This study was aimed at investigating the reliability and accuracy of cone-beam computed tomography (CBCT) diagnosis of contact between a temporary anchorage device (TAD) and tooth root and assessing any effect produced by metal brackets, imaging software program, and image segmentation or color enhancement tools.
Eighteen fresh pig mandibles were used. TADs (Vector, 1.4 × 8 mm) were placed at the buccal intermolar alveolar bone on both sides of the mandibles. With soft tissue kept intact, each mandible underwent CBCT scans (voxel size, 400 μm) before and after placing TADs, and after placing metal brackets on involved molars. Alveolar bone specimens containing the TADs were then exposed to microcomputed tomography (microCT) scans (voxel size, 27 μm) after TAD removal. Two independent raters, blinded of image identity, diagnosed TAD-root contact using ImageJ (National Institutes of Health and the Laboratory for Optical and Computational Instrumentation, University of Wisconsin, Madison, Wis) for microCT; Dolphin (Dolphin Imaging and Management Solutions, Chatsworth, Calif) and Anatomage software programs (Anatomage, Santa Clara, Calif) for CBCT images. Intrarater and interrater reliability and diagnostic accuracy were statistically assessed using Cohen kappa and McNemar tests.
Intrarater and interrater reliability of TAD-root contact diagnoses were perfect for microCT diagnoses (κ = 1), generally moderate to good (κ >0.5) for CBCT diagnoses except for the use of color enhancement tools (κ <0.25). For diagnostic accuracy, there was generally a low agreement (κ <0.45) between CBCT and microCT (gold standard). The percent accuracy ranged from 68.1% to 79.2% and was not different among raters, bracket presence/absence, or software choices (chi-square tests, P >0.05). Overall, diagnostic sensitivity was above 80%, whereas specificity was below 55%.
Despite good reliability, diagnoses of TAD-root contact using 400 μm voxel size CBCT imaging tend to be inaccurate, with a likelihood of high false-positive diagnoses.
Intra- and interrater reliability of CBCT diagnosis of TAD-root contact was fair.
Diagnosis with Anatomage (κ: 0.77∼0.93) was more reliable than with Dolphin.
CBCT diagnosis of TAD-root contact had a low agreement with microCT diagnosis.
The low diagnostic accuracy of CBCT was mostly because of a high false-positive rate.
Since the introduction of temporary anchorage devices (TADs) for the purpose of orthodontic anchorage about 2 decades ago, they have become increasingly popular in orthodontic treatment. , Although the overall success rate of TADs is high, ranging from 80% to 97%, several factors tend to bring an increased risk of TAD failure, including young age, poor oral hygiene, smoking, nonkeratinized gingival site, placement in the mandibular molar region, , and contact with tooth root. ,
Clinically, when the insertion site is narrow, the patient feels pain after TAD insertion, or the clinician is not certain of the designed placement site is achieved, 2-dimensional (2D) dental radiograph and cone-beam computed tomography (CBCT) are often used to assess TAD-root proximity. On the basis of imaging diagnosis and clinical observation, the clinician then decides whether the placed TAD needs to be removed and replaced to minimize adverse effects on the tooth root and TAD failure. Several recent studies have reported that CBCT offers more accurate observation of TAD-root relationship than 2D dental radiographs. , , , On the basis of a 46.5% concordance rate between periapical images and CBCT images for diagnosing TAD-root proximity, Watanabe et al recommended CBCT over conventional periapical radiographs for assisting TAD placement. However, it remains controversial whether this recommendation should be accepted, which is well reflected by the contradictory position statements issued by maxillofacial radiology societies in different countries. ,
A limited understanding of the benefits and limitations of CBCT in assessing TAD position is a cause of this controversy. To date, no study has systematically investigated the reliability and accuracy of CBCT-based diagnosis of TAD-root contact, especially with the use of a gold standard. As a metal device, TADs act as a filter positioned within the object to absorb the lower wavelength rays. This results in a beam of high energy being recorded at the detector. In the 3-dimensional (3D) reconstruction, this results in dark streaks. Streaking artifacts like these can potentially make it difficult to locate the end tips of TADs accurately. Low accuracy of CBCT diagnosis in the presence of metal has been reported in previous studies. In addition to CBCT imaging associated artifact factors, the diagnostic accuracy and reliability of TAD-root contact may also be affected by factors involved in the image analysis process, such as imaging software, rater experience, image segmentation tool, etc.
