Int. J. Oral Maxillofac. Surg.

2013; 42: 1159–1166

http://dx.doi.org/10.1016/j.ijom.2013.02.006, available online at http://www.sciencedirect.com

Clinical Paper
Research and Emerging Technologies – Imaging

The accuracy of three- E. Varga Jr.a, B. Hammerb,

B. M. Hardyc, L. Kamera
a
b
AO Research Institute, Davos, Switzerland;

dimensional model generation. CFC Hirslanden Cranio Facial Centre,

Hirslanden Medical Centre, Aarau,
Switzerland; cUniversity of Newcastle, New
South Wales, Australia

What makes it accurate to be

used for surgical planning?
E. Varga Jr., B. Hammer, B. M. Hardy, L. Kamer: The accuracy of three-dimensional
model generation. What makes it accurate to be used for surgical planning? Int. J.
Oral Maxillofac. Surg. 2013; 42: 1159–1166. # 2013 International Association of
Oral and Maxillofacial Surgeons. Published by Elsevier Ltd. All rights reserved.

Abstract. Computerized treatment planning is routinely used in cranio-maxillofacial

applications. For these therapeutic applications, it is of critical importance to have a
precise model of the structures in question. Improved imaging techniques and
advances in software engineering have moved three-dimensional (3D) computer
models from the research and development area into routine clinical application.
The importance of high-resolution source imaging is well understood by surgeons.
The influence of image processing is poorly understood in the surgical community
and we hypothesize that this may be a source of significant error. We evaluated the
workflow for creating a virtual model using computed tomography data, and the
impact that image processing decisions have on final virtual model accuracy. We
chose to create a model of the dental surface since it is one of the most complex
structures in the area. Individual image processing steps are explained and the
magnitudes of their influence on model quality are demonstrated and compared.
Keywords: 3D computer model; accuracy; ima-
This study demonstrates that inappropriate image processing can introduce errors of
ging; imaging modalities; imaging procedures;
similar magnitude as the use of inadequate source data. Finally, the study shows that dental occlusion; computer tomography.
errors caused by inappropriate image processing amplify the inaccuracies of low-
resolution source imagery and eliminate the benefits of high-resolution source Accepted for publication 13 February 2013
imaging. Available online 21 March 2013

Computerized treatment planning is inter-occlusal relationship between upper could potentially replace classical plaster
routinely used in dental and in cranio- and lower jaw.1–5 models, allowing new options for occlusal
maxillofacial applications. For these Advanced imaging techniques, soft- analysis and treatment workflows.9–12 The
therapeutic applications, it is of critical ware, and computerized manufacturing generation of a virtual model requires
importance to have a precise model of techniques have made three-dimensional imaging and subsequent image processing
the structures in question. This is espe- (3D) computer models available not only steps. Image capture is performed by
cially important with regard to the facial for research and development, but also for direct scanning of the patient’s occlusion
bones, the occlusal surfaces, and to the routine clinical applications.6–8 They or scanning of a plaster cast using either

0901-5027/0901159 + 08 $36.00/0 # 2013 International Association of Oral and Maxillofacial Surgeons. Published by Elsevier Ltd. All rights reserved.
1160 Varga et al.

