The aims of this study were to identify stable mandibular structures in 3 dimensions in growing patients using a regional implant technique and to test the reproducibility of mandibular regional superimposition in 3 dimensions using the regions identified.
Three-dimensional voxel-based regional mandibular registrations were performed on bone plates, and screws were placed in the anterior chin and symphysis regions of 20 growing patients (mean age, 12.1 ± 1.3 years). Three-dimensional models of the resulting superimpositions were built for the chin, symphysis, and third molar crypts. Absolute mean errors were calculated for each region to evaluate stability. Longitudinal cone-beam computed tomography scans were obtained of 25 patients (mean age, 12.7 ± 1.4 years) with different skeletal malocclusions (20 Class II, 5 Class III). To evaluate reliability of mandibular registrations using the chin and symphysis regions, voxel-based superimpositions were performed independently by 2 observers. The resulting superimpositions between the 2 examiners were overlaid, and the mean difference along the entire surface of the mandible was calculated.
The chin and symphysis regions showed high levels of precision (chin absolute mean error, 0.37 ± 0.16 mm; symphysis absolute mean error, 0.4 ± 0.15 mm). The third molar region had a high registration error (absolute mean error, 1.94 ± 0.06 mm). The voxel-based registrations using the chin and symphysis were reliable and reproducible between examiners (absolute mean error, 0.12 ± 1.1 mm). Intraclass correlation coefficient results showed a high degree of agreement between examiners.
The chin and symphysis regions are stable areas for 3-dimensional mandibular regional superimpositions.
Three-dimensional voxel registrations were performed in the mandible.
The anterior contour of the chin and the inner cortical plate of the symphysis are stable regions for registration.
The lower contour of the third molar crypt is not as stable as the chin or symphysis.
Registration using the combined chin and symphysis region is reliable.
The ability to quantify the influence of orthodontic and orthopedic interventions on craniofacial growth is important in orthodontics. Historically, superimpositions of 2-dimensional (2D) cephalometric radiographs using biologically stable structures have been used to differentiate between growth, displacement, and treatment effects.
To date, the most popular method used for regional mandibular superimposition is the method of Björk and Björk and Skieller for registering on the inner cortical surface of the inferior border of the symphysis, the anterior surface of the chin above pogonion, and the distinct trabecular pattern in the symphysis combined with the trabecular structure related to the inferior alveolar canal or the inferior contour developing third molar germ before root development. This method was developed based on classic investigations using tantalum implants to determine the areas of the mandible that are stable, despite craniofacial growth and remodeling.
Assessments of growth and changes caused by orthodontic treatment using conventional 2D radiographs raise 2 important issues: first, a 3-dimensional (3D) object is reduced to a 2D image, thereby limiting the regions or surfaces that can be evaluated and, second, errors in 2D cephalometrics are inherent and unavoidable. These errors include magnification and head positioning errors, landmark identification and tracing errors, and reproducibility of the reference lines or planes used for superimposition. Three-dimensional imaging can reduce many of these challenges and has been used for cranial base registration for the past decade.
The development of a reliable 3D mandibular superimposition method would provide valuable information that was previously not available using 2D techniques. The transverse dimension can be evaluated using posteroanterior cephalograms; however, there are currently no superimposition techniques with posteroanterior cephalograms to evaluate asymmetric growth. Furthermore, anatomic structures such as the condyles, which are often obstructed in 2D cephalograms, could be precisely measured and analyzed in all planes of space using 3D cone-beam computed tomgraphy (CBCT) images.
Although Cevidanes et al have published methods for cranial base superimpositions on growing and nongrowing patients using voxel-based registrations in 3 dimensions, studies evaluating 3D regional mandibular superimpositions have limitations. Koerich et al used the entire volume of the mandible for their registrations rather than specific regions. Furthermore, their sample consisted of nongrowing patients. Ruellas et al evaluated the reproducibility and accuracy of different regions for mandibular registration; however, their study lacked an external reference source such as mini-implants for a control. In addition, they did not evaluate the accuracy of fit in regions used for registration. Therefore, the aims of this study were to identify stable mandibular structures in 3 dimensions in growing patients using a modified regional implant technique, and to test the reproducibility of mandibular regional superimpositions in 3 dimensions using the regions identified.
Material and methods
CBCT scans for this study were derived from previous research and clinical databases. Patients with craniofacial syndromes or significant asymmetric growth were excluded, as were scans with double images due to motion artifacts. This study was approved by the Institutional Review Board for research on human subjects (IRB #12-1496) of the University of North Carolina.
All scans were acquired using either an i-CAT machine with a voxel size of 0.3 × 0.3 × 0.3 mm (Imaging Sciences International, Hatfield, Pa) or the NewTom 3G (at a resolution of 0.3 × 0.3 × 0.3 mm (AFP Imaging, Elmsford, NY). The settings for both scanners were 5 mA and 12 kV(p) with scan times of 20 to 25 secondss. Scans at pretreatment (T1) and 1 year (T2) for each subject were taken on the same CBCT machine.
