https://orcid.org/0000-0001-7304-044X
Summary
To explore profile shape variation in a large population of middle-aged individuals and investigate features of sexual dimorphism.
Facial profile photographs of 1776 individuals (964 females and 812 males; 46 years old), members of the Northern Finland Birth Cohort 1966 (NFBC1966), were uploaded and digitized with Viewbox software. Profile shape was defined with 47 landmarks (4 fixed and 43 sliding semi-landmarks). Digitized images were reoriented and scaled with Procrustes Superimposition, and shape variation was determined with a Principal Component Analysis.
More than 90 per cent of profile shape variation was explained by Principal Components (PC) 1–9. PC1, explaining the largest amount of variation (33.1 per cent) described changes in facial convexity, slope of the forehead, lip, and chin protuberance. PC2 (23.1 per cent variation) was more related to vertical changes of the lower facial third and PC3 (11 per cent variation) primarily described changes in lip protrusion and nasal projection. Shape analysis showed a significant difference between the average female and the average male profile shape (P < 0.001); however, this was not evident upon visual observation. The shape variable most associated with sex was PC3 (η 2 = 0.245; P < 0.001), which described changes in lip prominence and in projection of the dorsal nasal surface. An additional discriminant analysis showed that profile shape predicted sex in 76 per cent of males and 79.6 per cent of females.
There is significant sexual dimorphism in facial profile among middle-aged adults. Profile shape variation was associated to changes in lip protrusion, nasal protuberance, and chin projection.
Introduction
Facial appearance is a result of genetic and environmental factors acting throughout the entire life span. The most common facial changes occurring during growth and development are a vertical increase of the lower facial third, a progressively more prominent nose, and a flattening of the orbital and the midfacial region (1, 2). Males and females experience facial growth differently, partly based on their respective hormonal inputs (3, 4), and females also reach skeletal maturity earlier than males (5). However, after adolescence and throughout adulthood, sexual dimorphism in facial appearance tends to decline and differentiate (1, 6, 7). In mature adults, sex appears to have a smaller effect on facial dimorphism than age, mainly due to changes in bone metabolism and reduced hormonal expression occurring with aging (7). Declining levels of oestrogen and testosterone, in females and males, respectively, have a similar impact on both sexes, namely a decrease in mandibular size, a flattening of the face and thinning of the lips (8–11). Although these changes tend to occur almost twice as fast in females than in males (9), ultimately, facial differences tend to become more subtle with time.
As facial aesthetics continue to be the main motivational factor for patients seeking various treatments, including orthodontics, physiologic changes in facial morphology provide important information to clinicians, whose interventions aim to improve facial appearance. Diagnostic assessments of the face can be performed with direct anthropometric measurements through radiographic images also depicting soft tissues and with clinical photography (1, 6, 7, 12). These methods have been previously applied in order to explore human facial morphology and determine differences between sexes, ethnicities, and age groups. Sexual dimorphism, in particular, has been the primary focus of many investigations, the most recent of which have made assessments using 3D facial images of various population groups (1, 6, 12–16). In their vast majority, these studies show that the most distinct differences between males and females faces are seen in the slope of the forehead (16, 17), the size and shape of the nose (13, 15, 18, 19), the contour of the lips, and facial convexity (1,14,16).
Despite recent technological developments in 3D photography, a set of 2D photographs remain the most commonly used tool for facial evaluations in most clinical settings and has been found reliable for clinical purposes (20, 21). For this reason, 2D facial data have also been used extensively in the past (6, 18, 22, 23). Most of the literature, however, investigates populations of growing individuals in order to assess growth changes during adolescence or soft tissue changes after treatment (18, 22). Some studies have reported on soft tissue morphology or physiologic facial changes in older, adult populations (18, 23–25), but their sample populations are generally small or largely heterogeneous. The results of those studies indicate that aging of the human face is expressed mostly through changes in the lips, nose, and chin projection.
