Accepted Manuscript Describing Crouzon and Pfeiffer syndrome based on principal component analysis Femke C.R. Staal, B.Sc., Allan J.T. Ponniah, M.R.C.S. (Eng.), M.Sc., Freida Angullia, B.Sc., M.D., M.R.C.S. (Eng.), Clifford Ruff, B.Sc., M.Sc., Maarten J. Koudstaal, M.D., D.D.S., PhD, David Dunaway, F.D.S., R.C.S FRCS (plast.) PII:
S1010-5182(15)00027-X
DOI:
10.1016/j.jcms.2015.02.005
Reference:
YJCMS 1967
To appear in:
Journal of Cranio-Maxillo-Facial Surgery
Received Date: 1 January 2015 Accepted Date: 6 February 2015
Please cite this article as: Staal FCR, Ponniah AJT, Angullia F, Ruff C, Koudstaal MJ, Dunaway D, Describing Crouzon and Pfeiffer syndrome based on principal component analysis, Journal of CranioMaxillofacial Surgery (2015), doi: 10.1016/j.jcms.2015.02.005. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Describing Crouzon and Pfeiffer syndrome based on principal component analysis Femke C.R. Staal, B.Sc., Great Ormond Street Hospital, London, United Kingdom
RI PT
Allan J.T. Ponniah, M.R.C.S. (Eng.), M.Sc., Great Ormond Street Hospital, London, United Kingdom
Freida Angullia, B.Sc., M.D., M.R.C.S. (Eng.), Great Ormond Street Hospital, London, United Kingdom
Clifford Ruff, B.Sc., M.Sc., Medical Physics Department, University college London,
SC
London, United Kingdom
M AN U
Maarten J. Koudstaal, M.D., D.D.S., PhD, Erasmus Medical Center, Maxillofacial Surgery, Rotterdam, The Netherlands David Dunaway, F.D.S., R.C.S FRCS (plast.), Great Ormond Street Hospital, London,
AC C
EP
TE D
United Kingdom
1
ACCEPTED MANUSCRIPT
Corresponding author Femke C.R. Staal [email protected] Tel: +31 (0)6 253 30 443
RI PT
Craniofacial department FAO David Dunaway Great Ormond Street Hospital Great Ormond Street London WC1N 3JH
AC C
EP
TE D
M AN U
SC
Fax: +44 (0) 20 7813 8446
2
ACCEPTED MANUSCRIPT
Financial disclosure None of the authors has a financial interest in any of the products, devices, or drugs
AC C
EP
TE D
M AN U
SC
RI PT
mentioned in this article.
3
ACCEPTED MANUSCRIPT
Abstract Crouzon and Pfeiffer syndrome are syndromic craniosynostosis caused by specific mutations in the FGFR genes. Patients share the characteristics of a tall, flattened
RI PT
forehead, exorbitism, hypertelorism, maxillary hypoplasia and mandibular prognathism. Geometric morphometrics allows the identification of the global shape changes within and between the normal and syndromic population. Methods: Data from 27 Crouzon-
SC
Pfeiffer and 33 normal subjects were landmarked in order to compare both populations. With principal component analysis the variation within both groups was visualized and
M AN U
the vector of change was calculated. This model normalized a Crouzon-Pfeiffer skull and was compared to age-matched normative control data. Results: PCA defined a vector that described the shape changes between both populations. Movies showed how the normal skull transformed into a Crouzon-Pfeiffer phenotype and vice versa. Comparing these results to established age-matched normal control data confirmed that our model
TE D
could normalize a Crouzon-Pfeiffer skull. Conclusions: PCA was able to describe deformities associated with Crouzon-Pfeiffer syndrome and is a promising method to analyze variability in syndromic craniosynostosis. The virtual normalization of a Crouzon-
EP
Pfeiffer skull is useful to delineate the phenotypic changes required for correction, can help surgeons plan reconstructive surgery and is a potentially promising surgical
AC C
outcome measure.
Keywords: Crouzon syndrome; Pfeiffer syndrome; Principal component analysis; Craniosynostosis; FGFR2; Three-dimensional imaging
4
ACCEPTED MANUSCRIPT
Introduction Crouzon and Pfeiffer syndrome are autosomal dominant disorders associated with multisutural craniosynostosis.(Crouzon 1912; Pfeiffer 1964) Both syndromes have a
RI PT
large variation in phenotype in both the cranial and facial features.(Carinci, Pezzetti et al. 2005; Vogels and Fryns 2006) Patients share the characteristics of a tall, flattened
forehead (due to bicoronal synostosis), exorbitism, hypertelorism, maxillary hypoplasia,
SC
and relative mandibular prognatism.(Crouzon 1912; Renier, Lajeunie et al. 2000;
Lajeunie, Heuertz et al. 2006; Vogels and Fryns 2006; Cunningham, Seto et al. 2007)
M AN U
These anatomical abnormalities can contribute to functional problems including increased intracranial pressure (ICP), upper airway obstruction and ocular problems.(Renier, Lajeunie et al. 2000; Vogels and Fryns 2006) Some argue that Crouzon and Pfeiffer syndromes are part of the same entity as both are associated with FGFR2 mutations. This suggests that the same mutation can produce different
TE D
phenotypes.(Rutland, Pulleyn et al. 1995; Meyers, Day et al. 1996; Steinberger, Reinhartz et al. 1996; Lajeunie, Heuertz et al. 2006; Cunningham, Seto et al. 2007) Pfeiffer syndrome has therefore been included in this study.
EP
Previous methods used to analyse these complex deformities include clinical descriptions(Crouzon 1912; Renier, Lajeunie et al. 2000; Carinci, Pezzetti et al. 2005;
AC C
Lajeunie, Heuertz et al. 2006; Vogels and Fryns 2006; Cunningham, Seto et al. 2007), linear morphometrics(Richtsmeier 1987; Posnick, Lin et al. 1993; Farkas, Katic et al. 2005) and 3D-CT analysis.(Carr, Posnick et al. 1992; Kreiborg, Marsh et al. 1993) Most cephalometric methods analyse two-dimensional coordinates of a three-dimensional shape. Conventional 3D-CT analysis describes individual distances and angles between landmarks, rather than the skull shape as a whole unit. A more recent technique to analyse the size and shape of objects is geometric morphometrics.(Bookstein 1997) It has been used successfully in the analysis of craniofacial shapes, for example the face
5
ACCEPTED MANUSCRIPT
variation during growth in primates.(O'Higgins 2000) Principal component analysis (PCA) can describe the complex shape of skulls as a whole and allows for shape comparison between different skulls.
RI PT
The purpose of this study is to characterize the anatomical abnormalities in Crouzon and Pfeiffer skulls using PCA with the intention to develop a process that can
SC
describe how the Crouzon-Pfeiffer skull varies from the unaffected population.
Methods
M AN U
Data collection
Data was collected from a series of 27 Crouzon-Pfeiffer patients, 21 clinically diagnosed with Crouzon syndrome and 6 with Pfeiffer (genetic diagnosis was not available for all patients). The inclusion criteria were patients between ages 6 to 13 with no history of facial surgery. Patients with previous posterior cranial vault remodelling were included,
TE D
but all patients with any history of facial surgery were excluded as this would distort the landmarks. Anatomical normal pediatric data (n=33) was collected from a series epileptic patients undergoing CT scans for surgical planning and two historic collections of skulls:
EP
the Bosma collection (John Hopkins) (Shapiro and Richtsmeier 1997) and the Natural History Museum (London). Inclusion criteria were patients with an unaffected facial
AC C
skeleton, between ages 6 to 13. The scans were taken using a 16 slice Siemens somatom sensation spiral CT scanner set to 0,75 collimation (Siemens Medical Solutions, Malvern, Pa.) and saved as Digital Imaging and Communications in Medicine (DICOM) files. The DICOM files were then converted into a University College London (UCL) proprietary format and loaded into 3D voxel imaging software (Robin 3D 2006). The normal and patient CT images were thresholded for analysis of the hard tissue surface. Polygon mesh surfaces (stl) representing bone, were extracted from all scans for landmark placement.
6
ACCEPTED MANUSCRIPT
Landmarks The 60 scans were landmarked using 3D voxel imaging software (Robin 3D 2006). This
RI PT
software allowed visualization of changes between two scans by creating a thin-plate spline warp. In order to compare normal and patient scans a reliable set of homologous landmarks had to be determined. Landmarks were mainly placed on anatomical points
SC
including foramina, ridges and intersections of sutures for reliability and
repeatability.(Bookstein 1997; O'Higgins 2000) The identification of our landmark set
M AN U
was an iterative process of testing different points. Landmarks capture only the information of specific points and the space between them is interpolated. Therefore, a sufficient number of landmarks were needed to define all of the different features of the skulls. Most landmarks were placed on the frontofacial region where most surgical change occurs with fewer landmarks placed on the back of the skull.
