DIMENSIONAL EVALUATION OF TAILOR-MADE CRANIOPLASTY IMPLANTS

Dalberto Dias da Costa1, Sérgio Fernando Lajarin2.

Universidade Federal do Paraná, DEMEC, Av. Cel. Francisco H. dos Santos, 210 CEP 81531-990, Curitiba, Brasil,
1 [email protected]
2 [email protected]

Abstract: Cranioplasty is a medical technique used to The application of alloplastic materials has growing in
recover large cranial defects and has presented a great the last years, in particularly because the implants can be
development in last decades. This high evolution is credited previously produced and according to some researchers
to the discovery of new biomaterials and the efficiency of [1,2,3,4] contribute to the reduction of surgical time and
information systems, such as CT (Computed Tomography); infections risks.
MIP (Medical Image Processing); CAD (Computer-Aided Despite of alloplastic materials couldn’t replace bone
Design); CAM (Computer-Aided Manufacturing) and CNC tissues in all of its functions; some of them, in special
(Computer Numerical Control). However, the quality of the titanium, hydroxyapatite and polymethylmetacrilate
prefabricated implants, concerning to dimensional and form (PMMA) provide the necessary brain protection.
errors, still remains poorly investigated. The aim of this The enhancements of medical imaging technologies
paper is presented and discussed the method and results of a observed in the last decades, such as tomography, magnetic
dimensional evaluation of a cranioplasty implant resonance and ultrasound, have provide the basements for a
manufactured from a set of computer tomography (CT) new discipline named Medical Imaging Processing (MIP)
images. The presented article contains a short review which make intensive use of information systems to process
concerning about the manufacturing process used to produce an convert to a vetorized representation the images acquired
customized implants for medical application, highlighting from human organs.
the importance of the metrological resources available for Computer Tomography (CT), for example, is a MIP
the assessment of dimensional quality of this kind of variant that allows the modeling of hard tissues by creating a
product. The methodology contains the steps and resources geometrical representation of complex bones like those
employed in the dimensional evaluation of an acrylic found in craniofacial skeleton.
implant cast from a dried human skull. The influence of the The combination of Computer-Aided Design (CAD),
quality of CAD models was evaluated with a Coordinated Computer-Aided Manufacturing (CAM) systems, whose
Measuring Machine (CMM) by measuring a standard sphere knowledge is highly disseminated in the mechanical field,
used to probe tip calibration. Despite of the relative low and CT imaging have provide the basic resources to model
deviation found in the manufactured implant, as compared and manufacture tailor-made implants.
to the current reports in the specialized literature, it was However, the quality of the prefabricated implants,
observed that the quality of digital models and the concerning to dimensional and form errors remains poorly
localization procedures impact directly on the measured investigated. The aim of this paper is presented and
values. discussed the method and results of the dimensional
assessment of a cranioplasty implant cast from a set of
Keywords: three-dimensional modeling, coordinate computer tomography (CT) images.
measuring machine, localization, cranioplasty. A short review of the up-to-date processes applied to
manufacture cranioplasty implants is presented next with
special emphasis on the importance of the metrological
1. INTRODUCTION resources demanded to inspect this kind of complex product.
The restoration of large cranial defects caused by 1.1. Rapid Prototyping
traumas or congenital represents a difficult task in medicine.
Rapid Prototyping (RP) is a generic name given to a
This difficult is partially due the unavailability of a bonelike
set of processes based on the deposition, layer-by-layer, of a
material and the poor cosmetic results in case of autograft
thin slice material until a solid form is achieved [5].
implantation [1].
According to Leong et al. [5] and Dietmar et al. [6], this gases, like oxygen and nitrogen. This drawback results in
kind of processing has become a valuable alternative to more expensive implants.
produce customized craniofacial models.
1.4. Machining
In general, the digital model of injured region is a solid
model represented by a finite set of non-regular triangles. Machining processes are applied to produce very
Therefore, the RP software needs to slice it in a limited precise products from the progressive removing of an
number of contours before start the deposition phase, which unwanted portion of a blank material. Despite of their lager
can be accomplished, among other, by laser sintering (SLS), dissemination in metalworking industries, they present some
fusion and deposition (FDM) or three-dimensional printing limitation to fabricate smaller and complex geometries,
[5]. which are easily manufactured by casting or rapid
On the presence of surfaces with great curvatures, the prototyping techniques. Theses limitations arise from the
number of slice needs to be enlarged in order to reduce the interaction occurring between the cutting tool and the solid
