Optimization analysis for plane orientation in 3-dimensional

cephalometric analysis of serial cone-beam computerized

tomography images
Manuel O. Lagravère, DDS, PhD, FRCD(C),a Marc Secanell, PhD,b
Paul W. Major, DDS, MSc, FRCD(C),c and Jason P. Carey, PhD, PEng,dAlberta, Canada
UNIVERSITY OF ALBERTA

Objectives. The purpose of this study was to evaluate and reduce potential errors associated with superimposition of
serial cone-beam computerized tomography (CBCT) images using planes based on cranial base landmarks.
Methods. CBCTs from 10 patients were analyzed. The potential impact of errors in cranial base landmark
identification on assessment of the relative position of distant landmarks is mitigated by means of a mathematical
algorithm that ensures that the distances and angles between landmark identification points are maintained for
different images by readjusting the landmark coordinates.
Results. Significant improvement was observed after optimization. The errors found in a previous study were
significantly reduced, some by more than 90%. Errors found in the standardization were viewed in both
infraorbitals and mentons ranging from 1 to 3 mm.
Conclusions. The mathematical transformation to readjust the coordinates of ELSA, left and right auditory external
meatus (AEM), and dorsum foramen magnum (DFM) points significantly improves their use for image superimposition.
(Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2011;111:771-777)

Superimposition of anatomical structures or landmarks
in lateral cephalometric analysis has traditionally been
done for determining orthodontic treatment results and
growth. The accuracy of superimposition depends on
examiner experience and landmark location. Leonardi
et al.1 stated that landmarks automatically or manually
located and traced are not sufficiently accurate to be
used in general practice. Of the available structures for
superimposition, studies on implants and human skulls
have shown that superimposition on highly stable areas
of the cranial base is preferable.2,3
Two-dimensional (2D) imaging does not fully rep-
resent a 3D structure, because much of the information
is lost when 3D structures are converted to 2D images.4
With the appearance of cone-beam computerized to-
mography (CBCT), many of the problems that occur
when locating landmarks and evaluating changes can
a
Clinical Assistant Professor, Orthodontic Graduate Program, Faculty
of Medicine and Dentistry, University of Alberta, Alberta, Canada.
b
Assistant Professor, Mechanical Engineering, Faculty of Engineer-
ing, University of Alberta, Alberta, Canada.
c
Professor, Chair of Dentistry, Faculty of Medicine and Dentistry,
University of Alberta, Alberta, Canada.
d
Associate Professor, Mechanical Engineering, Faculty of Engineer-
ing, University of Alberta, Alberta, Canada.
Received for publication Dec 17, 2010; returned for revision Feb 11,
2011; accepted for publication Feb 11, 2011.
1079-2104/$ - see front matter
© 2011 Mosby, Inc. All rights reserved.
doi:10.1016/j.tripleo.2011.02.017
be overcome. Many studies5,6 have reported accuracy
and reliability of location of landmarks in CBCT im-
ages but few4,7-9 have reported using these landmarks
as reference planes for pre- and posttreatment image
superimposition. Some results show that this is not a
simple task.10 The purpose of this study was to evaluate
and reduce potential errors associated with superimpo-
sition of serial CBCT images using reference planes
based on cranial base landmarks.
MATERIAL AND METHODS
CBCT’s volumetric data (NewTom 3G Volumetric
Scanner, Aperio, Italy) taken at 110 kV, 6.19 mA, and
8 mm aluminum filtration from 10 patients participat-
ing in a maxillary expansion clinical trial were used for
this study. Images were taken using a 12-inch detector
field at baseline before treating the patients. Ethics
approval was obtained from the Health Research Ethics
Board at the University of Alberta.
Images were obtained and converted to DICOM
(Digital Imaging and Communications in Medicine)
format using the NewTom software (Aperio, Italy) with
a voxel size of 0.25 mm. Using AMIRA software
(AMIRA, Mercury Computer Systems, Inc., Berlin,
Germany), the DICOM-format images were rendered
into a volumetric image using 512 ⫻ 512 matrices
giving a range of 400 to 420 DICOM slices. Sagittal,
axial, and coronal volumetric slices, as well as the 3D
image reconstruction were used to determine landmark
positions.