Therefore, the purpose of this study was to investigate whether these factors affect the reliability and accuracy of CBCT-based assessment of TAD-tooth root contact. We hypothesized that the presence of metal could substantially decrease the reliability and accuracy of the diagnosis of TAD-root proximity using CBCT imaging. Whether the hypothesis is accepted or not, we expected that the findings of this study would help clinicians to understand better the benefits, limitations as well as indications of CBCT in diagnosing TAD placement.
Material and methods
Fresh mandibles from 5-month-old domestic pigs ( Sus Scrofa ), equivalent to adolescent humans in craniofacial skeletal maturity, were used for this study. The sample size was determined as follows. First, because a lack of existing report on the accuracy of CBCT diagnosis of TAD-root contact, we assumed that only one might correctly judge TAD-root contact 50% of the time when the TADs are near roots in CBCT images, but may correctly judge 100% of the time when the TADs are distant from roots. Next, on the basis of a recent report that the approximate prevalence of potential TAD-root contact was approximately 60% with the other 40% to be distanced from roots, we estimated that the overall accuracy of diagnosing TAD-root contact from CBCT images is 70%. Finally, we assumed that using image segmentation tool, the accuracy was improved from 70% to 95%. To detect this change of 25%, a total sample of 36 was required to achieve an 80% power at an α level of 0.05. With each side (right, left) treated as an individual sample, a total of 18 pig mandibles were used for the study.
Pigs that were killed for reasons unrelated to the study at Ohio State University laboratory animal resources, which otherwise would be discarded, were used for collecting mandibular specimens. Immediately after pig euthanasia, the pig mandibles were separated from the pig heads with all soft tissues covering the mandibles left intact. Each mandible then underwent 3 CBCT scans as detailed below, using a Planmeca Promax 3D machine (Planmeca, Helsinki, Finland) with a scan setting of 90 kVp, 12.5 mA, 13.5 sec, and 400 μm voxel size, which is commonly used for orthodontic patients. Because of the large size of the pig mandibles, which could not be completely contained by the “mandible” field of view designed for human patients, a “double jaw” field of view was used for the specimens. During each scan, the pig mandible was centered in the positioning stage with the snout facing forward. Two lines were scribed on the mandible, one vertically positioned between 2 mandibular incisors, the other positioned horizontally passing distal to the last erupted molar on both sides, to ensure the mandible has the same position and orientation for the repeated CBCT scans.
The first CBCT scan was used to assess root position and interradicular space for TAD placement. For each half mandible, an 8 mm (length) × 1.4 mm (diameter) Vector TAS TAD (Ormco, Orange, Calif) was placed at the level of mucogingival junction between the molars by 1 operator (S.S.). The TAD was angled approximately 40° to the occlusal plane during insertion. If during insertion of TAD, root contact was felt (as hard to penetrate surface); TAD was slowly retracted and redirected to avoid root contact. An attempt was made to not fully remove the TAD and reinsert at a different location to avoid error in detection of TAD-root contact. Immediately after TAD placement, the second CBCT scan was taken. Then, 2 orthodontic brackets (Forestadent, Pforzheim, Germany) were bonded using 3M etching gel, primer, and bonding reagent (3M, St Paul, Minn) on both molars adjacent to the TAD, followed immediately by the third CBCT scan.
Subsequently, at each TAD site, a 1 cm (width) × 1.5 cm (height) box osteotomy was performed with an electric hand piece (Brasseler USA, Savannah, Ga) with the TAD and adjacent roots contained in the separated bone specimen. Each of the sections was clearly labeled for specimen identification. TAD in each of these sections were removed by 1 operator (S.S.) to eliminate any metal artifacts in microcomputed tomography (microCT). Because care was executed to prevent any change to the bone. Each specimen was then scanned using SkyScan 1172 X-ray microtomograph microCT scanner (SkyScan, Aartselaar, Belgium) at a setting of 100 kV, 100 μA, and 27 μm voxel size. SkyScan’s volumetric NRecon reconstruction software was used to reconstruct cross-section slices from acquired microCT angular projections through the object. Reconstructed slices were rotated and images saved as 16-bit tagged image format files using ImageJ (National Institutes of Health and the Laboratory for Optical and Computational Instrumentation, University of Wisconsin, Madison, Wis).