In addition, 2D soft tissue outlines have mostly been described with linear and angular variables. These are heavily influenced by size and, therefore, fail to provide reliable shape information (26). In order to be able to better assess facial variation and sexual dimorphism, geometric morphometric methods can be applied using landmark configurations that are subsequently translated, rotated, and scaled to a common centroid size by means of Procrustes Superimposition (27). Such methods have been previously used to assess facial profiles in a sample of growing orthodontic patients (6, 28). However, the results of these studies are not representative of the general population and cannot be extrapolated to older groups of patients. Up to this date, large population data of facial profile shape in adults are not available and, thus, the present investigation aimed to establish patterns of shape variation and explore the presence of sexual dimorphism in a large homogenous sample of middle-aged adults. The study hypothesis was that there is significant sexual dimorphism presenting in the profiles of middle-aged adults.
Materials and methods
Study population and data collection
The sample for this investigation was part of the Northern Finland Birth Cohort 1966 (NFBC1966), a longitudinal research program developed to improve and promote the well-being of the population in Northern Finland. All members of the cohort were born in 1966 in the two northernmost provinces of Finland, Oulu and Lapland, and comprise 96.3 per cent (N = 12 058) of all births in this region within this year (29). In 2012, at the age of 46 years, 1964 of these individuals, who were all born and lived within a 100 km range from the city of Oulu, returned for a follow-up clinical examination and a series of diagnostic tests. For the purpose of this study, we included all participants who did not have recognizable craniofacial deformities, craniofacial syndromes, or a history of treatments that could have altered their facial appearance. Also, included extraoral images needed to be of diagnostic quality, with neutral facial expression and without any interruptions in the profile outline (e.g. presence of facial hair). After the review of the initial sample, 1776 individuals (964 females and 812 males) fulfilled these prerequisites and, thus, comprised the final study population (Supplementary Figure 1). The research protocol was approved by the Ethical Committee of the Northern Ostrobothnia Hospital District on September 2011 (EETTMK#: 74/2011), and written consent was obtained from all volunteers prior to participation.
A detailed description of the performed clinical examination has been reported elsewhere (30); and clinical photographs (frontal and profile) were obtained according to a predetermined protocol in order to assure standardization (31). Subject demographics were exported to an Excel worksheet (Microsoft Excel, Microsoft©, Richmond, Washington, USA) and facial profile images were imported into Viewbox 4.1 software (dHAL Software, Kifissia, Greece) for digitization.
Shape analysis
Shape analysis of all profile images was performed with geometric morphometric methods utilizing a predetermined digitization template with 48 landmarks (4 fixed and 44 sliding semi-landmarks). These methods aim to quantify the shape variation of anatomical structures using Cartesian coordinates of landmarks and after other sources of variation (e.g. size) have been mathematically controlled for (26). Here, digitization of each image was performed in three steps. First, the upper and lower lip vermillion borders were selected manually by an experienced operator (GK). Then, two curves were projected on the image, one between the upper stomion point to trichion, defining the upper profile outline, and one between the lower stomion point and the middle of the throat, defining the lower profile outline. This was preferred to a single curve outlining the entire profile in order to account for instances with lip incompetence. In the last step of the process, constructed landmarks were automatically distributed between the extremes of the upper and lower profile outlines on predetermined locations corresponding to a percentage distance along their respective curve. For example, Point 1 on the lower outline was automatically placed at 5 per cent of the distance between the lower vermillion border and the end of the lower outline curve, Point 2 was placed at 10 per cent of the distance, and Point 3 at 15 per cent of the distance until the entire lower profile was adequately described with points. The upper profile outline, the upper lip and the lower lip were defined in a similar manner. Upper and lower vermillion borders that were placed manually in the first step were represented on the profile outlines by their nearest point on the respective curves.
The use of traditional soft tissue landmarks was avoided in order to minimize the error of manual landmark identification and because soft tissue landmarks do not meet the definition of a true anatomical landmark. Anatomical landmarks are biologically homologous among humans and describe locations where there is a visible change in local anatomy, such as the extremes of a structure or the crossing point of two types of tissue or texture (32). Therefore, the only points that were treated as anatomical landmarks in this study were the upper and lower stomia and the vermillion borders of the lips. All other points were constructed points that were distributed automatically on the profile curves and were treated as semi-landmarks. By definition, semi-landmarks do not have a biological interpretation and do not show correspondence across samples; therefore, they cannot be used as homologous landmarks during subsequent shape analyses. In order to reduce shape variation related to the initial placement of curve semi-landmarks, following the digitization of all profile images, semi-landmarks were allowed to slide along their respective curves to optimize their position as compared to the average shape of the entire sample. This iterative process was repeated three times in order to minimize the bending energy between individual samples and the average shape and establish homology between corresponding landmarks (32, 33).
