Virtual and Physical Prototyping

ISSN: 1745-2759 (Print) 1745-2767 (Online) Journal homepage: http://www.tandfonline.com/loi/nvpp20

A method to design biomimetic scaffolds for bone

tissue engineering based on Voronoi lattices

M. Fantini, M. Curto & F. De Crescenzio

To cite this article: M. Fantini, M. Curto & F. De Crescenzio (2016): A method to design
biomimetic scaffolds for bone tissue engineering based on Voronoi lattices, Virtual and
Physical Prototyping, DOI: 10.1080/17452759.2016.1172301

To link to this article: http://dx.doi.org/10.1080/17452759.2016.1172301

Published online: 19 Apr 2016.

Submit your article to this journal

Article views: 32

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=nvpp20

Download by: [RMIT University Library] Date: 30 April 2016, At: 17:27
 VIRTUAL AND PHYSICAL PROTOTYPING, 2016
http://dx.doi.org/10.1080/17452759.2016.1172301

A method to design biomimetic scaffolds for bone tissue engineering based on

Voronoi lattices
M. Fantini, M. Curto and F. De Crescenzio
School of Engineering and Architecture, Alma Mater Studiorum University of Bologna, Italy

ABSTRACT ARTICLE HISTORY

In regenerative medicine, 3D scaffolds are used to sustain the regeneration of tissues in removed or Received 9 March 2016
damaged parts of the human body. As such practices are being widely experimented in clinical Accepted 27 March 2016
applications, the design, the materials and the manufacturing process to obtain efficient 3D
KEYWORDS
biocompatible lattices are being significantly investigated. Nevertheless, most of the proposed 3D scaffolds; Voronoi
designs are based on regular 3D shapes obtained from the repetition of unit cells disposed in a diagram; generative design
three-dimensional array. This approach does not exploit the whole potential of computer-aided
design tools coupled with manufacturing capabilities for freeform shapes. In this paper, we
Downloaded by [RMIT University Library] at 17:27 30 April 2016

propose a method to model biomimetic lattices controlling the porosity and the pores size of
scaffolds to be integrated with the anatomical shape of the defect. The method has been
implemented in bone tissue case study and implements a generative design approach based on
Voronoi diagrams.

1. Introduction Generally, for bone scaffolds, the pores size should be

maintained in a given range [150–600 µm] (Cornell 1999)
Designing synthetic scaffolds for guided tissue regener-
and the porosity of the bone can differ according to the
ation is a very challenging issue. The primary function
specific bone site considered. In the literature, it is
of a scaffold is to provide a structural template that
reported that the bone trabecular morphology is a
serves to bear loads of the surrounding tissues, while
porous environment variable in a range between 50%
the cells progressively regenerate in the scaffold
and 90%, whereas the compact bone tissue has a
volume (Chevallay and Herbage 2000, Yang et al. 2001,
lower porosity [<10%] (Hollister and Kikuchi 1994,
Hollister 2005, Saxena 2005). The success of the scaffold
Salgado et al. 2004) due to the Haversian channels.
depends on several factors which are related to the form
of the lattice and to the material used to manufacture it On the material side, the ideal material should be bio-
(Ramakrishna et al. 2001, Dee et al. 2002, Molly and compatible and biodegradable, in order to allow the
Stevens 2008, Shrivats et al. 2014). total replacement of biological tissue once the regener-
On the form side, the requirements that need to be ation of tissues is completed (Hollinger and Wong 1996).
met can be classified into two main levels: (i) the external Customised porous scaffolds can be obtained by Sub-
boundary surface of the scaffold must be coherent with tractive Manufacturing (Ciocca et al. 2013a, Ciocca et al.
the anatomy of the patient to be replaced and (ii) the 2013b), starting from blocks of porous material manufac-
internal porous and interconnected microstructure of tured by different technologies through conventional
the scaffold must favour the cells proliferation in the fabrication methods, including techniques such as
entire volume to be regenerated. Therefore, the scaffold solvent casting and particulate leaching, gas foaming,
must reproduce the shape of the tissue graft and the fibre meshes and fibre bonding, phase separation, melt
ingrowth of cells needs a trabecular architecture, in moulding, emulsion freeze drying, solution casting and
which the percentage porosity and the pores size must freeze drying (Sachlos and Czernuszka 2003). Also, custo-
be controllable (Hollister et al. 2000, Wettergreen et al. mised porous scaffolds can be directly manufactured by
2005, Van Cleynenbreugel et al. 2006). These features means of advanced fabrication methods, such as solid
of the scaffold, together with the degree of pores inter- freeform fabrication (SFF) or additive manufacturing
connection (or interconnectivity), have impact on the (AM) (Naing et al. 2005, Quadrani et al. 2005, Abdelaal
time and quality of the regeneration (Williams et al. and Darwish 2011) of biocompatible and biodegradable
2005). materials.

CONTACT M. Fantini [email protected]

© 2016 Informa UK Limited, trading as Taylor & Francis Group
 2 M. FANTINI ET AL.