771
 OOOOE
772 Lagravère et al. June 2011

Table I. Descriptive statistics of coordinates during both transformation procedures (mm)

First transformation Second transformation
Mean SD Minimum Maximum Mean SD Minimum Maximum
ELSA x 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
y 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
z 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
AEML x 0.997 1.323 –1.790 2.390 0.000 0.000 0.000 0.000
y –0.026 0.856 –1.190 0.940 0.000 0.000 0.000 0.000
z 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
AEMR x –0.098 1.254 –20.080 1.900 0.000 0.000 0.000 0.000
y 0.109 10.047 –1.260 2.140 0.000 0.000 0.000 0.000
z 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
DFM x 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
y 0.665 1.322 –0.530 3.850 0.000 0.000 0.000 0.000
z 0.681 1.731 –0.820 50.050 0.000 0.000 0.000 0.000
InfraOL x 0.061 10.071 –1.850 1.940 0.361 1.216 –1.240 2.480
y –0.033 0.766 –1.130 1.100 –0.170 0.860 –1.740 10.090
z –1.281 2.841 –8.330 1.650 –0.693 1.442 –3.360 10.010
InfraOR x –0.184 0.898 –20.000 1.490 0.128 10.010 –1.440 20.040
y 0.250 0.624 –0.570 1.490 0.343 0.613 –0.640 1.180
z –1.159 2.705 –7.940 1.610 –0.505 1.342 –2.530 0.890
MnL x 0.450 10.088 –1.170 1.790 0.868 1.575 –1.760 2.910
y 10.026 2.577 –1.380 7.580 0.291 1.381 –10.050 2.720
z –1.195 3.144 –8.980 20.080 –0.531 1.569 –3.170 1.460
MnR x 0.172 0.985 –1.250 1.550 0.595 1.387 –1.960 2.120
y 0.983 2.693 –10.050 80.060 0.483 1.174 –0.560 2.960
z –10.039 2.945 –8.410 1.610 –0.300 1.533 –2.460 1.770
AEML, auditory external meatus left; AEMR, auditory external meatus right; DFM, dorsum foramen magnum; ELSA, midpoint between both
foramen spinosums; InfraOL, infraorbitale left; InfraOR, infraorbitale right; MnL, menton left; MnR, menton right.

Four landmarks are required to define a 3D anatom-
ical reference coordinate system. The left and right
auditory external meatus (AEM) and dorsum foramen
magnum (DFM) were selected based on position and
early formation in skeletal growth. It has been reported
that most of the cranial base growth (⬎85%) occurs in
the first 5 years of life, presenting minor changes after
this age. The fourth point, ELSA, is defined in a pre-
vious publication as the midpoint between the left and
right foramen spinosum.11 ELSA was selected as the
origin of the new Cartesian coordinate system. From
the origin, 3D positional coordinates for AEM left,
AEM right, and DFM were determined. Intrareliability
values were determined using intraclass correlation co-
efficient for all 4 landmarks, repeating the process 3
times for each image. Intrareliability values for each
Cartesian coordinate for all landmarks were reported in
Lagravère et al.,6 showing the lowest value of 0.965 in
the x-axis for AEM left (AEML).
Landmarks used in the present study are defined in
Table I. The principal investigator located the land-
marks on each image 3 times. Spherical markers of
0.5-mm diameter were placed indicating the position of
the landmark, and the software used the center of these
spherical markers as coordinates. For purposes of stan-
dardization of the coordinate axial system, it was es-
tablished that both AEM right (AEMR) and AEML lie
in the XY plane and thus have a zero Z coordinate,
whereas DFM lies in the YZ plane and thus has a zero
X coordinate.
First coordinate transformation procedure
A detailed description of the transformation process
is given in Lagravère et al.10 Briefly, to transform all
global landmark coordinates to an ELSA coordinate
system, the vector describing the position of ELSA in
the AMIRA coordinate system was subtracted from all
anatomical landmarks, thus making ELSA the (0,0,0)
origin of the reference Cartesian coordinate system.
This is illustrated in Fig. 1.
Coordinate system transformations were performed
in 2 steps. The coordinate system was constructed using
2 planes defined by anatomical landmarks. The first
plane, which represents a new XY plane, was defined
using both AEM and ELSA; the 2 vectors were taken
with respect to the ELSA coordinate system. The sec-
ond plane, defined as a new YZ plane, was formed by
ELSA and DFM perpendicular to the new XY plane.
In this set of transformations, left and right AEM
coordinates had zero z-component, as they lie in the
X⬘⬘Y ⬘⬘ plane. The second transformation is a rotation of
the X⬘⬘Y ⬘⬘plane defined in the above steps to set the
OOOOE
Volume 111, Number 6 Lagravère et al. 773

Fig. 1. Coordinate system transformation proposal. ELSA, AEMR, AEML, and DFM are lying outside of the coordinate system;
thus, a transformation is performed to make ELSA the (0,0,0) center of the coordinate system and the other 3 points lie on their
respective 0 axis.

