ORIGINAL ARTICLE

Classification of facial asymmetry by

cluster analysis
Hyeon-Shik Hwang,a Il-Sun Youn,b Ki-Heon Lee,c and Hoi-Jeong Limd
Gwangju and Daejeon, Korea

Introduction: The purposes of this study were to classify patients with facial asymmetry by using the cluster
analysis and to evaluate the characteristics of the resultant groups. Methods: One hundred consecutive
orthodontic patients who had frontal cephalograms and photographs taken for the diagnosis of facial
asymmetry were included in the sample. A cluster analysis was used to classify the subjects, with craniofacial
measurements related to asymmetry obtained from the cephalograms and photographs. Results: The
subjects were divided into 5 clusters based on only 3 variables from the frontal cephalograms: menton
deviation, apical base midline discrepancy, and vertical difference of right and left antegonion. The results of
the ANOVA test according to the variables showed statistically significant differences between the groups in
all variables, indicating that each group can be easily identified with the variables. Conclusions: The results
suggest that patients with facial asymmetry can be classified simply based on some variables on frontal
cephalograms, and this classification can be helpful in proper diagnosis and treatment planning for these
patients. (Am J Orthod Dentofacial Orthop 2007;132:279.e1-279.e6)

I
n the past, mild facial asymmetry was disregarded other factors such as onset, individual growth, and com-
by clinicians because it was believed that normal pensation.
craniofacial skeletons had some asymmetry, and Although the nature of asymmetry is complex and its
that this was subclinical and could be compatible with characteristics are diverse, approaches to systematic clas-
normal dental occlusion.1-4 However, with the increas- sification of facial asymmetry have been few. Hinds et al9
ing concern about facial appearance, patients now classified mandibular asymmetry into 2 categories: unilat-
complain of even slight asymmetry. Some patients eral condylar hyperplasia and deviation prognathism.
believe asymmetry develops or progresses during or Rowe10 classified asymmetry into 3 groups: unilateral
after orthodontic treatment. A systematic and accurate condylar hyperplasia, unilateral macrognathia confined to
approach to facial asymmetry is mandatory for proper the skeletal element only, and unilateral macrognathia of
management in orthodontic practices. both osseous and muscular components. Bruce and Hay-
Many etiologic factors are related to the development ward11 classified mandibular asymmetry into deviation
of facial asymmetry.5-8 Bishara et al,7 in a review of dental prognathism, unilateral condylar hyperplasia, and unilat-
and facial asymmetries, showed that many factors are eral macrognathia. However, these classifications were
implicated in asymmetry: genetic or congenital malforma- mostly for severe asymmetry, which is the concern of
tions such as hemifacial microsomia, environmental fac- surgeons. Bishara et al7 classified dentofacial asymmetry
tors such as habits and trauma, functional deviations, and into dental, skeletal, muscular, and functional. However,
so on. Asymmetry can have different characteristics even that classification system was simply based on the struc-
with the same etiology because it can be influenced by tures involved and did not provide enough information for
diagnosis and treatment planning in orthodontic practices.
a
Professor and chairman, Department of Orthodontics, 2nd Stage of Brain The purposes of this study were to classify
Korea 21, School of Dentistry, Dental Science Research Institute, Chonnam patients with facial asymmetry by using the cluster
National University, Gwangju, Korea.
b
Private practice, Daejeon, Korea.
analysis technique on the basis of the craniofacial
c
Associate professor, Department of Orthodontics, School of Dentistry, measurements related to asymmetry and to evaluate
Chonnam National University, Gwangju, Korea. the characteristics of the resultant groups according
d
Professor, Department of Orthodontics, 2nd Stage of Brain Korea 21, School
of Dentistry, Chonnam National University, Gwangju, Korea.
to the measurements.
Reprint requests to: Hyeon-Shik Hwang, Department of Orthodontics, School
of Dentistry, Chonnam National University, Hak-Dong, Dong-Gu, Gwangju
501-757, Korea; e-mail, hhwang@chonnam.ac.kr.
MATERIAL AND METHODS
Submitted, September 2006; revised and accepted, January 2007. The sample consisted of 100 consecutive orthodon-
0889-5406/$32.00
Copyright © 2007 by the American Association of Orthodontists. tic patients (41 male, 59 female; mean age, 18.2 years;
doi:10.1016/j.ajodo.2007.01.017 SD, 7.5 years) who had frontal cephalograms and
279.e1
279.e2 Hwang et al American Journal of Orthodontics and Dentofacial Orthopedics
September 2007