TE D
The first 56 landmarks, which were based on an earlier study(Pluijmers, Ponniah et al. 2012), were located on a test data set using the 3D imaging software. The surface of the skull was then warped to the position of corresponding landmarks of a different
EP
skull in order to show the discrepancies between the skull shapes. The warped scan was superimposed to its actual counterpart and differences in surface were visualized in a
AC C
colour-coded map. This colour map located regions with varying degrees of discrepancy between the two skull surfaces (fig 1). If the landmarks captured all of the skull shape, no differences in surface would appear on the colour map. The landmarks were adjusted in these regions to ensure reliability and repeatability. Some landmarks were removed because they did not contribute to capturing the skull shape. Additional landmarks were placed to improve data capture. 66 landmarks were defined which described the frontofacial skeleton optimally. These landmarks allowed for the best possible comparison between the patient and normal scans (table 1, fig 2). The mandible was
7
ACCEPTED MANUSCRIPT
excluded from analysis because of variation in position and there was a low availability of scans of the whole facial skeleton in the normal group.
RI PT
Data analysis To determine the repeatability of the landmarks, a normal skull and a syndromic skull
were chosen at random and were landmarked ten times in different sittings, making it
SC
possible to obtain the intra-observer reliability.
As Crouzon and Pfeiffer syndrome are both associated with FGFR2 mutations,
M AN U
they are analysed as one group and the syndromes are not separately analysed. PCA is a statistical method that can extract the variations in shape within a population. Eigenvectors can be extracted from the spacial coordinates of the landmarks and these are the principal components shape variation.(O'Higgins 2000) PCA can describe a large amount of high dimensional data with a smaller number of relevant variables, the
TE D
principal components. Therefore the deformities in the Crouzon-Pfeiffer skull can be captured as a whole instead of comparing single measurements. The first principal component describes the largest variation within the population. The thin-plate spline
EP
(TPS) technique can interpolate changes between landmarks and uses minimum bending energy to estimate the surface between landmarks.(Bookstein 1997) We used
AC C
this method to visualize the changes and create movies that showed the variation within and between the populations. To validate the observations of the model, the normalising eigenvector was
applied to all patients, and six measurements were taken before and after. The selected bony measurements assess the orbital and upper midface (zygomatic) regions in the horizontal planes as previously described.(Waitzman, Posnick et al. 1992) To make equivalent measurements after normalizing the skull, 3D cuts were made to simulate two comparable CT slices. All patients were positioned in the orbitomeatal plane and a cut
8
ACCEPTED MANUSCRIPT
was made in the orbital region and a second cut in the upper midface (zygomatic) region. The 3D imaging software was used to perform all measurements and the data was compared to age-matched normative control data.(Waitzman, Posnick et al. 1992)
groups: 6 to 9, and 10 to 13 years of age.
SC
Results
RI PT
The means and standard deviations of the normative control data were split in two
Landmark reliability
M AN U
For each landmark the standard deviation (SD) was calculated (table 2). All landmarks used were within a SD value of 3 mm. Normal population
5 out of the 66 landmarks were outside of the 2 mm limit. 18 landmarks were between 1 mm and 2 mm and 43 landmarks had an SD < 1 mm. Thus, 43/66 (65%) of the
Patient population
TE D
landmarks placed were considered highly accurate and 92% were within 2 mm limit.
6 out of the 66 landmarks were outside of the 2 mm limit. 13 landmarks were between 1
EP
mm and 2 mm and 47 landmarks had an SD of < 1 mm. Thus, 47/66 (71%) of the
AC C
landmarks were considered highly accurate and 91% within the 2 mm limit.
Points placed around more distinct anatomical features like the infraorbital foramen had higher accuracy than points placed on curvatures (table 2). This was true for both normal and syndromic skulls. All of the landmarks have shown to be reliable to locate on the different skull shapes. Further points were considered, but were found to be unreliable and therefore excluded from the PCA.
Variation within the populations
9
ACCEPTED MANUSCRIPT
The PCA on each group defined the variability within the groups. The first principal component of variation in the normal group showed allometric growth within the age range of 6 to 13 years (fig 3).
RI PT
The second component shows how long faces tend to be thinner and short faces tend to be wider in the way they vary in relation to each other. Both forehead and maxilla contribute equally to the variation in lengthening and widening of the skull. There is little
SC
variation in the shape and position of the orbit (fig 4). Table 3 shows the age and sex distribution of the syndromic population. The first mode of variation in the syndromic
M AN U
group displayed the variation in the slope and height of the forehead, midfacial height and retrusion and variation in the orbital shape and orientation (fig 5). The second component primarily shows allometric growth within the age range of 6 to 13. There is little variation in the forehead, maxilla or orbital shape compared to the first component
TE D
(fig 6).
Variation between the populations
PCA was also performed between normal and syndromic subjects resulting in an
EP
‘average’ vector between all the warps. This averaged vector describes shape changes between the syndromic and normal skulls and represents a model for normalization of a
AC C
Crouzon-Pfeiffer skull. Resulting thin-plate spline movies of the principal components showed how a normal skull transformed into its predicted syndromic phenotype (fig 7). The shape of the orbit changed from a normal orientation to a wider laterally rotated ‘butterfly’ shape. Retrusion and shortening of the midface occurs while moving along the vector from normal to syndromic. This is made obvious by the elevation of the nasal floor towards the inferior orbital rim in a more syndromic phenotype. The forehead becomes more turricephalic as normal moves to syndromic.
10
ACCEPTED MANUSCRIPT
Linear measurements Then the ‘average normal’ vector was applied to all of the 27 syndromic patients to show their unique normalized skulls, an example is shown in figure 8. Linear measurements
RI PT
were taken from all syndromic scans as described earlier and compared to the normative control data (tables 4 and 5). In three measurements the mean values were out of the normal range in both age groups. After normalization, only zygomatic arch
SC
distance was outside normal boundaries in the younger group. The intertemporal
distance did not reach normal boundaries in the other group. However, the model drove
M AN U
the values closer to normal boundaries and this was true for both the zygomatic arch distance and the interzygomatic distance (table 4).
Discussion
With geometric morphometrics the skull is analysed as a whole instead of using multiple
TE D
linear measurements to compare it to an average normal skull. Although based on mathematical principals this study allows a holistic approach that is visual and anatomical. We find that PCA is complementary to conventional methods in describing
EP
differences in skull shape and it enables visualization of the skull as a whole. It is superior to traditional methods in a way that PCA uses all the available data instead of
AC C
performing linear and angular measurements on specific slices of a CT scan. Furthermore, with PCA the skull is compared to a unique skull shape within normal boundaries, rather than to average values taken from samples of a normal population. For further application normalized 3D values need to be defined in place of the
traditional cephalometrics. Analysing the intra-observer errors confirmed that the chosen landmarks are reproducible.
11
ACCEPTED MANUSCRIPT
The PCA showed the variation within the populations and could calculate a mathematically accurate mean shape. Visualization of these changes was possible with TPS movies, as seen previously in the analysis of Noonan(Hammond, Hutton et al.
RI PT
2004) and Apert(Pluijmers, Ponniah et al. 2012) syndrome. PCA is purely mathematical which means there is no focus on specific anatomical features. However, the observed allometric growth in the first component of the normal population shows that the model
SC
appears to mirror what we know about growth in the normal population. The second component displays long thin faces and short wide faces. The syndromic model
M AN U
expressed ‘Crouzonness’ in the first component and in the second component primarily allometric growth was shown. This demonstrated that variability in craniofacial phenotype has more effect than changes in age in this age group. The model demonstrated the deformities of Crouzon-Pfeiffer syndrome and defined a vector (composite of multiple 3D vectors) between the populations. This
TE D
allowed transforming a syndromic to its unique normalized skull. The calculated vector can either create a syndromic skull from a normal patient or normalize a syndromic skull, depending on its direction. Given the large phenotypic variation of Crouzon-Pfeiffer
EP
syndrome the ‘average normal’ vector might not be sufficient for all patients to produce a skull within normal boundaries. A skull with a severe phenotype will need to be moved
AC C
further along the normal vector to appear normal than a mild one. For example, a mild syndromic patient needs 70% of the vector to get within the normal boundaries, whereas more severe patients may need 120% along the vector, with 100% representing the average. Previous work showed that PCA was successful in Apert syndrome which has less genetic variation compared to Crouzon-Pfeiffer syndrome.(Pluijmers, Ponniah et al. 2012) Despite the large genetic variation in these syndromes, the model appears to be capable of capturing the differences accurately.
12
ACCEPTED MANUSCRIPT
By using TPS warping it appears that the model described Crouzon-Pfeiffer malformations, which could be used to plan and refine surgical correction for CrouzonPfeiffer syndrome. In order to validate our model we normalised a syndromic skull and
RI PT
compared this to the established normal reference data by Waitzman et al.(Waitzman, Posnick et al. 1992) (table 4 and 5). The lateral orbital distance and intertemporal distance were larger than normal, which is in agreement with clinically observed
SC
hypertelorism. The length of the zygomatic arch was smaller than normal, representing the clinically observed midfacial retrusion. These results concur with findings of a
M AN U
previous study on 1 year old patients with Crouzon syndrome.(Carr, Posnick et al. 1992) The measurements confirmed that our model can normalize a specific skull and shows how the skull needs to be changed. This can be used in cephalometric surgical planning. By normalizing the skull of a patient and visualizing the differences in a colour-coded map, it can show where to place the osteotomies or bony onlays in an individual patient.