Several limitations are related to the conventional fab- Voronoi diagrams received and receive so much atten-
rication methods. These are mainly concerned with the tion because of their presence in nature and they are
precise control of the pores size, followed by pores geo- being exploited for the generation of shapes in different
metry, pores interconnection, spatial distribution of fields, such as architecture, jewelry and industrial design
pores and construction of internal channels within the in general.
scaffold (Peltola et al. 2008). Through AM, on the other The paper is organised as follows. The next section
hand, it is virtually possible to control the internal mor- gives information on the Voronoi diagrams and on the
phology of the scaffold. Limitations concerned with the way they work. Afterward, the workflow for the gener-
3D-printed scaffolds are due to material inhomogeneity, ation of a Voronoi-based lattice is described. Results
limited choice of materials, minimum resolution and and the way for controlling the features parameters are
accuracy of machines and, in some cases, difficulty in discussed in the last section before the conclusions.
removing the support material in very deep and small
pores (Hollister et al. 2002).
For what concerns the biomedical field, AM has been 2. Methods: generation of biomimetic
widely used for the production of customised surgical scaffolds based on Voronoi lattice
guides, bone plates and anatomical prosthesis (Ciocca
2.1. Voronoi diagram
et al. 2011, Dobbe et al. 2011, Ciocca et al. 2012,
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Mazzoli 2013). Recently, advances in AM are also Since its introduction and practical application in the
opening the way to a better employment of this technol- early twentieth century, the Voronoi diagram, initially
ogy for the production of customised scaffolds and for described by Dirichlet (1850) and Voronoi (1908), has
the computer-aided design (CAD) modelling of anatomic been a popular topic not only in computational geome-
and porous structures to be printed in 3D. try but also in a variety of disciplines. Many natural pro-
A modern approach to scaffold design is based on the cesses can be used to define particular classes of Voronoi
use of hierarchical structures created by the repetition of diagrams. Since they are related to several well-known
a unit cell of known geometry and known properties and geometrical structures, various authors believe that the
on the possibility to predict the properties of the scaf- Voronoi diagram is one of the most fundamental con-
folds since the properties of the unit cell are known. A structs defined by a discrete set of points. Therefore,
variety of geometries and geometry libraries have been such diagrams have increasingly attracted the attention
proposed and their geometrical and mechanical proper- of researchers and computer scientists in the last few
ties have been compared and reported (Hoffmann and years (Aurenhammer 1991).
Arinyo 2002, Chua et al. 2003, Sun et al. 2005, Bucklen The ordinary Voronoi diagram, for a point set and its
et al. 2008, Gabbrielli et al. 2008). However, the unit cell construction, has been studied extensively in both two
approach produces trabecular architectures that do not and higher dimensions (Okabea et al. 1994). In a 2D
really mimic the morphology of porous skeletal struc- vector space, where the Voronoi diagram is modelled,
tures. These geometries are mainly based on a construc- points, lines and polygons are primitives. The primitives,
tive solid geometry (CSG) approach and reflect, once in higher dimensions, are built upon the lower dimen-
combined with the 3D model of the bone graft substi- sionality primitives. In our case, for the third dimension,
tute, a scarcely natural appearance. we consider the polyhedrons. The Voronoi theory can
To overcome this limitation, we propose a method to be thought of as below (Van der Putte 2009).
create customised biomimetic internal scaffolds with a Let the finite Euclidean space contain a certain
porous and interconnected microstructure, as it number (n) of points (with n > 1), called seeds. Around
appears in natural bone, based on a generative design each seed a circle is drawn, with a starting radius of 0,
approach. Since generative design is not about design- expanding at the same rate until two circles touch so
ing a shape, but it is about designing the process that that a line or boundary is formed. The circles keep
builds a shape, this approach allows the user to obtain expanding until they cannot expand any further (Figure
complex shapes by an automatic modelling process, 1). The resulting graph is the Voronoi diagram for that
and to customise the results by interactively modifying set of points. The diagram is built corresponding to the
certain parameters. areas in which each point location is closest to the
The proposed method starts with the generation of seed corresponding to that area.
Voronoi diagrams. Starting from a discrete number of The mathematical equation of such a polygon V(pi)
points called seeds, a Voronoi diagram is the result of a can be expressed as follows (Okabe et al. 1992):
specific space partition in which every cell is composed
V( pi ) = {p/d( p, pi ) ≤ d( p, pj ), j = i, j = 1, . . . , n},
of the points that are closer to the generating seed.
 VIRTUAL AND PHYSICAL PROTOTYPING 3

Figure 1. Graphical representation about the formation of a 2D Voronoi diagram.

Downloaded by [RMIT University Library] at 17:27 30 April 2016

where editor integrated into Rhinoceros 5. The first sub-circuit

works on the bounding box of the patient bone geome-
. p1, … , pn is a set of distinct seeds located in Rd; try, creating a set of points randomly distributed in the
. d(p, pi) represents the Euclidean distance between bounding box volume (Populate 3D). Such points will
location p and seed pi; be the seeds to generate the 3D Voronoi structure that
. V(pi) represents the ordinary Voronoi polygon associ- divides the bounding box volume into a set of polyhedral
ated with seed pi. cells (Figure 4). Each cell, being a closed polysurface,
encloses a volume and each face of the cell coincides
In 2D, Voronoi cells are separated from each other by with the boundary face of the adjacent cell. With the
boundaries that consist of line segments. The Voronoi aim to gather a porous and interconnected lattice struc-
cell consists of a convex polygon without holes. In 3D, ture, we need to generate a single closed polysurface,
these boundaries consist of planes, and the cells namely a solid, with the seeds at the centre of the
consist of convex polyhedrons without holes. In Figure pores and void channels to interconnect these pores.
2, an example of the 3D Voronoi diagram is shown. The method here described is based on the idea of
building the lattice as a network of solid beams along
the edges of the polyhedral cells generated above. Start-
2.2. Voronoi-based lattice generation
ing from these polyhedrons, a deconstruction operation
In this section, a generative design workflow implement- (DeBrep) is applied in order to obtain the boundary rep-
ing the 3D Voronoi diagram for modelling bone tissue- resentation of each polyhedral cell, in terms of a list of
engineering scaffolds is proposed (Figure 3). The aim is faces, edges and vertices (F, E, V ). Thus, the DeBrep com-
to automatically build a trabecular structure, starting ponent allows the separation of all the faces from the 3D
from a set of points randomly distributed in the bound- Voronoi diagram. Two parallel computations are then
ing box volume of the bone defective part to be performed. The first one starts from the faces of the poly-
replaced, hereafter called patient bone geometry. This hedral cells and scales each face assuming its centroid as
process allows us to obtain a customised scaffold con- reference point with a scale factor Sf. In parallel, each
trolling the percentage porosity and pores size in order starting polyhedral cell is scaled assuming its centroid
to reach the desired values. as reference point with a scale factor Sv (Figure 5).
The generative workflow allows us to design a porous Next, the faces, edges and vertices of such smaller
and interconnected lattice ready to be 3D printed, start- polyhedral cells are obtained again through the DeBrep
ing from a data set of four inputs: the patient bone geo- component. Hence, we have the edges of the scaled
metry acquired from computed tomography (CT) scan faces of the starting cells and the edges of the faces of
and surgery planning, the number of seeds (n) and two the scaled cells. These edges are the skeleton of the tra-
scale factors (Sf and Sv). The output is generated becular structure. Then, in order to obtain the beams of
through a series of sub-routines implemented in sub-cir- the trabeculae we need to build a solid on these edges.
cuits designed in Grasshopper, the graphical algorithm This is done by generating a series of quad meshes on
 4 M. FANTINI ET AL.