4. Horizontal difference of right and left Ag (X4):

difference between horizontal positions of the right
Ag and left Ag.
5. Maxillary base canting (X5): angle formed by the
line connecting the right and left jugal points and
the horizontal reference line, which is vertical to
MSR.
6. Maxillary alveolar canting (X6): angle formed by
the line connecting the right and left points con-
structed with the buccal contour of the first molar
and the adjacent alveolar bone contour and the
horizontal reference line.
7. Bulkiness difference of mandibular inferior bor-
der (X7): difference between the right and left
bulkiness of the mandible determined subjec-
tively as a number from 1 to 5, with a larger
number for more severe asymmetry.
8. Lip line canting (X8): angle formed by the line
connecting the right and left commissures of the lip
and the interpupillary line.
Fig 1. Linear and angular measurements on frontal
cephalogram: X1, Me deviation; X2, apical base midline Each measurement was recorded with a sign
discrepancy; X3, vertical difference of right and left Ag; (⫹ or –) for asymmetry. The direction of chin devia-
X4, horizontal difference of right and left Ag (X4L-X4R); tion was determined from each subject’s frontal
X5, maxillary base canting; X6, maxillary alveolar cant- photograph, and a sign was given to the measurement
ing. based on the direction of chin deviation. For X1, ⫹
was given to the measurement if the direction of Me
frontal photographs taken for diagnosis of facial asym- deviation was the same as the direction of chin
metry in Chonnam National University Hospital, deviation, and – was given if not the same. For X2, ⫹
Gwangju, Korea. A frontal cephalogram and a frontal was given if the midpoint of the mandibular incisors
photograph were taken of each subject, who was seated was positioned on the same side as the chin deviation
in an upright position with the head fixed by ear rods compared with the midpoint of the maxillary incisor
and with the Frankfort horizontal plane parallel to the roots, and – was given if not. For X3, ⫹ was given if
floor. Yashica Dental Eye III (Kyocera, Kyoto, Japan) the Ag on the side of the chin deviation was positioned
was used at magnification rate of 1:10 for the photo- superiorly compared with the opposite side of Ag. For
graphs. X4, ⫹ was given if the Ag on the side of the chin
Tracings were constructed of the frontal cephalo- deviation was positioned far from the MSR compared
grams, and traditional contours and points were digi- with the opposite side of Ag. For X5 and X6, ⫹ was
tized by an author (I.Y). For the construction of the given if the connecting line canted upward on the side
midsagittal reference line (MSR), crista galli and ante- of chin deviation, and – was given if the line canted
rior nasal spine were used. As the measurements related downward on the deviation side. For X7, ⫹ was given
to facial asymmetry, 7 measurements were made from if more bulkiness was seen on the deviation side,
the frontal cephalograms and 1 measurement from the and – was given if on the opposite side. For X8, ⫹ was
photographs (Fig 1). given if the lip line canted upward on the deviation
side, and – was given if the line canted downward on
1. Menton (Me) deviation (X1): angle formed by the
the deviation side.
crista galli-Me line and MSR.
2. Apical base midline discrepancy (X2): horizontal
distance between the midpoints of the maxillary Statistical analysis
central incisor roots and the mandibular central A cluster analysis is a multivariate procedure used
incisor roots. to detect natural data groupings on the basis of simi-
3. Vertical difference of right and left antegonion larity of the data.12 The K-means cluster analysis was
(X3): vertical distance between right antegonion used to classify the 100 subjects in this study with the
(Ag) and left Ag. 8 variables previously described. To assess the repro-
American Journal of Orthodontics and Dentofacial Orthopedics Hwang et al 279.e3
Volume 132, Number 3

Fig 2. Scattergram for 5 subgroups segregated by cluster analysis: X1, Me deviation; X2, apical
base midline discrepancy; X3, vertical difference of right and left Ag.