The resulting profile shapes were superimposed with Procrustes Superimposition (27) to minimize the sum of squared distances between all points and transform landmark coordinates into shape coordinates (Supplementary Figure 2). A Principal Component Analysis was used on shape coordinates for variable reduction and to determine the most significant principal components (PCs) of shape. In essence, the shape information provided through the shape coordinates is reduced to a smaller number of variables, PCs, in order to allow for further statistical analyses. This is a standard statistical process in geometric morphometrics in order to be able to interpret the results of the data analyses.
Statistical analysis
The variation of facial profile was explored with 3D plots of the PCs. In order to visualize the direction of variation explained by each PC, profile shapes were created using the mean shape coordinates and the shape coordinates corresponding to ±3 standard deviation (SD) from the mean (34). In this way, the direction of variation was visualized as change in profile shape configurations. This process was performed for the entire sample, as well as for the female and male subsamples separately. Shape differences between males and females were detected with permutation tests on the mean Procrustes distance between the male and female profile configurations.
In order to assess the effect of sex on profile shape, a multivariate regression model was developed with the first nine PCs as dependent factors and sex as the independent variable. An additional discriminant analysis with cross-validation was performed in the entire sample to investigate if profile shape could predict sex. The results of the discriminant analysis were cross-validated with the ‘leave one out’ method. A Type-1 error of 5 per cent was accepted for all statistical analyses.
Results
Error of the method
In order to evaluate the error of the method and its reliability, the above process was repeated for 120 randomly selected images, after a washout period of at least 2 weeks. The shape coordinates of these 120 subjects that were extracted from the second digitization were then compared to the shape coordinates from the first digitization. This comparison was done in two ways. The mean Procrustes distance between digitizations performed at two different time points was assessed through permutation tests (10 000 permutations) and was found to be non-significant, indicating no systematic error (P = 0.915). Random digitization error was determined as a percentage of total variance in shape space as explained by PC1–PC9. This resulted in a random error of 4.6 per cent, which represented the amount of variance attributed to the repetition of the entire digitization process and was considered to be low.
Shape analysis
More than 70 per cent of shape variation in our sample was explained by the first four PCs (PC1: 33.1 per cent; PC2: 23.1 per cent; PC3: 11 per cent; and PC4: 6.7 per cent) and more than 90 per cent of variation was explained by the first nine PCs. (Figure 1). In order to visualize profile shape variation, profile morphings were created based on the shape information provided by PC1–PC4, explaining more than 70 per cent of the sample variation (Figure 1). The amount and direction of shape variation within each PC axis are projected through profile outlines representing ±3 SDs away from the average profile (Figure 2c). PC1, explaining the largest amount of shape variation, described changes in the slope of the forehead, chin protuberance, and lip fullness. PC2 was more related to the size of the nose and the vertical dimension of the lower facial third; PC3 primarily described changes in lip fullness and projection of the bridge of the nose, and PC4 was mainly linked to the length of the upper lip and the depth of the labiomental sulcus.
Scree plot displaying the percentage of variation explained by each PC.
