Midface Hypoplasia and Cranial Base Morphology in Syndromic Craniosynostosis: A Comparative Analysis Study Using a Predictive Regression Model

Syndromic craniosynostosis (SC) is a complex congenital condition caused by the premature fusion of cranial sutures, resulting in restricted skull growth, cranial base deformities, and midface hypoplasia that often leads to functional and aesthetic complications^1,2. These abnormalities can compromise airway patency, visual integrity, and neurodevelopmental outcomes, underscoring the importance of early recognition and precise intervention^1,3. Surgical management in SC is particularly time-sensitive, as delayed correction increases the risk of raised intracranial pressure, obstructive sleep apnea (OSA), and ophthalmic complications such as proptosis and corneal exposure^1,3,4. Therefore, determining the optimal timing of cranial and midfacial advancement remains one of the most critical and challenging aspects of clinical management¹.

Conventional assessment methods for syndromic craniosynostosis often rely on descriptive growth analyses and functional thresholds to guide surgical intervention^5,6. While these approaches provide general reference points, they lack patient-specific predictive accuracy and are influenced by clinical subjectivity^7,8. In recent years, interest has grown in developing quantitative models that can anticipate craniofacial growth trajectories and inform individualized surgical planning. Among these, the Hariri-Ros-Nor regression model offers a reproducible and quantitative framework by establishing predictive relationships between cranial base landmarks and midfacial dimensions, providing clinicians with an objective tool to forecast growth and optimize surgical timing⁹.

The predictive strength of this model is particularly relevant to the most common syndromic subtypes, such as Apert and Crouzon syndromes, which exhibit distinct growth patterns and surgical requirements⁹. By using standard craniofacial CT imaging and reproducible anatomical landmarks, namely Sella, Nasion, Basion, ZMR, and ZML, the model can quantify midfacial growth with efficiency and interpretability, even in clinical settings where advanced geometric morphometric or finite-element analyses are unavailable⁹. It is especially valuable in early to mid-childhood (approximately 4-12 years), when facial growth is active and surgical timing decisions have lasting developmental implications⁹. However, the model's applicability is limited by its reliance on CT imaging, its validation within a single cohort, and potential variation in very young or late-adolescent patients, highlighting the need for its integration alongside multidisciplinary evaluation¹⁰.

Building on this foundation, the Hariri-Ros-Nor regression model provides a reproducible, patient-specific approach to predicting midfacial and cranial base growth, addressing the limitations of current descriptive frameworks. Quantifying anatomical relationships allows earlier identification of patients at risk of functional deterioration and supports proactive, data-driven surgical planning. These capabilities hold particular significance in preventing irreversible complications associated with delayed intervention, such as airway obstruction, intracranial hypertension, and ophthalmic morbidity. Therefore, this study aims to evaluate the accuracy and clinical applicability of the Hariri-Ros-Nor regression model in assessing midfacial and cranial base growth patterns among syndromic and non-syndromic pediatric populations.

This study was approved by the Medical Ethics Committee, Faculty of Dentistry, Universiti Malaya, Kuala Lumpur (DF OS1921/0082(P)). Since the research involved retrospective and anonymized computed tomography (CT) scan data without the use of any biological tissues or fluids, patient consent was not required. The methodology of this study followed the procedures outlined by Hariri et al.⁹ with several modifications. The equipment and software used are listed in the Table of Materials.

1. Subject selection and CT scan data retrieval

Retrospective cranial CT scans of pediatric patients diagnosed with syndromic craniosynostosis (SC) were retrieved from the digital imaging archive of the Craniofacial Clinic at University Malaya Medical Centre (UMMC), covering the period from November 2015 to December 2022. Additionally, prospective CT scans obtained between May 2023 and January 2024 were included.