Y ⬘⬘Z⬘⬘plane in which lies the DFM anatomical landmark. ELSA in images 1 and 2 have the same relative distance
After this transformation, the DFM had zero X-compo- between each other.
nent. To obtain the error correction for AEML, AEMR,
Using the initial coordinate transformation, values and DFM in image 2, we solved 9 equations:
were obtained for the new infraorbital (right and left)
㛳xAEML
1
㛳 ⫺ 㛳xAEML
2
㛳⫽0
and menton (right and left) landmarks. Images were
measured 3 times and thus it was expected that an error 㛳xAEMR
1
㛳 ⫺ 㛳xAEMR
2
㛳⫽0
close to zero would be obtained. After the analysis was 㛳xDFM
1
㛳 ⫺ 㛳xDFM
2
㛳⫽0
completed, a second transformation was performed on
㛳xAEML ⫺ xAEMR
1
㛳 ⫺ 㛳xAEML ⫺ xAEMR
2
㛳⫽0
the 4 initial points.
㛳xAEML ⫺ xDFM
1
㛳 ⫺ 㛳xAEML ⫺ xDFM
2
㛳⫽0 (1)
Second coordinate transformation procedure 㛳xAEMR ⫺ x 1
DFM 㛳 ⫺ 㛳xAEMR ⫺ x 2
DFM 㛳⫽0
involving the 4 initial landmarks
The relative position of the AEML, AEMR, and 共x · x 兲 ⫺ 共x · x 兲⫽0
1
AEML
1
AEMR
2
AEML
2
AEMR

DFM landmarks with respect to ELSA should be the 共xAEML

1 1
· xDFM 兲 ⫺ 共xAEML
2 2
· xDFM 兲⫽0
same between 2 images. However, because of operator 共xAEMR · xDFM兲 ⫺ 共xAEMR · xDFM兲⫽0
1 1 2 2

errors, small differences in the coordinates are observed

between images. To correct for the measurement errors,
an algorithm was developed to ensure that the distances
and angles between AEML, AEMR, DFM, and ELSA
are maintained. A system of equations is solved to find
the necessary perturbation so as to make all distances
and angles the same. The system of equations solved is
as follows.
The algorithm receives the coordinates of AEML,
AEMR, DFM, and ELSA for 2 different images, i.e.,
image 1 and image 2. Given that operator error usually
exists in recording these values, we would like to
guarantee that the relative angle and distances between
the 2 images are conserved. To achieve this, we ob-
tained an error correction vector, ␦ui, for every point in
image 2 such that the points AEML, AEMR, DFM, and
where㛳 㛳 represents the Euclidean distance, ⫻ repre-
1
sents the dot product, xAEML are the coordinates of
2
AEML in image 1 and xAEML are the corrected coordi-
nates of AEML in image 2, i.e.,
2
xAEML ⬅ xAEML,original
2
⫹ ␦uAEML (2)
where ␦uAEML is the unknown error correction vector.
The system has 9 unknowns, corresponding to the
measurement error corrections for AEML, DFM, and
AEMR and 9 equations; therefore, the system can be
solved using any nonlinear solver. In this case, MAT-
LAB’s (MathWorks, Natick, MA, USA) fsolve routine
is used.
Note that the first 3 equations guarantee that the
distance between ELSA and AEML, AEMR and DFM,
 OOOOE
774 Lagravère et al. June 2011

Fig. 2. Illustration of the correction of the landmark points between images. The 3 points with different colors represent the
following: original landmarks in image 1 (blue), original landmarks in image 2 (red), and corrected landmarks in image 2 (green).
ELSA is not modified between images; therefore, superposition is observed at point (0,0,0) (mm).

