- Several Pareto solutions are shown in Fig. 5 and
more, mutation took place at a probability of 0.01 in Table 2. Figure 5a shows the resulting Pareto solu-
the case of all generations. After the analysis of each tions in the two-dimensional objective function
generation, the values of objective functions were spaces. These solutions were based on an approxi-
plotted as a two-dimensional scatter chart. mate trade-off line of two objective functions (F1
and F2). In this chart, the strongest geometry for the
RESULTS proximal region is (1). At the same time, (1) is a
relatively worse geometry on the Pareto front for
Static stress analysis of pre-optimization the distal region. From this scatter chart (Fig. 5a),
The distribution of the maximum principal stress five dominant designs were selected that were on
acting in the coronal midplane of the cement mantle the Pareto front ([1][5]). Figure 5b shows the lin-
in the initial design is shown in Fig. 4. It was found earity of objective functions of F2 and F3. Clearly,
that the largest maximum principal stress in the these two objective functions are correlated and not
cement mantle was found near the tip of the stems for a trade-off. Figure 5c also shows the trade-off line of
both the walking (12.9 Mpa) and the stair-climbing two objective functions (F3 and F4). Each number
(15 Mpa) conditions, and another stress concentra- in the figure represents the same design as that in
tion was found in the proximal region of the cement another figure.
mantle. Note that these results were all bases of
objective functions.
Optimization and scatter diagram
Results of optimization are shown in Fig. 5. On all This study was an initial attempt to use a GA for the
charts, the direction of optimization is the lower left multiobjective optimization of the femoral stem of a
one. On each diagram, each point has the following cemented THA. Compared with a mathematical
properties: optimization method such as the steepest-descent

Artif Organs, Vol. 35, No. 4, 2011


a b

FIG. 5. Two-dimensional scatter chart of

optimization results. (a) The relationship
between walk-distal and walk-proximal, (b)
c walk-distal and stair-distal, and (c) stair-
distal and stair-proximal.

method, a searching method such as the GA has an long. Even with the simplified analysis performed in
advantage for complex multiparameter analysis this study, a searching method required more than
because it searches the optimum values without using 2 months of CPU time to complete all the calculations
a derivative function. Therefore, the GA will never (Intel Xeon 2.4 GHz, Windows 2000 PC). An attrac-
converge to a local extreme, and the solutions will tion of a searching method is that the individual cal-
cover a wide range. Searching methods, however, culation of each genetic parameter can easily be
could not be used until recently, because the calcula- carried out in parallel. Unlike in a mathematical opti-
tion time required to obtain good solutions was very mization method, in a searching method, each gene

TABLE 2. Stem geometry of Pareto solution on (1) to (5).

(1) (2) (3) (4) (5)
Proximal ML L1 (mm) 8.47 10.93 18.61 19.45 19.45
L2 (mm) 14.7 12.43 14 14.73 13.99
AP L3 (mm) 5.28 4.81 5.1 5.12 5.78
L4 (mm) 6.33 6.2 5.1 6.39 6.42
Middle ML L5 (mm) 13.1 13.46 13.2 13.48 10
AP L6 (mm) 6.18 5.15 6.47 6.31 6.32
L7 (mm) 5.95 6.27 4.63 4.67 4.66
Distal ML L8 (mm) 8.77 6.6 6.66 6.6 6.1
AP L9 (mm) 3.24 4.07 3.3 3.43 3.55
L10 (mm) 2.54 2.78 2.66 2.55 2.51
12.4 12.2 9.3 8.6 7
Walk-distal (Mpa)
Walk-proximal (Mpa) 3.3 3.5 3.7 3.8 4.9
Stair-distal (Mpa) 12.8 13.7 10.6 9.9 8.5
Stair-proximal (Mpa) 4.9 5.4 7.6 5.5 7.5

