We analyzed 440 archival CBCT scans from the University of California, San Francisco (UCSF) School of Dentistry, collected from 2004 to 2007, for planning and diagnostic purposes under informed consent and institutional review board (IRB) approval from all participants for use in research, as previously published. These imaging data were subsequently approved for additional archival research (UCSF IRB number: 11-06996), as described by Young et al The scans were originally acquired using a MercuRay CBCT scanner (Hitachi Medical, Tokyo, Japan) with an estimated radiation exposure of approximately 200 mSv. Participants were seated upright, with the x-ray tube and imaging screen rotating around their heads. They were instructed to remain still, keep their teeth in occlusion, lips relaxed, avoid swallowing, and maintain their tongue against the palate with their head in a natural position. Scanner settings were 110 kVp and 10 mA, producing 512 slices in a 10-second scan, with a 19 × 19 × 19 cm field of view (FOV) and a voxel size of 0.38 mm. Images were reconstructed using CBWorks (version 2.1; Cyber Med, Seoul, Korea) and Avia (Hitachi Medical Devices, Tokyo, Japan) software and saved in digital imaging and communications in medicine format. Each scan included the participant’s ANB angle classification, sex, and age, with all other personal information anonymized.

The CBCT scans were collected at a time when large-FOV CBCT was more commonly accepted for diagnosis. It was believed that because CBCT reduced distortion of craniofacial imaging, temporomandibular joint, airways, and facial soft tissue, it would also improve measurement and treatment planning. Moreover, the risk to benefit ratio was an active area of research, and this approach was considered within safe limits per the Declaration of Helsinki. However, with the benefit of hindsight and additional information, current as low as reasonably achievable (ALARA) guidelines favor even lower-dose alternatives, such as conventional 2D radiographs for routine orthodontics, much reduced FOV, or use of ultra-low dose CBCT, when warranted. We acknowledge that current ALARA principles prioritize lower-dose imaging (eg, conventional radiographs) for routine orthodontics, particularly for young patients, because of a greater appreciation of their increased sensitivity to ionizing radiation. Guidelines from the American Academy of Oral and Maxillofacial Radiology and recent reviews emphasize that although large-FOV CBCT should not be routine, the use of archival datasets—when IRB-approved and justified—aligns with ethical research practices to derive value from past exposures without promoting outdated protocols. ^, This study is consistent with these guidelines. We use archival data to analyze facial shape alone, and not to endorse historical imaging protocols. The current study received approval from the UCSF IRB (IRB number: 11-06996) to access and study this archival data expressly to maximize the usage of a now irreplaceable dataset. The findings remain relevant to understanding esthetic perceptions, independent of imaging methods.

From the initial pool of 440 scans, 82 were selected based on the following inclusion criteria ( Fig 1 ): (1) no apparent facial asymmetries, congenital anomalies, or known syndromes; (2) no significant vertical disproportions in the lower face, defined by the Eastman normal value for lower anterior face height or total anterior face height (55%) with a standard deviation of ±2%; (3) age range of 13-50 years at the time of scanning, and (4) ANB angle of 0°-10°, measured via Steiner cephalometric analysis on reconstructed lateral cephalograms.

Inclusion and exclusion criteria. Flow chart illustrating sample generation for esthetic rating. A snapshot of a video clip shown to assessors is included ( bottom left ).

Scans were grouped by the ANB angle : (1) orthognathic: 0° to <3.6° (n = 11; 6 females and 5 males); (2) mild SSD: ≥3.6° to <6° (n = 12; 7 females and 5 males); (3) moderate SSD: ≥6° to <8° (n = 12; 6 females and 6 males); and (4) severe SSD: ≥8° (n = 5; 3 females and 2 males).

To ensure a balanced distribution and limit the assessment time to 30-40 minutes, 40 CBCT scans were randomly selected from each SSD category ( Fig 1 ). In addition, 8 scans (20% of the sample) were duplicated to test rater reliability. Six scans were found to have imaging artifacts, so they were randomly replaced with an additional 6 scans. The final sample consisted of 48 CBCT scans from 40 patients: orthognathic (n = 11), mild SSD (n = 12), moderate SSD (n = 12), severe SSD (n = 5), and duplicates (n = 8; 2 per SSD category) ( Fig 1 ). This sample included 40 participants (22 females and 18 males), with a mean age of 22.3 years (range: 14-48 years) and a mean ANB angle of 5.1° (range: 0.1°-10.6°).

For each selected CBCT scan, a 3D model of the external soft tissue surface was generated by thresholding in Amira 3D software (Mercury Software, San Francisco, Calif), as described by Young et al One operator (J.D.) reviewed all models in MeshLab (version 1.2.1; Institute of Information Science and Technologies, Pisa, Italy) to identify imaging artifacts or distortions, excluding further 6 scans because of poor resolution.

A group of laypeople aged 18+ years, with no dental knowledge or prior experience evaluating facial esthetics or sexual dimorphism, was recruited. An a priori power analysis (G∗Power3) assumed a small effect size ( d = 0.15), α = 0.05, and power = 0.90, requiring 99 raters. We recruited 108 potential assessors, who completed a demographic questionnaire and the State-Trait Anxiety Inventory questionnaire to assess psychological status, as anxiety can influence attractiveness judgments. Assessors with State-Trait Anxiety Inventory questionnaire scores >52 (indicating potential anxiety disorder) were excluded. Three were excluded for high trait anxiety and 5 for incomplete scoring, leaving 100 assessors. This group was primarily self-identified white Canadians (53%), with bachelor’s degrees (81%) and a mean age of 30 years (range: 19-65 years). Of these, 61 were first-year dental students, and 39 had diverse backgrounds. Mean state and trait anxiety scores were 31.3 (± 8.4) and 34.4 (± 8.2), respectively, indicating no clinically significant anxiety.