respectively, are the same for the 2 images. The other 3
equations guarantee that the relative distances between
AEML and AEMR, AEML and DFM, and AEMR and
DFM are the same between images 1 and 2, respec-
tively. Finally, the last 3 equations guarantee that all
angles are the same.
To verify the effect of optimizing the distances and
angles of AEML, AEMR, and DFM with respect to
ELSA, landmarks from 2 images were input to the
algorithm. Fig. 2 shows the landmarks on image 1, the
original landmarks on image 2, and the corrected land-
marks after both transformations.
Once the new landmark coordinates are established,
the first coordinate transformation procedure is per-
formed to have a standardization of the global axis. To
verify the changes, 4 points outside of the coordinate
system formation were located. These points were in-
fraorbital left and right (InfraOL and InfraOR) and
menton left and right (ML and MR). Descriptive sta-
tistics on all new coordinates and pairwise comparisons
among the 3 measurement trials were made.
RESULTS
Once the raw data were transformed to a standard-
ized ELSA coordinate system using ELSA as refer-
ence (0,0,0), both AEM as (X, Y, 0) and DFM as (0,
Y, Z), the new values were used to determine the
mean measurement differences for the 3 trials for
each axis of every landmark and then averaged. The
equation used is
关ⱍ M2 ⫺ M1ⱍ ⫹ ⱍ M3 ⫺ M2ⱍ ⫹ ⱍ M3 ⫺ M1ⱍ兴
M⫽
3
(3)
where M1, M2, and M3 are each trial’s measurement on
an image and M is the mean measurement difference.
This calculation was done for each coordinate axis and
each image.
Values obtained for each transformation can be
viewed in Table I. It should be noted that after the first
transformation, mean differences are very large, in
some cases approximately 1 cm. Following the second
OOOOE
Volume 111, Number 6 Lagravère et al. 775

transformation, these values have improved greatly giv- Table II. Pairwise comparisons of means from 3 trials
ing mean differences of less than 1 mm. In the worst of after both transformations
cases, the range for the second transformation was Pairwise comparisons
approximately 4 mm. The standard deviations have also Mean difference Standard error
been reduced with respect to the points obtained only ELSA x1 x2 0.000 0.000
after the first transformation. x3 0.000 0.000
A pairwise comparison of the 3 measurement differ- x2 x3 0.000 0.000
ences obtained from each trial measurement after both y1 y2 0.000 0.000
transformations was also done. Results are presented in y3 0.000 0.000
y2 y3 0.000 0.000
Table II. Table II shows there is no significant variation z1 z2 0.000 0.000
between measurement differences. Measurement dif- z3 0.000 0.000
ferences of approximately 0 mm are observed for the z2 z3 0.000 0.000
points used to generate the coordinate system. Mea- AEML x1 x2 0.000 0.000
surement differences of 1 mm are observed for points x3 0.000 0.000
x2 x3 0.000 0.000
that were away from the origin of the coordinate sys- y1 y2 0.000 0.000
tem. y3 0.000 0.000
y2 y3 0.000 0.000
z1 z2 0.000 0.000
DISCUSSION z3 0.000 0.000
Superimposition of images is a tool used by ortho- z2 z3 0.000 0.000
dontists to aid in diagnosis, treatment planning, treat- AEMR x1 x2 0.000 0.000
ment follow-up, and growth. Many types of superim- x3 0.000 0.000
x2 x3 0.000 0.000
positioning methods have been used for 2D images; for
y1 y2 0.000 0.000
example, lateral cephalograms. With the advent of 3D y3 0.000 0.000
imaging, however, further development in methods for y2 y3 0.000 0.000
superimposition is required. Some work has been done z1 z2 0.000 0.000
on the development of ways to superimpose 3D imag- z3 0.000 0.000
z2 z3 0.000 0.000
ing. One method is using gray scales to superimpose
DFM x1 x2 0.000 0.000
images. Cevidanes et al.12,13 used this style of CBCT x3 0.000 0.000
image superimposition to determine treatment changes x2 x3 0.000 0.000
associated with orthognathic surgery. Their method y1 y2 0.000 0.000
consisted of identifying the cranial base structures in y3 0.000 0.000
y2 y3 0.000 0.000
the images. These images were later subjected to dif-
z1 z2 0.000 0.000
ferent software to obtain mean surface distance and to z3 0.000 0.000
quantify how much, on average, the 2 surfaces differ z2 z3 0.000 0.000
from each other using color-coded graphical displays. InfraOL x1 x2 –0.229 0.167
Changes seen after using the software can only be x3 0.541 0.577
x2 x3 0.771 0.527
interpreted in the sense of surface distance changes in
y1 y2 0.144 0.357
tendencies (outward or inward) of large anatomical y3 –0.255 0.408
structures (maxilla and mandible). For this reason, a y2 y3 –0.399 0.327
different approach was investigated in this work to z1 z2 0.097 0.663
obtain changes in anatomical structures (whatever their z3 –1.039 0.684
z2 z3 –1.136 0.693
size) and reporting these changes in terms of each
InfraOR x1 x2 –0.777 0.218
coordinate axis individually. x3 0.191 0.479
Kawamata et al.14 also tried to superimpose 3D im- x2 x3 0.969 0.524
ages. The images used in their study were obtained y1 y2 0.432 0.299
from a CT machine and their method involved manu- y3 0.514 0.291
y2 y3 0.082 0.216
ally superimposing anatomical structures to view con-
z1 z2 0.213 0.752
dyle displacement after surgery. The steps involved in z3 –0.757 0.637
this process were to create lateral, axial, and frontal 3D z2 z3 –0.970 0.765
CT images of the pre- and postoperative temporoman-
dibular joint region. Then, both lateral images (pre and
post) were superimposed and rotated until anatomic
structures, such as zygomatic arch, mastoid process,
and infraorbital foramen overlapped. Once this was
 OOOOE
776 Lagravère et al. June 2011

