Three-dimensional maxillary and mandibular regional superimposition using cone beam computed tomography: a validation study


This study aimed to validate a novel method for fast regional superimposition of cone beam computed tomography (CBCT) scans. The method can be used with smaller field of view scans, thereby allowing for a lower radiation dose. This retrospective study used two dry skulls and secondary data from 15 patients who had more than one scan taken using the same machine. Two observers tested two types of regional voxel-based superimposition: maxillary and mandibular. The registration took 10–15 s. Three-dimensional surface models of the maxillas and mandibles were generated via standardized threshold segmentation, and the accuracy and reproducibility of the superimpositions were assessed using the iterative closest point technique to measure the root mean square (RMS) distance between the images. Five areas were measured and a RMS ≤ 0.25 was considered successful. Descriptive statistics and the intra-class correlation coefficient (ICC) were used to compare the intra-observer measurement reproducibility. The ICC was ≥0.980 for all of the variables and the highest RMS found was 0.241. The inter-observer reproducibility was assessed case by case and was perfect (RMS 0) for 68% (23 out of 34) of the superimpositions done and not clinically significant (RMS ≤ 0.25) for the other 32%. The method is fast, accurate, and reproducible and is an alternative to cranial base superimposition.

Cone beam computed tomography (CBCT) has become a very popular diagnostic tool, with several applications in dentistry. One of these is the superimposition of CBCT scans, which has become the state-of-the-art technique for the assessment of treatment outcome, for which CBCT is indicated. It allows clinicians and researchers to better understand the treatment outcomes and improve techniques.

In medical imaging, the process of spatially superimposing three-dimensional (3D) images is called image superimposition, image registration, or fusion. There are three basic types of superimposition that clinicians need to know: (1) point–landmark-based, (2) surface-based, and (3) voxel-based. The latter and most efficient method compares non-changing reference structures in volumetric data voxel by voxel, does not depend on landmark identification as in the point–landmark-based method, and is not limited by segmentation errors as in surface-based methods.

In orthodontics and oral and maxillofacial surgery, the superimposition of CBCT scans with a large field of view (FOV) has been used to assess orthopedic and surgical outcomes. Cevidanes et al. were the first to introduce a voxel-based method for the superimposition of CBCT scans into dentistry; they used the cranial base as the reference to superimpose two or more CBCT scans obtained from non-growing patients. Despite its excellent research application, this method involves the use of different software programs and is time-consuming. Nada et al., using a different software program, tested voxel-based superimposition using either the anterior cranial base or the left zygomatic arch as the reference in non-growing patients. The FOV of the CBCT and the radiation exposure could be reduced slightly with the zygomatic arch superimposition. Despite the good results using each structure as the reference, the method used for each superimposition was also time-consuming (30–40 min).

Most of the studies mentioned above were performed to understand changes in the maxilla and/or the mandible in relation to the cranial base in large FOV scans. There are two problems with this technique: (1) a large FOV is needed to appreciate localized changes in the maxilla and (2) even with a large FOV, the changes in the mandible are not assessed accurately because the mandible can have a different position in each scan. The issue is that a large FOV exposes the patient to a higher radiation dose compared to the use of a medium or small FOV. Therefore, a different method that allows fast, reliable, and accurate 3D regional superimposition of CBCT scans with smaller FOVs and a lower radiation dose is needed.

As stated previously, the voxel-based technique is not new, however superimposition using the maxilla and the mandible as the reference is. Therefore, the aim of this study was to test the accuracy and the reproducibility of a regional superimposition method for the maxilla and mandible in non-growing patients using CBCT.

Materials and methods

Subjects and CBCT scan

The study was approved by the necessary ethics committee. The sample for this retrospective study comprised the CBCT files for two dry skulls obtained from the Oral Diagnostic Science Department of Virginia Commonwealth University and secondary data from 15 patients who had undergone either surgical treatment (coronectomy of wisdom teeth and bone grafts) and/or orthodontic treatment at a private practice. The CBCT scans were taken between April 2009 and March 2015 and the patients ranged in age from 27 to 65 years. All of the patients had either full dentitions or were partially edentulous. Inclusion criteria for the human subjects were (1) non-growing patient, with (2) two CBCT scans (T1 and T2) taken using the same machine and with the same voxel size (0.25 mm). Exclusion criteria were (1) same patient with CBCT scans from different machines, (2) CBCT scans with a different voxel size between T1 and T2.