Figure 2. (a) Reference cube, (b) 3D Voronoi polyhedron, and (c) Partition in 3D Voronoi polyhedrons.

the edges vertex, which can be extracted in the proper Edwin Catmull and Jim Clark devised it in 1978 as a gen-
list (Figure 6). eralisation of bi-cubic uniform B-spline surfaces to arbi-
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Before the final intersection for producing the custo- trary topology. By this component, which is contained
mised scaffold, there is another component that makes in the Rhinos plug-in Wavebird for shape reconstructing,
an essential task, based on the Catmull–Clark algorithm. a smoothed lattice mesh can be obtained (Figure 7).
This is a technique used in computer graphics to create Finally, a mesh Boolean intersection is performed
smooth surfaces by subdivision surface modelling. between the patient bone geometry mesh and the

Figure 3. Workflow for the 3D Voronoi scaffold generation.

 VIRTUAL AND PHYSICAL PROTOTYPING 5

Figure 4. Partition in Voronoi polyhedrons of the bounding box volume generated from the patient anatomy.

Voronoi lattice mesh in order to obtain the customised correlate the input parameters with the target ones. The
Downloaded by [RMIT University Library] at 17:27 30 April 2016

scaffold mesh of the bone graft (Figure 8). input parameters are:
It is clear that by varying the number of seeds (n) and
the scale factor for the faces (Sf) and for the volume (Sv) . the number of seeds (n) to be included in the bound-
of the polyhedral cells into a range from [0–1], a different ing box of the scaffold;
lattice is obtained in terms of percentage porosity and . the scale factors Sf and Sv.
pores size. In order to find the correlation between
these input parameters (number of seeds and scale The target parameters that we consider in this paper are:
factors) and the target parameters (percentage porosity
and pores size), in the next section, an analysis procedure . the percentage porosity P%;
is proposed. It is based on comparing the key parameters . the pores size.
for the scaffold evaluation such as percentage porosity,
pores size, pores interconnection and trabecular The percentage porosity P% can be easily computed as
thickness. follows:
Vbounding box − Vscaffold
P% = × 100.
Vbounding box
3. Results
Since
In order to control the final trabecular morphology of the
porous and interconnected Voronoi scaffold, we need to Vvoid = Vbounding box − Vscaffold .

Figure 6. (a) Generation and deconstruction of faces and polyhe-

Figure 5. (a) Each face is scaled respect the face’s centroid and drons, (b) particular of scaled faces and polyhedrons, and (c)
(b) each polyhedron is scaled respect the polyhedron’s centroid. quad meshes on the edges to construct the lattice.
 6 M. FANTINI ET AL.

Then, for each pair of values of the two scale factors (Sf
and Sv), we generated a sample of scaffold by varying the
number of seeds in a range of [100–1000], with incre-
mental steps of 50, and computed the percentage poros-
ity (P%). Results are shown in Figure 11 for the sample of
scaffold generated by setting Sf = Sv = 0.75.
The first observation on these results concerns the
impact of the number of seeds onto the percentage por-
osity. By varying the number of seeds, we can observe
that the variation in the percentage porosity of the scaf-
folds in a sample, characterised by constant values of Sf
and Sv, is very low. This can be explained considering
that the number of seeds determines both the number
Figure 7. (a) Trabecular mesh and (b) smoothed lattice mesh of pores and the number of trabeculae at the same
after Catmull–Clark algorithm. time. Thus, a larger number of seeds mean a larger
number of pores (voids), but also a larger number of tra-
beculae (solids), whose presence balances the total
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Actually, we can show how the number of seeds and amount of the void volume. Therefore, it can be
the scale factors Sf and Sv influence the percentage por- assumed that the scale factors Sf and Sv are related to
osity, where the expected percentage porosity (P%) of the final scaf-
fold with a very low influence of the number of seeds
(n): the greater are the scale factors the higher is the
. Sf is the polyhedrons faces scale factor; total porosity. To obtain, for example, an expected per-
. Sv is the polyhedrons volume scale factor. centage porosity P% ≈ 80%, both scale factors (Sf and
Sv) have to be set to 0.75. Of course, the geometries of
First of all, by varying each scale factor in a range of the scaffolds will differ for different combinations of Sf
[0–1] and maintaining the same number of seeds, we and Sv that allow obtaining the same value of percentage
derived a sample of scaffold and measured the percen- porosity.
tage porosity (P%), which is referred to each pair of Since for a given combination of Sf and Sv we can
values of the two scale factors. In Figure 9, three snap- control the porosity with a good accuracy, we can now
shots of a cubic scaffold (10 × 10 × 10 mm), with the observe the effect of a different number of seeds (n)
same number of seeds (n = 500), generated by setting on the morphology of the scaffold.
Sf = Sv = 0.25 for the first one, Sf = Sv = 0.50 for the The number of seeds (or the seeds density as the
second one and Sf = Sv = 0.75 for the third one, are number of seed points per unit volume) can be given
depicted. in order to gather a target pores size of the scaffold.
Afterward, we plotted the percentage porosity Hence, this is a significant parameter that has to be
obtained by varying the Sv scale factor for different identified and quantified being also an important index
values of the Sf scale factor. Results are shown in for cells habitability. The ideal populating input
Figure 10. number of seeds can be estimated in relation to the