![(a) Principal component analysis (PCA) graph displaying facial variation according to sex [in standard deviation (SD) units], in the entire sample, as explained by PC1 (33.1%), PC2 (23.1%), and PC3 (11%) (females: light red; males: blue). (b) Average and exaggerated profiles of female (red) and male (blue) subjects. c) Shape morphings displaying shape variation as explained by PC1–4. The middle column shows the average profile shape of the entire sample. Lines crossing the profile shape represent the amount and direction of variation at each profile landmark (blue: positive; red: negative). The left and right columns present morphings of shape extremes from −3 to +3 SDs of PC scores (top row: PC1; upper middle row: PC2; lower middle row: PC3; and bottom row: PC4).](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/ejo/44/1/10.1093_ejo_cjab004/2/m_cjab004_fig2.jpeg?Expires=1745057877&Signature=nnGCKas3EUKTSR8x7PHU70CvGxVAMwo~uBFh7S4TtIPhNq-bqXb4zHtDpf8oxMtCSQ7ZmfW7Q1rnN4suNuA-8qkL-TM1v537h-iZ0Fmq1C-3gm7G4W0rc3uic2rQeWDPKKuELSUce8wdUqi4GMTGnl8~eGwl93vTBngMGtkjdqznjKcjuMURKat1PYx8atw7KAJAiXxAmj9rmyw2jHhWeKleAx0Hd5ayApqecxRJ6jOziOHa-UKPKn01QN36e2kdae5NEqvHGTO1MwAbqVg6qg~kc0lpNRPyJ3HcFaHHAogJ0-Gg4~6uF68oU593ZwoTCmHoB6DtKVAx8Uch56nVoA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
(a) Principal component analysis (PCA) graph displaying facial variation according to sex [in standard deviation (SD) units], in the entire sample, as explained by PC1 (33.1%), PC2 (23.1%), and PC3 (11%) (females: light red; males: blue). (b) Average and exaggerated profiles of female (red) and male (blue) subjects. c) Shape morphings displaying shape variation as explained by PC1–4. The middle column shows the average profile shape of the entire sample. Lines crossing the profile shape represent the amount and direction of variation at each profile landmark (blue: positive; red: negative). The left and right columns present morphings of shape extremes from −3 to +3 SDs of PC scores (top row: PC1; upper middle row: PC2; lower middle row: PC3; and bottom row: PC4).
Profile shape exhibited statistically significant sexual dimorphism in the studied population. This was determined with permutation tests on the mean Procrustes distance between males (N = 812) and females (N = 964; Procrustes distance: 0.020, P < 0.001; 100 000 repetitions without replacement). The variation in profile shape within the entire sample is displayed in Figure 2a by means of a 3D plot. The average landmark configurations for males and females were transformed into average male and females profile shapes. In order to better identify sex differences, exaggerated versions of the average male and female profiles were created by multiplying the male and female average shape coordinates by 5, and transforming the new coordinates into profile outlines (Figure 2b). As shown in Figure 2b, males presented flatter lips, a more prominent nose, and more pronounced eyebrow ridges than females.
Due to the significant sexual dimorphism in profile, shape variation was also assessed separately for males and females, presenting similar morphing patterns for PC1–PC4 as the ones observed in the entire sample (Supplementary Figures 3–6).
A multivariate regression model was used to evaluate the effect of sex on shape and indicated a significant association (η 2 = 0.398; P < 0.001). Between-groups effect sizes revealed that males and females differed primarily in PC3 scores (η 2 = 0.245; P < 0.001), while, for other PCs, this effect was not strong (Table 1). In order to further investigate the association between sex and profile shape variation, a discriminant analysis was performed to assess if profile shape could predict the sex of an individual within the sample. The results showed that profile shape, as described by PC1–PC9, successfully predicted sex in 76 per cent of males and 79.6 per cent of females (Supplementary Table 1). Based on the discriminant function coefficients, shape variation explained by PC3 has the strongest predictive effect (Supplementary Table 2). This is in agreement with the previous results of the multivariate analysis, and, thus, it can be concluded that shape variation described by PC3 relates the most to the presence of sexual dimorphism in the sample.
Between groups effect sizes in profile shape variation as explained by PC1–PC9
| Profile shape PCs | Variation explained (%) | η 2 | P-value |
|---|---|---|---|
| PC1 | 33.1 | 0.003 | 0.029 |
| PC2 | 23.1 | 0.000 | 0.865 |
| PC3 | 11.0 | 0.245 | 0.000 |
| PC4 | 6.7 | 0.005 | 0.003 |
| PC5 | 5.1 | 0.055 | 0.000 |
| PC6 | 4.0 | 0.078 | 0.000 |
| PC7 | 3.2 | 0.002 | 0.050 |
| PC8 | 3.1 | 0.004 | 0.008 |
| PC9 | 2.2 | 0.006 | 0.001 |
| Profile shape PCs | Variation explained (%) | η 2 | P-value |
|---|---|---|---|
| PC1 | 33.1 | 0.003 | 0.029 |
| PC2 | 23.1 | 0.000 | 0.865 |
| PC3 | 11.0 | 0.245 | 0.000 |
| PC4 | 6.7 | 0.005 | 0.003 |
| PC5 | 5.1 | 0.055 | 0.000 |
| PC6 | 4.0 | 0.078 | 0.000 |
| PC7 | 3.2 | 0.002 | 0.050 |
| PC8 | 3.1 | 0.004 | 0.008 |
| PC9 | 2.2 | 0.006 | 0.001 |
Fixed factor: sex (male/female).