Artif Organs, Vol. 35, No. 4, 2011


parameter is independent from the others belonging this is the reason for the synchronized optimization
to the same generation. Thus, in this study, a 50-CPU direction.
PC cluster was used to perform all calculations in less The main limitation of this study was that the
than 2 days. failure criteria related to the boundary conditions.
For general single objective optimization, the We used the largest maximum principal stress as the
optimal implant is only one particular case, the one objective function. In some clinical reports, it has
represented in the FE model. In most cases, because of been suggested that the failure of the femoral stem is
the variations in shape and properties, the optimal attributable not only to the failure of the cement
implant will be suboptimal (19). Hence, a multiobjec- mantle but also to the failure of the cementimplant
tive approach was used for considering biomechanical interface (21,22). Another limitation was the accu-
conditions.With respect to the Pareto solutions shown racy of the finite element method model. Most prop-
in Fig. 5a, the geometries of (1) and (5) were the best erties and model shapes were derived from our
results for the two objective functions (Sp-walk and previous study (12) as a continued work. Therefore,
Sd-walk). However, both geometries have a risk for the model in this study did not include the latest
the other objective function. Therefore, geometries methods in modeling and material properties. Con-
(2), (3), and (4) were well balanced for the two objec- sidering these limitations, we can propose a new opti-
tive functions. In contrast, in the case of the stair- mization method in a future work.
climbing condition (Fig. 5c), (2) and (3) did not exist
on the Pareto front. Then, geometry (4) was the best-
balanced geometry for this analysis condition. Geom- CONCLUSION
etries (3) and (4) were almost the same except in the This study is the first report of the use of a genetic
case of the proximal cross-section. Therefore, the algorithm for optimizing the shape of the femoral
dimensions of the stem of the proximal cross-section stem in a cemented total hip arthroplasty. The opti-
should be carefully designed. mized shape was obtained considering multiobjective
Resolution of the genetic parameter was one of the functions. This design is not the best design for all
important factors that affected the accuracy of the objective functions, but it is better than the existing
solution. We used 10-bit strings as the genetic designs for all conditions because of the trade-off
parameters. The resolution of these strings was between the objective functions. Stem designers
approximately 1.5 10-2 mm in the case of the design should find the method presented in this study useful
parameter L1 (15/210). These resolutions had to be when modifying the shape of the stem.
varied according to the accuracy of stem casting.
Low-resolution bit strings increase the calculation Acknowledgment: This research was partially
speed but adversely affect the accuracy of the supported by Engineous Japan, Inc., Yokohama,
solution. In the future work, the relationship between Japan.
the lengths of the genetic bit strings and the accuracy
of the solution has to be investigated.
From Fig. 5b, it is seen that the optimization direc- REFERENCES
tion of the two boundary conditions is the same. In 1. Jasty M, Maloney WJ, Bragdon CR, OConnor DO, Haire T,
this study, the original values of all the loads were Harris WH. The initiation of failure in cemented femoral com-
obtained in an experimental study (20). In the case of ponents of hip arthroplasties. J Bone Joint Surg 1991;73-
stair climbing, the load was higher than that in the 2. Dall D, Learmonth I, Solomon M, Miles A, Davenport J. Frac-
case of the walking condition for the purpose of ture and loosening of Charnley femoral stems. J Bone Joint
rotation. Therefore, the cement stress value in the Surg 1993;75-B:25965.
3. Middleton R, Howie D, Costi K, Sharpe P. Effects of design
case of the stair-climbing condition was also higher changes on cemented tapered femoral stem fixation. Clin
than that in the case of the walking condition, but the Orthop 1998;355:4756.
tendency of results was the same in both cases. The 4. Mann KA, Bartel DL, Wright TM, Burstein AH. Coulomb
frictional interfaces in modeling cemented total hip
largest maximum principal stress value in the cement replacements. J Biomech 1995;28:126778.
mantle, which can lead to the fracture of the bulk of 5. Barrack RL. Early failure of modern cemented stems. J
the cement mantle, was higher in the case of the Arthroplasty 2000;15:103650.
6. Stolk J, Verdonschot N, Cristofolini L, Toni A, Huiskes R.
stair-climbing condition than in the case of the Finite element and experimental models of cemented
walking condition (Fig. 5). This tendency was hip joint reconstructions can produce similar bone and
observed only in the proximal region. Therefore, the cement strains in pre-clinical tests. J Biomech 2002;35:499
stair-climbing condition was detrimental to the 7. Yang RJ, Choi KK, Crowninshfield RD, Brand RA. Design
cement around the proximal region. We suggest that sensitivity analysis: a new method for implant design and ad

Artif Organs, Vol. 35, No. 4, 2011


comparison with parametric finite analysis. J Biomech 1984;17: oral stem design and centralizer. J Arthrop 2001;16-5:648
84954. 57.
8. Huiskes R, Boeklagen R. Mathematical shape optimization of 15. Mann KA, Gupta S, Race A, Miller MA, Cleary RJ, Ayers DC.
hip prosthesis design. J Biomech 1989;22:793804. Cement microcracks in thin-mantle regions after in vitro
9. Yoon SY, Jang GH, Kim YY. Shape optimal design of the fatigue loading. J Arthrop 2004;19-5:60512.
stem of a cemented hip prosthesis to minimize stress con- 16. Mann KA, Bartel DL, et al. Mechanical characteristics of the
centration in the cement layer. J Biomech 1989;22:1279 stem-cement inteface. J Orthop Res 1991;9:798808.
84. 17. Stolk J, Verdonschot N, Huiskes R. Stair climbing is more
10. Katoozian H, Davy DT. Effects of loading conditions and detrimental to the cement in hip replacement than walking.
objective function on three-dimensional shape optimization of Clin Orthop 2002;405:294305.
femoral components of hip endoprostheses. Med Eng Physics 18. Watanabe S, Hiroyasu T, et al. Neighborhood cultivation
2000;22:24351. genetic algorithm for multi-objective optimization problems.
11. Fernandes PR, Folgardo J, Ruben RB. Shape optimization of a IPSJ J 2002;43:No.SIG10 (TOM7);18398.
cementless hip stem for a minimum of interface stress and 19. Huiskes R, Hollister SJ. From structure to process, from organ
displacement. Comput Methods Biomech Biomed Engin to cell: recent developments of FE-analysis in orthopaedic
2004;7:5161. biomechanics. Trans ASME J Biomech Engin 1993;115:52027.
12. Tanino H, Ito H, Higa M, et al. Three-dimensional computer- 20. Bergmann G, Deuretzbacher G, Heller M, et al. Hip contact
aided design based design sensitivity analysis and shape opti- forces and gait patterns from routine activities. J Biomech
mization of the stem using adaptive p-method. J Biomech 2001;34:85972.
2006;39:194853. 21. Mohler CG, Callaghan JJ, Collis DK, Johnston RC. Early loos-
13. Estok DM, Orr TE, Harris WH. Factors affecting cement ening of the femoral component at the cement-prosthesis
strains near the tip of a cemented femoral component. J interface after total hip replacement. J Bone Joint Surg
Arthrop 1997;12-1:408. 1995;77-A:131522.
14. Breusch SJ, Lukoschek M, Kreutzer J, Brocai D, Gruen 22. Gardiner RC, Hozack WJ. Failure of the cement-bone
TA. Dependency of cement mantle thickness on fem- interface. J Bone Joint Surg 1994;76-B:4952.

Artif Organs, Vol. 35, No. 4, 2011