Quantitative Analysis of Male Versus Female Frontal Bone and Orbital Skeletal Morphology

Abstract

Background:

As the field of facial feminization surgery expands, plastic surgeons have increasingly sought to better understand differences in skeletal morphology between men and women. This study investigated sex-based differences in the frontal bone and orbital regions by quantitatively describing morphological variability in an adult cisgender population.

Methods:

We reviewed healthy adults who underwent head computerized tomography imaging before orthognathic surgery. Standardized sets of 501 digital landmarks were used to capture surface morphometrics of the forehead, supraorbital ridges, and orbital rims. All landmarked crania were aligned using Generalized Procrustes Analysis. Principal component analysis evaluated the contribution of sex to the overall variance in coordinates. Using robust rank aggregation, a ranked list of important landmarks or “features” most predictive of sex was generated after integrating mean feature differences, linear discriminant analysis, and random forest importance scores.

Results:

A total of 154 patients were analyzed, including 70 (45%) men and 84 (55%) women imaged at 23.6 ± 7.7 and 26.0 ± 11.1 years of age, respectively. A list of 119 important features with significant aggregate rank scores (P < 0.05) demonstrated the greatest sex-based variability at the (1) upper forehead, (2) glabella and medial brow prominences, and (3) outer infraorbital rims, primarily in the posterior/anterior direction.

Conclusions:

We identified key anatomical regions of the upper facial third with the greatest sex-based variability. Our findings provide an objective framework upon which continued investigations of skeletal dimorphisms may be performed and provide preliminary data to help guide preoperative counseling and surgical planning for facial feminization surgery.

Takeaways

Question: How can sex-based differences in the frontal bone and orbital regions be quantified in an adult cisgender population, and which anatomical areas exhibit the greatest morphological variability?

Findings: After integrating 3 ranked links of important features, we found the greatest sex-based variability at the (1) upper forehead, (2) glabella and medial brow prominences, and (3) outer infraorbital rims, primarily in the posterior/anterior direction.

Meaning: Our findings provide an objective framework upon which continued investigations of skeletal dimorphisms may be performed and provide preliminary data to help guide counseling and surgical planning for facial feminization surgery.

INTRODUCTION

Upper facial third morphology is a significant determinant of perceived gender by layperson observers,^1–3 with subjective femininity and masculinity dictated by an interplay of glabellar and supraorbital prominence, forehead dimensions, and orbital shape. The male forehead and brows are said to be more pronounced due to a thicker anterior table of the frontal sinus and supraorbital ridges,^3,4 producing the appearance of a more acute nasofrontal angle. Although orbital volume has been observed to be greater in men, women appear to display a rounder orbital aperture.⁵ With regard to the overlying soft tissues, eyebrow position/shape and hairline location come into play.³

Facial feminization surgery (FFS) can have a transformative impact on psychosocial well-being and quality of life in patients experiencing gender dysphoria.^6,7 The field of FFS is rapidly growing, owing to societal progress toward inclusivity and equity as well as health care legislation improving access to transgender health care.^8–10 In this shifting sociopolitical landscape, plastic surgeons have increasingly sought to better understand sex-based differences in facial morphology. However, current treatment techniques are largely based on subjective principles and assessments.^11,12

There is minimal evidence in the plastic surgery literature with which a patient’s anatomy can be objectively placed in the context of natural morphological variations within a population.¹³ Although identifying surgical candidates and determining operative goals are individualized decisions, it is important to establish normative data to help guide preoperative counseling and define optimal endpoints of intervention. Borrowing from the field of biological anthropology, this study uses 3-dimensional (3D) geomorphometry to investigate sex-based differences in the fronto-orbital region by quantitatively describing average morphology and variability in an adult cisgender population.

METHODS

Study Population

Following institutional review board approval at the Children’s Hospital of Philadelphia and Yale-New Haven Medical Center, we performed a retrospective review of patients of at least 18 years of age and therefore skeletally mature who underwent computed tomography (CT) imaging of the cranium before planned orthognathic surgery. Patients with a history of craniofacial trauma, genetic syndromes, metabolic growth disorders, or exogenous hormone exposure were excluded.