All CBCT DICOM files were first converted to gipl format to build 3D models of the mandible using ITK-SNAP (verion 3.6; open-source software, www.itksnap.org ).
For all 3D mandibular regional superimpositions, the T1 and T2 volumes were superimposed by first roughly approximating analogous anatomic regions and then performing the final registration using a validated automated voxel-based registration program, Slicer CMF (version 3.1; www.slicer.org ). After the T1 and T2 volumes were registered, 3D segmentation of the mandibles was performed using an automated thresholding algorithm in Slicer CMF (Intensity Segmenter module). Refinement of the segmented mandibles was done using the Active Contour and Adaptive segmentation tools in ITK-SNAP. The clean segmented mandibles were converted to 3D surface models using the Model Maker module in Slicer.
To identify stable anatomic regions in 3 dimensions (aim 1), 20 Class III Bone Anchor patients with longitudinal CBCT scans were examined (mean age at T1, 12.1 years [SD, 1.3 years]; mean age at T2, 13.3 years [SD, 1.1 years]). Bone plates and screws were placed around the apex of the incisors between the lateral incisor and canine ( Fig 1 ); an area described as stable acted as regional implants analogous to the tantalum implants originally used by Björk. The bone plates and screws were confirmed to be immobile at T1 and T2 clinically.
The 2D evidence currently available was used to identify reported regions of stability for validation in 3 dimensions ( Fig 2 ), specifically, (1) the anterior surface of the chin bounded vertically from pogonion to B-point and laterally at the distal-incisal point of the right and left lateral incisors (chin), (2) the internal cortical bone of the mandibular symphysis at the lateral limit of its lingual surface and from its inferior border to the level of the center of both mental foramina (symphysis), and (3) inferior contours of the third molar germs. Superimposition using the contour of the mandibular canal was excluded because in the original work of Björk, and more recently, it has been recognized that the mandibular canal changes with mandibular growth.
T1 and T2 mandibular models were superimposed by registering on the 3D volumes of the stable plates and screws. To examine the stability of the chin, symphysis, and third molar, the absolute mean surface error was calculated from registered structures, using Slicer (Model to Model Distance and Mesh Statistics modules), for these regions. Absolute mean surface error measures the average magnitude of the errors along the entire surface of the superimposed region without considering their direction. It is more stringent at detecting smaller errors compared with root mean square calculations, which give more weight to larger errors, skewing the error estimate toward the odd outliers.
Sample size calculation for 2-way random intraclass correlation coefficients was performed with a minimally acceptable level of reliability ( H 0 :ρ 0 ) set at 0.75, with the alternative hypothesis H 1 :ρ 1 set at 0.95, with α = 0.05 and power (1-β) of 0.80. With these parameters, a sample of 25 subjects was needed to be evaluated by 2 examiners. To test the reproducibility (aim 2) of mandibular superimposition using the stable area identified in aim 1, longitudinal CBCT scans were superimposed for 25 growing patients (mean age, 12.7 years; SD, 1.4 years; cervical vertebral maturation, 2-5). The average time interval between T1 and T2 was 12.6 months (SD, 0.9 months). This sample included 10 Class II patients treated with Herbst appliances, 10 Class II patients treated with elastics, and 5 Class III patients treated with bone anchors. T1 scans were oriented to the Frankfort horizontal plane by observer 1 to a known 3D coordinate system. Two independent observers performed voxel-based superimpositions by registering T2 to T1. The resulting superimpositions between the 2 examiners were overlaid to calculate the error difference along the entire surface of the independently registered mandibles. An intraclass correlation coefficient of 0.998 shows a high degree of reliability (upper bound of 95% confidence interval, 1.000; lower bound of 95% confidence interval, 0.995).
The absolute mean registration error for the chin, symphysis, and third molar regions when superimposing on bone plates and screws are shown in the Table and Figure 3 . Normality of the data was confirmed using Kolmogorov-Smirnov goodness of fit.
|Absolute mean error||0.37||0.40||1.94|
|P value||0.765||0.698||0.0025 ∗|
The chin region showed submillimeter precision (absolute mean error, 0.37 mm; SD, 0.16 mm) of registration from T1 to T2, indicating anatomic stability over time. Similarly, the symphysis region displayed excellent precision (absolute mean error, 0.4 mm; SD, 0.15 mm). The third molar region had a high registration error (absolute mean error, 1.94 mm; SD, 0.06 mm), suggesting anatomic instability over time.
Since the chin and symphysis regions each had excellent and almost identical stability values, both areas were combined to create a single region of mandibular registration to test interobserver reliability (aim 2). Using these regions, we found a high degree of reproducibility across observers, as indicated by a low error (absolute mean error, 0.12 mm; SD, 0.11 mm) whose range fell below the threshold of measurement, with a voxel size of 0.3 mm.