Potential subjects were identified from the Craniofacial Clinic records and screened manually by the investigator according to predefined eligibility criteria. The inclusion criteria comprised (1) a confirmed diagnosis of syndromic craniosynostosis by craniofacial surgeons, (2) availability of a complete medical record, and (3) age <12 years. Control subjects were also under 12 years of age and required to have complete cranial and facial CT scans without any history of craniofacial anomalies. The exclusion criteria applied to all cohorts included age ≥12 years, non-syndromic or isolated craniosynostosis, incomplete clinical or imaging documentation, a history of craniofacial surgery, or midface hypoplasia associated with other syndromes. Following screening, a finalized list of eligible subjects was submitted to the Research Unit of Biomedical Imaging to obtain non-contrast CT skull datasets. Imaging data were provided on compact discs (CDs) in Digital Imaging and Communications in Medicine (DICOM) format, with acquisition protocols standardized according to departmental guidelines (axial slice thickness ≤1.00 mm, full craniofacial coverage, and consistent voxel resolution). The Biomedical Imaging unit verified dataset completeness, absence of significant artifacts, and compliance with DICOM standards prior to release. All verified datasets were securely archived and subsequently prepared for import into a 3D medical image processing software for three-dimensional reconstruction and analysis. Thirty subjects were selected in total and divided equally into three cohorts: non-operated SCgroup (SCNO), operated SC group (SCO), and normal control group (n = 10 per group).

2. 3D reconstruction and landmark identification

All imported datasets were displayed in axial, sagittal, and coronal planes, together with a three-dimensional preview. Metadata within the DICOM files allowed automatic extraction of acquisition parameters (slice thickness, voxel resolution, and number of slices). To generate the three-dimensional volumetric model, a bone mask was first created and subsequently reconstructed by selecting Segmentation → Calculate 3D in Mimics. Segmentation was further refined using the thresholding function, accessible via Segmentation → Thresholding in the main toolbar. Variability in bone definition or the presence of excessive noise can be observed when inappropriate Hounsfield Unit (HU) ranges were applied during segmentation. This issue can be resolved by applying the standard "Bone (CT)" threshold preset available in 3D medical imaging software and refining the segmentation masks using the Edit Mask function to exclude artifacts and improve structural continuity. This process ensured that only osseous structures were highlighted in the mask, thereby excluding adjacent soft tissues and reducing image noise or artifacts.

Anatomical landmarks were identified and marked using the point creation tool, located under Measurements → Create Point. For each landmark, the investigator rotated and magnified the 3D reconstruction to optimize visibility and then placed a point directly on the anatomical site with a single mouse click. The following landmarks were recorded: the Sella (S) at the midpoint of the sella turcica, the Nasion (N) at the junction of the frontonasal suture, the Basion (Ba), defined as the anterior midpoint of the occipital bone at the spheno-occipital synchondrosis, and the right and left zygomaticomaxillary sutures (ZMR, ZML) at the infraorbital rim. Minor variability in landmark positioning may occur when points are defined from a single viewing perspective. To enhance accuracy, each landmark was confirmed across multiple viewing angles by rotating the three-dimensional reconstruction and cross-referencing with axial, sagittal, and coronal planes prior to final placement. Each landmark was automatically stored within the project tree under the "Measurements" list. The complete set of landmark coordinates and labels was exported using the Export Measurements function, producing datasets in .csv or .xls format. These landmarks were subsequently used as reference points for linear measurements of the cranial base and midfacial skeleton.

3. Linear measurement of cranial and midfacial variables

Linear craniofacial measurements were performed in 3D medical image processing software, using the distance measurement function. After three-dimensional reconstruction and landmark placement, the measurement tool was accessed through the Analysis → Measurements → Distance between points menus. In this function, two anatomical landmarks were manually selected on the 3D reconstruction, and the software automatically calculated the Euclidean distance between the selected coordinates and the results were displayed within the Measurements Results window.

The following linear measurements were obtained: SN (anterior cranial base length), defined as the distance between the sella (S) and nasion (N); SBa (posterior cranial base length), defined as the distance between the sella (S) and basion (Ba); NBa (total cranial base length), defined as the distance between the nasion (N) and basion (Ba); and ZMR - ZML (maxillary width), defined as the interzygomatic distance between the right and left zygomaticomaxillary sutures. Each measurement was performed directly on the 3D model with zoom and rotation tools applied as necessary to ensure accurate point placement.

