Ethical statement
This study protocol was approved by the Ethics Committee of the Institutional Review Board of the Faculty of Science of Charles University. The research was conducted in accordance with the ethical standards defined in the 1964 Declaration of Helsinki and its subsequent amendments. All subject's parents or legal representatives were instructed about the scanning procedure. By signing an informed consent form, they agreed to the anonymous use of 3D facial models for this research project.
Research participants
The analysed set of patients was part of an ongoing long-term research project investigating the modelled facial growth and development in individuals with or without craniofacial deformities during childhood and adolescence. The complete study sample included non-syndromic patients with isolated cleft lip (CL) and complete unilateral cleft lip and palate (UCLP) aged 0.2–2.0 years, compared to age-matched healthy children. The original set of respondents contained both right-sided and left-sided clefts. In subsequent analyses, they were all unified on the left side. All research participants were divided into followed time periods: T0 0.2–0.5 years; T1 0.6–1.0 years; T2 1.1–1.5 years; T3 1.6–2.0 years. The age distribution was created according to the recommendations of the World Health Organization. The total number of research subjects enrolled in the study was summarised in Table 1.
Facial scanning of all evaluated children was accomplished at the Laboratory of 3D Imaging and Analytical Methods, Faculty of Science, Charles University. Only cleft individuals meeting the following criteria were subsumed in the study: non-syndromic cleft patients (CL, UCLP) who underwent early neonatal cheiloplasty and were operated by MUDr. Jiří Borský, Ph.D., full-term neonates (37 + 1 up to 41 + 6 weeks) of Czech origin in excellent health condition without any life-threatening diseases. In contrast, children with other craniofacial deformities, operated with later cheiloplasty or another plastic surgeon, premature or sick children, or newborns with congenital syndromes were excluded from the study.
The control group consisted of healthy age-matched infants and toddlers (0.2–2.0 years of age) who were full term (37 + 1 up to 41 + 6 weeks), of Czech nationality, without craniofacial deformities, had not undergone any plastic surgery and had no impaired facial muscle functions. All individuals with visible facial malformations or trauma were excluded from the study.
Surgical interventions up to two years of age
According to Czech surgical protocol, early neonatal cheiloplasty (6 ± 4 days after birth) followed by palatoplasty (10.5 ± 3.5 months of age) was performed during the patient's first two years of life. All cleft individuals (CL, UCLP) included in the analyses underwent early neonatal cheiloplasty before T0. The UCLP patients were stratified such that T0 and T1 correspond to the state before palatoplasty, whereas T2 and T3 represent the condition after palatoplasty.
Early neonatal cheiloplasty (ENC) was accomplished in all analysed cleft patients using a uniform modified method according to Tennison [27] at the Motol University Hospital in Prague, Czech Republic. Early lip surgery may be achieved only in a hospital providing specialised technical equipment focused on the neonatal operation. An experienced multidisciplinary team of specialists with extensive knowledge in newborn care ensured the proper progress of the entire surgery. All specifics of the neonatal period should be fully respected during the operation.
The basic preconditions for the successful operation were excluding all life-threatening diseases and syndromes and quality analgesia, oxygenation, hydration, and sufficient thermal well-being of the newborn. Before the operation, laboratory and paediatric examinations were required and subsequently, the functionality of the cardiac system was checked. If the child was eligible for surgery, parents or legal representatives were informed of the risks and benefits of early surgery and then signed an informed consent form.
After the newborn was placed under general anaesthesia, the airways were secure. An otorhinolaryngologist initiated the operation, performed rhinoepipharyngoscopy, and evaluated the tubal torus, otomicroscopy and paracentesis. High-frequency tympanometry and determination of otoacoustic emission followed to prevent irreversible changes in the middle ear cavity. If secretion in the middle ear cavity was found, a spot tympanotomy was performed, and the secretion was eventually removed [28]. The next part of the surgical protocol was the collection of the jaw impression performed by a plastic surgeon. It was necessary to protect the tracheal tube through dislocation during this procedure.