ducibility of the cluster analysis, 4 subsamples, which minimal values. For apical base midline discrepancy,
consisted of approximately 70% of the patients ran- groups A and C showed the greatest values, whereas
domly selected from the original 100 facial asymmetry groups B, D, and E had minimal values. For vertical
patients, were subjected to the same clustering proce- difference of right and left Ag, groups A and D
dure as the original one. To investigate the character- showed large discrepancies, and groups C and E had
istics of each group, ANOVA was used according to minimal values. In group B, Ag on the chin deviation
the measurements, and the Duncan test was used as a side was positioned inferiorly; this was different
post-hoc comparison. Statistical evaluations were per- from groups A and D. For horizontal difference of
formed at the 5% level of significance with software right and left Ag, maxillary base canting, maxillary
(version 8.2, SAS, Cary, NC). alveolar canting, and lip line canting, group A had
greater values than the other groups. For the bulki-
RESULTS ness difference of the mandibular inferior border, the
Our subjects were divided into five clusters, A differences were not distinct in general, although
(n ⫽ 7), B (n ⫽ 16), C (n ⫽ 21), D (n ⫽ 28), and E group B had larger values than the other groups.
(n ⫽ 28), as illustrated in Figure 2. Three variables From the overall evaluation, group A showed
were used for the analysis: Me deviation, apical base greater values in all variables except bulkiness differ-
midline discrepancy, and vertical difference of right ence, and group E had small values in general. On the
and left Ag. The results of the reproducibility test other hand, group B had ⫺ values in many variables,
showed 95.3%, 93.8%, 94.3%, and 94.5% with including the vertical difference of right and left Ag;
subsamples of 64, 65, 70, and 73 subjects, respec- this indicated a specific type of asymmetry compared
tively. with the other groups.
Table I gives the results of ANOVA according to
the variables. All variables showed statistically sig- DISCUSSION
nificant differences between the groups. For Me The cluster analysis technique divided the 100
deviation, group A had greatest value, groups C and patients into 5 groups with different characteristics.
D had moderate values, and groups B and E had Interestingly, only 3 variables were selected in the
279.e4 Hwang et al American Journal of Orthodontics and Dentofacial Orthopedics
September 2007

Table I. Comparison of groups classified by cluster analysis according to measurements

Variable Mean ⫾ SD P value

X1 Group A Group C Group D Group B Group E ⬍.001

4.93 ⫾ 1.54 2.43 ⫾ 0.62 1.75 ⫾ 0.84 0.91 ⫾ 1.56 0.45 ⫾ 0.95

X2 Group A Group C Group B Group D Group E ⬍.001

5.93 ⫾ 1.48 3.93 ⫾ 1.32 1.56 ⫾ 2.07 1.34 ⫾ 1.55 0.70 ⫾ 0.97
X3 Group A Group D Group C Group E Group B ⬍.001
7.21 ⫾ 1.82 4.93 ⫾ 1.43 0.57 ⫾ 1.59 0.16 ⫾ 1.46 ⫺4.09 ⫾ 1.23
X4 Group A Group C Group D Group E Group B ⬍.001
9.79 ⫾ 4.33 3.42 ⫾ 2.62 2.16 ⫾ 3.90 0.63 ⫾ 3.53 ⫺0.72 ⫾ 2.83

X5 Group A Group D Group C Group E Group B .001

2.79 ⫾ 1.80 1.23 ⫾ 1.97 0.50 ⫾ 1.87 0.13 ⫾ 1.99 ⫺0.47 ⫾ 2.13

X6 Group A Group D Group C Group E Group B ⬍.001

4.29 ⫾ 2.61 2.05 ⫾ 1.61 1.50 ⫾ 1.30 0.11 ⫾ 1.71 ⫺0.13 ⫾ 2.07

X7 Group B Group C Group A Group E Group D ⬍.001

2.44 ⫾ 1.59 1.10 ⫾ 2.10 0.71 ⫾ 2.87 0.50 ⫾ 1.73 ⫺0.57 ⫾ 1.81

X8 Group A Group D Group C Group E Group B .001

2.64 ⫾ 2.19 1.45 ⫾ 1.63 0.85 ⫾ 1.41 0.41 ⫾ 1.48 ⫺0.19 ⫾ 1.74

X1, Menton deviation; X2, apical base midline discrepancy; X3, vertical difference of right and left antegonion; X4, horizontal difference of right
and left antegonion; X5, maxillary base canting; X6, maxillary alveolar canting; X7, bulkiness difference of mandibular inferior border; X8, lip line
canting in facial photograph. Groups connected by horizontal line were not significantly different.