Figure 8. Boolean intersection of the patient bone geometry and the lattices to obtain the porous scaffold.
 VIRTUAL AND PHYSICAL PROTOTYPING 7

Figure 9. Sample of scaffold (10 × 10 × 10 mm) with 500 seeds and setting: (a) Sf = Sv = 0.25 (P% = 17.4), (b) Sf = Sv = 0.50 (P% = 47.8),
(c) Sf = Sv = 0.75 (P% = 79.6).

target percentage porosity and the target pores size, by Then, to evaluate the actual size of the randomly
setting the relative desired values as input. created pores, we propose the following method. Since
Downloaded by [RMIT University Library] at 17:27 30 April 2016

To this aim, starting from the bounding box volume of the pores have irregular shapes, the idea is based on
the scaffold, the total void volume can be easily esti- the identification, for each polyhedron, of the sphere
mated as a function of the target percentage porosity that is centred at the centroid of the polyhedral cell
(P%). Assuming that all the pores have a spherical and has the same volume of the polyhedron scaled by
shape with the same diameter (Dp), the total number Sv. This will be recalled as the polyhedron–sphere
of pores is computed as the target total void volume sub- method. In Figure 12, the approximated spherical pores
divided by the volume of a single sphere of the given are depicted for a sample of scaffold with the shape of
diameter. a cube 10 mm sided (P% = 81.6 and 100 seeds). In this
By the proposed workflow for the 3D Voronoi scaffold way, we can evaluate the size of each pore as the diam-
generation, the pores (voids) can be represented by the eter of the related polyhedron–sphere, so that a compari-
polyhedral cells and, therefore, the number of pores cor- son with the literature values, included between [150–
responds to the number of seeds. 600 µm] can be provided, starting from the number of
In this way, the ideal populating input number of seeds set as input in the Voronoi diagram.
seeds (n) is estimated as the ratio of the target total Therefore, we generated a sample of scaffold (10 ×
void volume and the volume of the sphere with the 10 × 10 mm), with a bounding box volume = 1000 mm3,
diameter (Dp) representing the target pores size. setting Sf = Sv = 0.75 to obtain an expected percentage
porosity P% ≈ 80%. The number of Voronoi seeds was
Vbounding box × P%
n=
3 . estimated in relation to target percentage porosity (P%
4 Dp = 80%) and the target pores size (varying from 5.00 to
p
3 2 0.50 mm), as the ratio of the target total void volume
(800 mm3) to the volume of the sphere with the diam-
Since
eter representing the target pores size (Figure 13).
Vvoid = Vbounding box × P%. Then, the actual outcoming size of each pore of the
scaffold, generated with different target pores size

Figure 10. Effect of the scale factors Sf and Sv on the percentage Figure 11. Effect of the number of seeds on the percentage por-
porosity of the lattice (with 500 seeds). osity of the scaffold (with Sf = Sv = 0.75).
 8 M. FANTINI ET AL.

Figure 12. Polyhedron–sphere method: (a) polyhedrons scaled by Sv, (b) polyhedrons scaled by Sv and corresponding spheres, and (c)
spheres representing the pores.

(and, therefore, a different number of seeds), was generated scaffold, the maximum pores size approxi-
measured by applying the polyhedron–sphere method. mate with a good accuracy the target pores size, and
Downloaded by [RMIT University Library] at 17:27 30 April 2016

The results are reported in Figure 14. the variance is relatively low, remaining below 1 and
In Figure 15, two snapshots of a cubic scaffold 10 mm decreasing with the increase in the number of seeds.
sided, with the same scale factors (Sf = Sv = 0.75), gener- This can be explained considering that the total pores
ated with a different number of seeds (200, 800), are void, computed by the polyhedron–sphere method,
depicted. results, as expected, in 421.875 mm3, since the starting
The first observation on these results shows as the Voronoi polyhedrons (with total volume 1000 mm3) are
number of Voronoi seeds rapidly increases with the scaled by Sv = 0.75 in the three dimensions (1000
decrease in the target pores size: in the sample of scaf- mm3 × 0.75 × 0.75 × 0.75 = 421.875 mm3). The remaining
fold with the shape of a cube 10 mm sided, a target part of void (378,125 mm3) to reach the expected 80%
pores size of 5 mm corresponds to 12 seeds, while a porosity (800 mm3) is due to connective channels
target pores size of 0.5 mm correspond to 12,223 between the pores, essential to assure a correct pores
seeds. To obtain, for example, the actual size of each interconnection.
pore included in the range [150–600 µm], a suitable Moreover, the number of seeds can give us infor-
target pores size can be set to 0.6 or 0.5 mm. mation about the number of pores inside the customised
The second observation concerns the size of the scaffold. Since the input is given as the number of points
actual pores that is almost entirely smaller than the to be included in the bounding box of the patient bone
target one. Since on the generated scaffold the pores geometry, we initially assumed that the number of seeds
have different dimensions, their measures can slightly in the model of the customised scaffold depends on the
differ from the target value. We observed that in the ratio between the volume of this model and the volume
of its bounding box. Therefore, a rough estimation of the