PC, Principal Component.
Between groups effect sizes in profile shape variation as explained by PC1–PC9
| Profile shape PCs | Variation explained (%) | η 2 | P-value |
|---|---|---|---|
| PC1 | 33.1 | 0.003 | 0.029 |
| PC2 | 23.1 | 0.000 | 0.865 |
| PC3 | 11.0 | 0.245 | 0.000 |
| PC4 | 6.7 | 0.005 | 0.003 |
| PC5 | 5.1 | 0.055 | 0.000 |
| PC6 | 4.0 | 0.078 | 0.000 |
| PC7 | 3.2 | 0.002 | 0.050 |
| PC8 | 3.1 | 0.004 | 0.008 |
| PC9 | 2.2 | 0.006 | 0.001 |
| Profile shape PCs | Variation explained (%) | η 2 | P-value |
|---|---|---|---|
| PC1 | 33.1 | 0.003 | 0.029 |
| PC2 | 23.1 | 0.000 | 0.865 |
| PC3 | 11.0 | 0.245 | 0.000 |
| PC4 | 6.7 | 0.005 | 0.003 |
| PC5 | 5.1 | 0.055 | 0.000 |
| PC6 | 4.0 | 0.078 | 0.000 |
| PC7 | 3.2 | 0.002 | 0.050 |
| PC8 | 3.1 | 0.004 | 0.008 |
| PC9 | 2.2 | 0.006 | 0.001 |
Fixed factor: sex (male/female).
PC, Principal Component.
Discussion
This is the first study to investigate profile shape variation in a large, genetically homogenous population of middle-aged individuals. This is a unique group because all its members have the same exact age and ethnic identity. Due to these baseline characteristics, the observed phenotypic variation is mainly associated with sex differences; environmental and ethnic effects are less likely to have had an effect on our results.
Our results showed significant sexual dimorphism in facial profile; however, this was not directly obvious upon visual observation of the average male and female profiles (Figure 2b). Therefore, we exaggerated those average profile outlines in order to intensify the differences between males and females. The exaggerated profiles (Figure 2b) revealed that the areas contributing the most to the presence of sexual dimorphism are the lips, the nose, and the forehead region. Males had flatter lips, a more prominent nose, and more pronounced eyebrow ridges than females. Others have also reported that sexual dimorphism might not always be easily identifiable (14, 16) and have, thus, also used exaggerated versions of male and female faces to explore sex differences (14).
Facial appearance in mature adults is affected by change in hormonal input, decrease of bone metabolism, and reduction of tissue tonicity (4, 7, 35). It has been suggested that facial variation decreases in middle age (41–64 years) due to the lengthening of the lower face in women, and then tends to be reinstated at an older age due to a subsequent lengthening of the lower face in men (7). Although the present sample is not longitudinal and, thus, cannot reveal age-related changes, the visual presentation of our results shows that facial dimorphism among middle-aged adults is minor.
However, the statistical analysis revealed significant sexual dimorphism in our sample. The between-group analysis showed that PC3, describing changes in lip protrusion, slope of the forehead, and projection of the dorsal nasal surface had the largest effect in differentiating between sexes. Differences in these facial areas are clearly visualized in the exaggerated male and female profile shapes (Figure 2b and 2c). Considering that most of the facial aging process is also related to changes in the lips and the nose (24, 36), one may speculate that the facial areas with high sexual dimorphism are also areas related to facial neoteny. In fact, there is ample discussion about whether facial differences between males and females are not strictly sex dependent but are also related to factors influencing facial youthfulness (7, 37, 38).