In patients with UCLP, a modified Tennison method was used to add two mucosal lobes to deepen the vestibule and the bottom of the nasal cavity and complete the upper oral atrium. At the end of the surgery, soft tissues were used to close the alveolar cleft. The total length of the operation varied depending on the severity and extent of the cleft, and it usually took around 60–120 minutes. After the lip surgery, the patients were transferred to the neonatological ICU for further care, where they stayed for an average of 3 days [29].
The next step in treating UCLP patients was palatoplasty completed between the 0.6–1.0 years of a child's life, depending on the patient's health condition and the extent and severity of the cleft. In general, palatoplasty aimed to restore the function of the soft palatal muscles (levator veli palatini muscle and tensor veli palatini muscle) with a pathological course that prevented normal palatal function. Patients with UCLP underwent Furlow double reversing Z-plasty [30]. The soft palate was extended with mirror-inverted "Z plastics" and single-stage hard palate closure was performed simultaneously, along with the release of lateral incisions around the maxilla tubers to relieve excessive suture tension.
3D facial scanning
For morphometric analysis, three-dimensional facial models of cleft and healthy children from 0.2 to 2.0 years were used. The 3D stereophotogrammetry is a safe, non-invasive method based on the optical principle of scanning. Thus, it is suitable for imaging children. Scanning of cleft children began in 2010 and has continued to the present, while scanning of healthy children was conducted from June 2020 to June 2021. The 3D images of cleft children older than 2018 were acquired by a non-invasive optical Vectra 3D scanner (Canfield Scientific, Inc., Fairfield, NJ, USA), while newer 3D images of healthy and cleft children were created by high-resolution 3dMD Face System (3dMD Inc., Atlanta, GA, USA). Several studies confirmed that the data obtained through the Vectra 3D scanner and 3dMD Face System were generally comparable; thus, the 3D images obtained using both systems could be merged. The eye and ear regions were identified to have significant errors in the 3dMD and Vectra 3D representations; therefore, these regions were excluded from the further morphometric analysis [31, 32].
Scanning of such young children was technically quite demanding. However, the advantage of 3D scanners was a very fast capturing (2 ms), while the final images were merged within 7 s. All participants were scanned under the same conditions. The parents sat in front of the scanner with the children on their laps during the scanning. Two experienced researchers performed scanning, one of whom attracted the child's attention, and the other scanned the 3D facial model. A neutral expression on a child's face was required throughout the scanning, i.e., open eyes, closed mouth, head slightly tilted and without the expression of emotions. In the case of the youngest respondents (around 0.2 years of age), the child was leaning on the parents; in older individuals, there was an effort to keep the respondent's head in a natural frontal position. Obstructing hair distorting the 3D images was captured by the headband and removed during further processing of 3D images. The final high-resolution surface model represented as a triangular mesh with texture was created using the associated Mirror PhotoTools software or 3dMD Software.
3D data processing
For further processing, each surface model was imported into RapidForm 2006 software (INUS Technology Inc., Seoul, Korea). Unsuitable images for use in the study (children with strong facial emotions, missing parts of the facial scan etc.) were discarded. The manual adjustment included trimming the unnecessary parts of the face (hair, ears, lower part of the chin, neck area). In addition, minor surface model errors were corrected, the holes created in the 3D image due to insufficient scanning of some facial parts were filled, and scans were smoothed. All scans were aligned with the cleft located on the left side of the face. Eventually, each facial model was reduced to a surface mesh with 26k triangles.
Homology of 3D facial models
Subsequent editing and morphometric assessment of 3D facial models were processed in the Morphome3cs software (http://www.morphome3cs.com/). First, it was crucial to achieve homology representing vertices with the same index defining the same anatomical characteristics. The scanner's surface images did not manifest vertex homology even after editing in RapidForm 2006 software. Coherent point drift-dense correspondence analysis (CPD-DCA), a modification of the original dense correspondence analysis (DCA) algorithm by Hutton et al. [33], was used to create homologous representations of facial surfaces [34]. In contrast to the original DCA, CPD-DCA used the automatic non-rigid coherent point drift registration algorithm to find matching vertices resulting in more accurate correspondence outside the convex hull of the landmarks.