dimensional and form errors which are influenced by the blank. In addition, in order to produce lighter implants the
stair-case effect [7]. In addition, some material presents a volume of removed material (chips) needs to be very high,
high shrinkage after solidification, what implies in another what implies in a significant loss.
source of error. Some researchers have reported on the application of
Choi et al.[8] report about dimensional error on RP machine processes, in special the milling, to produce master
models of a craniomaxillary complex and a mandible. They templates [1] or molds [3] for further processing. However,
attribute to the average error the low spatial resolution the use of direct milling is unusual. Heissler et al. [12] have
achieved in CT phase and the difficulty in determine a mentioned the drawbacks of direct milling of titanium
correct threshold value during image processing. blocks, including the material waste and the long lead time.
Unfortunately, the measurement method was based on a Bazan [4] produce a milled implant from PMMA
simple vernier caliper positioned above craniometrical (acrylic) block. By choosing a simpler model and a cheaper
landmarks, which contribute for a poor repeatability of material, that author demonstrates some advantages of the
measured values. direct machining. Among them, it was highlighted a low
dimensional deviation between the digital model and the
1.2. Forming
produced implant.
When implants are mass produced, forming is
1.5. Assessment of Dimensional and Form Deviations
considered a very attractive process, due the low cost and
high precision achieved in the final product. By the other The evaluation of dimensional errors in customized
hand, the high investment required to project and build a implants is not a simple task. Some researchers [4,8,13]
specific die implies in a costly final product in the case of have presented the dimensional differences among the
tailor-made implants. original model, or the digital one, and the manufactured part
In addition, the selected biomaterial needs to present a basing on simpler techniques, such as the distances
considerable plasticity in order to deform in correspondence measured with calipers.
to the mechanical tension imposed by punch and the die In almost the cases, implants are constituted of
block. In almost the cases, forming is applied to produce complex surfaces, what implies in a higher degree of
auxiliary materials for general applications, such as difficulty for manual measurements. Despite of their
dynamical meshes or fixturing devices. relevance for a rough evaluation, there is a lack of
The application for forming to produce customized confidence concerning the real differences among the
implants is scarcely reported in literature. Carr et al. [9] models and the implants. In addition, there is an absence of
used a machined epoxy die to draw a titanium plate. Joffe et an ideal model, or nominal values, for reference purpose,
al. [10] made master templates from machined polyurethane since the source model is a digital representation constructed
foam in order to copy the defect borders and press them in from CT images.
titanium plates. Nowadays, there is a branch of dimensional metrology
specialized in the evaluation of complex surfaces, also
1.3. Casting
referred as “free-form inspection”. In that case, the
Traditionally, casting is considered a process where a instrumentation is more sophisticated than the manual
liquid material is solidified inside a preexisting mold. As the calipers. There are two distinct techniques: the contact and
material remains liquid during a relatively long time non-contact inspection. Contact measuring, which can be
interval, very intricate forms can be produced with low cost. performed in Coordinated Measuring Machines (MMC) are
Some kind of biopolymers like PMMA, for example, can be based on a probe triggering device. By the other hand, the
intraoperative, not requiring the construction of dedicated non-contact method allows a faster scanning of the
molds and has been claimed as an advantageous interested region by means of a laser technology [14,15].
cranioplasty technique [11]. Independently of the employed resources, there is a
In the case of titanium, a more controllable processes fundamental problem to solve before staring the measuring,
needs to be employed, because it is a very reactive metal, in which is concerned to how to localize a free-form part
special on higher temperature. In general, a vacuum casting related to a fixed measurement frame.
should be applied to avoid the reaction with the surrounding Initially, the digital model is defined in a properly
coordinate frame called Design Coordinated System (DCS).
By the other hand, the part to be measured is positioned in in X-ray scanning. This kind of error is more significant in
the Measurement Coordinate System (MCS). Therefore, the surfaces with higher curvature. In this situation, the last slice
great deal is to establish a relationship between these two will create a flat area at the ending of the scanning
systems, which is currently called localization problem [14]. processes. The presence of a undesired flat area can influence
The Figure 1 provides a graphical illustration of this the localization procedures and difficult the establishment of
problem. the relationship between the DCS and MCS.