The areas of largest shape variation in the sample are described by PC1–4 and shown in Figure 2. PC1 was mainly associated with changes in chin protuberance and lip protrusion and ranged from a convex profile with full lips to a concave profile with thinner lips and a more dominant chin (Figure 2c). PC2 was more related to changes in nasal projection and in the vertical dimension of the lower face. Taking into consideration that the total size of the profile shape is maintained constant from the Procrustes Superimposition, the variation in nasal appearance represents pure shape differences (Figure 2c). PC3, which had the strongest effect on sexual dimorphism, described changes in lip protrusion, slope of the forehead, and dorsal protuberance of the nose, as mentioned before, while PC4 was related to vertical changes of the upper lip. All the above patterns also represent facial changes occurring in the human face with aging, namely the flattening and lengthening of the lips, a downward and forward growth of the nose, and a more pronounced chin (18, 25) and, thus, agree with the assumption that the areas of highest profile variation are also areas related to a youthful appearance of the face.
Based on methodology, our results are comparable to previous studies using geometric morphometric methods to assess facial soft tissue variation and sexual dimorphism in humans (6, 14, 15, 16). Mydlová et al. (16) studied an ethnically homogenous population of individuals between 20 and 80 years and reported that, in all age groups, males presented with a more protruded forehead, eyebrow ridges, nose, and a longer upper lip. Although their observations agree with the results presented here, it is unlikely that they are representative of the general population due to the small sample size of that study in relation to the wide age range. Here, we report on a very large group of individuals, all of the same exact age, and are thus able to make safer conclusions about the general population. A similar study by Imaizumi et al. (14) studied a large homogenous population of 1000 individuals and identified the nose and the lips as the areas of greater sexual dimorphism. However, they included individuals between 19 and 59 years of age and reported their findings without controlling for the effect of age on facial shape. It is, therefore, expected that their results are heavily affected by the large age variation and are, therefore, not representative of the true sexual dimorphism in the population.
Furthermore, our discriminant analysis revealed that PC1–9, describing more than 90 per cent of shape variation, were able to predict sex in 76 per cent of males and 79 per cent of females in the present middle-aged adult population. Others have also used facial metric systems in order to address the same question and have reported predictive values higher than 90 per cent in some instances (13, 15, 39, 40). However, most previous studies also included size as a predictor in their discriminant analysis, influencing the results significantly as males have significantly larger faces than females. When size was excluded from the analysis, the predictive value of shape alone dropped to 73.3 per cent (15), which is closer to the percentages shown here. In addition, the vast majority of previously published studies included young adults or heterogeneous samples, where facial differences are expected to be more pronounced. A more recent 3D face study by Abbas et al. (41) correctly classified sex with an accuracy of 89 per cent in a sample of 4745 15-year-old individuals using Geodesic path curvature. This accuracy is 10 per cent higher than reported in this study and may reflect the differences between the use of 3D and 2D techniques in that 3D techniques may be more discriminatory, especially in a young cohort who have not yet reached full face maturity.
Clinically, facial profiles contribute significantly to a comprehensive diagnosis and treatment planning. Despite the vast developments in 3D technology, most orthodontic treatment planning is still largely performed with traditional diagnostic methods in daily practice. These include cephalometric images, 2D facial photographs, and plaster or digital models. Within this context, the information provided by this investigation applies to the current status of clinical practice. Therefore, the knowledge of normal profile shape variation at different stages of facial growth and development can improve treatment decisions by individualizing them to each patient. This is especially true for patients seeking plastic surgery or other interventions to improve their facial appearance.
Special considerations and limitations
The present study was conducted on a large homogenous Finnish population, representing Northern European populations of Caucasian ancestry. Despite the advantages that this sample presents, the results of this study should be interpreted within the context of the sample characteristics and, therefore, cannot be extrapolated to populations of different ethnic origin. Furthermore, we have no knowledge of the percentage of menopausal females in our sample. Menopause, especially at early stages, has been reported to affect facial shape in females and thereby disrupt the observed convergence of facial appearance between males and females after the age of 50 (9). It also needs to be noticed that there are no reports regarding the orthodontic history of the individuals in the sample. As part of a government-funded health care system, it is possible that they received some orthodontic intervention at a prepubertal or pubertal age. However, this is not expected to have altered the results of this study because cases that underwent orthognathic or other surgical interventions were excluded. Moreover, orthodontic treatment is very common in modern societies and can be considered part of a modern way of life. On the contrary, more invasive treatments that are performed solely to create change in facial appearance are not performed often, are elective, and are not part of a social health care program.