Landmarking Approach

Fine-cut (≤1 mm) head CT scans were imported into Mimics Version 23.0 (Materialise, Leuven, Belgium). After bony segmentation with maximal capture of the frontal bone and orbits, 3D reconstructions were exported to Checkpoint (Stratovan, Davis, CA). Following the landmarking approach detailed in Bardua et al,¹⁴ standardized sets of 501 digital landmarks were used to capture surface morphometrics of the forehead, supraorbital ridges, and orbital rims. Three templates were constructed using surface meshes onto which anatomically defined landmarks and sliding semilandmarks forming “curves” between landmarks were manually placed. Semilandmarks were semiautomatically projected in the patching step, during which templates are warped to the biological shape of each cranium and termed “patches.” All curve and surface points were then slid to optimize the geometric correspondence across all semilandmarks.¹⁴

Twelve landmarks were established for template placement based on standard definitions of anthropometric reference points. Landmarks used for the frontal patch template (Fig. 1) included bregma, forehead center (point of maximal convexity at the frontal bone midline), glabella, nasion, and outer frontozygomatic suture bilaterally (outer edge of the frontozygomatic suture). Bilateral landmarks used for the orbital patch templates included orbitale, supraorbital notch, and inner frontozygomatic suture (inner edge of the frontozygomatic suture) (Fig. 2). All landmarked crania were aligned using generalized Procrustes analysis to assess for differences in shape without the influence of scale.¹⁵

Fig. 1.

The 6 anatomical landmarks used for the frontal patch template were bregma, forehead center, glabella, nasion, and bilateral outer frontozygomatic sutures. Template control points were manually positioned upon placement of the anatomical landmarks, and semilandmarks were automatically generated by Checkpoint based on the anatomical landmark and template control point positions. The diagrams show a simplified version of the coordinate system applied, with the purple, green, and orange lines corresponding to each other in the 3D (A) and 2-dimensional (B) visualizations. L, left; R, right.

Fig. 2.

The 6 anatomical landmarks (red) used for the orbital patch templates bilaterally were orbitale, supraorbital notch, and inner frontozygomatic suture. Template control points (blue) were manually positioned along the orbital rim upon placement of the anatomical landmarks, and semilandmarks (purple) were automatically generated by Checkpoint based on the anatomical landmark and template control point positions. L, left; R, right.

Statistical Analysis

Analysis of aligned landmark coordinates was conducted using R 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). Using the geomorph R package (version 4.0.6),¹⁶ missing values were imputed via the thin-plate spline interpolation method. The overall significance of patient metadata (ie, sex, age, race, and ethnicity) was assessed using the Procrustes analysis of variance test (procD.lm function) including these demographic covariates with default parameters. Principal component analysis (PCA)¹⁵ assessed the contribution of demographic variables to the overall variance in landmark coordinates. Each landmark had a primitive x, y, and z coordinate for which negative and positive values corresponded to its position in the left/right, posterior/anterior, and inferior/superior directions, respectively. The PCA was computed with each point and corresponding axis as a separate variable without scaling. (See figure, Supplemental Digital Content 1, which displays the principal component [PC] loading scores across all subjects for the top 6 PCs [first tab], across only men for the top 6 PCs [second tab], and across only women for the top 6 PCs [third tab], https://links.lww.com/PRSGO/E278.)

Male versus female differences in means and variances were assessed using the 2-sample t test and F test, respectively, and corrected for multiple hypothesis testing using the Benjamini–Hochberg procedure. (See figure, Supplemental Digital Content 2, which displays the mean absolute values of each feature for men and women along with differences in mean values between men and women; significance was assessed using a 2-sample t test and adjusted for multiple hypothesis testing using the Benjamini–Hochberg method [first tab]. For variances of each feature between men and women, significance was assessed using an F test and adjusted for multiple hypothesis testing using the Benjamini–Hochberg method [second tab], https://links.lww.com/PRSGO/E279.) For each variable, a Euclidean distance was additionally computed to represent an overall male versus female distance in 3D space. (See figure, Supplemental Digital Content 3, which displays the Euclidean distance computed for each feature to represent the overall distance between average male and female coordinates in the 3D space, https://links.lww.com/PRSGO/E280.) To assess facial symmetry, bilateral landmarks were paired with each other for the calculation of Pearson correlation coefficients.