TE D
A limitation of this study is that it mainly analyses frontofacial region and ignores several craniofacial regions (i.e. mandible, cranial base and neurocranium). As we excluded al patients with any history of facial surgery this could bias the phenotype
EP
towards the less severe. Furthermore, the age range is limited, but the cranio-orbitozygomatic skeleton is at 85% of its eventual adult size by 5 years of age and surgical
AC C
intervention should preferably be done on children after this age.(Waitzman, Posnick et al. 1992) When there is too much variability between the sizes of the skulls due to growth, this variation can overshadow the subtle changes and the PCA analysis will conclude that growth is the biggest change in shape. It is important to limit variation such as age and therefore only patients between ages 6 to 13 were included. The demographics of the skulls used to model normal variation should correspond to the patients’ demographics. Ideally they would be matched for ethnicity and sex. The sample size of syndromic patients was small as the data was collected from only one hospital
13
ACCEPTED MANUSCRIPT
(Great Ormond Street Hospital) in the United Kingdom. However, the small numbers reflect the rarity of Crouzon and Pfeiffer syndrome.(Renier, Lajeunie et al. 2000; Kabbani and Talkad 2004; Ridgway and Weiner 2004) We suggest a more elaborate study with
RI PT
collaboration between national and international craniofacial centres to increase the numbers and therefore the sensitivity of PCA. Potential sources of normal paediatric CTscans would be radiotherapy planning scans and trauma cohorts. The advantage of the
SC
radiotherapy patients is that full diagnoses and demographics are available. The
problem with trauma cases is that it is difficult to get scans of a complete and unaffected
M AN U
facial skeleton.
Furthermore, the outcome of craniofacial surgery can be measured by comparing the normalized skull with the actual postoperative scan.
Conclusion
TE D
Using PCA to describe a syndrome is a promising method to analyse variability in several syndromic craniosynostosis. The virtual normalization of a Crouzon-Pfeiffer skull is useful to understand the phenotypic changes needed and can help us with planning
EP
reconstructive surgery. This method has great potential for measuring outcomes in
AC C
craniofacial surgery.
14
ACCEPTED MANUSCRIPT
References
AC C
EP
TE D
M AN U
SC
RI PT
Bookstein, F. L. (1997). "Shape and the Information in Medical Images: A Decade of the Morphometric Synthesis." Computer Vision And Image Understanding 66(2): 97-118. Carinci, F., F. Pezzetti, et al. (2005). "Apert and Crouzon Syndromes- Clinical Findings, Genes and Extracellular Matrix." The Journal Of Craniofacial Surgery 16(3): 361368. Carr, M., J. C. Posnick, et al. (1992). "Cranio-orbito-zygomatic measurements from standard CT scans in unoperated Crouzon and Apert infants: comparison with normal controls." Cleft Palate-craniofacial journal 29(2): 129-136. Crouzon, M. O. (1912). "Dysostose cranio-faciale héréditaire." ulletin de la Société des Médecins des Hôpitaux de Paris 33(4): 545–555. Cunningham, M. L., M. L. Seto, et al. (2007). "Syndromic craniosynostosis: from history to hydrogen bonds." Orthod Craniofacial Res 10: 67-81. Farkas, L. G., M. J. Katic, et al. (2005). "International Anthropometric Study of Facial Morphology in Various Ethnic Groups/Races." The Journal Of Craniofacial Surgery 16(4): 615-646. Hammond, P., T. J. Hutton, et al. (2004). "3D analysis of facial morphology." Am J Med Genet A 126A(4): 339-348. Kabbani, H. and S. R. Talkad (2004). "Craniosynostosis." American Family Physician 69(12): 2863-2870. Kreiborg, S., J. L. Marsh, et al. (1993). "Comparative three-dimensional analysis of CTscans of the calvaria and cranial base in Apert and Crouzon syndromes." Journal of Cranio-Maxillo-Facial Surgery 21: 181-188. Lajeunie, E., S. Heuertz, et al. (2006). "Mutation screening in patients with syndromic craniosynostoses indicates that a limited number of recurrent FGFR2 mutations accounts for severe forms of Pfeiffer syndrome." Eur J Hum Genet 14(3): 289298. Meyers, G. A., D. Day, et al. (1996). "FGFR2 Exon Ilia and IlIc Mutations in Crouzon, Jackson-Weiss, and Pfeiffer Syndromes- Evidence for Missense Changes, Insertions, and a Deletion Due to Alternative RNA Splicing." Am. J. Hum. Genet. 58: 491-498. O'Higgins, P. (2000). "The study of morphological variation in the hominid fossil recordbiology, landmarks and geometry." J. Anat. 197: 103-120. Pfeiffer, R. A. (1964). "Dominant erbliche akrocephalosyndaktylie." Zeitschrift fur Kinderheilkunde 90: 301-320. Pluijmers, B. I., A. J. T. Ponniah, et al. (2012). "Using principal component analysis to describe the Apert skull deformity and simulate its correction." Journal of Plastic, Reconstructive and Aesthetic Surgeons. Posnick, J. C., K. Y. Lin, et al. (1993). "Crouzon syndrome Quantitative assessment of presenting deformity and surgical results based on CT scans." Plastic and Reconstructive Surgery 92(6): 1027-1037. Renier, D., E. Lajeunie, et al. (2000). "Management of craniosynostosis." Child's Nerv Syst 16: 645-658. Richtsmeier, J. T. (1987). "Comparative Study of Normal, Crouzon, and Apert Craniofacial Morphology Using Finite Element Scaling Analysis." American journal of Physical Anthropology 74: 473-493. Ridgway, E. B. and H. L. Weiner (2004). "Skull deformities." Pediatr Clin North Am 51(2): 359-387.
15
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Rutland, P., L. J. Pulleyn, et al. (1995). "Identical mutions in the FGFR2 cause both Pfeiffer and Crouzon syndrome phenotypes." Nature Genetics 9: 173-176. Shapiro, D. and J. T. Richtsmeier (1997). "Brief communication: A sample of pediatric Skulls Available for study." American journal of Physical Anthropology 103: 415416. Steinberger, D., T. Reinhartz, et al. (1996). "FGFR2 mutation in clinically nonclassifiable autosomal dominant craniosynostosis with pronounced phenotypic variation." American Journal of Medical Genetics 66: 81-86. Vogels, A. and J. P. Fryns (2006). "Pfeiffer syndrome." Orphanet Journal Of Rare Diseases 1. Waitzman, A. A., J. C. Posnick, et al. (1992). "Craniofacial skeletal measurements based on computed tomography: Part II. Normal values and growth trends." Cleft Palate-craniofacial journal 29(2): 118-128.
16
ACCEPTED MANUSCRIPT
Figure legend
Figure 1 Colour-coded map showing the predicted (warped) skull compared to its actual
RI PT
counterpart. Frontal and 30-degree view showing positive and negative surface differences. Areas of light blue and green show good correspondence between the two scans, showing that the landmarks capture most of the skull shape.
SC
Figure 2 Landmark projected on the skull in frontal (left) and 30-degree (right) view, showing how the landmarks from table 1 are placed on the skull.
M AN U
Figure 3 First principal component of the normal group in frontal (top) and 30-degree (bottom) view. From left to right: minus two SD and plus two SD, showing allometric growth within the age range of 6 to 13 years.
Figure 4 Second principal component of the normal group in frontal (top) and 30 degree (bottom) view. From left to right: minus two SD and plus two SD, showing how short
TE D
faces tend to be wider (left) and long faces tend to be thinner (right) in the way they vary in relation to each other
Figure 5 First principal component of the syndromic group in frontal (top) and 30-degree
EP
(bottom) view. From left to right: minus two SD and plus two SD, showing the shape variation of the frontofacial region within the syndromic group. On the left: a tall
AC C
forehead, retruded and short midface. On the right: a shorter forehead with a less retruted and taller midface. Figure 6 Second principal component of the syndromic group in frontal (top) and 30degree (bottom) view. From left to right: minus two SD and plus two SD, showing allometric growth within the age range of 6 to 13. Figure 7 Normal skull transforming into a syndromic skull using the (PCA) model. From left to right: normal skull transforming into its (unique) predicted syndromic skull. Figure 8 Crouzon skull transforming into a normalized one using the (PCA) model. From
17
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
left to right: Crouzon skull transforming into its (unique) predicted normalized skull.
18
ACCEPTED MANUSCRIPT
Table legend
Table 1 Set of 66 landmarks used in this study.
RI PT
Table 2 Characters in the first column correspond to the landmarkset used in this study. The second and third column represent the intra-rater reliability. Table 3 Age and sex distribution of the syndromic population.
SC
Table 4 Measurements in the orbital and upper midface (zygomatic) regions for the syndromic and normal subjects between 6 and 9 years of age.
M AN U
Table 5 Measurements in the orbital and upper midface (zygomatic) regions for the
AC C
EP
TE D
syndromic and normal subjects between 10 and 13 years of age.