Table II. Characteristics of groups classified by cluster analysis

Variables Group A Group B Group C Group D Group E

X1 Severe Mild Moderate Moderate NS

X2 Severe Mild Moderate Mild NS
X3 Superiorly Inferiorly NS Superiorly NS

X1, Degree of menton deviation; X2, apical base midline discrepancy; X3, vertical position of antegonion on menton deviation side; NS, not
significant.

Fig 3. Schematic representation of groups: Mx, maxillary apical base midline; Mn, mandibular
apical base midline; Me, Me position.

analysis, whereas 8 variables were used. This indicates Group A showed a distinct difference of right and left
that facial asymmetry can be classified by using only 3 ramus length. Me was deviated to the side of the shorter
variables. Our results would simplify a diagnostic ramus. Mandibular apical base midline also was deviated
procedure for facial asymmetry. Table II summarizes to the same side. As shown in Table I, maxillary basal and
the characteristics of the groups, and Figure 3 presents alveolar bone had significant canting. Lip line canting also
a schematic illustration of each group. is distinct in this type of asymmetry. Group A has a typical
American Journal of Orthodontics and Dentofacial Orthopedics Hwang et al 279.e5
Volume 132, Number 3

type of asymmetry; its etiology is believed to be asym- made simply by determining whether there is a unilateral
metric condylar or whole mandibular growth.9-11 With chewing habit.
asymmetric mandibular growth, the chin deviates to the Group C had a specific characteristic: no ramus
side of the shorter ramus. Compensatory asymmetric length difference between the right and left sides but
vertical growth of the maxilla is common in this type of Me and lower apical base midline deviations to 1 side.
asymmetry, contributing to cants of the maxilla and the lip The reason for this asymmetry is a functional shift of
line. the mandible that results from various types of occlusal
Theoretically, a proper treatment modality for this interference such as a malposed tooth, a crossbite of at
type of asymmetry would be growth modification with least 1 tooth, or a constricted maxillary arch. Abnormal
functional jaw orthopedic appliances such as the hybrid tooth contact causes the subsequent mandibular dis-
appliance.13 However, perfect correction of facial placement in maximum intercuspation. Schmid et al18
asymmetry might not be obtained if growth is unfavor- described this type as displacement asymmetry; other
able. If the patient is an adult, surgical correction would asymmetries were structural. Joondeph,19 in a recent
be a treatment choice for this type of asymmetry as review, called this type functional asymmetry and
suggested by Obwegeser and Makek.14 suggested intercepting it at an early stage because the
Group B showed the same ramus length difference condyle and the fossa adapt easily to the deviated
between the right and left sides as group A, but Me mandibular position.20-22
deviation was different. Me was deviated in the oppo- The characteristics of group D were similar to
site direction to shorter ramus, indicating that the group A. The only difference was in magnitude. A
etiology of group B was not related to asymmetric separate cluster was formed in the analysis because all
condylar growth. Waugh,15 in a study on the influence variables showed small discrepancies. Although the
treatment modality for this group would be the same as
of diet on the jaws and faces of American Eskimos,
for group A, camouflage treatment could be a good
indicated that the difference of chewing force affects
option without correction of face, because the degree of
the form of the jaws and the face. Tay16 suggested that
asymmetry is not great.
unilateral mastication has a certain association with
Group E showed minimal values in all variables,
mandibular asymmetry. Ingervall,17 in an electromyo-
indicating that these subjects were within normal limits.
graphic study to investigate the correlation between
Relatively many subjects, 28%, were included in this
muscular activity and facial morphology, showed that
cluster, because our subjects were patients who had
subjects with marked activity in the masticatory mus-
frontal cephalograms taken for diagnosis, regardless of
cles had shorter face height—the so-called “low angle.”
existence or severity of facial asymmetry.
Considering all these studies, it is likely that the From the results of this study, it is recognized that
asymmetry in group B resulted from unilateral masti- each group has a specific etiology and a different
cation: the chewing side of the jaw develops a low treatment modality. Although facial asymmetry has
angle, and the nonchewing side develops a high angle. various etiologic factors and diverse characteristics, it
In addition to different angle shapes between the right can be classified into 4 or 5 groups. Classification is
and left sides, soft-tissue asymmetry can result from possible with only 3 variables from frontal cephalo-
unilateral muscle hypertrophy on the chewing side. grams. We believe that this classification system will be
Various causative factors such as missing teeth, scis- a great help in accurate diagnosis and proper treatment
sors-bite, occlusal interference, and faulty restorations planning for patients with facial asymmetry.
on 1 side might be responsible for this type of asym-
metry. CONCLUSIONS
Another possible cause of the asymmetry in group B These results suggest that patients with facial asym-
is deviation prognathism, described by Bruce and Hay- metry can be classified simply based on 3 variables on
ward.11 They suggested that an increase on 1 side of the a frontal cephalogram: Me deviation, apical base mid-
mandible could occur without asymmetric condylar line discrepancy, and vertical difference of Ag between
growth. Similarly, Obwegeser and Makek14 suggested the right and left sides.
differentiating hemimandibular hyperplasia from hemi-
mandibular elongation, although the classification was not REFERENCES
totally the same as the system proposed by Bruce and
1. Letzer GM, Kronman JH. A posteroanterior cephalometric eval-
Hayward.11 In any case, group B asymmetry might result uation of craniofacial asymmetry. Angle Orthod 1967;37:205-11.
from not only unilateral mastication but also aberrant 2. Vig PS, Hewitt AB. Asymmetry of the human facial skeleton.
growth. Distinction between 2 causative factors could be Angle Orthod 1975;45:125-9.
279.e6 Hwang et al American Journal of Orthodontics and Dentofacial Orthopedics
September 2007