Figure 14. Pores size (measured applying the polyhedron–

sphere method) of a cubic scaffold 10 mm sided, scale factors
Figure 13. Number of Voronoi seeds estimated in relation to the Sf = Sv = 0.75, with a target percentage porosity P% = 80%, gen-
target pores size for a cubic scaffold 10 mm sided with a target erated with different target pores size (and therefore a different
percentage porosity P% = 80%. number of seeds).
 VIRTUAL AND PHYSICAL PROTOTYPING 9

case (Figure 16(b)). Since the connective channels are

still departing from each face of the polyhedrons to
run out to the corresponding face of the neighbour,
the level of interconnectivity increases with the
number of Voronoi seeds. Moreover, we can also
observe that the size of each connective channel is rep-
resented by the area of the departing face of the polyhe-
dral cell.
Therefore, we can conclude that the mean number of
faces of the polyhedrons could give the level of intercon-
Figure 15. Sample of scaffold (10 × 10 × 10 mm) with Sf = Sv = nectivity of the generated Voronoi scaffold, and the
0.75 and setting: (a) number of seeds = 200 (target pores size mean area of the faces of the polyhedrons could give
= 2 mm) and (b) number of seeds = 800 (target pores size = 1
the mean size of the connective channels. The properties
mm).
of the connective channels have been computed as the
mean pores interconnection and mean channels size for
number of seeds (and, therefore, pores) in the final scaf- a Voronoi scaffold with a target percentage porosity (P% =
fold can be based on this proportion. Actually, depend- 80%) and varying the target pores size (from 5.00 to 0.50
Downloaded by [RMIT University Library] at 17:27 30 April 2016

ing on the shape of the patient bone geometry, a mm), as depicted in Figure 17.
different number of expected seeds (and therefore Finally, the trabecular thickness (that represents the
pores) can be included in the final mesh of the custo- solid part of the scaffold opposite to the total void
mised scaffold. volume) can be evaluated as the mean value of the dis-
Another key parameter for the scaffold evaluation is tance between the centroids of the coupled faces of
the pores interconnection (or interconnectivity) that is neighbouring polyhedral cells scaled by Sv. We have
the property, which allows the perfusion of nutrients seen that Voronoi polyhedrons, scaled by Sv, can rep-
and the extrusion of wasting material inside the cancel- resent the pores (voids) of the scaffold structure, evalu-
lous bone structure. This is the main reason for preferring ated by means of the polyhedron–sphere method,
scaffolds with a high permeability structure that also therefore, the trabecular skeleton is built up in the free
ensures the availability of higher surface area for space between these cells (Figure 18).
enhanced cell adhesion and proliferation. Therefore, The trabecular thickness has been computed for a
the degree of pores interconnection can be considered Voronoi scaffold with a target percentage porosity (P% =
as important as pores size. 80%) and by varying the target pores size (from 5.00 to
In the case of the generated Voronoi scaffold, we tried 0.50 mm), as depicted in Figure 19.
to understand how each pore communicates with its
neighbours, and in particular to find the number and
4. Discussion
the size of the connective channels departing from
each pore. The main findings of this work concern two aspects,
We have already seen that the pores (voids) can be which are the general design method and the CAD
represented by the polyhedral cells scaled by Sv and approach followed. The general design method resulted
that the actual size of the pores can be evaluated by very efficient since the generation of biomimetic porous
means of the polyhedron–sphere method. lattices, shaped on the patient anatomy, is very intuitive
If we look at the void component (pores plus connec- and fast. Moreover, great accuracy in the expected per-
tive channels) of a very simple cubic Voronoi scaffold (10 centage porosity and pores size has been achieved by
mm sided), generated with only eight seeds correspond- setting the suitable input parameters. The CAD approach
ing to the bounding box vertices and minimal scale also resulted positive since a closed and watertight mesh
factors (Sf = Sv = 0.25), we can observe how the eight of the customised porous scaffold, for different bone
pores are mutually interconnected (Figure 16(a)). grafts, have been obtained in a short time. This allows
For each of the eight polyhedrons (that are cubes in to directly interfacing the generative modelling module
this case), six connective channels depart, one from for the generation of porous interconnected scaffolds
each face of the cube, giving the impression that the with AM systems.
resulting structure is sixfold connected. If we add These findings can be explained by the fact that the
another seed (the centroid of the bounding box), we interactive generative design approach, which has
can now observe nine polyhedral cells with different been developed from the Voronoi diagram, is indepen-
shapes and more faces with respect to the previous dent by the final shape of the scaffold and allows to
 10 M. FANTINI ET AL.
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Figure 16. Void component (pores plus connective channels) for a Voronoi scaffold with minimal scale factors (Sf = Sv = 0.25) and (a)
number of seeds = 8 and (b) number of seeds = 9.

simply modify the input parameters in suitable ranges for based on CSG, boundary representation (B-Rep) and, in
satisfying the surgical requirements in terms of both per- the last few years, on implicit surfaces such as triply per-
centage porosity and pores size. iodic minimal surfaces (TPMS). Only a case study based
In a recent review about the design of scaffold for on Voronoi diagrams for general irregular porous struc-
bone tissue engineering, considering the emergence of tures modelling has been reported (Kou and Tan 2010).
SFF methods (Velasco et al. 2015), the main reported However, this paper does not investigate key parameters
design techniques, in previous relevant works, are for bone replacement scaffolds generation, such as

Figure 17. Mean pores interconnection and mean channels size of a cubic scaffold 10 mm sided, scale factors Sf = Sv = 0.75, with a
target percentage porosity P% = 80%, generated with different target pores size (and therefore different number of seeds).
 VIRTUAL AND PHYSICAL PROTOTYPING 11

Figure 18. Trabecular thickness evaluation: (a) Voronoi polyhedral cells scaled by Sv with the distance of the centroids of coupled faces,
(b) Voronoi polyhedral cells scaled by Sv with the trabecular mesh, and (c) the final scaffold (Sf = Sv = 0.75; number of seeds n = 12;
target pores size Dp = 5 mm).