Table II. Continued errors in determining X, Y, and Z coordinates of dis-

Pairwise Comparisons tance landmarks. Errors present in landmarks chosen to
Mean Difference Standard Error standardize the reference plane can be magnified to
MnL x1 x2 0.002 0.411 points that are farther from these reference points. A
x3 1.302 0.747 minor error in setting the axis results in minor devia-
x2 x3 1.300 0.767 tions for points near the origin of the coordinate system;
y1 y2 0.020 0.576 however, this results in large deviations for points in
y3 0.437 0.655
y2 y3 0.417 0.736
regions farther away from the origin. There is a possi-
z1 z2 0.351 0.706 bility that errors in one axis could be canceled by an
z3 –0.796 0.744 error in another axis or errors in other reference points,
z2 z3 –1.147 0.723 but there is also a possibility that these could increase
MnR x1 x2 –0.475 0.505 the error present in the region of interest.
x3 0.892 0.658
x2 x3 1.367 0.779 The basis of this analysis and point selection has
y1 y2 –0.114 0.593 been described in a previous article.10 It has been
y3 0.725 0.557 reported that intrareliability of the landmarks forming
y2 y3 0.839 0.757 the coordinate system (ELSA, AEML, AEMR, and
z1 z2 0.388 0.755
DFM) was excellent in all axes.10 Four points (infraor-
z3 –0.450 0.727
z2 z3 –0.838 0.826 bital left and right and menton left and right) located at
the maximum distance from the cranial base reference
AEML, auditory external meatus left; AEMR, auditory external me-
atus right; DFM, dorsum foramen magnum; ELSA, midpoint be- system were located and used to determine the trans-
tween both foramen spinosums; InfraOL, infraorbitale left; InfraOR, formation potential to produce error when superimpo-
infraorbitale right; MnL, menton left; MnR, menton right. sitioning. When the potential envelope of error for the
No statistically significant difference was found (P ⬎ .05) on all reference plane system produced by the compounding
comparisons.
error of the landmarks defining the reference points was
applied using mathematical transformation, the error in
locating distant landmarks was high, as shown in Table
I in the first transformation column. By applying a
done, a joint image was created where colors demon- second mathematical transformation, it can be seen that
strated the amount of displacement of the condyle. This this envelope of error reduces significantly (Table I).
method, although practical, can present several draw- An alternative technique for CBCT image superim-
backs, as trying to determine coordinate displacements position is best-fit analysis of multiple cranial base
of structures cannot be done because no reference sys- landmarks and computer-aided superimposition based
tem has been determined. To apply this method to on best fit of object shapes in the cranial base.13 Opti-
growing patients is also difficult because anatomical mization analysis is another alternative to use when
structures are still changing and using this method in trying to determine a standard reference system based
the cranial base can be complicated, as visualization of on specific landmarks; this was the alternative verified
these structures is difficult when using only 3D images. in this article. Some variations of this technique are
Another way to approach superimposition is with the planned for future research; one such variation is to use
use of an optimization analysis. This analysis would 6 landmarks instead of 4 to optimize the coordinate
involve minimizing the total root mean square error system. Another would be to replace the landmarks
found over a series of fixed landmark positions. This is most prone to identification error and use other land-
a best-fit type of analysis. There are many ways of marks located in the vicinity of the original 4. Future
producing optimization analysis.15,16 Cevidanes et research is needed to critically evaluate the errors as-
al.’s12 approach involved, in the initial steps, a type of sociated with these alternative techniques.
optimization analysis. The method presented in the
present article has the goal to be a simple, precise, and CONCLUSIONS