2. MATERIALS AND METHODS

The procedures proposed here started with the work
developed by Bazan [4], and also reported by Da Costa et al.
[18], about the direct machining and evaluation of one
PMMA implant. The primary information source of those
works was a dried human skull which was modeled from CT
scanning. The reconstructed three-dimensional CAD model
was generated in STL(Stereolithography) format. Obviously
that kind of representation is an approximation and its
quality depends strongly on the number of triangles.
In modelled area of interest (top calvaria, illustrated in
Figure 2, on top), 145696 triangles were used. That digital
model was used later to produce a physical part by direct
machining of a PMMA block.
Figure 1 – Schematic representation of the localization problem . With the purpose of manufacturing the same physical
model, however, by applying another manufacturing
Traditionally, localization is achieved by presenting the process, a dental gypsum mold was machined for further
part at a desired position and orientation, using special tools, PMMA polymerization.
fixtures or other part presentation/orientation devices totally In a second stage the influence of the model quality on
dedicated for specific products. In practical applications, the localization error using a standard sphere was
localization can be regarded as a two-step process: (i) find investigated. In the third stage, the implant produced in the
the point-to-point corresponding relationship between gypsum mold was evaluated by the procedures applied in
measurement and design surfaces; and, (ii) solve the 3D the sphere measurement. These three stages are described
rigid transformation between these two surfaces to bring next.
them into a common coordinate system [14]. 2.1. Tailor-made implant of PMMA
An interesting technique to correct the location is
presented by Spitz [16] and Lai and Chen [17]. This is A casting mold was machined from the digital model
known as principle 3-2-1, a group of data is inspected with of a dried human skull reconstructed by Bazan[4]. The mold
the intention of establishing a reference grating in relation to was composed of two halves. One is the punch (calvaria´s
the component. Three points are measured from the first inner surface) and other is the die (calvaria´s outer surface),
datum to establish a plane. Two points are measured from as showed in Figure 2.
the second datum to establish a second plane perpendicular
to the first. Finally, one point is measured from the last
datum perpendicular to the first two, establishing in that way
the origin and three-dimensional direction of the system of
coordinates. However, as Li and Gu [14] pointed out this
approach has shortcomings because the parts are required to
have plane surfaces, and the result is very sensitive to
measurement and dimensional errors. Besides, because of
the manufacturing errors on the datums, the localization
quality is sensitive to the selection of those points.
Many approaches and methods have been developed
for localization with higher accuracy, efficiency and
robustness. An example is presented by Lai and Chen [17]
that developed a method of refined positioning through
iterative measurements. That and other similar methods [14]
are applied after a rough localization by the 3-2-1 principle.
In the case of cranioplasty implants, besides the
difficulty related to implant localization, there is another
source of error due the previously methods applied to
produce the digital model. As pointed out by some authors
[1,4,8,13] computer tomography implies in a volume- Figure 2 – STL Digital model obtained by Bazan [4] (top), and the
averaging error, which arise from the slice thickness adopted die/punch of the mold (down).
 The external structures of the mold were made from
epoxy, which was further filled with dental gypsum. A
CAM system (EdgeCAM® from Pathtrace) was applied to
generate the tool path to mill the gypsum core. All machine
phase was undertaken in vertical axis machine center
(Discovery 4022 – Romi S.A). The milled surfaces were
impregnated with silicon demolder (UltraLub®).
An auto-polymerized acrylic resin composed of two
parts of liquid (monomer) and three parts of powder
(polymerized beads) was mixed at room temperature and
poured into the mold cavity. Before the polymerization
starting the mold was closed and punch pressed with a
hydraulic press. After thirty minutes of polymerization the
mold was opened and the implant was extracted.
Figure 4 - Staircase effect from CT scan
2.2. Evaluation
In order to verify the influence of the digital model All of the measurements were carried in a MMC
quality on the localization results, three different models of (Sheffield, Discovery D8) with spherical contact probe of