19
Label Landmark
Definition
Scan orientation
AB
Orbitale
The lowest part of the orbital margin. It is used to define the Frankfort Plane and to measure orbital height.
Frontal
CD
Superior ‘orbitale’
Exact point on inner superior orbital rim vertically above ‘orbitale’
Frontal
EF
Frontomalare anterior
Most anterior point on the fronto-malar suture Frontal
GH
Supraorbital notch
Most superior point of the supraorbital notch
IJ
Infraorbital foramen
Most superior point of the infraorbital foramen Frontal
KL
Infraorbital foramen / orbital rim
Point on the inner orbital rim directly above infraorbital foramen
Frontal
MN
Ectoconchion
The intersection of the most anterior surface of the lateral border of the orbit and a line bisecting the orbit along its long axis. To mark Ectoconchion, move a toothpick or other thin straight instrument up and down, keeping it parallel to the superior orbital border, until you divide the eye orbit into two equal halves. Mark the point on the anterior orbital margin
Frontal
OP
Superior orbital fissure The most superior lateral point of the superior Frontal orbital fissure
Q
Nasion
The point of intersection of the nasofrontal suture and the mid-sagittal plane. Nasion corresponds to the nasal root
RS
Nasomaxillary suture pinch
Narrowest portion of the midline of the face to Rotate 30 left/right the naso-maxillary suture.
TU
Dacryon
V
Inferior part of nasal bone
W
Anterior nasal spine
X
AC C
ACCEPTED MANUSCRIPT
YZ A1
TE D
M AN U
SC
RI PT
Rotate 20 left/right
Frontal
Rotate 10-20 left/right
The most inferior point of the junction between the nasal base in the midsaggital plane. The apex of the anterior nasal spine
20-30 chin down
EP
The point on the medial border of the orbit at which the frontal, lacrimal, and maxilla intersect. In other words, Dacryon lies at the intersection of the lacrimomaxillary suture and the frontal bone.
40 chin down
Subspinale
The deepest point of the profile below the anterior nasal spine.
Frontal
Alare
The most laterally positioned point on the anterior margin of the nasal aperture
Frontal
Prosthion
Most anterior midline point on the alveolar process of the maxilla between the central incissors.
Frontal
B1C1 Interdentale superior lateral1 (right + left)
Most anterior point on the alveolar process of Frontal the maxilla between the central incisor and the lateral incisors. (Between the 41,42 (right) and the 31,32 (left))
ACCEPTED MANUSCRIPT Most anterior point on the alveolar process of Rotate 20 left/right the maxilla between the lateral incisor and the canine. (Between the 12,13 (right) and the 22,23 (left))
F1G1 Maxillary angle
Most superior medial point on the curve in the maxilla from a frontal view
Frontal
H1I1
Ectomalare
Frontal and check 90 degree left/right
J1K1
Infraorbital foramen to alveolus
On the maxilla: positioned at the most lateral point on the lateral surface of the alveolar crest. Found along the second molar on the maxilla. Middle of the straight line down from the infraorbital foramen (GH) to the dentoalveolar junction
RI PT
D1E1 Interdentale superior lateral2 (right + left)
Frontal
Middle of the straight line from the infraorbital Frontal foramen (GH) to the maxillary angle (U2V2)
N1O1 Zygomaxillare
Intersection of zygomaxilllary suture and most medial masseter muscle attachment.
Rotate 20-40 left/right
P1Q1 Zygomatic angle
Angle between zygomatic ridge and orbital portion of the zygoma
Rotate 70-90 left/right
R1S1 Glenoid fossa
Most superior point of the glenoid fossa
Rotate 90 left/right
T1U1 Upper part of half zygomatic process (right + left)
Upper part in the middle of the glenoid fossa and zygomatic angle
Rotate 90 left/right
V1W1 Lower part of half zygomatic process (right + left)
Lower part in the middle of the glenoid fossa and zygomatic angle
Rotate 90 left/right
X1Y1 Porion
Highest point on the external auditory meatus Rotate 80-90 left/right
Zygomaticofrontal suture
B2C2 Mastoidale
F2G2 Pterion
Glabella
M AN U
AC C
H2
Most lateral point on the fronto-malar suture.
Rotate 100-140 left/right
The inferior tip of the mastoid process
Robins 3D
Most superior point of superior temporal line. (top of the lateral orbit bone)
Rotate 30-40 left/right
Region/point on the upper end of the greater wing of the sphenoid. Junction of the sphenoid, temporal, parietal and frontal sutures. The most forwardly projecting point in the mid-sagittal plane at the lower margin of the frontal bone, which lies above the nasal root and between the superciliary arches. Note that in juvenile skulls with strongly forwardly vaulted foreheads, the most projecting point of the curve of the forehead is not that of Glabella.
Robins 3D
EP
D2E2 Frontotemporale
TE D
Z1A2
SC
L1M1 Infraorbital foramen to maxillary angle
Frontal
I2
Bregma
The point where the sagittal and coronal sutures meet
Rotate 70 chin down.
J2
Lambda
The point where the two branches of the lambdoidal suture meet with the sagittal suture
Rotate 180 left/right, place the mark and 20 chinup to get the right angle for the mark
ACCEPTED MANUSCRIPT K2L2 Infraorbital foramen to frontal bone
Frontal: ‘Ortho reslice’ in Robins 3D
Frontal: ‘Ortho reslice’ in Robins 3D
AC C
EP
TE D
M AN U
SC
RI PT
M2N2 Lateral orbit to forehead
The most ventral part of the frontal bone (NOT part of the orbit what might be more ventral) on the straight line from the infraorbital foramen The most ventral part of the frontal bone on the straight line from the lateral part of the orbital rim
ACCEPTED MANUSCRIPT 0.91
B
0.945
1.002
C
1.086
1.118
D
0.935
1.189
E
0.768
0.564
F
0.368
0.646
G
0.374
0.369
H
0.434
0.517
I
0.412
0.666
J
0.354
0.497
K
0.615
0.664
L
0.473
0.432
M
1.106
1.264
N
0.769
0.878
O
0.436
0.425
P
0.458
0.527
Q
0.961
0.317
R
0.538
0.716
S
0.473
0.611
T
0.797
1.11
U
1.168
0.882
V
0.435
0.57
0.449
X
0.618
Y
0.462
Z
1.164
A1
0.364
B1
0.455
C1
0.49
0.479
D1
0.54
0.531
0.509
0.451
0.381 0.472 0.621 0.474
EP
0.812
J1
AC C
E1
0.46
TE D
W
0.597
1.199
K1
0.818
0.949
L1
0.487
0.721
M1
0.584
1.761
N1
1.764
0.956
O1
1.976
0.594
P1
0.643
0.403
Q1
0.786
0.425
F1 G1 H1 I1
SC
1.005
M AN U
A
RI PT
Label SD normals SD syndromic
0.576
0.363
0.682
0.344
2.105
0.648
1.633
0.921
0.691
S1
1.284
0.815
T1
0.957
0.635
U1
0.537
0.74
V1
1.001
0.792
W1
0.586
0.7
X1
1.612
2.269
0.788
0.43
0.782
1.329
A2
1.024
1.172
B2
0.523
0.512
C2
0.503
0.396
D2
0.739
0.489
E2
1.126
0.966
F2
2.596
2.296
G2
2.249
2.656
H2
1.165
1.035
I2
0.594
1.017
J2
2.123
1.395
K2
1.467
1.835
L2
1.621
2.531
M2
1.626
2.88
N2
1.801
2.955
AC C
EP
TE D
Y1 Z1
SC
0.878
M AN U
R1
RI PT
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 3: Age and sex distribution of the syndromic population Age in years Females Males Total 6 4 3 7 7 1 1 2 8 1 2 3 9 2 1 3 10 1 0 1 11 5 0 5 12 1 3 4 13 2 0 2 Total 17 10 27
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 4: Measurements in the orbital and upper midface (zygomatic) regions for the syndromic and normal Subjects between 6 and 9 years of age Syndromic Normalized syndromic Normal Controls Variable Mean (mm) SD Mean (mm) SD Mean (mm) SD Anterior interorbital distance 21.6 3.1 20.1 1.9 21.8 2.8 Lateral orbital distance 92.4 4.8 83.7 4.2 87.3 4.7 Intertemporal distance 85.3 7.4 77.9 6.7 72.7 6.0 Interzygomatic-buttress distance 80.2 6.2 78.8 5.2 83.0 4.7 Interzygomatic-arch distance 105.7 5.5 105.5 6.3 106.9 10.8 Zygomatic-arch length 43.0 5.9 46.4 6.3 51.0 3.7
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 5: Measurements in the orbital and upper midface (zygomatic) regions for syndromic and normal Subjects between 10 and 13 years of age Syndromic Normalized syndromic Normal Controls Variable Mean (mm) SD Mean (mm) SD Mean (mm) SD Anterior interorbital distance 23.4 4.3 20.9 1.8 22.3 1.8 Lateral orbital distance 100.1 3.8 90.9 3.7 90.3 4.3 Intertemporal distance 91.2 7.3 83.7 6.6 74.6 4.8 Interzygomatic-buttress distance 86.0 5.3 84.0 6.1 85.1 5.5 Interzygomatic-arch distance 115.0 5.5 114.4 5.9 112.1 7.1 Zygomatic-arch length 43.0 6.3 49.1 6.9 55.2 2.4
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
S1010-5182(15)00027-X
DOI:
10.1016/j.jcms.2015.02.005
Reference:
YJCMS 1967
To appear in:
Journal of Cranio-Maxillo-Facial Surgery
Received Date: 1 January 2015 Accepted Date: 6 February 2015
Please cite this article as: Staal FCR, Ponniah AJT, Angullia F, Ruff C, Koudstaal MJ, Dunaway D, Describing Crouzon and Pfeiffer syndrome based on principal component analysis, Journal of CranioMaxillofacial Surgery (2015), doi: 10.1016/j.jcms.2015.02.005. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Describing Crouzon and Pfeiffer syndrome based on principal component analysis Femke C.R. Staal, B.Sc., Great Ormond Street Hospital, London, United Kingdom
RI PT
Allan J.T. Ponniah, M.R.C.S. (Eng.), M.Sc., Great Ormond Street Hospital, London, United Kingdom
Freida Angullia, B.Sc., M.D., M.R.C.S. (Eng.), Great Ormond Street Hospital, London, United Kingdom
Clifford Ruff, B.Sc., M.Sc., Medical Physics Department, University college London,
SC
London, United Kingdom
M AN U
Maarten J. Koudstaal, M.D., D.D.S., PhD, Erasmus Medical Center, Maxillofacial Surgery, Rotterdam, The Netherlands David Dunaway, F.D.S., R.C.S FRCS (plast.), Great Ormond Street Hospital, London,
AC C
EP
TE D
United Kingdom
1
ACCEPTED MANUSCRIPT
Corresponding author Femke C.R. Staal [email protected] Tel: +31 (0)6 253 30 443
RI PT
Craniofacial department FAO David Dunaway Great Ormond Street Hospital Great Ormond Street London WC1N 3JH
AC C
EP
TE D
M AN U
SC
Fax: +44 (0) 20 7813 8446
2
ACCEPTED MANUSCRIPT
Financial disclosure None of the authors has a financial interest in any of the products, devices, or drugs
AC C
EP
TE D
M AN U
SC
RI PT
mentioned in this article.