3. Shah SM, Joshi MR. An assessment of asymmetry in the normal 14. Obwegeser HL, Makek MS. Hemimandibular hyperplasia—
craniofacial complex. Angle Orthod 1978;48:141-8. hemimandibular elongation. J Maxillofac Surg 1986;14:183-208.
4. Peck S, Peck L, Kataja M. Skeletal asymmetry in esthetically 15. Waugh LM. Influence of diet on the jaw and face of the
pleasing faces. Angle Orthod 1991;61:43-8. American Eskimo. J Am Dent Assoc 1937;24:1640-7.
5. Lundstrom A. Some asymmetries of the dental arches, jaws, and skull, 16. Tay DK. Physiognomy in the classification of individuals with a
and their etiological significance. Am J Orthod 1961;47:81-106. lateral preference in mastication. J Orofac Pain 1994;8:61-72.
6. Erickson GE, Waite DE. Mandibular asymmetry. J Am Dent 17. Ingervall B. Facial morphology and activity of temporal and lip
Assoc 1974;89:1369-73. muscles during swallowing and chewing. Angle Orthod 1976;
7. Bishara SE, Burkey PS, Kharouf JG. Dental and facial asymme- 46:372-80.
tries: a review. Angle Orthod 1994;64:89-98. 18. Schmid W, Mongini F, Felisio A. A computer-based assessment
8. Cohen MM. Perspectives on craniofacial asymmetry. III. Com-
of structural and displacement asymmetries of the mandible.
mon and/or well-known causes of asymmetry. Int J Oral Maxil-
Am J Orthod Dentofacial Orthop 1991;100:19-34.
lofac Surg 1995;24:127-33.
19. Joondeph DR. Mysteries of asymmetries. Am J Orthod Dento-
9. Hinds EC, Reid LC, Burch RJ. Classification and management of
facial Orthop 2000;117:577-9.
mandibular asymmetry. Am J Surg 1960;100:825-34.
10. Rowe NL. Aetiology, clinical features, and treatment of mandib- 20. Pirttiniemi P, Kantomaa T, Lahtela P. Relationship between
ular deformity. Br Dent J 1960;108:64-96. craniofacial and condyle path asymmetry in unilateral cross-bite
11. Bruce RA, Hayward JR. Condylar hyperplasia and mandibular patients. Eur J Orthod 1990;12:408-13.
asymmetry: a review. J Oral Surg 1968;26:281-90. 21. Hesse KL, Årtun J, Joondeph DR, Kennedy DB. Changes in
12. Johnson RA, Wichern DW. Applied multivariate statistical condylar position and occlusion associated with maxillary ex-
analysis. Englewood Cliffs, NJ: Prentice-Hall International; pansion for correction of functional unilateral posterior crossbite.
1988. p. 573-627. Am J Orthod Dentofacial Orthop 1997;111:410-8.
13. Vig PS, Vig KW. Hybrid appliances: a component approach to 22. Langberg BJ, Arai K, Miner RM. Transverse skeletal and dental
dentofacial orthopedics. Am J Orthod Dentofacial Orthop 1986; asymmetry in adults with unilateral lingual posterior crossbite.
90:273-85. Am J Orthod Dentofacial Orthop 2005;127:6-15.