The first stage of CPD-DCA was the application of 9 landmarks on 3D models before any morphometric analysis. The following landmarks were used in this study: exR – right exocanthion, exL – left exocanthion, enR – right endocanthion, enL – left endocanthion, N – nasion, Pn – pronasale, chR – right cheilion, chL – left cheilion and Pg – pogonion [35]. These landmarks were used for rigid pre-alignment, which accelerated convergence in the next stage. Before data evaluation, the measurement error was determined to 0.25 mm after reusing landmarks on 5 randomly selected 3D facial models, according to von Cramon-Taubadel et al. [36].
The next step was to select a base mesh according to which all other surface models (floating meshes) were aligned based on landmarks. The choice of the base mesh had very little influence on the subsequent statistics unless the surface model had many errors. The 3D models with many errors were excluded from the analysed sample. Thus, the base mesh selection may be randomised. An automatic non-rigid registration algorithm was then used to fit that base mesh on each other surface, which was homologous to the base mesh. Landmarks were no longer used in subsequent stages.
Finally, a generalised Procrustes analysis (GPA) was used to align the homologous surface representations rigidly by removing translational and rotational differences from the data. Using GPA, models can be normalised to an equal size or maintain the original size. In all subsequent morphometric analyses, the facial shape was examined in which the investigated facial segments' size was not preserved.
The visualisation of principal component analysis
In the first step, principal component analysis (PCA) was applied to the vertex coordinate matrix to quantify variability while reducing the dimensionality of the vectors. Two uncorrelated principal components (PCs) were created using a linear transformation from the original variables. Only PCs containing sufficient information to distinguish between variables were utilised in further statistical analyses. Determination of these PCs was possible from the scree plot, which displayed the percentage variability of each PC. The number of significant PCs was determined using the broken-stick rule of thumb [37]. Based on this, only the first two PCs that most affected the vector variability were included in the analyses.
A scatter plot with 95 % confidence intervals was used to visualise the PCA depicting the PC values of all objects. Applying a scatter plot, differences in variability between the cleft and control groups were observed. Each point on the scatter plot represented a specific individual and its position appointed within the PCs. The average face corresponding to the negative/positive PCs was interactively visualised in Morphome3cs software using the Principal Component Advanced function.
Superprojection
Superprojection resulted in colour-coded maps that visualised the average morphological differences of the facial surfaces of the cleft and control groups in the perpendicular direction to the base mesh. Average faces of cleft patients were first included in the superprojection analyses; therefore, morphological differences between cleft and healthy children may be observed. Generally, the shade of red on the colour-coded maps (Fig 2) indicated more retrusive facial parts located more posteriorly in cleft children than in the controls. In contrast, more prominent and anterior facial areas in cleft patients than in healthy children were described by shaded blue colour. The cool colour shades on the colour-coded maps (Fig 3) indicated more prominent facial areas in older children (T3) compared to newborns (T0). In contrast, cool coloured shades depicted facial areas that were more reduced in older children (T3) compared to infants (T0). The facial parts coloured green depicted areas where no morphological differences were detected.
Using shell difference significance, statistically significant facial morphological changes were revealed by the per-vertex t-test. The per-vertex t-test evaluated differences in distances between 2 corresponding vertices in polygon networks to identify regions where differences between two surfaces in the superprojection were significant. The visualisation resulted in a map of significance, where the blue colour indicated p-values and the differences in shades of blue marked the level of statistical significance [26].
Statistical analyses
Statistical analyses were evaluated based on the PC scores obtained from Morphome3cs software. Further assessment was completed in the Paleontological Statistics and R software. The student’s t-test was used to assess significant differences between the sexes. A mixed linear model was applied to test whether the first two PC scores were significantly related to cleft type or age. The significance level was set at 0.05.