3
ACCEPTED MANUSCRIPT
Abstract Crouzon and Pfeiffer syndrome are syndromic craniosynostosis caused by specific mutations in the FGFR genes. Patients share the characteristics of a tall, flattened
RI PT
forehead, exorbitism, hypertelorism, maxillary hypoplasia and mandibular prognathism. Geometric morphometrics allows the identification of the global shape changes within and between the normal and syndromic population. Methods: Data from 27 Crouzon-
SC
Pfeiffer and 33 normal subjects were landmarked in order to compare both populations. With principal component analysis the variation within both groups was visualized and
M AN U
the vector of change was calculated. This model normalized a Crouzon-Pfeiffer skull and was compared to age-matched normative control data. Results: PCA defined a vector that described the shape changes between both populations. Movies showed how the normal skull transformed into a Crouzon-Pfeiffer phenotype and vice versa. Comparing these results to established age-matched normal control data confirmed that our model
TE D
could normalize a Crouzon-Pfeiffer skull. Conclusions: PCA was able to describe deformities associated with Crouzon-Pfeiffer syndrome and is a promising method to analyze variability in syndromic craniosynostosis. The virtual normalization of a Crouzon-
EP
Pfeiffer skull is useful to delineate the phenotypic changes required for correction, can help surgeons plan reconstructive surgery and is a potentially promising surgical
AC C
outcome measure.
Keywords: Crouzon syndrome; Pfeiffer syndrome; Principal component analysis; Craniosynostosis; FGFR2; Three-dimensional imaging
4
ACCEPTED MANUSCRIPT
Introduction Crouzon and Pfeiffer syndrome are autosomal dominant disorders associated with multisutural craniosynostosis.(Crouzon 1912; Pfeiffer 1964) Both syndromes have a
RI PT
large variation in phenotype in both the cranial and facial features.(Carinci, Pezzetti et al. 2005; Vogels and Fryns 2006) Patients share the characteristics of a tall, flattened
forehead (due to bicoronal synostosis), exorbitism, hypertelorism, maxillary hypoplasia,
SC
and relative mandibular prognatism.(Crouzon 1912; Renier, Lajeunie et al. 2000;
Lajeunie, Heuertz et al. 2006; Vogels and Fryns 2006; Cunningham, Seto et al. 2007)
M AN U
These anatomical abnormalities can contribute to functional problems including increased intracranial pressure (ICP), upper airway obstruction and ocular problems.(Renier, Lajeunie et al. 2000; Vogels and Fryns 2006) Some argue that Crouzon and Pfeiffer syndromes are part of the same entity as both are associated with FGFR2 mutations. This suggests that the same mutation can produce different
TE D
phenotypes.(Rutland, Pulleyn et al. 1995; Meyers, Day et al. 1996; Steinberger, Reinhartz et al. 1996; Lajeunie, Heuertz et al. 2006; Cunningham, Seto et al. 2007) Pfeiffer syndrome has therefore been included in this study.
EP
Previous methods used to analyse these complex deformities include clinical descriptions(Crouzon 1912; Renier, Lajeunie et al. 2000; Carinci, Pezzetti et al. 2005;
AC C
Lajeunie, Heuertz et al. 2006; Vogels and Fryns 2006; Cunningham, Seto et al. 2007), linear morphometrics(Richtsmeier 1987; Posnick, Lin et al. 1993; Farkas, Katic et al. 2005) and 3D-CT analysis.(Carr, Posnick et al. 1992; Kreiborg, Marsh et al. 1993) Most cephalometric methods analyse two-dimensional coordinates of a three-dimensional shape. Conventional 3D-CT analysis describes individual distances and angles between landmarks, rather than the skull shape as a whole unit. A more recent technique to analyse the size and shape of objects is geometric morphometrics.(Bookstein 1997) It has been used successfully in the analysis of craniofacial shapes, for example the face
5
ACCEPTED MANUSCRIPT
variation during growth in primates.(O'Higgins 2000) Principal component analysis (PCA) can describe the complex shape of skulls as a whole and allows for shape comparison between different skulls.
RI PT
The purpose of this study is to characterize the anatomical abnormalities in Crouzon and Pfeiffer skulls using PCA with the intention to develop a process that can
SC
describe how the Crouzon-Pfeiffer skull varies from the unaffected population.
Methods
M AN U
Data collection
Data was collected from a series of 27 Crouzon-Pfeiffer patients, 21 clinically diagnosed with Crouzon syndrome and 6 with Pfeiffer (genetic diagnosis was not available for all patients). The inclusion criteria were patients between ages 6 to 13 with no history of facial surgery. Patients with previous posterior cranial vault remodelling were included,
TE D
but all patients with any history of facial surgery were excluded as this would distort the landmarks. Anatomical normal pediatric data (n=33) was collected from a series epileptic patients undergoing CT scans for surgical planning and two historic collections of skulls:
EP
the Bosma collection (John Hopkins) (Shapiro and Richtsmeier 1997) and the Natural History Museum (London). Inclusion criteria were patients with an unaffected facial
AC C
skeleton, between ages 6 to 13. The scans were taken using a 16 slice Siemens somatom sensation spiral CT scanner set to 0,75 collimation (Siemens Medical Solutions, Malvern, Pa.) and saved as Digital Imaging and Communications in Medicine (DICOM) files. The DICOM files were then converted into a University College London (UCL) proprietary format and loaded into 3D voxel imaging software (Robin 3D 2006). The normal and patient CT images were thresholded for analysis of the hard tissue surface. Polygon mesh surfaces (stl) representing bone, were extracted from all scans for landmark placement.
6
ACCEPTED MANUSCRIPT
Landmarks The 60 scans were landmarked using 3D voxel imaging software (Robin 3D 2006). This
RI PT
software allowed visualization of changes between two scans by creating a thin-plate spline warp. In order to compare normal and patient scans a reliable set of homologous landmarks had to be determined. Landmarks were mainly placed on anatomical points
SC
including foramina, ridges and intersections of sutures for reliability and
repeatability.(Bookstein 1997; O'Higgins 2000) The identification of our landmark set
M AN U
was an iterative process of testing different points. Landmarks capture only the information of specific points and the space between them is interpolated. Therefore, a sufficient number of landmarks were needed to define all of the different features of the skulls. Most landmarks were placed on the frontofacial region where most surgical change occurs with fewer landmarks placed on the back of the skull.