tomographic imaging modalities or optical high-resolution CBCT imaging and on varying image resolutions ranging from
surface scanning. The image data is then image processing parameters relevant to 0.4 mm to 0.8 mm in the z-axis or axial
computed to produce a 3D visualization. creating an accurate virtual model of the direction.
Computed tomography (CT) and cone occlusal surface. Duplication and down-sampling of the
beam computed tomography (CBCT) ima- models was used to simulate the lower
ging capture a consecutive series of two- resolution imaging capture, rather than
Materials and methods
dimensional (2D) images containing digi- scanning at lower resolution. Simulation
tized tissue densities of the scanned slice. Image acquisition eliminated the introduction of error due to
The data stack is stored as single image malpositioning of the real-world model
A standard dental plaster model was
slices in DICOM (Digital Imaging and during repeated scanning at different reso-
scanned using an experimental high-reso-
Communication in Medicine) format.13–17 lutions. This standardization meant that all
lution micro-CBCT scanner (SCANCO
Optical surface scanning captures 3D deviations from the gold standard could
Medical AG, Brüttisellen, Switzerland)
surface information via the generation of reliably be attributed to image processing
with an isotropic image resolution of
a point cloud rather than the interpolation steps, rather than errors in acquisition.
0.082 mm using standard CBCT image
of tissue densities. These points are extra- From each 2D image stack, an
acquisition parameters (60 kVp/40 keV
polated to a 3D surface mesh to form the unsmoothed model was generated using
(900 mA)). The DICOM information was
shape of an object.18 In contrast to CT or the LegoSurfaceGen procedure within
transferred to a desktop computer running
CBCT, optical scanning only captures Amira. This algorithm produces a surface
Windows XP (Microsoft Corporation, Red-
surface contours, without any subsurface mesh exactly identical to the voxels, with-
mond, CA, USA) and post-processed in
detail. out any automatic smoothing or change
Amira, a commercial software package
CT and CBCT imaging produces volu- involved. For each model an identical
for image visualization and data analysis
metric data that subsequently needs to be threshold window was chosen by an expert
(Visage Imaging GmbH, Berlin, Ger-
turned into a surface mesh. In both X-ray and the Amira LegoSurfaceGen applied.
many), and Geomagic Studio, a reverse
based imaging and optical surface scan- Thresholding defines which voxel is in the
engineering software package (Geomagic
ning, the final result is a 3D surface mesh region of interest on the grey scale. The
Corp., Research Triangle Park, NC, USA).
of the imaged structure. The surface end result of this procedure was eight
meshes can then be merged together to virtual models of the same physical object
create a composite skull model.19–23 This Image processing and analysis generated from the source image data, of
technique relies on both the accuracy of varying resolution. Each model was then
the models and the precision of the incor- The workflow for creating a virtual model compared with the GS1 model using the
poration procedures. was evaluated using the CBCT data and distance map calculation and visualization
When considering a virtual model, the divided into six major image processing procedure in Geomagic. The distance map
clinician must be aware of the need for steps. The first three steps, evaluating the procedure calculated the differences
adequate source imaging and the image impact of the image resolution (step 1), between the GS1 and the down-sampled
processing steps required to create the final thresholding procedure (step 2), and models. Distance mapping allowed calcu-
model. Some of these steps are performed smoothing of the labelled threshold data lation of maximum and mean differences
by user interaction, while some are ‘hidden’ (step 3), are performed on the 2D CT data between the two models (Fig. 1A and B).
procedures within the software. The impact before surface model generation, as pre- The down-sampled models were used in
of the image processing steps may subtly sented in Fig. 1. Then a 3D triangulated the further steps to compare each model of
influence the accuracy of the final model. surface model is generated to evaluate the the procedures with the models of its own
The generation of a virtual model from effect of different model generation tech- identical resolution. This enabled the
either 2D or 3D data is based on well- niques (step 4), smoothing the surface assessment of the impact of the further
established imaging and image processing model (step 5), and reducing the number procedures without the influence of image
techniques. Surgical planning requires of triangles of the surfaces (step 6) shown resolution. The down-sampled models
models of sufficient detail to enable the in Fig. 2. were termed ‘corresponding gold stan-
virtual plan to perform well when put into dards’ (CGS).
reality. Inadequate models may result in Every model used in the steps was made
Step 1—image resolution
misleading patient assessments, inap- from copies from the CGS in order to have
propriate treatment, and unsatisfactory A gold standard for step 1 (GS1) CT scan them in the exact same position in the
clinical outcomes. Conversely, accurate was captured with a 0.082 mm resolution. virtual space. The imaging procedures
virtual occlusion modelling may enable To simulate lower resolution CT acquisi- were applied to the models in the identical
more efficient and effective workflows tion, the gold standard image data stack position, thus avoiding the need for model
for patient assessment and treatment. was duplicated and down-sampled to a alignment, which could lead to false
The objective of this study was to show variety of resolutions of interest. A total deviations.
the importance and evaluate the impact of of eight down-sampling procedures were
different imaging modalities and image used to produce the following different
Step 2—threshold segmentation
processing steps on the accuracy of the image resolutions: four copies were
resulting virtual model, and to identify and down-sampled in an isotropic manner to This procedure defines the borderlines of
quantify the most critical steps to ensure represent an image resolution of 0.1– our models. It converts the continuous
accurate model generation. Further goals 0.4 mm. A further four datasets were greyscale 2D CT data into a binary repre-
were to educate clinicians of the implica- down-sampled to represent non-isotropic sentation of the pixels of CT data that
tions of these computational procedures scanning, all having the same image reso- represent the surface of the model being
and to discuss the clinical impact of lution of 0.4 mm in the x and y axes generated. In most cases it is difficult to
the results. We report on the use of representing the transverse plane, but decide which voxels belong in the grey
 The accuracy of three-dimensional model generation. What makes it accurate to be used for surgical planning? 1161