percentage porosity, pores size, pores interconnection AM is evolving from rapid prototyping to the end-of-
Downloaded by [RMIT University Library] at 17:27 30 April 2016

and trabecular thickness. Moreover, no practical rec- use product manufacturing and offers numerous
ommendations are provided for guiding the design of benefits for innovative design solutions. Nowadays,
internal trabecular morphology of the bone scaffold. design freedom has been significantly broadened,
Furthermore, the new and important aspect of this including shape complexity, material complexity, hier-
study is not the design of a specific scaffold, but the archical complexity and functional complexity (Gibson
development of a general interactive method to design et al. 2010, Yang and Zhao 2015).
different customised scaffolds according to the specific The emerging field of design for additive manufactur-
requirements provided by the clinical team. Therefore, ing (DFAM) is exploring new methodological frameworks
the proposed method can be a useful tool for the and design rules, but is also highlighting geometrical
design of biomimetic scaffolds for bone regeneration. limitations and constraints to be considered (Adam and
Nevertheless, together with the design, materials and Zimmer 2015, Kumke et al. 2016).
manufacturing processes also play fundamental roles in Therefore, to validate the manufacturability of the
obtaining efficient 3D biocompatible scaffolds with this proposed biomimetic scaffolds based on the Voronoi
internal trabecular microstructure. For this reason, the diagram, a review of possible methods that can be
capabilities of AM, according to the biomaterials used, used in the actual fabrication, in terms of AM systems
should allow the manufacturability of these scaffolds in and biomaterials, is reported. The three main classes of
terms of percentage porosity, pores size, pores intercon- synthetic biomaterials used today are metals, ceramics
nection and trabecular thickness. and polymers.
The class of biomaterials with a long history of appli-
cation in bone implants comprises metals and alloys.
Among them, the use of stainless steels, cobalt (Co), tita-
nium (Ti) and relative alloys are well proven due to their
good biocompatibility, satisfactory mechanical strength
and superior corrosion resistance. Therefore, metal-
based additive manufacturing (MAM) techniques, such
as selective laser melting (SLM) and electron beam
melting, are increasingly being used for the fabrication
of porous metals for bone scaffolds and orthopaedic
implants, with predefined external shape and internal
architecture (Wang et al. 2016).
For example, Yan et al. (2015) reported that Ti–6Al–4V
biomorphic lattices for bone implants were manufac-
tured by SLM. The design was based on gyroid and
Figure 19. Mean pores interconnection and mean channels size
diamond TPMS, having an interconnected high porosity
of a cubic scaffold 10 mm sided, scale factors Sf = Sv = 0.75, with
a target percentage porosity P% = 80%, generated with different of 80–95% and pores size in the range of 560–1600
target pores size (and therefore a different number of seeds). and 480–1450 μm, respectively. The manufacturability,
 12 M. FANTINI ET AL.

microstructure and mechanical properties were evalu- polymer designed for bone tissue-engineering appli-
ated and the comparison between 3D micro-CT recon- cations. EnvisionTEC Perfactory 3D Printer is a rapid pro-
structed models and original CAD models of the Ti– totyping manufacturing system, based on digital light
6Al–4V TPMS lattices showed excellent reproduction of processing technology, to create 3D models that range
the designs. from the conceptual to the fully functional. Some
Also, Taniguchi et al. (2016) studied the effect of pores examples of fabrication of porous PPF scaffolds, as a
size on bone ingrowth into porous titanium implants fab- huge opportunity for the regenerative medicine, have
ricated by SLM, since this AM technique has the ability to been proposed.
produce metallic scaffolds with accurately controlled Wang et al. (2015) reported the manufacturing of
porosity, pores size and interconnectivity for orthopaedic porous, cylindrical scaffolds, consisting of modular ring-
applications. Three porous titanium implants, based on a shaped bases, with 500 μm wall thickness, two pore
diamond lattice basic structure (with an intended poros- sizes (400 and 800 μm) and two different porosities
ity of 65% and pores size of 300, 600 and 900 μm), were (25% and 50%). More recently, Luo et al. (2016) have
successfully manufactured by the EOSINT-M270 SLM described the fabrication of a porous, cylindrical scaffold,
system. The porous implants were evaluated using CT by using the Schoen gyroid TPMS geometry, with 400
and the actual average pores size resulted of 309, 632 μm strut thickness, pore diameter of 1400 μm, and por-
and 956 μm, respectively. osity of 88.2%.
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Finally, Sing et al. (2015) investigated the effect of Actually, the next step of this work will be the con-
design and process parameters on the dimensional accu- crete fabrication of the proposed Voronoi-based lattices
racy of titanium lattice structures fabricated by using to directly assess the manufacturability. Moreover, as a
SLM. Geometrical design parameters, such as unit cell natural extension of this work could aim to find a
type and strut diameter, had not significant effects on proper seeds distribution to allow the generation of scaf-
the dimensional accuracy of the lattice structures. On folds with non-homogenous porosity, for example, with
the other hand, the processing parameters, such as higher density for the compact external part (cortical or
laser power and laser scan speed, had a significant compact bone tissue) and lower density for the trabecu-
effect on the powder adhesion on the struts and conse- lar internal one (cancellous or spongy bone tissue). Since
quentially affected the dimensional accuracy. the initial distribution of Voronoi seeds influences the
In the field of ceramic biomaterials, Sabree et al. (2015) final scaffold trabecular morphology, non-homogenous
reported on how highly porous ceramic scaffolds have scaffolds could be obtained with a higher concentration
been fabricated by using stereolithography and the of seeds distribution in the external part and lower in the
lost-mould process combined with gel-casting. All scaf- internal one. This method can also be refined and/or
folds had a simple cubic strut structure with an internal expanded on other requirements, such as the mechan-
porosity of approximately 42% and internal pore dimen- ical properties, once the density of seeds has been
sions in the range of 300–600 μm. related to such properties.
For what concerns the class of polymers as biomater- However, some limitations of this study should also be
ials, another method for the actual fabrication of considered, since a drawback of this approach is related
Voronoi-based lattices could be the manufacturing as to the generation of very large scaffolds due to the com-
poly(propylene fumarate) (PPF) scaffolds by using an putational resources required.
EnvisionTEC Perfactory 3D Printer. Poly(propylene fuma-
rate) is an important biodegradable and crosslinkable
5. Conclusions
In this paper, we proposed a method to design intercon-
nected porous lattices that mimic specific tissue charac-
teristics for the realisation of bone regenerative scaffolds.
The shape of reference is a healthy spongy bone, which
presents a heterogeneous structure to enhance the cells
proliferation and the tissue regeneration. The advantage
of this method, compared to other approaches, is to
provide geometrical heterogeneity thus resulting in a
really biomimetic shape (Figure 20).
The method is applied to provide an intuitive and fast
Figure 20. (a) 3D generated Voronoi scaffold and (b) healthy tool to create biomimetic scaffolds and is expandable to
spongy bone (Marinozzi et al. 2012). other tissue-engineering requirements or to other
 VIRTUAL AND PHYSICAL PROTOTYPING 13