Upon completion, the measurement list was automatically updated within the software's project tree. The complete dataset was exported using the Export Measurements function and saved in .xls format, including all landmark labels with their corresponding linear values. These exported values were subsequently used for statistical analysis.

4. Regression model application

The exported measurements were imported into a spreadsheet program for further analysis. Using the Hariri-Ros-Nor regression formula, predicted values were calculated for both the total cranial base length and maxillary width. The following equations were applied:

The predicted values derived from these formulas were then compared with the corresponding measured values for each subject to assess consistency and deviation in cranial morphology by calculating the standard deviation of the dataset. Following that, the dataset is now ready for statistical analysis. 

5. Statistical analysis

All measured and predicted values were compiled into a single dataset and analyzed using statistical analysis software. Descriptive statistics were first calculated to summarize cranial base and midfacial parameters across all groups. Inter-group differences were assessed using the non-parametric Mann–Whitney U test. Additionally, Pearson correlation analysis was conducted to evaluate the relationship between age and cranial measurements within each group. Visual representations of the data, including boxplots for group comparisons and line graphs for correlation trends, were generated to support and clarify statistical interpretations. 

According to Table 1, the measured NBa values were consistently higher than their predicted counterparts across all groups, with the greatest deviation observed in the operated syndromic craniosynostosis (SCO) cohort. In contrast, ZMR-ZML (maxillary width) values were significantly lower than predicted across all categories, demonstrating considerable variability.

Following Pearson correlation analysis (Table 2) and matrix scatter plot distribution (Figure 1) within the normal cohort, an exceptionally strong positive correlation was identified between NBa and ZMR-ZML (r = 0.992, p < 0.001), indicating a highly synchronized and proportional relationship between cranial base length and midfacial width during physiologic growth. This near-perfect association underscores the internal consistency of landmark identification and supports the robustness of the Hariri-Ros-Nor regression model under normal developmental conditions.

Additionally, Pearson correlation analysis (Table 3) revealed strong correlations between age and both NBa (r = 0.917) and ZMR-ZML (r = 0.728), further validating the model's applicability to normative craniofacial growth trajectories. Conversely, in the non-operated syndromic (SCNO) group, a significant correlation was found only for ZMR-ZML (r = 0.772), while the operated syndromic (SCO) group demonstrated weak correlations across all variables. These findings suggest that surgical intervention disrupts the inherent growth coordination between cranial base and midfacial structures, thereby reducing the model's predictive accuracy.

This interpretation is consistent with the boxplot distribution observed in Figure 2, which illustrates the distribution of measured and predicted values for NBa and ZMR-ZML across the normal control, operated syndromic craniosynostosis (SC operated), and non-operated syndromic craniosynostosis (SC non-operated) groups. Across all groups, measured NBa values were generally higher than predicted values, indicating that the Hariri-Ros-Nor regression model tended to underestimate cranial base length. This deviation was most pronounced in the SC operated group, where both measured NBa and ZMR-ZML values displayed greater variability and wider interquartile ranges compared to the normal cohort.

In Figure 3, age included as a factor is presented as the ratio distributions of measured-to-predicted values for both NBa and ZMR-ZML across the same groups. The ratios in the normal control group clustered around consistent central values, reinforcing the model's stability under normal developmental conditions. Conversely, the operated and non-operated SC groups showed markedly broader distributions, with several extreme values and visible outliers, suggesting increased prediction error and inter-individual variation. Notably, the SC operated group displayed a lower median ratio with reduced spread compared to the SC non-operated group, likely reflecting the restrictive influence of postoperative bony remodeling on craniofacial dimensional changes.