Fig. 1. Image processing steps performed on 2D CT data. The first line shows the impact of image resolution (A: 0.1 mm voxel size; B: 0.8 mm
non-isotropic). The second line highlights the threshold segmentation (C: high, exclusive settings; D: low, inclusive settings; resolution: 0.1 mm).
The third line demonstrates the smoothing procedure (E: without smoothing; F: smoothed border line; resolution: 0.3 mm).

values, whether they should be inside or Step 3—smoothing 2D CT data Gen procedure creates meshed surface
outside the borderline. To mimic this pro- models according to the marching cubes
A predefined smoothing procedure was
blem, two different threshold values were algorithm, smoothing the voxel structure
applied to the 2D CT data using the
defined, representing two different mesh into a triangulated meshed surface topol-
‘smooth label’ function (parameters: size
border alternatives: a low (inclusive) and a ogy. The deviations between the Surface-
5) in the Amira segmentation editor. A
high (exclusive) threshold value. For each Gen model and the identical resolution
LegoSurfaceGen model was created from
of the varying resolution CT scans gener- LegoSurfaceGen models obtained in step
each of the smoothed 2D CT datasets. The
ated in step 1, two additional models were 1 were compared using the Geomagic
deviations between the models were com-
generated using a high and a low threshold distance map calculation and visualization
pared with the models of corresponding
and the LegoSurfaceGen procedure. (Fig. 2A and B).
resolution obtained in step 1, using the
For each resolution, the high and low
Geomagic distance map calculation and
threshold models were separately com-
visualization (Fig. 1E and F). Step 5—smoothing of generated 3D
pared with the medium threshold model
models
of the corresponding resolution generated
in step 1 (CGS). This enabled the assess- Two different smoothing settings were
Step 4—3D model generation using the
ment of the impact of the threshold pro- chosen in Geomagic for each model
marching cubes algorithm
cedure without the influence of image obtained in step 4: setting 1, correspond-
resolution. The deviations between the For each of the resolutions established in ing to moderate smoothing (‘smoothness
models were calculated using the Geoma- step 1, a model was created using the level’ to a low level ‘3’ and ‘curvature
gic distance map calculation and visuali- Amira standard SurfaceGen module, priority’ to a high level ‘8’), and setting 2,
zation (Fig. 1C and D). rather than LegoSurfaceGen. The Surface- corresponding to extensive smoothing
1162 Varga et al.

Fig. 2. The effects of image processing steps made on 3D data. The first line shows the effect of 3D model generation from 2D data with different
algorithms (A: LegoSurfaceGen; B: marching cubes algorithm; resolution: 0.4 mm). Effects of smoothing the 3D surface are visualized in the
second line (C: moderate smoothing; D: extensive smoothing; resolution: 0.3 mm). The third line presents the result of reducing the number of
surface triangles (E: 41,000 triangles; F: 20,500 triangles; resolution 0.4 mm).