TE D
The first 56 landmarks, which were based on an earlier study(Pluijmers, Ponniah et al. 2012), were located on a test data set using the 3D imaging software. The surface of the skull was then warped to the position of corresponding landmarks of a different
EP
skull in order to show the discrepancies between the skull shapes. The warped scan was superimposed to its actual counterpart and differences in surface were visualized in a
AC C
colour-coded map. This colour map located regions with varying degrees of discrepancy between the two skull surfaces (fig 1). If the landmarks captured all of the skull shape, no differences in surface would appear on the colour map. The landmarks were adjusted in these regions to ensure reliability and repeatability. Some landmarks were removed because they did not contribute to capturing the skull shape. Additional landmarks were placed to improve data capture. 66 landmarks were defined which described the frontofacial skeleton optimally. These landmarks allowed for the best possible comparison between the patient and normal scans (table 1, fig 2). The mandible was
7
ACCEPTED MANUSCRIPT
excluded from analysis because of variation in position and there was a low availability of scans of the whole facial skeleton in the normal group.
RI PT
Data analysis To determine the repeatability of the landmarks, a normal skull and a syndromic skull
were chosen at random and were landmarked ten times in different sittings, making it
SC
possible to obtain the intra-observer reliability.
As Crouzon and Pfeiffer syndrome are both associated with FGFR2 mutations,
M AN U
they are analysed as one group and the syndromes are not separately analysed. PCA is a statistical method that can extract the variations in shape within a population. Eigenvectors can be extracted from the spacial coordinates of the landmarks and these are the principal components shape variation.(O'Higgins 2000) PCA can describe a large amount of high dimensional data with a smaller number of relevant variables, the
TE D
principal components. Therefore the deformities in the Crouzon-Pfeiffer skull can be captured as a whole instead of comparing single measurements. The first principal component describes the largest variation within the population. The thin-plate spline
EP
(TPS) technique can interpolate changes between landmarks and uses minimum bending energy to estimate the surface between landmarks.(Bookstein 1997) We used
AC C
this method to visualize the changes and create movies that showed the variation within and between the populations. To validate the observations of the model, the normalising eigenvector was
applied to all patients, and six measurements were taken before and after. The selected bony measurements assess the orbital and upper midface (zygomatic) regions in the horizontal planes as previously described.(Waitzman, Posnick et al. 1992) To make equivalent measurements after normalizing the skull, 3D cuts were made to simulate two comparable CT slices. All patients were positioned in the orbitomeatal plane and a cut
8
ACCEPTED MANUSCRIPT
was made in the orbital region and a second cut in the upper midface (zygomatic) region. The 3D imaging software was used to perform all measurements and the data was compared to age-matched normative control data.(Waitzman, Posnick et al. 1992)
groups: 6 to 9, and 10 to 13 years of age.
SC
Results
RI PT
The means and standard deviations of the normative control data were split in two
Landmark reliability
M AN U
For each landmark the standard deviation (SD) was calculated (table 2). All landmarks used were within a SD value of 3 mm. Normal population
5 out of the 66 landmarks were outside of the 2 mm limit. 18 landmarks were between 1 mm and 2 mm and 43 landmarks had an SD < 1 mm. Thus, 43/66 (65%) of the
Patient population
TE D
landmarks placed were considered highly accurate and 92% were within 2 mm limit.
6 out of the 66 landmarks were outside of the 2 mm limit. 13 landmarks were between 1
EP
mm and 2 mm and 47 landmarks had an SD of < 1 mm. Thus, 47/66 (71%) of the
AC C
landmarks were considered highly accurate and 91% within the 2 mm limit.
Points placed around more distinct anatomical features like the infraorbital foramen had higher accuracy than points placed on curvatures (table 2). This was true for both normal and syndromic skulls. All of the landmarks have shown to be reliable to locate on the different skull shapes. Further points were considered, but were found to be unreliable and therefore excluded from the PCA.
Variation within the populations
9
ACCEPTED MANUSCRIPT
The PCA on each group defined the variability within the groups. The first principal component of variation in the normal group showed allometric growth within the age range of 6 to 13 years (fig 3).
RI PT
The second component shows how long faces tend to be thinner and short faces tend to be wider in the way they vary in relation to each other. Both forehead and maxilla contribute equally to the variation in lengthening and widening of the skull. There is little
SC
variation in the shape and position of the orbit (fig 4). Table 3 shows the age and sex distribution of the syndromic population. The first mode of variation in the syndromic
M AN U
group displayed the variation in the slope and height of the forehead, midfacial height and retrusion and variation in the orbital shape and orientation (fig 5). The second component primarily shows allometric growth within the age range of 6 to 13. There is little variation in the forehead, maxilla or orbital shape compared to the first component
TE D
(fig 6).
Variation between the populations
PCA was also performed between normal and syndromic subjects resulting in an
EP
‘average’ vector between all the warps. This averaged vector describes shape changes between the syndromic and normal skulls and represents a model for normalization of a
AC C
Crouzon-Pfeiffer skull. Resulting thin-plate spline movies of the principal components showed how a normal skull transformed into its predicted syndromic phenotype (fig 7). The shape of the orbit changed from a normal orientation to a wider laterally rotated ‘butterfly’ shape. Retrusion and shortening of the midface occurs while moving along the vector from normal to syndromic. This is made obvious by the elevation of the nasal floor towards the inferior orbital rim in a more syndromic phenotype. The forehead becomes more turricephalic as normal moves to syndromic.
10
ACCEPTED MANUSCRIPT
Linear measurements Then the ‘average normal’ vector was applied to all of the 27 syndromic patients to show their unique normalized skulls, an example is shown in figure 8. Linear measurements
RI PT
were taken from all syndromic scans as described earlier and compared to the normative control data (tables 4 and 5). In three measurements the mean values were out of the normal range in both age groups. After normalization, only zygomatic arch
SC
distance was outside normal boundaries in the younger group. The intertemporal
distance did not reach normal boundaries in the other group. However, the model drove
M AN U
the values closer to normal boundaries and this was true for both the zygomatic arch distance and the interzygomatic distance (table 4).
Discussion
With geometric morphometrics the skull is analysed as a whole instead of using multiple
TE D
linear measurements to compare it to an average normal skull. Although based on mathematical principals this study allows a holistic approach that is visual and anatomical. We find that PCA is complementary to conventional methods in describing
EP
differences in skull shape and it enables visualization of the skull as a whole. It is superior to traditional methods in a way that PCA uses all the available data instead of
AC C
performing linear and angular measurements on specific slices of a CT scan. Furthermore, with PCA the skull is compared to a unique skull shape within normal boundaries, rather than to average values taken from samples of a normal population. For further application normalized 3D values need to be defined in place of the
traditional cephalometrics. Analysing the intra-observer errors confirmed that the chosen landmarks are reproducible.
11
ACCEPTED MANUSCRIPT
The PCA showed the variation within the populations and could calculate a mathematically accurate mean shape. Visualization of these changes was possible with TPS movies, as seen previously in the analysis of Noonan(Hammond, Hutton et al.
RI PT
2004) and Apert(Pluijmers, Ponniah et al. 2012) syndrome. PCA is purely mathematical which means there is no focus on specific anatomical features. However, the observed allometric growth in the first component of the normal population shows that the model
SC
appears to mirror what we know about growth in the normal population. The second component displays long thin faces and short wide faces. The syndromic model
M AN U
expressed ‘Crouzonness’ in the first component and in the second component primarily allometric growth was shown. This demonstrated that variability in craniofacial phenotype has more effect than changes in age in this age group. The model demonstrated the deformities of Crouzon-Pfeiffer syndrome and defined a vector (composite of multiple 3D vectors) between the populations. This
TE D
allowed transforming a syndromic to its unique normalized skull. The calculated vector can either create a syndromic skull from a normal patient or normalize a syndromic skull, depending on its direction. Given the large phenotypic variation of Crouzon-Pfeiffer
EP
syndrome the ‘average normal’ vector might not be sufficient for all patients to produce a skull within normal boundaries. A skull with a severe phenotype will need to be moved
AC C
further along the normal vector to appear normal than a mild one. For example, a mild syndromic patient needs 70% of the vector to get within the normal boundaries, whereas more severe patients may need 120% along the vector, with 100% representing the average. Previous work showed that PCA was successful in Apert syndrome which has less genetic variation compared to Crouzon-Pfeiffer syndrome.(Pluijmers, Ponniah et al. 2012) Despite the large genetic variation in these syndromes, the model appears to be capable of capturing the differences accurately.
12
ACCEPTED MANUSCRIPT
By using TPS warping it appears that the model described Crouzon-Pfeiffer malformations, which could be used to plan and refine surgical correction for CrouzonPfeiffer syndrome. In order to validate our model we normalised a syndromic skull and
RI PT
compared this to the established normal reference data by Waitzman et al.(Waitzman, Posnick et al. 1992) (table 4 and 5). The lateral orbital distance and intertemporal distance were larger than normal, which is in agreement with clinically observed
SC
hypertelorism. The length of the zygomatic arch was smaller than normal, representing the clinically observed midfacial retrusion. These results concur with findings of a
M AN U
previous study on 1 year old patients with Crouzon syndrome.(Carr, Posnick et al. 1992) The measurements confirmed that our model can normalize a specific skull and shows how the skull needs to be changed. This can be used in cephalometric surgical planning. By normalizing the skull of a patient and visualizing the differences in a colour-coded map, it can show where to place the osteotomies or bony onlays in an individual patient.
