Accuracy of linear measurements obtained from stitched cone beam computed tomography images versus direct skull measurements

Background: To assess whether the linear measurements obtained from stitched cone beam computed tomography (CBCT) images were as accurate as the direct skull measurements. Methods: Nine dry human skulls were marked with gutta-percha at reference points to obtain Twenty-two linear measurements on each skull. Ten measurements in the cranio-caudal plane, two measurements in the antero-posterior plane, and ten measurements in the medio-lateral plane. CBCT linear measurements obtained using stitching software were measured and compared with direct skull measurements. Results: The absolute Dahlberg error between direct linear measurements and linear measurements on stitched CBCT images ranged from (0.07 mm to 0.41 mm). The relative Dahlberg error ranged from (0.2% to 1.8%). Moreover, Intra-class Correlation Coefficient (ICC) ranged from (0.97 to 1.0) indicating excellent agreement. Conclusion: Stitched CBCT linear measurements were highly comparable to the direct skull measurements using a digital caliper.

The use of cone beam computed tomography (CBCT) machines in dentistry started in the second half of the 1990s 1 . Now, CBCT is extensively used in the dental field for implant planning, in endodontics, maxillofacial surgeries and orthodontics 2 .
In the field of orthodontics, analysis of cephalometric radiographs requires accurate identification of specific landmarks for precise measurements between these landmarks 3 . As a consequence, the small field of view (FOV) CBCT systems available in small clinics cannot yet satisfy the needs of maxillofacial surgeons or orthodontists 4 . Thus, visualizing all of these landmarks on the same scan is not always possible 5 .
In order to compensate for this shortcoming, small FOV images can be scanned and then fused together to produce a single large FOV image. However, there are few data to show whether this fused image is as precise as a single image of the whole area of interest 4,6 .
Therefore, the aim of the current study was to assess the diagnostic accuracy of stitched CBCT linear measurements versus direct measurements on skulls.

Methods
The current study was conducted on nine dry human skulls obtained from the Anatomy department, Faculty of Medicine, Cairo University to avoid the exposure of living humans to unnecessary radiation doses. 26 anatomical landmarks were identified on each skull (Table 1). Gutta percha cones (GE16121542, META BIOMED) were glued and used as radiopaque markers (Figure 1-Figure 3). 22 linear measurements were taken and recorded using a high precision sliding digital caliper (6400192, Allendale Electronics Ltd, Hertfordshire, UK). Ten measurements in the cranio-caudal plane (Table 2), two measurements in the antero-posterior plane (Table 3), and ten measurements in the medio-lateral plane (Table 4). 22 direct linear measurements were measured and were considered to be the gold standard in the study (Figure 4-Figure 6).
For soft tissue simulation, the skulls were covered with 20 layers of pink modelling wax (1mm thick each) (Tenatex eco, Kemdent) to achieve an average of 14-16 mm wax thickness 7 .
The skulls were stabilized in the Planmeca ProMax 3D Mid CBCT machine using a wooden stand passing through the foramen magnum and were oriented using the laser beams ( Figure 7). Three consecutive FOVs were scanned: two scans each of the Table 1. Showing the twenty-six anatomical landmarks identified on each skull.

Nasion (N)
The most anterior median point on the fronto-nasal suture. Infra-orbital foramen (ORF) R&L Foramen located below the infra-orbital margin of the right and left orbits.
Greater palatine foramen (GP) R&L An aperture on the right and left postero-lateral aspects of the hard palate.
Mental foramen (MF) R&L An aperture on the buccal surface of the mandible in the area of the mandibular premolars teeth on the right and left sides.
Anterior ramus (AR) R &L Point on the middle of the anterior border of the right and left ramus.
Posterior ramus (PR) R&L Point on the middle of the posterior border of the right and left ramus.

Amendments from Version 1
A new co-author has been added: Mostafa Mohamed El Dawlatly MSc., MOrth RCS(Ed), PhD, who was involved in setting the measuring protocol on the skulls, taking the study photos and in writing the article manuscript.
Any further responses from the reviewers can be found at the end of the article  size 160 ×60 mm (single arch) for the mandible and the maxilla separately ( Figure 8 and Figure 9) and one scan with a FOV size 200 ×100 mm for the upper third of the face ( Figure 10). Each one of the three FOVs was scanned separately using a voxel resolution 0.2 mm, 90 kVp and 10 mA, then stitching was performed using Romexis software (Planmeca Romexis Viewer Launcher 4.5.0.R) creating one large volume ( Figure 11).
After completion of the stitching procedure the linear measurements were obtained from the stitched CBCT images for a later comparison with the gold standard ( Figure 12 and Figure 13).

Statistical analysis
Statistical analysis was performed on SPSS (version 17). For assessment of the agreement between all measurements with     Table 3. Showing antero-posterior linear measurements.

ANS-PNS
Anterior nasal spine to posterior nasal spine.

AR(R)-PR(R)
Right anterior ramus to right posterior ramus.

Results
Error assessment of linear measurements conducted on stitched CBCT images versus direct skull measurements (the gold standard) (Table 5) The results of the current study showed that, the difference between the mean of the direct linear measurements and the mean of the linear measurements conducted on the stitched CBCT images ranged from (-0.25 mm to 0.5 mm), the  Mean absolute difference of all measurements was (0.11± 0.12 mm). Bland and Altman Lower limit of agreement ranged from (-0.92 mm to -0.06 mm). Bland and Altman Upper limit of agreement ranged from (0.1 mm to 1.24 mm).

Discussion
The smaller the scan FOV, the higher the spatial resolution of the image 8 . Stitching of small CBCT images to create a large image can be very useful to collect the needed craniomaxillofacial data with small FOV machines 9 . Increasing the FOV can be done by automatically fusing up to three small FOVs to obtain a larger FOV 10 .
CBCT ''Stitching'' option could be useful but whether it is precise enough to obtain accurate and reliable measurements remains The results of the current study showed that the relative Dahlberg error ranged from 0.2% to a maximum of 1.8%. Consequently, the error was considered small and clinically non-significant as the measurement error in craniofacial imaging is considered clinically acceptable up till the value of 5% 13,14 .
The Moreover, the results of Kopp and Ottl; 2010 10 further agree with those obtained from the current study. They used an automated method to increase the FOV of CBCT images by stitching three small FOV volumes to obtain a larger FOV one. They concluded that, the stitching software was accurate in the obtained linear measurements.
On the same line of agreement, a study was performed by Kim et al.; 2012 5 to investigate whether images of skulls obtained by both manual and automatic stitching of three CBCT images, provided accurate measurements as those obtained by multidetector computed tomography (MDCT). The results showed that the mean difference between automatically stitched CBCT images and the reference images ranged from (-0.8944 mm to -1.0628 mm).

Conclusion
Stitched CBCT linear measurements were highly comparable to the direct skull measurements. However, a percent of error should be expected from CBCT-derived measurements.

Data availability
Underlying data is available from Open Science Framework I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias You can publish traditional articles, null/negative results, case reports, data notes and more The peer review process is transparent and collaborative Your article is indexed in PubMed after passing peer review Dedicated customer support at every stage For pre-submission enquiries, contact research@f1000.com