(‘smoothness level’ to a high level ‘8’ and compared with the identical resolution image resolution, the maximum positive/
‘curvature priority’ to a low level ‘3’). For model obtained in step 4. The differences maximum negative values showed similar
each resolution, the models with moderate between models were analyzed using the absolute deviation values. There was a
and extensive smoothing were separately Geomagic distance map calculation and four-fold increase in the deviation
compared with the original unsmoothed visualization (Fig. 2E and F). between the highest and lowest resolution,
model of identical resolution generated in with maximum absolute values of about
step 4. The differences between models 0.45 mm (corresponding to +0.438/
Results
were analyzed using the Geomagic dis- 0.492 mm). There was a modest increase
tance map calculation and visualization The impact of CT scan resolution and the in the average values from 0.021 mm
(Fig. 2C and D). five subsequent workflow steps was reaching 0.127 mm. The distance maps
assessed (Fig. 3). demonstrated a uniform deviation pattern
on the occlusal surfaces when compared
Step 6—reducing the number of mesh
with the GS (Fig. 3, step 1).
triangles Step 1—resolution
In the eight surface models generated in Beginning with the highest resolution, the
Step 2a—threshold segmentation with
step 4, the number of mesh triangles was data revealed a linear increase in the max-
low threshold value
calculated in Geomagic. The number of imum deviations between the models gen-
triangles for each model was reduced by erated from reduced resolution CT scans The data showed that the maximum devia-
50% to represent a common complexity and the gold standard model. In the non- tion in the positive direction from the CGS
reduction procedure performed in model isotropic image resolution part (i.e. from increased as CT resolution decreased in
generation workflows. For each resolu- 0.5 to 0.8 mm) there was a less steep slope models with isotropic voxels, and did
tion, the resulting simplified model was compared to the isotropic part. For each not change in models with non-isotropic
 The accuracy of three-dimensional model generation. What makes it accurate to be used for surgical planning? 1163

Fig. 3. Deviations visualized at each step with their corresponding resolutions; resolution in x-axis (mm), deviation in y-axis (mm).
1164 Varga et al.

Fig. 4. Distance maps showing the different surface topology modifications. Left: uniform deviation pattern on the occlusal surface (Step 2a).
Right: overall flattening of the surface with filling of the fissures and cutting of the cusps (Step 3).