reliable instrument to analyze changes within an indi- The objective of this article was to improve a 3D
vidual over time for assessment of growth and treat- method to superimpose CBCT images mentioned in a
ment outcomes. previous study.10 By using an optimization analysis to
Initial evaluation of the system presented in this reduce the mean square root error of landmark location
article found good reference plane reliability, but sec- of ELSA, AEML, AEMR, and DFM (landmarks used
ondary sensitivity analysis identified that small errors in to establish the reference coordinate system), the mean
locating cranial base landmarks had large potential errors were greatly reduced. This analysis gives good
OOOOE
Volume 111, Number 6
prognosis on the establishment of an easy, accurate, and
reliable method to superimpose 3D images.
REFERENCES
1. Leonardi R, Giordano D, Maiorana F, Spampinato C. Automatic
cephalometric analysis. Angle Orthod 2008;78:145-51.
2. Bjork A. The use of metallic implants in the study of facial
growth in children: Method and application. Am J Phys Anthro-
pol 1968;29:243-54.
3. Melsen B, Melsen F. The postnatal development of the palato-
maxillary region studied on human autopsy material. Am J
Orthod 1982;82:329-42.
4. Berkowitz S. A multicenter retrospective 3D study of serial
complete unilateral cleft lip and palate and complete bilateral
cleft lip and palate casts to evaluate treatment: part 1—the
participating institutions and research aims. Cleft Palate Cranio-
fac J 1999;36:413-24.
5. Lagravère MO, Gordon JM, Guedes IH, Flores-Mir C, Carey JP,
Heo G, et al. Reliability of traditional cephalometric landmarks
as seen in three-dimensional analysis in maxillary expansion
treatments. Angle Orthod 2009;79:1047-56.
6. Lagravère MO, Low C, Flores-Mir C, Chung R, Carey JP, Heo
G, et al. Intraexaminer and interexaminer reliabilities of land-
mark identification on digitized lateral cephalograms and format-
ted 3-dimensional cone-beam computerized tomography images.
Am J Orthod Dentofac Orthop 2010;137:598-604.
7. Seckel NG, van der Tweel I, Elema GA, Specken TF. Landmark
positioning on maxilla of cleft lip and palate infant—a reality?
Cleft Palate Craniofac J 1995;32:434-41.
8. Sachdeva RC. SureSmile technology in a patient— centered orth-
odontic practice. J Clin Orthod 2001;35:245-53.
9. Ashmore JL, Kurland BF, King GJ, Wheeler TT, Ghafari J,
Ramsay DS. A 3-dimensional analysis of molar movement dur-
Lagravère et al. 777
ing headgear treatment. Am J Orthod Dentofac Orthop
2002;121:18-29; Discussion: 29-30.
10. Lagravère MO, Major PW, Carey J. Sensitivity analysis for plane
orientation in three-dimensional cephalometric analysis based on
superimposition of serial cone beam computed tomography im-
ages. Dentomaxillofac Radiol 2010;39:400-8.
11. Lagravère MO, Major PW. Proposed reference point for 3-di-
mensional cephalometric analysis with cone-beam computerized
tomography. Am J Orthod Dentofac Orthop 2005;128:657-60.
12. Cevidanes LH, Bailey LJ, Tucker GR Jr., Styner MA, Mol A,
Phillips CL, et al. Superimposition of 3D cone-beam CT models
of orthognathic surgery patients. Dentomaxillofac Radiol
2005;34:369-75.
13. Cevidanes LH, Styner MA, Proffit WR. Image analysis and
superimposition of 3-dimensional cone-beam computed tomog-
raphy models. Am J Orthod Dentofac Orthop 2006;129:611-8.
14. Kawamata A, Fujishita M, Nagahara K, Kanematu N, Niwa K,
Langlais RP. Three-dimensional computed tomography evalua-
tion of postsurgical condylar displacement after mandibular os-
teotomy. Oral Surg Oral Med Oral Pathol Oral Radiol Endod
1998;85:371-6.
15. Maes F, Collignon A, Vandermeulen D, Marchal G, Suetens P.
Multimodality image registration by maximization of mutual
information. IEEE Trans Med Imaging 1997;16:187-98.
16. Czaplicki A. Optimization solutions depend on the choice of
coordinate system. Acta Bioeng Biomech 2008;10:75-9.
Reprint requests:
Manuel O. Lagravère, DDS, PhD
Faculty of Medicine and Dentistry
Rm. 4051, Dentistry/Pharmacy Centre
University of Alberta
Edmonton, Alberta, Canada T6G 2N8
manuel@ualberta.ca