domains in which the design porous interconnected Hoffmann, C., and Arinyo, R.J., 2002. Parametric modelling. In: G.
structures is needed. Farin, J. Hoschek and M.S. Kim, eds. Handbook of CAGD.
Amsterdam: Elsevier, 519–541.
Hollinger, J., and Wong, M.E., 1996. The integrated processes of
hard tissue regeneration with special emphasis on fracture
Disclosure statement healing. Oral Surgery, Oral Medicine, Oral Pathology, Oral
Radiology, and Endodontology, 82 (6), 594–606.
No potential conflict of interest was reported by the authors. Hollister, S.J., 2005. Porous scaffold design for tissue engineer-
ing. Nature Materials, 4 (7), 518–524.
Hollister, S.J., and Kikuchi, N., 1994. Homogenization theory and
digital imaging: a basis for studying the mechanics and
References
design principles of bone tissue. Biotechnology and
Abdelaal, O.A., and Darwish, S.M., 2011. Fabrication of tissue engin- Bioengineering, 43 (7), 586–596.
eering scaffolds using rapid prototyping techniques. World Hollister, S.J., et al., 2000. An image based approach for design-
Academy of Science, Engineering and Technology, 59, 11–27. ing and manufacturing craniofacial scaffolds. International
Adam, G.A.O., and Zimmer, D., 2015. On design for additive Journal of Oral and Maxillofacial Surgery, 29 (1), 67–71.
manufacturing: evaluating geometrical limitations. Rapid Hollister, S.J., Maddox, R.D., and Taboas, J.M., 2002. Optimal
Prototyping Journal, 21 (6), 662–670. design and fabrication of scaffolds to mimic tissue properties
Aurenhammer, F., 1991. Voronoi diagrams – a survey of a fun- and satisfy biological constraints. Biomaterials, 23, 4095–
damental geometric data structure. AMC Computing 4103.
Downloaded by [RMIT University Library] at 17:27 30 April 2016

Surveys, 23 (3), 345–405. Kou, X.Y., and Tan, S.T., 2010. A simple and effective geometric
Bucklen, B.S., et al., 2008. Bone-derived CAD library for assembly representation for irregular porous structure modeling.
of scaffolds in computer-aided tissue engineering. Virtual Computer-Aided Design, 42 (10), 930–941.
and Physical Prototyping, 3 (1), 13–23. Kumke, M., Watschke, H., and Vietor, T., 2016. A new methodo-
Chevallay, B., and Herbage, D., 2000. Collagen-based biomater- logical framework for design for additive manufacturing.
ials as 3D scaffold for cell cultures: applications for tissue Virtual and Physical Prototyping, 11 (1), 3–19.
engineering and gene therapy. Medical & Biological Luo, Y., et al., 2016. Synthesis and biological evaluation of well-
Engineering & Computing, 38 (2), 211–218. defined poly(propylene fumarate) oligomers and their use in
Chua, C.K., et al., 2003. Development of a tissue engineering 3D printed scaffolds. Biomacromolecules, 17 (2), 690–697.
scaffold structure library for Rapid prototyping. Part 1: inves- Marinozzi, F., et al., 2012. Variability of morphometric par-
tigation and classification. The International Journal of ameters of human trabecular tissue from coxo-arthritis and
Advanced Manufacturing Technology, 21, 291–301. osteoporotic samples. Annali dell’Istituto Superiore di Sanita,
Ciocca, L., et al., 2013a. CAD-CAM generated hydroxyapatite scaf- 48 (1), 19–25.
fold to replace the mandibular condyle in sheep: preliminary Mazzoli, A., 2013. Selective laser sintering in biomedical engin-
results. Journal of Biomaterials Applications, 28 (2), 207–218. eering. Medical & Biological Engineering & Computing, 51 (3),
Ciocca, L., et al., 2013b. Mesenchymal stem cells and platelet gel 245–256.
improve bone deposition within CAD-CAM custom-made Molly, M., and Stevens, M.M., 2008. Biomaterials for bone tissue
ceramic HA scaffolds for condyle substitution. Biomed engineering. Materials Today, 11 (5), 18–25.
Research International. Article ID 549762, 10 pages. Naing, M.W., et al., 2005. Fabrication of customised scaffolds
Ciocca, L., et al., 2011. Direct metal laser sintering (DMLS) of a using computer-aided design and rapid prototyping tech-
customized titanium mesh for prosthetically guided bone niques. Rapid Prototyping Journal, 11 (4), 249–259.
regeneration of atrophic maxillary arches. Medical & Okabe, A., Boots, B., and Sugihara, K., 1992. Spatial tessellations –
Biological Engineering & Computing, 49 (11), 1347–1352. concepts and applications of voronoi diagrams. Chichester:
Ciocca, L., et al., 2012. A CAD/CAM-prototyped anatomical con- Wiley.
dylar prosthesis connected to a custom-made bone plate to Okabe, A., Boots, B., and Sugihara, K., 1994. Nearest neighbour-
support a fibula free flap. Medical & Biological Engineering & hood operations with generalized Voronoi diagrams: a
Computing, 50 (7), 743–749. review. International Journal of Geographical Information
Cornell, C.N., 1999. Osteoconductive materials and their role as Systems, 8 (1), 43–71.
substitutes for autogenous bone grafts. Orthopedic Clinics of Peltola, S.M., et al., 2008. A review of rapid prototyping tech-
North America, 30 (4), 591–598. niques for tissue engineering purposes. Annals of Medicine,
Dee, K.C., Puleo, D.A., and Bizios, R., 2002. An introduction to 40 (4), 268–280.
tissue–biomaterial interaction. Vol. 3, pp. 37–52. Wiley. Quadrani, P., et al., 2005. High-resolution 3D scaffold model for
Dobbe, J.D.D., et al., 2011. Patient-tailored plate for bone fix- engineered tissue fabrication using a rapid prototyping tech-
ation and accurate 3D positioning in corrective osteotomy. nique. Medical & Biological Engineering & Computing, 43 (2),
Medical & Biological Engineering & Computing, 51 (1), 19–27. 196–199.
Gabbrielli, R., Turner, I.G., and Bowen, C.R., 2008. Development Ramakrishna, S., et al., 2001. Biomedical applications of
of modelling methods for materials to be used as bone sub- polymer-composite materials: a review. Composites Science
stitutes. Key Engineering Materials, 361–363, 903–906. and Technology, 61, 1189–1224.
Gibson, I., Rosen, D.W., and Stucker, B., 2010. Additive manufac- Sabree, I., Gough, J.E., and Derby, B., 2015. Mechanical proper-
turing technologies: rapid prototyping to direct digital manu- ties of porous ceramic scaffolds: influence of internal dimen-
facturing. New York, NY: Springer. sions. Ceramics International, 41 (41), 8425–8432.
 14 M. FANTINI ET AL.