The dry skulls images were acquired with a Kodak Carestream 9300 (Carestream Health Inc., Rochester, NY, USA) and 13.5 × 17 cm FOV, scan time of 11.3 s, set at 85 kVp, 4 mA, and 0.3-mm voxel size. Two images of each dry skull were taken, modifying its position between T1 and T2. These images were used as a gold standard since there was no bony change between T1 and T2. The patient images were acquired with an i-CAT scanner (Imaging Sciences International LLC, Hatfield, PA, USA) and 16 × 13 cm FOV, scan time of 27 s, set at 120 kVp, 8 mA, and isotropic 0.25-mm voxel size. The DICOM (Digital Imaging and Communication in Medicine) files were imported into OnDemand 3D v1.0.10.5261 (Cybermed Inc., Seoul, Korea). The T2 scan was taken between 4 and 24 months (average 12.3 months) after T1.

3D image processing

A summary of the method is given in Fig. 1 . One observer cropped the CBCT files from T1 and T2 to simulate a 10 × 5 cm FOV scan, obtaining a significant amount of the maxillary and mandibular area. The crops were done as shown in Fig. 2 ; this resulted in a total of four images: T1 mandible, T1 maxilla, T2 mandible, and T2 maxilla. The software used allows the clinician to crop in any dimension, and the inferosuperior crops are done precisely by selecting the number of slices that the user wants to keep. In the present study, 200 slices were used to simulate 5 cm of height (200 × 0.25 mm = 5 cm). The software does not allow precise cropping in the anteroposterior dimension, therefore the CBCT scans were approximately 10 cm. The images were saved in the software database.

Fig. 1
Flowchart of the method. The blue boxes are steps done using OnDemand 3D and the green boxes are steps done using VAM (Md, mandible; Mx, maxilla; T2S, T2 superimposed).

Fig. 2
(A) Full skull before crop. (B) Maxilla cropped, and (C) mandible cropped. Each image was cropped to simulate a 10 × 5 cm CBCT. The maxillary crop included the upper teeth, alveolar process, and part of the zygomatic bone, avoiding the inclusion of the zygomatic arch as a whole. The mandibular crop included the lower teeth, corpus, angle, and part of the ramus.

Two observers (L.K. and A.W.) attempted to perform the regional superimposition independently. For the mandibular superimposition, the cropped mandibular files from T1 and T2 were opened using the ‘fusion’ tab of the software. The fusion module allows the observer to manually move T2 as close as possible to the position of T1 and also allows the observer to do an automatic voxel-based superimposition. The superimposition process took approximately 10–15 s. The software reads the voxels from the whole scan in T1 and tries to match them with a similar area in T2. Although the software had a tool to focus on the voxels of a specific region of interest, this was not needed in the present study. For the maxillary superimposition, the stable areas included in the crop were the zygomatic process of the maxilla and the palate. For the mandible, the stable areas were the symphysis, corpus, and part of the ramus.

After the superimposition had been done ( Fig. 3 ), the T2 file in its new orientation was saved (T2 superimposed, T2S). One observer (L.K.) was responsible for segmenting T1 and T2S mandibular files using the ‘3D picker’ tool inside the ‘3D’ module. All the segmentations were standardized at 381–382 grey levels and the segmented files were exported in STL format (Standard Tessellation Language) using the software parameters of 0.005 reduction error and a smooth of 1. The same steps were done for the maxillary cropped area.

Fig. 3
(A) Sagittal and (B) axial views of the maxilla before superimposition, and the same (C) sagittal and (D) axial views after the maxillary superimposition. Note that in the maxillary area, the T1 and T2 images match, while in the mandible (white arrows), they do not.

One observer imported all six STL files (T1 maxilla and mandible and T2S maxilla and mandible for each observer) into VAM (Canfield Scientific, Fairfield, NJ, USA) and performed measurements with the iterative closest point (ICP) technique. The ICP measures the smallest distance between two surfaces, providing the root mean square (RMS). A RMS value smaller than 0.25 mm (the voxel size) and errors for living subjects comparable to those found with the dry skulls were required to prove that the superimposition method is accurate. The aim was to perform the measurements in stable areas not influenced by the alveolar changes. In the maxilla, measurements were made at the lower border of the zygomatic alveolar crest, anterior and posterior to the zygomatic maxillary suture ( Fig. 4 A–C) . In the mandible, the measurements were made at the basal bone of the chin prominence and distal to the mental foramens on both sides ( Fig. 4 D, E).