Assessors were divided into groups of 15-20 and evaluated the 3D models in a quiet seminar room. For each model, a standardized video clip was created, showing the face rotating around a vertical axis (aligned with the Frankfort horizontal plane) from right to left profile, with a pause at the frontal view. These clips, mimicking real-life perception, were projected in random order on a 120-in × 90-in screen using PowerPoint 2016 (version 16.36; Microsoft, Redmond, Wash). Assessors rated each face for facial attractiveness and sexual dimorphism using a 100-mm VAS with anchors: extremely unattractive to extremely attractive for esthetics and masculine to feminine for dimorphism. They had 20 seconds to view and rate each clip, with 5-second transitions, completing the process in approximately 40 minutes. No talking, eating, or drinking was allowed, and assessors could not revisit prior clips. Scores were measured (J.D.), averaged per model, and analyzed using descriptive statistics and intraclass correlation coefficients to assess rater reliability.

Shape data were derived using a semi-automated landmarking method ( Fig 2 ; Hallgrímsson et al ). An atlas was created from 1 scan, cropped, and decimated to approximately 2,500 points for computational efficiency. This atlas was nonlinearly registered to 10 random scans, then to all scans, using the optimal step nonrigid iterative closest point algorithm. Sixty-five 3D (x, y, and z) landmarks were calculated ( Fig 2 ; Hallgrímsson et al ). Procrustes superimposition aligned these landmarks into a common shape space by scaling to a centroid size of 1, rotating to minimize squared deviations, and centering on the group mean. Asymmetry was assessed by calculating deviations of left-right paired landmarks from midline landmarks, yielding symmetrical and asymmetrical components. Procrustes distances from the population mean were computed, in which lower values indicated closer alignment to the average shape, to test whether attractiveness correlates with proximity to the mean.

Soft tissue landmarks (lateral, angle, and frontal view). Landmarks ( blue , n = 65) generated using Hallgrímsson et al from an atlas-based automated registration (lateral, angle, and frontal views).

Principal components analysis was used to identify major shape variation axes in the full sample (n = 440), with analysis of variance testing for shape differences between the full and rated samples (n = 40). Multivariate regressions examined relationships between Procrustes data (and asymmetry) and continuous variables: facial attractiveness, sexual dimorphism, age, and Procrustes distance. Independent variables included size, age, and VAS scores; dependent variables were Procrustes or asymmetry data. Significance was assessed via permutation (1000 replicates). Regression coefficients quantified shape-variable relationships, and linear regression tested associations with ANB angle, age, and VAS scores for attractiveness and sexual dimorphism. Shape vectors were visualized by warping an average face (minimum Procrustes distance) into extreme configurations using a thin-plate spline algorithm in Landmark Editor (University of California Davis, Davis, Calif). Analyses were performed in MorphoJ (version 1.08.02; Klingenberg , Chris Klingenberg, Zurich, Switzerland).

The rated subsample (n = 40) showed no significant shape difference from the overall sample (n = 324, Procrustes distance = 0.002, Hotelling’s T ² = 0.27, P = 0.257; Fig 3 ). Perceived facial attractiveness was unrelated to age (r ² = 0.091, P = 0.604), asymmetry ( P = 0.131), or Procrustes distance from the mean (r ² = 0.028, P >0.05). However, shape correlated with attractiveness (r ² = 0.473, P = 0.058, 4.5% variation; Fig 4 ) and sexual dimorphism (masculinity or femininity) (r ² = 0.558, P <0.001, 9.0% variation; Fig 5 ). ANB angle influenced shape (r ² = 0.574, P <0.001, explaining 11.3% of variation; Fig 6 ) and was negatively associated with attractiveness-related shapes (r ² = 0.494, P <0.001; Fig 7 ).

Principal components analysis of facial shape. PC1-3 of shape variation ordination for the total sample ( red , n = 324) and assessed sample ( blue , n = 40). Each axis shows the population-calculated mean shape warped to the extreme using the associated eigenvector (lateral, angle, and frontal views). PC1-3 , Principal components 1-3.

Multivariate regression of averaged perceived esthetics on shape. Population-averaged esthetic scores predicted 4.5% of facial shape variation (r ² = 0.473, P = 0.058). Faces represent the sample mean warped along the shape vector to extreme values. More attractive faces have prominent chins and vertical profiles; less attractive faces have retrusive chins and convex profiles. Sex-stratified vectors for females ( circles , r ² = 0.688, P <0.0001) and males ( squares , r ² = 0.410, P <0.0001) are shown.

The distribution of perceived femininity to masculinity shapes scores binned by actual biological sex. Multivariate regression of femininity to masculinity perceptions on shape shows overlap across sexes despite significant mean differences.

Multivariate regression of ANB on shape. ANB predicted 11.3% of total shape variation ( P <0.001). Higher ANB, as independently measured from 2D radiographs, is associated with a retrusive chin and severe SSD in 3D CBCT.

Stay updated, free dental videos. Join our Telegram channel

Join