To identify landmarks with a significant contribution to sex-based differences in frontal bone and orbital morphology, 3 methods were integrated. First, for each variable or “feature,” we computed the absolute value of the mean difference between men and women and ranked the features by magnitude of difference. Second, we applied a linear discriminant analysis, which finds a linear combination of features to separate men and women along a single axis. Each feature received a weight indicating its contribution to the separation, and the magnitude of these weights was used to rank the importance of features. Third, we used supervised machine learning and trained a random forest classification model via the randomForest function¹⁷ to use its feature selection function. The model parameters included 1000 trees, 40 variables randomly sampled at each split, and a node size of 10. The classification accuracy was evaluated via cross-validation with 500 iterations, in which the data were randomly split with 70% used for training and 30% used for testing, resulting in a mean accuracy of 81.7% (95% CI: 71.7–91.3). The mean decrease in Gini impurity value for 1 iteration of the random forest training model was used to rank the importance of each variable for gender classification. Finally, robust rank aggregation was used to integrate the 3 ranked lists via the aggregateRanks function¹⁸ to identify the features ranked toward the top of each of the 3 lists. An aggregated rank significance score of a P value of less than 0.05 was used to define important features. (See figure, Supplemental Digital Content 4, which displays the feature selection used to determine important distinguishing features between men and women. The table shows the relative and absolute differences in male and female mean values, the linear discriminant analysis weights and magnitudes, random forest feature importance scores, and the aggregate rank scores, https://links.lww.com/PRSGO/E281.)

RESULTS

Study Population

Medical records were screened for 331 adults who underwent preoperative CT imaging for planned orthognathic surgery. Of these patients, 106 were excluded due to a metabolic growth disorder or genetic syndrome, 7 were excluded due to craniofacial trauma, and 3 were excluded due to the use of exogenous hormones. After excluding an additional 61 patients whose scans had inadequate visualization of the cranium, 154 patients were included in the study population: 70 (45%) men and 84 (55%) women who underwent imaging at a mean age of 23.6 ± 7.7 and 26.0 ± 11.1 years, respectively (Table 1). Procrustes analysis of variance testing demonstrated that sex and race were significantly associated with the population variability (P = 0.001 and P = 0.001, respectively).

Table 1.

Demographic Characteristics of the Study Population (N = 154)

	Male (N = 70)	Female (N = 84)	P
Age at CT scan, y	23.6 ± 7.7	26.0 ± 11.1	0.249
Race, n (%)
White	54 (77)	65 (77)	0.987
Asian	10 (14)	13 (15)
Black or African American	2 (3)	2 (2)
Other	4 (6)	4 (5)
Ethnicity, n (%)
Hispanic/Latino	5 (7)	6 (7)	1.000
Not Hispanic/Latino	65 (93)	78 (93)

Population Variability by Sex

The Procrustes averages of landmarks in men and women demonstrated an appreciable difference in forehead inclination, degree of glabellar and brow prominence, nasofrontal angle, and orbital contour between sexes (Fig. 3). The PCA revealed that this variability was captured in PC5 and PC6 (Supplemental Digital Content 1, https://links.lww.com/PRSGO/E278), with scatterplots illustrating a notable separation between sexes (Fig. 4). These 2 PCs collectively separated sex with high accuracy, as indicated by an area under the curve of 0.84, and accounted for 9.9% of the overall population variance. The top contributors to these PCs included the y coordinates of the orbitale and adjacent points along the outer infraorbital rims as well as between the glabella and nasion. (See table, Supplemental Digital Content 5, which displays the top variables contributing to PC5, https://links.lww.com/PRSGO/E282.) (See table, Supplemental Digital Content 6, which displays the top variables contributing to PC6, https://links.lww.com/PRSGO/E283.) From the full list of 1527 analyzed features, a ranked list of 119 important features, defined as those with significant aggregate rank scores (P < 0.05), was generated to identify features most predictive of sex in our population. These features were geographically concentrated around the (1) upper forehead, (2) glabella and medial brow prominences, and (3) outer infraorbital rims (Fig. 3). Importantly, these features were predominantly along the y axis with some differences along the z axis, suggesting high sex-based variability in anteroinferior projection in these regions, with negligible differences along the x axis. There was a significant positive correlation between sides for the 96 paired landmarks in both men (r = 0.704, P < 0.0001) and women (r = 0.704, P < 0.0001).