voxels. The maximum positive and max- of areas with small-radius curvatures. The maximum positive deviation reached up
imum negative values showed different loss of detail in the area of tight curvature to 1.6 mm. The average deviation also
absolute values. The maximum negative led to a flattening of the occlusal cusps approached 0.486 mm.
values showed an approximately linear (thus reducing the height of the occlusal
increase up to 0.8 mm, regardless of cusps) and of the occlusal fissures (thus
Step 6—reducing the number of triangles
model isotropy. The maximum positive filling the occlusal fissures), leading to an
values showed a similar increase from overall flattening of the occlusal surface Reducing the triangle number by 50% did
0.1 to 0.4 mm, but afterwards the values topology as shown in Fig. 1(E and F) and not change the maximum positive devia-
remained virtually unchanged between the on the distance maps in Fig. 4. tions. Below a CT resolution of 0.3 mm,
0.4, 0.5, 0.6, 0.7, and 0.8 mm resolution the reduction in model complexity led to
CT scans. There was virtually no change deviations of between 0.1 and 0.2 mm.
Step 4—model generation using the
in the average positive values, and a slight The average deviation showed no
marching cubes algorithm
decreasing was shown in average negative change.
values from 0.037 mm to 0.086 mm. For both the maximum positive and max-
imum negative values and for both the
Discussion
average positive and average negative
Step 2b—threshold segmentation with
values, there was a linear increase in The objective of this study was to assess the
high threshold value
deviation from the CGS until 0.4 mm impact of the different imaging modalities
The data illustrated maximum deviations image resolution. The rate of deviation and different image processing steps on the
from the CGS in the positive directions, accelerated at resolutions lower than accuracy of a virtual occlusion model and
increasing linearly as CT resolution 0.4 mm. Large maximum deviations were to guide clinicians in understanding the
decreased. The maximum negative devia- observed for values at a non-isotropic most critical steps for accurate model gen-
tion increased with decreasing resolution image resolution of 0.7 and 0.8 mm, eration. The following image processing
in the models with isotropic voxels, but deviating >1.6 mm from the CGS. The steps were assessed: image resolution,
did not change significantly with the non- marching cubes algorithm resulted in a manual threshold segmentation, smoothing
isotropic voxels. This is the inverse of the similar non-uniform change in the surface the labelled threshold data, the marching
results described in section 2a. The max- topology, as observed in step 3. cubes algorithm,24,25 smoothing of the sur-
imum positive values exhibited a linear face mesh,26,27 and the number of
increase up to 0.8 mm and the maximum meshes.28–30 Intuitively, resolution is the
Step 5a—smoothing of generated
negative values remained stable at resolu- essence of an accurate model. No image
surface models with moderate settings
tions above 0.4 mm. The average negative processing step can compensate for a lack
value showed no change and the average The data demonstrated linear growth in of resolution in the original data. Inade-
positive value was slightly increased up to deviation until the 0.4 mm resolution quate resolution amplifies the negative
0.078 mm. group, and then ceased to change. This effects of image processing. However,
In both assessments 2a and 2b, the procedure did not change the maximum or image processing can severely degrade
distance maps showed a uniform deviation average values. Maximum deviations the quality of a model generated from
pattern within the overall occlusal surface approached 0.12 mm, average values high-resolution data. The evaluations show
when compared with the CGS. reached 0.012 mm. that each processing step affects the quality
and accuracy of the 3D model. Each step is
dependent on the image processing techni-
Step 3—smoothing 2D CT data Step 5b—smoothing of generated
ques chosen in previous steps. Errors intro-
surface models with extensive settings
The data displayed a linear increase in duced early in the workflow are multiplied
model deviation as CT resolution was The data displayed growth in deviation. In by subsequent image processing steps. Par-
reduced. In contrast to assessment of steps contrast to 5a, when using these settings ticularly, early decisions made during
1, 2a, and 2b, this smoothing procedure led the maximum positive value increased manipulation of label field data in 2D scans
to a characteristic change in the surface with lower image resolution. At the non- have a dramatic effect on the accuracy of
topology of the occlusal surface with loss isotropic level of 0.7 and 0.8 mm, the the final model.
 The accuracy of three-dimensional model generation. What makes it accurate to be used for surgical planning? 1165