Sachlos, E., and Czernuszka, J.T., 2003. Making tissue engineer- Master thesis. Geographical Information Management and
ing scaffolds work. Review on the application of solid free- Applications. Utrecht University, Netherlands.
form fabrication technology to the production of tissue Velasco, M.A., Narváez-Tovar, C.A., and Garzón-Alvarado, D.A.,
engineering scaffolds. European Cell Materials, 5, 29–40. 2015. design, materials, and mechanobiology of biodegrad-
Salgado, A.J., Coutinho, O.P., and Reis, R.L., 2004. Bone tissue able scaffolds for bone tissue engineering. Biomed Research
engineering: state of the art and future trends. International. Article ID 729076, 21 pages.
Macromolecular Bioscience, 4 (8), 743–765. Wang, M.O., et al., 2015. Evaluating 3D-printed biomaterials as
Saxena, A.K., 2005. Tissue engineering: present concepts and strat- scaffolds for vascularized bone tissue engineering.
egies. Journal of Indian Association of Pediatric Surgeons, 10, 14–19. Advanced Materials, 27 (1), 138–144.
Shrivats, A.R., McDermott, M.C., and Hollinger, J.O., 2014. Bone Wang, X., et al., 2016. Topological design and additive manufac-
tissue engineering: state of the union. Drug Discovery turing of porous metals for bone scaffolds and orthopaedic
Today, 19 (6), 781–786. implants: A review. Biomaterials, 83, 127–141.
Sing, S.L., et al., 2015. characterization of titanium lattice struc- Wettergreen, M.A., et al., 2005. Computer-aided tissue engin-
tures fabricated by selective laser melting using an adapted eering of a human vertebral body. Annals of Biomedical
compressive test method. Experimental Mechanics. Advance Engineering, 33 (10), 1333–1343.
online publication. doi:10.1007/s11340-015-0117-y Williams, J.M., et al., 2005. Bone tissue engineering using poly-
Sun, W., et al., 2005. Bio-CAD modeling and its applications in caprolactone scaffolds fabricated via selective laser sintering.
computer-aided tissue engineering. Computer-Aided Design, Biomaterials, 26 (23), 4817–4827.
37, 1097–1114. Yan, C., et al., 2015. Ti-6Al-4V triply periodic minimal surface
Taniguchi, N., et al., 2016. Effect of pore size on bone ingrowth structures for bone implants fabricated via selective laser
Downloaded by [RMIT University Library] at 17:27 30 April 2016

into porous titanium implants fabricated by additive manu- melting. Journal of the Mechanical Behavior of Biomedical
facturing: an in vivo experiment. Materials Science and Materials, 51, 61–73.
Engineering: C, 59, 690–701. Yang, S., et al., 2001. The design of scaffolds for use in tissue
Van Cleynenbreugel, T., et al., 2006. Micro-CT-based screening engineering. Part I traditional factors. Tissue Engineering, 7
of biomechanical and structural properties of bone tissue (6), 679–689.
engineering scaffolds. Medical & Biological Engineering & Yang, S., and Zhao, Y.F., 2015. Additive manufacturing-enabled
Computing, 44 (7), 517–525. design theory and methodology: a critical review. The
Van der Putte, T., 2009. Using the discrete 3D Voronoi diagram for International Journal of Advanced Manufacturing Technology,
the modelling of 3D continuous information in geosciences. 80, 327–342.