a hemisphere were built, as presented in the Figure 3. The 2mm and uncertainty of 5 + L/200 (mm). This kind of
spherical form was chosen due its simplicity and existence machine is numerically controlled and its software
of an ideal (analytical) model for further comparisons. (PCDIMS CAD++) offers resources for rough localization
The real hemisphere, Figure 3a, was the stainless steel through the “3-2-1 principle” and refined, which is called as
ball with 19,05mm in diameter used for CMM probe “Best-Fit”.
calibration. The first CAD model (Figure 3b) was created in Following the hemisphere evaluation, the cast implant
IGES (Initial Graphics Exchange Specification). The second was measured with the same procedure. The implant was set
(Figure 3c) was built from 10 images of the circumferences in CMM table from its bottom planar surface. A rough
resulting from the intersection of sphere with parallel planes alignment was made based on the “3-2-1 principle” followed
equally spaced. The images were created inside a squared by the refinement after scanning the whole surface with 164
field of view with a pixel/millimeter ratio below 0.04 mm. points.
After the exportation of images bitmaps, their borders were
segmented and the correspondent surface reconstructed in a 3. RESULTS AND DISCUSSION
STL representation using specialized software (Mimics®).
The machined gypsum mold presented an excellent
surface finish and a detailing level similar to the digital
Flat area model. The polymerization and withdrawing were done
without problems. The cast PMMA shows a good match
between the external surface and its corresponding one at
the skull.
The results obtained after compare the steel
hemisphere with three different digital models (see figure 3
for details) allows to affirm that exists a considerable
(a) (b) (c) (d) influence of the quality model on the total deviation (radial
deviation). The model with a flat area at the hemisphere top
Figure 3 – (a) calibration standard sphere; (b) model 1; (c) model 2; (d)
model 3 with deformation details. produced the greatest deviation values, independently of the
applied localization method, as shown in tables 1 and 2.
As the presence of flat areas is a consequence of a
The digital model generated by Bazan [4] presented a normal CT reconstruction which contributes to lower the
flat area at the top of the outer surface, resulting from the model quality, was decided to eliminate it in order to reduce
intersection of the last CT slice with the scanned surface. A the localization error. Accordingly, in the implant
schematic representation of this failure is shown in Figure 4, measuring, the touched points were acquired outside the flat
which is related to the “staircase effect” whose magnitude is area.
dependent on the slice thickness and surface curvature. The The Figure 5 shows the distribution of the measured
hemisphere model (figure 3d) was built in order to evaluate points on the implant surface. It was observed that the points
the effect of this deformation on the localization quality. It were distributed between 0,298 (MX) and -0,243 (MI),
was obtained by lofting 10 circumferences, similar to those resulting in a total deviation (DV) equal to 0,541 mm.
used in figure 3c, with another specialized CAD system It was also observed a concentration of positive values
(Solidworks®). This lofting process results in a spherical at the implant extremities and a more concentration of
surface plus a planar one corresponding to the intersection negative differences at the implant central area. This type of
of the 10th plane. pattern could be influenced by the localization error other
than by manufacturing ones.
 The traditional 3-2-1 principle, what is widely applied Table 1– Sphere center and radial deviation in models measured, after
principle 3-2-1.
in the localization of prismatic parts, reveals to be inefficient
Radial deviation
to localize the inspected surface when the results of tables 1 Model X error Y error Z error
(mean values)
and 2 are compared and should be consider as a rough 1 0,042 0,011 -0,403 -0,778
localization. In order to minimize the localization error it’s 2 0,087 -0,077 -0,084 0,663
necessary to enlarge the number of measured points and 3 0,144 -0,136 -0,034 1,024
carry out successive refinements. The procedure adopted
here was that provided by PC-DMIS package, which is Table 2– Sphere and radial deviation in the models measured after Best-Fit.
highly dependent of quantity of measured points and also of Radial deviation
Model X error Y error Z error
the number of iterations, i. e., the subsequent localization (mean values)
refines the preceding one. 1 -0,004 0,030 0,002 0,003
2 0,006 -0,014 -0,214 -0,402
3 0,047 -0,041 -0,345 -0,860