The use of virtual 3D models for treat- using average resolution CT scanners as 10. Plooij JM, Maal TJ, Haers P, Borstlap WA,
ment planning and computer-aided surgery long as the importance of image proces- Kuijpers-Jagtman AM, Bergé SJ. Digital
(CAS) systems has several advantages. sing is understood. The surgeons must three-dimensional image fusion processes
Virtual treatment planning can make sur- understand the implications of each step for planning and evaluating orthodontics
gery less invasive, operating times shorter, of the model creation workflow to ensure and orthognathic surgery. A systematic
and postoperative outcomes more accurate. that their model truly reflects reality. review. Int J Oral Maxillofac Surg
This technique relies on the accuracy of the 2011;40:341–52.
virtual model to ensure a successful sur- Funding 11. Swennen GR, Mollemans W, Schutyser F.
Three-dimensional treatment planning of
gery. Models produced from high-resolu-
This research project was supported by the orthognathic surgery in the era of virtual
tion source data can appear accurate on
AOCMF Specialty of the AO Foundation imaging. J Oral Maxillofac Surg
superficial inspection, despite key inac-
(AOCMF Research Grant C-10-25Z). 2009;67:2080–92.
curacies introduced by image processing. 12. O’Neil M, Khambay B, Bowman A, Moos
Clinicians planning surgery with virtual KF, Barbenel J, Walker F, et al. Validation of
models must be familiar with each step Competing interests a new method for building a three-dimen-
of the workflow. The production of accu- sional physical model of the skull and denti-
None declared.
rate 3D models from imaging data requires tion. Br J Oral Maxillofac Surg 2012;50:49–
specific knowledge of the techniques 54.
involved. Ethical approval 13. Hounsfield GN. Computed medical imaging.
This study demonstrates that there are Science 1980;210:22–8.
two distinct classes of error introduced by Not required. 14. De Vos W, Casselman J, Swennen GR. Cone-
image processing. One class makes a uni- beam computerized tomography (CBCT)
form change to the entire occlusal surface, Acknowledgements. The authors are grate- imaging of the oral and maxillofacial region:
making it smaller or bigger. The other ful to Hansruedi Noser, Thomas Heldstab, a systematic review of the literature. Int J
alters the surface topology in a non-uni- Balázs Erdőhelyi, Balázs Bagó, and Oral Maxillofac Surg 2009;38:609–25.
form manner, particularly impacting areas Nicole Haldi for their valuable support 15. Jahne B. Digital image processing. New
of high geometric complexity (Figs. 1E York: Springer-Verlag; 1995.
for the project. Special thanks to Geoff
and F, and 4). Both effects lead to inac- 16. Pratt WK. Digital image processing. New
R. Richards for the AO fellowship.
curacies in virtual model generation. York: John Wiley & Sons; 1991.
17. Loubele M, Maes F, Schutyser F, Marchal G,
We have to note that comparing differ-
Jacobs R, Suetens P. Assessment of bone
ent models is only possible by showing the References segmentation quality of cone-beam CT ver-
deviation between the nearest neighbour- sus multislice spiral CT: a pilot study. Oral
ing points. Deviation between correspond- 1. Jacobs R. Dental cone beam CT and its
justified use in oral health care. JBR-BTR Surg Oral Med Oral Pathol Oral Radiol
ing points on both images is closer to Endod 2006;102:225–34.
2011;94:254–65.
reality, but it can only be shown on two 18. Moerenhout BA, Gelaude F, Swennen GR,
2. Ogawa T, Ogimoto T, Koyano K. Validity of
completely identical models. the examination method of occlusal contact Casselman JW, Van Der Sloten J, Mom-
A potentially satisfactory alternative to pattern relating to mandibular position. J maerts MY. Accuracy and repeatability of
3D model generation from CT data is Dent 2000;28:23–9. cone-beam computed tomography (CBCT)
optical scanning of dental structures. 3. Paquette DE. Importance of the occlusal measurements used in the determination of
The clinical use of 3D surface laser scan- plane in virtual treatment planning. J Clin facial indices in the laboratory setup. J Cra-
ners is increasing rapidly,31 but we have to Orthod 2011;45:217–21. niomaxillofac Surg 2009;37:18–23.
keep in mind that this technique does not 4. Watamoto T, Egusa H, Mizumori T, Yashiro 19. Gateno J, Xia J, Teichgraeber JF, Rosen A. A
eliminate the need for CT in many cases. If K, Takada K, Yatani H. Restoration of occlu- new technique for the creation of a compu-
detail of the underlying structures is sal and proximal contacts by a single molar terized composite skull model. J Oral Max-
needed for surgical planning, CT scanning crown improves the smoothness of the mas- illofac Surg 2003;61:222–7.
is still required, as is often the case in ticatory movement. J Dent 2008;36:984–92. 20. Noh H, Nabha W, Cho JH, Hwang HS.
orthognathic and surgical procedures 5. Smith BG. Occlusion: 1. General considera- Registration accuracy in the integration of
affecting dental structures. An additional tions. Dent Update 1991;18:141–5. laser-scanned dental images into maxillofa-
source of error within optical scanning is 6. Trefný P, Tauferová E, Bálková S. Three- cial cone-beam computed tomography
dimensional visualisation and analysis of images. Am J Orthod Dentofacial Orthop
the merging and alignment of CT-derived
post-operative changes in the size and shape 2011;140:585–91.
and optical scan-derived model data.
of the dental arch and palate. Acta Chir Plast 21. Swennen GR, Mommaerts MY, Abeloos J,
Finally, models generated from optical De Clercq C, Lamoral P, Neyt N, et al. A
2005;47:124–8.
surface scanning might still undergo cone-beam CT based technique to augment
7. Marchetti C, Bianchi A, Bassi M, Gori R,
image processing like in steps 5 and 6, the 3D virtual skull model with a detailed
Lamberti C, Sarti A. Mathematical modeling
potentially introducing the inaccuracies dental surface. Int J Oral Maxillofac Surg
and numerical simulation in maxillo-facial
detailed in this study. virtual surgery (VISU). J Craniofac Surg 2009;38:48–57.
In conclusion, accurate 3D models 2006;17:661–7. 22. Swennen GR, Barth EL, Eulzer C, Schutyser
require a combination of appropriate reso- 8. Schutyser F, Swennen G, Suetens P. Robust F. The use of a new 3D splint and double CT
lution imaging and image processing. The visualization of the dental occlusion by a scan procedure to obtain an accurate ana-
importance of resolution in the source double scan procedure. Med Image Comput tomic virtual augmented model of the skull.
imaging is well understood by surgeons. Comput Assist Interv 2005;8:368–74. Int J Oral Maxillofac Surg 2007;36:146–52.
The critical nature of image processing is 9. Favero L, Terrazzani C, Favero V, Stellini E, 23. Swennen GR, Mollemans W, De Clercq C,
poorly recognized and this study demon- Cocilovo F. Virtual study models: a compar- Abeloos J, Lamoral P, Lippens F, et al. A
strates it to be a source of significant error. ison of modular application systems. Prog cone-beam computed tomography triple
Adequate quality models can be produced Orthod 2009;10:16–25. scan procedure to obtain a three-dimensional
1166 Varga et al.