Fig. 3.

The frontal and lateral views of the Procrustes average male (blue) and female (pink) specimens are overlaid to demonstrate global morphological variations between sexes. Correlating with these images are graphs illustrating male vs female average differences in positioning of landmarks along the 3 axes. Landmarks were categorized as upper forehead, forehead center, glabella, medial brow ridges, lateral brow ridges, upper orbits, lower orbits, and orbitale. Positions along the y axis (top left graph) represent differences in the posterior (negative) or anterior (positive) direction. Positions along the x axis (bottom left graph) represent differences in the left (negative) or right (positive) direction. Positions along the z axis (right graph) represent differences in the inferior (negative) or superior (positive) direction.

Fig. 4.

PCA of morphometric data with PCA projections shown for the top 6 PCs, colored by demographic characteristics of the population (sex, race, and age). The percentage variance contributing to each PC is shown in parentheses and represented in a scree plot across the top 20 PCs.

Euclidean distances were calculated to represent overall male versus female differences in the 3D space (Supplemental Digital Content 3, https://links.lww.com/PRSGO/E280). Only landmarks for which mean differences achieved statistical significance (Supplemental Digital Content 2, https://links.lww.com/PRSGO/E279) are highlighted here. Starting with the frontal patch, the overall distance between the average male and female forehead center was 2.80 mm; broken down by the y and z axes, this point was positioned 2.20 mm superiorly and 1.72 mm posteriorly in men compared with women. Moving cephalically toward bregma, overall distances incrementally decreased. The overall distance between the average male and female glabella was 2.12 mm; broken down by the y and z axes, this point was positioned 1.60 mm anteriorly and 1.43 mm inferiorly in men compared with women. Moving laterally from the glabella along the brow ridges bilaterally, overall distances incrementally decreased from 1.94 mm adjacent to the glabella to 0.35 mm closer to the outer aspects of the frontozygomatic sutures. In the orbital patches, distances between points on the right and left sides were averaged. The overall distance between the average male and female orbitale was 1.98 mm, which represented the greatest male versus female difference along the entire orbital rim; broken down by the y and z axes, this point was positioned 1.90 mm posteriorly and 0.17 mm superiorly in men compared with women.

Within-sex Variability

When analyzing morphological variability within the male and female populations separately, range was similar between sexes (Supplemental Digital Content 1, https://links.lww.com/PRSGO/E278). The PCA projections and corresponding scree plots are shown for men and women separately, and similar distributions of variability across the PCs are observed (Fig. 5). A PC1 loading correlation plot demonstrated that the magnitude of the contribution of each feature to PC1 in men and women, respectively, was highly correlated (r = 0.92, P < 2.22e − 16, Fig. 5), suggesting that morphological variability is explained by similar features in both sexes.

Fig. 5.

Comparison of patterns of morphological variability within the male and female populations. PCA projections and corresponding scree plots are shown for male and female populations analyzed separately, demonstrating similar distributions of variability across PCs. The magnitude of the contribution of each feature to PC1 is highly correlated between men and women, demonstrating that variability in both sexes is determined by similar features.

Population Variability by Race and Age

The PCA revealed that population variability based on race was captured in PC3, with a scatterplot illustrating a notable separation between the White and Asian populations (Fig. 4). The top contributors to this PC included landmarks at the inner and outer frontozygomatic sutures as well as lateral and outer supraorbital rims along the z axis, suggesting high race-based variability in inferior/superior position. (See table, Supplemental Digital Content 7, which displays the top variables contributing to PC3, https://links.lww.com/PRSGO/E284.) Population variability based on age was captured in PC4 (Fig. 4), and the top contributors to this PC largely fell along the outer border of the frontal patch with no clinically meaningful geographic patterns. (See table, Supplemental Digital Content 8, which displays the top variables contributing to PC4, https://links.lww.com/PRSGO/E285.)