Our shape analysis also revealed some outliers as seen in Figure 2a. These were all individually checked post hoc in order to assess if there were errors in the digitization process. All outliers were individuals with extremely thin lips, which significantly affected the entire profile. However, they were not excluded from the sample because they were true data. In addition, it is unlikely that they influenced our results because they were less than 1 per cent of the entire sample.
Conclusions
The results of this study show significant sexual dimorphism in facial profile among middle-aged adults and, thus, we are able to accept the initial study hypothesis. The performed shape analyses revealed that profile shape variation in this age group is primarily associated with changes in lip protrusion, nasal protuberance, and chin projection. Furthermore, profile shape was able to predict sex in 76 and 79 per cent of males and females, respectively.
Supplementary material
Supplementary material is available at European Journal of Orthodontics online.
Supplementary Table 1: Discriminant analysis of the predictive value of profile shape on sex.
Supplementary Table 2: Individual contribution of each of the predicting factors to the discriminant function.
Supplementary Figure 1: Flow chart displaying the selection process for the final study population.
Supplementary Figure 2: Schematic displaying the process of Procrustes Superimposition. In order to compare the two shapes (red and black) that are defined with 5 landmarks each, those shapes are first centered (b) to a common centroid. Then, they are scaled to a fixed centroid size in order to eliminate size differences between the shapes (c) and finally the shapes are rotated so that the sum of the squared distances between corresponding landmarks is minimized (d). This process allows for pure shape comparisons between landmark configurations. It is noted that this schematic is purely descriptive and only serves to facilitate the readers’ understanding of Procrustes Superimposition.
Supplementary Figure 3: a) PCA graph displaying facial variation in females (in SD units), as explained by PC1 (35.7%) and PC2 (23.4%). b) The middle column depicts the average female facial shape, as well as the amount and direction of variation at each profile landmark (blue: positive, red: negative). Profile morphings showing the shape extremes from -3 to +3 standard deviations of PC scores (top row: PC1, bottom row: PC2).
Supplementary Figure 4: a) PCA graph displaying facial variation in females (in SD units), as explained by PC3 (8.9%) and PC4 (6.2%). b) The middle column depicts the average female facial shape, as well as the amount and direction of variation at each profile landmark (blue: positive, red: negative). Profile morphings showing the shape extremes from -3 to +3 standard deviations of PC scores (top row: PC3, bottom row: PC4).
Supplementary Figure 5: a) PCA graph displaying facial variation in males (in SD units), as explained by PC1 (34.1%) and PC2 (23.8%). b) The middle column depicts the average male facial shape, as well as the amount and direction of variation at each profile landmark (blue: positive, red: negative). Profile morphings showing the shape extremes from -3 to +3 standard deviations of PC scores (top row: PC1, bottom row: PC2).
Supplementary Figure 6: a) PCA graph displaying facial variation in males (in SD units), as explained by PC3 (9.1%) and PC4 (7.6%). b) The middle column depicts the average male facial shape, as well as the amount and direction of variation at each profile landmark (blue: positive, red: negative). Profile morphings showing the shape extremes from -3 to +3 standard deviations of PC scores (top row: PC3, bottom row: PC4).
Acknowledgements
We thank all cohort members and researchers who participated in the 46 years study. We also wish to acknowledge the work of the NFBC Project Center.
Funding
NFBC1966 received financial support from the University of Oulu grant no. 24000692, Oulu University Hospital grant no. 24301140, and European Regional Development Fund grant no. 539/2010 A31592.
Conflict of interest
D.H. owns stock in dHAL Software, the company that markets Viewbox 4. The other authors have nothing to declare.
Data availability
NFBC data is available from the University of Oulu, Infrastructure for Population Studies. Permission to use the data can be applied for research purposes via electronic material request portal. In the use of data, we follow the EU general data protection regulation (679/2016) and Finnish Data Protection Act. The use of personal data is based on cohort participant’s written informed consent at his/her latest follow-up study, which may cause limitations to its use. Please contact NFBC Project Center ([email protected]) and visit the cohort website (www.oulu.fi/nfbc) for more information.
References