augmented virtual skull model appropriate 27. Taubin G. Curve and surface smoothing 31. Chang YB, Xia JJ, Gateno J, Xiong Z,
for orthognathic surgery planning. J Cranio- without shrinkage. Proceedings of the Fifth Teichgraeber JF, Lasky RE, et al. In vitro
fac Surg 2009;20:297–307. International Conference on Computer evaluation of new approach to digital dental
24. Cline HE, Lorensen WE. System and Vision. 1995:852. model articulation. J Oral Maxillofac Surg
method for the display of surface structures 28. Luebke DP. A developer’s survey of poly- 2012;70:952–62.
contained within the interior region of a solid gonal simplification algorithms. IEEE Com-
body. Patent Number 4710876; 1987. put Graph Appl 2001;21:24–35. Address:
25. Lorensen WE, Cline HE. Marching cubes: a 29. Garland M, Heckbert PS. Surface simplifica- Endre Varga
high resolution 3D surface construction tion using quadric error metrics. SIGGRAPH AO Research Institute
algorithm. Comput Graph 1987;21:163–9. 97 Conference Proceedings. 1997:209–16. Davos
26. Belyaev A. A comparison of mesh smooth- 30. Schroeder WJ, Zarge JA, Lorensen WE. Clavadelerstrasse 8
ing methods. Proceedings of the Israel- Decimation of triangle meshes. SIGGRAPH 7270 Davos Platz
Korea BiNational Conference on Geometric ’92 Proceedings of the 19th annual confer- Switzerland
Tel: +36 30 5020368
Modeling and Computer Graphics. ence on Computer graphics and interactive
E-mail: [email protected]
2003:83–7. techniques. 1992:65–70.