Figure 5 - Measurement results: (left) cast PMMA implant, (center) distribution of the measured points on STL model, (right) errors indications.

investigation conducted here it was possible to observe that

4. CONCLUSIONS the quality of digital model influence the localization results.
The flat areas, which are resulting from tomography
The development of specialized information systems
process, should not be scanned for localization purpose,
for the acquisition and processing of medical images has
since they produce a considerable noise in the localization
become possible the creation of geometrical models of
procedures. The better results were achieved without
humans organs. Thanks to these new technologies
considering this kind of plateau and carry on a refined
cranioplasty implants can be prefabricated for repair large
localization with a large number of measured points in an
defects, contributing to a shorter surgical time and a better
iterative process.
cosmetic result.
Despite of a better localization achieved with
Unfortunately, from the geometrical point of view,
successive refinements, the STL model is an approximation
almost types of craniofacial implants are compound of very
of the original dried skull and its planar faces also impacts
complex surfaces. This high degree of complexity limits the
on the global error. Therefore, a separation of the
available measuring methods which can be applied in their
manufacturing error from that induced by localization
inspection. The use of vernier calipers as sometimes
procedures was not possible.
reported in the current literature [4,8,13] should be
considered as a roughly assessment of real difference among
the manufactured implant and original model.
5. ACKNOWLEDGMENTS
The main conclusion of this work points to the choice
of a more reliable technology, such those found in
We thanks the UFPR/FUNPAR and CAPES for their
Coordinated Measuring Machines (CMM) equipped with
financial support.
dedicated inspection software.
However, despite of their ability in conducting
automated inspection of three-dimensional parts, there is a
fundamental problem, called localization, concerning how to
establish a relationship between the DCS and MCS. In the
6. REFERENCES [12] Heissler E., Fischer F. S., Bolouri, S., Lehmann, T.,
Mathar, W., Gebhardt, A., Lanksch, W., Bier, J.,
Custom-made cast titanium implants produced with
[1] Eufinger H., Saylor B., Computer-assisted Prefabrication
CAD/CAM for the reconstruction of cranium defects,
of Individual Craniofacial Implants, AORN Journal, v.
International Journal Oral Maxillofacial Surgery, 27, p.
74, n. 5, nov. 2001.
334-338, 1998.
[2] Hieu L.C., Bohez E., Vander Sloten J., Phien H.N.,
[13] Bouyssié J. F., Bouyssié S., Sharrock P., Duran D.,
Vatcharaporn E., Binh P.H., An P.V.,To N.C. e Oris P.,
Stereolithographic models derived from X-ray
Design and manufacturing of personalized implants and
computed tomography - Reproduction accuracy,
standardized templates for cranioplasty applications,
Surgery Radiologic Anatomy, Vol. 19, p. 193 – 199,
IEEE ICIT’02, Bangkok, THAILAND, p. 1025-1030,
1997.
2002.
[14] Li Y., Gu P., Free-form surface inspection techniques
[3] Schiller C., Rasche C., Wehmöller, Beckmann F.,
state of the art review, Computer-Aided Design, Vol.
Eufinger H., Epple M. e Weihe S., Geometrically
36, pp. 1395–1417, 2004.
structured implants for cranial reconstruction made of
biodegradable polyesters and calcium phosphate/calcium
[15] Lin Y. P, Wang C. T., Dai K. R., Reverse engineering
carbonate, Biomaterials, 25, p. 1239–1247, 2004.
in CAD model reconstruction of customized artificial
joint, Medical Engineering & Physics, Vol. 27, pp.
[4] Bazan O., Machining of the cranioplasty prosthesis from
189–193, 2005.
tomographic images, Master in Engineering Dissertation,
Mechanical Engineering Faculty, Federal University of
[16] Spitz S.N. Dimensional inspection planning for
Paraná, Curitiba, Brazil, 2004 [in Portuguese].
coordinate measuring machines, PhD Dissertation,
University of Southern California; 1999.
[5] Leong K. F., Cheah C. M., Chua C. K., Solid freeform
fabrication of three-dimensional scaffolds for
[17] Lai J. Y., Chen K. J., Localization of parts with
engineering replacement tissues and organs,
irregular shape for CMM inspection, Int J Adv Manuf
Biomaterials, 24, p. 2363–2378, 2003.
Technol, Vol. 32, pp. 1188-1200, 2007.
[6] Dietmar W., Sittinger M., Risbud M. V., Scaffold-based
[18] Da Costa, D. ; Pedrini, H. ; Bazan, Prosthesis
tissue engineering: rationale for computer-aided design
machining from tomographic images. In: Congresso
and solid free-form fabrication systems, Trends in
Brasileiro de Engenharia de Fabricação, 2005,
Biotechnology, v. 22, n. 7, p. 354-362, Jul. 2004.
Joinville. Anais do 3º COBEF, 2005. [in Portuguese].
[7] Kulkarni P., Marsan A., Dutta D., A review of process
planning techniques in layered manufacturing, Rapid
Prototyping Journal, Vol. 6, N. 1, p. 18-35, 2000.

[8] Choi J.-Y., Choi J.-H., Kim N.-K., Kim Y., Lee J.-K.,
Kim M.-K., Lee J.-H., Kim M.-J., Analysis of errors in
medical rapid prototyping models. Int. Jornual Oral
Maxillofacial Surgery 2002; 31: 23–32.

[9] Carr J. C., Fright W. R., Beatson R. K., Surface

Interpolation with Radial Basis Functions for Medical
Imaging, IEEE Transactions on Medical Imaging, v. 16,
n.1, p.96-107, feb. 1997.

[10] Joffe J. M., Aghabcigi B., Davies E. H., Harris M. A

retrospective study of 66 titanium cranioplasties,British
Journal of Oral and Maxillofacial surgery, vol. 31, p.
144-148, 1993.

[11] Sherburn E. W., Silbergeld D. L., A new method of

acrylic cranioplasty, Surgery Neurology, Vol. 46, pp.
292-294, 1996.