TE D
A limitation of this study is that it mainly analyses frontofacial region and ignores several craniofacial regions (i.e. mandible, cranial base and neurocranium). As we excluded al patients with any history of facial surgery this could bias the phenotype
EP
towards the less severe. Furthermore, the age range is limited, but the cranio-orbitozygomatic skeleton is at 85% of its eventual adult size by 5 years of age and surgical
AC C
intervention should preferably be done on children after this age.(Waitzman, Posnick et al. 1992) When there is too much variability between the sizes of the skulls due to growth, this variation can overshadow the subtle changes and the PCA analysis will conclude that growth is the biggest change in shape. It is important to limit variation such as age and therefore only patients between ages 6 to 13 were included. The demographics of the skulls used to model normal variation should correspond to the patients’ demographics. Ideally they would be matched for ethnicity and sex. The sample size of syndromic patients was small as the data was collected from only one hospital
13
ACCEPTED MANUSCRIPT
(Great Ormond Street Hospital) in the United Kingdom. However, the small numbers reflect the rarity of Crouzon and Pfeiffer syndrome.(Renier, Lajeunie et al. 2000; Kabbani and Talkad 2004; Ridgway and Weiner 2004) We suggest a more elaborate study with
RI PT
collaboration between national and international craniofacial centres to increase the numbers and therefore the sensitivity of PCA. Potential sources of normal paediatric CTscans would be radiotherapy planning scans and trauma cohorts. The advantage of the
SC
radiotherapy patients is that full diagnoses and demographics are available. The
problem with trauma cases is that it is difficult to get scans of a complete and unaffected
M AN U
facial skeleton.
Furthermore, the outcome of craniofacial surgery can be measured by comparing the normalized skull with the actual postoperative scan.
Conclusion
TE D
Using PCA to describe a syndrome is a promising method to analyse variability in several syndromic craniosynostosis. The virtual normalization of a Crouzon-Pfeiffer skull is useful to understand the phenotypic changes needed and can help us with planning
EP
reconstructive surgery. This method has great potential for measuring outcomes in
AC C
craniofacial surgery.
14
ACCEPTED MANUSCRIPT
References
AC C
EP
TE D
M AN U
SC
RI PT
Bookstein, F. L. (1997). "Shape and the Information in Medical Images: A Decade of the Morphometric Synthesis." Computer Vision And Image Understanding 66(2): 97-118. Carinci, F., F. Pezzetti, et al. (2005). "Apert and Crouzon Syndromes- Clinical Findings, Genes and Extracellular Matrix." The Journal Of Craniofacial Surgery 16(3): 361368. Carr, M., J. C. Posnick, et al. (1992). "Cranio-orbito-zygomatic measurements from standard CT scans in unoperated Crouzon and Apert infants: comparison with normal controls." Cleft Palate-craniofacial journal 29(2): 129-136. Crouzon, M. O. (1912). "Dysostose cranio-faciale héréditaire." ulletin de la Société des Médecins des Hôpitaux de Paris 33(4): 545–555. Cunningham, M. L., M. L. Seto, et al. (2007). "Syndromic craniosynostosis: from history to hydrogen bonds." Orthod Craniofacial Res 10: 67-81. Farkas, L. G., M. J. Katic, et al. (2005). "International Anthropometric Study of Facial Morphology in Various Ethnic Groups/Races." The Journal Of Craniofacial Surgery 16(4): 615-646. Hammond, P., T. J. Hutton, et al. (2004). "3D analysis of facial morphology." Am J Med Genet A 126A(4): 339-348. Kabbani, H. and S. R. Talkad (2004). "Craniosynostosis." American Family Physician 69(12): 2863-2870. Kreiborg, S., J. L. Marsh, et al. (1993). "Comparative three-dimensional analysis of CTscans of the calvaria and cranial base in Apert and Crouzon syndromes." Journal of Cranio-Maxillo-Facial Surgery 21: 181-188. Lajeunie, E., S. Heuertz, et al. (2006). "Mutation screening in patients with syndromic craniosynostoses indicates that a limited number of recurrent FGFR2 mutations accounts for severe forms of Pfeiffer syndrome." Eur J Hum Genet 14(3): 289298. Meyers, G. A., D. Day, et al. (1996). "FGFR2 Exon Ilia and IlIc Mutations in Crouzon, Jackson-Weiss, and Pfeiffer Syndromes- Evidence for Missense Changes, Insertions, and a Deletion Due to Alternative RNA Splicing." Am. J. Hum. Genet. 58: 491-498. O'Higgins, P. (2000). "The study of morphological variation in the hominid fossil recordbiology, landmarks and geometry." J. Anat. 197: 103-120. Pfeiffer, R. A. (1964). "Dominant erbliche akrocephalosyndaktylie." Zeitschrift fur Kinderheilkunde 90: 301-320. Pluijmers, B. I., A. J. T. Ponniah, et al. (2012). "Using principal component analysis to describe the Apert skull deformity and simulate its correction." Journal of Plastic, Reconstructive and Aesthetic Surgeons. Posnick, J. C., K. Y. Lin, et al. (1993). "Crouzon syndrome Quantitative assessment of presenting deformity and surgical results based on CT scans." Plastic and Reconstructive Surgery 92(6): 1027-1037. Renier, D., E. Lajeunie, et al. (2000). "Management of craniosynostosis." Child's Nerv Syst 16: 645-658. Richtsmeier, J. T. (1987). "Comparative Study of Normal, Crouzon, and Apert Craniofacial Morphology Using Finite Element Scaling Analysis." American journal of Physical Anthropology 74: 473-493. Ridgway, E. B. and H. L. Weiner (2004). "Skull deformities." Pediatr Clin North Am 51(2): 359-387.
15
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Rutland, P., L. J. Pulleyn, et al. (1995). "Identical mutions in the FGFR2 cause both Pfeiffer and Crouzon syndrome phenotypes." Nature Genetics 9: 173-176. Shapiro, D. and J. T. Richtsmeier (1997). "Brief communication: A sample of pediatric Skulls Available for study." American journal of Physical Anthropology 103: 415416. Steinberger, D., T. Reinhartz, et al. (1996). "FGFR2 mutation in clinically nonclassifiable autosomal dominant craniosynostosis with pronounced phenotypic variation." American Journal of Medical Genetics 66: 81-86. Vogels, A. and J. P. Fryns (2006). "Pfeiffer syndrome." Orphanet Journal Of Rare Diseases 1. Waitzman, A. A., J. C. Posnick, et al. (1992). "Craniofacial skeletal measurements based on computed tomography: Part II. Normal values and growth trends." Cleft Palate-craniofacial journal 29(2): 118-128.
16
ACCEPTED MANUSCRIPT
Figure legend
Figure 1 Colour-coded map showing the predicted (warped) skull compared to its actual
RI PT
counterpart. Frontal and 30-degree view showing positive and negative surface differences. Areas of light blue and green show good correspondence between the two scans, showing that the landmarks capture most of the skull shape.
SC
Figure 2 Landmark projected on the skull in frontal (left) and 30-degree (right) view, showing how the landmarks from table 1 are placed on the skull.
M AN U
Figure 3 First principal component of the normal group in frontal (top) and 30-degree (bottom) view. From left to right: minus two SD and plus two SD, showing allometric growth within the age range of 6 to 13 years.
Figure 4 Second principal component of the normal group in frontal (top) and 30 degree (bottom) view. From left to right: minus two SD and plus two SD, showing how short
TE D
faces tend to be wider (left) and long faces tend to be thinner (right) in the way they vary in relation to each other
Figure 5 First principal component of the syndromic group in frontal (top) and 30-degree
EP
(bottom) view. From left to right: minus two SD and plus two SD, showing the shape variation of the frontofacial region within the syndromic group. On the left: a tall
AC C
forehead, retruded and short midface. On the right: a shorter forehead with a less retruted and taller midface. Figure 6 Second principal component of the syndromic group in frontal (top) and 30degree (bottom) view. From left to right: minus two SD and plus two SD, showing allometric growth within the age range of 6 to 13. Figure 7 Normal skull transforming into a syndromic skull using the (PCA) model. From left to right: normal skull transforming into its (unique) predicted syndromic skull. Figure 8 Crouzon skull transforming into a normalized one using the (PCA) model. From
17
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
left to right: Crouzon skull transforming into its (unique) predicted normalized skull.
18
ACCEPTED MANUSCRIPT
Table legend
Table 1 Set of 66 landmarks used in this study.
RI PT
Table 2 Characters in the first column correspond to the landmarkset used in this study. The second and third column represent the intra-rater reliability. Table 3 Age and sex distribution of the syndromic population.
SC
Table 4 Measurements in the orbital and upper midface (zygomatic) regions for the syndromic and normal subjects between 6 and 9 years of age.
M AN U
Table 5 Measurements in the orbital and upper midface (zygomatic) regions for the
AC C
EP
TE D
syndromic and normal subjects between 10 and 13 years of age.
19
Label Landmark
Definition
Scan orientation
AB
Orbitale
The lowest part of the orbital margin. It is used to define the Frankfort Plane and to measure orbital height.