DISCUSSION

Anthropologists, developmental biologists, and forensic scientists have long used historical specimens, primarily in the museum space,^19–21 to understand the evolution of human biological form. The present study applies these established methods of shape analysis to create a normative anatomical “map” with which skeletal dimorphisms can be quantitatively studied at the population level. A comparative analysis of frontal bone and orbital morphology revealed the most notable sex-based variability at the upper forehead, glabella and medial brow prominences, and outer infraorbital rims, whereas the spectrum of morphological variability within sexes was comparable. These findings provide an objective framework upon which continued investigations of skeletal dimorphisms focused on key anatomical regions may be performed. Our preliminary data may also help guide preoperative counseling and surgical planning for facial feminization.

The Frontal Bone

It has been widely accepted that the male forehead overlying the frontal sinus is more prominent than that of females,¹¹ rendering frontal sinus setback and contouring foundational facial feminizing procedures. In his seminal 1987 article, Ousterhout¹¹ proposed a classification system to guide surgical management based on frontal sinus morphology. For mild brow prominences and/or frontal sinus agenesis, burring alone opens the nasofrontal angle. For moderate bossing with a normal frontal sinus, augmentation of the concavity superior to the protrusion with a prosthesis is recommended in addition to bony contouring. Finally, prominent bossing requires an anterior wall osteotomy to achieve adequate bone reduction. However, even the development of these ubiquitously referenced guidelines was based on purely subjective observations of preserved skulls. As such, true craniometric studies of these tenants are lacking.

Several anthropological studies have found that the glabella and mastoid process show the greatest sexual dimorphism,^20–22 and later imaging studies have reported that men display a markedly more acute nasofrontal angle by 14.5 degrees, greater posterior forehead inclination by 3.7 degrees, and wider glabella by 10.5 mm compared with women.²³ Building on these assessments, our analysis of frontal bone morphology found the most pronounced sex-based variability along the middle of the forehead, glabella, and medial brow prominences. Near the frontal midline, men exhibit a 1.60-mm anterior and 1.43-mm inferior projection of the glabella, transitioning to a 1.72-mm posterior and 2.20-mm superior retrusion of the forehead center compared with women. These differences align with several studies detecting a flatter, taller forehead with greater posterior inclination in men compared with a more vertical, steeper forehead inclination with a smoother arch observed in women.^24,25 As these patterns represent population trends, any or all may be present to varying degrees in a given patient population. Importantly, the morphological features of patients seeking FFS might be more “masculine” than our cohort’s average men, as they desire surgical modification. Our findings confirm the principles of Ousterhout¹¹ and support interventions aimed at reducing the glabellar prominence to soften the nasofrontal angle and, in conjunction with soft tissue procedures such as hairline advancement, augment the appearance of a shorter, more convex forehead.^5,26–30

The Orbits

Women are often said to have a larger orbital aperture with less pronounced brows.³¹ As such, the supraorbital rim has become a frequent target of intervention in FFS. As described by Sykes and Moore,²⁶ this involves burring of the outer supraorbital rims during forehead reshaping.^5,26,32,33 Supporting these interventions, we found more inferior projection of the lateral brow prominences in men compared with women. However, these differences were relatively minimal, with average male versus female differences reaching less than 1 mm when approaching the frontozygomatic sutures. Interestingly, shaving the upper aspect of the orbital rim has sometimes been perceived by patients to provide inadequate feminization,^5,34 prompting strategies such as more radical reshaping via ostectomy of the superolateral rim and burring beneath the superior border of the aperture to elevate the bony margin and expand the space of the upper eyelid sulcus.⁵ However, drastically contouring this region to “feminize” the brow is not strongly supported by our data. Rather, feminization may be wholly a product of soft tissue manipulation (ie, lifting of the lateral brows), which may contribute to the appearance of a larger orbital aperture,³⁴ a notion that itself requires more quantitative evidence. In our population, the female orbit is slightly more oblong, whereas the male orbit appears rounder. Overall, greater attention should be paid to other areas of the orbits to explain these observations.