Fig. 4
Areas of interest measured and colour-coded map ranging from 0.4 to −0.4 mm. (A) Lateral view of the maxilla; (B) latero-inferior view of the maxilla; (C) anterior view of the maxilla; (D) lateral view of the mandible; (E) anterior view of the mandible.

To prove that the method is reproducible, the ICP was used to measure the distances between T2S of operator 1 and T2S of operator 2. The RMS value was obtained: the distance between the models should be smaller than 0.25, while 0 would be considered perfect. The measurements were repeated by the same operator after 10 days to ensure reproducibility. The results were exported to an excel spreadsheet.

Statistical analysis

The statistical analyses were done using IBM SPSS Statistics software version 22.0 (IBM Corp., Armonk, NY, USA). Descriptive statistics for the RMS were obtained separately for the dry skulls and the human subjects. Intra-examiner agreement for the measurements was assessed by means of the intra-class correlation coefficient (ICC) and descriptive statistics with mean differences and confidence intervals set at 95%, and included the living subjects and dry skulls. The values for inter-examiner reproducibility are reported individually.


Table 1 shows the descriptive analysis of the cases. The highest RMS found for the dry skulls was 0.195 and the highest mean RMS was 0.184. For the human subjects, the highest RMS was 0.241 and the highest mean was 0.105.

Table 1
Descriptive statistics for the dry skulls and human subjects. a
Min Max Mean SD
Dry skulls
Mandible right 0.075 0.099 0.087 0.017
Chin 0.021 0.152 0.087 0.093
Mandible left 0.017 0.178 0.098 0.114
Maxilla right 0.176 0.192 0.184 0.011
Maxilla left 0.170 0.195 0.183 0.018
Human subjects
Mandible right 0.040 0.241 0.105 0.070
Chin 0.031 0.154 0.100 0.044
Mandible left 0.042 0.176 0.087 0.041
Maxilla right 0.023 0.160 0.072 0.038
Maxilla left 0.045 0.160 0.092 0.040
Min, minimum; Max, maximum; SD, standard deviation.

a Numbers are the root mean square (RMS) in millimeters.

Table 2 shows the ICC and descriptive results. All the values for the ICC were higher than 98%, showing the excellent reproducibility of the measurements. The descriptive results confirmed the excellent reproducibility of the measurements, with all mean values smaller than −0.005 ± −0.013.

Table 2
Intra-class correlation coefficient (ICC) and Bland–Altman to test the reproducibility of the measurements.
ICC Bland–Altman
Mean ± SD 95% CI
Mandible right 0.994 −0.003 ± 0.011 −0.008 to 0.003
Chin 0.996 0.001 ± 0.006 −0.001 to 0.004
Mandible left 0.980 −0.005 ± 0.013 −0.012 to 0.001
Maxilla right 0.994 −0.002 ± 0.009 −0.006 to 0.003
Maxilla left 0.987 0.002 ± 0.011 −0.004 to 0.008
SD, standard deviation; CI, confidence interval.

Table 3 shows the case-by-case analysis of method reproducibility; RMS was compared between the two observers. The results for the two dry skulls were perfect for the mandible and maxilla (RMS = 0). For the human subjects, the result was perfect for 67% (10 out of 15) of the mandibles and 60% (9/15) of the maxillas. Overall the result was perfect for 68% (23/34) of the superimpositions. For the cases in which the superimposition was not perfect, the highest RMS found was 0.026 in the mandible (case 12) and 0.047 in the maxilla (case 1).

Table 3
Case-by-case analysis showing the RMS difference in each area of interest between T2S by observer 1 and T2S by observer 2.
Case Mandible right Chin Mandible left Maxilla right Maxilla left
DS1 0 0 0 0 0
DS2 0 0 0 0 0
1 0 0 0 0.047 0.030
2 0 0 0 0 0
3 0 0 0 0 0
4 0 0 0 0 0
5 0.007 0.014 0.008 0 0
6 0 0 0 0 0
7 0 0 0 0 0
8 0 0 0 0 0
9 0 0 0 0.011 0.004
10 0 0 0 0.009 0.011
11 0 0 0 0.009 0.024
12 0.026 0.012 0.020 0 0
13 0.008 0.012 0.019 0 0
14 0.003 0.005 0.009 0.008 0.010
15 0.006 0.009 0.008 0.015 0.032
Only gold members can continue reading. Log In or Register to continue

Jan 16, 2018 | Posted by in Oral and Maxillofacial Surgery | Comments Off on Three-dimensional maxillary and mandibular regional superimposition using cone beam computed tomography: a validation study
Premium Wordpress Themes by UFO Themes