Frontal
CD
Superior ‘orbitale’
Exact point on inner superior orbital rim vertically above ‘orbitale’
Frontal
EF
Frontomalare anterior
Most anterior point on the fronto-malar suture Frontal
GH
Supraorbital notch
Most superior point of the supraorbital notch
IJ
Infraorbital foramen
Most superior point of the infraorbital foramen Frontal
KL
Infraorbital foramen / orbital rim
Point on the inner orbital rim directly above infraorbital foramen
Frontal
MN
Ectoconchion
The intersection of the most anterior surface of the lateral border of the orbit and a line bisecting the orbit along its long axis. To mark Ectoconchion, move a toothpick or other thin straight instrument up and down, keeping it parallel to the superior orbital border, until you divide the eye orbit into two equal halves. Mark the point on the anterior orbital margin
Frontal
OP
Superior orbital fissure The most superior lateral point of the superior Frontal orbital fissure
Q
Nasion
The point of intersection of the nasofrontal suture and the mid-sagittal plane. Nasion corresponds to the nasal root
RS
Nasomaxillary suture pinch
Narrowest portion of the midline of the face to Rotate 30 left/right the naso-maxillary suture.
TU
Dacryon
V
Inferior part of nasal bone
W
Anterior nasal spine
X
AC C
ACCEPTED MANUSCRIPT
YZ A1
TE D
M AN U
SC
RI PT
Rotate 20 left/right
Frontal
Rotate 10-20 left/right
The most inferior point of the junction between the nasal base in the midsaggital plane. The apex of the anterior nasal spine
20-30 chin down
EP
The point on the medial border of the orbit at which the frontal, lacrimal, and maxilla intersect. In other words, Dacryon lies at the intersection of the lacrimomaxillary suture and the frontal bone.
40 chin down
Subspinale
The deepest point of the profile below the anterior nasal spine.
Frontal
Alare
The most laterally positioned point on the anterior margin of the nasal aperture
Frontal
Prosthion
Most anterior midline point on the alveolar process of the maxilla between the central incissors.
Frontal
B1C1 Interdentale superior lateral1 (right + left)
Most anterior point on the alveolar process of Frontal the maxilla between the central incisor and the lateral incisors. (Between the 41,42 (right) and the 31,32 (left))
ACCEPTED MANUSCRIPT Most anterior point on the alveolar process of Rotate 20 left/right the maxilla between the lateral incisor and the canine. (Between the 12,13 (right) and the 22,23 (left))
F1G1 Maxillary angle
Most superior medial point on the curve in the maxilla from a frontal view
Frontal
H1I1
Ectomalare
Frontal and check 90 degree left/right
J1K1
Infraorbital foramen to alveolus
On the maxilla: positioned at the most lateral point on the lateral surface of the alveolar crest. Found along the second molar on the maxilla. Middle of the straight line down from the infraorbital foramen (GH) to the dentoalveolar junction
RI PT
D1E1 Interdentale superior lateral2 (right + left)
Frontal
Middle of the straight line from the infraorbital Frontal foramen (GH) to the maxillary angle (U2V2)
N1O1 Zygomaxillare
Intersection of zygomaxilllary suture and most medial masseter muscle attachment.
Rotate 20-40 left/right
P1Q1 Zygomatic angle
Angle between zygomatic ridge and orbital portion of the zygoma
Rotate 70-90 left/right
R1S1 Glenoid fossa
Most superior point of the glenoid fossa
Rotate 90 left/right
T1U1 Upper part of half zygomatic process (right + left)
Upper part in the middle of the glenoid fossa and zygomatic angle
Rotate 90 left/right
V1W1 Lower part of half zygomatic process (right + left)
Lower part in the middle of the glenoid fossa and zygomatic angle
Rotate 90 left/right
X1Y1 Porion
Highest point on the external auditory meatus Rotate 80-90 left/right
Zygomaticofrontal suture
B2C2 Mastoidale
F2G2 Pterion
Glabella
M AN U
AC C
H2
Most lateral point on the fronto-malar suture.
Rotate 100-140 left/right
The inferior tip of the mastoid process
Robins 3D
Most superior point of superior temporal line. (top of the lateral orbit bone)
Rotate 30-40 left/right
Region/point on the upper end of the greater wing of the sphenoid. Junction of the sphenoid, temporal, parietal and frontal sutures. The most forwardly projecting point in the mid-sagittal plane at the lower margin of the frontal bone, which lies above the nasal root and between the superciliary arches. Note that in juvenile skulls with strongly forwardly vaulted foreheads, the most projecting point of the curve of the forehead is not that of Glabella.
Robins 3D
EP
D2E2 Frontotemporale
TE D
Z1A2
SC
L1M1 Infraorbital foramen to maxillary angle
Frontal
I2
Bregma
The point where the sagittal and coronal sutures meet
Rotate 70 chin down.
J2
Lambda
The point where the two branches of the lambdoidal suture meet with the sagittal suture
Rotate 180 left/right, place the mark and 20 chinup to get the right angle for the mark
ACCEPTED MANUSCRIPT K2L2 Infraorbital foramen to frontal bone
Frontal: ‘Ortho reslice’ in Robins 3D
Frontal: ‘Ortho reslice’ in Robins 3D
AC C
EP
TE D
M AN U
SC
RI PT
M2N2 Lateral orbit to forehead
The most ventral part of the frontal bone (NOT part of the orbit what might be more ventral) on the straight line from the infraorbital foramen The most ventral part of the frontal bone on the straight line from the lateral part of the orbital rim
ACCEPTED MANUSCRIPT 0.91
B
0.945
1.002
C
1.086
1.118
D
0.935
1.189
E
0.768
0.564
F
0.368
0.646
G
0.374
0.369
H
0.434
0.517
I
0.412
0.666
J
0.354
0.497
K
0.615
0.664
L
0.473
0.432
M
1.106
1.264
N
0.769
0.878
O
0.436
0.425
P
0.458
0.527
Q
0.961
0.317
R
0.538
0.716
S
0.473
0.611
T
0.797
1.11
U
1.168
0.882
V
0.435
0.57
0.449
X
0.618
Y
0.462
Z
1.164
A1
0.364
B1
0.455
C1
0.49
0.479
D1
0.54
0.531
0.509
0.451
0.381 0.472 0.621 0.474
EP
0.812
J1
AC C
E1
0.46
TE D
W
0.597
1.199
K1
0.818
0.949
L1
0.487
0.721
M1
0.584
1.761
N1
1.764
0.956
O1
1.976
0.594
P1
0.643
0.403
Q1
0.786
0.425
F1 G1 H1 I1
SC
1.005
M AN U
A
RI PT
Label SD normals SD syndromic
0.576
0.363
0.682
0.344
2.105
0.648
1.633
0.921
0.691
S1
1.284
0.815
T1
0.957
0.635
U1
0.537
0.74
V1
1.001
0.792
W1
0.586
0.7
X1
1.612
2.269
0.788
0.43
0.782
1.329
A2
1.024
1.172
B2
0.523
0.512
C2
0.503
0.396
D2
0.739
0.489
E2
1.126
0.966
F2
2.596
2.296
G2
2.249
2.656
H2
1.165
1.035
I2
0.594
1.017
J2
2.123
1.395
K2
1.467
1.835
L2
1.621
2.531
M2
1.626
2.88
N2
1.801
2.955
AC C
EP
TE D
Y1 Z1
SC
0.878
M AN U
R1
RI PT
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 3: Age and sex distribution of the syndromic population Age in years Females Males Total 6 4 3 7 7 1 1 2 8 1 2 3 9 2 1 3 10 1 0 1 11 5 0 5 12 1 3 4 13 2 0 2 Total 17 10 27
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 4: Measurements in the orbital and upper midface (zygomatic) regions for the syndromic and normal Subjects between 6 and 9 years of age Syndromic Normalized syndromic Normal Controls Variable Mean (mm) SD Mean (mm) SD Mean (mm) SD Anterior interorbital distance 21.6 3.1 20.1 1.9 21.8 2.8 Lateral orbital distance 92.4 4.8 83.7 4.2 87.3 4.7 Intertemporal distance 85.3 7.4 77.9 6.7 72.7 6.0 Interzygomatic-buttress distance 80.2 6.2 78.8 5.2 83.0 4.7 Interzygomatic-arch distance 105.7 5.5 105.5 6.3 106.9 10.8 Zygomatic-arch length 43.0 5.9 46.4 6.3 51.0 3.7
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Table 5: Measurements in the orbital and upper midface (zygomatic) regions for syndromic and normal Subjects between 10 and 13 years of age Syndromic Normalized syndromic Normal Controls Variable Mean (mm) SD Mean (mm) SD Mean (mm) SD Anterior interorbital distance 23.4 4.3 20.9 1.8 22.3 1.8 Lateral orbital distance 100.1 3.8 90.9 3.7 90.3 4.3 Intertemporal distance 91.2 7.3 83.7 6.6 74.6 4.8 Interzygomatic-buttress distance 86.0 5.3 84.0 6.1 85.1 5.5 Interzygomatic-arch distance 115.0 5.5 114.4 5.9 112.1 7.1 Zygomatic-arch length 43.0 6.3 49.1 6.9 55.2 2.4
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