The Mann-Whitney U test (Table 4) further evaluated the median ratios between measured and predicted values within each group. For the Normal vs. SC Non-operated comparison, no statistically significant differences were observed across any parameters: RatioObservedNBa (U = 48.00, p = 0.912, r = 0.03), RatioPredictedNBa (U = 45.00, p = 0.739, r = 0.08), RatioObservedZMR-ZML (U = 48.00, p = 0.912, r = 0.03), and RatioPredictedZMR-ZML (U = 47.00, p = 0.853, r = 0.05). All results indicated negligible effect sizes, suggesting comparable distributions between the normal and non-operated syndromic groups. Similarly, comparisons between the Normal and SC Operated groups revealed no significant differences: RatioObservedNBa (U = 65.00, p = 0.280, r = 0.25), RatioPredictedNBa (U = 66.00, p = 0.247, r = 0.27), RatioObservedZMR-ZML (U = 66.00, p = 0.247, r = 0.27), and RatioPredictedZMR-ZML (U = 65.00, p = 0.280, r = 0.25). These small-to-moderate effect sizes indicate slight directional tendencies but do not reach statistical significance.

DATA AVAILABILITY:
All raw data are summarized in Supplementary File 1. The results of the Mann-Whitney U test are provided in Supplementary File 2.

Figure 1: The Matrix scatter plot revealed the midface and cranial base variables in the normal cohort. A strong correlation with a positive linear relationship was observed between the NBa and ZMR-ZML. Please click here to view a larger version of this figure.

Figure 2: Boxplot comparing measured and predicted ZMR-ZML values across all cohorts. SCO: operated syndromic craniosynostosis, SCNO: non-operated syndromic craniosynostosis. Please click here to view a larger version of this figure.

Figure 3: Boxplot of the ratio of age with the variables across all cohorts. SCO: operated syndromic craniosynostosis, SCNO: non-operated syndromic craniosynostosis. Please click here to view a larger version of this figure.

Table 1: Analysis of demographic data. SCO: operated syndromic craniosynostosis, SCNO: non-operated on syndromic craniosynostosis. Please click here to download this Table.

Table 2: Pearson Correlation coefficient value of the midface and cranial base variable in the normal cohort (**p < 0.01). Please click here to download this Table.

Table 3: Pearson Correlation coefficient value of the midface and cranial base variable in all study cohorts (**p < 0.01). SCO: operated syndromic craniosynostosis, SCNO: non-operated syndromic craniosynostosis. Please click here to download this Table.

Table 4: Mann-Whitney U results of the ratio of age with the variables across all cohorts. SCO: operated syndromic craniosynostosis, SCNO: non-operated syndromic craniosynostosis. Please click here to download this Table.

Supplementary File 1: Raw data generated during the study. Please click here to download this File.

Supplementary File 2: The results of the Mann-Whitney U test. Please click here to download this File.

This pilot study demonstrates a structured and reproducible imaging protocol that integrates landmark-based analysis with predictive modeling for quantitative craniofacial evaluation. The protocol emphasizes critical methodological steps such as standardized segmentation, precise landmark placement, and consistent measurement procedures to ensure accuracy and reproducibility. Image segmentation and three-dimensional reconstruction were performed in the 3D medical image processing software using consistent thresholding parameters to isolate bone tissue and generate volumetric models. Landmark placement accuracy was optimized using zoom, rotation, and multi-planar viewing tools, allowing each anatomical point to be verified across axial, sagittal, and coronal planes prior to measurement. Linear distances between predefined cranial and midfacial landmarks were obtained using the "Distance between points" function, with results automatically stored and exported in .xls format for statistical evaluation. Reproducibility was maintained through identical parameter settings, a single calibrated workstation, and standardized data-export procedures.

Modifications and troubleshooting were implemented to minimize measurement variability and improve segmentation reliability. In cases of inconsistent bone delineation or image noise, the Bone (CT) threshold preset was applied, and the segmentation mask was refined using the Edit Mask function to exclude artifacts and restore surface continuity. For minor three-dimensional reconstruction errors, the Calculate 3D function was reapplied following manual correction. Landmark variability, particularly in complex sutural regions, was reduced by confirming placement across multiple planes prior to finalization.