Moving to the lower orbits, our analysis highlighted relatively greater sex-based differences along the outer infraorbital rims. The sagittal relationship between the globe and infraorbital rim is an important component of facial contour. Any sagittal imbalance or deformity of the infraorbital rim, as can be caused by traumatic injury or age-related skeletal retrusion,³⁵ may influence perceived gender. Moreover, increased anterior projection of the bony orbit in relation to the globe is often accentuated in patients with reduced orbital and eyelid fat, resulting in a hollow, deep-set appearance of the eyes.²⁶ Although we found greater inferior projection of the lateral brow prominences and upper orbits in men compared with women, albeit minimal, there was relatively greater sex-based variability at the lower orbits. Specifically, the orbitale were projected 1.90 mm anteriorly in women compared with men. Although bony contouring of the lower orbits is not typically performed for facial feminization, malar augmentation for midface projection via alloplastic malar implants has become increasingly popular,^36,37 with long-term studies demonstrating safety and favorable aesthetic outcomes.³⁸ A fuller malar region with prominent malar eminences may be achieved.³⁷ Our results newly suggest that malar projection at the lower orbits may contribute more significantly than supraorbital morphology to sex-based differences in orbital shape. The influence of this anatomical region on perceived femininity requires further investigation.

Demographic Considerations

Although not a focus of this study due to small subgroup sample sizes, we found the most notable race-based variability between the White and Asian populations in the inferior/superior directions. This variability was observed around the frontozygomatic sutures and lateral and outer supraorbital rims, which may justify differences in midfacial reconstruction approaches between these 2 populations. Although White patients receive malar implants to increase zygomatic projection, East Asian patients more often desire reduction malarplasty to reduce anterolateral protrusion of the zygomatic body and process and may seek further surgery when high cheekbones are accompanied by prominent upward-slanting palpebral fissures.³⁹ This is because the East Asian face has been described as mesocephalic, and a prominent zygoma combined with a flat nose may make the face appear even flatter.⁴⁰ We refrain from drawing definitive conclusions without larger subgroup sizes. Likewise, the relationship between age and fronto-orbital morphology requires further investigation to corroborate prior observations such as increased supraorbital projection in women with age.⁴¹ Indeed, youth and femininity are tied together, and so there is a significant overlap in surgical interventions for facial feminization and rejuvenation.

Study Limitations

This study has limitations, most notably the lack of racial and ethnic heterogeneity among our study population. As morphological variations between groups of different genetic ancestry have been described,⁴² a more demographically diverse cohort is necessary to generalize conclusions. Moreover, patients seeking orthognathic surgery for isolated class III malocclusion or aesthetic purposes may still exhibit atypical skeletal features. Despite our strict exclusion criteria including craniofacial trauma and syndromes, there are inherent limitations to a surgical cohort. Importantly, although we studied skeletal morphology, the soft tissues also play a major role in facial contour, and plastic surgeons should consider selective alterations to the skeletal framework, either alone or in concert with the overlying soft tissues, to optimize the results of feminizing procedures of the upper facial third and orbits. Therefore, our findings may not fully capture the variability present in a true normative population. Future studies using datasets from broader, nonsurgical populations could help validate and expand upon these results.

Despite these caveats, this study uses rigorous methods of shape analysis to compare frontal bone and orbital morphology between sexes without a priori assumptions about expected differences. Mapping a patient’s features against population averages can help identify areas that deviate from desired endpoints and justify the need for and extent of, in this case, frontal sinus setback and contouring, orbital rim contouring, and/or malar augmentation. Overall, our findings offer a data-driven approach to supplement existing subjective assessments. Integrating these measurements into virtual surgical planning may further enhance the precision and predictability of outcomes. The ultimate goal of FFS is to improve congruity between self-perceived gender and perception by others, so there is also a crucial role for research on subjective gender perception. Future studies should aim to integrate layperson assessments or patient-reported outcomes with objective evaluations of skeletal changes, taking into consideration the normative values established in this study.

CONCLUSIONS

A geomorphometric analysis of frontal bone and orbital dimorphism revealed the greatest degrees of sex-based variability around the upper forehead, glabella and medial brow prominences, and outer infraorbital rims. Our findings may help inform preoperative evaluation and planning for feminizing procedures of the upper facial third and orbits, and constitute a firm basis upon which more focused research on skeletal dimorphisms in these key regions may be conducted.

DISCLOSURE

The authors have no financial interest to declare in relation to the content of this article.

ACKNOWLEDGMENT

The authors would like to thank Samantha L. Cox, PhD, for her assistance in developing the landmarking methodology and China N. Byrns, MS, PhD, for her assistance with data visualization.