All reconstructed CBCT and microCT in digital imaging and communications in medicine file format were randomly ordered and assigned codes to conceal the identity of the specimens. Two independent, blinded raters (S.S. and A.W.) subsequently analyzed all CBCT images using Dolphin 3D (Dolphin Imaging and Management Solutions, Chatsworth, Calif) and InVivo Dental (version 5.0; Anatomage, San Jose, Calif) software programs and all microCT images using ImageJ software. One of the raters (S.S.) had previous exposure to reading CBCT images, whereas the other (A.W.) did not. Before launching the analysis, both raters received training on using the software programs, and a trial session was conducted in implementing a diagnosis protocol. The raters were then given randomly coded 3D image files for analysis. More specifically, in the sagittal view, the rater first rotated the images so that the occlusal plane tangent to the molar cusp tips were oriented parallel to the horizontal plane, then serial 2D slices at 3 orthogonal planes were viewed for the diagnosis. The slice thickness was set at 0.5 mm.
During diagnosis, the raters were allowed to freely change the contrast and magnification, the same way a clinician would be allowed to do during diagnosis. If the TAD and an adjacent root overlapped or had a less than 0.25 mm of separation (based on an average 0.25 mm periodontal ligament [PDL] thickness), it was diagnosed as the TAD being in-contact with the root, and, otherwise, it was diagnosed as being out-of-contact. If an in-contact diagnosis was made from at least 2 planes, a positive (in-contact) diagnosis was given to that sample. Otherwise, the diagnosis was negative (out-of-contract). To assess whether the image segmentation/color enhancement tool in Anatomage (Anatomage) enhanced diagnosis, reliability, and accuracy, we based the diagnosis on the same criteria after automatic application of color enhancement to the 2D slices. The 2 raters conducted the diagnosis independently. One week after the initial diagnosis, the images were randomly reordered and assigned new codes, and then the raters conducted another round of diagnosis, which were used to assess intrarater reliability.
Intrarater and interrater reliability was assessed using Cohen kappa tests. Diagnostic accuracy was evaluated in 2 ways. First, the agreement between CBCT and microCT diagnoses for each condition was assessed by Cohen kappa tests. With the microCT diagnosis used as the gold standard, diagnostic sensitivity, specificity, and accuracy of CBCT diagnoses were calculated. The differences in the percent accuracy between the presence and absence of metal brackets, between 2 software programs, between 2 raters, and between diagnoses with and without image segmentation/color enhancement tool were assessed with McNemar tests. All statistical analyses were conducted using SPSS software (version 24; IBM, Armonk, NY).
Representative CBCT images of TADs in and out-of-contact are presented in Figure 1 . Qualitatively, images in Anatomage visually appeared to be clearer, although both showed blur boundaries around TADs. Representative microCT images of TADs in- and out-of-contact of roots are presented in Figure 2 . Qualitatively, microCT images were substantially clearer and more detailed than CBCT images. Representative CBCT images after application of color enhancement tool are presented in Figure 2 . Despite the seemingly clearer outline of teeth, boundaries around TADs were still blurry.
Intrarater reliability of microCT and CBCT diagnoses under varied conditions is shown in Figure 3 . The diagnosis made from microCT imaging showed perfect intrarater reliability for both raters (κ = 1). CBCT diagnosis made in Anatomage showed substantially stronger intrarater reliability (κ range, 0.771-0.929) than those made in Dolphin (κ range, 0.366-0.536) regardless of rater and bracket presence/absence factors. Using the segmentation/color enhancement tool in Anatomage did not change the intrareliability for rater 1 but drastically decreased the reliability for rater 2 with both the presence and absence of brackets on involved molars.
Interrater reliability data are presented in Figure 4 . Like intrarater reliability, there was perfect interrater reliability for microCT images (κ = 1). Interrater reliability of CBCT diagnosis made in Anatomage tended to be higher than that made in Dolphin (κ = 0.85 and κ = 0.536, respectively). The presence of brackets in the view substantially decreased reliability in Dolphin (κ decreased from 0.536 to 0.113). Using the color enhancement tool in Anatomage also substantially decreased the interrater reliability regardless of bracket presence.