The Hariri-Ros-Nor regression model was applied to craniofacial CT data from operated, non-operated, and normal pediatric subjects to evaluate its validity across varying surgical conditions. Predicted values were most consistent in the normal cohort, confirming the model's robustness under physiologic growth patterns. In contrast, operated syndromic craniosynostosis (SC) patients exhibited the largest discrepancies between measured and predicted values, particularly in the ZMR-ZML dimension, where postoperative changes in growth trajectories introduced non-linear variability not captured by static regression equations. This finding aligns with previous reports showing that postoperative remodeling can alter expected craniofacial growth vectors and reduce the predictability of regression-based models¹⁰. Incorporating age as a covariate improved prediction accuracy, underscoring its importance as a modifying factor. Future refinements of the Hariri-Ros-Nor model should therefore integrate age dynamically and expand landmark sets to account for regional variation across syndromic subtypes such as Apert and Crouzon. Moreover, the incorporation of three-dimensional morphometric analysis and machine learning-based landmark detection could further enhance predictive accuracy and adaptability¹¹.

The Pearson correlation analysis and matrix scatter plot further demonstrated a strong proportional relationship between cranial base length and midfacial width in the normal cohort. As the cranial base elongates, the midfacial complex expands correspondingly, maintaining overall facial harmony and balance. This pattern is consistent with established craniofacial growth principles, where coordinated expansion of the cranial base provides a structural foundation for midfacial projection and symmetry^12,13. The exceptionally high correlation coefficient observed (r > 0.99) indicates minimal inter-individual variability, reflecting a stable and predictable growth trajectory in the absence of syndromic or surgical disruption.

Complementary Mann-Whitney U analyses further supported these findings, showing no statistically significant differences in either measured or model-predicted craniofacial ratios (NBa and ZMR-ZML) between the normal and syndromic cohorts. The consistently high p-values (p = 0.247-0.912) and small effect sizes (r = 0.03-0.27) indicate that cranial base and midfacial proportionality were largely preserved regardless of surgical status. This stability suggests that the regression model maintained predictive reliability across morphologically variable groups, reflecting the preservation of cranial base-midface proportional relationships even in syndromic cases. These findings correspond with prior studies demonstrating partial retention of cranial base-midfacial proportionality following reconstructive surgery, where postoperative remodeling normalizes global craniofacial proportions despite localized asymmetry^14,15. The negligible to small effect sizes observed here may also reflect adaptive remodeling and growth compensation mechanisms inherent in the pediatric craniofacial complex^16,17. Collectively, these results reaffirm the stability of craniofacial ratios across normal and SC groups and highlight the Hariri-Ros-Nor regression model's potential as a clinically reliable predictive tool adaptable to diverse developmental and surgical contexts.

This method demonstrates several advantages over conventional cephalometric and geometric morphometric techniques. Unlike two-dimensional cephalometry, which is limited by projection distortion and landmark overlap, the three-dimensional Mimics-based workflow enables true anatomical visualization and quantitative precision¹⁸. The integration of segmentation, landmarking, and measurement within a single software environment improves workflow efficiency, minimizes data transfer errors, and enhances reproducibility through standardized parameter settings. Automated Euclidean distance computation reduces observer-dependent variability, while exportable datasets ensure transparency for longitudinal and multi-operator studies. Collectively, this structured protocol establishes a reliable, efficient, and reproducible framework for craniofacial morphometric analysis, with potential applications in surgical planning, postoperative evaluation, and predictive growth modeling.

Evidently, the proposed imaging and measurement framework provides a reliable and reproducible platform for quantitative craniofacial assessment, bridging the gap between morphologic description and predictive modeling. By integrating standardized segmentation, landmark-based analysis, and regression-driven prediction, this approach enhances diagnostic precision and facilitates objective evaluation of surgical outcomes. Its efficiency and usability allow straightforward implementation in clinical workflows, supporting multidisciplinary decision-making in craniofacial centers. Importantly, the protocol's reproducibility enables longitudinal monitoring of postoperative growth and model validation across diverse patient populations, establishing a scalable foundation for future predictive and machine learning-based advancements in craniofacial surgery.